Informix Product Family

Informix
Version 11.70

IBM Informix Performance Guide



SC27-3544-06
Informix Product Family
Informix
Version 11.70

IBM Informix Performance Guide



SC27-3544-06
 Note
Before using this information and the product it supports, read the information in “Notices” on page C-1.

This edition replaces SC27-3544-05.

This document contains proprietary information of IBM. It is provided under a license agreement and is protected
by copyright law. The information contained in this publication does not include any product warranties, and any
statements provided in this manual should not be interpreted as such.
When you send information to IBM, you grant IBM a nonexclusive right to use or distribute the information in any
way it believes appropriate without incurring any obligation to you.
© Copyright IBM Corporation 1996, 2014.
US Government Users Restricted Rights – Use, duplication or disclosure restricted by GSA ADP Schedule Contract
with IBM Corp.
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
About this publication . . . . . . . . . . . . . . . . . . .
xiii . . . . . . . . . . . .
Topics beyond the scope of this publication . . . . . . . . . . .
xiii . . . . . . . . . . . .
Types of users . . . . . . . . . . . . . . . . . . . . .
xiv . . . . . . . . . . . .
Software dependencies . . . . . . . . . . . . . . . . . .
xiv . . . . . . . . . . . .
Assumptions about your locale . . . . . . . . . . . . . . .
xiv . . . . . . . . . . . .
Demonstration databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
What's new in performance for Informix, version 11.70 . . . . . . . . . . . . . . . . . . . . . xv
Example code conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxi
Additional documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxii
Compliance with industry standards . . . . . . . . . . . . . . . . . . . . . . . . . . xxii
How to provide documentation feedback . . . . . . . . . . . . . . . . . . . . . . . . . xxii

Chapter 1. Performance basics . . . . . . . . . . . . . . . . . . . . . . . . . 1-1

Developing a basic approach to performance measurement and tuning . . . . . . . . . . . . . . . 1-1
Quick start for acceptable performance on a small database . . . . . . . . . . . . . . . . . . . 1-2
Performance goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-2
Measurements of performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-3
Throughput . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-3
Response time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-5
Cost per transaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-7
Resource utilization and performance . . . . . . . . . . . . . . . . . . . . . . . . . . 1-7
Resource utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-8
CPU utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-9
Memory utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-10
Disk utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-11
Factors that affect resource utilization . . . . . . . . . . . . . . . . . . . . . . . . . . 1-12
Maintenance of good performance . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-14

Chapter 2. Performance monitoring and the tools you use . . . . . . . . . . . . . 2-1

Evaluate the current configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-1
Create a performance history . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-2
The importance of a performance history . . . . . . . . . . . . . . . . . . . . . . . . 2-2
Tools that create a performance history . . . . . . . . . . . . . . . . . . . . . . . . . 2-3
Monitor performance with the OpenAdmin Tool (OAT) for Informix . . . . . . . . . . . . . . . . 2-7
Monitor database server resources . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-7
Monitor resources that impact CPU utilization . . . . . . . . . . . . . . . . . . . . . . 2-7
Monitor memory utilization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-8
Monitor disk I/O utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-10
Monitor transactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-12
Using the onlog utility to monitor transactions . . . . . . . . . . . . . . . . . . . . . . 2-12
Using the onstat utility to monitor transactions . . . . . . . . . . . . . . . . . . . . . . 2-13
Monitor sessions and queries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-13
Monitoring memory usage for each session . . . . . . . . . . . . . . . . . . . . . . . 2-13
Using the SET EXPLAIN statement . . . . . . . . . . . . . . . . . . . . . . . . . . 2-14

Chapter 3. Effect of configuration on CPU utilization. . . . . . . . . . . . . . . . 3-1

UNIX configuration parameters that affect CPU utilization . . . . . . . . . . . . . . . . . . . 3-1
UNIX semaphore parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-1
UNIX file-descriptor parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-2
UNIX memory configuration parameters . . . . . . . . . . . . . . . . . . . . . . . . 3-3
Windows configuration parameters that affect CPU utilization . . . . . . . . . . . . . . . . . . 3-3
Configuration parameters and environment variables that affect CPU utilization . . . . . . . . . . . . 3-3
Specifying virtual processor class information. . . . . . . . . . . . . . . . . . . . . . . 3-4
Setting the MULTIPROCESSOR configuration parameter when using multiple CPU VPs . . . . . . . . 3-10

© Copyright IBM Corp. 1996, 2014 iii

 Setting the SINGLE_CPU_VP configuration parameter when using one CPU VP. . . . . . . . . . . 3-10
Optimizing access methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-10
Limiting PDQ resources in queries . . . . . . . . . . . . . . . . . . . . . . . . . . 3-11
Limiting the performance impact of CPU-intensive queries . . . . . . . . . . . . . . . . . . 3-12
Limiting the number of PDQ scan threads that can run concurrently . . . . . . . . . . . . . . 3-12
Configuring poll threads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-13
Enabling fast polling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-17
Network buffer pools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-17
Network buffers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-18
Support for private network buffers . . . . . . . . . . . . . . . . . . . . . . . . . 3-18
Network buffer size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-19
Virtual processors and CPU utilization. . . . . . . . . . . . . . . . . . . . . . . . . . 3-20
Adding virtual processors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-20
Automatic addition of CPU virtual processors . . . . . . . . . . . . . . . . . . . . . . 3-20
Monitoring virtual processors. . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-21
CPU virtual processor private memory caches . . . . . . . . . . . . . . . . . . . . . . 3-24
Connections and CPU utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-25
Multiplexed connections and CPU utilization . . . . . . . . . . . . . . . . . . . . . . 3-25
MaxConnect for multiple connections UNIX . . . . . . . . . . . . . . . . . . . . . . . 3-26

Chapter 4. Effect of configuration on memory utilization . . . . . . . . . . . . . . 4-1

Shared memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-1
Resident portion of shared memory . . . . . . . . . . . . . . . . . . . . . . . . . . 4-2
Virtual portion of shared memory . . . . . . . . . . . . . . . . . . . . . . . . . . 4-2
Message portion of shared memory . . . . . . . . . . . . . . . . . . . . . . . . . . 4-3
Estimating the size of the resident portion of shared memory . . . . . . . . . . . . . . . . . 4-3
Estimating the size of the virtual portion of shared memory . . . . . . . . . . . . . . . . . . 4-4
Estimating the size of the message portion of shared memory . . . . . . . . . . . . . . . . . 4-6
Configuring UNIX shared memory . . . . . . . . . . . . . . . . . . . . . . . . . . 4-6
Freeing shared memory with onmode -F . . . . . . . . . . . . . . . . . . . . . . . . 4-7
Configuration parameters that affect memory utilization . . . . . . . . . . . . . . . . . . . . 4-8
Setting the size of the buffer pool, logical-log buffer, and physical-log buffer . . . . . . . . . . . . 4-9
The LOCKS configuration parameter and memory utilization . . . . . . . . . . . . . . . . . 4-15
The RESIDENT configuration parameter and memory utilization . . . . . . . . . . . . . . . . 4-17
The SHMADD and EXTSHMADD configuration parameters and memory utilization . . . . . . . . . 4-17
The SHMTOTAL configuration parameter and memory utilization . . . . . . . . . . . . . . . 4-18
The SHMVIRTSIZE configuration parameter and memory utilization . . . . . . . . . . . . . . 4-19
The SHMVIRT_ALLOCSEG configuration parameter and memory utilization. . . . . . . . . . . . 4-19
The STACKSIZE configuration parameter and memory utilization . . . . . . . . . . . . . . . 4-20
Configure and monitor memory caches . . . . . . . . . . . . . . . . . . . . . . . . . 4-21
Data-dictionary cache . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-22
Data-distribution cache . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-23
Monitor and tune the SQL statement cache . . . . . . . . . . . . . . . . . . . . . . . 4-25
Session memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-38
Data-replication buffers and memory utilization . . . . . . . . . . . . . . . . . . . . . . 4-39
Memory latches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-39
Monitoring latches with command-line utilities . . . . . . . . . . . . . . . . . . . . . . 4-39
Monitoring latches with SMI tables . . . . . . . . . . . . . . . . . . . . . . . . . . 4-40
Encrypted values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-40

Chapter 5. Effect of configuration on I/O activity . . . . . . . . . . . . . . . . . 5-1

Chunk and dbspace configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-1
Associate disk partitions with chunks . . . . . . . . . . . . . . . . . . . . . . . . . 5-2
Associate dbspaces with chunks . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-2
Placing system catalog tables with database tables . . . . . . . . . . . . . . . . . . . . . 5-2
I/O for cooked files for dbspace chunks . . . . . . . . . . . . . . . . . . . . . . . . . 5-2
Direct I/O (UNIX) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-3
Direct I/O (Windows) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-3
Concurrent I/O (AIX only) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-3
Enabling the direct I/O or concurrent I/O option (UNIX). . . . . . . . . . . . . . . . . . . 5-4

iv IBM Informix Performance Guide

 Confirming the use of the direct or concurrent I/O option (UNIX) . . . . . . . . . . . . . . . . 5-4
Placement of critical data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-5
Consider separate disks for critical data components . . . . . . . . . . . . . . . . . . . . 5-5
Consider mirroring for critical data components . . . . . . . . . . . . . . . . . . . . . . 5-5
Configuration parameters that affect critical data . . . . . . . . . . . . . . . . . . . . . . 5-8
Configure dbspaces for temporary tables and sort files. . . . . . . . . . . . . . . . . . . . . 5-8
Creating temporary dbspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-10
Specify temporary tables in the DBSPACETEMP configuration parameter . . . . . . . . . . . . . 5-11
Override the DBSPACETEMP configuration parameter for a session . . . . . . . . . . . . . . . 5-11
Estimating temporary space for dbspaces and hash joins. . . . . . . . . . . . . . . . . . . 5-12
PSORT_NPROCS environment variable . . . . . . . . . . . . . . . . . . . . . . . . 5-13
Configure sbspaces for temporary smart large objects. . . . . . . . . . . . . . . . . . . . . 5-14
Creating temporary sbspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-14
Specify which sbspaces to use for temporary storage . . . . . . . . . . . . . . . . . . . . 5-15
Placement of simple large objects . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-15
Advantage of blobspaces over dbspaces . . . . . . . . . . . . . . . . . . . . . . . . 5-16
Blobpage size considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-16
Factors that affect I/O for smart large objects . . . . . . . . . . . . . . . . . . . . . . . 5-20
Disk layout for sbspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-20
Configuration parameters that affect sbspace I/O . . . . . . . . . . . . . . . . . . . . . 5-20
onspaces options that affect sbspace I/O . . . . . . . . . . . . . . . . . . . . . . . . 5-21
How the Optical Subsystem affects performance . . . . . . . . . . . . . . . . . . . . . . 5-24
Environment variables and configuration parameters for the Optical Subsystem . . . . . . . . . . . . 5-25
STAGEBLOB, an Optical Subsystem configuration parameter . . . . . . . . . . . . . . . . . 5-25
OPCACHEMAX, an Optical Subsystem configuration parameter . . . . . . . . . . . . . . . . 5-26
INFORMIXOPCACHE, an Optical Subsystem environment variable . . . . . . . . . . . . . . . 5-26
Table I/O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-26
Sequential scans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-26
Light scans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-27
Unavailable data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-28
Configuration parameters that affect table I/O . . . . . . . . . . . . . . . . . . . . . . . 5-28
How DATASKIP affects table I/O . . . . . . . . . . . . . . . . . . . . . . . . . . 5-29
Background I/O activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-29
Configuration parameters that affect checkpoints . . . . . . . . . . . . . . . . . . . . . 5-30
Configuration parameters that affect logging. . . . . . . . . . . . . . . . . . . . . . . 5-34
Configuration parameters that affect page cleaning . . . . . . . . . . . . . . . . . . . . 5-39
Configuration parameters that affect backup and restore. . . . . . . . . . . . . . . . . . . 5-41
Configuration parameters that affect rollback and recovery . . . . . . . . . . . . . . . . . . 5-42
Configuration parameters that affect data replication and auditing . . . . . . . . . . . . . . . 5-43
LRU tuning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-45

Chapter 6. Table performance considerations . . . . . . . . . . . . . . . . . . . 6-1

Placing tables on disk. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-1
Isolating high-use tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-2
Placing high-use tables on middle partitions of disks . . . . . . . . . . . . . . . . . . . . 6-3
Using multiple disks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-3
Backup and restore considerations when placing tables on disks . . . . . . . . . . . . . . . . 6-4
Factors affecting the performance of nonfragmented tables and table fragments . . . . . . . . . . . 6-5
Estimating table size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-5
Estimating data pages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-5
Estimating pages that simple large objects occupy . . . . . . . . . . . . . . . . . . . . . 6-8
Managing the size of first and next extents for the tblspace tblspace . . . . . . . . . . . . . . . . 6-10
Managing sbspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-10
Estimating pages that smart large objects occupy . . . . . . . . . . . . . . . . . . . . . 6-10
Improving metadata I/O for smart large objects . . . . . . . . . . . . . . . . . . . . . 6-12
Monitoring sbspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-13
Changing storage characteristics of smart large objects . . . . . . . . . . . . . . . . . . . 6-17
Managing extents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-19
Choosing table extent sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-20
Monitoring active tblspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-23
Monitoring the upper limit on extents and extent interleaving . . . . . . . . . . . . . . . . . 6-23

Contents v
 Reclaiming unused space within an extent . . . . . . . . . . . . . . . . . . . . . . . 6-26
Managing extent deallocation with the TRUNCATE keyword . . . . . . . . . . . . . . . . . 6-27
Defragment partitions to merge extents . . . . . . . . . . . . . . . . . . . . . . . . 6-28
Storing multiple table fragments in a single dbspace . . . . . . . . . . . . . . . . . . . . . 6-29
Displaying a list of table and index partitions . . . . . . . . . . . . . . . . . . . . . . . 6-29
Changing tables to improve performance . . . . . . . . . . . . . . . . . . . . . . . . . 6-29
Loading and unloading tables . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-29
Dropping indexes for table-update efficiency . . . . . . . . . . . . . . . . . . . . . . 6-32
Creating and enabling referential constraints efficiently . . . . . . . . . . . . . . . . . . . 6-32
Attaching or detaching fragments . . . . . . . . . . . . . . . . . . . . . . . . . . 6-34
Altering a table definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-35
Denormalize the data model to improve performance . . . . . . . . . . . . . . . . . . . . 6-42
Shortening rows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-42
Expelling long strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-43
Splitting wide tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-44
Redundant data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-45
Reduce disk space in tables with variable length rows . . . . . . . . . . . . . . . . . . . . 6-46
Reduce disk space by compressing tables and fragments. . . . . . . . . . . . . . . . . . . . 6-46

Chapter 7. Indexes and index performance considerations . . . . . . . . . . . . . 7-1

Types of indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-1
B-tree indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-1
Forest of trees indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-2
R-tree indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-4
Indexes that DataBlade modules provide . . . . . . . . . . . . . . . . . . . . . . . . 7-4
Estimating index pages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-4
Index extent sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-4
Estimating conventional index pages. . . . . . . . . . . . . . . . . . . . . . . . . . 7-5
Managing indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-8
Space costs of indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-8
Time costs of indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-8
Unclaimed index space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-9
Indexes on columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-10
Nonunique indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-12
Improve query performance with a forest of trees index . . . . . . . . . . . . . . . . . . . . 7-13
Detecting root node contention . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-13
Creating a forest of trees index . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-14
Disabling and enabling a forest of trees index . . . . . . . . . . . . . . . . . . . . . . 7-15
Performing a range scan on a forest of trees index . . . . . . . . . . . . . . . . . . . . . 7-15
Determining if you are using a forest of trees index . . . . . . . . . . . . . . . . . . . . 7-16
Finding the number of hashed columns and subtrees in a forest of trees index . . . . . . . . . . . 7-16
Creating and dropping an index in an online environment . . . . . . . . . . . . . . . . . . . 7-16
When you cannot create or drop indexes online . . . . . . . . . . . . . . . . . . . . . 7-17
Creating attached indexes in an online environment . . . . . . . . . . . . . . . . . . . . 7-18
Limiting memory allocation while creating indexes online . . . . . . . . . . . . . . . . . . 7-18
Improving performance for index builds . . . . . . . . . . . . . . . . . . . . . . . . . 7-18
Estimating memory needed for sorting . . . . . . . . . . . . . . . . . . . . . . . . 7-19
Estimating temporary space for index builds . . . . . . . . . . . . . . . . . . . . . . 7-20
Storing multiple index fragments in a single dbspace . . . . . . . . . . . . . . . . . . . . . 7-20
Improving performance for index checks . . . . . . . . . . . . . . . . . . . . . . . . . 7-20
Indexes on user-defined data types . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-21
Defining indexes for user-defined data types . . . . . . . . . . . . . . . . . . . . . . 7-22
Using an index that a DataBlade module provides . . . . . . . . . . . . . . . . . . . . . 7-27
Choosing operator classes for indexes . . . . . . . . . . . . . . . . . . . . . . . . . 7-27

Chapter 8. Locking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-1

Locks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-1
Locking granularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-1
Row and key locks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-1
Page locks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-2

vi IBM Informix Performance Guide

 Table locks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-3
Database locks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-3
Configuring the lock mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-4
Setting the lock mode to wait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-5
Locks with the SELECT statement . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-5
Isolation level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-5
Locking nonlogging tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-8
Update cursors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-8
Locks placed with INSERT, UPDATE, and DELETE statements . . . . . . . . . . . . . . . . . 8-10
The internal lock table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-10
Monitoring locks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-11
Configuring and managing lock usage . . . . . . . . . . . . . . . . . . . . . . . . . 8-12
Monitoring lock waits and lock errors . . . . . . . . . . . . . . . . . . . . . . . . . 8-13
Monitoring the number of free locks . . . . . . . . . . . . . . . . . . . . . . . . . 8-14
Monitoring deadlocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-15
Monitoring isolation levels that sessions use . . . . . . . . . . . . . . . . . . . . . . . 8-15
Locks for smart large objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-16
Byte-range locking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-16
Lock promotion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-19
Dirty Read isolation level and smart large objects . . . . . . . . . . . . . . . . . . . . . 8-20

Chapter 9. Fragmentation guidelines . . . . . . . . . . . . . . . . . . . . . . 9-1

Planning a fragmentation strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-1
Fragmentation goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-2
Examining your data and queries . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-4
Considering physical fragmentation factors . . . . . . . . . . . . . . . . . . . . . . . 9-5
Distribution schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-6
Choosing a distribution scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-7
Designing an expression-based distribution scheme . . . . . . . . . . . . . . . . . . . . . 9-8
Suggestions for improving fragmentation . . . . . . . . . . . . . . . . . . . . . . . . 9-9
Strategy for fragmenting indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-10
Attached indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-10
Detached indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-12
Restrictions on indexes for fragmented tables . . . . . . . . . . . . . . . . . . . . . . 9-13
Strategy for fragmenting temporary tables . . . . . . . . . . . . . . . . . . . . . . . . 9-13
Distribution schemes that eliminate fragments . . . . . . . . . . . . . . . . . . . . . . . 9-14
Fragmentation expressions for fragment elimination . . . . . . . . . . . . . . . . . . . . 9-14
Query expressions for fragment elimination . . . . . . . . . . . . . . . . . . . . . . . 9-15
Effectiveness of fragment elimination . . . . . . . . . . . . . . . . . . . . . . . . . 9-16
Improve the performance of operations that attach and detach fragments . . . . . . . . . . . . . . 9-19
Improving ALTER FRAGMENT ATTACH performance . . . . . . . . . . . . . . . . . . . 9-20
Improving ALTER FRAGMENT DETACH performance . . . . . . . . . . . . . . . . . . . 9-26
Forcing out transactions when altering table fragments . . . . . . . . . . . . . . . . . . . 9-27
Monitoring fragment use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-28
Monitoring fragmentation with the onstat -g ppf command. . . . . . . . . . . . . . . . . . 9-28
Monitoring fragmentation with SET EXPLAIN output . . . . . . . . . . . . . . . . . . . 9-29

Chapter 10. Queries and the query optimizer . . . . . . . . . . . . . . . . . . 10-1

The query plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-1
The access plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-1
The join plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-2
Example of query-plan execution . . . . . . . . . . . . . . . . . . . . . . . . . . 10-4
Query plans that include an index self-join path . . . . . . . . . . . . . . . . . . . . . 10-8
Query plan evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-9
Report that shows the query plan chosen by the optimizer . . . . . . . . . . . . . . . . . . 10-9
Sample query plan reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-12
XML query plans in IBM Data Studio . . . . . . . . . . . . . . . . . . . . . . . . 10-19
Factors that affect the query plan . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-19
Statistics held for the table and index . . . . . . . . . . . . . . . . . . . . . . . . . 10-20
Filters in the query . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-20

Contents vii
 Indexes for evaluating a filter . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-21
Effect of PDQ on the query plan . . . . . . . . . . . . . . . . . . . . . . . . . . 10-22
Effect of OPTCOMPIND on the query plan . . . . . . . . . . . . . . . . . . . . . . . 10-22
Effect of available memory on the query plan . . . . . . . . . . . . . . . . . . . . . . 10-23
Time costs of a query . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-23
Memory-activity costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-24
Sort-time costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-24
Row-reading costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-25
Sequential access costs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-26
Nonsequential access costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-26
Index lookup costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-26
In-place ALTER TABLE costs . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-27
View costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-27
Small-table costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-28
Data-mismatch costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-28
Encrypted-value costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-29
GLS functionality costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-29
Network-access costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-29
Optimization when SQL is within an SPL routine. . . . . . . . . . . . . . . . . . . . . . 10-31
SQL optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-31
Execution of an SPL routine . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-33
SPL routine executable format stored in UDR cache . . . . . . . . . . . . . . . . . . . . 10-33
Trigger execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-34
Performance implications for triggers . . . . . . . . . . . . . . . . . . . . . . . . . 10-35

Chapter 11. Optimizer directives . . . . . . . . . . . . . . . . . . . . . . . . 11-1

What optimizer directives are . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-1
Optimizer directives that are embedded in queries. . . . . . . . . . . . . . . . . . . . . 11-1
External optimizer directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-1
Reasons to use optimizer directives . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-2
Preparation for using directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-3
Guidelines for using directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-3
Types of optimizer directives that are supported in SQL statements . . . . . . . . . . . . . . . . 11-3
Access-method directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-4
Join-order directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-5
Join-method directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-6
Optimization-goal directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-7
Star-join directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-8
EXPLAIN directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-8
Example of directives that can alter a query plan . . . . . . . . . . . . . . . . . . . . . 11-10
Configuration parameters and environment variables for optimizer directives . . . . . . . . . . . . 11-12
Optimizer directives and SPL routines . . . . . . . . . . . . . . . . . . . . . . . . . 11-13
Forcing reoptimization to avoid an index and previously prepared statement problem . . . . . . . . . 11-13
External optimizer directives. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-15
Creating and saving external directives . . . . . . . . . . . . . . . . . . . . . . . . 11-15
Enabling external directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-15
Deleting external directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-16

Chapter 12. Parallel database query (PDQ) . . . . . . . . . . . . . . . . . . . 12-1

What PDQ is . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-1
Structure of a PDQ query . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-1
Database server operations that use PDQ . . . . . . . . . . . . . . . . . . . . . . . . . 12-2
Parallel update and delete operations . . . . . . . . . . . . . . . . . . . . . . . . . 12-2
Parallel insert operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-2
Parallel index builds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-3
Parallel user-defined routines . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-4
Hold cursors that use PDQ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-4
SQL operations that do not use PDQ . . . . . . . . . . . . . . . . . . . . . . . . . 12-4
Update statistics operations affected by PDQ . . . . . . . . . . . . . . . . . . . . . . 12-5
SPL routines and triggers and PDQ . . . . . . . . . . . . . . . . . . . . . . . . . . 12-5

viii IBM Informix Performance Guide

 Correlated and uncorrelated subqueries . . . . . . . . . . . . . . . . . . . . . . . .12-5
OUTER index joins and PDQ . . . . . . . . . . . . . . . . . . . . . . . . . . . .12-5
Remote tables used with PDQ . . . . . . . . . . . . . . . . . . . . . . . . . . .12-6
The Memory Grant Manager . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12-6
The allocation of resources for parallel database queries . . . . . . . . . . . . . . . . . . . .12-7
Limiting the priority of decision-support queries . . . . . . . . . . . . . . . . . . . . .12-8
Adjusting the amount of memory for DSS and PDQ queries . . . . . . . . . . . . . . . . . 12-11
Limiting the number of concurrent scans . . . . . . . . . . . . . . . . . . . . . . . 12-11
Limiting the maximum number of PDQ queries . . . . . . . . . . . . . . . . . . . . . 12-12
Managing PDQ queries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-12
Analyzing query plans with SET EXPLAIN output . . . . . . . . . . . . . . . . . . . . 12-12
Influencing the choice of a query plan . . . . . . . . . . . . . . . . . . . . . . . . 12-13
Setting the PDQ priority dynamically. . . . . . . . . . . . . . . . . . . . . . . . . 12-13
Enabling the database server to allocate PDQ memory . . . . . . . . . . . . . . . . . . . 12-13
User control of PDQ resources . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-15
DBA control of resources for PDQ and DSS queries . . . . . . . . . . . . . . . . . . . . 12-15
Monitoring resources used for PDQ and DSS queries . . . . . . . . . . . . . . . . . . . . 12-16
Monitoring PDQ resources by using the onstat Utility . . . . . . . . . . . . . . . . . . . 12-16
Identifying parallel scans in SET EXPLAIN output . . . . . . . . . . . . . . . . . . . . 12-18

Chapter 13. Improving individual query performance . . . . . . . . . . . . . . . 13-1

Test queries using a dedicated test system . . . . . . . . . . . . . . . . . . . . . . . . 13-1
Display the query plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-1
Improve filter selectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-2
Filters with user-defined routines . . . . . . . . . . . . . . . . . . . . . . . . . . 13-2
Avoid some filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-3
Use join filters and post-join filters . . . . . . . . . . . . . . . . . . . . . . . . . . 13-3
Automatic statistics updating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-6
How AUS works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-7
AUS expiration policies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-8
Viewing AUS statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-10
Rescheduling AUS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-10
Disabling AUS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-11
Update statistics when they are not generated automatically . . . . . . . . . . . . . . . . . . 13-11
Update the statistics for the number of rows . . . . . . . . . . . . . . . . . . . . . . 13-12
Drop data distributions if necessary when upgrading . . . . . . . . . . . . . . . . . . . 13-13
Creating data distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-13
Updating statistics for join columns . . . . . . . . . . . . . . . . . . . . . . . . . 13-16
Updating statistics for columns with user-defined data types . . . . . . . . . . . . . . . . . 13-16
Update statistics in parallel on very large databases . . . . . . . . . . . . . . . . . . . . 13-17
Adjust the amount of memory and disk space for UPDATE STATISTICS . . . . . . . . . . . . . 13-17
Data sampling during update statistics operations . . . . . . . . . . . . . . . . . . . . 13-18
Display data distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-18
Improve performance by adding or removing indexes . . . . . . . . . . . . . . . . . . . . 13-20
Replace autoindexes with permanent indexes . . . . . . . . . . . . . . . . . . . . . . 13-20
Use composite indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-20
Indexes for data warehouse applications. . . . . . . . . . . . . . . . . . . . . . . . 13-21
Configure B-tree scanner information to improve transaction processing . . . . . . . . . . . . . 13-22
Determine the amount of free space in an index page . . . . . . . . . . . . . . . . . . . 13-29
Optimizer estimates of distributed queries . . . . . . . . . . . . . . . . . . . . . . . . 13-29
Buffer data transfers for a distributed query . . . . . . . . . . . . . . . . . . . . . . 13-30
The query plan of a distributed query . . . . . . . . . . . . . . . . . . . . . . . . 13-30
Improve sequential scans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-31
Enable view folding to improve query performance . . . . . . . . . . . . . . . . . . . . . 13-31
Reduce the join and sort operations . . . . . . . . . . . . . . . . . . . . . . . . . . 13-32
Avoid or simplify sort operations . . . . . . . . . . . . . . . . . . . . . . . . . . 13-32
Use parallel sorts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-32
Use temporary tables to reduce sorting scope . . . . . . . . . . . . . . . . . . . . . . 13-33
Configuring memory for queries with hash joins, aggregates, and other memory-intensive elements. . . . 13-34
Optimize user-response time for queries . . . . . . . . . . . . . . . . . . . . . . . . . 13-35
Optimization level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-35

Contents ix
 Optimization goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-35
Optimize queries for user-defined data types . . . . . . . . . . . . . . . . . . . . . . . 13-38
Parallel UDRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-38
Selectivity and cost functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-39
User-defined statistics for UDTs . . . . . . . . . . . . . . . . . . . . . . . . . . 13-40
Negator functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-40
Optimize queries with the SQL statement cache . . . . . . . . . . . . . . . . . . . . . . 13-40
When to use the SQL statement cache . . . . . . . . . . . . . . . . . . . . . . . . 13-41
Using the SQL statement cache . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-41
Monitoring memory usage for each session . . . . . . . . . . . . . . . . . . . . . . . 13-43
Monitoring usage of the SQL statement cache . . . . . . . . . . . . . . . . . . . . . . 13-46
Monitor sessions and threads . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-47
Monitor sessions and threads with onstat commands . . . . . . . . . . . . . . . . . . . 13-48
Monitor sessions and threads with ON-Monitor (UNIX) . . . . . . . . . . . . . . . . . . 13-53
Monitor sessions and threads with SMI tables . . . . . . . . . . . . . . . . . . . . . . 13-54
Monitor transactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-55
Display information about transactions . . . . . . . . . . . . . . . . . . . . . . . . 13-55
Display information about transaction locks . . . . . . . . . . . . . . . . . . . . . . 13-56
Display statistics on user sessions . . . . . . . . . . . . . . . . . . . . . . . . . . 13-57
Display statistics on sessions executing SQL statements. . . . . . . . . . . . . . . . . . . 13-58

Chapter 14. The onperf utility on UNIX . . . . . . . . . . . . . . . . . . . . . 14-1

Overview of the onperf utility . . . . . . . . . . . . . . . . . . . . . . . . . . . .14-1
Basic onperf utility functions . . . . . . . . . . . . . . . . . . . . . . . . . . . .14-1
onperf utility tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .14-3
Requirements for running the onperf utility . . . . . . . . . . . . . . . . . . . . . . . .14-3
Starting the onperf utility and exiting from it . . . . . . . . . . . . . . . . . . . . . . .14-4
The onperf user interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .14-5
Graph tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .14-5
Query-tree tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-11
Status tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-11
Activity tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-12
Why you might want to use onperf . . . . . . . . . . . . . . . . . . . . . . . . . . 14-12
Routine monitoring with onperf . . . . . . . . . . . . . . . . . . . . . . . . . . 14-13
Diagnosing sudden performance loss . . . . . . . . . . . . . . . . . . . . . . . . . 14-13
Diagnosing performance degradation . . . . . . . . . . . . . . . . . . . . . . . . . 14-13
onperf utility metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-13
Database server metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-13
Disk-chunk metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-15
Disk-spindle metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-15
Physical-processor metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-15
Virtual-processor metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-16
Session metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-16
Tblspace metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-17
Fragment metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-18

Appendix A. Case studies and examples . . . . . . . . . . . . . . . . . . . . A-1

Case study of a situation in which disks are overloaded . . . . . . . . . . . . . . . . . . . . A-1

Appendix B. Accessibility . . . . . . . . . . . . . . . . . . . . . . . . . . . B-1

Accessibility features for IBM Informix products. . . . . . . . . . . . . . . . . . . . . . . B-1
Accessibility features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-1
Keyboard navigation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-1
Related accessibility information . . . . . . . . . . . . . . . . . . . . . . . . . . . B-1
IBM and accessibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-1
Dotted decimal syntax diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-1

Notices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C-1
Privacy policy considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C-3
Trademarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C-3

x IBM Informix Performance Guide

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . X-1

Contents xi
xii IBM Informix Performance Guide
Introduction
This introduction provides an overview of the information in this publication and
describes the conventions it uses.

About this publication

This publication provides information about how to configure and operate IBM®
Informix® to improve overall system throughput and to improve the performance
of SQL queries. This publication also includes information about performance
tuning issues and methods, fragmentation guidelines, and a complete description
of the onperf utility

Information in this publication can help you perform the following tasks:
v Monitor system resources that are critical to performance
v Identify database activities that affect these critical resources
v Identify and monitor queries that are critical to performance
v Use the database server utilities (especially onperf, ISA and onstat) for
performance monitoring and tuning
v Eliminate performance bottlenecks by:
– Balancing the load on system resources
– Adjusting the configuration parameters or environment variables of your
database server
– Adjusting the arrangement of your data
– Allocating resources for decision-support queries
– Creating indexes to speed up retrieval of your data

Performance measurement and tuning encompass a broad area of research and

practice and can involve information beyond the scope of this publication.

Performance issues related to dimensional databases and data warehouse queries

are described in the IBM Informix Data Warehouse Guide.
Related concepts:
Performance tuning dimensional databases (Data Warehouse Guide)

Topics beyond the scope of this publication

Attempts to balance the workload often produce a succession of moderate
performance improvements. Sometimes the improvements are dramatic. However,
in some situations a load-balancing approach is not enough. The following types of
situations might require measures beyond the scope of this publication:
v Application programs that require modification to make better use of database
server or operating-system resources
v Applications that interact in ways that impair performance
v A host computer that might be subject to conflicting uses
v A host computer with capacity that is inadequate for the evolving workload
v Network performance problems that affect client/server or other applications

© Copyright IBM Corp. 1996, 2014 xiii

 No amount of database tuning can correct these situations. Nevertheless, they are
easier to identify and resolve when the database server is configured properly.

Important: Although broad performance considerations also include reliability and

data availability as well as improved response time and efficient use of system
resources, this publication discusses only response time and system resource use.
For discussions of improved database server reliability and data availability, see
information about switchover, mirroring, and high availability in your IBM
Informix Administrator's Guide. For information about backup and restore, see the
IBM Informix Backup and Restore Guide.

Types of users
This publication is written for the following users:
v Database administrators
v Database server administrators
v Database-application programmers
v Performance engineers

This publication assumes that you have the following background:

v A working knowledge of your computer, your operating system, and the utilities
that your operating system provides
v Some experience working with relational databases or exposure to database
concepts
v Some experience with computer programming
v Some experience with database server administration, operating-system
administration, or network administration

Software dependencies
This publication assumes that you are using IBM Informix, Version 11.70.

Assumptions about your locale

IBM Informix products can support many languages, cultures, and code sets. All
the information related to character set, collation and representation of numeric
data, currency, date, and time that is used by a language within a given territory
and encoding is brought together in a single environment, called a Global
Language Support (GLS) locale.

The IBM Informix OLE DB Provider follows the ISO string formats for date, time,
and money, as defined by the Microsoft OLE DB standards. You can override that
default by setting an Informix environment variable or registry entry, such as
DBDATE.

If you use Simple Network Management Protocol (SNMP) in your Informix

environment, note that the protocols (SNMPv1 and SNMPv2) recognize only
English code sets. For more information, see the topic about GLS and SNMP in the
IBM Informix SNMP Subagent Guide.

The examples in this publication are written with the assumption that you are
using one of these locales: en_us.8859-1 (ISO 8859-1) on UNIX platforms or
en_us.1252 (Microsoft 1252) in Windows environments. These locales support U.S.
English format conventions for displaying and entering date, time, number, and
currency values. They also support the ISO 8859-1 code set (on UNIX and Linux)

xiv IBM Informix Performance Guide

 or the Microsoft 1252 code set (on Windows), which includes the ASCII code set
plus many 8-bit characters such as é, è, and ñ.

You can specify another locale if you plan to use characters from other locales in
your data or your SQL identifiers, or if you want to conform to other collation
rules for character data.

For instructions about how to specify locales, additional syntax, and other
considerations related to GLS locales, see the IBM Informix GLS User's Guide.

Demonstration databases
The DB-Access utility, which is provided with your IBM Informix database server
products, includes one or more of the following demonstration databases:
v The stores_demo database illustrates a relational schema with information about
a fictitious wholesale sporting-goods distributor. Many examples in IBM
Informix publications are based on the stores_demo database.
v The superstores_demo database illustrates an object-relational schema. The
superstores_demo database contains examples of extended data types, type and
table inheritance, and user-defined routines.

For information about how to create and populate the demonstration databases,
see the IBM Informix DB-Access User's Guide. For descriptions of the databases and
their contents, see the IBM Informix Guide to SQL: Reference.

The scripts that you use to install the demonstration databases are in the
$INFORMIXDIR/bin directory on UNIX platforms and in the %INFORMIXDIR%\bin
directory in Windows environments.

What's new in performance for Informix, version 11.70

This publication includes information about new features and changes in existing
functionality.

The following changes and enhancements are relevant to this publication. For a
complete list of what's new in this release, see the release notes or the information
center at http://pic.dhe.ibm.com/infocenter/idshelp/v117/topic/com.ibm.po.doc/
new_features.htm.

Introduction xv
Table 1. What's new in the IBM Informix Performance Guide for version 11.70.xC8
Overview Reference
Dynamic private memory caches for CPU virtual “CPU virtual processor private memory caches” on page
processors 3-24

Private memory caches for CPU virtual processors now

change size automatically as needed. You create private
memory caches by setting the VP_MEMORY_CACHE_KB
configuration parameter to the initial total size of the
caches. The size of a private memory cache increases and
decreases automatically, depending on the needs of the
associated CPU virtual processor. Previously, the size of
private memory caches was limited to the value of the
VP_MEMORY_CACHE_KB configuration parameter. You
can preserve the previous behavior by including a comma
and the word STATIC after the size value of the
VP_MEMORY_CACHE_KB configuration parameter.

The onstat -g vpcache command now shows the target

size for each bin in the cache before draining starts and
the last time that each bin was drained.
In-place alter operations on serial data types “Conditions for in-place alter operations” on page 6-36

The ALTER TABLE statement converts the following

column data types with in-place alter operations:
v SERIAL to SERIAL8
v SERIAL to BIGSERIAL
v SERIAL8 to BIGSERIAL
v BIGSERIAL to SERIAL8

Previously such data types were converted with slow

alter operations. In-place alter operations require less
space than slow alter operations and make the table
available to other sessions faster.

xvi IBM Informix Performance Guide

Table 1. What's new in the IBM Informix Performance Guide for version 11.70.xC8 (continued)
Overview Reference
Temporarily prevent constraint validation “Creating and enabling referential constraints efficiently”
on page 6-32
You can significantly increase the speed of loading or
migrating large tables by temporarily preventing the
database server from validating foreign-key referential
constraints. You can disable the validation of constraints
when you create constraints or change the mode of
constraints to ENABLED or FILTERING.
v You include the NOVALIDATE keyword in an ALTER
TABLE ADD CONSTRAINT statement or in a SET
CONSTRAINTS ENABLED or SET CONSTRAINTS
FILTERING statement.
v If you plan to run multiple ALTER TABLE ADD
CONSTRAINT or SET CONSTRAINTS statements, run
the SET ENVIRONMENT NOVALIDATE ON statement
to disable the validation of foreign-key constraints
during the current session.

When you migrate data, include the -nv option in the

dbimport command.

The NOVALIDATE keyword prevents the database server

from checking every row for referential integrity during
ALTER TABLE ADD CONSTRAINT and SET
CONSTRAINTS operations on foreign-key constraints.
When those statements finish running, the database
server automatically resumes referential-integrity
enforcement of those constraints in subsequent DML
operations.

Use this feature only on tables whose enabled foreign-key

constraints are free of violations, or when the referential
constraints can be validated after the tables are loaded or
migrated to the target database.
Faster creation of foreign-key constraints “Creating and enabling referential constraints efficiently”
on page 6-32
When you run the ALTER TABLE ADD CONSTRAINT
statement, some foreign-key constraints can be created
faster if the table has a unique index or a primary-key
constraint that is already defined on the columns in the
foreign-key constraint.

Foreign-key constraints are not created faster, however, if

the constraint key or index key includes columns of
user-defined or opaque data types, including BOOLEAN
and LVARCHAR, or if other restrictions are true for the
foreign-key constraint or for the referenced table.

Introduction xvii
Table 2. What's new in the IBM Informix Performance Guide for version 11.70xC4
Overview Reference
Data sampling for update statistics operations “Data sampling during update statistics operations” on
page 13-18
If you have a large index with more than 100 000 leaf
pages, you can generate index statistics based on
sampling when you run UPDATE STATISTICS statements
in LOW mode. Gathering index statistics from sampled
data can increase the speed of the update statistics
operations. To enable sampling, set the
USTLOW_SAMPLE configuration parameter or the
USTLOW_SAMPLE option of the SET ENVIRONMENT
statement.

Table 3. What's new in the IBM Informix Performance Guide for version 11.70xC3
Overview Reference
Automatic read-ahead operations “Configuration parameters that affect table I/O” on page
5-28
You can enable the database server to use read-ahead
operations automatically to improve performance. Most AUTO_READAHEAD configuration parameter
queries can benefit from processing the query while (Administrator's Reference)
asynchronously retrieving the data required by the query.
The database server can automatically use asynchronous
operations for data or it can avoid them if the data for
the query is already cached. Use the
AUTO_READAHEAD configuration parameter to
configure automatic read-ahead operations for all queries,
and use the SET ENVIRONMENT AUTO_READAHEAD
statement to configure automatic read-ahead operations
for a particular session.

The RA_THRESHOLD configuration parameter is

deprecated with this release.

The RA_THRESHOLD configuration parameter is

deprecated with this release.
Reserving memory for critical activities “The LOW_MEMORY_RESERVE configuration parameter
and memory utilization” on page 4-15
You can enable the server to reserve a specific amount of
memory for use when critical activities (such as rollback LOW_MEMORY_RESERVE configuration parameter
activities) are needed and the server has limited free (Administrator's Reference)
memory. If you enable the new
LOW_MEMORY_RESERVE configuration parameter by
setting it to a specified value in kilobytes, the critical
activities can complete even when you get out-of-memory
errors. You can also dynamically adjust the value of the
LOW_MEMORY_RESERVE configuration parameter with
the onmode -wm or -wf command.
Configuring the server response to low memory Configure the server response when memory is critically
low (Administrator's Guide)
You can configure the actions that the server takes to
continue processing when memory is critically low. You LOW_MEMORY_MGR configuration parameter
can specify the criteria for terminating sessions based on (Administrator's Reference)
idle time, memory usage, and other factors so that the
targeted application can continue and avoid
out-of-memory problems. Configuring the low memory
response is useful for embedded applications that have
memory limitations.

xviii IBM Informix Performance Guide

Table 4. What's new in the IBM Informix Performance Guide for version 11.70xC2
Overview Reference
Improve network connection performance and scalability “Improve connection performance and scalability” on
page 3-16
You can improve the performance and scalability of
network connections on UNIX operating systems by NUMFDSERVERS configuration parameter
using the NUMFDSERVERS configuration parameter. Use (Administrator's Reference)
the configuration parameter to adjust the maximum
number of poll threads for the server to use when
migrating TCP/IP connection across virtual processors
(VPs). Adjusting the number of poll threads, for example,
by increasing the number of poll threads beyond the
default number, is useful if the database server has a high
rate of new connect and disconnect requests or if you
find a high amount of contention between network
shared file (NSF) locks. For example, if you have multiple
CPU VPs and poll threads and this results in NSF
locking, you can increase the value of NUMFDSERVERS
(and you can increase the number of poll threads
specified in the NETTYPE configuration parameter) to
reduce NSF lock contention.

Table 5. What's new in the IBM Informix Performance Guide for version 11.70xC1
Overview Reference
Less root node contention with forest of trees indexes “Forest of trees indexes” on page 7-2

If you have many concurrent users who routinely “Improve query performance with a forest of trees index”
experience delays due to root node contention, you might on page 7-13
improve query performance if you convert your B-tree
index to a forest of trees index. A forest of trees index is CREATE INDEX statement (SQL Syntax)
similar to a B-tree index, but has multiple root nodes and
potentially fewer levels. You create forest of trees indexes HASH ON clause (SQL Syntax)
with the new HASH ON clause of the CREATE INDEX
statement of SQL.
Automatic light scans on tables “Light scans” on page 5-27

Informix now automatically performs light scans when

appropriate. In the previous release, you had to set
configuration parameters to enable Informix to perform
these scans.
Defragmenting partitions “Defragment partitions to merge extents” on page 6-28

A frequently updated table can become fragmented over

time and this fragmentation degrades performance when
the table is accessed by the server. You can improve
performance by defragmenting partitions to merge
noncontiguous extents. Defragmenting a table brings data
rows closer together and avoids partition header page
overflow problems. Use the SQL administration API
task() or admin() function with the defragment argument
and specify the table name or partition number that you
want to defragment.

Introduction xix
Table 5. What's new in the IBM Informix Performance Guide for version 11.70xC1 (continued)
Overview Reference
Automatically add CPU virtual processors “Automatic addition of CPU virtual processors” on page
3-20
When the database server starts, it checks that the
number of CPU virtual processors is at least half the
number of CPU processors on the database server
computer. This ratio of CPU processors to CPU virtual
processors is a recommended minimum to ensure that the
database server performs optimally in most situations. If
necessary, the database server automatically increases the
number of CPU virtual processors to half the number of
CPU processors.
Automatically optimize data storage “Defragment partitions to merge extents” on page 6-28

You can automatically compress, shrink, repack, and

defragment tables and fragments by enabling the
auto_crsd Scheduler task. You can disable or update the
thresholds for each of these operations.
Large pages support on Linux “Resident portion of shared memory” on page 4-2

Large pages for non-message shared memory segments “Virtual portion of shared memory” on page 4-2
that reside in physical memory are now enabled by
default on Linux platforms. Previously, large pages were
supported only on AIX® and Solaris systems. The use of
large pages can provide performance benefits in large
memory configurations. To enable or disable support for
large pages, use the IFX_LARGE_PAGES environment
variable.
Improving performance by reducing buffer reads “Improve performance by adding or removing indexes”
on page 13-20
If you enable the new BATCHEDREAD_INDEX
configuration parameter, the optimizer automatically BATCHEDREAD_INDEX configuration parameter
chooses to fetch a set of keys from an index buffer, (Administrator's Reference)
reducing the number of buffer times a buffer is read.
Alerts for tables with in-place alter operations “In-place alter” on page 6-35

You can see which tables have outstanding in-place alter

operations, which can cause slight performance
degradation. Each table with an outstanding in-place alter
operation has an informative alert row in the ph_alert
table in the sysadmin database.
Automatic allocation of secondary partition header pages “Managing extents” on page 6-19
for extending the extent map

If you have a table that needs more extents and the

database server runs out of space on the partition header
page, the database server now automatically allocates
extended secondary partition header pages to
accommodate new extent entries. The database server can
now allocate an unlimited number of extents for any
partition, unless the size of a table dictates a limit to the
number of extents.

xx IBM Informix Performance Guide

Table 5. What's new in the IBM Informix Performance Guide for version 11.70xC1 (continued)
Overview Reference
Partitioning table and index storage by an INTERVAL “Distribution schemes” on page 9-6
strategy

You can define a storage distribution strategy for tables or

indexes that partitions data into a set of fragments that
are each based on an interval value of the fragment key,
which must be a column expression that references a
single column of a numeric, DATE, or DATETIME data
type. When rows are inserted that do not fit in the range
fragments, the database server automatically creates new
interval fragments without DBA intervention.

This kind of fragmentation strategy is useful when all

possible fragment key values in a growing table are not
known, and the DBA does not want to allocate fragments
for data that is not yet loaded.
Partitioning table and index storage by a LIST strategy “Distribution schemes” on page 9-6

You can define a storage distribution strategy for tables or

indexes that partitions data into a set of fragments that
are each based on a list of discrete values of the fragment
key. Each value in the list must be unique among the lists
for fragments of the same object. Query performance can
improve through fragment elimination when the
fragment key for a table has a finite set of values, and
queries on the table specify equality predicates on the
fragment key.
New editions and product names For more information about the Informix product family,
go to http://www.ibm.com/software/data/informix/.
IBM Informix Dynamic Server editions were withdrawn
and new Informix editions are available. Some products
were also renamed. The publications in the Informix
library pertain to the following products:
v IBM Informix database server, formerly known as IBM
Informix Dynamic Server (IDS)
v IBM OpenAdmin Tool (OAT) for Informix, formerly
known as OpenAdmin Tool for Informix Dynamic
Server (IDS)
v IBM Informix SQL Warehousing Tool, formerly known
as Informix Warehouse Feature

Example code conventions

Examples of SQL code occur throughout this publication. Except as noted, the code
is not specific to any single IBM Informix application development tool.

If only SQL statements are listed in the example, they are not delimited by
semicolons. For instance, you might see the code in the following example:
CONNECT TO stores_demo
...

DELETE FROM customer

WHERE customer_num = 121

Introduction xxi
 ...

COMMIT WORK
DISCONNECT CURRENT

To use this SQL code for a specific product, you must apply the syntax rules for
that product. For example, if you are using an SQL API, you must use EXEC SQL
at the start of each statement and a semicolon (or other appropriate delimiter) at
the end of the statement. If you are using DB–Access, you must delimit multiple
statements with semicolons.

Tip: Ellipsis points in a code example indicate that more code would be added in
a full application, but it is not necessary to show it to describe the concept that is
being discussed.

For detailed directions on using SQL statements for a particular application

development tool or SQL API, see the documentation for your product.

Additional documentation
Documentation about this release of IBM Informix products is available in various
formats.

You can access Informix technical information such as information centers,

technotes, white papers, and IBM Redbooks® publications online at
http://www.ibm.com/software/data/sw-library/.

Compliance with industry standards

IBM Informix products are compliant with various standards.

IBM Informix SQL-based products are fully compliant with SQL-92 Entry Level
(published as ANSI X3.135-1992), which is identical to ISO 9075:1992. In addition,
many features of IBM Informix database servers comply with the SQL-92
Intermediate and Full Level and X/Open SQL Common Applications Environment
(CAE) standards.

How to provide documentation feedback

You are encouraged to send your comments about IBM Informix user
documentation.

Use one of the following methods:

v Send email to [email protected].
v In the Informix information center, which is available online at
http://www.ibm.com/software/data/sw-library/, open the topic that you want
to comment on. Click the feedback link at the bottom of the page, complete the
form, and submit your feedback.
v Add comments to topics directly in the information center and read comments
that were added by other users. Share information about the product
documentation, participate in discussions with other users, rate topics, and
more!

Feedback from all methods is monitored by the team that maintains the user
documentation. The feedback methods are reserved for reporting errors and

xxii IBM Informix Performance Guide

omissions in the documentation. For immediate help with a technical problem,
contact IBM Technical Support at http://www.ibm.com/planetwide/.

We appreciate your suggestions.

Introduction xxiii
xxiv IBM Informix Performance Guide
Chapter 1. Performance basics
Performance measurement and tuning issues and methods are relevant to daily
database server administration and query execution.

Performance measurement and tuning encompass a broad area of research and

practice and can involve information beyond the scope of this publication. For
more information about the types of information contained in the publication and
the topics that are beyond the scope of this publication, see “About this
publication” on page xiii.

These topics:
v Describe a basic approach for performance measurement and tuning
v Provide guidelines for a quick start to obtain acceptable initial performance on a
small database
v Describe roles in maintaining good performance

Developing a basic approach to performance measurement and tuning

To maintain optimum performance for your database applications, develop a plan
for measuring system performance, making adjustments to maintain good
performance and taking corrective measures when performance degrades. Regular,
specific measurements can help you to anticipate and correct performance
problems.

By recognizing problems early, you can prevent them from affecting users
significantly. Early indications of a performance problem are often vague; users
might report that the system seems sluggish. Users might complain that they
cannot get all their work done, that transactions take too long to complete, that
queries take too long to process, or that the application slows down at certain
times during the day.

To determine the nature of the problem, you must measure the actual use of
system resources and evaluate the results.

Users typically report performance problems in the following situations:

v Response times for transactions or queries take longer than expected.
v Transaction throughput is insufficient to complete the required workload.
v Transaction throughput decreases.

An iterative approach to optimizing database server performance is recommended.

If repeating the steps found in the following list does not produce the desired
improvement, insufficient hardware resources or inefficient code in one or more
client applications might be causing the problem.

To optimize performance:
1. Establish performance objectives.
2. Take regular measurements of resource utilization and database activity.
3. Identify symptoms of performance problems: disproportionate utilization of
CPU, memory, or disks.

© Copyright IBM Corp. 1996, 2014 1-1

 4. Tune the operating-system configuration.
5. Tune the database server configuration.
6. Optimize the chunk and dbspace configuration, including placement of logs,
sort space, and space for temporary tables and sort files.
7. Optimize the table placement, extent sizing, and fragmentation.
8. Improve the indexes.
9. Optimize background I/O activities, including logging, checkpoints, and page
cleaning.
10. Schedule backup and batch operations for off-peak hours.
11. Optimize the implementation of the database application.
12. Repeat steps 2 through 11.

Quick start for acceptable performance on a small database

If you have a small database with each table residing on only one disk and using
only one CPU virtual processor, you can take specific measurements to help you
anticipate and correct performance problems.

To achieve acceptable initial performance on a small database:

1. Generate statistics of your tables and indexes to provide information to the
query optimizer to enable it to choose query plans with the lowest estimated
cost.
These statistics are a minimum starting point to obtain good performance for
individual queries. For guidelines, see “Update statistics when they are not
generated automatically” on page 13-11. To see the query plan that the
optimizer chooses for each query, see “Display the query plan” on page 13-1.
2. If you want a query to run in parallel with other queries, you must turn on the
Parallel Database Query (PDQ) feature.
Without table fragmentation across multiple disks, parallel scans do not occur.
With only one CPU virtual processor, parallel joins or parallel sorts do not
occur. However, PDQ priority can obtain more memory to perform the sort. For
more information, see Chapter 12, “Parallel database query (PDQ),” on page
12-1.
3. If you want to mix online transaction processing (OLTP) and decision-support
system (DSS) query applications, you can control the amount of resources a
long-running query can obtain so that your OLTP transactions are not affected.
For information about how to control PDQ resources, see “The allocation of
resources for parallel database queries” on page 12-7.
4. Monitor sessions and drill down into various details to improve the
performance of individual queries.
For information about the various tools and session details to monitor, see
“Monitoring memory usage for each session” on page 13-43 and “Monitor
sessions and threads” on page 13-47.

Performance goals
When you plan for measuring and tuning performance, you should consider
performance goals and determine which goals are the most important.

Many considerations go into establishing performance goals for the database server
and the applications that it supports. Be clear and consistent about articulating
performance goals and priorities, so that you can provide realistic and consistent

1-2 IBM Informix Performance Guide

 expectations about the performance objectives for your application. Consider the
following questions when you establish performance goals:
v Is your top priority to maximize transaction throughput, minimize response time
for specific queries, or achieve the best overall mix?
v What sort of mix between simple transactions, extended decision-support
queries, and other types of requests does the database server typically handle?
v At what point are you willing to trade transaction-processing speed for
availability or the risk of loss for a particular transaction?
v Is this database server instance used in a client/server configuration? If so, what
are the networking characteristics that affect its performance?
v What is the maximum number of users that you expect?
v Is your configuration limited by memory, disk space, or CPU resources?

The answers to these questions can help you set realistic performance goals for
your resources and your mix of applications.

Measurements of performance
You can use throughput, response time, cost per transaction, and resource
utilization measures to evaluate performance.

Throughput, response time, and cost per transaction are described in the topics
that follow.

Resource utilization can have one of two meanings, depending on the context. The
term can refer to the amount of a resource that a particular operation requires or
uses, or it can refer to the current load on a particular system component. The term
is used in the former sense to compare approaches for accomplishing a given task.
For instance, if a given sort operation requires 10 megabytes of disk space, its
resource utilization is greater than another sort operation that requires only 5
megabytes of disk space. The term is used in the latter sense to refer, for instance,
to the number of CPU cycles that are devoted to a particular query during a
specific time interval.

For a discussion about the performance impact of different load levels on various
system components, see “Resource utilization and performance” on page 1-7.

Throughput
Throughput measures the overall performance of the system. For transaction
processing systems, throughput is typically measured in transactions per second
(TPS) or transactions per minute (TPM).

Throughput depends on the following factors:

v The specifications of the host computer
v The processing overhead in the software
v The layout of data on disk
v The degree of parallelism that both hardware and software support
v The types of transactions being processed

Ways to measure throughput

The best way to measure throughput for an application is to include code in the
application that logs the time stamps of transactions as they commit.

Chapter 1. Performance basics 1-3

 If your application does not provide support for measuring throughput directly,
you can obtain an estimate by tracking the number of COMMIT WORK statements
that the database server logs during a given time interval. You can use the onlog
utility to obtain a listing of logical-log records that are written to log files. You can
use information from this command to track insert, delete, and update operations
as well as committed transactions. However, you cannot obtain information stored
in the logical-log buffer until that information is written to a log file.

If you need more immediate feedback, you can use onstat -p to gather an estimate.
You can use the SET LOG statement to set the logging mode to unbuffered for the
databases that contain tables of interest. You can also use the trusted auditing
facility in the database server to record successful COMMIT WORK events or other
events of interest in an audit log file. Using the auditing facility can increase the
overhead involved in processing any audited event, which can reduce overall
throughput.
Related information:
Auditing data security (Security Guide)

Standard throughput benchmarks

The Transaction Processing Performance Council (TPC) provides standard
benchmarks that allow reasonable throughput comparisons across hardware
configurations and database servers. IBM is an active member in good standing of
the TPC.

The TPC provides the following standardized benchmarks for measuring

throughput:
v TPC-A
This benchmark is used for simple online transaction-processing (OLTP)
comparisons. It characterizes the performance of a simple transaction-processing
system, emphasizing update-intensive services. TPC-A simulates a workload that
consists of multiple user sessions connected over a network with significant disk
I/O activity.
v TPC-B
This benchmark is used for stress-testing peak database throughput. It uses the
same transaction load as TPC-A but removes any networking and interactive
operations to provide a best-case throughput measurement.
v TPC-C
This benchmark is used for complex OLTP applications. It is derived from
TPC-A and uses a mix of updates, read-only transactions, batch operations,
transaction rollback requests, resource contentions, and other types of operations
on a complex database to provide a better representation of typical workloads.
v TPC-D
This benchmark measures query-processing power in terms of completion times
for very large queries. TPC-D is a decision-support benchmark built around a set
of typical business questions phrased as SQL queries against large databases (in
the gigabyte or terabyte range).

Because every database application has its own particular workload, you cannot
use TPC benchmarks to predict the throughput for your application. The actual
throughput that you achieve depends largely on your application.

1-4 IBM Informix Performance Guide

 Response time
Response time measures the performance of an individual transaction or query.
Response time is typically treated as the elapsed time from the moment that a user
enters a command or activates a function until the time that the application
indicates that the command or function has completed.

The response time for a typical Informix application includes the following
sequence of actions. Each action requires a certain amount of time. The response
time does not include the time that it takes for the user to think of and enter a
query or request:
1. The application forwards a query to the database server.
2. The database server performs query optimization and retrieves any
user-defined routines (UDRs). UDRs include both SPL routines and external
routines.
3. The database server retrieves, adds, or updates the appropriate records and
performs disk I/O operations directly related to the query.
4. The database server performs any background I/O operations, such as logging
and page cleaning, that occur during the period in which the query or
transaction is still pending.
5. The database server returns a result to the application.
6. The application displays the information or issues a confirmation and then
issues a new prompt to the user.

Figure 1-1 contains a diagram that shows how the actions just described in steps 1
through 6 contribute to the overall response time.

DB-Access Database server

custno custname
SELECT*in 1234 XYZLTD
Database Background 1235 XSPORTS

User enters Application Database Database Database Database Client application

request (not forwards server server server server receives, processes,
included in request to optimizes retrieves or performs modifies data and displays results
response database query and adds selected background I/O values and from database server.
time). server. retrieves records. (sometimes sends results
user-defined affects to client.
routines. response time).

Overall response time

Figure 1-1. Components of the response time for a single transaction

Response time and throughput

Response time and throughput are related. The response time for an average
transaction tends to decrease as you increase overall throughput.

However, you can decrease the response time for a specific query, at the expense of
overall throughput, by allocating a disproportionate amount of resources to that
query. Conversely, you can maintain overall throughput by restricting the resources
that the database allocates to a large query.

Chapter 1. Performance basics 1-5

 The trade-off between throughput and response time becomes evident when you
try to balance the ongoing need for high transaction throughput with an
immediate need to perform a large decision-support query. The more resources
that you apply to the query, the fewer you have available to process transactions,
and the larger the impact your query can have on transaction throughput.
Conversely, the fewer resources you allow the query, the longer the query takes.

Response-time measurement
To measure the response time for a query or application, you can use the timing
commands and performance monitoring and timing functions that your operating
system provides.

Operating-system timing commands:

Your operating system typically has a utility that you can use to time a command.
You can often use this timing utility to measure the response times to SQL
statements that a DB-Access command file issues.
UNIX Only
If you have a command file that performs a standard set of SQL
statements, you can use the time command on many systems to obtain an
accurate timing for those commands.
The following example shows the output of the UNIX time command:
time commands.dba
...
4.3 real 1.5 user 1.3 sys
The time output lists the amount of elapsed time (real), the user CPU time,
and the system CPU time. If you use the C shell, the first three columns of
output from the C shell time command show the user, system, and elapsed
times, respectively. In general, an application often performs poorly when
the proportion of system CPU time exceeds one-third of the total elapsed
time.
The time command gathers timing information about your application. You
can use this command to invoke an instance of your application, perform a
database operation, and then exit to obtain timing figures, as the following
example illustrates:
time sqlapp
(enter SQL command through sqlapp, then exit)
10.1 real 6.4 user 3.7 sys
You can use a script to run the same test repeatedly, which allows you to
obtain comparable results under different conditions. You can also obtain
estimates of your average response time by dividing the elapsed time for
the script by the number of database operations that the script performs.

Operating-system tools for monitoring performance:

Operating systems usually have a performance monitor that you can use to
measure response time for a query or process.
Windows Only
You can often use the Performance Logs and Alerts that the Windows
operating system supplies to measure the following times:
v User time
v Processor time

1-6 IBM Informix Performance Guide

 v Elapsed time

Timing functions within your application:

Most programming languages have a library function for the time of day. If you
have access to the source code, you can insert pairs of calls to this function to
measure the elapsed time between specific actions.
ESQL/C Only
For example, if the application is written in IBM Informix ESQL/C, you
can use the dtcurrent() function to obtain the current time. To measure
response time, you can call dtcurrent() to report the time at the start of a
transaction and again to report the time when the transaction commits.

Elapsed time, in a multiprogramming system or network environment where

resources are shared among multiple processes, does not always correspond to
execution time. Most operating systems and C libraries contain functions that
return the CPU time of a program.

Cost per transaction

The cost per transaction is a financial measure that is typically used to compare
overall operating costs among applications, database servers, or hardware
platforms. You can measure the cost per transaction.

To measure the cost per transaction:

1. Calculate all the costs associated with operating an application. These costs can
include the installed price of the hardware and software; operating costs,
including salaries; and other expenses. These costs can include the installed
price of the hardware and software; operating costs, including salaries; and
other expenses.
2. Project the total number of transactions and queries for the effective life of an
application.
3. Divide the total cost over the total number of transactions.

Although this measure is useful for planning and evaluation, it is seldom relevant
to the daily issues of achieving optimum performance.

Resource utilization and performance

A typical transaction-processing application undergoes different demands
throughout its various operating cycles. Peak loads during the day, week, month,
and year, as well as the loads imposed by decision-support (DSS) queries or
backup operations, can significantly impact any system that is running near
capacity. You can use direct historical data derived from your particular system to
pinpoint this impact.

You must take regular measurements of the workload and performance of your
system to predict peak loads and compare performance measurements at different
points in your usage cycle. Regular measurements help you to develop an overall
performance profile for your database server applications. This profile is critical in
determining how to improve performance reliably.

Chapter 1. Performance basics 1-7

 For the measurement tools that the database server provides, see “Database server
tools” on page 2-3. For the tools that your operating system provides for
measuring performance impacts on system and hardware resources, see
“Operating-system tools” on page 2-3.

Utilization is the percentage of time that a component is actually occupied, as

compared with the total time that the component is available for use. For instance,
if a CPU processes transactions for a total of 40 seconds during a single minute, its
utilization during that interval is 67 percent.

Measure and record utilization of the following system resources regularly:

v CPU
v Memory
v Disk

A resource is said to be critical to performance when it becomes overused or when

its utilization is disproportionate to that of other components. For instance, you
might consider a disk to be critical or overused when it has a utilization of 70
percent and all other disks on the system have 30 percent. Although 70 percent
does not indicate that the disk is severely overused, you can improve performance
by rearranging data to balance I/O requests across the entire set of disks.

How you measure resource utilization depends on the tools that your operating
system provides for reporting system activity and resource utilization. After you
identify a resource that seems overused, you can use the performance-monitoring
utilities that the database server provides to gather data and make inferences about
the database activities that might account for the load on that component. You can
adjust your database server configuration or your operating system to reduce those
database activities or spread them among other components. In some cases, you
might need to provide additional hardware resources to resolve a performance
bottleneck.

Resource utilization
Whenever a system resource, such as a CPU or a particular disk, is occupied by a
transaction or query, the resource is unavailable for processing other requests.
Pending requests must wait for the resources to become available before they can
complete.

When a component is too busy to keep up with all its requests, the overused
component becomes a bottleneck in the flow of activity. The higher the percentage
of time that the resource is occupied, the longer each operation must wait for its
turn.

You can use the following formula to estimate the service time for a request based
on the overall utilization of the component that services the request. The expected
service time includes the time that is spent both waiting for and using the resource
in question. Think of service time as that portion of the response time accounted
for by a single component within your computer, as the following formula shows:
S= P/(1-U)
S is the expected service time.
P is the processing time that the operation requires after it obtains the
resource.
U is the utilization for the resource (expressed as a decimal).

1-8 IBM Informix Performance Guide

 As Figure 1-2 shows, the service time for a single component increases dramatically
as the utilization increases beyond 70 percent. For instance, if a transaction requires
1 second of processing by a given component, you can expect it to take 2 seconds
on a component at 50 percent utilization and 5 seconds on a component at 80
percent utilization. When utilization for the resource reaches 90 percent, you can
expect the transaction to take 10 seconds to make its way through that component.

Elapsed 12
time (as a 10
multiple of 8
processing 6
time) in 4
minutes
2
0
0 10 20 30 40 50 60 70 80 90 100
Resource utilization (%)

Figure 1-2. Service Time for a Single Component as a Function of Resource Utilization

If the average response time for a typical transaction soars from 2 or 3 seconds to
10 seconds or more, users are certain to notice and complain.

Important: Monitor any system resource that shows a utilization of over 70

percent or any resource that exhibits symptoms of overuse as described in the
following sections.

When you consider resource utilization, also consider whether increasing the page
size of a standard or temporary dbspace is beneficial in your environment. If you
want a longer key length than is available for the default page size of a standard
or temporary dbspace, you can increase the page size.

CPU utilization
Estimates of CPU utilization and response time can help you determine if you
need to eliminate or reschedule some activities.

You can use the resource-utilization formula in the previous topic (“Resource
utilization” on page 1-8) to estimate the response time for a heavily loaded CPU.
However, high utilization for the CPU does not always indicate a performance
problem. The CPU performs all calculations that are needed to process
transactions. The more transaction-related calculations that it performs within a
given period, the higher the throughput will be for that period. As long as
transaction throughput is high and seems to remain proportional to CPU
utilization, a high CPU utilization indicates that the computer is being used to the
fullest advantage.

On the other hand, when CPU utilization is high but transaction throughput does
not keep pace, the CPU is either processing transactions inefficiently or it is
engaged in activity not directly related to transaction processing. CPU cycles are
being diverted to internal housekeeping tasks such as memory management.

You can easily eliminate the following activities:

v Large queries that might be better scheduled at an off-peak time

Chapter 1. Performance basics 1-9

 v Unrelated application programs that might be better performed on another
computer

If the response time for transactions increases to such an extent that delays become
unacceptable, the processor might be swamped; the transaction load might be too
high for the computer to manage. Slow response time can also indicate that the
CPU is processing transactions inefficiently or that CPU cycles are being diverted.

When CPU utilization is high, a detailed analysis of the activities that the database
server performs can reveal any sources of inefficiency that might be present due to
improper configuration. For information about analyzing database server activity,
see “Database server tools” on page 2-3.

Memory utilization
Memory is not managed as a single component, such as a CPU or disk, but as a
collection of small components called pages.

The size of a typical page in memory can range from 1 to 8 kilobytes, depending
on your operating system. A computer with 64 megabytes of memory and a page
size of 2 kilobytes contains approximately 32,000 pages.

When the operating system needs to allocate memory for use by a process, it
scavenges any unused pages within memory that it can find. If no free pages exist,
the memory-management system has to choose pages that other processes are still
using and that seem least likely to be needed in the short run. CPU cycles are
required to select those pages. The process of locating such pages is called a page
scan. CPU utilization increases when a page scan is required.

Memory-management systems typically use a least recently used algorithm to select

pages that can be copied out to disk and then freed for use by other processes.
When the CPU has identified pages that it can appropriate, it pages out the old
page images by copying the old data from those pages to a dedicated disk. The
disk or disk partition that stores the page images is called the swap disk, swap space,
or swap area. This paging activity requires CPU cycles as well as I/O operations.

Eventually, page images that have been copied to the swap disk must be brought
back in for use by the processes that require them. If there are still too few free
pages, more must be paged out to make room. As memory comes under increasing
demand and paging activity increases, this activity can reach a point at which the
CPU is almost fully occupied with paging activity. A system in this condition is
said to be thrashing. When a computer is thrashing, all useful work comes to a halt.

To prevent thrashing, some operating systems use a coarser memory-management

algorithm after paging activity crosses a certain threshold. This algorithm is called
swapping. When the memory-management system resorts to swapping, it
appropriates all pages that constitute an entire process image at once, rather than a
page at a time.

Swapping frees up more memory with each operation. However, as swapping

continues, every process that is swapped out must be read in again, dramatically
increasing disk I/O to the swap device and the time required to switch between
processes. Performance is then limited to the speed at which data can be
transferred from the swap disk back into memory. Swapping is a symptom of a
system that is severely overloaded, and throughput is impaired.

1-10 IBM Informix Performance Guide

 Many systems provide information about paging activity that includes the number
of page scans performed, the number of pages sent out of memory (paged out), and
the number of pages brought in from memory (paged in):
v Paging out is the critical factor because the operating system pages out only
when it cannot find pages that are free already.
v A high rate of page scans provides an early indicator that memory utilization is
becoming a bottleneck.
v Pages for terminated processes are freed in place and simply reused, so
paging-in activity does not provide an accurate reflection of the load on memory.
A high rate of paging in can result from a high rate of process turnover with no
significant performance impact.

Although the principle for estimating the service time for memory is the same as
that described in “Resource utilization and performance” on page 1-7, you use a
different formula to estimate the performance impact of memory utilization than
you do for other system components.

You can use the following formula to calculate the expected paging delay for a
given CPU utilization level and paging rate:
PD= (C/(1-U)) * R * T
PD is the paging delay.
C is the CPU service time for a transaction.
U is the CPU utilization (expressed as a decimal).
R is the paging-out rate.
T is the service time for the swap device.

As paging increases, CPU utilization also increases, and these increases are
compounded. If a paging rate of 10 per second accounts for 5 percent of CPU
utilization, increasing the paging rate to 20 per second might increase CPU
utilization by an additional 5 percent. Further increases in paging lead to even
sharper increases in CPU utilization, until the expected service time for CPU
requests becomes unacceptable.

Disk utilization
Because transfer rates vary among disks, most operating systems do not report
disk utilization directly. Instead, they report the number of data transfers per
second (in operating-system memory-page-size units.)

Because each disk acts as a single resource, you can use the following basic
formula to estimate the service time, which is described in detail in “Resource
utilization” on page 1-8:
S= P/(1-U)

To compare the load on disks with similar access times, simply compare the
average number of transfers per second.

If you know the access time for a given disk, you can use the number of transfers
per second that the operating system reports to calculate utilization for the disk. To
do so, multiply the average number of transfers per second by the access time for
the disk as listed by the disk manufacturer. Depending on how your data is laid

Chapter 1. Performance basics 1-11

 out on the disk, your access times can vary from the rating of the manufacturer. To
account for this variability, you should add 20 percent to the access-time
specification of the manufacturer.

The following example shows how to calculate the utilization for a disk with a
30-millisecond access time and an average of 10 transfer requests per second:
U = (A * 1.2) * X
= (.03 * 1.2) * 10
= .36
U is the resource utilization (this time of a disk).
A is the access time (in seconds) that the manufacturer lists.
X is the number of transfers per second that your operating system reports.

You can use the utilization to estimate the processing time at the disk for a
transaction that requires a given number of disk transfers. To calculate the
processing time at the disk, multiply the number of disk transfers by the average
access time. Include an extra 20 percent to account for access-time variability:
P = D (A * 1.2)
P is the processing time at the disk.
D is the number of disk transfers.
A is the access time (in seconds) that the manufacturer lists.

For example, you can calculate the processing time for a transaction that requires
20 disk transfers from a 30-millisecond disk as follows:
P = 20 (.03 * 1.2)
= 20 * .036
= .72

Use the processing time and utilization values that you calculated to estimate the
expected service time for I/O at the particular disk, as the following example
shows:
S = P/(1-U)
= .72 / (1 - .36)
= .72 / .64
= 1.13

Factors that affect resource utilization

The performance of your database server application depends many factors,
including hardware and software configuration, your network configuration, and
the design of your database.

You must consider these factors when you attempt to identify performance
problems or make adjustments to your system:
v Hardware resources
As discussed earlier in this chapter, hardware resources include the CPU,
physical memory, and disk I/O subsystems.
v Operating-system configuration
The database server depends on the operating system to provide low-level
access to devices, process scheduling, interprocess communication, and other
vital services.

1-12 IBM Informix Performance Guide

 The configuration of your operating system has a direct impact on how well the
database server performs. The operating-system kernel takes up a significant
amount of physical memory that the database server or other applications
cannot use. However, you must reserve adequate kernel resources for the
database server to use.
v Network configuration and traffic
Applications that depend on a network for communication with the database
server, and systems that rely on data replication to maintain high availability, are
subject to the performance constraints of that network. Data transfers over a
network are typically slower than data transfers from a disk. Network delays
can have a significant impact on the performance of the database server and
other application programs that run on the host computer.
v Database server configuration
Characteristics of your database server instance, such as the number of CPU
virtual processors (VPs), the size of your resident and virtual shared-memory
portions, and the number of users, play an important role in determining the
capacity and performance of your applications.
v Dbspace, blobspace, and chunk configuration
The following factors can affect the time that it takes the database server to
perform disk I/O and process transactions:
– The placement of the root dbspace, physical logs, logical logs, and
temporary-table dbspaces
– The presence or absence of mirroring
– The use of devices that are buffered or unbuffered by the operation system
v Database and table placement
The placement of tables and fragments within dbspaces, the isolation of high-use
fragments in separate dbspaces, and the spreading of fragments across multiple
dbspaces can affect the speed at which the database server can locate data pages
and transfer them to memory.
v Tblspace organization and extent sizing
Fragmentation strategy and the size and placement of extents can affect the
ability of the database server to scan a table rapidly for data. Avoid interleaved
extents and allocate extents that are sufficient to accommodate growth of a table
to prevent performance problems.
v Query efficiency
Proper query construction and cursor use can decrease the load that any one
application or user imposes. Remind users and application developers that
others require access to the database and that each person's activities affect the
resources that are available to others.
v Scheduling background I/O activities
Logging, checkpoints, page cleaning, and other operations, such as making
backups or running large decision-support queries, can impose constant
overhead and large temporary loads on the system. Schedule backup and batch
operations for off-peak times whenever possible.
v Remote client/server operations and distributed join operations
These operations have an important impact on performance, especially on a host
system that coordinates distributed joins.
v Application-code efficiency
Application programs introduce their own load on the operating system, the
network, and the database server. These programs can introduce performance
problems if they make poor use of system resources, generate undue network

Chapter 1. Performance basics 1-13

 traffic, or create unnecessary contention in the database server. Application
developers must make proper use of cursors and locking levels to ensure good
database server performance.

Maintenance of good performance

Performance is affected in some way by all system users: the database server
administrator, the database administrator, the application designers, and the client
application users.

The database server administrator usually coordinates the activities of all users to
ensure that system performance meets overall expectations. For example, the
operating-system administrator might need to reconfigure the operating system to
increase the amount of shared memory. Bringing down the operating system to
install the new configuration requires bringing the database server down. The
database server administrator must schedule this downtime and notify all affected
users when the system will be unavailable.

The database server administrator should:

v Be aware of all performance-related activities that occur.
v Educate users about the importance of performance, how performance-related
activities affect them, and how they can assist in achieving and maintaining
optimal performance.

The database administrator should pay attention to:

v How tables and queries affect the overall performance of the database server
v The placement of tables and fragments
v How the distribution of data across disks affects performance

Application developers should:

v Carefully design applications to use the concurrency and sorting facilities that
the database server provides, rather than attempt to implement similar facilities
in the application.
v Keep the scope and duration of locks to the minimum to avoid contention for
database resources.
v Include routines within applications that, when temporarily enabled at runtime,
allow the database server administrator to monitor response times and
transaction throughput.

Database users should:

v Pay attention to performance and report problems to the database server
administrator promptly.
v Be courteous when they schedule large, decision-support queries and request as
few resources as possible to get the work done.

1-14 IBM Informix Performance Guide

Chapter 2. Performance monitoring and the tools you use
You can use performance monitoring tools to create a performance history, to
monitor database resources at scheduled times, or to monitor ongoing transaction
or query performance.

This chapter also contains cross-references to topics that about how to interpret the
results of performance monitoring

The kinds of data that you need to collect depend on the kinds of applications that
you run on your system. The causes of performance problems on OLTP (online
transaction processing) systems are different from the causes of problems on
systems that are used primarily for DSS query applications. Systems with mixed
use provide a greater performance-tuning challenge and require a sophisticated
analysis of performance-problem causes.

Evaluate the current configuration

Before you begin to adjust the configuration of your database server, evaluate the
performance of your current configuration. You can view the contents of your
configuration file with onstat commands or IBM OpenAdmin Tool (OAT) for
Informix.

To alter certain database server characteristics, you must bring down the database
server, which can affect your production system. Some configuration adjustments
can unintentionally decrease performance or cause other negative side effects.

If your database applications satisfy user expectations, avoid frequent adjustments,

even if those adjustments might theoretically improve performance. If your users
are reasonably satisfied, take a measured approach to reconfiguring the database
server. When possible, use a test instance of the database server to evaluate
configuration changes before you reconfigure your production system.

When performance problems relate to backup operations, you might also examine
the number or transfer rates for tape drives. You might need to alter the layout or
fragmentation of your tables to reduce the impact of backup operations. For
information about disk layout and table fragmentation, see Chapter 6, “Table
performance considerations,” on page 6-1 and Chapter 7, “Indexes and index
performance considerations,” on page 7-1.

For client/server configurations, consider network performance and availability.

Evaluating network performance is beyond the scope of this publication. For
information about monitoring network activity and improving network availability,
see your network administrator or see the documentation for your networking
package.

Determine whether you want to set the configuration parameters that help
maintain server performance by automatically adjusting properties of the database
server while it is running, for example:
v AUTO_AIOVPS: Adds AIO virtual processors when I/O workload increases.
v AUTO_CKPTS: Increases the frequency of checkpoints to avoid transaction
blocking.

© Copyright IBM Corp. 1996, 2014 2-1

 v AUTO_LRU_TUNING: Manages cached data flushing as the server load
changes.
v AUTO_READAHEAD: Changes the automatic read-ahead mode or disables
automatic read-ahead operations for a query.
v AUTO_REPREPARE: Reoptimizes SPL routines and reprepares prepared objects
after a schema change.
v AUTO_STAT_MODE: Enables or disables the mode for selectively updating only
stale or missing data distributions in UPDATE STATISTICS operations.
v DYNAMIC_LOGS: Allocates additional log files when necessary.
v LOCKS: Allocates additional locks when necessary.
v RTO_SERVER_RESTART: Provides the best performance possible while meeting
the recovery time objective after a problem.
Related reference:
onstat -c command: Print ONCONFIG file contents (Administrator's Reference)

onstat -g cfg command: Print the current values of configuration parameters

(Administrator's Reference)

Create a performance history

As soon as you set up your database server and begin to run applications on it,
you should begin scheduled monitoring of resource use. As you accumulate data,
you can analyze performance information.

To accumulate data for performance analysis, use the command-line utilities

described in “Database server tools” on page 2-3 and “Operating-system tools” on
page 2-3 in operating scripts or batch files.

The importance of a performance history

If you have a history of the performance of your system, you can begin to track
the cause of problems as soon as users report slow response or inadequate
throughput.

If a history is not available, you must start tracking performance after a problem
arises, and you might not be able to tell when and how the problem began. Trying
to identify problems after the fact significantly delays resolution of a performance
problem.

To build a performance history and profile of your system, take regular snapshots
of resource-utilization information.

For example, if you chart the CPU utilization, paging-out rate, and the I/O transfer
rates for the various disks on your system, you can begin to identify peak-use
levels, peak-use intervals, and heavily loaded resources.

If you monitor fragment use, you can determine whether your fragmentation
scheme is correctly configured. Monitor other resource use as appropriate for your
database server configuration and the applications that run on it.

2-2 IBM Informix Performance Guide

 Choose tools from those described in the following sections, and create jobs that
build up a history of disk, memory, I/O, and other database server resource use.
To help you decide which tools to use to create a performance history, this chapter
briefly describes the output of each tool.

Tools that create a performance history

When you monitor database server performance, you use tools from the host
operating system and command-line utilities that you can run at regular intervals
from scripts or batch files.

You also use performance monitoring tools with a graphical interface to monitor
critical aspects of performance as queries and transactions are performed.

Operating-system tools
The database server relies on the operating system of the host computer to provide
access to system resources such as the CPU, memory, and various unbuffered disk
I/O interfaces and files. Each operating system has its own set of utilities for
reporting how system resources are used.

Different implementations of some operating systems have monitoring utilities

with the same name but different options and informational displays.

UNIX Only

The following table lists some UNIX utilities that monitor system resources.

UNIX Utility Description

vmstat utility Displays virtual-memory statistics
iostat utility Displays I/O utilization statistics
sar utility Displays a variety of resource statistics
ps utility Displays active process information

For details on how to monitor your operating-system resources, consult the

reference manual or your system administration guide.

To capture the status of system resources at regular intervals, use scheduling tools
that are available with your host operating system (for example, cron) as part of
your performance monitoring system.

Windows Only

You can often use the Performance Logs and Alerts that the Windows operating
system supplies to monitor resources such as processor, memory, cache, threads,
and processes. The Performance Logs and Alerts also provide charts, alerts,
reports, and the ability to save information to log files for later analysis.

For more information about how to use the Performance Logs and Alerts, consult
your operating-system manuals.

Database server tools

The database server provides tools and utilities that capture snapshot information
about your configuration and performance.

Chapter 2. Performance monitoring and the tools you use 2-3

 You can use these utilities regularly to build a historical profile of database activity,
which you can compare with current operating-system resource-utilization data.
These comparisons can help you discover which database server activities have the
greatest impact on system-resource utilization. You can use this information to
identify and manage your high-impact activities or adjust your database server or
operating-system configuration.

The database server tools and utilities that you can use for performance
monitoring include:
v IBM OpenAdmin Tool (OAT) for Informix
v The onstat utility
v The onlog utility
v The oncheck utility
v The ON-Monitor utility (on UNIX only)
v The onperf utility (on UNIX only)
v DB-Access and the system-monitoring interface (SMI), which you can use to
monitor performance from within your application
v SQL administration API commands

You can use onstat, onlog, or oncheck commands invoked by the cron scheduling
facility to capture performance-related information at regular intervals and build a
historical performance profile of your database server application. The following
sections describe these utilities.

You can use SQL SELECT statements to query the system-monitoring interface
(SMI) from within your application.

The SMI tables are a collection of tables and pseudo-tables in the sysmaster
database that contain dynamically updated information about the operation of the
database server. The database server constructs these tables in memory but does
not record them on disk. The onstat utility options obtain information from these
SMI tables.

You can use cron and SQL scripts with DB-Access or onstat utility options to
query SMI tables at regular intervals.

Tip: The SMI tables are different from the system catalog tables. System catalog
tables contain permanently stored and updated information about each database
and its tables (sometimes referred to as metadata or a data dictionary).

You can use ON-Monitor to check the current database server configuration.

You can use onperf to display database server activity with the Motif window
manager.
Related concepts:
The onlog utility (Administrator's Reference)
The oncheck Utility (Administrator's Reference)
DB-Access User's Guide (DB-Access Guide)
System catalog tables (SQL Reference)
Chapter 14, “The onperf utility on UNIX,” on page 14-1
Related reference:

2-4 IBM Informix Performance Guide

 The onstat utility (Administrator's Reference)
The System-Monitoring Interface Tables (Administrator's Reference)
SQL administration API portal: Arguments by functional category
(Administrator's Reference)

Performance information that IBM Informix Server Administrator provides:

IBM Informix Server Administrator (ISA) is a browser-based tool that provides

Web-based system administration for the entire range of IBM Informix database
servers.

ISA is the first in a new generation of browser-based, cross-platform administrative

tools. It provides access to every Informix database server command-line function
and presents the output in an easy-to-read format.

The database server CD-ROM distributed with your product includes ISA. For
information on how to install ISA, see the following file on the CD-ROM.
Table 2-1. Operating system file
Operating System File
UNIX /SVR_ADM/README
Windows \SVR_ADM\readme.txt

With ISA, you can use a browser to perform these common database server
administrative tasks:
v Change configuration parameters temporarily or permanently
v Change the database server mode between online and offline and its
intermediate states
v Modify connectivity information in the sqlhosts file
v Check dbspaces, sbspaces, logs, and other objects
v Manage logical and physical logs
v Examine memory use and adding and freeing memory segments
v Read the message log
v Back up and restore dbspaces and sbspaces
v Run various onstat commands to monitor performance
v Enter simple SQL statements and examine database schemas
v Add and remove chunks, dbspaces, and sbspaces
v Examine and manage user sessions
v Examine and manage virtual processors (VPs)
v Use the High-Performance Loader (HPL), dbimport, and dbexport
v Manage Enterprise Replication
v Manage an IBM Informix MaxConnect server
v Use the following utilities: dbaccess, dbschema, onbar, oncheck, ondblog,
oninit, onlog, onmode, onparams, onspaces, and onstat

You also can enter any Informix utility, UNIX shell command, or Windows
command (for example, oncheck -cd; ls -l).

Chapter 2. Performance monitoring and the tools you use 2-5

 Performance information that the onstat utility displays:

The onstat utility displays a wide variety of performance-related and status

information contained within the SMI tables. You can use the onstat utility to
check the current status of the database server and monitor the activities of the
database server.

For a complete list of all onstat options, use the onstat - - command. For a
complete display of all the information that onstat gathers, use the onstat -a
command.

Tip: Profile information displayed by onstat commands, such as onstat -p,

accumulates from the time the database server was started. To clear performance
profile statistics so that you can create a new profile, run the onstat -z. If you use
onstat -z to reset statistics for a performance history or appraisal, ensure that other
users do not also enter the command at different intervals.

The following table lists some of the onstat commands that display general
performance-related information.
Table 2-2. onstat commands that display performance information
onstat command Description
onstat -p Displays a performance profile that includes
the number of reads and writes, the number
of times that a resource was requested but
was not available, and other miscellaneous
information
onstat -b Displays information about buffers currently
in use
onstat -l Displays information about the physical and
logical logs
onstat -x Displays information about transactions,
including the thread identifier of the user
who owns the transaction
onstat -u Displays a user activity profile that provides
information about user threads including the
thread owner's session ID and login name
onstat -R Displays information about buffer pools,
including information about buffer pool page
size.
onstat -F Displays page-cleaning statistics that include
the number of writes of each type that
flushes pages to disk
onstat -g Requires an additional argument that
specifies the information to be displayed

For example, onstat -g mem displays

memory statistics.

For more information about options that provide performance-related information,

see “Monitoring fragmentation with the onstat -g ppf command” on page 9-28 and
“Monitor database server resources” on page 2-7.

2-6 IBM Informix Performance Guide

 Related reference:
onstat -g monitoring options (Administrator's Reference)

Monitor performance with the OpenAdmin Tool (OAT) for Informix

The IBM OpenAdmin Tool (OAT) for Informix provides multiple ways to gather,
view, and analyze performance data.

With OAT, you can:

v Collect performance statistics.
v Find and eliminate database server performance bottlenecks.
v Identify and monitor queries that are critical to performance.
v Improve checkpoint performance and manage LRU queues.
v Manage compression of data in tables and table-fragment rows
v Monitor critical system resources (CPU, memory, disk, virtual processors).
v Monitor and track locking.
v Optimize the disk layout.
v Tune the buffer cache.
v Use query drill-down.
v View the SQL statement cache.
v Use automatic statistics update.
v View historical performance graphs.
v Explore user sessions.

Monitor database server resources

Monitor specific database server resources to identify performance bottlenecks and
potential trouble spots and to improve resource use and response time.

One of the most useful commands for monitoring system resources is onstat -g
and its many options.

Monitor resources that impact CPU utilization

Threads, network communications, and virtual processors impact CPU utilization.
You can use onstat -g arguments to monitor threads, network communications,
and virtual processors.

Use the following onstat -g command options to monitor threads.

onstat -g Option Description

act Displays active threads.
ath Displays all threads.

The sqlexec threads represent portions of

client sessions; the rstcb value corresponds to
the user field of the onstat -u command.
cpu Displays the last time the thread ran, how
much CPU time the thread used, the number
of times the thread ran, and other statistics
about all the threads running in the server.

Chapter 2. Performance monitoring and the tools you use 2-7

 onstat -g Option Description
rea Displays ready threads.
sle Displays all sleeping threads.
sts Displays maximum and current stack use per
thread.
tpf tid Displays a thread profile for tid.

If tid is 0, this argument displays profiles for

all threads.
wai Displays waiting threads, including all
threads waiting on mutex or condition, or
yielding.

Use the following onstat -g command options to monitor the network.

onstat -g Command Option Description

ntd Displays network statistics by service.
ntt Displays network user times.
ntu Displays network user statistics.
qst Displays queue statistics.

Use the following onstat -g command options to monitor virtual processors.

onstat -g Command Option Description

glo Displays global multithreading information,
including CPU-use information about virtual
processors, the total number of sessions, and
other multithreading global counters.
sch Displays the number of semaphore
operations, spins, and busy waits for each
VP.
spi Displays spin locks that are acquired by
virtual processors after they have spun more
than 10,000 times.

To reduce contention, reduce the number of

virtual processors, reduce the load on the
computer, or, on some platforms, use the
no-age or processor affinity options of
virtual processors. If sh_lock mutexes have
highly contended spin locks, create private
memory caches for CPU virtual processors
by setting the VP_MEMORY_CACHE_VP
configuration parameter.
wst Displays wait statistics.

Monitor memory utilization

You can use some specific onstat -g command options to monitor memory
utilization.

2-8 IBM Informix Performance Guide

Use the following onstat -g options to monitor memory utilization. For overall
memory information, omit table name, pool name, or session id from the commands
that permit those optional parameters.
Table 2-3. onstat -g Options for monitoring memory utilization
Argument Description
ffr pool name | session id Displays free fragments for a pool of shared
memory or by session
dic table name Displays one line of information for each
table cached in the shared-memory
dictionary

If you provide a specific table name as a

parameter, this argument displays internal
SQL information about that table.
dsc Displays one line of information for each
column of distribution statistics cached in the
data distribution cache.
mem pool name | session id Displays memory statistics for the pools that
are associated with a session

If you omit pool_name | session id, this

argument displays pool information for all
sessions.
mgm Displays Memory Grant Manager resource
information, including:
v The values of the PDQ configuration
parameters
v Memory and scan information
v Load information, such as the number of
queries that are waiting for memory, the
number of queries that are waiting for
scans, the number of queries that are
waiting for queries with higher PDQ
priority to run, and the number of queries
that are waiting for a query slot
v Active queries and the number of queries
at each gate
v Statistics on free resources
v Statistics on queries
v The resource/lock cycle prevention count,
which shows the number of times the
system immediately activated a query to
avoid a potential deadlock
nsc client id Displays shared-memory status by client ID

If you omit client id, this argument displays

all client status areas.
nsd Displays network shared-memory data for
poll threads
nss session id Displays network shared-memory status by
session id

If you omit session id, this argument

displays all session status areas.

Chapter 2. Performance monitoring and the tools you use 2-9

 Table 2-3. onstat -g Options for monitoring memory utilization (continued)
Argument Description
osi Displays information about your operating
system resources and parameters, including
shared memory and semaphore parameters,
the amount of memory currently configured
on the computer, and the amount of memory
that is unused

Use this option when the server is not online.

prc Displays one line of information for each
user-defined routine (SPL routine or external
routine written in C or Java™ programming
language) cached in the UDR cache
seg Displays shared-memory-segment statistics

This argument shows the number and size of

all attached segments.
ses session id Displays memory usage for session id

If you omit session id, this argument

displays memory usage for all sessions.
ssc Displays one line of information for each
query cached in the SQL statement cache
stm session id Displays memory usage of each SQL
statement for session id

If you omit session id, this argument

displays memory usage for all sessions.
ufr pool name | session id Displays allocated pool fragments by user or
session

Related reference:
onstat -g monitoring options (Administrator's Reference)

Monitor disk I/O utilization

You can use some specific onstat -g arguments and the oncheck utility to
determine if your disk I/O operations are efficient for your applications.

Using onstat -g to monitor I/O utilization

You can use some specific onstat -g command arguments to monitor disk IO.

Use the following onstat -g command arguments to monitor disk I/O utilization.

onstat -g Argument Description

iof Displays asynchronous I/O statistics by
chunk or file

This argument is similar to the onstat -d,

except that information about nonchunk files
also appears. This argument displays
information about temporary dbspaces and
sort files.
iog Displays asynchronous I/O global
information

2-10 IBM Informix Performance Guide

onstat -g Argument Description
ioq Displays asynchronous I/O queuing statistics
iov Displays asynchronous I/O statistics by
virtual processor

For a detailed case study that uses various onstat outputs, see Appendix A, “Case
studies and examples,” on page A-1.

Using the oncheck utility to monitor I/O utilization

Disk I/O operations are usually the longest component of the response time for a
query. You can use the oncheck Utility to monitor disk I/O operations.

Contiguously allocated disk space improves sequential disk I/O operations,

because the database server can read in larger blocks of data and use the
read-ahead feature to reduce the number of I/O operations.

The oncheck utility displays information about storage structures on a disk,

including chunks, dbspaces, blobspaces, extents, data rows, system catalog tables,
and other options. You can also use oncheck to determine the number of extents
that exist within a table and whether or not a table occupies contiguous space.

The oncheck utility provides the following options and information that apply to
contiguous space and extents.

Option Information
-pB Blobspace simple large object (TEXT or BYTE data)

For information about how to use this option to determine the efficiency of
blobpage size, see “Determine blobpage fullness with oncheck -pB output”
on page 5-18.
-pe Chunks and extents

For information about how to use this option to monitor extents, see
“Checking for extent interleaving” on page 6-24 and “Eliminating interleaved
extents” on page 6-24.
-pk Index key values.

For information about how to improve the performance of this option, see
“Improving performance for index checks” on page 7-20.
-pK Index keys and row IDs

For information about how to improve the performance of this option, see
“Improving performance for index checks” on page 7-20.
-pl Index-leaf key values

For information about how to improve the performance of this option, see
“Improving performance for index checks” on page 7-20.
-pL Index-leaf key values and row IDs

For information about how to improve the performance of this option, see
“Improving performance for index checks” on page 7-20.
-pp Pages by table or fragment

For information about how to use this option to monitor space, see
“Considering the upper limit on extents” on page 6-24.

Chapter 2. Performance monitoring and the tools you use 2-11

 Option Information
-pP Pages by chunk

For information about how to use this option to monitor extents, see
“Considering the upper limit on extents” on page 6-24.
-pr Root reserved pages

For information about how to use this option, see “Estimating tables with
fixed-length rows” on page 6-5.
-ps Space used by smart large objects and metadata in sbspace.
-pS Space used by smart large objects and metadata in sbspace and storage
characteristics

For information about how to use this option to monitor space, see
“Monitoring sbspaces” on page 6-13.
-pt Space used by table or fragment

For information about how to use this option to monitor space, see
“Estimating table size” on page 6-5.
-pT Space used by table, including indexes

For information about how to use this option to monitor space, see
“Performance of in-place alters for DDL operations” on page 6-40.

For more information about using oncheck to monitor space, see “Estimating table
size” on page 6-5. For more information about concurrency during oncheck
execution, see “Improving performance for index checks” on page 7-20.
Related concepts:
The oncheck Utility (Administrator's Reference)

Monitor transactions
You can use the onlog and onstat utilities to monitor transactions.

Using the onlog utility to monitor transactions

The onlog utility displays all or selected portions of the logical log. This utility can
help you identify a problematic transaction or gauge transaction activity that
corresponds to a period of high utilization, as indicated by your periodic snapshots
of database activity and system-resource consumption.

This onlog utility can take input from selected log files, the entire logical log, or a
backup tape of previous log files.

Use onlog with caution when you read logical-log files still on disk, because
attempting to read unreleased log files stops other database activity. For greatest
safety, back up the logical-log files first and then read the contents of the backup
files. With proper care, you can use the onlog -n option to restrict onlog only to
logical-log files that have been released.

To check on the status of logical-log files, use onstat -l.

Related concepts:
The onlog utility (Administrator's Reference)

2-12 IBM Informix Performance Guide

 Using the onstat utility to monitor transactions
If the throughput of transactions is not very high, you can use some onstat utility
commands to identify a transaction that might be a bottleneck.

Use the following onstat utility commands to monitor transactions.

onstat command Description

onstat -x Displays transaction information such as
number of locks held and isolation level.
onstat -u Displays information about each user thread
onstat -k Displays locks held by each session
onstat -g sql Displays last SQL statement this session
executed

Related reference:
The onstat utility (Administrator's Reference)

Monitor sessions and queries

Monitoring sessions and threads is important for sessions that perform queries as
well as sessions that perform inserts, updates, and deletes. Some of the information
that you can monitor for sessions and threads allows you to determine if an
application is using a disproportionate amount of the resources.

To monitor database server activity, you can view the number of active sessions
and the amount of resources that they are using.

Monitoring memory usage for each session

You can use some specific onstat -g command arguments to get memory
information for each session.

Use the following command arguments to get memory information for each
session.

onstat -g command argument Description

ses Displays one-line summaries of all active
sessions
ses session id Displays session information by session id
sql session id Displays SQL information by session

If you omit session id, this argument displays

summaries of all sessions.
stm session id Displays amount of memory used by each
prepared SQL statement in a session

If you omit session id, this argument displays

information for all prepared statements.

For examples and discussions of session-monitoring command-line utilities, see

“Monitoring memory usage for each session” on page 13-43 and “Monitor sessions
and threads” on page 13-47.

Chapter 2. Performance monitoring and the tools you use 2-13

 Using the SET EXPLAIN statement
You can use the SET EXPLAIN statement or the EXPLAIN directive to display the
query plan that the optimizer creates for an individual query.

For more information, see “Display the query plan” on page 13-1.

2-14 IBM Informix Performance Guide

Chapter 3. Effect of configuration on CPU utilization
The combination of operating-system and Informix configuration parameters can
affect CPU utilization. You can change the settings of the Informix configuration
parameters that directly affect CPU utilization, and you can adjust the settings for
different types of workloads.

Multiple database server instances that run on the same host computer perform
poorly when compared with a single database server instance that manages
multiple databases. Multiple database server instances cannot balance their loads
as effectively as a single database server. Avoid multiple residency for production
environments in which performance is critical.

UNIX configuration parameters that affect CPU utilization

Your database server distribution includes a machine notes file that contains
recommended values for UNIX configuration parameters. Because the UNIX
parameters affect CPU utilization, you should compare the values in the machine
notes file with your current operating-system configuration.

The following UNIX parameters affect CPU utilization:

v Semaphore parameters
v Parameters that set the maximum number of open file descriptors
v Memory configuration parameters

UNIX semaphore parameters

Semaphores are kernel resources with a typical size of 1 byte each. Semaphores for
the database server are in addition to any that you allocate for other software
packages. You can set some UNIX semaphore parameters.

Each instance of the database server requires the following semaphore sets:
v One set for each group of up to 100 virtual processors (VPs) that are started with
the database server
v One set for each additional VP that you might add dynamically while the
database server is running
v One set for each group of 100 or fewer user sessions connected through the
shared-memory communication interface

Tip: For best performance, allocate enough semaphores for double the number of
ipcshm connections that you expect. Use the NETTYPE configuration parameter to
configure database server poll threads for this doubled number of connections.

Because utilities such as onmode use shared-memory connections, you must

configure a minimum of two semaphore sets for each instance of the database
server: one for the initial set of VPs and one for the shared-memory connections
that database server utilities use. The SEMMNI operating-system configuration
parameter typically specifies the number of semaphore sets to allocate. For
information about how to set semaphore-related parameters, see the configuration
instructions for your operating system.

© Copyright IBM Corp. 1996, 2014 3-1

 The SEMMSL operating-system configuration parameter typically specifies the
maximum number of semaphores per set. Set this parameter to at least 100.

Some operating systems require that you configure a maximum total number of
semaphores across all sets, which the SEMMNS operating-system configuration
parameter typically specifies. Use the following formula to calculate the total
number of semaphores that each instance of the database server requires:
SEMMNS = init_vps + added_vps + (2 * shmem_users) + concurrent_utils
init_vps
is the number of virtual processors (VPs) that are started with the database
server. This number includes CPU, PIO, LIO, AIO, SHM, TLI, SOC, and
ADM VPs. The minimum value is 15.
added_vps
is the number of VPs that you intend to add dynamically.
shmem_users
is the number of shared-memory connections that you allow for this
instance of the database server.
concurrent_utils
is the number of concurrent database server utilities that can connect to
this instance. It is suggested that you allow for a minimum of six utility
connections: two for ON-Archive or ON-Bar and four for other utilities
such as onstat, and oncheck.

If you use software packages that require semaphores, the SEMMNI configuration
parameter must include the total number of semaphore sets that the database
server and your other software packages require. You must set the SEMMSL
configuration parameter to the largest number of semaphores per set that any of
your software packages require. For systems that require the SEMMNS
configuration parameter, multiply SEMMNI by the value of SEMMSL to calculate
an acceptable value.
Related concepts:
“Configuring poll threads” on page 3-13

UNIX file-descriptor parameters

Some operating systems require you to specify a limit on the number of file
descriptors that a process can have open at any one time. To specify this limit, use
an operating-system configuration parameter, typically NOFILE, NOFILES, NFILE,
or NFILES.

The number of open file descriptors that each instance of the database server needs
depends on the number of chunks in your database, the number of VPs that you
run, and the number of network connections that your database server instance
must support.

Use the following formula to calculate the number of file descriptors that your
instance of the database server requires:
NFILES = (chunks * NUMAIOVPS) + NUMBER_of_CPU_VPS + net_connections
chunks is the number of chunks to be configured.
net_connections
is the number of network connections that you specify in either of the
following places:

3-2 IBM Informix Performance Guide

 v sqlhosts file
v NETTYPE configuration entries
Network connections include all but those specified as the ipcshm
connection type.

Each open file descriptor is about the same length as an integer within the kernel.
Allocating extra file descriptors is an inexpensive way to allow for growth in the
number of chunks or connections on your system.

UNIX memory configuration parameters

The configuration of memory in the operating system can affect other resources,
including CPU and I/O.

Insufficient physical memory for the overall system load can lead to thrashing, as
“Memory utilization” on page 1-10 describes. Insufficient memory for the database
server can result in excessive buffer-management activity. For more information
about configuring memory, see “Configuring UNIX shared memory” on page 4-6.

Windows configuration parameters that affect CPU utilization

The Informix distribution includes a machine notes file that contains recommended
values for Informix configuration parameters on Windows. Compare the values in
this file with your current ONCONFIG configuration file settings.

Informix runs in the background. For best performance, give the same priority to
foreground and background applications.

On Windows, to change the priorities of foreground and background applications,

go to Start > Settings > Control Panel, open the System icon, and click the
Advanced Tab. Select the Performance Options button and select either the
Applications or Background Services radio button.

The configuration of memory in the operating system can impact other resources,
including CPU and I/O. Insufficient physical memory for the overall system load
can lead to thrashing, as “Memory utilization” on page 1-10 describes. Insufficient
memory for Informix can result in excessive buffer-management activity. When you
set the Virtual Memory values in the System icon on the Control Panel, ensure
that you have enough paging space for the total amount of physical memory.

Configuration parameters and environment variables that affect CPU

utilization
Some configuration parameters and environment variables affect CPU utilization.
You might need to adjust the settings of these parameters and variables when you
consider methods of improving performance.

The following configuration parameters in the database server configuration file

have a significant impact on CPU utilization:
v DS_MAX_QUERIES
v DS_MAX_SCANS
v FASTPOLL
v MAX_PDQPRIORITY
v MULTIPROCESSOR

Chapter 3. Effect of configuration on CPU utilization 3-3

 v NETTYPE
v OPTCOMPIND
v SINGLE_CPU_VP
v VPCLASS
v VP_MEMORY_CACHE_KB

The following environment variables affect CPU utilization:

v OPTCOMPIND
v PDQPRIORITY
v PSORT_NPROCS

The OPTCOMPIND environment variable, when set in the environment of a client

application, indicates the preferred way to perform join operations. This variable
overrides the value that the OPTCOMPIND configuration parameter sets. For
details on how to select a preferred join method, see “Optimizing access methods”
on page 3-10.

The PDQPRIORITY environment variable, when set in the environment of a client

application, places a limit on the percentage of CPU VP utilization, shared memory,
and other resources that can be allocated to any query that the client starts.

A client can also use the SET PDQPRIORITY statement in SQL to set a value for
PDQ priority. The actual percentage allocated to any query is subject to the factor
that the MAX_PDQPRIORITY configuration parameter sets. For more information
about how to limit resources that can be allocated to a query, see “Limiting PDQ
resources in queries” on page 3-11.

PSORT_NPROCS, when set in the environment of a client application, indicates

the number of parallel sort threads that the application can use. The database
server imposes an upper limit of 10 sort threads per query for any application. For
more information about parallel sorts and PSORT_NPROCS, see “Configure
dbspaces for temporary tables and sort files” on page 5-8.
Related concepts:
Environment variables (SQL Reference)
Related reference:
Database configuration parameters (Administrator's Reference)

Specifying virtual processor class information

Use the VPCLASS configuration parameter to specify a class of virtual processors,
the number of virtual processors that the database server should start for a specific
class, and the maximum number allowed.

To execute user-defined routines (UDRs), you can define a new class of virtual
processors to isolate UDR execution from other transactions that execute on the
CPU virtual processors. Typically you write user-defined routines to support
user-defined data types.

If you do not want a user-defined routine to affect the normal processing of user
queries in the CPU class, you can use the CREATE FUNCTION statement to assign
the routine to a user-defined class of virtual processors. The class name that you
specify in the VPCLASS configuration parameter must match the name specified in
the CLASS modifier of the CREATE FUNCTION statement.

3-4 IBM Informix Performance Guide

For guidelines, on using the cpu and num options of the VPCLASS configuration
parameter, see “Setting the number of CPU VPs.”
Related reference:
VPCLASS configuration parameter (Administrator's Reference)
CREATE FUNCTION statement (SQL Syntax)

Setting the number of CPU VPs

You can configure the number of CPU virtual processors (VPs) that the database
server uses. Do not allocate more CPU VPs than there are CPU processors
available to service them.

When the database server starts, the number of CPU VPs is automatically
increased to half the number of CPU processors on the database server computer,
unless the SINGLE_CPU_VP configuration parameter is enabled. However, you
might want to change the number of CPU VPs based on your performance needs.

Use the following guidelines to set the number of CPU VPs:

Uniprocessor computers
For uniprocessor computers, specify one CPU VP:
VPCLASS cpu,num=1
Dual-processor computers
For dual-processor systems, you might improve performance by running
with two CPU VPs. To test if performance improves, set the num field of
the VPCLASS configuration parameter to 1 in the onconfig file and then
add a CPU VP dynamically at run time by running the onmode -p
command.
Multiprocessor computers that are primarily database servers
For multiprocessor systems with four or more CPUs that are primarily
used as database servers, set the num option of the VPCLASS
configuration parameter in the onconfig file to one less than the total
number of processors. For example, if you have four CPUs, use the
following specification:
VPCLASS cpu,num=3

When you use this setting, one processor is available to run the database
server utilities or the client application.
Multiprocessor computers that are not primarily database servers
For multiprocessor systems that you do not use primarily to support
database servers, you can start with somewhat fewer CPU VPs to allow for
other activities on the system and then gradually add more if necessary.
Multi-core or hardware multithreading computers with logical CPUs
For multiprocessor systems that use multi-core processors or hardware
multithreading to support more logical CPUs than physical processors, you
can assign the number of CPU VPs according to the number of logical
CPU VPs available for that purpose. The amount of processing that an
additional logical CPU can provide might be only a fraction of what a
dedicated physical processor can support.
On systems, where multi-core processors are installed, the optimal
configuration in most cases is the same as for systems with a number of

Chapter 3. Effect of configuration on CPU utilization 3-5

 individual processors equal to the total number of cores. Setting the
number of CPU VPs to N-1, where N is number of cores is close to optimal
for CPU-intensive workloads.
On computers where the CPU uses multiple threads per core, operating
systems show more logical processors than actual processing cores. To take
advantage of more CPU threads, the database server must be configured
with the number of CPU VPs in the range between N and M, where N is
number of cores and M is total number of logical CPUs reported by
system. The number of CPU VPs where optimal performance is achieved
depends on the workload.
When increasing the number of CPU VPs to use more threads per core, the
expected gain in performance is only a fraction of what dedicated physical
processor or core can provide.
If you are migrating Informix from multi-CPU/multicore systems to
systems with multiple threads per core, take special care in regard to
processor affinity. When binding Informix CPU VPs to the logical
processors of the operating system, you must be aware of the architecture
for the CPU. If you are not sure, do not use the CPU affinity so that the
operating system schedules CPU VPs to logical processors with available
resources. Using affinity without understanding the relationship between
the logical CPUs and processing cores can result in severe performance
degradation.
For example, to bind each of 8 configured CPU VPs to a separate core on
an 8-core system with two threads per core (16 logical CPUs), use the
following setting:
VPCLASS cpu,num=8,aff=(0-14/2)
Related concepts:
“Automatic addition of CPU virtual processors” on page 3-20
Related reference:
VPCLASS configuration parameter (Administrator's Reference)

Disabling process priority aging for CPU VPs

Use the noage option of the VPCLASS configuration parameter to disable process
priority aging for database server CPU VPs on operating systems that support this
feature. Priority aging occurs when the operating system lowers the priority of
long-running processes as they accumulate processing time. You might want to
disable priority aging because it can cause the performance of the database server
processes to decline over time.

Your database server distribution includes a machine notes file that contains
information about whether your version of the database server supports this
feature. For information about where to find this machine notes file, see the
Introduction to this guide.

Specify the noage option of VPCLASS if your operating system supports this
feature.
Related reference:
VPCLASS configuration parameter (Administrator's Reference)

3-6 IBM Informix Performance Guide

Specifying processor affinity
Use the aff option of the VPCLASS parameter to specify the processors to which
you want to bind CPU VPs or AIO VPs. When you assign a CPU VP to a specific
CPU, the VP runs only on that CPU. However, other processes can also run on that
CPU.

The database server supports automatic binding of CPU VPs to processors on

multiprocessor host computers that support processor affinity. Your database server
distribution includes a machine notes file that contains information about whether
your version of the database server supports this feature.

You can use processor affinity for the purposes that the following sections describe.
Related reference:
VPCLASS configuration parameter (Administrator's Reference)

Distributing computation impact:

You can use processor affinity to distribute the computation impact of CPU virtual
processors (VPs) and other processes. On computers that are dedicated to the
database server, assigning CPU VPs to all but one of the CPUs achieves maximum
CPU utilization.

On computers that support both database server and client applications, you can
bind applications to certain CPUs through the operating system. By doing so, you
effectively reserve the remaining CPUs for use by database server CPU VPs, which
you bind to the remaining CPUs with the VPCLASS configuration parameter. Set
the aff option of the VPCLASS configuration parameter to the numbers of the
CPUs on which to bind CPU VPs. For example, the following VPCLASS setting
assigns CPU VPs to processors 4 to 7:
VPCLASS cpu,num=4,aff=(4-7)

When specifying a range of processors, you can also specify an incremental value
with the range that indicates which CPUs in the range should be assigned to the
virtual processors. For example, you can specify that the virtual processors are
assigned to every other CPU in the range 0-6, starting with CPU 0.
VPCLASS CPU,num=4,aff=(0-6/2)

The virtual processors are assigned to CPUs 0, 2, 4, 6.

If you specify VPCLASS CPU,num=4,aff=(1-10/3), the virtual processors are assigned

to every third CPU in the range 1-10, starting with CPU 1. The virtual processors
are assigned to CPUs 1, 4, 7, 10.

When you specify more than one value or range, the values and ranges do not
have to be incremental or in any particular order. For example you can specify
aff=(8,12,7-9,0-6/2).

The database server assigns CPU virtual processors to CPUs in a circular pattern,
starting with the first processor number that you specify in the aff option. If you
specify a larger number of CPU virtual processors than physical CPUs, the
database server continues to assign CPU virtual processors starting with the first
CPU. For example, suppose you specify the following VPCLASS settings:
VPCLASS cpu,num=8,aff=(4-7)

The database server makes the following assignments:

Chapter 3. Effect of configuration on CPU utilization 3-7
 v CPU virtual processor number 0 to CPU 4
v CPU virtual processor number 1 to CPU 5
v CPU virtual processor number 2 to CPU 6
v CPU virtual processor number 3 to CPU 7
v CPU virtual processor number 4 to CPU 4
v CPU virtual processor number 5 to CPU 5
v CPU virtual processor number 6 to CPU 6
v CPU virtual processor number 7 to CPU 7
Related reference:
VPCLASS configuration parameter (Administrator's Reference)

Isolating AIO VPs from CPU VPs:

On a system that runs database server and client (or other) applications, you can
bind asynchronous I/O (AIO) VPs to the same CPUs to which you bind other
application processes through the operating system. In this way, you isolate client
applications and database I/O operations from the CPU VPs.

This isolation can be especially helpful when client processes are used for data
entry or other operations that require waiting for user input. Because AIO VP
activity usually comes in quick bursts followed by idle periods waiting for the
disk, you can often interweave client and I/O operations without their unduly
impacting each other.

Binding a CPU VP to a processor does not prevent other processes from running
on that processor. Application (or other) processes that you do not bind to a CPU
are free to run on any available processor. On a computer that is dedicated to the
database server, you can leave AIO VPs free to run on any processor, which
reduces delays on database operations that are waiting for I/O. Increasing the
priority of AIO VPs can further improve performance by ensuring that data is
processed quickly once it arrives from disk.

Avoiding a certain CPU:

The database server assigns CPU VPs to CPUs serially, starting with the CPU
number you specify in this parameter. You might want to avoid assigning CPU
VPs to a certain CPU that has a specialized hardware or operating-system function
(such as interrupt handling).

Setting the number of AIO VPs

Use the aio and num options of the VPCLASS configuration parameter to indicate
the number of AIO virtual processors that the database server starts initially.

If your operating system does not support kernel asynchronous I/O (KAIO), the
database server uses AIO virtual processors (VPs) to manage all database I/O
requests.

If the VPCLASS configuration parameter does not specify the number of AIO VPs
to start in the onconfig file, then the setting of the AUTO_AIOVPS configuration
parameter controls the number of AIO VPs:
v If AUTO_AIOVPS is set to 1 (on), the number of AIO VPs initially started is
equal to the number of AIO chunks, up to a maximum of 128.

3-8 IBM Informix Performance Guide

v If AUTO_AIOVPS is set to 0 (off), the number of AIO VPs started is equal to the
greater of 6 or twice the number of AIO chunks, up to a maximum of 128.

The recommended number of AIO virtual processors depends on how many disks
your configuration supports. If KAIO is not implemented on your platform, you
should allocate one AIO virtual processor for each disk that contains database
tables. You can add an additional AIO virtual processor for each chunk that the
database server accesses frequently.

You can use the AUTO_AIOVPS configuration parameter to enable the database
server to automatically increase the number of AIO virtual processors and
page-cleaner threads when the server detects that AIO virtual processors are not
keeping up with the I/O workload.

The machine notes file for your version of the database server indicates whether
the operating system supports KAIO. If KAIO is supported, the machine notes
describe how to enable KAIO on your specific operating system.

If your operating system supports KAIO, the CPU VPs make asynchronous I/O
requests to the operating system instead of AIO virtual processors. In this case,
configure only one AIO virtual processor, plus two additional AIO virtual
processor for every file chunk that does not use KAIO.

If you use cooked files and if you enable direct I/O using the DIRECT_IO
configuration parameter, you can reduce the number of AIO virtual processors. If
the database server implements KAIO and if direct I/O is enabled, the database
server will attempt to use KAIO, so you probably do not need more than one AIO
virtual processor. Temporary dbspaces do not use direct I/O. If you have
temporary dbspaces, you will probably need more than one AIO virtual processors.

Even when direct I/O is enabled with the DIRECT_IO configuration parameter, if
the file system does not support either direct I/O or KAIO, you still must allocate
two additional AIO virtual processors for every active dbspace chunk that is not
using KAIO.

The goal in allocating AIO virtual processors is to allocate enough of them so that
the lengths of the I/O request queues are kept short (that is, the queues have as
few I/O requests in them as possible). When the I/O request queues remain
consistently short, I/O requests are processed as fast as they occur. Use the onstat
-g ioq command to monitor the length of the I/O queues for the AIO virtual
processors.

Allocate enough AIO VPs to accommodate the peak number of I/O requests.
Generally, allocating a few extra AIO VPs is not detrimental. To start additional
AIO VPs while the database server is in online mode, use the onmode -p
command. You cannot drop AIO VPs in online mode.
Related reference:
AUTO_AIOVPS configuration parameter (Administrator's Reference)
VPCLASS configuration parameter (Administrator's Reference)

Chapter 3. Effect of configuration on CPU utilization 3-9

 Setting the MULTIPROCESSOR configuration parameter when
using multiple CPU VPs
If you are running multiple CPU VPs, set the MULTIPROCESSOR configuration
parameter to 1. When you set MULTIPROCESSOR to 1, the database server
performs locking in a manner that is appropriate for a multiprocessor. Otherwise,
set this parameter to 0.

The number of CPU VPs is used as a factor in determining the number of scan
threads for a query. Queries perform best when the number of scan threads is a
multiple (or factor) of the number of CPU VPs. Adding or removing a CPU VP can
improve performance for a large query because it produces an equal distribution of
scan threads among CPU VPs. For instance, if you have 6 CPU VPs and scan 10
table fragments, you might see a faster response time if you reduce the number of
CPU VPs to 5, which divides evenly into 10. You can use onstat -g ath to monitor
the number of scan threads per CPU VP or use onstat -g ses to focus on a
particular session.
Related reference:
MULTIPROCESSOR configuration parameter (Administrator's Reference)

Setting the SINGLE_CPU_VP configuration parameter when

using one CPU VP
If you are running only one CPU VP, set the SINGLE_CPU_VP configuration
parameter to 1. Otherwise, set this parameter to 0.

Important: If you set the SINGLE_CPU_VP parameter to 1, the value of the num
option of the VPCLASS configuration parameter must also be 1.

Note: The database server treats user-defined virtual-processor classes (that is, VPs
defined with VPCLASS) as if they were CPU VPs. Thus, if you set
SINGLE_CPU_VP to nonzero, you cannot create any user-defined classes.

When you set the SINGLE_CPU_VP parameter to 1, you cannot add CPU VPs
while the database server is in online mode.
Related reference:
SINGLE_CPU_VP configuration parameter (Administrator's Reference)
VPCLASS configuration parameter (Administrator's Reference)

Optimizing access methods

The OPTCOMPIND configuration parameter helps the query optimizer choose an
appropriate access method for your application. When the optimizer examines join
plans, OPTCOMPIND indicates the preferred method for performing the join
operation for an ordered pair of tables.

If OPTCOMPIND is equal to 0, the optimizer gives preference to an existing index

(nested-loop join) even when a table scan might be faster. If OPTCOMPIND is set
to 1 and the isolation level for a given query is set to Repeatable Read, the
optimizer uses nested-loop joins.

When OPTCOMPIND is equal to 2, the optimizer selects a join method based on

cost alone even though table scans can temporarily lock an entire table. For more
information about OPTCOMPIND and the different join methods, see “Effect of
OPTCOMPIND on the query plan” on page 10-22.
3-10 IBM Informix Performance Guide
 To set the value for OPTCOMPIND for specific applications or user sessions, set
the OPTCOMPIND environment variable for those sessions. Values for this
environment variable have the same range and semantics as for the configuration
parameter.
Related reference:
OPTCOMPIND configuration parameter (Administrator's Reference)

Setting the value of OPTCOMPIND within a session

You can set or change the value of OPTCOMPIND within a session for different
kinds of queries. To do this, use the SET ENVIRONMENT OPTCOMPIND
command, not the OPTCOMPIND configuration parameter.

For a DSS query, you should set the value of OPTCOMPIND to 2 or 1, and you
should be sure that the isolation level is not set to Repeatable Read. For an OLTP
query, you could set the value to 0 or 1 with the isolation level not set to
Repeatable Read.

The value that you enter using the SET ENVIRONMENT OPTCOMPIND
command takes precedence over the default setting specified in the ONCONFIG
file. The default OPTCOMPIND setting is restored when the current session
terminates. No other user sessions are affected by SET ENVIRONMENT
OPTCOMPIND statements that you execute.
Related concepts:
OPTCOMPIND Environment Option (SQL Syntax)

Limiting PDQ resources in queries

The MAX_PDQPRIORITY configuration parameter limits the percentage of parallel
database query (PDQ) resources that a query can use. Use MAX_PDQPRIORITY to
limit the impact of large CPU-intensive queries on transaction throughput.

To limit the impact of large CPU-intensive queries on transaction throughput

Set the value of the MAX_PDQPRIORITY configuration parameter to an integer

that represents a percentage of the following PDQ resources that a query can
request:
v Memory
v CPU VPs
v Disk I/O
v Scan threads

When a query requests a percentage of PDQ resources, the database server

allocates the MAX_PDQPRIORITY percentage of the amount requested, as the
following formula shows:
Resources allocated = PDQPRIORITY/100 * MAX_PDQPRIORITY/100

For example, if a client uses the SET PDQPRIORITY 80 statement to request 80

percent of PDQ resources, but MAX_PDQPRIORITY is set to 50, the database
server allocates only 40 percent of the resources (50 percent of the request) to the
client.

For decision support and online transaction processing (OLTP), setting

MAX_PDQPRIORITY allows the database server administrator to control the
impact that individual decision-support queries have on concurrent OLTP

Chapter 3. Effect of configuration on CPU utilization 3-11

 performance. Reduce the value of MAX_PDQPRIORITY when you want to allocate
more resources to OLTP processing. Increase the value of MAX_PDQPRIORITY
when you want to allocate more resources to decision-support processing.

For more information about how to control the use of PDQ resources, see “The
allocation of resources for parallel database queries” on page 12-7.
Related reference:
MAX_PDQPRIORITY configuration parameter (Administrator's Reference)

Limiting the performance impact of CPU-intensive queries

The DS_MAX_QUERIES configuration parameter specifies a maximum number of
decision-support queries that can run at any one time. Queries with a low PDQ
priority use proportionally fewer resources, so a larger number of those queries
can run simultaneously. You can use the DS_MAX_QUERIES configuration
parameter to limit the performance impact of CPU-intensive queries.

The DS_MAX_QUERIES configuration parameter controls only queries with a PDQ

priority that is nonzero.

The database server uses the value of DS_MAX_QUERIES with

DS_TOTAL_MEMORY to calculate quantum units of memory to allocate to a
query. For more information about how the database server allocates memory to
queries, see “The DS_TOTAL_MEMORY configuration parameter and memory
utilization” on page 4-12.
Related concepts:
“The DS_TOTAL_MEMORY configuration parameter and memory utilization” on
page 4-12
Related reference:
DS_MAX_QUERIES configuration parameter (Administrator's Reference)

Limiting the number of PDQ scan threads that can run

concurrently
The DS_MAX_SCANS configuration parameter limits the number of PDQ scan
threads that can run concurrently. This configuration parameter prevents the
database server from being flooded with scan threads from multiple
decision-support queries.

To calculate the number of scan threads allocated to a query, use the following
formula:
scan_threads = min (nfrags, (DS_MAX_SCANS * pdqpriority / 100
* MAX_PDQPRIORITY / 100) )
nfrags is the number of fragments in the table with the largest number of
fragments.
pdqpriority
is the PDQ priority value set by either the PDQPRIORITY environment
variable or the SQL statement SET PDQPRIORITY.

Reducing the number of scan threads can reduce the time that a large query waits
in the ready queue, particularly when many large queries are submitted
concurrently. However, if the number of scan threads is less than nfrags, the query
takes longer once it is underway.

3-12 IBM Informix Performance Guide

 For example, if a query needs to scan 20 fragments in a table, but the scan_threads
formula lets the query begin when only 10 scan threads are available, each scan
thread scans two fragments serially. Query execution takes approximately twice as
long as if 20 scan threads were used.
Related reference:
DS_MAX_SCANS configuration parameter (Administrator's Reference)

Configuring poll threads

The NETTYPE configuration parameter configures poll threads for each connection
type that your instance of the database server supports. If your database server
instance supports connections over more than one interface or protocol, you must
specify a separate NETTYPE configuration parameter for each connection type.

You typically include a separate NETTYPE parameter for each connection type that
is associated with a dbservername. You list dbservernames in the
DBSERVERNAME and DBSERVERALIASES configuration parameters. You
associate connection types with dbservernames in the sqlhosts information. For
details about connection types and the sqlhosts information, see connectivity
information in your.IBM Informix Administrator's Guide.
Related reference:
“UNIX semaphore parameters” on page 3-1
NETTYPE configuration parameter (Administrator's Reference)

Specifying the connection protocol

The first NETTYPE entry, which specifies the protocol for a given connection type,
applies to all dbservernames associated with that type. Subsequent NETTYPE
entries for that connection type are ignored.

NETTYPE entries are required for connection types that are used for outgoing
communication only even if those connection types are not listed in the sqlhosts
information.
UNIX Only
The following protocols apply to UNIX platforms:
v IPCSHM
v TLITCP
v IPCSTR
v SOCTCP
v TLIIMC
v SOCIMC
v SQLMUX
v SOCSSL
Windows Only
The following protocols apply to Windows platforms:
v SOCTCP
v IPCNMP
v SQLMUX
v SOCSSL
Related reference:

Chapter 3. Effect of configuration on CPU utilization 3-13

 NETTYPE configuration parameter (Administrator's Reference)

Specifying virtual-processor classes for poll threads

Each poll thread configured or added dynamically by a NETTYPE entry runs in a
separate VP. A poll thread can run in one of two VP classes: NET and CPU. For
best performance, use a NETTYPE entry to assign only one poll thread to the CPU
VP class. Then assign all additional poll threads to NET VPs.
Related reference:
NETTYPE configuration parameter (Administrator's Reference)

Specifying the number of connections and poll threads

The optimum number of connections per poll thread is approximately 300 for
uniprocessor computers and up to 350 for multiprocessor computers, although this
can vary depending on the platform and database server workload.

A poll thread can support 1024 or perhaps more connections. If the FASTPOLL
configuration parameter is enabled, you might be able to configure fewer poll
threads, but should test the performance to determine the optimal configuration
for your environment.

Each NETTYPE entry configures the number of poll threads for a specific
connection type, the number of connections per poll thread, and the
virtual-processor class in which those poll threads run, using the following
comma-separated fields. There can be no white space within or between these
fields.
NETTYPE connection_type,poll_threads,conn_per_thread,vp_class
connection_type
identifies the protocol-interface combination to which the poll threads are
assigned. You typically set this field to match the connection_type field of a
dbservername entry that is in the sqlhosts information.
poll_threads
is the number of poll threads assigned to the connection type.
conn_per_thread
is the number of connections per poll thread. Use the following formula to
calculate this number:
conn_per_thread = connections / poll_threads
connections
is the maximum number of connections that you expect the
indicated connection type to support. For shared-memory
connections (ipcshm), double the number of connections for best
performance.
This field is used only for shared memory connections on
Windows. Other connection methods on Windows ignore this
value.
For shared memory connection, the value of conn_per_thread is the
maximum number of connections per thread. For network
connections, the value of conn_per_thread can be exceeded.
vp_class
is the class of virtual processor that can run the poll threads. Specify CPU
if you have a single poll thread that runs on a CPU VP. For best
performance, specify NET if you require more than one poll thread. If you

3-14 IBM Informix Performance Guide

 are running Windows, specify NET in all cases. The default value for this
field depends on the following conditions:
v If the connection type is associated with the dbservername that is listed
in the DBSERVERNAME parameter, and no previous NETTYPE
parameter specifies CPU explicitly, the default VP class is CPU. If the
CPU class is already taken, the default is NET.
v If the connection type is associated with a dbservername that the
DBSERVERALIASES parameter specifies, the default VP class is NET.

If the value of conn_per_thread exceeds 350 and the number of poll threads for the
current connection type is less than the number of CPU VPs, you can improve
performance by specifying the NET CPU class, adding poll threads (do not exceed
the number of CPU VPs), and recalculating the value of conn_per_thread. The
default value for conn_per_thread is 50.

Important: Each ipcshm connection requires a semaphore. Some operating systems

require that you configure a maximum number of semaphores that can be
requested by all software packages that run on the computer. For best
performance, double the number of actual ipcshm connections when you allocate
semaphores for shared-memory communications. See “UNIX semaphore
parameters” on page 3-1.

If your computer is a uniprocessor and your database server instance is configured

for only one connection type, you can omit the NETTYPE parameter. The database
server uses the information provided in the sqlhosts information to establish
client/server connections.

If your computer is a uniprocessor and your database server instance is configured

for more than one connection type, include a separate NETTYPE entry for each
connection type. If the number of connections of any one type significantly exceeds
300, assign two or more poll threads, up to a maximum of the number of CPU
VPs, and specify the NET VP class, as the following example shows:
NETTYPE ipcshm,1,200,CPU
NETTYPE tlitcp,2,200,NET # supports 400 connections

For ipcshm, the number of poll threads correspond to the number of memory
segments. For example, if NETTYPE is set to 3,100 and you want one poll thread,
set the poll thread to 1,300.

If your computer is a multiprocessor, your database server instance is configured

for only one connection type, and the number of connections does not exceed 350,
you can use NETTYPE to specify a single poll thread on either the CPU or the
NET VP class. If the number of connections exceeds 350, set the VP class to NET,
increase the number of poll threads, and recalculate conn_per_thread.

Important: You should carefully distinguish between poll threads for network
connections and poll threads for shared memory connections, which should run
one per CPU virtual processor. TCP connections should only be in network virtual
processors, and you should only have the minimum needed to maintain
responsiveness. Shared memory connections should only be in CPU virtual
processors and should run in every CPU virtual processor.
Related concepts:
“Improve connection performance and scalability” on page 3-16
Related reference:

Chapter 3. Effect of configuration on CPU utilization 3-15

 NETTYPE configuration parameter (Administrator's Reference)
VPCLASS configuration parameter (Administrator's Reference)

Improve connection performance and scalability

You can improve connection performance and scalability by specifying information
in the NUMFDSERVERS and NS_CACHE configuration parameters and by using
multiple listen threads.

Informix SQL sessions can migrate across CPU VPs. You can improve the
performance and scalability of network connections on UNIX by using the
NUMFDSERVERS configuration parameter to specify a number for the poll threads
to use when distributing a TCP/IP connection across VPs. Specifying
NUMFDSERVERS information is useful if the database server has a high rate of
new connect and disconnect requests or if you find a high amount of contention
between network shared file (NSF) locks.

You should also review and, if necessary, change the information in the NETTYPE
configuration parameter, which defines the number of poll threads for a specific
connection type, the number of connections per poll thread, and the
virtual-processor class in which those poll threads run. You specify NETTYPE
configuration parameter information as follows:
NETTYPE connection_type,poll_threads,conn_per_thread,vp_class

On UNIX, if vp_class is NET, poll_threads can be a value that is greater than or

equal to 1. If vp_class is CPU, the number of poll_threads can be 1 through the
number of CPU VPs. On Windows, poll_threads can be value that is greater than
or equal to 1.

For example, suppose you specify 8 poll threads in the NETTYPE configuration
parameter, as follows:
NETTYPE soctcp,8,300,NET

You can also specify 8 in the NUMFDSERVERS configuration parameter to enable

the server to use all 8 poll thread to handle network connections migrating
between VPs.

You can use the NS_CACHE configuration parameter to define the maximum
retention time for an individual entry in the host name/IP address cache, the
service cache, the user cache, and the group cache. The server can get information
from the cache faster than it does when querying the operating system.

You can improve service for connection requests by using multiple listen threads.
When you specify DBSERVERNAME and DBSERVERALIASES configuration
parameter information for onimcsoc or onsoctcp protocols, you can specify the
number of multiple listen threads for the database server aliases in your sqlhosts
information. The default value of number is 1.

The DBSERVERNAME and DBSERVERALIASES configuration parameters define

database server names (dbservernames) that have corresponding entries in the
sqlhosts information. Each dbservername parameter in the sqlhosts information has
a nettype entry that specifies an interface/protocol combination. The database
server runs one or more poll threads for each unique nettype entry.

You can use the onstat -g ath command to display information about all threads.
Related concepts:
3-16 IBM Informix Performance Guide
 Name service maximum retention time set in the NS_CACHE configuration
parameter (Administrator's Guide)
“Specifying the number of connections and poll threads” on page 3-14
“Monitor threads with onstat -g ath output” on page 13-49
Related reference:
NETTYPE configuration parameter (Administrator's Reference)
NUMFDSERVERS configuration parameter (Administrator's Reference)
NS_CACHE configuration parameter (Administrator's Reference)
DBSERVERNAME configuration parameter (Administrator's Reference)
DBSERVERALIASES configuration parameter (Administrator's Reference)
Multiple listen threads (Administrator's Guide)

Enabling fast polling

You can use the FASTPOLL configuration parameter to enable or disable fast
polling of your network, if your operating-system platform supports fast polling.
Fast polling is beneficial if you have a large number of connections.

For example, if you have more than 300 concurrent connections with the database
server, you can enable the FASTPOLL configuration parameter for better
performance.
Related reference:
FASTPOLL configuration parameter (Administrator's Reference)

Network buffer pools

The sizes of buffers for TCP/IP connections affect memory and CPU utilization.
Sizing these buffers to accommodate a typical request can improve CPU utilization
by eliminating the need to break up requests into multiple messages.

However, you must use this capability with care; the database server dynamically
allocates buffers of the indicated sizes for active connections. Unless you carefully
size buffers, they can use large amounts of memory. For details on how to size
network buffers, see “Network buffer size” on page 3-19.

The database server dynamically allocates network buffers from the global memory
pool for request messages from clients. After the database server processes client
requests, it returns buffers to a common network buffer pool that is shared among
sessions that use SOCTCP, IPCSTR, or TLITCP network connections.

This common network buffer pool provides the following advantages:

v Prevents frequent allocations and deallocations from the global memory pool
v Uses fewer CPU resources to allocate and deallocate network buffers to and
from the common network buffer pool for each network transfer
v Reduces contention for allocation and deallocation of shared memory

The free network buffer pool can grow during peak activity periods. To prevent
large amounts of unused memory from remaining in these network buffer pools
when network activity is no longer high, the database server returns free buffers
when the number of free buffers reaches specific thresholds.

Chapter 3. Effect of configuration on CPU utilization 3-17

 The database server provides the following features to further reduce the allocation
and deallocation of and contention for the free network buffers:
v A private free network buffer pool for each session to prevent frequent
allocations and deallocations of network buffers from the common network
buffer pool or from the global memory pool in shared memory
v Capability to specify a larger than 4-kilobyte buffer size to receive network
packets or messages from clients

As the system administrator, you can control the free buffer thresholds and the size
of each buffer with the following methods:
v NETTYPE configuration parameter
v IFX_NETBUF_PVTPOOL_SIZE environment variable
v IFX_NETBUF_SIZE environment variable and b (client buffer size) option in the
sqlhosts information

Network buffers
The database server implements a threshold of free network buffers to prevent
frequent allocations and deallocations of shared memory for the network buffer
pool. This threshold enables the database server to correlate the number of free
network buffers with the number of connections that you specify in the NETTYPE
configuration parameter.

The database server dynamically allocates network buffers for request messages
from clients. After the database server processes client requests, it returns buffers
to the network free-buffer pool.

If the number of free buffers is greater than the threshold, the database server
returns the memory allocated to buffers over the threshold to the global pool.

The database server uses the following formula to calculate the threshold for the
free buffers in the network buffer pool:
free network buffers threshold =
100 + (0.7 * number_connections)

The value for number_connections is the total number of connections that you
specified in the third field of the NETTYPE entry for the different type of network
connections (SOCTCP, IPCSTR, or TLITCP). This formula does not use the
NETTYPE entry for shared memory (IPCSHM).

If you do not specify a value in the third field of the NETTYPE parameter, the
database server uses the default value of 50 connections for each NETTYPE entry
corresponding to the SOCTCP, TLITCP, and IPCSTR protocols.

Support for private network buffers

The database server provides support for private network buffers for each session
that uses SOCTCP, IPCSTR, or TLITCP network connections.

For situations in which many connections and sessions are constantly active, these
private network buffers have the following advantages:
v Less contention for the common network buffer pool
v Fewer CPU resources to allocate and deallocate network buffers to and from the
common network buffer pool for each network transfer

3-18 IBM Informix Performance Guide

 The IFX_NETBUF_PVTPOOL_SIZE environment variable specifies the size of the
private network buffer pool for each session. The default size is one buffer.

Use the onstat utility commands in the following table to monitor the network
buffer usage.

Command Output Field Description

onstat -g ntu q-pvt The current number and highest number of
buffers that are free in the private pool for this
session
onstat -g ntm q-exceeds The number of times that the free buffer
threshold was exceeded

The onstat -g ntu command displays the following format for the q-pvt output
field:
current number / highest number

If the number of free buffers (value in q-pvt field) is consistently 0, you can
perform one of the following actions:
v Increase the number of buffers with the environment variable
IFX_NETBUF_PVTPOOL_SIZE.
v Increase the size of each buffer with the environment variable
IFX_NETBUF_SIZE.

The q-exceeds field indicates the number of times that the threshold for the shared
network free-buffer pool was exceeded. When this threshold is exceeded, the
database server returns the unused network buffers (over this threshold) to the
global memory pool in shared memory. Optimally, this value should be 0 or a low
number so that the server is not allocating or deallocating network buffers from
the global memory pool.
Related reference:
IFX_NETBUF_PVTPOOL_SIZE environment variable (UNIX) (SQL Reference)
IFX_NETBUF_SIZE environment variable (SQL Reference)

Network buffer size

The IFX_NETBUF_SIZE environment variable specifies the size of each network
buffer in the common network buffer pool and the private network buffer pool.

The default buffer size is 4 kilobytes.

The IFX_NETBUF_SIZE environment variable allows the database server to

receive messages longer than 4 kilobytes in one system call. The larger buffer size
reduces the amount of overhead required to receive each packet.

Increase the value of IFX_NETBUF_SIZE if you know that clients send greater
than 4-kilobyte packets. Clients send large packets during any of the following
situations:
v Loading a table
v Inserting rows greater than 4 kilobytes
v Sending simple large objects

Chapter 3. Effect of configuration on CPU utilization 3-19

 The b option for sqlhosts allows the client to send and receive greater than 4
kilobytes. The value for the sqlhosts option should typically match the value for
IFX_NETBUF_SIZE.

You can use the following onstat command to see the network buffer size:
onstat -g afr global | grep net

The size field in the output shows the network buffer size in bytes.
Related reference:
Connectivity configuration (Administrator's Guide)
IFX_NETBUF_SIZE environment variable (SQL Reference)

Virtual processors and CPU utilization

While the database server is online, you can start and stop virtual processors (VPs)
that belong to certain classes.

You can use onmode -p or ON-Monitor to start additional VPs for the following
classes while the database server is online: CPU, AIO, PIO, LIO, SHM, TLI, and
SOC. You can drop VPs of the CPU class only while the database server is online.

You should carefully distinguish between poll threads for network connections and
poll threads for shared memory connections, which should run one per CPU
virtual processor. TCP connections should only be in network virtual processors,
and you should only have the minimum needed to maintain responsiveness.
Shared memory connections should only be in CPU virtual processors and should
run in every CPU virtual processor

Adding virtual processors

Whenever you add a network VP (SOC or TLI), you also add a poll thread. Every
poll thread runs in a separate VP, which can be either a CPU VP or a network VP
of the appropriate network type.

Adding more VPs can increase the load on CPU resources, so if the NETTYPE
value indicates that an available CPU VP can handle the poll thread, the database
server assigns the poll thread to that CPU VP. If all the CPU VPs have poll threads
assigned to them, the database server adds a second network VP to handle the poll
thread.

Automatic addition of CPU virtual processors

When the database server starts, the number of CPU virtual processors (VPs) is
automatically increased to half the number of CPU processors on the computer.
This ratio of CPU processors to CPU VPs is a recommended minimum to ensure
that the database server performs optimally in most situations.

During start up, the database server calculates a target number of CPU VPs that
represents an even number equal to or greater than half the number of CPU
processors and compares the target number with the currently allocated number of
CPU VPs. The database server adds the necessary number of CPU VPs to equal the
target number.

If fewer than eight CPU VPs are configured, the server can dynamically add CPU
VPs to a total (configured plus added) of eight.

3-20 IBM Informix Performance Guide

 The SINGLE_CPU_VP configuration parameter must be set to 0 for CPU VPs to be
automatically added. When CPU VPs are automatically added, the value of the
VPCLASS configuration parameter is not updated in the onconfig file; therefore,
the value of the VPCLASS configuration parameter for CPU VPs might not be the
same as the actual number of configured CPU VPs.

Use the auto_tune_cpu_vps task in the Scheduler to control the automatic addition
of CPU VPs. To prevent the automatic addition of CPU VPs, disable the
auto_tune_cpu_vps task in the ph_task table in the sysadmin database:
UPDATE ph_task
SET tk_enable = ’F’
WHERE tk_name = ’auto_tune_cpu_vps’;

The following table shows possible configurations and how many CPU VPs would
be added automatically in each situation.
Table 3-1. Example of how CPU VPs are automatically added
Automatically added
CPU processors Target CPU VPs Allocated CPU VPs CPU VPs
8 4 3 1
3 2 2 0
24 8 6 2

Related concepts:
“Setting the number of CPU VPs” on page 3-5

Monitoring virtual processors

Monitor the virtual processors to determine if the number of virtual processors
configured for the database server is optimal for the current level of activity.

To monitor virtual processors:

v Use command-line utilities, such as onstat-g ioq to view information. See “Using
some onstat-g commands to monitor virtual processors”
v Use the AUTO_AIOVPS configuration parameter to enable the database server
to automatically increase the number of AIO virtual processors and page-cleaner
threads when the server detects that AIO virtual processors are not keeping up
with the I/O workload.
v Query SMI tables. See “Using SMI tables to monitor virtual processors” on page
3-23.

Using some onstat-g commands to monitor virtual processors

You can use the onstat-g glo, onstat-g rea, and onstat-g ioq commands to monitor
virtual processors.

Monitor virtual processors with the onstat-g glo command:

Use the onstat-g glo command to display information about each virtual processor
that is running and to display cumulative statistics for each virtual-processor class.

The onstat -g glo command provides the following types of information:

v How many session threads that are running
v How often threads switch, yield, or need to spin many times to obtain a latch or
resource

Chapter 3. Effect of configuration on CPU utilization 3-21

 v The virtual processor classes that are running and how much time each class
spent running
v The number of virtual processors that are running for each virtual processor
class
v The virtual processors that are running and how much time each virtual
processor spent running
v The efficiency of each virtual processor

Use the onstat -g rea command to determine whether you need to increase the
number of virtual processors.
Related concepts:
“Monitor virtual processors with the onstat-g rea command”
Related reference:
onstat -g glo command: Print global multithreading information
(Administrator's Reference)

Monitor virtual processors with the onstat-g rea command:

Use the onstat-g rea command to monitor the number of threads in the ready
queue.

onstat-g rea displays this information:

v The status field in the output shows the value ready when the thread is in the
ready queue.
v The vp-class output field shows the virtual processor class on which the thread
executes.

If the number of threads in the ready queue is growing for a class of virtual
processors (for example, the CPU class), you might have to add more of those
virtual processors to your configuration.

Ready threads:
tid tcb rstcb prty status vp-class name

6 536a38 406464 4 ready 3cpu main_loop()

28 60cfe8 40a124 4 ready 1cpu onmode_mon
33 672a20 409dc4 2 ready 3cpu sqlexec

Figure 3-1. onstat-g rea output

Related concepts:
“Monitor virtual processors with the onstat-g glo command” on page 3-21
Related reference:
onstat -g rea command: Print ready threads (Administrator's Reference)

Monitor virtual processors with the onstat-g ioq command:

Use the onstat-g ioq command to determine whether you need to allocate
additional AIO virtual processors.

The onstat-g ioq command displays the length of the I/O queues under the
column len, as the figure below shows. You can also see the maximum queue

3-22 IBM Informix Performance Guide

length (since the database server started) in the maxlen column. If the length of the
I/O queue is growing, I/O requests are accumulating faster than the AIO virtual
processors can process them. If the length of the I/O queue continues to show that
I/O requests are accumulating, consider adding AIO virtual processors.

onstat -g ioq

AIO I/O queues:

q name/id len maxlen totalops dskread dskwrite dskcopy
adt 0 0 0 0 0 0 0
msc 0 0 1 12 0 0 0
aio 0 0 4 89 68 0 0
pio 0 0 1 1 0 1 0
lio 0 0 1 17 0 17 0
kio 0 0 0 0 0 0 0
gfd 3 0 3 254 242 12 0
gfd 4 0 17 614 261 353 0

onstat -d
Dbspaces
address number flags fchunk nchunks flags owner name
a1de1d8 1 1 1 1 N informix rootdbs
a1df550 2 1 2 1 N informix space1
2 active, 32,678 maximum
Chunks
address chk/dbs offset size free bpages flags pathname
a1de320 1 1 0 75000 66447 PO- /ix/root_chunk
a1df698 2 2 0 500 447 PO- /ix//chunk1
2 active, 32,678 maximum

Figure 3-2. onstat-g ioq and onstat -d output

Each chunk serviced by the AIO virtual processors has one line in the onstat-g ioq
output, identified by the value gfd in the q name column. You can correlate the
line in onstat -g ioq with the actual chunk because the chunks are in the same
order as in the onstat -d output. For example, in the onstat-g ioq output, there are
two gfd queues. The first gfd queue holds requests for root_chunk because it
corresponds to the first chunk shown in the onstat -d output. Likewise, the second
gfd queue holds requests for chunk1 because it corresponds to the second chunk
in the onstat -d output.

If the database server has a mixture of raw devices and cooked files, the gfd
queues correspond only to the cooked files in onstat -d output.
Related reference:
onstat -g ioq command: Print I/O queue information (Administrator's
Reference)

Using SMI tables to monitor virtual processors

You can get information from system-monitoring interface (SMI) tables to use to
monitor virtual processors.

You must connect to the sysmaster database to query the SMI tables. Query the
sysvpprof SMI table to obtain information about the virtual processors that are
currently running. This table contains the following columns.

Column Description
vpid ID number of the virtual processor

Chapter 3. Effect of configuration on CPU utilization 3-23

 Column Description
class Class of the virtual processor
usercpu Seconds of user CPU consumed
syscpu Seconds of system CPU consumed

CPU virtual processor private memory caches

Each CPU virtual processor (VP) can optionally have a private memory cache to
speed access time to memory blocks.

All memory allocations that are requested by threads in the database server are
fulfilled by memory pools. When a memory pool has insufficient memory blocks to
satisfy a memory allocation request, blocks are allocated from the global memory
pool. Because all threads use the same global memory pool, contention can occur.
Private memory caches allow each CPU virtual processor to retain its own set of
memory blocks that can be used to bypass the global memory pool. The initial
allocation for private memory caches is from the global memory pool. When the
blocks are freed, they are freed to the private memory cache on a specific CPU
virtual process. When a memory allocation is requested, the thread first checks
whether the allocation can be satisfied by blocks in the private memory cache.
Otherwise, the thread requests memory from the global memory pool.

To determine whether private memory caches might improve performance for your
database server, run the onstat -g spi command and look for the sh_lock mutex. If
the values from the onstat -g spi command output show contention for the
sh_lock mutex, try creating private memory caches.

You set the VP_MEMORY_CACHE_KB configuration parameter to enable private

memory caches by specifying the initial size of all private memory caches. The size
of each private memory cache increases or decreases automatically based on the
workload of the associated CPU VP. By default, the total size of private memory
caches can exceed the value of the VP_MEMORY_CACHE_KB configuration
parameter, but cannot exceed the value of the SHMTOTAL configuration
parameter. Alternatively, you can limit the total size of private memory caches to
the size value of the VP_MEMORY_CACHE_KB configuration parameter.

You can view statistics about CPU VP private memory caches by running the
onstat -g vpcache command. You can view statistics about memory pools by
running the onstat -g mem command.

Attention: If you have multiple CPU VPs, private memory caches can increase
the amount of memory that the database server uses.
Related reference:
VP_MEMORY_CACHE_KB configuration parameter (Administrator's
Reference)
onstat -g vpcache command: Print CPU VP private memory cache statistics
(Administrator's Reference)
onstat -g mem command: Print pool memory statistics (Administrator's
Reference)
onstat -g spi command: Print spin locks with long spins (Administrator's
Reference)

3-24 IBM Informix Performance Guide

Connections and CPU utilization
Some applications have a large number of client/server connections. Opening and
closing connections can consume a large amount of system CPU time.

The following topics describe ways that you might be able to reduce the system
CPU time required to open and close connections.

Multiplexed connections and CPU utilization

Many traditional nonthreaded SQL client applications use multiple database
connections to perform work for a single user. Each database connection
establishes a separate network connection to the database server. The multiplexed
connection facility provides the ability for one network connection in the database
server to handle multiple database connections from a client application.

Multiplexed connections enable the database server to create multiple database

connections without consuming the additional computer resources that are
required for additional network connections.

When a nonthreaded client uses a multiplexed connection, the database server still
creates the same number of user sessions and user threads as with a
nonmultiplexed connection. However, the number of network connections
decreases when you use multiplexed connections. Instead, the database server uses
a multiplex listener thread to allow the multiple database connections to share the
same network connection.

To improve response time for nonthreaded clients, you can use multiplexed
connections to execute SQL queries. The amount of performance improvement
depends on the following factors:
v The decrease in total number of network connections and the resulting decrease
in system CPU time
The usual cause for a large amount of system CPU time is the processing of
system calls for the network connection. Therefore, the maximum decrease in
system CPU time is proportional to the decrease in the total number of network
connections.
v The ratio of this decrease in system CPU time to the user CPU time
If the queries are simple and use little user CPU time, you might experience a
sizable reduction in response time when you use a multiplexed connection. But
if the queries are complex and use a large amount of user CPU time, you might
not experience a performance improvement.
To get an idea of the amounts of system CPU time and user CPU times per
virtual processor, use the onstat -g glo option.

To use multiplexed connections for a nonthreaded client application, you must take
the following steps before you bring up the database server:
1. Define an alias using the DBSERVERALIASES configuration parameter. For
example, specify:
DBSERVERALIASES ids_mux
2. Add an SQLHOSTS entry for the alias using sqlmux as the nettype entry, which
is the second column in the SQLHOSTS file. For example, specify:
ids_mux onsqlmux ......
The other fields in this entry, the hostname and servicename, must be present,
but they are ignored.

Chapter 3. Effect of configuration on CPU utilization 3-25

 3. Enable multiplexing for the selected connection types by specifying m=1 in the
sqlhosts file or registry that the client uses for the database server connection.
4. On Windows platforms, you must also set the IFX_SESSION_MUX
environment variable.

Warning: On Windows, a multithreaded application must not use the multiplexed

connection feature. If a multithreaded application enables the multiplexing option
in the sqlhosts registry entry and also defines the IFX_SESSION_MUX
environment variable, it can produce disastrous results, including crashing and
data corruption.
Related concepts:
Multiplexed connections (ESQL/C Guide)
Related tasks:
Supporting multiplexed connections (Administrator's Guide)

MaxConnect for multiple connections UNIX

IBM Informix MaxConnect is a networking product for Informix database server
environments on UNIX. You can use Informix MaxConnect to manage large
numbers (from several hundred to tens of thousands) of client/server connections.
Informix MaxConnect is best for OLTP data transfers, but is not recommended for
large multimedia data transfers.

Informix MaxConnect provides the following performance advantages for medium

to large OLTP configurations:
v Reduces CPU requirements on the database server by reducing the number of
physical connections.
Informix MaxConnect multiplexes connections so that the ratio of client
connections to database connections can be 100:1 or higher.
v Improves end-user response time by increasing system scalability to many
thousands of connections
v Reduces operating-system overhead by aggregating multiple small packets into
one transfer operation

To obtain maximum performance benefit, install Informix MaxConnect on either a

dedicated computer to which Informix clients connect or on the client application
server. Either of these configurations offloads the CPU requirements of handling a
large number of connections from the database server computer.

To monitor Informix MaxConnect, use the onstat -g imc command on the database
server computer and use the imcadmin command on the computer where Informix
MaxConnect is located.

For more information about installing, configuring, monitoring, and tuning

Informix MaxConnect, see the IBM Informix MaxConnect User's Guide.

Important: Informix MaxConnect and the IBM Informix MaxConnect User's Guide
ship separately from IBM Informix.

3-26 IBM Informix Performance Guide

Chapter 4. Effect of configuration on memory utilization
The combination of operating-system and Informix configuration parameters can
affect memory utilization.

You can change the settings of the Informix configuration parameters that directly
affect memory utilization, and you can adjust the settings for different types of
workloads.

Consider the amount of physical memory that is available on your host when you
allocate shared memory for the database server by setting operating-system
configuration parameters. In general, if you increase space for database server
shared memory, you can enhance the performance of your database server. You
must balance the amount of shared memory that is dedicated to the database
server against the memory requirements for VPs and other processes.
Related concepts:
“The Memory Grant Manager” on page 12-6

Shared memory
You must configure adequate shared-memory resources for the database server in
your operating system. Insufficient shared memory can adversely affect
performance.

The database server threads and processes require shared memory to share data by
sharing access to segments of memory.

The shared memory that Informix uses can be divided into the following parts,
each of which has one or more shared memory segments:
v Resident portion
v Virtual portion
v Message portion

The resident and message portions are static; you must allocate sufficient memory
for them before you bring the database server into online mode. (Typically, you
must reboot the operating system to reconfigure shared memory.) The virtual
portion of shared memory for the database server grows dynamically, but you
must still include an adequate initial amount for this portion in your allocation of
operating-system shared memory.

The amount of space that is required is the total that all portions of database server
shared memory need. You specify the total amount of shared memory with the
SHMTOTAL configuration parameter.

The LOCKS configuration parameter specifies the initial size of the lock table. If
the number of locks that sessions allocate exceeds the value of LOCKS, the
database server dynamically increases the size of the lock table. If you expect the
lock table to grow dynamically, set SHMTOTAL to 0. When SHMTOTAL is 0, there
is no limit on total memory (including shared memory) allocation.
Related reference:
LOCKS configuration parameter (Administrator's Reference)

© Copyright IBM Corp. 1996, 2014 4-1

 SHMTOTAL configuration parameter (Administrator's Reference)

Resident portion of shared memory

The resident portion of shared memory includes areas of shared memory that
record the state of the database server, including buffers, locks, log files, and the
locations of dbspaces, chunks, and tblspaces.

The settings that you use for the LOCKS, LOGBUFF, and PHYSBUFF configuration
parameters help determine the size of the resident portion.

The BUFFERPOOL configuration parameter determines the number of buffers

allocated to the resident segment when the database server is started. Subsequent
buffer pools that are added while the database server is running are moved into
virtual memory until the database server is restarted.

In addition to these configuration parameters, which affect the size of the resident
portion, the RESIDENT configuration parameter can affect memory use. When a
computer supports forced residency and the RESIDENT configuration parameter is
set to a value that locks the resident or resident and virtual portions, the resident
portion is never paged out.

The machine notes file for your database server indicates whether your operating
system supports forced residency.

On AIX, Solaris, and Linux systems that support large pages, the
IFX_LARGE_PAGES environment variable can enable the use of large pages for
non-message shared memory segments that are locked in physical memory. If large
pages are configured by operating system commands and the RESIDENT
configuration parameter specifies that some or all of the resident and virtual
portions of shared memory are locked in physical memory, Informix uses large
pages for the corresponding shared memory segments, provided sufficient large
pages are available. The use of large pages can offer significant performance
benefits in large memory configurations.
Related reference:
“Configuration parameters that affect memory utilization” on page 4-8
IFX_LARGE_PAGES environment variable (SQL Reference)

Virtual portion of shared memory

Informix uses the virtual portion of shared memory to allocate memory to each
database server subsystem, as needed.

The virtual portion of shared memory for the database server includes the
following components:
v Large buffers, which are used for large read and write I/O operations
v Sort-space pools
v Active thread-control blocks, stacks, and heaps
v User-session data
v Caches for SQL statements, data-dictionary information, and user-defined
routines
v A global pool for network-interface message buffers and other information

4-2 IBM Informix Performance Guide

 The SHMVIRTSIZE configuration parameter in the onconfig file provides the
initial size of the virtual portion. As the need for additional space in the virtual
portion arises, the database server adds shared memory in increments that the
SHMADD configuration parameter specifies. The EXTSHMADD configuration
parameter configures the size of the virtual-extension shared memory segments
that are added for user-defined routines and DataBlade routines. The limit on the
total shared memory allocated to the database server is specified by the
SHMTOTAL parameter.

The size of the virtual portion depends primarily on the types of applications and
queries that you are running. Depending on your application, an initial estimate
for the virtual portion might be as low as 100 KB per user or as high as 500 KB per
user, plus an additional 4 megabytes if you intend to use data distributions.

When a computer supports forced residency and the RESIDENT configuration

parameter is set to a value that locks virtual segments, the virtual segments that
are locked are never paged out.

On AIX, Solaris, and Linux systems that support large pages, the
IFX_LARGE_PAGES environment variable can enable the use of large pages for
non-message shared memory segments that are locked in physical memory. If large
pages are configured by operating system commands and the RESIDENT
configuration parameter specifies that some or all of the resident and virtual
portions of shared memory are locked in physical memory, Informix uses large
pages for the corresponding shared memory segments, provided sufficient large
pages are available. The use of large pages can offer significant performance
benefits in large memory configurations.
Related tasks:
“Creating data distributions” on page 13-13
Related reference:
“Configuration parameters that affect memory utilization” on page 4-8
IFX_LARGE_PAGES environment variable (SQL Reference)
EXTSHMADD configuration parameter (Administrator's Reference)
SHMADD configuration parameter (Administrator's Reference)
SHMTOTAL configuration parameter (Administrator's Reference)
SHMVIRTSIZE configuration parameter (Administrator's Reference)

Message portion of shared memory

The message portion of shared memory contains the message buffers that the
shared-memory communication interface uses. The amount of space required for
these buffers depends on the number of user connections that you allow using a
given networking interface.

If a particular interface is not used, you do not need to include space for it when
you allocate shared memory in the operating system.

Estimating the size of the resident portion of shared memory

You can use formulas to estimate the size of the resident portion (in KB) of shared
memory when you allocate operating-system shared memory.

Chapter 4. Effect of configuration on memory utilization 4-3

 The result of your calculations is an estimate that normally, slightly exceeds the
actual memory that is used for the resident portion of shared memory.

The following estimate was calculated to determine the resident portion of shared
memory on a 64-bit server. The sizes that are shown are subject to change, and the
calculation is approximate.

To estimate the size of the resident portion of shared memory

1. Estimate the size of the data buffer, using the following formula:
buffer_value = (BUFFERS * pagesize) + (BUFFERS * 254) + 250000
pagesize
is the shared-memory page size, as onstat -b shows it on the last line in
the buffer size field.
If you have multiple buffer pools, add the buffer sizes for each buffer pool
together.
2. Calculate the values in the following formulas:
locks_value = LOCKS * 44 44->128
logbuff_value = LOGBUFF * 1024 * 3
physbuff_value = PHYSBUFF * 1024 * 2
3. Calculate the estimated size of the resident portion in KB, using the following
formula:
rsegsize = 1.02 * (buffer_value + locks_value + logbuff_value
+ physbuff_value + 51,200 51,200->1,200,000) / 1024

Estimating the size of the virtual portion of shared memory

You can use a formula to estimate the initial size of the virtual portion of shared
memory. You specify the initial size in the SHMVIRTSIZE configuration parameter.

The formula for estimating an initial size of the virtual portion of shared memory
is as follows:
shmvirtsize = fixed overhead + shared structures +
(mncs * private structures) +
other buffers

To estimate an SHMVIRTSIZE value with the preceding formula:

1. Estimate the value for the fixed overhead portion of the formula as follows:
fixed overhead = global pool +
thread pool after booting
a. Run the onstat -g mem command to obtain the pool sizes allocated to
sessions.
b. Subtract the value in the freesize field from the value in the totalsize to
obtain the number of bytes allocated per session.
c. Estimate a value for the thread pool after booting variable. This variable is
partially dependent on the number of virtual processors.
2. Estimate the value of shared structures with the following formula:
shared structures = AIO vectors + sort memory +
dbspace backup buffers +
data-dictionary cache size +
size of user-defined routine cache +
histogram pool +
STMT_CACHE_SIZE (SQL statement cache) +
other pools (See onstat display.)
3. Estimate the next part of the formula, as follows:

4-4 IBM Informix Performance Guide

 a. Estimate the value of mncs (which is the maximum number of concurrent
sessions) with the following formula:
mncs = number of poll threads *
number connections per poll thread
The value for number of poll threads is the value that you specify in the
second field of the NETTYPE configuration parameter.
The value for number of connections per poll thread is the value that you
specify in the third field of the NETTYPE configuration parameter.
You can also obtain an estimate of the maximum number of concurrent
sessions when you run the onstat -u command during peak processing. The
last line of the onstat -u output contains the maximum number of
concurrent user threads.
b. Estimate the value of private structures, as follows:
private structures = stack + heap +
session control-block structures
stack Generally 32 KB but dependent on recursion in user-defined
routines. You can obtain the stack size for each thread with the
onstat -g sts option.
heap About 15 KB. You can obtain the heap size for an SQL statement
when you use the onstat -g stm option.
session control-block structures
The amount of memory used per session. The onstat -g ses option
displays the amount of memory, in bytes, in the total memory
column listed for each session id.
c. Multiply the results of steps 3a and 3b to obtain the following part of the
formula:
mncs * private structures
4. Estimate the value of other buffers to account for private buffers allocated for
features such as lightweight I/O operations for smart large objects (about 180
KB per user).
5. Add the results of steps 1 through 4 to obtain an estimate for the
SHMVIRTSIZE configuration parameter.

Tip: When the database server is running with a stable workload, you can use
onstat -g seg to obtain a precise value for the actual size of the virtual portion of
shared memory. You can then use the value for shared memory that this command
reports to reconfigure SHMVIRTSIZE.

To specify the size of segments that are added later to the virtual shared memory,
set the SHMADD configuration parameter. Use the EXTSHMADD configuration
parameter to specify the size of virtual-extension segments that are added for
user-defined routines and DataBlade routines.

The following table contains a list of additional topics for estimating the size of
shared structures in memory.
Table 4-1. Information for shared-memory structures
Shared-Memory Structure More Information
Sort memory “Estimating memory needed for sorting” on
page 7-19
Data-dictionary cache “Data-dictionary configuration” on page 4-22

Chapter 4. Effect of configuration on memory utilization 4-5

 Table 4-1. Information for shared-memory structures (continued)
Shared-Memory Structure More Information
Data-distribution cache (histogram pool) “Data-distribution configuration” on page 4-23
User-defined routine (UDR) cache “SPL routine executable format stored in UDR
cache” on page 10-33
SQL statement cache “Enabling the SQL statement cache” on page
13-42 “Monitor and tune the SQL statement
cache” on page 4-25
Other pools To see how much memory is allocated to the
different pools, use the onstat -g mem command.

Related concepts:
“Session memory” on page 4-38
Related reference:
SHMVIRTSIZE configuration parameter (Administrator's Reference)
NETTYPE configuration parameter (Administrator's Reference)
onstat -g mem command: Print pool memory statistics (Administrator's
Reference)

Estimating the size of the message portion of shared memory

You can estimate the size of the message portion of shared memory in kilobytes.

Estimate the size of the message portion of shared memory, using the following
formula:
msegsize = (10,531 * ipcshm_conn + 50,000)/1024
ipcshm_conn
is the number of connections that can be made using the shared-memory
interface, as determined by the NETTYPE parameter for the ipcshm
protocol.
Related reference:
NETTYPE configuration parameter (Administrator's Reference)

Configuring UNIX shared memory

On UNIX, you can configure shared-memory segments for the database server.

On UNIX, perform the following steps to configure the shared-memory segments

that your database server configuration needs. For information about how to set
parameters related to shared memory, see the configuration instructions for your
operating system.

To configure shared-memory segments for the database server:

1. If your operating system does not have a size limit for shared-memory
segments, take the following actions:
a. Set the operating-system configuration parameter for maximum segment
size, typically SHMMAX or SHMSIZE, to the total size that your database
server configuration requires. This size includes the amount of memory that
is required to start your database server instance and the amount of shared
memory that you allocate for dynamic growth of the virtual portion.

4-6 IBM Informix Performance Guide

 b. Set the operating-system configuration parameter for the maximum number
of segments, typically SHMMNI, to at least 1 per instance of the database
server.
2. If your operating system has a segment-size limit, take the following actions:
a. Set the operating-system configuration parameter for the maximum segment
size, typically SHMMAX or SHMSIZE, to the largest value that your system
allows.
b. Use the following formula to calculate the number of segments for your
instance of the database server. If there is a remainder, round up to the
nearest integer.
SHMMNI = total_shmem_size / SHMMAX
total_shmem_size
is the total amount of shared memory that you allocate for the
database server use.
3. Set the operating-system configuration parameter for the maximum number of
segments, typically SHMMNI, to a value that yields the total amount of shared
memory for the database server when multiplied by SHMMAX or SHMSIZE. If
your computer is dedicated to a single instance of the database server, that total
can be up to 90 percent of the size of virtual memory (physical memory plus
swap space).
4. If your operating system uses the SHMSEG configuration parameter to indicate
the maximum number of shared-memory segments that a process can attach,
set this parameter to a value that is equal to or greater than the largest number
of segments that you allocate for any instance of the database server.

For additional tips on configuring shared memory in the operating system, see the
machine notes file for UNIX or the release notes file for Windows.
Related concepts:
“The SHMADD and EXTSHMADD configuration parameters and memory
utilization” on page 4-17

Freeing shared memory with onmode -F

You can run the onmode -F command to free shared-memory segments that are
unavailable or no longer needed for a process.

The database server does not automatically free the shared-memory segments that
it adds during its operations. After memory has been allocated to the database
server virtual portion, the memory remains unavailable for use by other processes
running on the host computer. When the database server runs a large
decision-support query, it might acquire a large amount of shared memory. After
the query completes, the database server no longer requires that shared memory.
However, the shared memory that the database server allocated to service the
query remains assigned to the virtual portion even though it is no longer needed.

The onmode -F command locates and returns unused 8-kilobyte blocks of shared
memory that the database server still holds. Although this command runs only
briefly (one or two seconds), onmode -F dramatically inhibits user activity while it
runs. Systems with multiple CPUs and CPU VPs typically experience less
degradation while this utility runs.

You should run onmode -F during slack periods with an operating-system

scheduling facility (such as cron on UNIX). In addition, consider running this

Chapter 4. Effect of configuration on memory utilization 4-7

 utility after you perform any task that substantially increases the size of database
server shared memory, such as large decision-support queries, index builds, sorts,
or backup operations.
Related reference:
onmode -F: Free unused memory segments (Administrator's Reference)

Configuration parameters that affect memory utilization

A large number of configuration parameters in the ONCONFIG file affect memory
utilization and performance.

The following configuration parameters significantly affect memory utilization:

v BUFFERPOOL
v DS_NONPDQ_QUERY_MEM
v DS_TOTAL_MEMORY
v EXTSHMADD
v LOCKS
v LOGBUFF
v LOW_MEMORY_MGR
v LOW_MEMORY_RESERVE
v PHYSBUFF
v RESIDENT
v SHMADD
v SHMBASE
v SHMTOTAL
v SHMVIRTSIZE
v SHMVIRT_ALLOCSEG
v STACKSIZE
v Memory cache parameters (see “Configure and monitor memory caches” on
page 4-21)
v Network buffer size (see “Network buffer pools” on page 3-17)

The SHMBASE parameter indicates the starting address for database server shared
memory. When set according to the instructions in the machine notes file or release
notes file, this parameter has no appreciable effect on performance. For the path
name of each file, see the Introduction to this guide.

The DS_NONPDQ_QUERY_MEM parameter increases the amount of memory that

is available for non-PDQ queries. You can only use this parameter if PDQ priority
is set to zero. For more information, see “Configuring memory for queries with
hash joins, aggregates, and other memory-intensive elements” on page 13-34.

The following sections describe the performance effects and considerations

associated with some of the configuration parameters that are listed at the
beginning of this section.
Related concepts:
“Resident portion of shared memory” on page 4-2
“Virtual portion of shared memory” on page 4-2
Related reference:

4-8 IBM Informix Performance Guide

 LOW_MEMORY_MGR configuration parameter (Administrator's Reference)
LOW_MEMORY_RESERVE configuration parameter (Administrator's
Reference)

Setting the size of the buffer pool, logical-log buffer, and

physical-log buffer
The values that you specify for the BUFFERPOOL, DS_TOTAL_MEMORY,
LOGBUFF, and PHYSBUFF configuration parameters depend on the type of
applications that you are using (OLTP or DSS) and the page size.

Table 4-2 lists suggested settings for these parameters or guidelines for setting the
parameters.

For information about estimating the size of the resident portion of shared
memory, see “Estimating the size of the resident portion of shared memory” on
page 4-3. This calculation includes figuring the size of the buffer pool, logical-log
buffer, physical-log buffer, and lock table.
Table 4-2. Guidelines for OLTP and DSS applications
Configuration Parameter OLTP Applications DSS Applications
BUFFERPOOL The percentage of physical Set to a small buffer value and
memory that you need for increase the
buffer space depends on the DS_TOTAL_MEMORY value
amount of memory that is for light scans, queries, and
available on your system and sorts.
the amount of memory that is
used for other applications. For operations such as index
builds that read data through
the buffer pool, configure a
larger number of buffers.
DS_TOTAL_MEMORY Set to a value from 20 to 50 Set to a value from 50 to 90
percent of the value of percent of SHMTOTAL.
SHMTOTAL, in kilobytes.
LOGBUFF The default value for the Because database or table
logical log buffer size is 64 logging is usually turned off
KB. for DSS applications, you can
set LOGBUFF to 32 KB.
If you decide to use a smaller
value, the database server
generates a message a
message that indicates that
optimal performance might
not be obtained. Using a
logical log buffer smaller than
64 KB, impacts performance,
not transaction integrity.

If the database or application

is defined to use buffered
logging, increasing the
LOGBUFF size beyond 64 KB
improves performance.

Chapter 4. Effect of configuration on memory utilization 4-9

 Table 4-2. Guidelines for OLTP and DSS applications (continued)
Configuration Parameter OLTP Applications DSS Applications
PHYSBUFF The default value for the Because most DSS applications
physical log buffer size is 128 do not physically log, you can
KB. set PHYSBUFF to 32 KB.

If the
RTO_SERVER_RESTART
configuration parameter is
enabled, use the 512 kilobyte
default value for PHYSBUFF.

If you decide to use a value

that is smaller than the
default value, the database
server generates a message
that indicates that optimal
performance might not be
obtained. Using a physical log
buffer that is smaller than the
default size impacts
performance, not transaction
integrity.

Related reference:
BUFFERPOOL configuration parameter (Administrator's Reference)
DS_TOTAL_MEMORY configuration parameter (Administrator's Reference)
LOGBUFF configuration parameter (Administrator's Reference)
PHYSBUFF configuration parameter (Administrator's Reference)
RTO_SERVER_RESTART configuration parameter (Administrator's Reference)

The BUFFERPOOL configuration parameter and memory

utilization
The BUFFERPOOL configuration parameter specifies the properties of buffer pools.
The information that you define in the BUFFERPOOL configuration parameter
fields affects memory use.

You can have multiple buffer pools if you have dbspaces that use different page
sizes. The onconfig configuration file contains a BUFFERPOOL line for each page
size. For example, on a computer with a 2 KB page size, the onconfig file can
contain up to nine lines, including the default specification. When you create a
dbspace with a different page size, a buffer pool for that page size is created
automatically, if it does not exist. A BUFFERPOOL entry for the page size is added
to the onconfig file. The values of the BUFFERPOOL configuration parameter
fields are the same as the default specification.

The BUFFERPOOL configuration parameter controls the number of data buffers

available to the database server. These buffers are in the resident portion of shared
memory (buffer pool) and are used to cache database data pages in memory.

Increasing the number of buffers increases the likelihood that a needed data page
might already be in memory as the result of a previous request. However,
allocating too many buffers can affect the memory-management system and lead to

4-10 IBM Informix Performance Guide

excess operating system paging activity. To take advantage of the large memory
available on 64-bit addressing machines, you can increase the size of the buffer
pool.

The size of the buffer pool has a significant effect on database I/O and transaction
throughput.

The size of the buffer pool is equal to the number of buffers multiplied by the page
size. The percentage of physical memory that you need for buffer space depends
on the amount of memory that you have available on your system and the amount
that is used for other applications. For systems with a large amount of available
physical memory (4 GB or more), buffer space might be as much as 90 percent of
physical memory. For systems with smaller amounts of available physical memory,
buffer space might range from 20 - 25 percent of physical memory.

For example, suppose that your system has a page size of 2 KB and 100 MB of
physical memory. You can set the value in the buffers field to 10,000 - 12,500,
which allocates 20 - 25 MB of memory.

Calculate all other shared-memory parameters after you specify the size of the
buffer pool.

Note: If you use non-default page sizes, you might need to increase the size of
your physical log. If you frequently update non-default pages, you might need a
150 - 200 percent increase of the physical log size. Some experimentation might be
needed to tune the physical log. You can adjust the size of the physical log as
necessary according to how frequently the filling of the physical log triggers
checkpoints.

You can use onstat -g buf to monitor buffer pool statistics, including the
read-cache rate of the buffer pool. This rate represents the percentage of database
pages that are already present in a shared-memory buffer when a query requests a
page. (If a page is not already present, the database server must copy it into
memory from disk.) If the database server finds the page in the buffer pool, it
spends less time on disk I/O. Therefore, you want a high read-cache rate for good
performance. For OLTP applications where many users read small sets of data, the
goal is to achieve a read cache rate of 95 percent or better.

If the read-cache rate is low, you can repeatedly increase buffers and restart the
database server. As you increase the BUFFERPOOL value of buffers, you reach a
point at which increasing the value no longer produces significant gains in the
read-cache rate, or you reach the upper limit of your operating-system
shared-memory allocation.

Use the memory-management monitor utility in your operating system (such as

vmstat or sar on UNIX) to note the level of page scans and paging-out activity. If
these levels rise suddenly or rise to unacceptable levels during peak database
activity, reduce the size of the buffer pool.

Smart large objects and buffers

Depending upon your situation, you can take one of the following actions to
achieve better performance for applications that use smart large objects:
v If your applications frequently access smart large objects that are 2 KB or 4 KB
in size, use the buffer pool to keep them in memory longer. Use the following
formula to increase the value of the buffers field:

Chapter 4. Effect of configuration on memory utilization 4-11

 Additional_buffers = numcur_open_lo *
(lo_userdata / pagesize)
In this formula:
– numcur_open_lo is the number of concurrently opened smart large objects that
you can obtain from the onstat -g smb fdd command.
– lo_userdata is the number of bytes of smart-large-object data that you want to
buffer.
– pagesize is the default page size in bytes for the computer.
As a rule, try to have enough buffers to hold two smart-large-object pages for
each concurrently open smart large object. The additional page is available for
read-ahead purposes.
v Use lightweight I/O buffers in the virtual portion of shared memory.
Use lightweight I/O buffers only when you read or write smart large objects in
operations greater than 8000 bytes and seldom access them. That is, if the read
or write function calls read large amounts of data in a single-function
invocation, use lightweight I/O buffers.
When you use lightweight I/O buffers, you can prevent the flood of smart large
objects into the buffer pool and leave more buffers available for other data pages
that multiple users frequently access.
Related concepts:
“Lightweight I/O for smart large objects” on page 5-23
“BUFFERPOOL and its effect on page cleaning” on page 5-40
Related reference:
BUFFERPOOL configuration parameter (Administrator's Reference)
Monitor buffers (Administrator's Guide)

The DS_TOTAL_MEMORY configuration parameter and memory

utilization
The DS_TOTAL_MEMORY configuration parameter places a ceiling on the amount
of shared memory that a query can obtain. You can use this parameter to limit the
performance impact of large, memory-intensive queries. The higher you set this
parameter, the more memory a large query can use, and the less memory is
available for processing other queries and transactions.

For OLTP applications, set DS_TOTAL_MEMORY to 20 - 50 percent of the value of

SHMTOTAL, in KB. For applications that involve large decision-support (DSS)
queries, increase the value of DS_TOTAL_MEMORY to 50 - 80 percent of
SHMTOTAL. If you use your database server instance exclusively for DSS queries,
set this parameter to 90 percent of SHMTOTAL.

A quantum unit is the minimum increment of memory that is allocated to a query.

The Memory Grant Manager (MGM) allocates memory to queries in quantum
units. The database server uses the value of DS_MAX_QUERIES with the value of
DS_TOTAL_MEMORY to calculate a quantum of memory, according to the
following formula:
quantum = DS_TOTAL_MEMORY / DS_MAX_QUERIES

The database server can adjust the size of the quantum dynamically when it grants
memory. To allow for more simultaneous queries with smaller quanta each,
increase the value of the DS_MAX_QUERIES configuration parameter.
Related concepts:

4-12 IBM Informix Performance Guide

“The Memory Grant Manager” on page 12-6
“Limiting the performance impact of CPU-intensive queries” on page 3-12
Related reference:
DS_TOTAL_MEMORY configuration parameter (Administrator's Reference)

Algorithm for determining DS_TOTAL_MEMORY:

The database server derives a value for DS_TOTAL_MEMORY if you do not set
the DS_TOTAL_MEMORY configuration parameter or if you set this configuration
parameter to an inappropriate value.

Whenever the database server changes the value that you assigned to
DS_TOTAL_MEMORY, it sends the following message to your console:
DS_TOTAL_MEMORY recalculated and changed from old_value Kb
to new_value Kb

The variable old_value represents the value that you assigned to

DS_TOTAL_MEMORY in your configuration file. The variable new_value represents
the value that the database server derived.

When you receive the preceding message, you can use the algorithm to investigate
what values the database server considers inappropriate. You can then take
corrective action based on your investigation.

The following sections document the algorithm that the database server uses to
derive the new value for DS_TOTAL_MEMORY.

Deriving a minimum for decision-support memory:

In the first part of the algorithm that the database server uses to derive the new
value for the DS_TOTAL_MEMORY configuration parameter, the database server
establishes a minimum amount for decision-support memory.

When you assign a value to the DS_MAX_QUERIES configuration parameter, the

database server sets the minimum amount of decision-support memory according
to the following formula:
min_ds_total_memory = DS_MAX_QUERIES * 128 kilobytes

When you do not assign a value to the DS_MAX_QUERIES configuration

parameter, the database server uses the following formula instead, which is based
on the value of information in the VPCLASS configuration parameter:
min_ds_total_memory = NUMBER_CPUVPS * 2 * 128 kilobytes

Deriving a working value for decision-support memory:

In the second part of the algorithm that the database server uses to derive the new
value for the DS_TOTAL_MEMORY configuration parameter, the database server
establishes a working value for the amount of decision-support memory.

The database server verifies this amount in the third and final part of the
algorithm.

Chapter 4. Effect of configuration on memory utilization 4-13

 When the DS_TOTAL_MEMORY configuration parameter is set:

When the DS_TOTAL_MEMORY configuration parameter is set, the database

server checks whether the SHMTOTAL configuration parameter is set and then
determines which formula to use to calculate the amount of decision-support
memory.

When SHMTOTAL is set, the database server uses the following formula to
calculate the amount of decision-support memory:
IF DS_TOTAL_MEMORY <= SHMTOTAL - nondecision_support_memory THEN
decision_support_memory = DS_TOTAL_MEMORY
ELSE
decision_support_memory = SHMTOTAL -
nondecision_support_memory

This algorithm effectively prevents you from setting DS_TOTAL_MEMORY to

values that the database server cannot possibly allocate to decision-support
memory.

When SHMTOTAL is not set, the database server sets decision-support memory
equal to the value that you specified in DS_TOTAL_MEMORY.
Related reference:
DS_TOTAL_MEMORY configuration parameter (Administrator's Reference)

When the DS_TOTAL_MEMORY configuration parameter is not set:

When the DS_TOTAL_MEMORY configuration parameter is not set, the database

server uses other sources to calculate a value for the amount of decision-support
memory.

When SHMTOTAL is set, the database server uses the following formula to
calculate the amount of decision-support memory:
decision_support_memory = SHMTOTAL -
nondecision_support_memory

When the database server finds that you did not set SHMTOTAL, it sets
decision-support memory as in the following example:
decision_support_memory = min_ds_total_memory

For a description of the variable min_ds_total_memory, see “Deriving a minimum

for decision-support memory” on page 4-13.

Checking the derived value for decision-support memory:

In the final part of the algorithm that the database server uses to derive the new
value for the DS_TOTAL_MEMORY configuration parameter, the database server
verifies that the amount of shared memory is greater than min_ds_total_memory and
less than the maximum possible memory space for your computer.

When the database server finds that the derived value for decision-support
memory is less than the value of the min_ds_total_memory variable, it sets
decision-support memory equal to the value of min_ds_total_memory.

When the database server finds that the derived value for decision-support
memory is greater than the maximum possible memory space for your computer, it
sets decision-support memory equal to the maximum possible memory space.

4-14 IBM Informix Performance Guide

 The LOGBUFF configuration parameter and memory utilization
The LOGBUFF configuration parameter determines the amount of shared memory
that is reserved for each of the three buffers that hold the logical-log records until
they are flushed to the logical-log file on disk. The size of a buffer determines how
often it fills and therefore how often it must be flushed to the logical-log file on
disk.

If you log smart large objects, increase the size of the logical-log buffers to prevent
frequent flushing to the logical-log file on disk.
Related reference:
LOGBUFF configuration parameter (Administrator's Reference)

The LOW_MEMORY_RESERVE configuration parameter and

memory utilization
The LOW_MEMORY_RESERVE configuration parameter reserves a specific amount
of memory, in kilobytes, for the database server to use when critical activities are
needed and the server has limited free memory.

If you enable the new LOW_MEMORY_RESERVE configuration parameter by

setting it to a specified value in kilobytes, critical activities, such as rollback
activities, can complete even when you receive out-of-memory errors.
Related reference:
LOW_MEMORY_RESERVE configuration parameter (Administrator's
Reference)
onstat -g seg command: Print shared memory segment statistics
(Administrator's Reference)

The PHYSBUFF configuration parameter and memory utilization

The PHYSBUFF configuration parameter determines the amount of shared memory
that is reserved for each of the two buffers that serve as temporary storage space
for data pages that are about to be modified. The size of a buffer determines how
often it fills and therefore how often it must be flushed to the physical log on disk.

Choose a value for PHYSBUFF that is an even increment of the system page size.
Related reference:
PHYSBUFF configuration parameter (Administrator's Reference)

The LOCKS configuration parameter and memory utilization

The LOCKS configuration parameter specifies the initial size of the lock table. The
lock table holds an entry for each lock that a session uses. Each lock uses 120 bytes
within a lock table. You must provide for this amount of memory when you
configure shared memory.

If the number of locks needed by sessions exceeds the value set in the LOCKS
configuration parameter, the database server attempts to increase the lock table by
doubling its size. Each time that the lock table overflows (when the number of
locks needed is greater than the current size of the lock table), the database server
increases the size of the lock table, up to 99 times. Each time that the database
server increases the size of the lock table, the server attempts to double its size.
However, the server will limit each actual increase to no more than the maximum
number of added locks shown in Table 4-3 on page 4-16. After the 99th time that

Chapter 4. Effect of configuration on memory utilization 4-15

 the database server increases the lock table, the server no longer increases the size
of the lock table, and an application needing a lock receives an error.

The following table shows the maximum number of locks allowed on 32-bit and
64-bit platforms
Table 4-3. Maximum number of locks on 32-bit and 64-bit platforms
Maximum
Maximum Number of
Maximum Number of Locks Added Maximum
Number of Dynamic Lock Per Lock Table Number of
Platform Initial Locks Table Extensions Extension Locks Allowed
32-bit 8,000,000 99 100,000 8,000,000 + (99 x
100,000)
64-bit 500,000,000 99 1,000,000 500,000,000 + (99
x 1,000,000)

The default value for the LOCKS configuration parameter is 20,000.

To estimate a different value for the LOCKS configuration parameter, estimate the
maximum number of locks that a query needs and multiply this estimate by the
number of concurrent users. You can use the guidelines in the following table to
estimate the number of locks that a query needs.

Locks per TEXT or BYTE

Statement Isolation Level Table Row Key Data CLOB or BLOB Data
SELECT Dirty Read 0 0 0 0 0
SELECT Committed Read 1 0 0 0 0
SELECT Cursor Stability 1 1 0 0 1 lock for the CLOB or
BLOB value or (if
byte-range locking is
used) 1 lock for each
range
SELECT Indexed 1 Number of Number of 0 1 lock for the CLOB or
Repeatable Read rows that rows that BLOB value or (if
satisfy satisfy byte-range locking is
conditions conditions used) 1 lock for each
range
SELECT Sequential 1 0 0 0 1 lock for the CLOB or
Repeatable Read BLOB value or (if
byte-range locking is
used) 1 lock for each
range
INSERT Not applicable 1 1 Number of Number of 1 lock for the CLOB or
indexes pages in TEXT BLOB value
or BYTE data
DELETE Not applicable 1 1 Number of Number of 1 lock for the CLOB or
indexes pages in TEXT BLOB value
or BYTE data
UPDATE Not applicable 1 1 2 per Number of 1 lock for the CLOB or
changed key pages in old BLOB value or (if
value plus new byte-range locking is
TEXT or BYTE used) 1 lock for each
data range

4-16 IBM Informix Performance Guide

 Important: During the execution of the SQL statement DROP DATABASE, the
database server acquires and holds a lock on each table in the database until the
entire DROP operation completes. Make sure that the value for LOCKS is large
enough to accommodate the largest number of tables in a database.
Related concepts:
“Configuring and managing lock usage” on page 8-12
Related reference:
LOCKS configuration parameter (Administrator's Reference)

The RESIDENT configuration parameter and memory

utilization
The RESIDENT configuration parameter specifies whether shared-memory
residency is enforced for the resident portion of database server shared memory.
This configuration parameter works only on computers that support forced
residency.

The resident portion in the database server contains the buffer pools that are used
for database read and write activity. Performance improves when these buffers
remain in physical memory.

You should set the RESIDENT parameter to 1. If forced residency is not an option
on your computer, the database server issues an error message and ignores this
configuration parameter.

On machines that support 64-bit addressing, you can have a very large buffer pool
and the virtual portion of database server shared memory can also be very large.
The virtual portion contains various memory caches that improve performance of
multiple queries that access the same tables (see “Configure and monitor memory
caches” on page 4-21). To make the virtual portion resident in physical memory in
addition to the resident portion, set the RESIDENT parameter to -1.

If your buffer pool is very large, but your physical memory is not very large, you
can set RESIDENT to a value greater than 1 to indicate the number of memory
segments to stay in physical memory. This specification makes only a subset of the
buffer pool resident.

You can turn residency on or off for the resident portion of shared memory in the
following ways:
v Use the onmode utility to reverse temporarily the state of shared-memory
residency while the database server is online.
v Change the RESIDENT parameter to turn shared-memory residency on or off the
next time that you start database server shared memory.
Related reference:
RESIDENT configuration parameter (Administrator's Reference)

The SHMADD and EXTSHMADD configuration parameters and

memory utilization
The SHMADD configuration parameter specifies the size of each increment of
shared memory that the database server dynamically adds to the virtual portion.
The EXTSHMADD configuration parameter specifies the size of a virtual-extension

Chapter 4. Effect of configuration on memory utilization 4-17

 segment that is added when user-defined routines or DataBlade routines run in
user-defined virtual processors. Trade-offs are involved in determining the size of
an increment.

Adding shared memory uses CPU cycles. The larger each increment, the fewer
increments are required, but less memory is available for other processes. Adding
large increments is generally preferred; but when memory is heavily loaded (the
scan rate or paging-out rate is high), smaller increments allow better sharing of
memory resources among competing programs.

The range of values for SHMADD is 1024 through 4294967296 KB for a 64-bit
operating system and 1024 through 524288 KB for a 32-bit operating system. The
following table contains recommendations for setting SHMADD according to the
size of physical memory.

Memory Size SHMADD Value

256 MB or less 8192 KB (the default)
257 - 512 MB 16,384 KB
Larger than 512 MB 32,768 KB

The range of values for EXTSHMADD is the same as the range of values of
SHMADD.

Note: A shared memory segment can be as large as 4 terabytes, depending on

platform limits and the value of the SHMMAX kernel parameter. Use the onstat -g
seg command to display the number of shared-memory segments that the database
server is currently using.
Related tasks:
“Configuring UNIX shared memory” on page 4-6
Related reference:
SHMADD configuration parameter (Administrator's Reference)
EXTSHMADD configuration parameter (Administrator's Reference)

The SHMTOTAL configuration parameter and memory

utilization
The SHMTOTAL configuration parameter places an absolute upper limit on the
amount of shared memory that an instance of the database server can use.

If the SHMTOTAL configuration parameter is set to 0 or left unassigned, the

database server continues to attach additional shared memory as needed until no
virtual memory is available on the system.

You can usually set the SHMTOTAL configuration parameter to 0, except in the
following cases:
v You must limit the amount of virtual memory that the database server uses for
other applications or other reasons.
v Your operating system runs out of swap space and performs abnormally. In this
case, you can set SHMTOTAL to a value that is a few megabytes less than the
total swap space that is available on your computer.
v You are using automatic low memory management.
Related reference:
4-18 IBM Informix Performance Guide
 SHMTOTAL configuration parameter (Administrator's Reference)

The SHMVIRTSIZE configuration parameter and memory

utilization
The SHMVIRTSIZE parameter specifies the size of the virtual portion of shared
memory to allocate when you start the database server. The virtual portion of
shared memory holds session- and request-specific data as well as other
information.

Although the database server adds increments of shared memory to the virtual
portion as needed to process large queries or peak loads, allocation of shared
memory increases time for transaction processing. Therefore, you should set
SHMVIRTSIZE to provide a virtual portion large enough to cover your normal
daily operating requirements. The size of SHMVIRTSIZE can be as large the
SHMMAX configuration parameter allows.

The maximum value of SHMVIRTSIZE, which must be a positive integer, is:

v 4 terabytes on a 64-bit database server
v 2 gigabytes on a 32-bit database server

For an initial setting, it is suggested that you use the larger of the following values:
v 8000
v connections * 350

The connections variable is the number of connections for all network types that are
specified in the sqlhosts information by one or more NETTYPE configuration
parameters. (The database server uses connections * 200 by default.)

Once system utilization reaches a stable workload, you can reconfigure a new
value for SHMVIRTSIZE. As noted in “Freeing shared memory with onmode -F”
on page 4-7, you can instruct the database server to release shared-memory
segments that are no longer in use after a peak workload or large query.
Related reference:
SHMVIRTSIZE configuration parameter (Administrator's Reference)

The SHMVIRT_ALLOCSEG configuration parameter and

memory utilization
The SHMVIRT_ALLOCSEG configuration parameter specifies a threshold at which
the database server should allocate memory. This configuration parameter also
defines an alarm event security-code that is activated if the server cannot allocate
the new memory segment, thus ensuring that the database server never runs out of
memory.

When you set the SHMVIRT_ALLOCSEG configuration parameter, you must:

v Specify the percentage of memory used or the whole number of kilobytes
remaining on the server. You cannot use negative values and values between 0
and .39.
v Specify the alarm event-security code, which is a value ranging from 1 (not
noteworthy) to 5 (fatal). If you do not specify an event-security code, the server
sets the value to 3, which is the default value.

Example 1:

Chapter 4. Effect of configuration on memory utilization 4-19

 SHMVIRT_ALLOCSEG 3000, 4

This specifies that if the database serve has 3000 kilobytes remaining in virtual
memory and additional kilobytes of memory cannot be allocated, the server raises
an alarm level of 4.

Example 2:
SHMVIRT_ALLOCSEG .8, 4

This specifies that if the database server has twenty percent remaining in virtual
memory and additional kilobytes of memory cannot be allocated, the server raises
an alarm level of 4.
Related reference:
Event Alarm Parameters (Administrator's Reference)
SHMVIRT_ALLOCSEG configuration parameter (Administrator's Reference)

The STACKSIZE configuration parameter and memory

utilization
The STACKSIZE configuration parameter indicates the initial stack size for each
thread. The database server assigns the amount of space that this parameter
indicates to each active thread. This space comes from the virtual portion of
database server shared memory. You can reduce the amount of shared memory
that the database server adds dynamically.

To reduce the amount of shared memory that the database server adds
dynamically, estimate the amount of the stack space required for the average
number of threads that your system runs and include that amount in the value
that you set for the SHMVIRTSIZE configuration parameter.

To estimate the amount of stack space that you require, use the following formula:
stacktotal = STACKSIZE * avg_no_of_threads
avg_no_of_threads
is the average number of threads. You can monitor the number of active
threads at regular intervals to determine this amount. Use onstat -g sts to
check the stack use of threads. A general estimate is between 60 and 70
percent of the total number of connections (specified in the NETTYPE
parameters in your ONCONFIG file), depending on your workload.

The database server also executes user-defined routines (UDRs) with user threads
that use this stack. Programmers who write user-defined routines should take the
following measures to avoid stack overflow:
v Do not use large automatic arrays.
v Avoid excessively deep calling sequences.
v For DB-Access only: Use mi_call to manage recursive calls.

If you cannot avoid stack overflow with these measures, use the STACK modifier
of the CREATE FUNCTION statement to increase the stack for a particular routine.
Related reference:
STACKSIZE configuration parameter (Administrator's Reference)

4-20 IBM Informix Performance Guide

Configure and monitor memory caches
The database server uses caches to store information in memory instead of
performing a disk read or another operation to obtain the information. These
memory caches improve performance for multiple queries that access the same
tables. You can set some configuration parameters to increase the effectiveness of
each cache. You can view information about memory caches by running onstat
commands.

The following table lists the main memory caches that have the greatest effect on
performance and how to configure and monitor those caches.
Table 4-4. Main memory caches
onstat
Cache Name Cache Description Configuration Parameters command
Data Dictionary Stores information about the table DD_HASHSIZE: The maximum onstat -g dic
definition (such as column names and number of buckets in the cache.
data types).
DD_HASHMAX: The number of
tables in each bucket
Data Distribution Stores distribution statistics for a column. DS_POOLSIZE: The maximum onstat -g dsc
number of entries in the cache.

DS_HASHSIZE: The number of

buckets in the cache.
SQL Statement Stores parsed and optimized SQL STMT_CACHE: Enable the SQL onstat -g ssc
statements. statement cache.

STMT_CACHE_HITS: The number

of times anSQL statement is run
before it is cached.

STMT_CACHE_NOLIMIT: Prohibit
entries into the SQL statement cache
when allocated memory exceeds the
value of the STMT_CACHE_SIZE
configuration parameter.

STMT_CACHE_NUMPOOL: The
number of memory pools for the
SQL statement cache.

STMT_CACHE_SIZE: The size of

the SQL statement cache, in KB.
UDR Stores frequently used user-defined PC_POOLSIZE: The maximum onstat -g prc
routines and SPL routines. number of user-defined routines
and SPL routines in the cache.

PC_HASHSIZE: The number of

buckets in the UDR cache.

Related concepts:
“SPL routine executable format stored in UDR cache” on page 10-33
Related reference:
onstat -g cac command: Print information about caches (Administrator's
Reference)

Chapter 4. Effect of configuration on memory utilization 4-21

 onstat -g dsc command: Print distribution cache information (Administrator's
Reference)
onstat -g prc command: Print sessions using UDR or SPL routines
(Administrator's Reference)
onstat -g ssc command: Print SQL statement occurrences (Administrator's
Reference)
Database configuration parameters (Administrator's Reference)

Data-dictionary cache
The first time that the database server accesses a table, it retrieves the information
that it needs about the table (such as the column names and data types) from the
system catalog tables on disk. After the database server has accessed the table, it
places that information in the data-dictionary cache in shared memory.

Figure 4-1 shows how the database server uses this cache for multiple users. User 1
accesses the column information for tabid 120 for the first time. The database
server puts the column information in the data-dictionary cache. When user 2, user
3 and user 4 access the same table, the database server does not have to read from
disk to access the data-dictionary information for the table. Instead, it reads the
dictionary information from the data-dictionary cache in memory.

User 2 User 3 User 4 User 1

Shared-memory data-dictionary cache

syscolumns
syscolumns
tabid colname
tabid colname
120 lname
120 lname
120 fname
120 fname

Figure 4-1. Data-dictionary cache

The database server still places pages for system catalog tables in the buffer pool,
as it does all other data and index pages. However, the data-dictionary cache offers
an additional performance advantage, because the data-dictionary information is
organized in a more efficient format and organized to allow fast retrieval.

Data-dictionary configuration
The database server uses a hashing algorithm to store and locate information
within the data-dictionary cache. The DD_HASHSIZE and DD_HASHMAX
configuration parameters control the size of the data-dictionary cache.

To modify the number of buckets in the data-dictionary cache, use DD_HASHSIZE

(must be a prime number). To modify the number of tables that can be stored in
one bucket, use DD_HASHMAX.

For medium to large systems, you can start with the following values for these
configuration parameters:
v DD_HASHSIZE: 503
v DD_HASHMAX: 4

4-22 IBM Informix Performance Guide

 With these values, you can potentially store information about 2012 tables in the
data-dictionary cache, and each hash bucket can have a maximum of 4 tables.

If the bucket reaches the maximum size, the database server uses a least recently
used mechanism to clear entries from the data dictionary.
Related reference:
DD_HASHSIZE configuration parameter (Administrator's Reference)
DD_HASHMAX configuration parameter (Administrator's Reference)

Data-distribution cache
The query optimizer uses distribution statistics generated by the UPDATE
STATISTICS statement in the MEDIUM or HIGH mode to determine the query
plan with the lowest cost. The first time that the optimizer accesses the distribution
statistics for a column, the database server retrieves the statistics from the
sysdistrib system catalog table on disk and places that information in the
data-distribution cache in memory.

Figure 4-2 shows how the database server accesses the data-distribution cache for
multiple users. When the optimizer accesses the column distribution statistics for
User 1 for the first time, the database server puts the distribution statistics in the
data-distribution cache. When the optimizer determines the query plan for user 2,
user 3 and user 4 who access the same column, the database server does not have
to read from disk to access the data-distribution information for the table. Instead,
it reads the distribution statistics from the data-distribution cache in shared
memory.

User 2 User 3 User 4 User 1

Shared-memory data-dictionary cache

sysdistrib
sysdistrib
tabid colno … endcat
120 1 tabid colno …
120 2 endcat

Figure 4-2. Data-distribution cache

The database server initially places pages for the sysdistrib system catalog table in
the buffer pool as it does all other data and index pages. However, the
data-distribution cache offers additional performance advantages. It:
v Is organized in a more efficient format
v Is organized to allow fast retrieval
v Bypasses the overhead of the buffer pool management
v Frees more pages in the buffer pool for actual data pages rather than system
catalog pages
v Reduces I/O operations to the system catalog table

Data-distribution configuration
The database server uses a hashing algorithm to store and locate information
within the data-distribution cache. The DS_POOLSIZE configuration parameter

Chapter 4. Effect of configuration on memory utilization 4-23

 controls the size of the data-distribution cache and controls the total number of
column distributions that can be stored in the data-distribution cache.

To modify the number of buckets in the data-distribution cache, use the

DS_HASHSIZE configuration parameter.

The following formula determines the number of column distributions that can be
stored in one bucket.
Distributions_per_bucket = DS_POOLSIZE / DS_HASHSIZE

To modify the number of distributions per bucket, change either the

DS_POOLSIZE or DS_HASHSIZE configuration parameter.

For example, with the default values of 127 for DS_POOLSIZE and 31 for
DS_HASHSIZE, you can potentially store distributions for about 127 columns in
the data-distribution cache. The cache has 31 hash buckets, and each hash bucket
can have an average of 4 entries.

The values that you set for DS_HASHSIZE and DS_POOLSIZE, depend on the
following factors:
v The number of columns for which you run the UPDATE STATISTICS statement
in HIGH or MEDIUM mode and you expect to be used most often in frequently
run queries.
If you do not specify columns when you run UPDATE STATISTICS for a table,
the database server generates distributions for all columns in the table.
You can use the values of DD_HASHSIZE and DD_HASHMAX as guidelines for
DS_HASHSIZE and DS_POOLSIZE. The DD_HASHSIZE and DD_HASHMAX
specify the size for the data-dictionary cache, which stores information and
statistics about tables that queries access.
For medium to large systems, you can start with the following values:
– DD_HASHSIZE 503
– DD_HASHMAX 4
– DS_HASHSIZE 503
– DS_POOLSIZE 2000
Monitor these caches by running the onstat -g dsc command to see the actual
usage, and you can adjust these parameters accordingly.
v The amount of memory available
The amount of memory that is required to store distributions for a column
depends on the level at which you run UPDATE STATISTICS. Distributions for a
single column might require between 1 KB and 2 MB, depending on whether
you specify medium or high mode or enter a finer resolution percentage when
you run UPDATE STATISTICS.

If the size of the data-distribution cache is too small, the following performance
problems can occur:
v The database server uses the DS_POOLSIZE value to determine when to remove
entries from the data-distribution cache. However, if the optimizer needs the
dropped distributions for another query, the database server must reaccess them
from the sysdistrib system catalog table on disk. The additional I/O and buffer
pool operations to access sysdistrib on disk adds to the total response time of
the query.
The database server tries to maintain the number of entries in data-distribution
cache at the DS_POOLSIZE value. If the total number of entries reaches within
4-24 IBM Informix Performance Guide
 an internal threshold of DS_POOLSIZE, the database server uses a least recently
used mechanism to remove entries from the data-distribution cache. The number
of entries in a hash bucket can go past this DS_POOLSIZE value, but the
database server eventually reduces the number of entries when memory
requirements drop.
v If DS_HASHSIZE is small and DS_POOLSIZE is large, overflow lists can be long
and require more search time in the cache.
Overflow occurs when a hash bucket already contains an entry. When multiple
distributions hash to the same bucket, the database server maintains an overflow
list to store and retrieve the distributions after the first one.
If DS_HASHSIZE and DS_POOLSIZE are approximately the same size, the
overflow lists might be smaller or even nonexistent, which might waste memory.
However, the amount of unused memory is insignificant overall.

You might want to change the values of the DS_HASHSIZE and DS_POOLSIZE
configuration parameters if you see the following situations:
v If the data-distribution cache is full most of the time and commonly used
columns are not listed in the distribution name field, try increasing the values
of the DS_HASHSIZE and DS_POOLSIZE configuration parameters.
v If the total number of entries is much lower than the value of the DS_POOLSIZE
configuration parameter, you can reduce the values of the DS_HASHSIZE and
DS_POOLSIZE configuration parameters.
Related reference:
DD_HASHSIZE configuration parameter (Administrator's Reference)
DD_HASHMAX configuration parameter (Administrator's Reference)
DS_POOLSIZE configuration parameter (Administrator's Reference)
onstat -g dsc command: Print distribution cache information (Administrator's
Reference)

Monitor and tune the SQL statement cache

The SQL statement cache stores optimized SQL statements so that multiple users
who run the same SQL statement can achieve some performance improvements.

These performance improvements are:

v Reduced response times because they bypass the optimization step, as Figure 4-3
on page 4-26 shows
v Reduced memory usage because the database server shares query data
structures among users

For more information about the effect of the SQL statement cache on the
performance of individual queries, see “Optimize queries with the SQL statement
cache” on page 13-40.

Figure 4-3 on page 4-26 shows how the database server accesses the SQL statement
cache for multiple users.
v When the database server runs an SQL statement for User 1 for the first time,
the database server checks whether the same exact SQL statement is in the SQL
statement cache. If it is not in the cache, the database server parses the
statement, determines the optimal query plan, and runs the statement.
v When User 2 runs the same SQL statement, the database server finds the
statement in the SQL statement cache and does not optimize the statement.

Chapter 4. Effect of configuration on memory utilization 4-25

 v Similarly, if User 3 and User 4 run the same SQL statement, the database server
does not optimize the statement. Instead, it uses the query plan in the SQL
statement cache in memory.

User 1 User 2 User 3 User 4

1. Check Cache 1. Check Cache

2. Parse 2. Parse

3. Execute 3. Optimize

4. Execute

SQL Statement Cache SCC Memory Pool

Query Structures Query Structures

Figure 4-3. Database server actions when using the SQL statement cache

Prepared statements and the statement cache

Prepared statements are inherently cached for a single session. This means that if a
prepared statement is executed many times or if a single cursor is opened many
times, the session uses the same prepared query plan.

If a session prepares a statement and then executes it many times, the SQL
statement cache does not affect performance, because the statement is optimized
just once during the PREPARE statement.

However, if other sessions also prepare that same statement, or if the first session
prepares the statement several times, the statement cache usually provides a direct
performance benefit, because the database server only calculates the query plan
once. Of course, the original session might gain a small benefit from the statement
cache, even if it only prepares the statement once, because other sessions use less
memory, and the database server does less work for the other sessions

4-26 IBM Informix Performance Guide

SQL statement cache configuration
The value of the STMT_CACHE configuration parameter enables or disables the
SQL statement cache.

For more information about how the value of the STMT_CACHE configuration
parameter enables the SQL statement cache, see “Enabling the SQL statement
cache” on page 13-42 describes.

Figure 4-4 shows how the database server uses the values of the pertinent
configuration parameters for the SQL statement cache. Further explanation follows
the figure.

User 1 User 4

A
A Allocate memory from Pools
1. Check Cache 1. Check Cache specified by parameter:
- Match - No Match - STMT_CACHE_NUMPOOL

B Before Add Entry, check parameters:

2. Parse 2. Parse
- STMT_CACHE_HITS
- STMT_CACHE_SIZE
B - STMT_CACHE_NOLIMIT
3. Execute 3. Optimize

SQL Statement Cache Pool

4. Execute
Query Structures

SQL Statement Cache

Full Entries on SSC LRU Queues:

Query Structures Query Structures

Query Structures Query Structures

Figure 4-4. Configuration parameters that affect the SQL statement cache

When the database server uses the SQL statement cache for a user, it means the
database server takes the following actions:
v Checks the SQL statement cache first for a match of the SQL statement that the
user is executing
v If the SQL statement matches an entry, executes the statement using the query
memory structures in the SQL statement cache (User 2 in Figure 4-4)
v If the SQL statement does not match an entry, the database server checks if it
qualifies for the cache.
For information about what qualifies an SQL statement for the cache, see SQL
statement cache qualifying criteria.

Chapter 4. Effect of configuration on memory utilization 4-27

 v If the SQL statement qualifies, inserts an entry into the cache for subsequent
executions of the statement.

The following parameters affect whether or not the database server inserts the SQL
statement into the cache (User 1 in Figure 4-4 on page 4-27):
v STMT_CACHE_HITS specifies the number of times the statement executes with
an entry in the cache (referred to as hit count). The database server inserts one of
the following entries, depending on the hit count:
– If the value of STMT_CACHE_HITS is 0, inserts a fully cached entry, which
contains the text of the SQL statement plus the query memory structures
– If the value of STMT_CACHE_HITS is not 0 and the statement does not exist
in the cache, inserts a key-only entry that contains the text of the SQL
statement. Subsequent executions of the SQL statement increment the hit
count.
– If the value of STMT_CACHE_HITS is equal to the number of hits for a
key-only entry, adds the query memory structures to make a fully cached
entry.
v STMT_CACHE_SIZE specifies the size of the SQL statement cache, and
STMT_CACHE_NOLIMIT specifies whether or not to limit the memory of the
cache to the value of STMT_CACHE_SIZE. If you do not specify the
STMT_CACHE_SIZE parameter, it defaults to 524288 (512 * 1024) bytes.
The default value for STMT_CACHE_NOLIMIT is 1, which means the database
server will insert entries into the SQL statement cache even though the total
amount of memory might exceed the value of STMT_CACHE_SIZE.
When STMT_CACHE_NOLIMIT is set to 0, the database server inserts the SQL
statement into the cache if the current size of the cache will not exceed the
memory limit.

The following sections on STMT_CACHE_HITS, STMT_CACHE_SIZE,

STMT_CACHE_NOLIMIT, STMT_CACHE_NUMPOOL and provide more details
on how the following configuration parameters affect the SQL statement cache and
reasons why you might want to change their default values.

Number of SQL statement executions

When the SQL statement cache is enabled, the database server inserts a qualified
SQL statement and its memory structures immediately in the SQL statement cache
by default.

If your workload has a disproportionate number of ad hoc queries, use the

STMT_CACHE_HITS configuration parameter to specify the number of times an
SQL statement is executed before the database server places a fully cached entry in
the statement cache.

When the STMT_CACHE_HITS configuration parameter is greater than 0 and the

number of times the SQL statement has been executed is less than
STMT_CACHE_HITS, the database server inserts key-only entries in the cache.
This specification minimizes unshared memory structures from occupying the
statement cache, which leaves more memory for SQL statements that applications
use often.

Monitor the number of hits on the SQL statement cache to determine if your
workload is using this cache effectively. The following sections describe ways to
monitor the SQL statement cache hits.
Related concepts:

4-28 IBM Informix Performance Guide

 “Too many single-use queries in the SQL statement cache” on page 4-32
Related reference:
STMT_CACHE_HITS configuration parameter (Administrator's Reference)

Monitoring the number of hits on the SQL statement cache:

To monitor the number of hits in the SQL statement cache, run the onstat -g ssc
command.

The onstat -g ssc command displays fully cached entries in the SQL statement
cache. Figure 4-5 shows sample output for onstat -g ssc.

onstat -g ssc

Statement Cache Summary:

#lrus currsize maxsize Poolsize
#hits nolimit
4 49456 524288 57344
0  1

Statement Cache Entries:

lru hash ref_cnt
hits flag heap_ptr database user

-----------------
------------------------------------------------------------
0 153 0
0 -F a7e4690 vjp_stores virginia
SELECT * FROM customer, orders
WHERE customer.customer_num = orders.customer_num

AND order_date > "01/01/07"

1 259 0
0  -F aa58c20 vjp_stores virginia

SELECT * FROM customer, orders
WHERE customer.customer_num = orders.customer_num

AND order_date > "01/01/2007"

2 232 0
1  DF aa3d020 vjp_stores virginia

SELECT C.customer_num, O.order_num
FROM customer C, orders O, items I
WHERE C.customer_num = O.customer_num

AND O.order_num = I.order_num

3 232 1
1  -F aa8b020 vjp_stores virginia

SELECT C.customer_num, O.order_num
FROM customer C, orders O, items I
WHERE C.customer_num = O.customer_num
AND O.order_num = I.order_num

Total number of entries: 4.

Figure 4-5. onstat -g ssc output

To monitor the number of times that the database server reads the SQL statement
within the cache, look at the following output columns:
v In the Statement Cache Summary portion of the onstat -g ssc output, the #hits
column is the value of the SQL_STMT_HITS configuration parameter.
In Figure 4-5, the #hits column in the Statement Cache Summary portion of the
output has a value of 0, which is the default value of the STMT_CACHE_HITS
configuration parameter.

Chapter 4. Effect of configuration on memory utilization 4-29

 Important: The database server uses entries in the SQL statement cache only if
the statements are exactly the same. The first two entries in Figure 4-5 on page
4-29 are not the same because each contains a different literal value in the
order_date filter.
v In the Statement Cache Entries portion of the onstat -g ssc output, the hits
column shows the number of times that the database server ran each individual
SQL statement from the cache. In other words, the column shows the number of
times that the database server uses the memory structures in the cache instead
of optimizing the statements to generate them again.
The first time that it inserts the statement in the cache, the hits value is 0.
– The first two SQL statements in Figure 4-5 on page 4-29 have a hits column
value of 0, which indicates that each statement is inserted into the cache but
not yet run from the cache.
– The last two SQL statements in Figure 4-5 on page 4-29 have a hits column
value of 1, which indicates that these statements ran once from the cache.
The hits value for individual entries indicates how much sharing of memory
structures is done. Higher values in the hits column indicate that the SQL
statement cache is useful in improving performance and memory usage.

For a complete description of the output fields that onstat -g ssc displays, see
“SQL statement cache information in onstat -g ssc output” on page 4-36.

Determining the number of nonshared entries in the SQL statement cache:

To determine how many nonshared entries exist in the SQL statement cache, run
onstat -g ssc all.

The onstat -g ssc all option displays the key-only entries in addition to the fully
cached entries in the SQL statement cache.

To determine how many nonshared entries exist in the cache:

1. Compare the onstat -g ssc all output with the onstat -g ssc output.
2. If the difference between these two outputs shows that many nonshared entries
exist in the SQL statement cache, increase the value of the STMT_CACHE_HITS
configuration parameter to allow more shared statements to reside in the cache
and reduce the management overhead of the SQL statement cache.

You can use one of the following methods to change the STMT_CACHE_HITS
parameter value:
v Update the ONCONFIG file to specify the STMT_CACHE_HITS configuration
parameter. You must restart the database server for the new value to take effect.
You can use a text editor to edit the ONCONFIG file. Then bring down the
database server with the onmode -ky command and restart with the oninit
command.
v Increase the STMT_CACHE_HITS configuration parameter dynamically while
the database server is running:
You can use any of the following methods to reset the STMT_CACHE_HITS
value at run time:
– Issue the onmode -W command. The following example specifies that three
(3) instances are required before a new query is added to the statement cache:
onmode -W STMT_CACHE_HITS 2

4-30 IBM Informix Performance Guide

 – Call the ADMIN or TASK function of the SQL administration API. The
following example is equivalent to the onmode command in the previous
example:
EXECUTE FUNCTION TASK("ONMODE", "W", "STMT_CACHE_HITS", "2");
If you increase STMT_CACHE_HITS dynamically without updating the
configuration file, and the database server is subsequently restarted, the
STMT_CACHE_HITS setting reverts the value in the ONCONFIG file. Therefore,
if you want the setting to persist after subsequent restarts, modify the
ONCONFIG file.

Monitoring and tuning the size of the SQL statement cache

If the size of the SQL statement cache is too small, performance problems can
occur. You can monitor the effectiveness of the size of the SQL statement cache.

The following performance problems can occur:

v Frequently executed SQL statements are not in the cache
The statements used most often should remain in the SQL statement cache. If the
SQL statement cache is not large enough, the database server might not have
enough room to keep these statements when other statements come into the
cache. For subsequent executions, the database server must reparse, reoptimize,
and reinsert the SQL statement into the cache. Try increasing
STMT_CACHE_SIZE.
v The database server spends a lot of time cleaning the SQL statement cache
The database server tries to prevent the SQL statement cache from allocating
large amounts of memory by using a threshold (70 percent of the
STMT_CACHE_SIZE parameter) to determine when to remove entries from the
SQL statement cache. If the new entry causes the size of the SQL statement
cache to exceed the threshold, the database server removes least recently used
entries (that are not currently in use) before inserting the new entry.
However, if a subsequent query needs the removed memory structures, the
database server must reparse and reoptimize the SQL statement. The additional
processing time to regenerate these memory structures adds to the total response
time of the query.

You can set the size of the SQL statement cache in memory with the
STMT_CACHE_SIZE configuration parameter. The value of the parameter is the
size in kilobytes. If STMT_CACHE_SIZE is not set, the default value is 512
kilobytes.

The onstat -g ssc output shows the value of STMT_CACHE_SIZE in the maxsize
column. In Figure 4-5 on page 4-29, this maxsize column has a value of 524288,
which is the default value (512 * 1024 = 524288).

Use the onstat -g ssc and onstat -g ssc all options to monitor the effectiveness of
size of the SQL statement cache. If you do not see cache entries for the SQL
statements that applications use most, the SQL statement cache might be too small
or too many unshared SQL statement occupy the cache. The following sections
describe how to determine these situations.
Related reference:
STMT_CACHE_SIZE configuration parameter (Administrator's Reference)

Chapter 4. Effect of configuration on memory utilization 4-31

 Changing the size of the SQL statement cache:

You can analyze onstat -g ssc all output to determine if the SQL statement cache is
too small. If the size of the cache is too small, you can change it.

To determine if the size of the SQL statement cache is too small:

1. Run onstat -g ssc all to determine if the cache is too small.
2. Look at the values in the following output columns in the Statement Cache
Entries portion of the onstat -g ssc all output:
v The flags column shows the current status of an SQL statement in the cache.
A value of F in the second position indicates that the statement is currently
fully cached.
A value of - in the second position indicates that only the statement text
(key-only entry) is in the cache. Entries with this - value in the second
position appear in the onstat -g ssc all output, but not in the onstat -g ssc
output.
v The hits column shows the number of times the SQL statement has been
executed, excluding the first time it is inserted into the cache.
If you do not see fully cached entries for statements that applications use most
and the value in the hits column is large for the entries that do occupy the
cache, then the SQL statement cache is too small.

To change the size of the SQL statement cache:

1. Update the value of the STMT_CACHE_SIZE configuration parameter.
2. Restart the database server for the new value to take effect.
Related reference:
STMT_CACHE_SIZE configuration parameter (Administrator's Reference)

Too many single-use queries in the SQL statement cache:

When the database server places many queries that are only used once in the
cache, they might replace statements that other applications use often. You can
view onstat -g ssc all output to determine if too many unshared SQL statements
occupy the cache. If so, you can prevent unshared SQL statements from being fully
cached.

Look at the values in the following output columns in the Statement Cache
Entries portion of the onstat -g ssc all output. If you see a lot of entries that have
both of the following values, too many unshared SQL statements occupy the cache:
v flags column value of F in the second position
A value of F in the second position indicates that the statement is currently fully
cached.
v hits column value of 0 or 1
The hits column shows the number of times the SQL statement has been
executed, excluding the first time it is inserted into the cache.

Increase the value of the STMT_CACHE_HITS configuration parameter to prevent

unshared SQL statements from being fully cached.
Related concepts:
“Number of SQL statement executions” on page 4-28
Related reference:

4-32 IBM Informix Performance Guide

 STMT_CACHE_HITS configuration parameter (Administrator's Reference)

Memory limit and size

Although the database server tries to clean the SQL statement cache, sometimes
entries cannot be removed because they are currently in use. In this case, the size
of the SQL statement cache can exceed the value of the STMT_CACHE_SIZE
configuration parameter.

The default value of the STMT_CACHE_NOLIMIT configuration parameter is 1,

which means the database server inserts the statement even though the current
size of the cache might be greater than the value of the STMT_CACHE_SIZE
parameter.

If the value of the STMT_CACHE_NOLIMIT configuration parameter is 0, the

database server does not insert either a fully-qualified or key-only entry into the
SQL statement cache if the size will exceed the value of STMT_CACHE_SIZE.

Use the onstat -g ssc option to monitor the current size of the SQL statement
cache. Look at the values in the following output columns of the onstat -g ssc
output:
v The currsize column shows the number of bytes currently allocated in the SQL
statement cache.
In Figure 4-5 on page 4-29, the currsize column has a value of 11264.
v The maxsize column shows the value of STMT_CACHE_SIZE.
In Figure 4-5 on page 4-29, the maxsize column has a value of 524288, which is
the default value (512 * 1024 = 524288).

When the SQL statement cache is full and users are currently executing all
statements within it, any new SQL statements that a user executes can cause the
SQL statement cache to grow beyond the size that STMT_CACHE_SIZE specifies.
When the database server is no longer using an SQL statement within the SQL
statement cache, it frees memory in the SQL statement cache until the size reaches
a threshold of STMT_CACHE_SIZE. However, if thousands of concurrent users are
executing several ad hoc queries, the SQL statement cache can grow very large
before any statements are removed. In such cases, take one of the following
actions:
v Set the STMT_CACHE_NOLIMIT configuration parameter to 0 to prevent
insertions when the cache size exceeds the value of the STMT_CACHE_SIZE
parameter.
v Set the STMT_CACHE_HITS parameter to a value greater than 0 to prevent
caching unshared SQL statements.

You can use one of the following methods to change the STMT_CACHE_NOLIMIT
configuration parameter value:
v Update the ONCONFIG file to specify the STMT_CACHE_NOLIMIT
configuration parameter. You must restart the database server for the new value
to take effect.
v Use the onmode -W command to override the STMT_CACHE_NOLIMIT
configuration parameter dynamically while the database server is running.
onmode -W STMT_CACHE_NOLIMIT 0
If you restart the database server, the value reverts the value in the ONCONFIG
file. Therefore, if you want the setting to remain for subsequent restarts, modify
the ONCONFIG file.

Chapter 4. Effect of configuration on memory utilization 4-33

 Related reference:
STMT_CACHE_HITS configuration parameter (Administrator's Reference)

Multiple SQL statement cache pools

Under some circumstances when the SQL statement cache is enabled, the database
server allocates memory from one pool for query structures.

These circumstances are:

v When the database server does not find a matching entry in the cache
v When the database server finds a matching key-only entry in the cache and the
hit count reaches the value of the STMT_CACHE_HITS configuration parameter

This one pool can become a bottleneck as the number of users increases. The
STMT_CACHE_NUMPOOL configuration parameter allows you to configure
multiple sscpools.

You can monitor the pools in the SQL statement cache to determine the following
situations:
v The number of SQL statement cache pools is sufficient for your workload.
v The size or limit of the SQL statement cache is not causing excessive memory
management.
Related reference:
STMT_CACHE_NUMPOOL configuration parameter (Administrator's
Reference)

Number of SQL statement cache pools:

When the SQL statement cache (SSC) is enabled, the database server allocates
memory from the SSC pool for unlinked SQL statements. The default value for the
STMT_CACHE_NUMPOOL configuration parameter is 1. As the number of users
increases, this one SSC pool might become a bottleneck.

The number of longspins on the SSC pool indicates whether or not the SSC pool is
a bottleneck.

Use the onstat -g spi option to monitor the number of longspins on an SSC pool.
The onstat -g spi command displays a list of the resources in the system for which
a wait was required before a latch on the resource could be obtained. During the
wait, the thread spins (or loops), trying to acquire the resource. The onstat -g spi
output displays the number of times a wait (Num Waits column) was required for
the resource and the number of total loops (Num Loops column). The onstat -g spi
output displays only resources that have at least one wait.

Figure 4-6 shows an excerpt of sample output for onstat -g spi. Figure 4-6 indicates
that no waits occurred for any SSC pool (the Name column does not list any SSC
pools).

Spin locks with waits:

Num Waits Num Loops Avg Loop/Wait Name
34477 387761 11.25 mtcb sleeping_lock
312 10205 32.71 mtcb vproc_list_lock

Figure 4-6. onstat -g spi output

4-34 IBM Informix Performance Guide

If you see an excessive number of longspins (Num Loops column) on an SSC pool,
increase the number of SSC pools in the STMT_CACHE_NUMPOOL configuration
parameter to improve performance.
Related reference:
STMT_CACHE_NUMPOOL configuration parameter (Administrator's
Reference)

Size of SQL statement cache pools and the current cache:

Use the onstat -g ssc pool option to monitor the usage of each SQL statement
cache (SSC) pool.

The onstat -g ssc pool command displays the size of each pool. The onstat -g ssc
option displays the cumulative size of the SQL statement cache in the currsize
column. This current size is the size of memory allocated from the SSC pools by
the statements that are inserted into the cache. Because not all statements that
allocate memory from the SSC pools are inserted into the cache, the current cache
size could be smaller than the total size of the SSC pools. Normally, the total size
of all SSC pools does not exceed the STMT_CACHE_SIZE value.

Figure 4-7 shows sample output for onstat -g ssc pool.

onstat -g ssc pool

Pool Summary:
name class addr totalsize freesize #allocfrag #freefrag
sscpool0 V a7e4020 57344 2352 52 7

Blkpool Summary:
name class addr size #blks

Figure 4-7. onstat -g ssc pool output

The Pool Summary section of the onstat -g ssc pool output lists the following
information for each pool in the cache.

Column Description
name The name of the SQL statement cache (SSC)
pool
class The shared-memory segment type in which
the pool has been created. For SSC pools, this
value is always “V” for the virtual portion of
shared-memory.
addr The shared-memory address of the SSC pool
structure
totalsize The total size, in bytes, of this SSC pool
freesize the number of free bytes in this SSC pool
#allocfrag The number of contiguous areas of memory
in this SSC pool that are allocated
#freefrag The number of contiguous areas of memory
that are not used in this SSC pool

Chapter 4. Effect of configuration on memory utilization 4-35

 The Blkpool Summary section of the onstat -g ssc pool output lists the following
information for all pools in the cache.

Column Description
name The name of the SSC pool
class The shared-memory segment type in which
the pool has been created. For SSC pools, this
value is always “V” for the virtual portion of
shared-memory.
addr The shared-memory address of the SSC pool
structure
totalsize The total size, in bytes, of this SSC pool
#blks The number of 8-kilobyte blocks that make
up all the SSC pools

SQL statement cache information in onstat -g ssc output

The onstat -g ssc command displays summary information for the SQL statement
cache.

The onstat -g ssc command displays the following information for the SQL
statement cache.
Table 4-5. SQL statement cache information in onstat -g ssc output
Column Description
#lrus The number of LRU queues. Multiple LRU
queues facilitate concurrent lookup and
insertion of cache entries.
currsize The number of bytes currently allocated to
entries in the SQL statement cache
maxsize The number of bytes specified in the
STMT_CACHE_SIZE configuration
parameter
poolsize The cumulative number of bytes for all pools
in the SQL statement cache. Use the onstat -g
ssc pool option to monitor individual pool
usage.
#hits Setting of the STMT_CACHE_HITS
configuration parameter, which specifies the
number of times that a query is executed
before it is inserted into the cache
nolimit Setting of STMT_CACHE_NOLIMIT
configuration parameter

The onstat -g ssc command lists the following information for each fully cached
entry in the cache. The onstat -g ssc all option lists the following information for
both the fully cached entries and key-only entries.

Column Description
lru The LRU identifier
hash The hash-bucket identifier

4-36 IBM Informix Performance Guide

Column Description
ref_cnt The number of sessions currently using this
statement
hits The number of times that users read the
query from the cache, excluding the first time
the statement entered the cache
flags Shows flag codes.

The flag codes for position 1 are:

D Indicates that the statement has
been dropped
A statement in the cache can be
dropped (not used any more) when
one of its dependencies has
changed. For example, when you
run UPDATE STATISTICS for the
table, the optimizer statistics might
change, making the query plan for
the SQL statement in the cache
obsolete. In this case, the database
server marks the statement as
dropped the next time that it tries to
use it.
- Indicates that the statement has not
been dropped

The flag codes for position 2 are:

F Indicates that the cache entry is
fully cached and contains the
memory structures for the query
- Indicates that the statement is not
fully cached
A statement is not fully cached
when the number of times the
statement has been executed is less
than the value of the
STMT_CACHE_HITS configuration
parameter. Entries with this - value
in the second position appear in the
onstat -g ssc all but not in the
onstat -g ssc output.
heap_ptr Pointer to the associated heap for the
statement
database Database against which the SQL statement is
executed
user User executing the SQL statement
statement Statement text as it would be used to test for
a match

Chapter 4. Effect of configuration on memory utilization 4-37

Session memory
The database server uses the virtual portion of shared memory mainly for user
sessions. The majority of the memory that each user session allocates is for SQL
statements. The amount of used memory can vary from one statement to another.
You can determine which session and which statements have high memory
utilization.

Use the following utility options to determine which session and prepared SQL
statements have high memory utilization:
v onstat -g mem
v onstat -g stm

The onstat -g mem option displays memory usage of all sessions. You can find the
session that is using the most memory by looking at the totalsize and freesize
output columns. The following figure shows sample output for onstat -g mem.
This sample output shows the memory utilization of three user sessions with the
values 14, 16, 17 in the names output column.

onstat -g mem

Pool Summary:
name class addr totalsize freesize #allocfrag #freefrag
...
14 V a974020 45056 11960 99 10
16 V a9ea020 90112 10608 159 5
17 V a973020 45056 11304 97 13
...
Blkpool Summary:
name class addr size #blks
mt V a235688 798720 19
global V a232800 0 0

Figure 4-8. onstat -g mem output

To display the memory allocated by each prepared statement, use the onstat -g stm
option. The following figure shows sample output for onstat -g stm.

onstat -g stm

session 25 --------------------------------------------------
sdblock heapsz statement ('*’ = Open cursor)
d36b018 9216 select sum(i) from t where i between -1 and ?
d378018 6240 *select tabname from systables where tabid=7
d36b114 8400 <SPL statement>

Figure 4-9. onstat -g stm output

The heapsz column in the output in Figure 4-9 shows the amount of memory used
by the statement. An asterisk (*) precedes the statement text if a cursor is open on
the statement. The output does not show the individual SQL statements in an SPL
routine.

To display the memory for only one session, specify the session ID in the onstat -g
stm option. For an example, see “Monitor session memory with onstat -g mem
and onstat -g stm output” on page 13-52.

4-38 IBM Informix Performance Guide

 Related tasks:
“Estimating the size of the virtual portion of shared memory” on page 4-4

Data-replication buffers and memory utilization

Data replication requires two instances of the database server, a primary one and a
secondary one, running on two computers. If you implement data replication for
your database server, the database server holds logical-log records in the
data-replication buffer before it sends them to the secondary database server.

The data-replication buffer is always the same size as the logical-log buffer.

Memory latches
The database server uses latches to control access to shared memory structures
such as the buffer pool or the memory pools for the SQL statement cache. You can
obtain statistics on latch use and information about specific latches. These statistics
provide a measure of the system activity.

The statistics include the number of times that threads waited to obtain a latch. A
large number of latch waits typically results from a high volume of processing
activity in which the database server is logging most of the transactions.

Information about specific latches includes a listing of all the latches that are held
by a thread and any threads that are waiting for latches. This information allows
you to locate any specific resource contentions that exist.

You, as the database administrator, cannot configure or tune the number of latches.
However, you can increase the number of memory structures on which the
database server places latches to reduce the number of latch waits. For example,
you can tune the number of SQL statement cache memory pools or the number of
SQL statement cache LRU queues. For more information, see “Multiple SQL
statement cache pools” on page 4-34.

Warning: Never stop a database server process that is holding a latch. If you do,
the database server immediately initiates an abort.

Monitoring latches with command-line utilities

You can obtain information about latches by running onstat -p or onstat -s.

Monitoring latches with onstat -p

Run onstat -p to obtain the values in the lchwaits field. This field stores the
number of times that a thread was required to wait for a shared-memory latch.

Figure 4-10 shows an excerpt of sample onstat -p output that shows the lchwaits
field.

...
ixda-RA idx-RA da-RA logrec-RA RA-pgsused
lchwaits
5 0 204 0 148
12

Figure 4-10. Partial onstat -p output showing the lchwaits field

Related reference:

Chapter 4. Effect of configuration on memory utilization 4-39

 onstat -p command: Print profile counts (Administrator's Reference)

Monitoring latches with onstat -s

Run onstat -s to obtain general latch information. The output includes the
userthread column, which lists the address of any user thread that is waiting for a
latch.

You can compare this address with the user addresses in the onstat -u output to
obtain the user-process identification number.

Figure 4-11 shows sample onstat -s output.

...
Latches with lock or userthread set
name address lock wait userthread
LRU1 402e90 0 0 6b29d8
bf[34] 4467c0 0 0 6b29d8
...

Figure 4-11. onstat -s output

Monitoring latches with SMI tables

You can query the sysprofile SMI table to obtain the number of times a thread
waited for a latch.

The latchwts column of the sysprofile table contains the number of times that a
thread waited for a latch.

Encrypted values
An encrypted value uses more storage space than the corresponding plain text
value because all of the information needed to decrypt the value except the
encryption key is stored with the value.

Omitting the hint used with the password can reduce encryption overhead by up
to 50 bytes. If you are using encrypted values, you must make sure that you have
sufficient space available for the values.

Note: Embedding zero bytes in the encrypted result is not recommended.

Related concepts:
Column-level encryption (Security Guide)
Related reference:
Calculating storage requirements for encrypted data (SQL Syntax)

4-40 IBM Informix Performance Guide

Chapter 5. Effect of configuration on I/O activity
The configuration of your database server affects I/O activity.

The following factors affect I/O activity:

v The assignment of chunks and dbspaces can create I/O hot spots, or disk
partitions with a disproportionate amount of I/O activity.
v Your allocation of critical data, sort areas, and areas for temporary files and
index builds can place intermittent loads on various disks.
v How you configure read-ahead can increase the effectiveness of individual I/O
operations.
v How you configure the background I/O tasks, such as logging and page
cleaning, can affect I/O throughput.

Chunk and dbspace configuration

The number of disks that you use and the configuration of your chunks, dbspaces,
and blobspaces affect the performance of your database server. You can improve
performance by planning disk use and the configuration of chunks, dbspaces, and
blobspaces.

All the data that resides in a database is stored on disk. The speed at which the
database server can copy the appropriate data pages to and from disk determines
how well your application performs.

All the data that resides in a database is stored on disk. The Optical Subsystem
also uses a magnetic disk to access TEXT or BYTE data that is retrieved from
optical media. The speed at which the database server can copy the appropriate
data pages to and from disk determines how well your application performs.

Disks are typically the slowest component in the I/O path for a transaction or
query that runs entirely on one host computer. Network communication can also
introduce delays in client/server applications, but these delays are typically
outside the control of the database server administrator. For information about
actions that the database server administrator can take to improve network
communications, see “Network buffer pools” on page 3-17 and “Connections and
CPU utilization” on page 3-25.

Disks can become overused or saturated when users request pages too often.
Saturation can occur in the following situations:
v You use a disk for multiple purposes, such as for both logging and active
database tables.
v Disparate data resides on the same disk.
v Table extents become interleaved.

The various functions that your application requires, as well as the

consistency-control functions that the database server performs, determine the
optimal disk, chunk, and dbspace layout for your application. The more disks that
you make available to the database server, the easier it is to balance I/O across
them. For more information about these factors, see Chapter 6, “Table performance
considerations,” on page 6-1.

© Copyright IBM Corp. 1996, 2014 5-1

 This section outlines important issues for the initial configuration of your chunks,
dbspaces, and blobspaces. Consider the following issues when you decide how to
lay out chunks and dbspaces on disks:
v Placement and mirroring of critical data
v Load balancing
v Reduction of contention
v Ease of backup and restore

Together with round-robin fragmentation, you can balance chunks over disks and
controllers, saving time and handling errors. Placing multiple chunks on a single
disk can improve throughput.

Associate disk partitions with chunks

You should assign chunks to entire disk partitions. When a chunk coincides with a
disk partition (or device), it is easy to track disk-space use, and you avoid errors
caused by miscalculated offsets.

The maximum size for a chunk is 4 terabytes.

Associate dbspaces with chunks

You should associate a single chunk with a dbspace, especially when that dbspace
is to be used for a table fragment.

For more information about table placement and layout, see Chapter 6, “Table
performance considerations,” on page 6-1.

Placing system catalog tables with database tables

When a disk that contains the system catalog for a particular database fails, the
entire database remains inaccessible until the system catalog is restored. Because of
this potential inaccessibility, do not cluster the system catalog tables for all
databases in a single dbspace. Instead place the system catalog tables with the
database tables that they describe.

To create a system catalog table in the table dbspace:

1. Create a database in the dbspace in which the table is to reside.
2. Use the SQL statements DATABASE or CONNECT to make that database the
current database.
3. Enter the CREATE TABLE statement to create the table.

I/O for cooked files for dbspace chunks

On UNIX, you can control the use of direct I/O for cooked files used for dbspace
chunks.

On UNIX, you can allocate disk space in two ways:

v Use files that are buffered through the operating system. These files are often
called cooked files.
v Use unbuffered disk access, also called raw disk space.

When dbspaces reside on raw disk devices (also called character-special devices), the
database server uses unbuffered disk access. A raw disk directly transfers data
between the database server memory and disk without also copying data.

5-2 IBM Informix Performance Guide

 While you should generally use raw disk devices on UNIX systems to achieve
better performance, you might prefer to use cooked files, which are easier to
allocate and manage than raw devices. If you use cooked files, you might be able
to get better performance by enabling the Informix direct I/O option.

In addition, Informix supports a separate concurrent I/O option on AIX operating

systems. If you enable concurrent I/O on AIX, you get both unbuffered I/O and
concurrent I/O. With concurrent I/O, writing to two parts of a file can occur
concurrently. (On some other operating systems and file systems, enabling direct
I/O also enables concurrent I/O as part of the same file system direct I/O feature.)

To determine the best performance, perform benchmark testing for the dbspace
and table layout on your system.

Direct I/O (UNIX)

On UNIX, you can use direct I/O to improve the performance of cooked files.
Direct I/O can be beneficial because it avoids file system buffering. Because direct
I/O uses unbuffered I/O, it is more efficient for reads and writes that go to disk
(as opposed to those reads and writes that merely access the file system buffers).

Direct I/O generally requires data to be aligned on disk sector boundaries.

Direct I/O also allows the use of kernel asynchronous I/O (KAIO), which can
further improve performance. By using direct I/O and KAIO where available, the
performance of cooked files used for dbspace chunks can approach the
performance of raw devices.

If your file system supports direct I/O for the page size used for the dbspace
chunk, the database server operates as follows:
v Does not use direct I/O by default.
v Uses direct I/O if the DIRECT_IO configuration parameter is set to 1.
v Uses KAIO (if the file system supports it) with direct I/O by default.
v Does not use KAIO with direct I/O if the environment variable KAIOOFF is set.

If Informix uses direct I/O for a chunk, and another program tries to open the
chunk file without using direct I/O, the open will normally succeed, but there can
be a performance penalty. The penalty can occur because the file system attempts
to ensure that each open sees the same file data, either by switching to buffered
I/O and not using direct I/O for the duration of the conflicting open, or by
flushing the file system cache before each direct I/O operation and invalidating the
file system cache after each direct write.

Informix does not use direct I/O for temporary dbspaces.

Related reference:
DIRECT_IO configuration parameter (UNIX) (Administrator's Reference)

Direct I/O (Windows)

Direct I/O is used for dbspace chunks on Windows platforms regardless of the
value of the DIRECT_IO configuration parameter.

Concurrent I/O (AIX only)

On AIX operating systems, you can use concurrent I/O in addition to direct I/O
for chunks that use cooked files. Concurrent I/O can improve performance,

Chapter 5. Effect of configuration on I/O activity 5-3

 because it allows multiple reads and writes to a file to occur concurrently, without
the usual serialization of noncompeting read and write operations.

Concurrent I/O can be especially beneficial when you have data in a single chunk
file striped across multiple disks.

Concurrent I/O, which you enable by setting the DIRECT_IO configuration

parameter to 2, includes the benefit of avoiding file system buffering and is subject
to the same limitations and use of KAIO as occurs if you use direct I/O without
concurrent I/O. Thus, when concurrent I/O is enabled, you get both unbuffered
I/O and concurrent I/O.

If Informix uses concurrent I/O for a chunk, and another program (such as an
external backup program) tries to open the same chunk file without using
concurrent I/O, the open operation will fail.

Informix does not use direct or concurrent I/O for cooked files used for temporary
dbspace chunks.
Related reference:
DIRECT_IO configuration parameter (UNIX) (Administrator's Reference)

Enabling the direct I/O or concurrent I/O option (UNIX)

Use the DIRECT_IO configuration parameter to enable the direct I/O option on
UNIX or the concurrent I/O option on AIX.

Prerequisites:
v You must log on as user root or informix.
v Direct I/O or concurrent I/O must be available and the file system must
support direct I/O for the page size used for the dbspace chunk.

To enable direct I/O, set the DIRECT_IO configuration parameter to 1.

To enable concurrent I/O with direct I/O on AIX operating systems, set the
DIRECT_IO configuration parameter to 2.

If you do not want to enable direct I/O or concurrent I/O, set the DIRECT_IO
configuration parameter to 0.
Related reference:
DIRECT_IO configuration parameter (UNIX) (Administrator's Reference)

Confirming the use of the direct or concurrent I/O option

(UNIX)
You can confirm and monitor the use of direct I/O or concurrent I/O (on AIX) for
cooked file chunks.

You can confirm the use of direct I/O or concurrent I/O by:
v Displaying onstat -d information.
The onstat -d command displays information that includes a flag that identifies
whether direct I/O, concurrent I/O (on AIX), or neither is used for cooked file
chunks.
v Verifying that the DIRECT_IO configuration parameter is set to 1 (for direct I/O)
or 2 (for concurrent I/O).

5-4 IBM Informix Performance Guide

 Related reference:
DIRECT_IO configuration parameter (UNIX) (Administrator's Reference)
onstat -d command: Print chunk information (Administrator's Reference)

Placement of critical data

The disk or disks that contain the system reserved pages, the physical log, and the
dbspaces that contain the logical-log files are critical to the operation of the
database server. The database server cannot operate if any of these elements
becomes unavailable. By default, the database server places all three critical
elements in the root dbspace.

To arrive at an appropriate placement strategy for critical data, you must make a
trade-off between the availability of data and maximum logging performance.

The database server also places temporary table and sort files in the root dbspace
by default. You should use the DBSPACETEMP configuration parameter and the
DBSPACETEMP environment variable to assign these tables and files to other
dbspaces. For details, see “Configure dbspaces for temporary tables and sort files”
on page 5-8.

Consider separate disks for critical data components

If you place the root dbspace, logical log, and physical log in separate dbspaces on
separate disks, you can obtain some distinct performance advantages. The disks
that you use for each critical data component should be on separate controllers.

This approach has the following advantages:

v Isolates logging activity from database I/O and allows physical-log I/O requests
to be serviced in parallel with logical-log I/O requests
v Reduces the time that you need to recover from a failure
However, unless the disks are mirrored, there is an increased risk that a disk
that contains critical data might be affected in the event of a failure, which will
bring the database server to a halt and require the complete restoration of all
data from a level-0 backup.
v Allows for a relatively small root dbspace that contains only reserved pages, the
database partition, and the sysmaster database
In many cases, 10,000 kilobytes is sufficient.

The database server uses different methods to configure various portions of critical
data. To assign an appropriate dbspace for the root dbspace and physical log, set
the appropriate database server configuration parameters. To assign the logical-log
files to an appropriate dbspace, use the onparams utility.

For more information about the configuration parameters that affect each portion
of critical data, see “Configuration parameters that affect critical data” on page 5-8.

Consider mirroring for critical data components

Consider mirroring for the dbspaces that contain critical data. Mirroring these
dbspaces ensures that the database server can continue to operate even when a
single disk fails.

Chapter 5. Effect of configuration on I/O activity 5-5

 However, depending on the mix of I/O requests for a given dbspace, a trade-off
exists between the fault tolerance of mirroring and I/O performance. You obtain a
marked performance advantage when you mirror dbspaces that have a
read-intensive usage pattern and a slight performance disadvantage when you
mirror write-intensive dbspaces.

Most modern storage devices have excellent mirroring capabilities, and you can
use those devices instead of the mirroring capabilities of the database server.

When mirroring is in effect, two disks are available to handle read requests, and
the database server can process a higher volume of those requests. However, each
write request requires two physical write operations and does not complete until
both physical operations are performed. The write operations are performed in
parallel, but the request does not complete until the slower of the two disks
performs the update. Thus, you experience a slight performance penalty when you
mirror write-intensive dbspaces.

Consider mirroring the root dbspace

You can achieve a certain degree of fault tolerance with a minimum performance
penalty if you mirror the root dbspace and restrict its contents to read-only or
seldom-accessed tables.

When you place update-intensive tables in other, nonmirrored dbspaces, you can
use the database server backup-and-restore facilities to perform warm restores of
those tables in the event of a disk failure. When the root dbspace is mirrored, the
database server remains online to service other transactions while the failed disk is
being repaired.

When you mirror the root dbspace, always place the first chunk on a different
device than that of the mirror. The MIRRORPATH configuration parameter should
have a different value than ROOTPATH.
Related reference:
MIRRORPATH configuration parameter (Administrator's Reference)
ROOTPATH configuration parameter (Administrator's Reference)

Consider mirroring smart-large-object chunks

You can achieve higher availability and faster access if you mirror chunks that
contain metadata pages.

An sbspace is a logical storage unit composed of one or more chunks that store
smart large objects, which consist of CLOB (character large object) or BLOB (binary
large object) data.

The first chunk of an sbspace contains a special set of pages, called metadata, which
is used to locate smart large objects in the sbspace. Additional chunks that are
added to the sbspace can also have metadata pages if you specify them on the
onspaces command when you create the chunk.

Consider mirroring chunks that contain metadata pages for the following reasons:
v Higher availability
Without access to the metadata pages, users cannot access any smart large
objects in the sbspace. If the first chunk of the sbspace contains all of the
metadata pages and the disk that contains that chunk becomes unavailable, you
cannot access a smart large object in the sbspace, even if it resides on a chunk on

5-6 IBM Informix Performance Guide

 another disk. For high availability, mirror at least the first chunk of the sbspace
and any other chunk that contains metadata pages.
v Faster access
By mirroring the chunk that contains the metadata pages, you can spread read
activity across the disks that contain the primary chunk and mirror chunk.
Related concepts:
Sbspaces (Administrator's Guide)

Mirroring and its effect on the logical log

The logical log is write intensive. If the dbspace that contains the logical-log files is
mirrored, you encounter a slight double-write performance penalty. However, you
can adjust the rate at which logging generates I/O requests to a certain extent by
choosing an appropriate log buffer size and logging mode.

For details on the slight double-write performance penalty, see “Consider

mirroring for critical data components” on page 5-5.

With unbuffered and ANSI-compliant logging, the database server requests a flush
of the log buffer to disk for every committed transaction (two when the dbspace is
mirrored). Buffered logging generates far fewer I/O requests than unbuffered or
ANSI-compliant logging.

With buffered logging, the log buffer is written to disk only when it fills and all
the transactions that it contains are completed. You can reduce the frequency of
logical-log I/O even more if you increase the size of your logical-log buffers.
However, buffered logging leaves transactions in any partially filled buffers
vulnerable to loss in the event of a system failure.

Although database consistency is guaranteed under buffered logging, specific

transactions are not guaranteed against a failure. The larger the logical-log buffers,
the more transactions you might need to reenter when service is restored after a
failure.

Unlike the physical log, you cannot specify an alternative dbspace for logical-log
files in your initial database server configuration. Instead, use the onparams utility
first to add logical-log files to an alternative dbspace and then drop logical-log files
from the root dbspace.
Related reference:
The onparams Utility (Administrator's Reference)

Mirroring and its effect on the physical log

The physical log is write intensive, with activity occurring at checkpoints and
when buffered data pages are flushed. I/O to the physical log also occurs when a
page-cleaner thread is activated. If the dbspace that contains the physical log is
mirrored, you encounter a slight double-write performance penalty.

For details on the slight double-write performance penalty, see “Consider

mirroring for critical data components” on page 5-5.

To keep I/O to the physical log at a minimum, you can adjust the checkpoint
interval and the LRU minimum and maximum thresholds. (See “CKPTINTVL and
its effect on checkpoints” on page 5-31 and “BUFFERPOOL and its effect on page
cleaning” on page 5-40.)

Chapter 5. Effect of configuration on I/O activity 5-7

Configuration parameters that affect critical data
The configuration parameters that configure the root dbspace and the logical and
physical logs affect critical data.

You can use the following configuration parameters to configure the root dbspace:
v ROOTNAME
v ROOTOFFSET
v ROOTPATH
v ROOTSIZE
v MIRROR
v MIRRORPATH
v MIRROROFFSET

These parameters determine the location and size of the initial chunk of the root
dbspace and configure mirroring, if any, for that chunk. (If the initial chunk is
mirrored, all other chunks in the root dbspace must also be mirrored). Otherwise,
these parameters have no major impact on performance.

The following configuration parameters affect the logical logs:

v LOGSIZE
v LOGBUFF

LOGSIZE determines the size of each logical-log files. LOGBUFF determines the
size of the three logical-log buffers that are in shared memory. For more
information about LOGBUFF, see “The LOGBUFF configuration parameter and
memory utilization” on page 4-15.

The following configuration parameters determine the location and size of the
physical log:
v PHYSDBS
v PHYSFILE

Configure dbspaces for temporary tables and sort files

Applications that use temporary tables or large sort operations require a large
amount of temporary space. To improve performance of these applications, use the
DBSPACETEMP configuration parameter or the DBSPACETEMP environment
variable to designate one or more dbspaces for temporary tables and sort files.

Depending on how the temporary space is created, the database server uses the
following default locations for temporary table and sort files when you do not set
DBSPACETEMP:
v The dbspace of the current database, when you create an explicit temporary
table with the TEMP TABLE clause of the CREATE TABLE statement and do not
specify a dbspace for the table either in the IN dbspace clause or in the
FRAGMENT BY clause
This action can severely affect I/O to that dbspace. If the root dbspace is
mirrored, you encounter a slight double-write performance penalty for I/O to
the temporary tables and sort files.
v The root dbspace when you create an explicit temporary table with the INTO
TEMP option of the SELECT statement

5-8 IBM Informix Performance Guide

 This action can severely affect I/O to the root dbspace. If the root dbspace is
mirrored, you encounter a slight double-write performance penalty for I/O to
the temporary tables and sort files.
v The operating-system directory or file that you specify in one of the following
variables:
– In UNIX, the operating-system directory or directories that the
PSORT_DBTEMP environment variable specifies, if it is set
If PSORT_DBTEMP is not set, the database server writes sort files to the
operating-system file space in the /tmp directory.
– In Windows, the directory specified in TEMP or TMP in the User
Environment Variables window on Control Panel > System.
The database server uses the operating-system directory or files to direct any
overflow that results from the following database operations:
– SELECT statement with GROUP BY clause
– SELECT statement with ORDER BY clause
– Hash-join operation
– Nested-loop join operation
– Index builds

Warning: If you do not specify a value for the DBSPACETEMP configuration

parameter or the DBSPACETEMP environment variable, the database server uses
this operating-system file for implicit temporary tables. If this file system has
insufficient space to hold a sort file, the query that performs the sort returns an
error. Meanwhile, the operating system might be severely impacted until you
remove the sort file.

You can improve performance with the use of temporary dbspaces that you create
exclusively to store temporary tables and sort files. Use the DBSPACETEMP
configuration parameter and the DBSPACETEMP environment variable to assign
these tables and files to temporary dbspaces.

When you specify dbspaces in either the DBSPACETEMP configuration parameter

or the DBSPACETEMP environment variable, you gain the following performance
advantages:
v Reduced I/O impact on the root dbspace, production dbspaces, or
operating-system files
v Use of parallel sorts into the temporary files (to process query clauses such as
ORDER BY or GROUP BY, or to sort index keys when you execute CREATE
INDEX) when you specify more than one dbspace for temporary tables and
PDQ priority is set to greater than 0.
v Improved speed with which the database server creates temporary tables when
you assign two or more temporary dbspaces on separate disks
v Automatic fragmentation of the temporary tables across dbspaces when
SELECT....INTO TEMP statements are run

The following table shows statements that create temporary tables and information
about where the temporary tables are created.

Chapter 5. Effect of configuration on I/O activity 5-9

Statement That
Creates Temporary FRAGMENT BY Where Temp Table
Table Database Logged WITH NO LOG clause clause Created
CREATE TEMP Yes No No Root dbspace
TABLE
CREATE TEMP Yes Yes No One of dbspaces that
TABLE are specified in
DBSPACETEMP
CREATE TEMP Yes No Yes Cannot create temp
TABLE table. Error 229/196
SELECT ..INTO TEMP Yes Yes No Fragmented by
round-robin only in
the non-logged
dbspaces that are
specified in
DBSPACETEMP

Important: Use the DBSPACETEMP configuration parameter or the

DBSPACETEMP environment variable for better performance of sort operations
and to prevent the database server from unexpectedly filling file systems. The
dbspaces that you list must be composed of chunks that are allocated as
unbuffered devices.
Related concepts:
“Specify temporary tables in the DBSPACETEMP configuration parameter” on
page 5-11
Related reference:
DBSPACETEMP configuration parameter (Administrator's Reference)
CREATE TEMP TABLE statement (SQL Syntax)
INTO TEMP clause (SQL Syntax)

Creating temporary dbspaces

You can create a dbspace for the exclusive use of temporary tables and sort files.
The database server does not perform logical or physical logging of temporary
dbspaces, and temporary dbspaces are never backed up as part of a full-system
backup.

To create a dbspace for the exclusive use of temporary tables and sort files, use
onspaces -t. For best performance, use the following guidelines:
v If you create more than one temporary dbspace, create each dbspace on a
separate disk to balance the I/O impact.
v Place no more than one temporary dbspace on a single disk.

You cannot mirror a temporary dbspace that you create with onspaces -t.

Important: In the case of a database with logging, you must include the WITH NO
LOG clause in the SELECT... INTO TEMP statement to place the explicit temporary
tables in the dbspaces listed in the DBSPACETEMP configuration parameter and
the DBSPACETEMP environment variable. Otherwise, the database server stores
the explicit temporary tables in the root dbspace.
Related reference:

5-10 IBM Informix Performance Guide

 DBSPACETEMP configuration parameter (Administrator's Reference)
create tempdbspace argument: Create a temporary dbspace (SQL
administration API) (Administrator's Reference)
onspaces -c -d: Create a dbspace (Administrator's Reference)

Specify temporary tables in the DBSPACETEMP configuration

parameter
The DBSPACETEMP configuration parameter specifies a list of dbspaces in which
the database server places temporary tables and sort files by default. Some or all of
the dbspaces that you list in this configuration parameter can be temporary
dbspaces, which are reserved exclusively to store temporary tables and sort files.

If the database server inserts data into a temporary table through a SELECT INTO
TEMP operation that creates the TEMP table, that temporary table uses
round-robin distributed storage. Its fragments are created in the temporary
dbspaces that are listed in the DBSPACETEMP configuration parameter or in the
DBSPACETEMP environment variable. For example, the following query uses
round-robin distributed storage:
SELECT col1 FROM tab1
INTO TEMP temptab1 WITH NO LOG;

The DBSPACETEMP configuration parameter lets the database administrator

restrict which dbspaces the database server uses for temporary storage.

Important: The DBSPACETEMP configuration parameter is not set in the

onconfig.std file. For best performance with temporary tables and sort files, use
DBSPACETEMP to specify two or more dbspaces on separate disks.
Tips:
v If you work on a small system with a limited number of disks and
cannot place temporary dbspaces on different disk drives, you might
consider using 1 (or possibly 2) temporary dbspaces. This can reduce the
logging that is associated with the temporary dbspaces.
v If you have many disk drives, you can parallelize many operations (such
as sorts, joins, and temporary tables) without having multiple temporary
dbspaces. The number of temporary dbspaces that you have relates to
how much you want to spread the I/O out. A good starting place is 4
temporary dbspaces. If you create too many small temporary dbspaces,
you will not have enough space for nonparallel creation of large objects.
Related concepts:
“Configure dbspaces for temporary tables and sort files” on page 5-8
“Distribution schemes” on page 9-6
Related reference:
DBSPACETEMP configuration parameter (Administrator's Reference)
CREATE TEMP TABLE statement (SQL Syntax)

Override the DBSPACETEMP configuration parameter for a

session
To override the DBSPACETEMP configuration parameter, you can use the
DBSPACETEMP environment variable for both temporary tables and sort files.

Chapter 5. Effect of configuration on I/O activity 5-11

 This environment variable specifies a comma- or colon-separated list of dbspaces
in which to place temporary tables for the current session.

Important: Use the DBSPACETEMP configuration parameter or the

DBSPACETEMP environment variable for better performance of sort operations
and to prevent the database server from unexpectedly filling file systems.

You should use DBSPACETEMP rather than the PSORT_DBTEMP environment

variable to specify sort files for the following reasons:
v DBSPACETEMP typically yields better performance.
When dbspaces reside on character-special devices (also known as raw disk
devices), the database server uses unbuffered disk access. I/O is faster to
unbuffered devices than to regular (buffered) operating-system files because the
database server manages the I/O operation directly.
v PSORT_DBTEMP specifies one or more operating-system directories in which to
place sort files.
These operating-system files can unexpectedly fill on your computer because the
database server does not manage them.

Estimating temporary space for dbspaces and hash joins

You can estimate and increase the amount of temporary space for dbspaces and for
hash joins. If you do this, you can prevent the possible overflow of memory to
temporary space on disk.

You can use the following guidelines to estimate the amount of temporary space to
allocate:
v For OLTP applications, allocate temporary dbspaces that equal at least 10 percent
of the table.
v For DSS applications, allocate temporary dbspaces that equal at least 50 percent
of the table.

A hash join, which works by building a table (the hash table) from the rows in one
of the tables in a join, and then probing it with rows from the other table, can use
a significant amount of memory and can potentially overflow to temporary space
on disk. The hash table size is governed by the size of the table used to build the
hash table (which is often the smaller of the two tables in the join), after applying
any filters, which can reduce the number of rows and possibly reduce the number
of columns.

Hash-join partitions are organized into pages. Each page has a header. The header
and tuples are larger in databases on 64-bit platforms than in builds on 32-bit
platforms. The size of each page is the base page size (2K or 4K depending on
system) unless a single row needs more space. If you need more space, you can
add bytes to the length of your rows.

You can use the following formula to estimate the amount of memory that is
required for the hash table in a hash join:
hash_table_size = (32 bytes + row_size_smalltab) * num_rows_smalltab

where row_size_smalltab and num_rows_smalltab refer to the row size and the
number of rows, respectively, in the smaller of the two tables participating in the
hash join.

5-12 IBM Informix Performance Guide

 For example, suppose you have a page head that is 80 bytes in length and a row
header that is 48 bytes in length. Because each row must be aligned to 8 bytes, you
might need to add up to 7 bytes to the row length, as shown in these formulas:
per_row_size = 48 bytes + rowsize + mod(rowsize, 8)
page_size = base_page_size (2K or 4K)
rows_per_page = round_down_to_integer((page_size - 80 bytes) / per_row_size)

If the value of rows_per_page is less than one, increase the page_size value to the
smallest multiple of the base_page_size, as shown in this formula:
size = (numrows_smalltab / rows_per_page) * page_size

You can use the DS_NONPDQ_QUERY_MEM configuration parameter to

configure sort memory for all queries except PDQ queries. Its setting has no effect,
however, if the PDQ priority setting is greater than zero.

For more information, see “Hash join” on page 10-3 and “Configuring memory for
queries with hash joins, aggregates, and other memory-intensive elements” on
page 13-34.
Related reference:
DS_NONPDQ_QUERY_MEM configuration parameter (Administrator's
Reference)

PSORT_NPROCS environment variable

The PSORT_NPROCS environment variable specifies the maximum number of
threads that the database server can use to sort a query. When a query involves a
large sort operation, multiple sort threads can execute in parallel to improve the
performance of the query.

When the value of PDQ priority is 0 and PSORT_NPROCS is greater than 1, the
database server uses parallel sorts. The management of PDQ does not limit these
sorts. In other words, although the sort is executed in parallel, the database server
does not regard sorting as a PDQ activity. When PDQ priority is 0, the database
server does not control sorting by any of the PDQ configuration parameters.

When PDQ priority is greater than 0 and PSORT_NPROCS is greater than 1, the
query benefits both from parallel sorts and from PDQ features such as parallel
scans and additional memory. Users can use the PDQPRIORITY environment
variable to request a specific proportion of PDQ resources for a query. You can use
the MAX_PDQPRIORITY parameter to limit the number of such user requests. For
more information about MAX_PDQPRIORITY, see “Limiting PDQ resources in
queries” on page 3-11.

The database server allocates a relatively small amount of memory for sorting, and
that memory is divided among the PSORT_NPROCS sort threads. Sort processes
use temporary space on disk when not enough memory is allocated. For more
information about memory allocated for sorting, see “Estimating memory needed
for sorting” on page 7-19.

Important: For better performance for a sort operation, set PSORT_NPROCS

initially to 2 if your computer has multiple CPUs. If the subsequent CPU activity is
lower than I/O activity, you can increase the value of PSORT_NPROCS.

For more information about sorts during index builds, see “Improving
performance for index builds” on page 7-18.

Chapter 5. Effect of configuration on I/O activity 5-13

Configure sbspaces for temporary smart large objects
Applications can use temporary smart large objects for text, image, or other
user-defined data types that are only required during the life of the user session.
These applications do not require logging of the temporary smart large objects.
Logging adds I/O activity to the logical log and increases memory utilization.

You can store temporary smart large objects in a permanent sbspace or a

temporary sbspace.
v Permanent sbspaces
If you store the temporary smart large objects in a regular sbspace and keep the
default no logging attribute, changes to the objects are not logged, but the
metadata is always logged.
v Temporary sbspaces
Applications that update temporary smart large objects stored in temporary
sbspaces are significantly faster because the database server does not log the
metadata or the user data in a temporary sbspace.

To improve performance of applications that update temporary smart large objects,

specify the LOTEMP flag in the mi_lo_specset_flags or ifx_lo_specset_flags API
function and specify a temporary sbspace for the temporary smart large objects.
The database server uses the following order of precedence for locations to place
temporary smart large objects:
v The sbspace you specify in the mi_lo_specset_sbspace or ifx_lo_specset_sbspace
API function when you create the smart large object
Specify a temporary sbspace in the API function so that changes to the objects
and the metadata are not logged. The sbspace you specify in the API function
overrides any default sbspaces that the SBSPACETEMP or SBSPACENAME
configuration parameters might specify.
v The sbspace you specify in the IN Sbspace clause when you create an explicit
temporary table with the TEMP TABLE clause of the CREATE TABLE statement
Specify a temporary sbspace in the IN Sbspace clause so that changes to the
objects and the metadata are not logged.
v The permanent sbspace you specify in the SBSPACENAME configuration
parameter, if you do not specify an sbspace in the SBSPACETEMP configuration
parameter

If no temporary sbspace is specified in any of the above methods, then the

database server issues the following error message when you try to create a
temporary smart large object:
-12053 Smart Large Objects: No sbspace number specified.

Creating temporary sbspaces

To create an sbspace for the exclusive use of temporary smart large objects, use
onspaces -c -S with the -t option.

For best performance, use the following guidelines:

v If you create more than one temporary sbspace, create each sbspace on a
separate disk to balance the I/O impact.
v Place no more than one temporary sbspace on a single disk.

5-14 IBM Informix Performance Guide

 The database server does not perform logical or physical logging of temporary
sbspaces, and temporary sbspaces are never backed up as part of a full-system
backup. You cannot mirror a temporary sbspace that you create with onspaces -t.

Important: In the case of a database with logging, you must include the WITH NO
LOG clause in the SELECT... INTO TEMP statement to place the temporary smart
large objects in the sbspaces listed in the SBSPACETEMP configuration parameter.
Otherwise, the database server stores the temporary smart large objects in the
sbspace listed in the SBSPACENAME configuration parameter.
Related tasks:
Creating a temporary sbspace (Administrator's Guide)
Related reference:
onspaces -c -S: Create an sbspace (Administrator's Reference)

Specify which sbspaces to use for temporary storage

The SBSPACETEMP configuration parameter specifies a list of sbspaces in which
the database server places temporary smart large objects by default. Some or all of
the sbspaces that you list in this configuration parameter can be temporary
sbspaces, which are reserved exclusively to store temporary smart large objects.

Important: The SBSPACETEMP configuration parameter is not set in the

onconfig.std file. For best performance with temporary smart large objects, use
SBSPACETEMP to specify two or more sbspaces on separate disks.
Related reference:
SBSPACETEMP configuration parameter (Administrator's Reference)

Placement of simple large objects

You can store simple large objects in either the same dbspace in which the table
resides or in a blobspace.

A blobspace is a logical storage unit composed of one or more chunks that store
only simple large objects (TEXT or BYTE data). For information about sbspaces,
which store smart large objects (such as BLOB, CLOB, or multirepresentational
data), see “Factors that affect I/O for smart large objects” on page 5-20.

If you use a blobspace, you can store simple large objects on a separate disk from
the table with which the data is associated. You can store simple large objects
associated with different tables in the same blobspace.

You can create a blobspace with the onspaces utility or with an SQL administration
API command that uses the create blobspace argument with the admin() or task()
function.

You assign simple large objects to a blobspace when you create the tables with
which simple large objects are associated, using the CREATE TABLE statement.

Simple large objects are not logged and do not pass through the buffer pool.
However, frequency of checkpoints can affect applications that access TEXT or
BYTE data. For more information, see “LOGSIZE and LOGFILES and their effect
on checkpoints” on page 5-32.
Related reference:

Chapter 5. Effect of configuration on I/O activity 5-15

 CREATE TABLE statement (SQL Syntax)
create blobspace argument: Create a blobspace (SQL administration API)
(Administrator's Reference)
onspaces -c -b: Create a blobspace (Administrator's Reference)

Advantage of blobspaces over dbspaces

If you store simple large objects in a blobspace on a separate disk from the table
with which it is associated, instead of storing the objects in a dbspace, you can
obtain some performance advantages.

The performance advantages of storing simple large objects in a blobspace are:

v You have parallel access to the table and simple large objects.
v Unlike simple large objects stored in a dbspace, blobspace data is written
directly to disk. Simple large objects do not pass through resident shared
memory, which leaves memory pages free for other uses.
v Simple large objects are not logged, which reduces logging I/O activity for
logged databases.

For more information, see “Storing simple large objects in the tblspace or a
separate blobspace” on page 6-8.

Blobpage size considerations

Blobspaces are divided into units called blobpages. The database server retrieves
simple large objects from a blobspace in blobpage-sized units. You specify the size
of a blobpage in multiples of a disk page when you create the blobspace.

The optimal blobpage size for your configuration depends on the following factors:
v The size distribution of the simple large objects
v The trade-off between retrieval speed for your largest simple large object and the
amount of disk space that is wasted by storing simple large objects in large
blobpages

To retrieve simple large objects as quickly as possible, use the size of your largest
simple large object rounded up to the nearest disk-page-sized increment. This
scheme guarantees that the database server can retrieve even the largest simple
large object in a single I/O request. Although this scheme guarantees the fastest
retrieval, it has the potential to waste disk space. Because simple large objects are
stored in their own blobpage (or set of blobpages), the database server reserves the
same amount of disk space for every blobpage even if the simple large object takes
up a fraction of that page. Using a smaller blobpage allows you to make better use
of your disk, especially when large differences exist in the sizes of your simple
large objects.

To achieve the greatest theoretical utilization of space on your disk, you can make
your blobpage the same size as a standard disk page. Then many, if not most,
simple large objects would require several blobpages. Because the database server
acquires a lock and issues a separate I/O request for each blobpage, this scheme
performs poorly.

In practice, a balanced scheme for sizing uses the most frequently occurring
simple-large-object size as the size of a blobpage. For example, suppose that you
have 160 simple-large-object values in a table with the following size distribution:

5-16 IBM Informix Performance Guide

v Of these values, 120 are 12 kilobytes each.
v The other 40 values are 16 kilobytes each.

You can choose one of the following blobpage sizes:

v The 12-kilobyte blobpage size provides greater storage efficiency than a
16-kilobyte blobpage size, as the following two calculations show:
– 12 kilobytes
This configuration allows the majority of simple-large-object values to require
a single blobpage and the other 40 values to require two blobpages. In this
configuration, 8 kilobytes is wasted in the second blobpage for each of the
larger values. The total wasted space is as follows:
wasted-space = 8 kilobtyes * 40
= 329 kilobytes
– 16 kilobytes
In this configuration, 4 kilobytes is wasted in the extents of 120 simple large
objects. The total wasted space is as follows:
wasted-space = 4 kilobtyes * 120
= 640 kilobytes
v If your applications access the 16-kilobyte simple-large-object values more
frequently, the database server must perform a separate I/O operation for each
blobpage. In this case, the 16-kilobyte blobpage size provides better retrieval
speed than a 12-kilobyte blobpage size.

The maximum number of pages that a blobspace can contain is 2147483647.

Therefore, the size of the blobspace is limited to the blobpage size x 2147483647.
This includes blobpages in all chunks that make up the blobspace.

Tip: If a table has more than one simple-large-object column and the data values are not
close in size, store the data in different blobspaces, each with an appropriately sized
blobpage.

Optimize blobspace blobpage size

When you are evaluating blobspace storage strategy, you can measure efficiency by
two criteria: blobpage fullness and the blobpages required per simple large object.

Blobpage fullness refers to the amount of data within each blobpage. TEXT and
BYTE data stored in a blobspace cannot share blobpages. Therefore, if a single
simple large object requires only 20 percent of a blobpage, the remaining 80
percent of the page is unavailable for use.

However, avoid making the blobpages too small. When several blobpages are
needed to store each simple large object, you increase the overhead cost of storage.
For example, more locks are required for updates, because a lock must be acquired
for each blobpage.

Obtain blobspace storage statistics

To help you determine the optimal blobpage size for each blobspace, use the
oncheck -pB command.

The command lists the following statistics for each table (or database):
v The number of blobpages used by the table (or database) in each blobspace
v The average fullness of the blobpages used by each simple large object stored as
part of the table (or database)

Chapter 5. Effect of configuration on I/O activity 5-17

 Determine blobpage fullness with oncheck -pB output
The oncheck -pB command displays statistics that describe the average fullness of
blobpages. These statistics provide a measure of storage efficiency for individual
simple large objects in a database or table.

If you find that the statistics for a significant number of simple large objects show
a low percentage of fullness, the database server might benefit from changing the
size of the blobpage in the blobspace.

Both the oncheck -pB and onstat -d update commands display the same
information about the number of free blobpages. The onstat -d update command
displays the same information as onstat -d and an accurate number of free
blobpages for each blobspace chunk.

Execute oncheck -pB with either a database name or a table name as a parameter.
The following example retrieves storage information for all simple large objects
stored in the table sriram.catalog in the stores_demo database:
oncheck -pB stores_demo:sriram.catalog

oncheck -pB Output

Figure 5-1 shows the output of this command.

BLOBSpace Report for stores_demo:sriram.catalog

Total pages used by table 7

BLOBSpace usage:
Space Page Percent Full
Name Number Pages 0-25% 26-50% 51-75 76-100%
-------------------------------------------------------------
blobPIC 0x300080 1 x
blobPIC 0x300082 2 x
------
Page Size is 6144 3

bspc1 0x2000b2 2 x
bspc1 0x2000b6 2 x
------
Page Size is 2048 4

Figure 5-1. Output of oncheck -pB

Space Name is the name of the blobspace that contains one or more simple large
objects stored as part of the table (or database).

Page Number is the starting address in the blobspace of a specific simple large
object.

Pages is the number of the database server pages required to store this simple
large object.

Percent Full is a measure of the average blobpage fullness, by blobspace, for each
blobspace in this table or database.

Page Size is the size in bytes of the blobpage for this blobspace. Blobpage size is
always a multiple of the database server page size.

5-18 IBM Informix Performance Guide

The example output indicates that four simple large objects are stored as part of
the table sriram.catalog. Two objects are stored in the blobspace blobPIC in
6144-byte blobpages. Two more objects are stored in the blobspace bspc1 in
2048-byte blobpages.

The summary information that appears at the top of the display, Total pages used
by table is a simple total of the blobpages needed to store simple large objects. The
total says nothing about the size of the blobpages used, the number of simple large
objects stored, or the total number of bytes stored.

The efficiency information displayed under the Percent Full heading is imprecise,
but it can alert an administrator to trends in the storage of TEXT and BYTE data.

Interpreting blobpage average fullness:

You can analyze the output of the oncheck -pB command to calculate average
fullness.

The first simple large object listed in “Determine blobpage fullness with oncheck
-pB output” on page 5-18 is stored in the blobspace blobPIC and requires one
6144-byte blobpage. The blobpage is 51 to 75 percent full, meaning that the size is
between 0.51 * 6144 = 3133 bytes and 0.75 * 6144 = 4608. The maximum size of this
simple large object must be less than or equal to 75 percent of 6144 bytes, or 4608
bytes.

The second object listed under blobspace blobPIC requires two 6144-byte
blobpages for storage, or a total of 12,288 bytes. The average fullness of all
allocated blobpages is 51 to 75 percent. Therefore, the minimum size of the object
must be greater than 50 percent of 12,288 bytes, or 6144 bytes. The maximum size
of the simple large object must be less than or equal to 75 percent of 12,288 bytes,
or 9216 bytes. The average fullness does not mean that each page is 51 to 75
percent full. A calculation would yield 51 to 75 percent average fullness for two
blobpages where the first blobpage is 100 percent full and the second blobpage is 2
to 50 percent full.

Now consider the two simple large objects in blobspace bspc1. These two objects
appear to be nearly the same size. Both objects require two 2048-byte blobpages,
and the average fullness for each is 76 to 100 percent. The minimum size for these
simple large objects must be greater than 75 percent of the allocated blobpages, or
3072 bytes. The maximum size for each object is slightly less than 4096 bytes
(allowing for overhead).

Analyzing efficiency criteria with oncheck -pB output:

You can analyze the output of the oncheck -pB command to determine if there is a
more efficient storage strategy.

Looking at the efficiency information for that is shown for blobspace bspc1 in
Figure 5-1 on page 5-18, a database server administrator might decide that a better
storage strategy for TEXT and BYTE data would be to double the blobpage size
from 2048 bytes to 4096 bytes. (Blobpage size is always a multiple of the database
server page size.) If the database server administrator made this change, the
measure of page fullness would remain the same, but the number of locks needed
during an update of a simple large object would be reduced by half.

Chapter 5. Effect of configuration on I/O activity 5-19

 The efficiency information for blobspace blobPIC reveals no obvious suggestion
for improvement. The two simple large objects in blobPIC differ considerably in
size, and there is no optimal storage strategy. In general, simple large objects of
similar size can be stored more efficiently than simple large objects of different
sizes.

Factors that affect I/O for smart large objects

An sbspace is a logical storage unit, composed of one or more chunks, in which
you can store smart large objects (such as BLOB, CLOB, or multi representational
data). Disk layout for sbspaces, the settings of certain configuration parameters,
and some onspaces utility options affect I/O for smart large objects.

The DataBlade® API and the Informix ESQL/C application programming interface
also provide functions that affect I/O operations for smart large objects.

Important: For most applications, you should use the values that the database
server calculates for the disk-storage information.
Related concepts:
Sbspaces (Administrator's Guide)
Related reference:
What is Informix ESQL/C? (ESQL/C Guide)
DataBlade API overview (DataBlade API Guide)

Disk layout for sbspaces

You create sbspaces on separate disks from the table with which the data is
associated. You can store smart large objects associated with different tables within
the same sbspace. When you store smart large objects in an sbspace on a separate
disk from the table with which it is associated, the database server provides some
performance advantages.

These performance advantages are:

v You have parallel access to the table and smart large objects.
v When you choose not to log the data in an sbspace, you reduce logging I/O
activity for logged databases.

To create an sbspace, use the onspaces utility. You assign smart large objects to an
sbspace when you use the CREATE TABLE statement to create the tables with
which the smart large objects are associated.
Related reference:
onspaces -c -S: Create an sbspace (Administrator's Reference)
CREATE TABLE statement (SQL Syntax)

Configuration parameters that affect sbspace I/O

The SBSPACENAME, BUFFERPOOL, and LOGBUFF configuration parameters
affect the I/O performance of sbspaces.

The SBSPACENAME configuration parameter indicates the default sbspace name if

you do not specify the sbspace name when you define a column of data type
CLOB or BLOB. To reduce disk contention and provide better load balancing, place
the default sbspace on a separate disk from the table data.

5-20 IBM Informix Performance Guide

 The BUFFERPOOL configuration parameter specifies the default values for buffers
and LRU queues in a buffer pool for both the default page size buffer pool and for
any non-default pages size buffer pools. The size of your memory buffer pool
affects I/O operations for smart large objects because the buffer pool is the default
area of shared memory for these objects. If your applications frequently access
smart large objects, it is advantageous to have these objects in the buffer pool.
Smart large objects only use the default page size buffer pool. For information
about estimating the amount to increase your buffer pool for smart large objects,
see “The BUFFERPOOL configuration parameter and memory utilization” on page
4-10.

By default, the database server reads smart large objects into the buffers in the
resident portion of shared memory. For more information on using lightweight I/O
buffers, see “Lightweight I/O for smart large objects” on page 5-23.

The LOGBUFF configuration parameter affects logging I/O activity because it

specifies the size of the logical-log buffers that are in shared memory. The size of
these buffers determines how quickly they fill and therefore how often they need
to be flushed to disk.

If you log smart-large-object user data, increase the size of your logical-log buffer
to prevent frequent flushing to these log files on disk.
Related reference:
SBSPACENAME configuration parameter (Administrator's Reference)
BUFFERPOOL configuration parameter (Administrator's Reference)
LOGBUFF configuration parameter (Administrator's Reference)

onspaces options that affect sbspace I/O

When you create an sbspace with the onspaces utility, you specify information that
affects I/O performance. This information includes the size of extents, the
buffering mode (and whether you want the server to use lightweight I/O), and
logging.

Sbspace extents
As you add smart large objects to a table, the database server allocates disk space
to the sbspace in units called extents. Each extent is a block of physically
contiguous pages from the sbspace.

Even when the sbspace includes more than one chunk, each extent is allocated
entirely within a single chunk so that it remains contiguous. Contiguity is
important to I/O performance.

When the pages of data are contiguous, disk-arm motion is minimized when the
database server reads the rows sequentially. The mechanism of extents is a
compromise between the following competing requirements:
v The size of some smart large objects is not known in advance.
v The number of smart large objects in different tables can grow at different times
and different rates.
v All the pages of a single smart large object should ideally be adjacent for best
performance when you retrieve the entire object.

Because you might not be able to predict the number and size of smart large
objects, you cannot specify the extent length of smart large objects. Therefore, the

Chapter 5. Effect of configuration on I/O activity 5-21

 database server adds extents only as they are needed, but all the pages in any one
extent are contiguous for better performance. In addition, when the database server
creates a new extent that is adjacent to the previous extent, it treats both extents as
a single extent.

The number of pages in an sbspace extent is determined by one of the following

methods:
v The database server calculates the extent size for a smart large object from a set
of heuristics, such as the number of bytes in a write operation. For example, if
an operation asks to write 30 kilobytes, the database server tries to allocate an
extent the size of 30 kilobytes.
v The final size of the smart large object as indicated by one of the following
functions when you open the sbspace in an application program:
– For DB-Access: the DataBlade API mi_lo_specset_estbytes function. For more
information about the DataBlade API functions to open a smart large object
and set the estimated number of bytes, see the IBM Informix DataBlade API
Programmer's Guide.
– For ESQL/C: the Informix ESQL/C ifx_lo_specset_estbytes function. For more
information about the Informix ESQL/C functions to open a smart large
object and set the estimated number of bytes, see the IBM Informix ESQL/C
Programmer's Manual.

These functions are the best way to set the extent size because they reduce the
number of extents in a smart large object. The database server tries to allocate the
entire smart large object as one extent (if an extent of that size is available in the
chunk).
v The EXTENT_SIZE flag in the -Df option of the onspaces command when you
create or alter the sbspace
Most administrators do not use the onspaces EXTENT_SIZE flag because the
database server calculates the extent size from heuristics. However, you might
consider using the onspaces EXTENT_SIZE flag in the following situations:
– Many one-page extents are scattered throughout the sbspace.
– Almost all smart large objects are the same length.
v The EXTENT SIZE keyword of the CREATE TABLE statement when you define
the CLOB or BLOB column
Most administrators do not use the EXTENT SIZE keyword when they create or
alter a table because the database server calculates the extent size from
heuristics. However, you might consider using this EXTENT SIZE keyword if
almost all smart large objects are the same length.

Important: For most applications, you should use the values that the database
server calculates for the extent size. Do not use the DataBlade API
mi_lo_specset_extsz function or the Informix ESQL/C ifx_lo_specset_extsz
function to set the extent size of the smart large object.

If you know the size of the smart large object, it is recommended that you specify
the size in the DataBlade API mi_lo_specset_estbytes() function or Informix
ESQL/C ifx_lo_specset_estbytes() function instead of in the onspaces utility or the
CREATE TABLE or the ALTER TABLE statement. These functions are the best way
to set the extent size because the database server allocates the entire smart large
object as one extent (if it has contiguous storage in the chunk).

Extent sizes over one megabyte do not provide much I/O benefit because the
database server performs read and write operations in multiples of 60 kilobytes at

5-22 IBM Informix Performance Guide

the most. However, the database server registers each extent for a smart large
object in the metadata area; therefore, especially large smart large objects might
have many extent entries. Performance of the database server might degrade when
it accesses these extent entries. In this case, you can reduce the number of extent
entries in the metadata area if you specify the eventual size of the smart large
object in the mi_lo_specset_estbytes() function or ifx_lo_specset_estbytes()
function.

For more information, see “Improving metadata I/O for smart large objects” on
page 6-12.

Lightweight I/O for smart large objects

Instead of using the buffer pool, the administrator and programmer have the
option to use lightweight I/O. Lightweight I/O operations use private buffers in the
session pool of the virtual portion of shared memory.

By default, smart large objects pass through the buffer pool in the resident portion
of shared memory. Although smart large objects have lower priority than other
data, the buffer pool can become full when an application accesses many smart
large objects. A single application can fill the buffer pool with smart large objects
and leave little room for data that other applications might need. In addition, when
the database server performs scans of many pages into the buffer pool, the
overhead and contention associated with checking individual pages in and out
might become a bottleneck.

Important: Use private buffers only when you read or write smart large objects in
read or write operations greater than 8080 bytes and you seldom access them. That
is, if you have infrequent read or write function calls that read large amounts of
data in a single function invocation, lightweight I/O can improve I/O
performance.
Related concepts:
“The BUFFERPOOL configuration parameter and memory utilization” on page
4-10

Advantages of lightweight I/O for smart large objects:

Lightweight I/O provides some performance advantages, because the database

server is not using the buffer pool.

Lightweight I/O provides the following advantages:

v Transfers larger blocks of data in one I/O operation
These I/O blocks can be as large as 60 kilobytes. But the bytes must be adjacent
for the database server to transfer them in a single I/O operation.
v Bypasses the overhead of the buffer pool when many pages are read
v Prevents frequently accessed pages from being forced out of the buffer pool
when many sequential pages are read for smart large objects

When you use lightweight I/O buffers for smart large objects, the database server
might read several pages with one I/O operation. A single I/O operation reads in
several smart-large-object pages, up to the size of an extent. For information about
when to specify extent size, see “Sbspace extents” on page 5-21.

Chapter 5. Effect of configuration on I/O activity 5-23

 Specifying lightweight I/O for smart large objects:

To specify the use of lightweight I/O when creating the sbspace, use the
BUFFERING tag of the -Df option in the onspaces -c -S command.

The default value for BUFFERING is ON, which means to use the buffer pool. The
buffering mode that you specify (or the default, if you do not specify) in the
onspaces command is the default buffering mode for all smart large objects stored
within the sbspace.

Important: In general, if read and write operations to the smart large objects are
less than 8080 bytes, do not specify a buffering mode when you create the sbspace.
If you are reading or writing short blocks of data, such as 2 kilobytes or 4
kilobytes, leave the default of “buffering=ON” to obtain better performance.

Programmers can override the default buffering mode when they create, open, or
alter a smart-large-object instance with DataBlade API and the Informix ESQL/C
functions. The DataBlade API and the Informix ESQL/C application programming
interface provide the LO_NOBUFFER flag to allow lightweight I/O for smart large
objects.

Important: Use the LO_NOBUFFER flag only when you read or write smart large
objects in operations greater than 8080 bytes and you seldom access them. That is,
if you have infrequent read or write function calls that read large amounts of data
in a single function invocation, lightweight I/O can improve I/O performance.
Related reference:
onspaces -c -S: Create an sbspace (Administrator's Reference)
What is Informix ESQL/C? (ESQL/C Guide)
DataBlade API overview (DataBlade API Guide)

Logging
If you decide to log all write operations on data stored in sbspaces, logical-log I/O
activity and memory utilization increases.

For more information, see “Configuration parameters that affect sbspace I/O” on
page 5-20.

How the Optical Subsystem affects performance

The Optical Subsystem extends the storage capabilities of the database server for
simple large objects (TEXT or BYTE data) to write-once-read-many (WORM)
optical subsystems. The database server uses a cache in memory to buffer initial
TEXT or BYTE data pages requested from the Optical Subsystem.

The memory cache is a common storage area. The database server adds simple
large objects requested by any application to the memory cache if the cache has
space. To free space in the memory cache, the application must release the TEXT or
BYTE data that it is using.

A significant performance advantage occurs when you retrieve TEXT or BYTE data
directly into memory instead of buffering that data on disk. Therefore, proper
cache sizing is important when you use the Optical Subsystem. You specify the
total amount of space available in the memory cache with the OPCACHEMAX
configuration parameter. Applications indicate that they require access to a portion

5-24 IBM Informix Performance Guide

 of the memory cache when they set the INFORMIXOPCACHE environment variable. For
details, see “INFORMIXOPCACHE, an Optical Subsystem environment variable”
on page 5-26.

Simple large objects that cannot fit entirely into the space that remains in the cache
are stored in the blobspace that the STAGEBLOB configuration parameter names.
This staging area acts as a secondary cache on disk for blobpages that are retrieved
from the Optical Subsystem. Simple large objects that are retrieved from the
Optical Subsystem are held in the staging area until the transactions that requested
them are complete.

The database server administrator creates the staging-area blobspace with the
onspaces utility or with ON-Monitor (UNIX only.

You can use onstat -O to monitor utilization of the memory cache and
STAGEBLOB blobspace. If contention develops for the memory cache, increase the
value listed in the configuration file for OPCACHEMAX. (The new value takes
effect the next time that the database server starts shared memory.) For a complete
description of the Optical Subsystem, see the IBM Informix Optical Subsystem Guide.

Environment variables and configuration parameters for the Optical

Subsystem
The STAGEBLOB and OPCACHEMAX configuration parameters and the
INFORMIXOPCACHE environment variable affect the performance of the Optical
Subsystem.

STAGEBLOB, an Optical Subsystem configuration parameter

The STAGEBLOB configuration parameter identifies the blobspace that is to be
used as a staging area for TEXT or BYTE data that is retrieved from the Optical
Subsystem, and it activates the Optical Subsystem.

If the configuration file does not list the STAGEBLOB parameter, the Optical
Subsystem does not recognize the optical-storage subsystem.

The structure of the staging-area blobspace is the same as all other database server
blobspaces. When the database server administrator creates the staging area, it
consists of only one chunk, but you can add more chunks as desired. You cannot
mirror the staging-area blobspace. The optimal size for the staging-area blobspace
depends on the following factors:
v The frequency of simple-large-object storage
v The frequency of simple-large-object retrieval
v The average size of the simple large object to be stored

To calculate the size of the staging-area blobspace, you must estimate the number
of simple large objects that you expect to reside there simultaneously and multiply
that number by the average simple-large-object size.
Related reference:
STAGEBLOB configuration parameter (Administrator's Reference)

Chapter 5. Effect of configuration on I/O activity 5-25

 OPCACHEMAX, an Optical Subsystem configuration
parameter
The OPCACHEMAX configuration parameter specifies the total amount of space
that is available for simple-large-object retrieval in the memory cache that the
Optical Subsystem uses.

Until the memory cache fills, it stores simple large objects that are requested by
any application. Simple large objects that cannot fit in the cache are stored on disk
in the blobspace that the STAGEBLOB configuration parameter indicates. You can
increase the size of the cache to reduce contention among simple-large-object
requests and to improve performance for requests that involve the Optical
Subsystem.
Related reference:
OPCACHEMAX configuration parameter (UNIX) (Administrator's Reference)

INFORMIXOPCACHE, an Optical Subsystem environment

variable
The INFORMIXOPCACHE environment variable sets the size of the memory cache that
a given application uses for simple-large-object retrieval.

If the value of this variable exceeds the maximum that the OPCACHEMAX
configuration parameter specifies, OPCACHEMAX is used instead. If
INFORMIXOPCACHE is not set in the environment, the cache size is set to
OPCACHEMAX by default.
Related reference:
INFORMIXOPCACHE environment variable (SQL Reference)

Table I/O
One of the most frequent functions that the database server performs is to bring
data and index pages from disk into memory. Pages can be read individually for
brief transactions and sequentially for some queries. You can configure the number
of pages that the database server brings into memory, and you can configure the
timing of I/O requests for sequential scans.

You can also indicate how the database server is to respond when a query requests
data from a dbspace that is temporarily unavailable.

The following sections describe these methods of reading pages.

For information about I/O for smart large objects, see “Factors that affect I/O for
smart large objects” on page 5-20.

Sequential scans
When the database server performs a sequential scan of data or index pages, most
of the I/O wait time is caused by seeking the appropriate starting page. To
dramatically improve performance for sequential scans, you can bring in a number
of contiguous pages with each I/O operation.

The action of bringing additional pages along with the first page in a sequential
scan is called read ahead.

5-26 IBM Informix Performance Guide

 The timing of I/O operations that are needed for a sequential scan is also
important. If the scan thread must wait for the next set of pages to be brought in
after working its way through each batch, a delay occurs. Timing second and
subsequent read requests to bring in pages before they are needed provides the
greatest efficiency for sequential scans. The number of pages to bring in and the
frequency of read-ahead I/O requests depends on the availability of space in the
memory buffers. Read-ahead operations can increase page cleaning to unacceptable
levels if too many pages are brought in with each batch or if batches are brought
in too often.
Related reference:
Read-ahead operations (Administrator's Guide)

Light scans
Some sequential scans of tables can use light scans to read the data. A light scan
bypasses the buffer pool by utilizing session memory to read directly from disk.

Light scans can provide performance advantages over use of the buffer pool for
sequential scans and skip scans of large tables. These advantages include:
v Bypassing the overhead of the buffer pool when many data pages are read
v Preventing frequently accessed pages from being forced out of the buffer pool
when many sequential pages are read for a single query.

Light scans occur under these conditions:

v The optimizer chooses a sequential scan or a skip-scan of the table.
v The amount of data in the table exceeds one MiB.
v The query meets one of the following locking conditions:
– The isolation level is Dirty Read (or the database has no transaction logging).
– The table has at least a shared lock on the entire table and the isolation level
is not Cursor Stability.

Note: A sequential scan in Repeatable Read isolation automatically acquires a

share lock on the table.

Tables that cannot be accessed by light scans

Light scans are only performed on user tables whose data rows are stored in
tblspaces. Light scans are not used to access indexes, or to access data stored in
blobspaces, smart blob spaces, or partition blobs. Similarly, light scans are not used
to access data in the system catalog tables, nor in the tables and pseudotables of
system databases like sysadmin, sysmaster, sysuser, and sysutils.

Configuration settings that affect light scans

If the BATCHEDREAD_TABLE configuration parameter or the

IFX_BATCHEDREAD_TABLE session environment option to the SET
ENVIRONMENT statement is set to 0, light scans are not used to access tables that
have variable length rows, or tables where the row length is greater than the
pagesize of the dbspace in which the table is contained. A variable length row
includes tables that have a variable length column, such as VARCHAR,
LVARCHAR or NVARCHAR, as well as tables that are compressed.

You can use the IFX_BATCHEDREAD_TABLE session environment option of the

SET ENVIRONMENT statement, or the onmode -wm command, to override the

Chapter 5. Effect of configuration on I/O activity 5-27

 setting of the BATCHEDREAD_TABLE configuration parameter for the current
session. You can use the onmode -wf command to change the value of
BATCHEDREAD_TABLE in the ONCONFIG file.

Example of onstat output during a light scan

If you have a long-running scan, you can view output from the onstat -g scn
command to check the progress of the scan, to determine how long the scan will
take before it completes, and to see whether the scan is a light scan or a bufferpool
scan.

The following example shows some of the output from onstat -g scn for a light
scan. The word Light in the Scan Type field identifies the scan as a light scan.
SesID Thread Partnum Rowid Rows Scan’d Scan Type Lock Mode Notes
17 48 300002 207 15 Light Forward row lookup
Related reference:
BATCHEDREAD_TABLE configuration parameter (Administrator's Reference)
onstat -g scn command: Print scan information (Administrator's Reference)

Unavailable data
Another aspect of table I/O pertains to situations in which a query requests access
to a table or fragment in a dbspace that is temporarily unavailable. When the
database server determines that a dbspace is unavailable as the result of a disk
failure, queries directed to that dbspace fail by default. The database server allows
you to specify dbspaces that, when unavailable, can be skipped by queries,

For information about specifying dbspaces that, when unavailable, can be skipped
by queries, see “How DATASKIP affects table I/O” on page 5-29.

Warning: If a dbspace containing data that a query requests is listed in the

DATASKIP configuration parameter and is currently unavailable because of a disk
failure, the data that the database server returns to the query can be inconsistent
with the actual contents of the database.

Configuration parameters that affect table I/O

The AUTO_READAHEAD configuration parameter changes the automatic
read-ahead mode or disables automatic read-ahead for a query. In addition, the
DATASKIP configuration parameter enables or disables data skipping.

Automatic read-ahead processing helps improve query performance by issuing

asynchronous page requests when Informix detects that the query is encountering
I/O. Asynchronous page requests can improve query performance by overlapping
query processing with the processing necessary to retrieve data from disk and put
it in the buffer pool. You can also use the AUTO_READAHEAD environment
option of the SET ENVIRONMENT statement of SQL to enable or disable the value
of the AUTO_READAHEAD configuration parameter for a session.
Related reference:
AUTO_READAHEAD configuration parameter (Administrator's Reference)

5-28 IBM Informix Performance Guide

 How DATASKIP affects table I/O
The DATASKIP configuration parameter allows you to specify which dbspaces, if
any, queries can skip when those dbspaces are unavailable as the result of a disk
failure. You can list specific dbspaces and turn data skipping on or off for all
dbspaces.

When data skipping is enabled, the database server sets the sixth character in the
SQLWARN array to W..

Warning: The database server cannot determine whether the results of a query are
consistent when a dbspace is skipped. If the dbspace contains a table fragment, the
user who executes the query must ensure that the rows within that fragment are
not needed for an accurate query result. Turning DATASKIP on allows queries
with incomplete data to return results that can be inconsistent with the actual state
of the database. Without proper care, that data can yield incorrect or misleading
query results.
Related concepts:
SQLWARN array (SQL Tutorial)
Related reference:
DATASKIP Configuration Parameter (Administrator's Reference)

Background I/O activities

Background I/O activities do not service SQL requests directly. Many of these
activities are essential to maintain database consistency and other aspects of
database server operation. However, they create overhead in the CPU and take up
I/O bandwidth.

These overhead activities take time away from queries and transactions. If you do
not configure background I/O activities properly, too much overhead for these
activities can limit the transaction throughput of your application.

The following list shows some background I/O activities:

v Checkpoints
v Logging
v Page cleaning
v Backup and restore
v Rollback and recovery
v Data replication
v Auditing

Checkpoints occur regardless of whether much database activity occurs; however,

they can occur with greater frequency as activity increases. Other background
activities, such as logging and page cleaning, occur more frequently as database
use increases. Activities such as backups, restores, or fast recoveries occur only as
scheduled or under exceptional circumstances.

For the most part, tuning your background I/O activities involves striking a
balance between appropriate checkpoint intervals, logging modes and log sizes,
and page-cleaning thresholds. The thresholds and intervals that trigger background
I/O activity often interact; adjustments to one threshold might shift the
performance bottleneck to another.

Chapter 5. Effect of configuration on I/O activity 5-29

 The following sections describe the performance effects and considerations that are
associated with the configuration parameters that affect these background I/O
activities.

Configuration parameters that affect checkpoints

The RTO_SERVER_RESTART, CKPTINTVL, LOGSIZE, LOGFILES, PHYSFILE, and
ONDBSPACEDOWN configuration parameters affect checkpoints.

RTO_SERVER_RESTART and its effect on checkpoints

The RTO_SERVER_RESTART configuration parameter specifies the amount of time,
in seconds, that Informix has to recover from an unplanned outage.

The performance advantage of enabling this configuration parameter is:

v Enabling fast recovery to meet the RTO_SERVER_RESTART policy by seeding
the buffer pool with the data pages required by log replay.

The performance disadvantages of enabling this configuration parameter are:

v Increased physical log activity which might slightly impact transaction
performance
v Increased checkpoint frequency, because the physical log space is depleted more
quickly (You can increase the size of the physical log to avoid the increase in
checkpoint frequency.)

When RTO_SERVER_RESTART is enabled, the database server:

v Attempts to make sure nonblocking checkpoints do not run out of critical
resources during checkpoint processing by triggering more frequent checkpoints
if transactions might run out of physical or logical log resources, which would
cause transaction blocking.
v Ignores the CKPTINTVL configuration parameter.
v Automatically controls checkpoint frequency to meet the RTO policy and to
prevent the server from running out of log resources.
v Automatically adjusts the number of AIO virtual processors and cleaner threads
and automatically tunes LRU flushing.

The database server prints warning messages in the message log if the server
cannot meet the RTO_SERVER_RESTART policy.
Related reference:
RTO_SERVER_RESTART configuration parameter (Administrator's Reference)

Automatic checkpoints, LRU tuning, and AIO virtual processor tuning:

The database server automatically adjusts checkpoint frequency to avoid

transaction blocking. The server monitors physical and logical log consumption
along with information about past checkpoint performance. Then, if necessary, the
server triggers checkpoints more frequently to avoid transaction blocking.

You can turn off automatic checkpoint tuning by setting onmode -wf
AUTO_CKPTS to 0, or setting the AUTO_CKPTS configuration parameter to 0.

Because the database server does not block transactions during checkpoint
processing, LRU flushing is relaxed. If the server is not able to complete checkpoint
processing before the physical log is full (which causes transaction blocking), and if
you cannot increase the size of the physical log, you can configure the server for

5-30 IBM Informix Performance Guide

more aggressive LRU flushing. The increase in LRU flushing impacts transaction
performance, but reduces transaction blocking. If you do not configure the server
for more aggressive flushing, the server automatically adjusts LRU flushing to be
more aggressive only when the server is unable to find a low priority buffer for
page replacement.

When the AUTO_AIOVPS configuration parameter is enabled, the database server

automatically increases the number of AIO virtual processors and page-cleaner
threads when the server detects that AIO virtual processors are not keeping up
with the I/O workload.

Automatic LRU tuning affects all buffer pools and adjusts lru_min_dirty and
lru_max_dirty values in the BUFFERPOOL configuration parameter.
Related concepts:
“LRU tuning” on page 5-45
Related reference:
AUTO_CKPTS configuration parameter (Administrator's Reference)
AUTO_AIOVPS configuration parameter (Administrator's Reference)
BUFFERPOOL configuration parameter (Administrator's Reference)

CKPTINTVL and its effect on checkpoints

If the RTO_SERVER_RESTART configuration parameter is not on, the CKPTINTVL
configuration parameter specifies the frequency, in seconds, at which the database
server checks to determine whether a checkpoint is needed.

When the RTO_SERVER_RESTART configuration parameter is on, the database

server ignores the CKPTINTVL configuration parameter. Instead, the server
automatically triggers checkpoints in order to maintain the
RTO_SERVER_RESTART policy.

The database server can skip a checkpoint if all data is physically consistent when
the checkpoint interval expires.

Checkpoints also occur in either of these circumstances:

v Whenever the physical log becomes 75 percent full
v If a high number of dirty partitions exist, even if the physical log is not 75
percent full.
This occurs because when the database server checks if the physical log is 75
percent full, the server also checks if the following condition is true:
(Physical Log Pages Used + Number of Dirty Partitions) >=
(Physical Log Size * 9) /10)
A partition, which represents one page going into the physical log during
checkpoint processing and has a page that maintains information (such as the
number of rows and number of data pages) about the partition, becomes dirty
when the partition is updated.

If you set CKPTINTVL to a long interval, you can use physical-log capacity to
trigger checkpoints based on actual database activity instead of an arbitrary time
unit. However, a long checkpoint interval can increase the time needed for
recovery in the event of a failure. Depending on your throughput and
data-availability requirements, you can choose an initial checkpoint interval of 5,
10, or 15 minutes, with the understanding that checkpoints might occur more
often, depending on physical-logging activity.
Chapter 5. Effect of configuration on I/O activity 5-31
 The database server writes a message to the message log to note the time that it
completes a checkpoint. To read these messages, use onstat -m.
Related reference:
CKPTINTVL configuration parameter (Administrator's Reference)

LOGSIZE and LOGFILES and their effect on checkpoints

The LOGSIZE and LOGFILES configuration parameters indirectly affect
checkpoints because they specify the size and number of logical-log files. A
checkpoint can occur when the database server detects that the next logical-log file
to become current contains the most-recent checkpoint record.

If you need to free the logical-log file that contains the last checkpoint, the
database server must write a new checkpoint record to the current logical-log file.
If the frequency with which logical-log files are backed up and freed increases, the
frequency at which checkpoints occur increases. Although checkpoints block user
processing, they no longer last as long. Because other factors (such as the
physical-log size) also determine the checkpoint frequency, this effect might not be
significant.

When the dynamic log allocation feature is enabled, the size of the logical log does
not affect the thresholds for long transactions as much as it did in previous
versions of the database server. For details, see “LTXHWM and LTXEHWM and
their effect on logging” on page 5-38.

The LOGSIZE, LOGFILES, and LOGBUFF configuration parameters also affect

logging I/O activity and logical backups. For more information, see “Configuration
parameters that affect logging” on page 5-34.
Related concepts:
Estimate the number of logical-log files (Administrator's Guide)
Related reference:
LOGFILES configuration parameter (Administrator's Reference)
LOGSIZE configuration parameter (Administrator's Reference)

Checkpoints and the physical log

The PHYSFILE configuration parameter specifies the size of the initial physical log.
A checkpoint occurs when either the physical log becomes 75 percent full or a high
number of dirty partitions exist.

The rate at which transactions generate physical log activity can affect checkpoint
performance. To avoid transaction blocking during checkpoint processing, consider
the size of the physical log and how quickly it fills.

For example, operations that do not perform updates do not generate

before-images. If the size of the database is growing, but applications rarely update
the data, little physical logging occurs. In this situation, you might not need a large
physical log.

Similarly, you can define a smaller physical log if your application updates the
same pages. The database server writes the before-image of only the first update
that is made to a page for the following operations:
v Inserts, updates, and deletes for rows that contain user-defined data types
(UDTs), smart large objects, and simple large objects
v ALTER statements
5-32 IBM Informix Performance Guide
v Operations that create or modify indexes (B-tree, R-tree, or user-defined indexes)

Because the physical log is recycled after each checkpoint, the physical log must be
large enough to hold before-images from changes between checkpoints. If the
database server frequently triggers checkpoints because it runs out of physical log
space, consider increasing the size of the physical log.

If you increase the checkpoint interval or if you anticipate increased update

activity, you might want to increase the size of the physical log.

The physical log is an important part of maintaining RTO_SERVER_RESTART

policy. To ensure that you have an abundance of space, set the size of the physical
log to at least 110 percent of the size of all buffer pools.

You can use the onparams utility to change the physical log location and size. You
can change the physical log while transactions are active and without restarting the
database server.
Related concepts:
Strategy for estimating the size of the physical log (Administrator's Guide)
Related reference:
PHYSFILE configuration parameter (Administrator's Reference)
Change the physical-log location and size (Administrator's Guide)

ONDBSPACEDOWN and its effect on checkpoints

The ONDBSPACEDOWN configuration parameter specifies the response that the
database server makes when an I/O error indicates that a dbspace is down. By
default, the database server identifies any dbspace that contains no critical data as
down and continues processing. Critical data includes the root dbspace, the logical
log, or the physical log.

To restore access to that database, you must back up all logical logs and then
perform a warm restore on the down dbspace.

The database server halts operation whenever a disabling I/O error occurs on a
nonmirrored dbspace that contains critical data, regardless of the setting for
ONDBSPACEDOWN. In such an event, you must perform a cold restore of the
database server to resume normal database operations.

The value of ONDBSPACEDOWN has no affect on temporary dbspaces. For

temporary dbspaces, the database server continues processing regardless of the
ONDBSPACEDOWN setting. If a temporary dbspace requires fixing, you can drop
and recreate it.

When ONDBSPACEDOWN is set to 2, the database server continues processing to

the next checkpoint and then suspends processing of all update requests. The
database server repeatedly retries the I/O request that produced the error until the
dbspace is repaired and the request completes or the database server administrator
intervenes. The administrator can use onmode -O to mark the dbspace down and
continue processing while the dbspace remains unavailable or use onmode -k to
halt the database server.

Chapter 5. Effect of configuration on I/O activity 5-33

 Important: This 2 setting for ONDBSPACEDOWN can affect the performance for
update requests severely because they are suspended due to a down dbspace.
When you use this setting for ONDBSPACEDOWN, be sure to monitor the status
of the dbspaces.

When you set ONDBSPACEDOWN to 1, the database server treats all dbspaces as
though they were critical. Any nonmirrored dbspace that becomes disabled halts
normal processing and requires a cold restore. The performance impact of halting
and performing a cold restore when any dbspace goes down can be severe.

Important: If you decide to set ONDBSPACEDOWN to 1, consider mirroring all

your dbspaces.
Related reference:
ONDBSPACEDOWN configuration parameter (Administrator's Reference)

Configuration parameters that affect logging

The LOGBUFF, PHYSBUFF, LOGFILES, LOGSIZE, DYNAMIC_LOGS, LTXHWM,
LTXEHWM and TEMPTAB_NOLOG configuration parameters affect logging.

Logging, checkpoints, and page cleaning are necessary to maintain database

consistency. A direct trade-off exists between the frequency of checkpoints or the
size of the logical logs and the time that it takes to recover the database in the
event of a failure. Therefore, a major consideration when you attempt to reduce the
overhead for these activities is the delay that you can accept during recovery.

LOGBUFF and PHYSBUFF and their effect on logging

The LOGBUFF and PHYSBUFF configuration parameters affect logging I/O
activity because they specify the respective sizes of the logical-log and physical-log
buffers that are in shared memory. The size of these buffers determines how
quickly the buffers fill and therefore how often they need to be flushed to disk.
Related reference:
LOGBUFF configuration parameter (Administrator's Reference)
PHYSBUFF configuration parameter (Administrator's Reference)

LOGFILES and its effect on logging

The LOGFILES configuration parameter, which specifies the number of logical-log
files, affects logging.

When you initialize or restart the database server, it creates the number of
logical-log files that you specify in the LOGFILES configuration parameter.

You might add logical-log files for the following reasons:

v To increase the disk space allocated to the logical log
v To change the size of your logical-log files
v To enable an open transaction to roll back
v As part of moving logical-log files to a different dbspace
Related concepts:
Estimate the number of logical-log files (Administrator's Guide)
Related reference:
LOGFILES configuration parameter (Administrator's Reference)

5-34 IBM Informix Performance Guide

Calculating the space allocated to logical log files:

If all of your logical log files are the same size, you can calculate the total space
allocated to the logical log files.

To calculate the space allocated to these files, use the following formula:
total logical log space = LOGFILES * LOGSIZE

If you add logical-log files that are not the size specified by the LOGSIZE
configuration parameter, you cannot use the LOGFILES * LOGSIZE expression to
calculate the size of the logical log. Instead, you need to add the sizes for each
individual log file on disk.

Use the onstat -l utility to monitor logical-log files.

LOGSIZE and its effect on logging

The LOGSIZE configuration parameter specifies the size of each logical log file. It
is difficult to predict how much logical-log space your database server system
requires until the system is fully in use.

The size of the logical log space (LOGFILES * LOGSIZE) is determined by these
policies:
Recovery time objective (RTO)
This is the length of time you can afford to be without your systems. If
your only objective is failure recovery, the total log space only needs to be
large enough to contain all the transactions for two checkpoint cycles.
When the RTO_SERVER_RESTART configuration parameter is enabled and
the server has a combined buffer pool size of less that four gigabytes, you
can configure the total log space to 110% of the combined buffer pool sizes.
Too much log space does not impact performance; however, too little log
space can cause more frequent checkpoints and transaction blocking.
Recovery point objective (RPO)
This describes the age of the data you want to restore in the event of a
disaster. If the objective is to make sure transactional work is protected, the
optimum LOGSIZE should be a multiple of how much work gets done per
RPO unit. Because the database server supports partial log backup, an
optimal log size is not critical and a non-optimal log size simply means
more frequent log file changes. RPO is measured in units of time. If the
business rule is that the system cannot lose more than ten minutes of
transactional data if a complete site disaster occurs, then a log backup
should occur every ten minutes.
You can use the Scheduler, which manages and executes scheduled
administrative tasks, to set up automatic log backup.
Long Transactions
If you have long transactions that require a large amount of log space, you
should allocate that space for the logs. Inadequate log space impacts
transaction performance.

Choose a log size based on how much logging activity occurs and the amount of
risk in case of catastrophic failure. If you cannot afford to lose more than an hour's
worth of data, create many small log files that each hold an hour's worth of
transactions. Turn on continuous-log backup. Small logical-log files fill sooner,
which means more frequent logical-log backups.

Chapter 5. Effect of configuration on I/O activity 5-35

 If your system is stable with high logging activity, choose larger logs to improve
performance. Continuous-log backups occur less frequently with large log files.
Also consider the maximum transaction rates and speed of the backup devices. Do
not let the whole logical log fill. Turn on continuous-log backup and leave enough
room in the logical logs to handle the longest transactions.

The backup process can hinder transaction processing that involves data located on
the same disk as the logical-log files. If enough logical-log disk space is available,
however, you can wait for periods of low user activity before you back up the
logical-log files.
Related concepts:
The Scheduler (Administrator's Guide)
Related reference:
LOGSIZE configuration parameter (Administrator's Reference)

Estimating logical-log size when logging dbspaces:

To estimate the size of logical logs, use a formula or onstat -u information.

Use the following formula to obtain an initial estimate for LOGSIZE in kilobytes:
LOGSIZE = (connections * maxrows * rowsize) / 1024) / LOGFILES

In this formula:
v connections is the maximum number of connections for all network types
specified in the sqlhosts information by one or more NETTYPE parameters. If
you configured more than one connection by setting multiple NETTYPE
configuration parameters in your configuration file, sum the users fields for each
NETTYPE parameter, and substitute this total for connections in the preceding
formula.
v maxrows is the largest number of rows to be updated in a single transaction.
v rowsize is the average size of a row in bytes. You can calculate rowsize by adding
up the length (from the syscolumns system catalog table) of the columns in a
row.
v 1024 is a necessary divisor because you specify LOGSIZE in kilobytes.

To obtain a better estimate during peak activity periods, execute the onstat -u
command. The last line of the onstat -u output contains the maximum number of
concurrent connections.

You need to adjust the size of the logical log when your transactions include
simple large objects or smart large objects, as the following sections describe.

You also can increase the amount of space devoted to the logical log by adding
another logical-log file.
Related tasks:
Adding logical-log files manually (Administrator's Guide)

Estimating the logical-log size when logging simple large objects:

To obtain better overall performance for applications that perform frequent updates
of TEXT or BYTE data in blobspaces, reduce the size of the logical log.

5-36 IBM Informix Performance Guide

Blobpages cannot be reused until the logical log to which they are allocated is
backed up. When TEXT or BYTE data activity is high, the performance impact of
more frequent checkpoints is balanced by the higher availability of free blobpages.

When you use volatile blobpages in blobspaces, smaller logs can improve access to
simple large objects that must be reused. Simple large objects cannot be reused
until the log in which they are allocated is flushed to disk. In this case, you can
justify the cost in performance because those smaller log files are backed up more
frequently.

Estimating the logical-log size when logging smart large objects:

If you plan to log smart-large-object user data, you must ensure that the log size is
considerably larger than the amount of data being written. Smart-large-object
metadata is always logged even if the smart large objects are not logged.

Use the following guidelines when you log smart large objects:
v If you are appending data to a smart large object, the increased logging activity
is roughly equal to the amount of data written to the smart large object.
v If you are updating a smart large object (overwriting data), the increased logging
activity is roughly twice the amount of data written to the smart large object.
The database server logs both the before-image and after-image of a smart large
object for update transactions. When updating the smart large objects, the
database server logs only the updated parts of the before and after image.
v Metadata updates affect logging less. Even though metadata is always logged,
the number of bytes logged is usually much smaller than the smart large objects.

DYNAMIC_LOGS and its effect on logging

The dynamic log file allocation feature prevents hanging problems that are caused
by rollbacks of a long transaction because the database server does not run out of
log space. The DYNAMIC_LOGS configuration parameter specifies whether the
dynamic log file allocation feature is off, on, or causes the server to pause to allow
the manual addition of a logical log file.

Dynamic log allocation allows you to do the following actions:

v Add a logical log file while the system is active, even during fast recover.
v Insert a logical log file immediately after the current log file, instead of
appending it to the end.
v Immediately access the logical log file even if the root dbspace is not backed up.

The default value for the DYNAMIC_LOGS configuration parameter is 2, which

means that the database server automatically allocates a new logical log file after
the current log file when it detects that the next log file contains an open
transaction. The database server automatically checks if the log after the current
log still contains an open transaction at the following times:
v Immediately after it switches to a new log file while writing log records (not
while reading and applying log records)
v At the beginning of the transaction cleanup phase which occurs as the last phase
of logical recovery
Logical recovery happens at the end of fast recovery and at the end of a cold
restore or roll forward.
v During transaction cleanup (rollback of open transactions), a switch to a new log
file log might occur

Chapter 5. Effect of configuration on I/O activity 5-37

 The database server also checks after this switch because it is writing log records
for the rollback.

When you use the default value of 2 for DYNAMIC_LOGS, the database server
determines the location and size of the new logical log for you:
v The database server uses the following criteria to determine on which disk to
allocate the new log file:
– Favor mirrored dbspaces
– Avoid root dbspace until no other critical dbspace is available
– Least favored space is unmirrored and noncritical dbspaces
v The database server uses the average size of the largest log file and the smallest
log file for the size of the new logical log file. If not enough contiguous disk
space is available for this average size, the database server searches for space for
the next smallest average size. The database server allocates a minimum of 200
kilobytes for the new log file.

If you want to control the location and size of the additional log file, set
DYNAMIC_LOGS to 1. When the database server switches log files, it still checks
if the next active log contains an open transaction. If it does find an open
transaction in the next log to be active, it does the following actions:
v Issues alarm event 27 (log required)
v Writes a warning message to the online log
v Pauses to wait for the administrator to manually add a log with the onparams -a
-i command-line option

You can write a script that will execute when alarm event 27 occurs to execute
onparams -a -i with the location you want to use for the new log. Your script can
also execute the onstat -d command to check for adequate space and execute the
onparams -a -i command with the location that has enough space. You must use
the -i option to add the new log right after the current log file.

If you set DYNAMIC_LOGS to 0, the database server still checks whether the next
active log contains an open transaction when it switches log files. If it does find an
open transaction in the next log to be active, it issues the following warning:
WARNING: The oldest logical log file (%d) contains records
from an open transaction (0x%p), but the Dynamic Log
Files feature is turned off.
Related concepts:
Fast recovery (Administrator's Guide)
Related reference:
DYNAMIC_LOGS configuration parameter (Administrator's Reference)

LTXHWM and LTXEHWM and their effect on logging

With the dynamic log file feature, the long transaction high watermarks are no
longer as critical as in database server versions prior to Version 9.3, because the
server does not run out of log space unless you use up the physical disk space
available on the system. The LTXHWM and LTXEHWM configuration parameters
define long transaction watermarks.

The LTXHWM parameter still indicates how full the logical log is when the
database server starts to check for a possible long transaction and to roll it back.
LTXEHWM still indicates the point at which the database server suspends new

5-38 IBM Informix Performance Guide

 transaction activity to locate and roll back a long transaction. These events should
be rare, but if they occur, they can indicate a serious problem within an
application.

Under normal operations, use the default values for LTXHWM and LTXEHWM.
However, you might want to change these default values for one of the following
reasons:
v To allow other transactions to continue update activity (which requires access to
the log) during the rollback of a long transaction
In this case, you increase the value of LTXEHWM to raise the point at which the
long transaction rollback has exclusive access to the log.
v To perform scheduled transactions of unknown length, such as large loads that
are logged
In this case, you increase the value of LTXHWM so that the transaction has a
chance to complete before reaching the high watermark.
Related reference:
LTXEHWM configuration parameter (Administrator's Reference)
LTXHWM configuration parameter (Administrator's Reference)

TEMPTAB_NOLOG and its effect on logging

The TEMPTAB_NOLOG configuration parameter allows you to disable logging on
temporary tables. You can do this to improve performance and to prevent Informix
from transferring temporary tables when using High-Availability Data Replication
(HDR).

To disable logging on temporary tables, set the TEMPTAB_NOLOG configuration

parameter to 1.
Related reference:
TEMPTAB_NOLOG configuration parameter (Administrator's Reference)

Configuration parameters that affect page cleaning

Several configuration parameters, including the CLEANERS and
RTO_SERVER_RESTART configuration parameters, affect page cleaning. If pages
are not cleaned often enough, an sqlexec thread that performs a query might be
unable to find the available pages that it needs.

If the sqlexec thread cannot find the available pages that it needs, the thread
initiates a foreground write and waits for pages to be freed. Foreground writes
impair performance, so you should avoid them. To reduce the frequency of
foreground writes, increase the number of page cleaners or decrease the threshold
for triggering a page cleaning.

Use onstat -F to monitor the frequency of foreground writes.

The following configuration parameters affect page cleaning:

v BUFFERPOOL, which contains lrus, lru_max_dirty, and lru_min_dirty values
Information that was specified with the BUFFERS, LRUS, LRU_MAX_DIRTY,
and LRU_MIN_DIRTY configuration parameters before Version 10.0 is now
specified using the BUFFERPOOL configuration parameter.
v CLEANERS
v RTO_SERVER_RESTART

Chapter 5. Effect of configuration on I/O activity 5-39

 CLEANERS and its effect on page cleaning
The CLEANERS configuration parameter indicates the number of page-cleaner
threads to run. For installations that support fewer than 20 disks, one page-cleaner
thread is recommended for each disk that contains database server data. For
installations that support between 20 and 100 disks, one page-cleaner thread is
recommended for every two disks. For larger installations, one page-cleaner thread
is recommended for every four disks.

If you increase the number of LRU queues, you must increase the number of
page-cleaner threads proportionally.
Related reference:
CLEANERS configuration parameter (Administrator's Reference)

BUFFERPOOL and its effect on page cleaning

The BUFFERPOOL configuration parameter specifies the number of least recently
used (LRU) queues to set up within the shared-memory buffer pool. The buffer
pool is distributed among LRU queues. Configuring more LRU queues allows
more page cleaners to operate and reduces the size of each LRU queue.

For a single-processor system, set the lrus field of the BUFFERPOOL configuration
parameter to a minimum of 4. For multiprocessor systems, set the lrus field to a
minimum of 4 or to the number of CPU VPs, whichever is greater.

The lrus, lru_max_dirty, and lru_min_dirty values control how often pages are
flushed to disk between checkpoints. Automatic LRU tuning, as set by the
AUTO_LRU configuration parameter, affects all buffer pools and adjusts the
lru_min_dirty and lru_max_dirty values in the BUFFERPOOL configuration
parameter.

If you increase the lru_max_dirty and lru_min_dirty values to improve transaction

throughput, do not change the gap between the lru_max_dirty and lru_min_dirty.

When the buffer pool is very large and transaction blocking is occurring during
checkpoint processing, look in the message log to determine which resource is
triggering transaction blocking. If the physical or logical log is critically low and
triggers transaction blocking, increase the size of the resource that is causing the
transaction blocking. If you cannot increase the size of the resource, consider
making LRU flushing more aggressive by decreasing the lru_min_dirty and
lru_max_dirty settings so that the server has fewer pages to flush to disk during
checkpoint processing.

To monitor the percentage of dirty pages in LRU queues, use the onstat -R
command. When the number of dirty pages consistently exceeds the lru_max_dirty
limit, you have too few LRU queues or too few page cleaners. First, use the
BUFFERPOOL configuration parameter to increase the number of LRU queues. If
the percentage of dirty pages still exceeds the lru_max_dirty limit, update the
CLEANERS configuration parameter to increase the number of page cleaners.
Related concepts:
“The BUFFERPOOL configuration parameter and memory utilization” on page
4-10
Number of LRU queues to configure (Administrator's Guide)
Related reference:
BUFFERPOOL configuration parameter (Administrator's Reference)

5-40 IBM Informix Performance Guide

 RTO_SERVER_RESTART and its effect on page cleaning
The RTO_SERVER_RESTART configuration parameter allows you to use recovery
time objective (RTO) standards to set the amount of time, in seconds, that Informix
has to recover from a problem after you restart Informix and bring it into online or
quiescent mode.

When this configuration parameter is enabled, the database server automatically

adjusts the number of AIO virtual processors and cleaner threads and
automatically tunes LRU flushing.

Use the AUTO_LRU_TUNING configuration parameter to specify whether

automatic LRU tuning is enabled or disabled when the server starts.
Related reference:
RTO_SERVER_RESTART configuration parameter (Administrator's Reference)
AUTO_LRU_TUNING configuration parameter (Administrator's Reference)

Configuration parameters that affect backup and restore

Four configuration parameters that affect backup and restore on all operating
systems also affect background I/O. Additional configuration parameters affect
backup and restore on UNIX.

The following configuration parameters affect backup and restore on all operating
systems:
v BAR_MAX_BACKUP
v BAR_NB_XPORT_COUNT
v BAR_PROGRESS_FREQ
v BAR_XFER_BUF_SIZE

In addition, the following configuration parameters affect backup and restore on

UNIX:
v LTAPEBLK
v LTAPEDEV
v LTAPESIZE
v TAPEBLK
v TAPEDEV
v TAPESIZE

ON-Bar configuration parameters

BAR_MAX_BACKUP, BAR_NB_XPORT_COUNT, BAR_PROGRESS_FREQ, and
BAR_XFER_BUF_SIZE are some ON-Bar configuration parameters that affect
background I/O.

The BAR_MAX_BACKUP configuration parameter specifies the maximum number

of backup processes per ON-Bar command. This configuration parameter also
defines the degree of parallelism, determining how many processes start to run
concurrently, including processes for backing up and restoring a whole system.
When the number of running processes is reached, further processes start only
when a running process completes its operation.

BAR_NB_XPORT_COUNT specifies the number of shared-memory data buffers for

each backup or restore process.

Chapter 5. Effect of configuration on I/O activity 5-41

 BAR_PROGRESS_FREQ specifies, in minutes, how frequently the backup or restore
progress messages display in the activity log.

BAR_XFER_BUF_SIZE specifies the size, in pages, of the buffers.

Related reference:
BAR_MAX_BACKUP configuration parameter (Backup and Restore Guide)
BAR_NB_XPORT_COUNT configuration parameter (Backup and Restore
Guide)
BAR_PROGRESS_FREQ configuration parameter (Backup and Restore Guide)
BAR_XFER_BUF_SIZE configuration parameter (Backup and Restore Guide)

ontape configuration parameters (UNIX)

On UNIX, LTAPEBLK, LTAPEDEV, LTAPESIZE, TAPEBLK, TAPEDEV, and
TAPESIZE are configuration parameters that affect the ontape utility.

On UNIX, the LTAPEBLK, LTAPEDEV, and TAPESIZE configuration parameters

specify the block size, device, and tape size for logical-log backups made with
ontape. The TAPEBLK configuration parameter specifies the block size for database
backups made with ontape, onload, and onunload.

TAPEDEV specifies the tape device. TAPESIZE specifies the tape size for these
backups.
Related reference:
ON-Bar and ontape configuration parameters and environment variable
(Backup and Restore Guide)

Configuration parameters that affect rollback and recovery

The OFF_RECVRY_THREADS, ON_RECVRY_THREADS,
PLOG_OVERFLOW_PATH, and RTO_SERVER_RESTART configuration parameters
affect recovery. The LOW_MEMORY_RESERVE configuration parameter reserves a
specific amount of memory, in kilobytes, for the database server to use when
critical activities, such as rollback activities, are needed.

OFF_RECVRY_THREADS and ON_RECVRY_THREADS and their

effect on fast recovery
The OFF_RECVRY_THREADS and ON_RECVRY_THREADS configuration
parameters specify the number of recovery threads that operate when the database
server performs a cold restore, warm restore, or fast recovery. The setting of
ON_RECVRY_THREADS controls warm restores, and the setting of
OFF_RECVRY_THREADS controls cold restores and fast recovery.

To improve the performance of fast recovery, increase the number of fast-recovery

threads with the OFF_RECVRY_THREADS configuration parameter. The number
of threads should usually match the number of tables or fragments that are
frequently updated to roll forward the transactions recorded in the logical log.

Another estimate is the number of tables or fragments that experience frequent

updates. On a single-CPU host, the number of threads should be no fewer than 10
and no more than 30 or 40. At a certain point, the overhead that is associated with
each thread outweighs the advantages of parallel threads.

5-42 IBM Informix Performance Guide

 A warm restore takes place concurrently with other database operations. To reduce
the impact of the warm restore on other users, you can allocate fewer threads to it
than you would allocate to a cold restore. However, to replay logical-log
transactions in parallel during a warm restore, specify more threads in the
ON_RECVRY_THREADS configuration parameter.
Related reference:
OFF_RECVRY_THREADS configuration parameter (Administrator's Reference)

ON_RECVRY_THREADS configuration parameter (Administrator's Reference)

PLOG_OVERFLOW_PATH and its effect on fast recovery

The PLOG_OVERFLOW_PATH configuration parameter specifies the location of a
disk file (named plog_extend.servernum) that the database server uses if the
physical log file overflows during fast recovery.

The database server removes the plog_extend.servernum file when the first
checkpoint is performed during a fast recovery.
Related reference:
PLOG_OVERFLOW_PATH configuration parameter (Administrator's Reference)

RTO_SERVER_RESTART and its effect on fast recovery

The RTO_SERVER_RESTART configuration parameter enables you to use recovery
time objective (RTO) standards to set the amount of time, in seconds, that Informix
has to recover from a problem after you restart Informix and bring it into online or
quiescent mode.
Related reference:
RTO_SERVER_RESTART configuration parameter (Administrator's Reference)

The LOW_MEMORY_RESERVE configuration parameter and

memory utilization
The LOW_MEMORY_RESERVE configuration parameter reserves a specific amount
of memory, in kilobytes, for the database server to use when critical activities are
needed and the server has limited free memory.

If you enable the new LOW_MEMORY_RESERVE configuration parameter by

setting it to a specified value in kilobytes, critical activities, such as rollback
activities, can complete even when you receive out-of-memory errors.
Related reference:
LOW_MEMORY_RESERVE configuration parameter (Administrator's
Reference)
onstat -g seg command: Print shared memory segment statistics
(Administrator's Reference)

Configuration parameters that affect data replication and

auditing
Data replication and auditing are optional. If you use these features, you can set
configuration parameters that affect data-replication performance and auditing
performance.

Chapter 5. Effect of configuration on I/O activity 5-43

 To obtain immediate performance improvements, you can disable these features,
provided that the operating requirements for your system allow you to do so.

Configuration parameters that affect data replication

Synchronized data replication can increase the amount of time it take longer to free
the log buffer after a log flush. The DRINTERVAL, DRTIMEOUT, and
HDR_TXN_SCOPE configuration parameters can adjust synchronization and
system performance.

The DRINTERVAL configuration parameter indicates whether the data-replication

buffer is flushed synchronously or asynchronously to the secondary database
server. If this parameter is set to flush asynchronously, it specifies the interval
between flushes. Each flush impacts the CPU and sends data across the network to
the secondary database server.

If the DRINTERVAL configuration parameter is set to 0, the synchronization mode

that is specified by the HDR_TXN_SCOPE configuration parameter is used. The
HDR_TXN_SCOPE configuration parameter specifies whether HDR replication is
fully synchronous, nearly synchronous, or asynchronous.
v In fully synchronous mode, transactions require acknowledgement of completion
on the HDR secondary server before they can complete.
v In asynchronous mode, transactions do not require acknowledgement of being
received or completed on the HDR secondary server before they can complete.
v In nearly synchronous mode, transactions require acknowledgement of being
received on the HDR secondary server before they can complete.

The DRTIMEOUT configuration parameter specifies the interval for which either
database server waits for a transfer acknowledgment from the other. If the primary
database server does not receive the expected acknowledgment, it adds the
transaction information to the file named in the DRLOSTFOUND configuration
parameter. If the secondary database server receives no acknowledgment, it
changes the data-replication mode as the DRAUTO configuration parameter
specifies.
Related concepts:
Replication of primary-server data to secondary servers (Administrator's
Guide)
Fully synchronous mode for HDR replication (Administrator's Guide)
Nearly synchronous mode for HDR replication (Administrator's Guide)
Asynchronous mode for HDR replication (Administrator's Guide)
Related reference:
DRINTERVAL configuration parameter (Administrator's Reference)
DRTIMEOUT configuration parameter (Administrator's Reference)
DRLOSTFOUND configuration parameter (Administrator's Reference)
DRAUTO configuration parameter (Administrator's Reference)
HDR_TXN_SCOPE configuration parameter (Administrator's Reference)
onstat -g dri command: Print high-availability data replication information
(Administrator's Reference)

5-44 IBM Informix Performance Guide

 Configuration parameters that affect auditing
The ADTERR and ADTMODE configuration parameters affect auditing
performance.

The ADTERR configuration parameter specifies whether the database server is to

halt processing for a user session for which an audit record encounters an error.
When ADTERR is set to halt such a session, the response time for that session
appears to degrade until one of the successive attempts to write the audit record
succeeds.

The ADTMODE configuration parameter enables or disables auditing according to

the audit records that you specify with the onaudit utility. Records are written to
files in the directory that the AUDITPATH parameter specifies. The AUDITSIZE
parameter specifies the size of each audit-record file.

The effect of auditing on performance is largely determined by the auditing events

that you choose to record. Depending on which users and events are audited, the
impact of these configuration parameters can vary widely.

Infrequent events, such as requests to connect to a database, have low performance

impact. Frequent events, such as requests to read any row, can generate a large
amount of auditing activity. The more users for whom such frequent events are
audited, the greater the impact on performance.
Related reference:
ADTERR configuration parameter (Security Guide)
ADTMODE configuration parameter (Security Guide)
Related information:
Auditing data security (Security Guide)

LRU tuning
The LRU settings for flushing each buffer pool between checkpoints are not critical
to checkpoint performance. The LRU settings are necessary only for maintaining
enough clean pages for page replacement.

The default settings for LRU flushing are 50 percent for lru_min_dirty and 60
percent for lru_max_dirty.

If your database server has been configured for more aggressive LRU flushing
because of checkpoint performance, you can decrease the LRU flushing at least to
the default values.

The database server automatically tunes LRU flushing when the

AUTO_LRU_TUNING configuration parameter is on and in the following cases:
v A page replacement is forced to perform a foreground write in order to find an
empty page. In this case, LRU flushing is adjusted to be 5 percent more
aggressive for the specific bufferpool where the foreground write took place.
v A page replacement is forced to use a buffer that is marked as high priority,
meaning it is frequently accessed. In this case, LRU flushing is adjusted to be
one (1) percent more aggressive for the specific bufferpool where the page
replacement using high priority buffer took place.

Chapter 5. Effect of configuration on I/O activity 5-45

 v If the RTO_SERVER_RESTART configuration parameter is on and the time it
takes to flush the bufferpool is longer than the recovery time objective, LRU
flushing is adjusted to be 10 percent more aggressive for all bufferpools.

After a checkpoint has occurred, if a page replacement performed a foreground

write during the previous checkpoint interval, the database server increases the
LRU settings by 5 percent and continues to increase the LRU flushing at each
subsequent checkpoint until the foreground write stops or until the lru_max_dirty
for a given buffer pool falls below 10 percent. For example, if a page replacement
performs a foreground write and the LRU settings for a buffer pool are 80 and 90,
the database server adjusts these to 76 and 85.5.

In addition to foreground writes, LRU flushing is tuned more aggressively

whenever a page fault replaces high priority buffers and non-high priority buffers
are on the modified LRU queue. Automatic LRU adjustments only make LRU
flushing more aggressive; they do not decrease LRU flushing. Automatic LRU
adjustments are not permanent and are not recorded in the ONCONFIG file.

LRU flushing is reset to the values contained in the ONCONFIG file on which the
database server starts.

The AUTO_LRU_TUNING configuration parameter specifies whether automatic

LRU tuning is enabled or disabled when the server starts.
Related concepts:
“Automatic checkpoints, LRU tuning, and AIO virtual processor tuning” on page
5-30
Related reference:
AUTO_LRU_TUNING configuration parameter (Administrator's Reference)
RTO_SERVER_RESTART configuration parameter (Administrator's Reference)

5-46 IBM Informix Performance Guide

Chapter 6. Table performance considerations
Some performance issues are associated with unfragmented tables and table
fragments.

Issues include:
v Table placement on disk to increase throughput and reduce contention
v Space estimates for tables, blobpages, sbspaces, and extents
v Changes to tables that add or delete historical data
v Denormalization of the database to reduce overhead

Placing tables on disk

Tables that the database server supports reside on one or more portions of one or
more disks. You control the placement of a table on disk when you create it by
assigning it to a dbspace.

Tables that the database server supports reside on one or more portions of a disk
or disks. You control the placement of a table on disk when you create it by
assigning it to a dbspace. A dbspace consists of one or more chunks. Each chunk
corresponds to all or part of a disk partition. When you assign chunks to dbspaces,
you make the disk space in those chunks available for storing tables or table
fragments.

When you configure chunks and allocate them to dbspaces, you must relate the
size of the dbspaces to the tables or fragments that each dbspace is to contain. To
estimate the size of a table, follow the instructions in “Estimating table size” on
page 6-5.

The database administrator (DBA) who is responsible for creating a table assigns
that table to a dbspace in one of the following ways:
v By using the IN DBSPACE clause of the CREATE TABLE statement
v By using the dbspace of the current database
The most recent DATABASE or CONNECT statement that the DBA issues before
issuing the CREATE TABLE statement sets the current database.

The DBA can fragment a table across multiple dbspaces, as described in “Planning
a fragmentation strategy” on page 9-1, or use the ALTER FRAGMENT statement to
move a table to another dbspace. The ALTER FRAGMENT statement provides the
simplest method for altering the placement of a table. However, the table is
unavailable while the database server processes the alteration. Schedule the
movement of a table or fragment at a time that affects the fewest users.

Other methods exist for moving tables between dbspaces:

v You can unload the data from a table and then move that data to another
dbspace with the SQL statements LOAD and UNLOAD, the onload and
onunload utilities or the High-Performance Loader (HPL).
v You can load data into and unload data from external tables.

Moving tables between databases with LOAD and UNLOAD, onload and
onunload, or HPL involves periods in which data from the table is copied to tape

© Copyright IBM Corp. 1996, 2014 6-1

 and then reloaded onto the system. These periods present windows of
vulnerability during which a table can become inconsistent with the rest of the
database. To prevent the table from becoming inconsistent, you must restrict access
to the version that remains on disk while the data transfers occur.

Depending on the size, fragmentation strategy, and indexes that are associated
with a table, it can be faster to unload a table and reload it than to alter
fragmentation. For other tables, it can be faster to alter fragmentation. You can
experiment to determine which method is faster for a table that you want to move
or re-partition.
Related concepts:
The onunload and onload utilities (Migration Guide)
Related tasks:
Moving data with external tables (Administrator's Guide)
Related reference:
ALTER FRAGMENT statement (SQL Syntax)
LOAD statement (SQL Syntax)
UNLOAD statement (SQL Syntax)
CREATE EXTERNAL TABLE Statement (SQL Syntax)

Isolating high-use tables

You can place a table with high I/O activity on a dedicated disk device. Doing this
reduces contention for the data that is stored in that table.

When disk drives have different performance levels, you can put the tables with
the highest use on the fastest drives. Placing two high-use tables on separate disk
devices reduces competition for disk access when the two tables experience
frequent, simultaneous I/O from multiple applications or when joins are formed
between them.

To isolate a high-use table on its own disk device, assign the device to a chunk,
assign that chunk to a dbspace, and then place the table in the dbspace that you
created. Figure 6-1 shows three high-use tables, each in a separate dbspace, placed
on three disks.

Locate each high-use table

in a separate dbspace, each
on its own partition or disk.

High-use High-use High-use Database

table 1 table 2 table 3

Dbspace 1 Dbspace 2 Dbspace 3

Figure 6-1. Isolating high-use tables

6-2 IBM Informix Performance Guide

Placing high-use tables on middle partitions of disks
To minimize disk-head movement, place the most frequently accessed data on
partitions close to the middle band of the disk (not near the center and not near
the edge). This approach minimizes disk-head movement to reach data in the
high-demand table.

The following figure shows the placement of the most frequently accessed data on
partitions close to the middle band of the disk.

Single chunk in a Create high-use

dbspace table in dbspace.

Locate high-use table on the

middle partitions of the disk.

Disk platter

Figure 6-2. Disk platter with high-use table located on middle Partitions

To place high-use tables on the middle partition of the disk, create a raw device
composed of cylinders that reside midway between the spindle and the outer edge
of the disk. (For instructions on how to create a raw device, see the IBM Informix
Administrator's Guide for your operating system.) Allocate a chunk, associating it
with this raw device, as your IBM Informix Administrator's Reference describes. Then
create a dbspace with this same chunk as the initial and only chunk. When you
create a high-use table, place the table in this dbspace.

Using multiple disks

You can use multiple disks for dbspaces, logical logs, temporary tables, and sort
files.

Using multiple disks for a dbspace

Using multiple disks for a dbspace helps to distribute I/O across dbspaces that
contain several small tables.

A dbspace can include multiple chunks, and each chunk can represent a different
disk. The maximum size for a chunk is 4 terabytes. This arrangement allows you
to distribute data in a dbspace over multiple disks. Figure 6-3 shows a dbspace
distributed over three disks.

Device 0x21 Device 0x27 Device 0x22

Dbspace three_arms

Figure 6-3. A dbspace distributed over three disks

Chapter 6. Table performance considerations 6-3

 Because you cannot use this type of distributed dbspace for parallel database
queries (PDQ), you should use the table-fragmentation techniques described in
“Distribution schemes” on page 9-6 to partition large, high-use tables across
multiple dbspaces.

Using multiple disks for logical logs

You can distribute logical logs in different dbspaces on multiple disks in
round-robin fashion to improve logical backup performance. This scheme allows
the database server to back up logs on one disk, while performing logging
operations on the other disks.

Keep your logical logs and the physical log on separate devices to improve
performance by decreasing I/O contention on a single device. The logical and
physical logs are created in the root dbspace when the database server is
initialized. After initialization, you can move them to other dbspaces.

Spreading temporary tables and sort files across multiple disks

You can spread the I/O associated with temporary tables and sort files across
multiple disks, after defining dbspaces for temporary tables and sort files. This can
improve performance for applications that require a large amount of temporary
space for temporary tables or large sort operations.

To define several dbspaces for temporary tables and sort files, use onspaces -t.
When you place these dbspaces on different disks and list them in the
DBSPACETEMP configuration parameter, you spread the I/O associated with
temporary tables and sort files across multiple disks, as Figure 6-4 illustrates. You
can list dbspaces that contain regular tables in DBSPACETEMP.

Device 0x21 Device 0x27 Device 0x22

Dbspace tmpdbs1 Dbspace tmpdbs2 Dbspace tmpdbs3

DBSPACETEMP= tmpdbs1,tmpdbs2,tmpdbs3

Figure 6-4. Dbspaces for temporary tables and sort files

Users can specify their own lists of dbspaces for temporary tables and sort files
with the DBSPACETEMP environment variable. For details, see “Configure
dbspaces for temporary tables and sort files” on page 5-8.

Backup and restore considerations when placing tables on

disks
When you decide where to place your tables or fragments, remember that if a
device that contains a dbspace fails, all tables or table fragments in that dbspace
are rendered inaccessible, even though tables and fragments in other dbspaces are
accessible. The need to limit data unavailability in the event of a disk failure might
influence which tables you group together in a particular dbspace.

6-4 IBM Informix Performance Guide

 Although you must perform a cold restore if a dbspace that contains critical data
fails, you need only perform a warm restore if a noncritical dbspace fails. The
desire to minimize the impact of cold restores might influence the dbspace that
you use to store critical data.

Factors affecting the performance of nonfragmented tables

and table fragments
Numerous factors affect the performance of an individual table or table fragment.
These include the placement of the table or fragment, the size of the table or
fragment, the indexing strategy that was used, the size and placement of table
extents with respect to one another, and the frequency of access to the table.

Estimating table size

You can calculate the approximate sizes (in disk pages) of tables.

For a description of size calculations for indexes, see “Estimating index pages” on
page 7-4.

The disk pages allocated to a table are collectively referred to as a tblspace. The
tblspace includes data pages. A separate tblspace includes index pages. If simple
large objects (TEXT or BYTE data) are associated with a table that is not stored in
an alternative dbspace, pages that hold simple large objects are also included in the
tblspace.

The tblspace does not correspond to any fixed region within a dbspace. The data
extents and indexes that make up a table can be scattered throughout the dbspace.

The size of a table includes all the pages within the tblspace: data pages and pages
that store simple large objects. Blobpages that are stored in a separate blobspace
are not included in the tblspace and are not counted as part of the table size.

The size of a table includes all the pages within the tblspace: data pages and pages
that store simple large objects. Blobpages that are stored in a separate blobspace or
on an optical subsystem are not included in the tblspace and are not counted as
part of the table size.

The following sections describe how to estimate the page count for each type of
page within the tblspace.

Tip: If an appropriate sample table exists, or if you can build a sample table of
realistic size with simulated data, you do not need to make estimates. You can run
oncheck -pt to obtain exact numbers.

Estimating data pages

How you estimate the data pages of a table depends on whether that table
contains fixed-length or variable-length rows.

Estimating tables with fixed-length rows

You can estimate the size (in pages) of a table with fixed-length rows. A table with
fixed-length rows has no columns of the VARCHAR or NVARCHAR data type.

Perform the following steps to estimate the size (in pages) of a table with
fixed-length rows.

Chapter 6. Table performance considerations 6-5

 To estimate the page size, row size, number of rows, and number of data pages:
1. Use onstat -b to obtain the size of a page.
The buffer size field in the last line of this output displays the page size.
2. Subtract 28 from this amount to account for the header that appears on each
data page. The resulting amount is referred to as pageuse.
3. To calculate the size of a row, add the widths of all the columns in the table
definition. TEXT and BYTE columns each use 56 bytes. If you have already
created your table, you can use the following SQL statement to obtain the size
of a row:
SELECT rowsize FROM systables WHERE tabname =
’table-name’;
4. Estimate the number of rows that the table is expected to contain. This number
is referred to as rows. The procedure for calculating the number of data pages
that a table requires differs depending on whether the row size is less than or
greater than pageuse.
5. If the size of the row is less than or equal to pageuse, use the following formula
to calculate the number of data pages. The trunc() function notation indicates
that you are to round down to the nearest integer.
data_pages = rows / trunc(pageuse/(rowsize + 4))
The maximum number of rows per page is 255, regardless of the size of the
row.

Important: Although the maximum size of a row that the database server
accepts is approximately 32 kilobytes, performance degrades when a row
exceeds the size of a page. For information about breaking up wide tables for
improved performance, see “Denormalize the data model to improve
performance” on page 6-42.
6. If the size of the row is greater than pageuse, the database server divides the
row between pages. The page that contains the initial portion of a row is called
the home page. Pages that contains subsequent portions of a row are called
remainder pages. If a row spans more than two pages, some of the remainder
pages are completely filled with data from that row. When the trailing portion
of a row uses less than a page, it can be combined with the trailing portions of
other rows to fill out the partial remainder page. The number of data pages is
the sum of the home pages, the full remainder pages, and the partial remainder
pages.
a. Calculate the number of home pages.
The number of home pages is the same as the number of rows:
homepages = rows
b. Calculate the number of full remainder pages.
First calculate the size of the row remainder with the following formula:
remsize = rowsize - (pageuse + 8)
If remsize is less than pageuse - 4, you have no full remainder pages.
If remsize is greater than pageuse - 4, use remsize in the following formula to
obtain the number of full remainder pages:
fullrempages = rows * trunc(remsize/(pageuse - 8))
c. Calculate the number of partial remainder pages.
First calculate the size of a partial row remainder left after you have
accounted for the home and full remainder pages for an individual row. In
the following formula, the remainder() function notation indicates that you
are to take the remainder after division:

6-6 IBM Informix Performance Guide

 partremsize = remainder(rowsize/(pageuse - 8)) + 4
The database server uses certain size thresholds with respect to the page
size to determine how many partial remainder pages to use. Use the
following formula to calculate the ratio of the partial remainder to the page:
partratio = partremsize/pageuse
Use the appropriate formula in the following table to calculate the number
of partial remainder pages.

Formula to Calculate the Number of Partial Remainder

partratio Value Pages
Less than .1 partrempages = rows/(trunc((pageuse/10)/remsize) + 1)
Less than .33 partrempages = rows/(trunc((pageuse/3)/remsize) + 1)
.33 or larger partrempages = rows

d. Add up the total number of pages with the following formula:

tablesize = homepages + fullrempages + partrempages

Estimating tables with variable-length rows

You can estimate the size of a table with variable-length rows with columns of the
VARCHAR or NVARCHAR data type.

When a table contains one or more VARCHAR or NVARCHAR columns, its rows
can have varying lengths. These varying lengths introduce uncertainty into the
calculations. You must form an estimate of the typical size of each VARCHAR
column, based on your understanding of the data, and use that value when you
make the estimates.

Important: When the database server allocates space to rows of varying size, it
considers a page to be full when no room exists for an additional row of the
maximum size.

To estimate the size of a table with variable-length rows, you must make the
following estimates and choose a value between them, based on your
understanding of the data:
v The maximum size of the table, which you calculate based on the maximum
width allowed for all VARCHAR or NVARCHAR columns
v The projected size of the table, which you calculate based on a typical width for
each VARCHAR or NVARCHAR column

To estimate the maximum number of data pages:

1. To calculate rowsize, add together the maximum values for all column widths.
2. Use this value for rowsize and perform the calculations described in “Estimating
tables with fixed-length rows” on page 6-5. The resulting value is called
maxsize.

To estimate the projected number of data pages:

1. To calculate rowsize, add together typical values for each of your variable-width
columns. It is suggested that you use the most frequently occurring width
within a column as the typical width for that column. If you do not have access
to the data or do not want to tabulate widths, you might choose to use some
fractional portion of the maximum width, such as 2/3 (.67).
2. Use this value for rowsize and perform the calculations described in “Estimating
tables with fixed-length rows” on page 6-5. The resulting value is called projsize.

Chapter 6. Table performance considerations 6-7

 Selecting an intermediate value for the size of the table
The actual table size should fall somewhere between the projected number of data
pages (projsize) and the maximum number of data pages (maxsize).

Based on your knowledge of the data, choose a value within that range that seems
most reasonable to you. The less familiar you are with the data, the more
conservative (higher) your estimate should be.

Estimating pages that simple large objects occupy

You can estimate the total number of pages for all simple large objects, or you can
estimate the number of pages based on the median size of the simple large objects.

The blobpages can reside in either the dbspace where the table resides or in a
blobspace. For more information about when to use a blobspace, see “Storing
simple large objects in the tblspace or a separate blobspace.”

The following methods for estimating blobpages yield a conservative (high)

estimate because a single TEXT or BYTE column does not necessarily occupy the
entire blobpage within a tblspace. In other words, a blobpage in a tblspace can
contain multiple TEXT or BYTE columns.

To estimate the number of blobpages:

1. Obtain the page size with onstat -b.
2. Calculate the usable portion of the blobpage with the following formula:
bpuse = pagesize - 32
3. For each byte of blobsize n, calculate the number of pages that the byte
occupies (bpages_n) with the following formula:
bpages1 = ceiling(bytesize1/bpuse)
bpages2 = ceiling(bytesize2/bpuse)
...
bpages_n = ceiling(bytesize_n/bpuse)
The ceiling() function indicates that you should round up to the nearest integer
value.
4. Add up the total number of pages for all simple large objects, as follows:
blobpages = bpages1 + bpages2 + ... + bpagesn

Alternatively, you can base your estimate on the median size of simple large
objects (TEXT or BYTE data); that is, the simple-large-object data size that occurs
most frequently. This method is less precise, but it is easier to calculate.

To estimate the number of blobpages based on the median size of simple large
objects:
1. Calculate the number of pages required for simple large objects of median size,
as follows:
mpages = ceiling(mblobsize/bpuse)
2. Multiply this amount by the total number of simple large objects, as follows:
blobpages = blobcount * mpages

Storing simple large objects in the tblspace or a separate

blobspace
When you create a simple-large-object column on magnetic disk, you have the
option of storing the column data in the tblspace or in a separate blobspace. You

6-8 IBM Informix Performance Guide

can often improve performance by storing simple-large-object data in a separate
blobspace, and by storing smart large objects and user-defined data in sbspaces.

You can also store simple large objects on optical media, but this discussion does
not apply to simple large objects stored in this way.

In the following example, a TEXT value is stored in the tblspace, and a BYTE value
is stored in a blobspace named rasters:
CREATE TABLE examptab
(
pic_id SERIAL,
pic_desc TEXT IN TABLE,
pic_raster BYTE IN rasters
)

For information about storing simple-large-object data in a separate blobspace, see

“Estimating pages that simple large objects occupy” on page 6-8.

A TEXT or BYTE value is always stored apart from the rows of the table; only a
56-byte descriptor is stored with the row. However, a simple large object occupies
at least one disk page. The simple large object to which the descriptor points can
reside in the same set of extents on disk as the table rows (in the same tblspace) or
in a separate blobspace.

When simple large objects are stored in the tblspace, the pages of their data are
interspersed among the pages that contain rows, which can greatly increase the
size of the table. When the database server reads only the rows and not the simple
large objects, the disk arm must move farther than when the blobpages are stored
apart. The database server scans only the row pages in the following situations:
v When it performs any SELECT operation that does not retrieve a
simple-large-object column
v When it uses a filter expression to test rows

Another consideration is that disk I/O to and from a dbspace is buffered in shared
memory of the database server. Pages are stored in case they are needed again
soon, and when pages are written, the requesting program can continue before the
actual disk write takes place. However, because blobspace data is expected to be
voluminous, disk I/O to and from blobspaces is not buffered, and the requesting
program is not allowed to proceed until all output has been written to the
blobspace.

For best performance, store a simple-large-object column in a blobspace in either of

the following circumstances:
v When single data items are larger than one or two pages each
v When the number of pages of TEXT or BYTE data is more than half the number
of pages of row data

Estimating tblspace pages for simple large objects

In your estimate of the space required for a table, include blobpages for any simple
large objects that are to be stored in that tblspace. For a table that is both relatively
small and nonvolatile, you can achieve the effect of a dedicated blobspace by
separating row pages and blobpages.

To separate row pages from blobpages within a dbspace:

Chapter 6. Table performance considerations 6-9

 1. Load the entire table with rows in which the simple-large-object columns are
null.
2. Create all indexes. The row pages and the index pages are now contiguous.
3. Update all the rows to install the simple large objects. The blobpages now
appear after the pages of row and index data within the tblspace.

Managing the size of first and next extents for the tblspace tblspace
The tblspace tblspace is a collection of pages that describe the location and
structure of all tblspaces in a dbspace. Each dbspace has one tblspace tblspace.
When you create a dbspace, you can use the TBLTBLFIRST and TBLTBLNEXT
configuration parameters to specify the first and next extent sizes for the tblspace
tblspace in a root dbspace.

You can use the onspaces utility to specify the initial and next extent sizes for the
tblspace tblspace in non-root dbspaces.

Specify the initial and next extent sizes if you want to reduce the number of
tblspace tblspace extents and reduce the frequency of situations when you need to
place the tblspace tblspace extents in non-primary chunks.

The ability to specify a first extent size that is larger than the default provides
flexibility for managing space. When you create an extent, you can reserve space
during creation of the dbspace, thereby decreasing the risk of needing additional
extents created in chunks that are not initial chunks.

You can only specify the first and next extent sizes when you create a dbspace. You
cannot alter the specification of the first and next extents sizes after the creation of
the dbspace. In addition, you cannot specify extent sizes for temporary dbspaces,
sbspaces, blobspaces, or external spaces.

If you do not specify first and next extent sizes for the tblspace tblspace, Informix
uses the existing default extent sizes.
Related tasks:
Specifying the first and next extent sizes for the tblspace tblspace
(Administrator's Guide)
Related reference:
TBLTBLFIRST configuration parameter (Administrator's Reference)
TBLTBLNEXT configuration parameter (Administrator's Reference)

Managing sbspaces
An sbspace is a logical storage unit composed of one or more chunks that store
smart large objects. You can estimate the amount of storage needed for smart large
objects, improve metadata I/O, monitor sbspaces, and change storage
characteristics.

Estimating pages that smart large objects occupy

In your estimate of the space required for a table, you should also consider the
amount of sbspace storage for any smart large objects (such as CLOB, BLOB, or
multi-representative data types) that are part of the table. An sbspace contains
user-data areas and metadata areas.

6-10 IBM Informix Performance Guide

CLOB and BLOB data is stored in sbpages that reside in the user-data area. The
metadata area contains the smart-large-object attributes, such as average size and
whether or not the smart large object is logged. For more information about
sbspaces, see your IBM Informix Administrator's Guide.

Estimating the size of the sbspace and metadata area

The first chunk of an sbspace must have a metadata area. When you add smart
large objects, the database server adds more control information to this metadata
area.

If you add a chunk to the sbspace after the initial allocation, you can take one of
the following actions for metadata space:
v Allocate another metadata area on the new chunk by default.
This action provides the following advantages:
– It is easier because the database server automatically calculates and allocates a
new metadata area on the added chunk based on the average smart large
object size
– Distributes I/O operations on the metadata area across multiple disks
v Use the existing metadata area
If you specify the onspaces -U option, the database server does not allocate
metadata space in the new chunk. Instead it must use a metadata area in one of
the other chunks.

In addition, the database server reserves 40 percent of the user area to be used in
case the metadata area runs out of space. Therefore, if the allocated metadata
becomes full, the database server starts using this reserved space in the user area
for additional control information.

You can let the database server calculate the size of the metadata area for you on
the initial chunk and on each added chunks. However, you might want to specify
the size of the metadata area explicitly, to ensure that the sbspace does not run out
of metadata space and the 40 percent reserve area. You can use one of the
following methods to explicitly specify the amount of metadata space to allocate:
v Specify the AVG_LO_SIZE tag on the onspaces -Df option.
The database server uses this value to calculate the size of the metadata area to
allocate when the -Ms option is not specified. If you do not specify
AVG_LO_SIZE, the database server uses the default value of 8 kilobytes to
calculate the size of the metadata area.
v Specify the metadata area size in the -Ms option of the onspaces utility.
Use the procedure that “Sizing the metadata area manually for a new chunk”
describes to estimate a value to specify in the onspaces -Ms option.

Sizing the metadata area manually for a new chunk

Each chunk can contain metadata, but the sum total must accommodate enough
room for all LO headers (average length 570 bytes each) and the chunk free list
(which lists all the free extents in the chunk).

The following procedure assumes that you know the sbspace size and need to
allocate more metadata space.

To size the metadata area manually for a new chunk:

1. Use the onstat -d option to obtain the size of the current metadata area from
the Metadata size field.

Chapter 6. Table performance considerations 6-11

 2. Estimate the number of smart large objects that you expect to reside in the
sbspace and their average size.
3. Use the following formula to calculate the total size of the metadata area:
Total metadata kilobytes = (LOcount*570)/1024 +
(numchunks*800) + 100
LOcount
is the number of smart large objects that you expect to have in all
sbspace chunks, including the new one.
numchunks
is the total number of chunks in the sbspace.
4. To obtain the additional required area for metadata, subtract the current
metadata size that you obtained in step 1 from the value that you obtained in
step 3.
5. When you add another chunk, specify in the -Ms option of the onspaces -a
command the value that you obtained in step 4.

Example of calculating the metadata area for a new chunk:

This topic contains an example showing how to estimate the metadata size
required for two sbspaces chunks.

Suppose the Metadata size field in the onstat -d option shows that the current
metadata area is 1000 pages. If the system page size is 2048 bytes, the size of this
metadata area is 2000 kilobytes, as the following calculation shows:
current metadata = (metadata_size * pagesize) / 1024
= (1000 * 2048) / 1024
= 2000 kilobytes

Suppose you expect 31,000 smart large objects in the two sbspace chunks. The
following formula calculates the total size of metadata area required for both
chunks, rounding up fractions:
Total metadata = (LOcount*570)/1024 + (numchunks*800) + 100
= (31,000 * 570)/1024 + (2*800) + 100
= 17256 + 1600 + 100
= 18956 kilobytes

To obtain the additional area that is required for metadata:

1. Subtract the current metadata size from the total metadata value.
Additional metadata = Total metadata - current metadata
= 18956 - 2000
= 16956 kilobytes
2. When you add the chunk to the sbspace, use the -Ms option of the onspaces -a
command to specify a metadata area of 16,956 kilobytes.
% onspaces -a sbchk2 -p /dev/raw_dev1 -o 200 -Ms 16956

Improving metadata I/O for smart large objects

The metadata pages in an sbspace contain information about the location of the
smart large objects in the sbspace. Typically, these pages are read intensive. You
can improve metadata I/O by redistributing it.

You can distribute I/O to these pages in one of the following ways:
v Mirror the chunks that contain metadata.

6-12 IBM Informix Performance Guide

 For more information about the implications of mirroring, see “Consider
mirroring for critical data components” on page 5-5.
v Position the metadata pages on the fastest portion of the disk.
Because the metadata pages are the most read-intensive part of an sbspace, place
the metadata pages toward the middle of the disk to minimize disk seek time.
To position metadata pages, use the -Mo option when you create the sbspace or
add a chunk with the onspaces utility.
v Spread metadata pages across disks.
To spread metadata pages across disks, create multiple chunks in an sbspace,
with each chunk residing on a separate disk. When you add a chunk to the
sbspace with the onspaces utility, specify the -Ms option to allocate pages for
the metadata information.
Although the database server attempts to keep the metadata information with its
corresponding data in the same chunk, it cannot guarantee that they will be
together.
v Decrease the number of extents each smart large object occupies.
When a smart large object spans multiple extents, the metadata area contains a
separate descriptor for each extent. To decrease the number of descriptor entries
that must be read for each smart large object, specify the expected final size of
the smart large object when you create the smart large object.
The database server allocates the smart large object as a single extent (if it has
contiguous storage in the chunk) when you specify the final size in either of the
following functions:
– The DataBlade API mi_lo_specset_estbytes function
– The Informix ESQL/C ifx_lo_specset_estbytes function
For more information about the functions to open a smart large object and to set
the estimated number of bytes, see the IBM Informix ESQL/C Programmer's
Manual and IBM Informix DataBlade API Programmer's Guide.
For more information about sizing extents, see “Sbspace extents” on page 5-21.

Important: For highest data availability, mirror all sbspace chunks that contain
metadata.

Monitoring sbspaces
You can monitor the effectiveness of I/O operations on smart large objects. For
better I/O performance, all smart large objects should be allocated in one extent to
be contiguous.

For more information about sizing extents, see “Sbspace extents” on page 5-21.

Contiguity provides the following I/O performance benefits:

v Minimizes the disk-arm motion
v Requires fewer I/O operations to read the smart large object
v When doing large sequential reads, can take advantage of lightweight I/O,
which reads in larger blocks of data (60 kilobytes or more, depending on your
platform) in a single I/O operation

You can use the following command-line utilities to monitor the effectiveness of
I/O operations on smart large objects:
v oncheck -cS, -pe and -pS
v onstat -g smb s option

Chapter 6. Table performance considerations 6-13

 The following sections describe how to use these utility options to monitor
sbspaces.

Monitoring sbspaces with oncheck -cS

The oncheck -cS option checks smart-large-object extents and the sbspace
partitions in the user-data area.

Figure 6-5 shows an example of the output from the -cS option for s9_sbspc.

The values in the Sbs#, Chk#, and Seq# columns correspond to the Space Chunk
Page value in the -pS output. The Bytes and Pages columns display the size of
each smart large object in bytes and pages.

To calculate the average size of smart large objects, you can total the numbers in
the Size (Bytes) column and then divide by the number of smart large objects. In
Figure 6-5, the average number of bytes allocated is 2690, as the following
calculation shows:
Average size in bytes = (15736 + 98 + 97 + 62 + 87 + 56) / 6
= 16136 / 6
= 2689.3

For information about how to specify smart large object sizes to influence extent
sizes, see “Sbspace extents” on page 5-21.

Validating space ’s9_sbspc’ ...

Large Objects
ID Ref Size Allocced Creat Last
Sbs# Chk# Seq# Cnt (Bytes) Pages Extns Flags Modified
---- ---- ----- ---- ---------- -------- ----- ----- ------------------------
2 2 1 1 15736 8 1 N-N-H Thu Jun 21 16:59:12 2007
2 2 2 1 98 1 1 N-K-H Thu Jun 21 16:59:12 2007
2 2 3 1 97 1 1 N-K-H Thu Jun 21 16:59:12 2007
2 2 4 1 62 1 1 N-K-H Thu Jun 21 16:59:12 2007
2 2 5 1 87 1 1 N-K-H Thu Jun 21 16:59:12 2007
2 2 6 1 56 1 1 N-K-H Thu Jun 21 16:59:12 2007

Figure 6-5. oncheck -cS output

The Extns field shows the minimum extent size, in number of pages, allocated to
each smart large object.

Monitoring sbspaces with oncheck -pe

The oncheck -pe option displays information that includes the size in pages of the
chunk, the number of pages used, the number of pages that are free, and a list of
all the tables in the chunk, with the initial page number and the length of the table
in pages. This option also shows if smart large objects occupy contiguous space
within an sbspace.

Execute oncheck -pe to display the following information to determine if the smart
large objects occupy contiguous space within an sbspace:
v Identifies each smart large object with the term SBLOBSpace LO
The three values in brackets following SBLOBSpace LO correspond to the Sbs#,
Chk#, and Seq# columns in the -cS output.
v Offset of each smart large object
v Number of disk pages (not sbpages) used by each smart large object

6-14 IBM Informix Performance Guide

Tip: The oncheck -pe option provides information about sbspace use in terms of
database server pages, not sbpages.

Figure 6-6 shows sample output. In this example, the size field shows that the first
smart large object occupies eight pages. Because the offset field shows that the first
smart large object starts at page 53 and the second smart large object starts at page
61, the first smart large object occupies contiguous pages.

Chunk Pathname Size Used Free

1000 940 60

Description Offset Size

-------------------------------------------------- -------- --------
RESERVED PAGES 0 2
CHUNK FREELIST PAGE 2 1
s9_sbspc:’informix’.TBLSpace 3 50
SBLOBSpace LO [2,2,1] 53 8
SBLOBSpace LO [2,2,2] 61 1
SBLOBSpace LO [2,2,3] 62 1
SBLOBSpace LO [2,2,4] 63 1
SBLOBSpace LO [2,2,5] 64 1
SBLOBSpace LO [2,2,6] 65 1
...

Figure 6-6. oncheck -pe output that shows contiguous space use

Monitoring sbspaces with oncheck -pS

The oncheck -pS option displays information about smart-large-object extents and
metadata areas in sbspace partitions. If you do not specify an sbspace name on the
command line, oncheck checks and displays the metadata for all sbspaces.

Figure 6-7 on page 6-16 shows an example of the -pS output for s9_sbspc.

To display information about smart large objects, execute the following command:
oncheck -pS spacename

The oncheck -pS output displays the following information for each smart large
object in the sbspace:
v Space chunk page
v Size in bytes of each smart large object
v Object ID that DataBlade API and Informix ESQL/C functions use
v Storage characteristics of each smart large object

When you use onspaces -c -S to create an sbspace, you can use the -Df option to
specify various storage characteristics for the smart large objects. You can use
onspaces -ch to change attributes after the sbspace is created. The Create Flags
field in the oncheck -pS output displays these storage characteristics and other
attributes of each smart large object. In Figure 6-7 on page 6-16, the Create Flags
field shows LO_LOG because the LOGGING tag was set to ON in the -Df option.

Chapter 6. Table performance considerations 6-15

 Space Chunk Page = [2,2,2] Object ID = 987122917
LO SW Version 4
LO Object Version 1
Created by Txid 7
Flags 0x31 LO_LOG LO_NOKEEP_LASTACCESS_TIME LO_HIGH_INTEG
Data Type 0
Extent Size -1
IO Size 0
Created Thu Apr 12 17:48:35 2007
Last Time Modified Thu Apr 12 17:48:43 2007
Last Time Accessed Thu Apr 12 17:48:43 2007
Last Time Attributes Modified Thu Apr 12 17:48:43 2007
Ref Count 1
Create Flags 0x31 LO_LOG LO_NOKEEP_LASTACCESS_TIME LO_HIGH_INTEG
Status Flags 0x0 LO_FROM_SERVER
Size (Bytes) 2048
Size Limit -1
Total Estimated Size -1
Deleting TxId -1
LO Map Size 200
LO Map Last Row -1
LO Map Extents 2
LO Map User Pages 2

Figure 6-7. oncheck -pS output

Monitoring sbspaces with onstat -g smb

The onstat -g smb s option displays sbspace attributes.

Use the onstat -g smb s option to display the following characteristics that affect
the I/O performance of each sbspace:
v Logging status
If applications are updating temporary smart large objects, logging is not
required. You can turn off logging to reduce the amount of I/O activity to the
logical log, CPU utilization, and memory resources.
v Average smart-large-object size
Average size and extent size should be similar to reduce the number of I/O
operations required to read in an entire smart large object. The avg s/kb output
field shows the average smart-large-object size in kilobytes. In Figure 6-8 on
page 6-17, the avg s/kb output field shows the value 30 kilobytes.
Specify the final size of the smart large object in either of the following functions
to allocate the object as a single extent:
– The DataBlade API mi_lo_specset_estbytes function
– The Informix ESQL/C ifx_lo_specset_estbytes function
For more information about the functions to open a smart large object and to set
the estimated number of bytes, see the IBM Informix ESQL/C Programmer's
Manual and IBM Informix DataBlade API Programmer's Guide.
v First extent size, next extent size, and minimum extent size
The 1st sz/p, nxt sz/p, and min sz/p output fields show these extent sizes if you
set the extent tags in the -Df option of onspaces. In Figure 6-8 on page 6-17,
these output fields show values of 0 and -1 because these tags are not set in
onspaces.

6-16 IBM Informix Performance Guide

 sbnum 7 address 2afae48
Space : flags nchk owner sbname
-------- 1 informix client
Defaults : LO_LOG LO_KEEP_LASTACCESS_TIME

LO : ud b/pg flags flags avg s/kb max lcks

2048 0 -------- 30 -1
Ext/IO : 1st sz/p nxt sz/p min sz/p mx io sz
4 0 0 -1

HdrCache : max free

512 0

Figure 6-8. onstat -g smb s output

Changing storage characteristics of smart large objects

When you create an sbspace, but do not specify values in the -Df option of the
onspaces -c -S command, you use the defaults for the storage characteristics and
attributes (such as logging and buffering). After you monitor sbspaces, you might
want to change the storage characteristics, logging status, lock mode, or other
attributes for new smart large objects.

The database administrator or programmer can use the following methods to

override these default values for storage characteristics and attributes:
v The database administrator can use one of the following onspaces options:
– Specify values when the sbspace is first created with the onspaces -c -S
command.
– Change values after the sbspace is created with the onspaces -ch command.
Specify these values in the tag options of the -Df option of onspaces. For more
information about the onspaces utility, see the IBM Informix Administrator's
Reference.
v The database administrator can specify values in the PUT clause of the CREATE
TABLE or ALTER TABLE statements.
These values override the values in the onspaces utility and are valid only for
smart large objects that are stored in the associated column of the specific table.
Other smart large objects (from columns in other tables) might also reside in this
same sbspace. These other columns continue to use the storage characteristics
and attributes of the sbspace that onspaces defined (or the default values, if
onspaces did not define them) unless these columns also used a PUT clause to
override them for a particular column.
If you do not specify the storage characteristics for a smart-large-object column
in the PUT clause, they are inherited from the sbspace.
If you do not specify the PUT clause when you create a table with
smart-large-object columns, the database server stores the smart large objects in
the system default sbspace, which is specified by the SBSPACENAME
configuration parameter in the ONCONFIG file. In this case, the storage
characteristics and attributes are inherited from the SBSPACENAME sbspace.
v Programmers can use functions in the DataBlade API and Informix ESQL/C to
alter storage characteristics for a smart-large-object column.

Chapter 6. Table performance considerations 6-17

 For information about the DataBlade API functions for smart large objects, see
the IBM Informix DataBlade API Programmer's Guide. For information about the
Informix ESQL/C functions for smart large objects, see the IBM Informix ESQL/C
Programmer's Manual.

Table 6-1 summarizes the ways to alter the storage characteristics for a smart large
object.
Table 6-1. Altering storage characteristics and other attributes of an sbspace
System-Specified Column-Level Storage
Storage Characteristics Storage Storage
Characteristics Specified by PUT Characteris-tics Characteris-tics
Storage Specified by -Df clause of CREATE Specified by a Specified by an
Character-istic System Default Option in onspaces TABLE or ALTER DataBlade API ESQL/C
or Attribute Value Utility TABLE Function Function
Last-access OFF ACCESSTIME KEEP ACCESS TIME, Yes Yes
time NO KEEP ACCESS
TIME
Lock mode BLOB LOCK_MODE No Yes Yes
Logging status OFF LOGGING LOG, NO LOG Yes Yes
Data integrity HIGH INTEG No HIGH INTEG, Yes No
MODERATE INTEG
Size of extent None EXTENT_SIZE EXTENT SIZE Yes Yes
Size of next None NEXT_SIZE No No No
extent
Minimum 2 kilobytes on MIN_EXT_SIZE No No No
extent size Windows 4
kilobytes on
UNIX
Size of smart 8 kilobytes Average size of all No Estimated size of Estimated size
large object smart large objects a particular smart of a particular
in sbspace: large object smart large
AVG_LO_SIZE Maximum size of object Maximum
a particular smart size of a
large object particular smart
large object
Buffer pool ON BUFFERING No LO_BUFFER and LO_BUFFER
usage LO_ NOBUFFER and LO_
flags NOBUFFER
flags
Name of SBSPACE- Not in -Df option. Name of an existing Yes Yes
sbspace NAME Name specified in sbspace in which a
onspaces -S option. smart large object
resides: PUT ... IN
clause
Fragmenta- None No Round-robin Round-robin or Round-robin or
tion across distribution scheme: expression-based expression-based
multiple PUT ... IN clause distribution distribution
sbspaces scheme scheme
Last-access OFF ACCESSTIME KEEP ACCESS TIME, Yes Yes
time NO KEEP ACCESS
TIME

6-18 IBM Informix Performance Guide

 Altering smart-large-object columns
When you create or modify a table, you have several options for choosing storage
characteristics and other attributes (such as logging status, buffering, and lock
mode) for specific smart-large-object columns.

When you create or modify a table, you can:

v Use the values that were set when the sbspace was created. These values are
specified in one of the following ways:
– With the various tags of the -Df option of the onspaces -c -S command
– With the system default value for any specific tag that was not specified
For guidelines to change the default storage characteristics of the -Df tags, see
“onspaces options that affect sbspace I/O” on page 5-21.
v Use the PUT clause of the CREATE TABLE statement to specify non-default
values for particular characteristics or attributes.
When you do not specify particular characteristics or attributes in the PUT
clause, they default to the values set in the onspaces -c -S command.

Later, you can use the PUT clause of the ALTER TABLE statement to change the
optional storage characteristics of these columns. Table 6-1 on page 6-18 shows
which characteristics and attributes you can change.

You can use the PUT clause of the ALTER TABLE statement to perform the
following actions:
v Specify the smart-large-object characteristics and storage location when you add
a new column to a table.
The smart large objects in the new columns can have different characteristics
than those in the existing columns.
v Change the smart-large-object characteristics of an existing column.
The new characteristics of the column apply only to new smart large objects
created for that column. The characteristics of existing smart large objects remain
the same.

For example, the BLOB data in the catalog table in the superstores_demo database
is stored in s9_sbspc with logging turned off and has an extent size of 100
kilobytes. You can use the PUT clause of the ALTER TABLE statement to turn on
logging and store new smart large objects in a different sbspace.

For information about changing sbspace extents with the CREATE TABLE
statement, see “Extent sizes for smart large objects in sbspaces” on page 6-22.
Related reference:
Sbspace logging (Administrator's Guide)
CREATE TABLE statement (SQL Syntax)

Managing extents
As you add rows to a table, the database server allocates disk space in units called
extents. Each extent is a block of physically contiguous pages from the dbspace.
Even when the dbspace includes more than one chunk, each extent is allocated
entirely within a single chunk, so that it remains contiguous.

Contiguity is important to performance. When the pages of data are contiguous,

and when the database server reads the rows sequentially during read-ahead, light

Chapter 6. Table performance considerations 6-19

 scans, or lightweight I/O operations, disk-arm motion is minimized. For more
information about these operations, see “Sequential scans” on page 5-26, “Light
scans” on page 5-27, and “Configuration parameters that affect sbspace I/O” on
page 5-20.

The mechanism of extents is a compromise between the following competing

requirements:
v Most dbspaces are shared among several tables.
v The size of some tables is not known in advance.
v Tables can grow at different times and different rates.
v All the pages of a table should be adjacent for best performance.

If you have a table that needs more extents and the database server runs out of
space on the partition header page, the database server automatically allocates
extended secondary partition header pages to accommodate new extent entries.
The database server can allocate up to 32767 extents for any partition, unless the
size of a table dictates a limit to the number of extents.

Because table sizes are not known, the database server cannot preallocate table
space. Therefore, the database server adds extents only as they are needed, but all
the pages in any one extent are contiguous for better performance. In addition,
when the database server creates an extent that is next to the previous one, it treats
both as a single extent.

A frequently updated table can become fragmented over time which degrades the
performance every time the table is accessed by the server. Defragmenting a table
brings data rows closer together and avoids partition header page overflow
problems.

Choosing table extent sizes

When you create a table, you can specify extent sizes for the data rows of a table
in a dbspace and for each fragment of a fragmented table, and the smart large
objects in an sbspace. The database server calculates extent sizes for smart large
objects in sbspaces.

Extent sizes for tables in a dbspace

When you create a table, you can specify the size of the first extent, as well as the
size of the extents to be added as the table grows. You can also modify the size of
the first extent in a table in a dbspace, and you can modify the size of new
subsequent extents.

The following sample SQL statement creates a table with a 512-kilobyte initial
extent and 100-kilobyte added extents:
CREATE TABLE big_one (...column specifications...)
IN big_space
EXTENT SIZE 512
NEXT SIZE 100

The default value for the extent size and the next-extent size is eight times the disk
page size on your system. For example, if you have a 2-kilobyte page, the default
length is 16 kilobytes.

You can use the ALTER TABLE statement with the MODIFY EXTENT SIZE clause
to change the size of the first extent of a table in a dbspace. When you change the

6-20 IBM Informix Performance Guide

size of the first extent, Informix records the change in the system catalog and on
the partition page, but only makes the actual change when the table is rebuilt or a
new partition or fragment is created.

You might want to change the size of the first extent of a table in a dbspace in
either of these situations:
v If a table was created with small first extent size and you need to keep adding
a lot of next extents, the table becomes fragmented across multiple extents and
the data is scattered.
v If a table was created with a first extent that is much larger than the amount of
data that is stored, space is wasted.

The following example changes the size of the first extent of a table in a dbspace to
50 kilobytes:
ALTER TABLE customer MODIFY EXTENT SIZE 50;

Changes to the first extent size are recorded into the system catalog table and on
the partition page on the disk. However, changes to the first extent size do not take
effect immediately. Instead, whenever a change that rebuilds the table occurs, the
server uses the new first extent size.

For example, if a table has a first extent size of 8 kilobytes and you use the ALTER
TABLE statement to change this to 16 kilobytes, the server does not drop the
current first extent and recreate it with the new size. Instead, the new first extent
size of 16 kilobytes takes effect only when the server rebuilds the table after actions
such as creating a cluster index on the table or detaching a fragment from the
table.

If a TRUNCATE TABLE statement without the REUSE option is executed before

the ALTER TABLE statement with the MODIFY EXTENT SIZE clause, there is no
change in the current first extent.

Use the MODIFY NEXT SIZE clause to change the size of the next extent to be
added. This change does not affect next extents that already exist.

The following example changes the size of the next extent of a table to 50
kilobytes:
ALTER TABLE big_one MODIFY NEXT SIZE 50;

The next extent sizes of the following kinds of tables do not affect performance
significantly:
v A small table is defined as a table that has only one extent. If such a table is
heavily used, large parts of it remain buffered in memory.
v An infrequently used table is not important to performance no matter what size
it is.
v A table that resides in a dedicated dbspace always receives new extents that are
adjacent to its old extents. The size of these extents is not important because,
being adjacent, they perform as one large extent.

Avoid creating large numbers of extents

When you assign an extent size to these kinds of tables, the only consideration is
to avoid creating large numbers of extents. A large number of extents causes the
database server to spend extra time finding the data. In addition, an upper limit

Chapter 6. Table performance considerations 6-21

 exists on the number of extents allowed. (“Considering the upper limit on extents”
on page 6-24 covers this topic.)

Tips for allocating space for table extents

No upper limit exists on extent sizes except the size of the chunk. The maximum
size for a chunk is 4 terabytes. When you know the final size of a table (or can
confidently predict it within 25 percent), allocate all its space in the initial extent.
When tables grow steadily to unknown size, assign them next-extent sizes that let
them share the dbspace with a small number of extents each.

Allocating space for table extents

To allocate space for table extents:

1. Decide how to allocate space among the tables.
For example, you might divide the dbspace among three tables in the ratio 0.4:
0.2: 0.3 (reserving 10 percent for small tables and overhead).
2. Give each table one-fourth of its share of the dbspace as its initial extent.
3. Assign each table one-eighth of its share as its next-extent size.
4. Monitor the growth of the tables regularly with oncheck.

As the dbspace fills up, you might not have enough contiguous space to create an
extent of the specified size. In this case, the database server allocates the largest
contiguous extent that it can.
Related reference:
TBLTBLFIRST configuration parameter (Administrator's Reference)
TBLTBLNEXT configuration parameter (Administrator's Reference)
MODIFY EXTENT SIZE (SQL Syntax)

Extent sizes for table fragments

When you fragment an existing table, you might want to adjust the next-extent
size because each fragment requires less space than the original, unfragmented
table.

If the unfragmented table was defined with a large next-extent size, the database
server uses that same size for the next-extent on each fragment, which results in
over-allocation of disk space. Each fragment requires only a proportion of the
space for the entire table.

For example, if you fragment the preceding big_one sample table across five disks,
you can alter the next-extent size to one-fifth the original size. The following
example changes the next-extent size to one-fifth of the original size:
ALTER TABLE big_one MODIFY NEXT SIZE 2;
Related reference:
MODIFY NEXT SIZE clause (SQL Syntax)

Extent sizes for smart large objects in sbspaces

When you create a table, you should use the extent size that the database server
calculates for smart large objects in sbspaces. Alternatively, you can use the final
size of the smart large object, as indicated by a particular function when you open
the sbspace in an application program.

6-22 IBM Informix Performance Guide

 You can use the final size of the smart large object when you open one of the
following application programs:
v For DB-Access: Use the DataBlade API mi_lo_specset_estbytes function. For
more information about the DataBlade API functions to open a smart large
object and set the estimated number of bytes, see the IBM Informix DataBlade API
Programmer's Guide.
v For ESQL/C: Use the Informix ESQL/C ifx_lo_specset_estbytes function. For
more information about the Informix ESQL/C functions to open a smart large
object and set the estimated number of bytes, see the IBM Informix ESQL/C
Programmer's Manual.

For more information about sizing extents, see “Sbspace extents” on page 5-21. For
more information, see “Monitoring sbspaces” on page 6-13.

Monitoring active tblspaces

Monitor tblspaces to determine which tables are active. Active tables are those that
a thread has currently opened.

Output from the onstat -t option includes the tblspace number and the following
four fields.
Field Description
npages
Pages allocated to the tblspace
nused Pages used from this allocated pool
nextns Number of extents used
npdata
Number of data pages used

If a specific operation needs more pages than are available (npages minus nused),
a new extent is required. If enough space is available in this chunk, the database
server allocates the extent here; if not, the database server looks for space in other
available chunks. If none of the chunks contains adequate contiguous space, the
database server uses the largest block of contiguous space that it can find in the
dbspace. Figure 6-9 shows an example of the output from this option.

Tblspaces
n address flgs ucnt tblnum physaddr npages nused npdata nrows nextns
0 422528 1 1 100001 10000e 150 124 0 0 3
1 422640 1 1 200001 200004 50 36 0 0 1
54 426038 1 6 100035 1008ac 3650 3631 3158 60000 3
62 4268f8 1 6 100034 1008ab 8 6 4 60 1
63 426a10 3 6 100036 1008ad 368 365 19 612 3
64 426b28 1 6 100033 1008aa 8 3 1 6 1
193 42f840 1 6 10001b 100028 8 5 2 30 1
7 active, 200 total, 64 hash buckets

Figure 6-9. onstat -t output

Monitoring the upper limit on extents and extent interleaving

You can monitor the upper limit on the number of extents. You can also check for
and eliminate extent interleaving.

The maximum number of extents for a partition is 32767.

Chapter 6. Table performance considerations 6-23

 Considering the upper limit on extents
Do not allow a table to acquire a large number of extents because an upper limit
exists on the number of extents allowed. Trying to add an extent after you reach
the limit causes error -136 (No more extents) to follow an INSERT request.

To help ensure that the limit is not exceeded, the database server performs the
following actions:
v The database server checks the number of extents each time that it creates an
extent. If the number of the extent being created is a multiple of 16, the database
server automatically doubles the next-extent size for the table. Therefore, at
every 16th creation, the database server doubles the next-extent size.
v When the database server creates an extent next to the previous extent, it treats
both extents as a single extent.

Checking for extent interleaving

When two or more growing tables share a dbspace, extents from one tblspace can
be placed between extents from another tblspace. When this situation occurs, the
extents are said to be interleaved. Performance suffers when disk seeks for a table
must span more than one extent, particularly for sequential scans.

Interleaving creates gaps between the extents of a table. Figure 6-10 shows gaps
between table extents.

Table 1 extents
Table 2 extents
Table 3 extents

Figure 6-10. Interleaved table extents

Try to optimize the table-extent sizes to allocate contiguous disk space, which
limits head movement. Also consider placing the tables in separate dbspaces.

Check periodically for extent interleaving by monitoring chunks. Execute oncheck

-pe to obtain the physical layout of information in the chunk. The following
information appears:
v Dbspace name and owner
v Number of chunks in the dbspace
v Sequential layout of tables and free space in each chunk
v Number of pages dedicated to each table extent or free space

This output is useful for determining the degree of extent interleaving. If the
database server cannot allocate an extent in a chunk despite an adequate number
of free pages, the chunk might be badly interleaved.

Eliminating interleaved extents

You can eliminate interleaved extents by reorganizing the tables with the
UNLOAD and LOAD statements, creating or altering an index to cluster, or using
the ALTER TABLE statement.

6-24 IBM Informix Performance Guide

Reorganizing dbspaces and tables to eliminate extent interleaving:

You can rebuild a dbspace to eliminate interleaved extents so that the extents for
each table are contiguous.

The order of the reorganized tables within the dbspace is not important, but the
pages of each reorganized table should be contiguous so that no lengthy seeks are
required to read the table sequentially. When the disk arm reads a table
nonsequentially, it ranges only over the space that table occupies.

Table 1 extents
Table 2 extents
Table 3 extents

Figure 6-11. A dbspace reorganized to eliminate interleaved extents

To reorganize tables in a dbspace:

1. For DB-Access users: Copy the tables in the dbspace individually to tape with
the UNLOAD statement in DB-Access.
2. Drop all the tables in the dbspace.
3. Re-create the tables with the LOAD statement or the dbload utility.

The LOAD statement re-creates the tables with the same properties they had
before, including the same extent sizes.

You can also unload a table with the onunload utility and reload the table with the
companion onload utility.
Related concepts:
The onunload and onload utilities (Migration Guide)
Related reference:
LOAD statement (SQL Syntax)
UNLOAD statement (SQL Syntax)

Creating or altering an index to cluster:

Depending on the circumstances, you can eliminate extent interleaving if you

create a clustered index or alter a clustered index. When you use the TO CLUSTER
clause of the CREATE INDEX or ALTER INDEX statement, the database server
sorts and reconstructs the table.

The TO CLUSTER clause reorders rows in the physical table to match the order in
the index. For more information, see “Clustering” on page 7-11.

The TO CLUSTER clause eliminates interleaved extents under the following

conditions:
v The chunk must contain contiguous space that is large enough to rebuild each
table.
v The database server must use this contiguous space to rebuild the table.

Chapter 6. Table performance considerations 6-25

 If blocks of free space exist before this larger contiguous space, the database
server might allocate the smaller blocks first. The database server allocates space
for the ALTER INDEX process from the beginning of the chunk, looking for
blocks of free space that are greater than or equal to the size that is specified for
the next extent. When the database server rebuilds the table with the smaller
blocks of free space that are scattered throughout the chunk, it does not
eliminate extent interleaving.

To display the location and size of the blocks of free space, execute the oncheck
-pe command.

To use the TO CLUSTER clause of the ALTER INDEX statement:

1. For each table in the chunk, drop all fragmented or detached indexes except the
one that you want to cluster.
2. Cluster the remaining index with the TO CLUSTER clause of the ALTER
INDEX statement. This step eliminates interleaving the extents when you
rebuild the table by rearranging the rows.
3. Re-create all the other indexes.

You do not need to drop an index before you cluster it. However, the ALTER
INDEX process is faster than CREATE INDEX because the database server reads
the data rows in cluster order using the index. In addition, the resulting indexes
are more compact.

To prevent the problem from recurring, consider increasing the size of the tblspace
extents.

Using ALTER TABLE to eliminate extent interleaving:

If you use the ALTER TABLE statement to add or drop a column or to change the
data type of a column, the database server copies and reconstructs the table. When
the database server reconstructs the entire table, it rewrites the table to other areas
of the dbspace. However, if other tables are in the dbspace, no guarantee exists
that the new extents will be adjacent to each other.

Important: For certain types of operations that you specify in the ADD, DROP,
and MODIFY clauses, the database server does not copy and reconstruct the table
during the ALTER TABLE operation. In these cases, the database server uses an
in-place alter algorithm to modify each row when it is updated (rather than during
the ALTER TABLE operation). For more information about the conditions for this
in-place alter algorithm, see “In-place alter” on page 6-35.

Reclaiming unused space within an extent

After the database server allocates disk space to a tblspace as part of an extent,
that space remains dedicated to the tblspace. Even if all extent pages become
empty after you delete data, the disk space remains unavailable for use by other
tables unless you reclaim the space.

Important: When you delete rows in a table, the database server reuses that space
to insert new rows into the same table. This section describes the procedures for
reclaiming unused space for use by other tables.

You might want to resize a table that does not require the entire amount of space
that was originally allocated to it. You can reallocate a smaller dbspace and release
the unneeded space for other tables to use.

6-26 IBM Informix Performance Guide

 As the database server administrator, you can reclaim the disk space in empty
extents and make it available to other users by rebuilding the table. To rebuild the
table, use any of the following SQL statements:
v ALTER INDEX
v UNLOAD and LOAD
v ALTER FRAGMENT

Reclaiming space in an empty extent with ALTER INDEX

If the table with the empty extents includes an index, you can run the ALTER
INDEX statement with the TO CLUSTER clause. Clustering an index rebuilds the
table in a different location within the dbspace.

When you run the ALTER INDEX statement with the TO CLUSTER clause, all of
the extents associated with the previous version of the table are released. Also, the
newly built version of the table has no empty extents.
Related concepts:
Clustering (Performance Guide)
Related reference:
ALTER INDEX statement (SQL Syntax)

Reclaiming space in an empty extent by unloading and

re-creating or reloading a table
If the table does not include an index, you can unload the table, re-create the table
(either in the same dbspace or in another one), and reload the data with the
UNLOAD and LOAD statements or the onunload and onload utilities.
Related concepts:
The onunload and onload utilities (Migration Guide)
Related reference:
LOAD statement (SQL Syntax)
UNLOAD statement (SQL Syntax)

Releasing space in an empty extent with ALTER FRAGMENT

You can use the ALTER FRAGMENT statement to rebuild a table. When you run
this statement, it releases space within the extents that were allocated to that table.

For more information about the syntax of the ALTER FRAGMENT statement, see
the IBM Informix Guide to SQL: Syntax.

Managing extent deallocation with the TRUNCATE keyword

TRUNCATE is an SQL keyword that quickly deletes active rows from a table and
the b-tree structures of its indexes, without dropping the table or its schema, access
privileges, triggers, constraints, and other attributes. With this SQL data-definition
language statement, you can depopulate a local table and reuse the table without
re-creating it, or you can release the storage space that formerly held its data rows
and b-tree structures.

Two implementations of TRUNCATE exist:

v The first implementation, called "fast truncate," operates on most tables.
v The second implementation, called "slow truncate," operates on tables that
include opaque or smart large object data types, or inherited indexes that are
defined on ROW types within data type hierarchies.
Chapter 6. Table performance considerations 6-27
 The performance advantages of using the TRUNCATE TABLE statement instead of
the DELETE statement are much better for the fast truncate implementation,
because this implementation does not examine or run all of the rows in a table.
Slow truncation implementation occurs on tables that include opaque or smart
large object data types or inherited indexes that are defined on ROW types within
data types, because the truncate operation examines each row containing these
items.

For more information about using TRUNCATE, see the IBM Informix Guide to SQL:
Syntax.

Defragment partitions to merge extents

You can improve performance by defragmenting partitions to merge
non-contiguous extents.

A frequently updated table can become fragmented over time which degrades the
performance every time the table is accessed by the server. Defragmenting a table
brings data rows closer together and avoids partition header page overflow
problems.

Defragmenting an index brings the entries closer together which improves the
speed at which the table information is accessed.

You cannot stop a defragment request after the request has been submitted.
Additionally, there are specific objects that cannot be defragmented and you cannot
defragment a partition if another operation is running that conflicts with the
defragment request.

Tip: Before you defragment a partition:

v Review the information about important limitations and considerations in
Defragment partitions.
v Run the oncheck -pt and pT command to determine the number of extents for a
specific table or fragment.

To defragment a table, index, or partition, run the EXECUTE FUNCTION

command with the defragment argument. You can specify the table name, index
name, or partition number that you want to defragment.

You can use the onstat -g defragment command to display information about the
active defragment requests.
Related tasks:
Scheduling data optimization (Administrator's Guide)
Related reference:
onstat -g defragment command: Print defragment partition extents
(Administrator's Reference)
oncheck -pt and -pT: Display tblspaces for a Table or Fragment
(Administrator's Reference)
defragment argument: Dynamically defragment partition extents (SQL
administration API) (Administrator's Reference)

6-28 IBM Informix Performance Guide

Storing multiple table fragments in a single dbspace
You can store multiple fragments of the same table or index in a single dbspace,
thus reducing the total number of dbspaces needed for a fragmented table. You
must specify a name for each fragment that you want to store in the same dbspace.
Storing multiple table or index fragments in a single dbspace simplifies the
management of dbspaces.

You can also use this feature to improve query performance over storing each
fragment in a different dbspace when a dbspace is located on a faster device.

For more information, see information about managing partitions in the IBM
Informix Administrator's Guide.

Displaying a list of table and index partitions

Use the onstat -g opn option to display a list of the table and index partitions, by
thread ID, that are currently open in the system.

For an example of onstat -g opn output and an explanation of output fields, see
the IBM Informix Administrator's Reference.

Changing tables to improve performance

You can change tables to improve performance by dropping indexes, attaching or
detaching fragments, and altering table definitions. You can also create databases
for decision-support applications by unloading and loading tables in OLTP
databases.

You might want to change an existing table for various reasons:

v To refresh large decision-support tables with data periodically
v To add or drop historical data from a certain time period
v To add, drop, or modify columns in large decision-support tables when the need
arises for different data analysis

Loading and unloading tables

You can create databases for decision-support applications by periodically loading
tables that have been unloaded from active OLTP databases.

You can use one or more of the following methods to load large tables quickly:
v External tables
v Nonlogging tables
The database server provides support to:
– Create nonlogging or logging tables in a logging database.
– Alter a table from nonlogging to logging and vice versa.
The two table types are STANDARD (logging tables) and RAW (nonlogging
tables). You can use any loading utility such as dbimport or HPL to load raw
tables.
v High-Performance Loader (HPL)
You can use HPL in express mode to load tables quickly.

The following sections describe:

Chapter 6. Table performance considerations 6-29

 v Advantages of logging and nonlogging tables
v Step-by-step procedures to load data using nonlogging tables
Related tasks:
Moving data with external tables (Administrator's Guide)
Related reference:
CREATE EXTERNAL TABLE Statement (SQL Syntax)

Advantages of logging tables

Logging type options specify the logging characteristics that can improve
performance in various bulk operations on the table.

STANDARD, which corresponds to a table in a logged database of previous

versions, is the default logging type that is used when you issue the CREATE
TABLE statement without specifying the table type.

Standard tables have the following features:

v Logging to allow rollback, recovery, and restoration from archives.
v Recovery from backups
v All insert, delete, and update operations
v Constraints to maintain the integrity of your data
v Indexes to quickly retrieve a small number of rows

OLTP applications usually use standard tables. OLTP applications typically have
the following characteristics:
v Real-time insert, update, and delete transactions
Logging and recovery of these transactions is critical to preserve the data.
Locking is critical to allow concurrent access and to ensure the consistency of the
data selected.
v Update, insert, or delete one row or a few rows at a time
Indexes speed access to these rows. An index requires only a few I/O operations
to access the pertinent row, but scanning a table to find the pertinent row might
require many I/O operations.

Advantages of nonlogging tables

Nonlogging tables, which are also called raw tables, have characteristics that
enable you to load very large data warehousing tables quickly.

Raw tables have following characteristics:

v They do not use CPU and I/O resources for logging.
v They avoid problems such as running out of logical-log space.
v They are locked exclusively during an express load so that no other user can
access the table during the load.
v They do not support referential constraints and unique constraints, so overhead
for constraint-checking is eliminated.

Quickly loading a large standard table:

You can change a large, existing standard table into a nonlogging table and then
load the table.

To quickly load a large, existing standard table:

6-30 IBM Informix Performance Guide

1. Drop indexes, referential constraints, and unique constraints.
2. Change the table to nonlogging.
The following sample SQL statement changes a STANDARD table to
nonlogging:
ALTER TABLE largetab TYPE(RAW);
3. Load the table using a load utility such as dbexport or the High-Performance
Loader (HPL). For more information about dbexport and dbload, see the IBM
Informix Migration Guide. For more information about HPL, see the IBM Informix
High-Performance Loader User's Guide.
4. Perform a level-0 backup of the nonlogging table. You must make a level-0
backup of any nonlogging table that has been modified before you convert it to
STANDARD type. The level-0 backup provides a starting point from which to
restore the data.
5. Change the nonlogging table to a logging table before you use it in a
transaction. The following sample SQL statement changes a raw table to a
standard table:
ALTER TABLE largetab TYPE(STANDARD);

Warning: Do not use nonlogging tables within a transaction where multiple

users can modify the data. If you need to use a nonlogging table within a
transaction, either set Repeatable Read isolation level or lock the table in
exclusive mode to prevent concurrency problems.
For more information about standard tables, see the previous section,
“Advantages of logging tables” on page 6-30.
6. Re-create indexes, referential constraints, and unique constraints.

Quickly loading a new nonlogging table:

You quickly create a new nonlogging table and load the table.

To quickly create and load a new, large table:

1. Create a nonlogging table in a logged database.
The following sample SQL statements create a nonlogging table:
CREATE DATABASE history WITH LOG;
CONNECT TO DATABASE history;
CREATE RAW TABLE history (...
);
2. Load the table using a load utility such as dbexport or the High-Performance
Loader (HPL). For more information about dbexport and dbload, see the IBM
Informix Migration Guide. For more information about HPL, see the IBM Informix
High-Performance Loader User's Guide.
3. Perform a level-0 backup of the nonlogging table.
You must make a level-0 backup of any nonlogging table that has been
modified before you convert it to STANDARD type. The level-0 backup
provides a starting point from which to restore the data.
4. Change the nonlogging table to a logging table before you use it in a
transaction.
The following sample SQL statement changes a raw table to a standard table:
ALTER TABLE largetab TYPE(STANDARD);

Warning: Do not use nonlogging tables within a transaction where multiple

users can modify the data. If you need to use a nonlogging table within a

Chapter 6. Table performance considerations 6-31

 transaction, either set Repeatable Read isolation level or lock the table in
exclusive mode to prevent concurrency problems.
For more information about standard tables, see the previous section,
“Advantages of logging tables” on page 6-30.
5. Create indexes on columns most often used in query filters.
6. Create any referential constraints and unique constraints, if needed.

Dropping indexes for table-update efficiency

In some applications, you can confine most table updates to a single time period.
You can set up your system so that all updates are applied overnight or on
specified dates. When updates are performed as a batch, you can drop all
nonunique indexes while you make updates and then create new indexes
afterward.

This strategy can have two positive effects:

v The updating program runs much faster if it does not need to update indexes at
the same time that it updates tables.
v Re-created indexes are more efficient.

For more information about when to drop indexes, see “Nonunique indexes” on
page 7-12.

To load a table that has no indexes:

1. Drop the table (if it exists).
2. Create the table without specifying any unique constraints.
3. Load all rows into the table.
4. Alter the table to apply the unique constraints.
5. Create the nonunique indexes.

If you cannot guarantee that the loaded data satisfies all unique constraints, you
must create unique indexes before you load the rows. You save time if the rows are
presented in the correct sequence for at least one of the indexes. If you have a
choice, make it the row with the largest key. This strategy minimizes the number
of leaf pages that must be read and written.

Creating and enabling referential constraints efficiently

When you create or enable foreign-key constraints on existing tables that contain
data, you can sometimes achieve better performance by reducing the time that the
database server spends searching for violating rows.

By maintaining the referential integrity of the database during DML operations,

and by supporting efficient join-query execution paths on tables that are related by
a star schema, foreign-key constraints can improve the performance of DML
operations in databases where the primary key of each dimension table
corresponds to a foreign key of the fact table.

When you use the ALTER TABLE ADD CONSTRAINT or ALTER TABLE MODIFY
statement to define a foreign-key constraint on an existing table, you might be able
to reduce the time required to validate of the new foreign-key constraint, if the
referenced table already has a unique index or a primary-key constraint on the
column corresponding to the key of the foreign-key constraint. When it creates a
foreign-key constraint on a table that already contains data, the database server

6-32 IBM Informix Performance Guide

checks the table for any rows that violate the constraint. If an index exists, the
database server makes a cost-based decision whether to scan every row in the table
for violations, or to scan only the index valses.

For large tables, scanning only the index values can provide substantial
performance improvement, unless one of the following requirements is not
satisfied:
v The ALTER TABLE statement is creating only one foreign-key constraint.
v The ALTER TABLE statement is not also creating or enabling a CHECK
constraint.
v The ALTER TABLE statement is not also changing the data type of any existing
column in the table.
v The foreign-key columns do not include user-defined data types (UDTs) or
built-in opaque data types.
v The new mode of the foreign-key constraint is not DISABLED.
v The table is not associated with an active violation table.

Except in the case of one or more violating rows, the ALTER TABLE ADD
CONSTRAINT or ALTER TABLE MODIFY statement can create and validate a
foreign-key constraint when some of these requirements are not satisfied, but the
database server will not consider using the index-key algorithm to validate the
foreign-key constraint. The additional validation costs to scan the entire table tend
to be proportional to the size of the table.

Enabling a foreign-key constraint using index-scan validation

To validate the enabled foreign-key constraint, the database server performs a

full-table scan to search for violating rows, unless a unique index or a primary-key
constraint already exists on the foreign-key column values. In that case, the
database server consider using an index scan for validation, unless one or more of
the following requirements is not satisfied:
v The SET CONSTRAINTS statement is enabling only one foreign-key constraint.
v The same statement is not enabling a CHECK constraint.
v The foreign-key columns do not include user-defined data types (UDTs) or
built-in opaque data types.
v The new mode of the foreign-key constraint is not DISABLED.
v The table is not associated with an active violation table.

Unless the table has one or more violating rows, the SET CONSTRAINTS
statement can enable and validate a foreign-key constraint when some of these
requirements are not satisfied, but the database server will not consider using the
index-key algorithm to validate the foreign-key constraint. The additional
validation costs for a full table scan can be substantial for very large tables.

Skipping validation of foreign-key constraints

In both the ALTER TABLE and SET CONSTRAINTS operations described above,
the goal was to use a more efficient algorithm for validating the referential
constraint. Greater efficiencies can be achieved, at least temporarily, by postponing
or avoiding the validation of ENABLED or FILTERING foreign-key constraints that
are being created by ALTER TABLE ADD CONSTRAINT statements, or while a
DISABLED foreign-key constraint is being reset to an ENABLED or FILTERING
mode.

Chapter 6. Table performance considerations 6-33

 This feature can be useful when tables that enforced referential constraints need to
be moved from an OLTP environment to another database or to a data warehouse.
To export the tables and restore their constraints without validation might be
necessary if the time available for relocation is insufficient for violations checking,
The tables might seem unlikely to include violating rows, if the constraints were
dropped or disabled immediately before the tables were exported.

Three alternative mechanisms are available for bypassing the validation of enabled
or filtering foreign-key constraints while they are being created, or while they are
being exported, or while their mode is being changed from DISABLED:
v You can include the NOVALIDATE keyword in the constraint mode specification
– of the ALTER TABLE ADD CONSTRAINT statement,
– or of the SET CONSTRAINTS ENABLED statement,
– or of the SET CONSTRAINTS FILTERING WITH ERROR statement,
– or of the SET CONSTRAINTS FILTERING WITHOUT ERROR statements.
v If you plan to run multiple ALTER TABLE ADD CONSTRAINT or SET
CONSTRAINTS statements, run the SET ENVIRONMENT NOVALIDATE ON
statement to disable the validation of foreign-key constraints during the current
session.
Setting this session environment option makes NOVIOLATE the default mode
for enabled or filtering referential constraints while the DDL statement is
running.
v If you are migrating data, include the -nv option in the dbimport command.
The effect of the -nv option is that the constraint modes of any ALTER TABLE
ADD CONSTRAINT or SET CONSTRAINTS statements that create or enable
foreign-key constraints are processed so that the ENABLED, or FILTERING
WITH ERROR, or FILTERING WITHOUT ERROR constraint mode specifications
are instead implemented (respectively) as the ENABLED NOVALIDATE, or
FILTERING WITH ERROR NOVALIDATE, or FILTERING WITHOUT ERROR
NOVALIDATE modes.

In each case, no constraint validation of existing rows occurs during the DDL
statement.

The effect of the NOVALIDATE keyword or of the -nv command-line flag does not
persist outside the operation that created or changed the mode of the foreign-key
constraint. The same constraint enforces referential integrity during subsequent
DELETE, INSERT, MERGE, and UPDATE operations.

If the NOVALIDATE feature is used on a table that might already contains rows
that violate the foreign-key constraint, it is the responsibility of the user to verify
that no violating rows exist in the data.

Attaching or detaching fragments

You can use ALTER FRAGMENT ATTACH and DETACH statements to perform
data warehouse-type operations. ALTER FRAGMENT DETACH provides a way to
delete a segment of the table data rapidly. Similarly, ALTER FRAGMENT ATTACH
provides a way to load large amounts of data into an existing table incrementally
by taking advantage of the fragmentation technology.

6-34 IBM Informix Performance Guide

 For more information about how to take advantage of the performance
enhancements for the ATTACH and DETACH options of the ALTER FRAGMENT
statement, see “Improve the performance of operations that attach and detach
fragments” on page 9-19.

Altering a table definition

The database server uses one of these algorithms to process an ALTER TABLE
statement in SQL: slow alter, in-place alter, or fast alter.

Slow alter
When the database server uses the slow alter algorithm to process an ALTER
TABLE statement, the table can be unavailable to other users for a long period of
time.

The table might be unavailable because the database server:

v Locks the table in exclusive mode for the duration of the ALTER TABLE
operation
v Makes a copy of the table in order to convert the table to the new definition
v Converts the data rows during the ALTER TABLE operation
v Can treat the ALTER TABLE statement as a long transaction and abort it if the
LTXHWM threshold is exceeded

Because the database server makes a copy of the table to convert the table to the
new definition, a slow alter operation requires space at least twice the size of the
original table plus log space.

The database server uses the slow alter algorithm when the ALTER TABLE
statement makes column changes that it cannot perform in place:
v Adding or dropping a column created with the ROWIDS keyword
v Adding or dropping a column created with the REPLCHECK keyword
v Dropping a column of the TEXT or BYTE data type
v Modifying a SMALLINT column to SERIAL, SERIAL8, or BIGSERIAL
v Converting an INT column to SERIAL, SERIAL8, or BIGSERIAL
v Modifying the data type of a column so that some possible values of the old
data type cannot be converted to the new data type (For example, if you modify
a column of data type INTEGER to CHAR(n), the database server uses the slow
alter algorithm if the value of n is less than 11. An INTEGER requires 10
characters plus one for the minus sign for the lowest possible negative values.)
v Modifying the data type of a fragmentation column in a way that value
conversion might cause rows to move to another fragment
v Adding, dropping or modifying any column when the table contains
user-defined data types, smart large objects, or LVARCHAR, SET, MULTISET,
ROW, or COLLECTION data types
v Modifying the original size or reserve specifications of VARCHAR or
NVARCHAR columns
v Adding ERKEY shadow columns

In-place alter
The in-place alter algorithm provides numerous performance advantages over the
slow alter algorithm

The in-place alter algorithm:

Chapter 6. Table performance considerations 6-35

 v Increases table availability
Other users can access the table sooner when the ALTER TABLE operation uses
the in-place alter algorithm, because the database server locks the table for only
the time that it takes to update the table definition and rebuild indexes that
contain altered columns.
This increase in table availability can increase system throughput for application
systems that require 24 by seven operations.
When the database server uses the in-place alter algorithm, it locks the table for
a shorter time than the slow alter algorithm because the database server:
– Does not make a copy of the table to convert the table to the new definition
– Does not convert the data rows during the ALTER TABLE operation
– Alters the physical columns in place with the latest definition after the alter
operation when you later update or insert rows. The database server converts
the rows that reside on each page that you updated.
v Requires less space than the slow alter algorithm
When the ALTER TABLE operation uses the slow alter algorithm, the database
server makes a copy of the table to convert the table to the new definition. The
ALTER TABLE operation requires space at least twice the size of the original
table plus log space.
When the ALTER TABLE operation uses the in-place alter algorithm, the space
savings can be substantial for very large tables.
v Improves system throughput during the ALTER TABLE operation
The database server does not log any changes to the table data during the
in-place alter operation. Not logging changes has the following advantages:
– Log space savings can be substantial for very large tables.
– The alter operation is not a long transaction.

If the check_for_ipa Scheduler task is enabled, each table that has one or more
outstanding in-place alter operations is listed in the ph_alert table in the sysadmin
database. The alert text is: Table database:owner.table_name has outstanding in
place alters. The alert type is informative.
Related reference:
The ph_alert Table (Administrator's Reference)

Conditions for in-place alter operations:

The database server can use the in-place alter algorithm to process only certain
ADD, DROP, or MODIFY operations of the ALTER TABLE statement, and only if
the table schema or the ALTER TABLE statement does not require a slow alter
algorithm.

ALTER TABLE operations that can be done in place

The database server can use the in-place alter algorithm in the following ALTER
TABLE operations:
v Add columns of built-in data types, except the data types that are listed in
“Conditions that prevent in-place alter operations” on page 6-38.
v Drop a column of built-in data types, except a column that contains TEXT or
BYTE data types, or a column that was created with the ROWIDS keyword.
v In Enterprise Replication, add or drop a column that is created with the
CRCOLS keyword.

6-36 IBM Informix Performance Guide

v Modify a column for which the database server can convert all possible values
of the old data type to the new data type.
v Modify a column that is part of the fragmentation expression for its table, only if
value changes do not require any data row to move from one fragment to
another fragment after data type conversion.

The following table shows the conditions under which the ALTER TABLE MODIFY
statement uses the in-place alter algorithm to convert columns of supported data
types.

Key:
All = The database server uses the in-place alter algorithm for all cases of the
specific column operation.
nf = The database server uses the in-place alter algorithm when the modified
column is not part of the table fragmentation expression.
Table 6-2. MODIFY operations and conditions that use the in-place alter algorithm
Operation on Column Condition
Convert a SMALLINT column to an INTEGER column All
Convert a SMALLINT column to a BIGINT column All
Convert a SMALLINT column to an INT8 column All
Convert a SMALLINT column to a DEC(p2,s2) column p2-s2 >= 5
Convert a SMALLINT column to a DEC(p2) column p2-s2 >= 5 OR nf
Convert a SMALLINT column to a SMALLFLOAT column All
Convert a SMALLINT column to a FLOAT column All
Convert a SMALLINT column to a CHAR(n) column n >= 6 AND nf
Convert an INT column to an INT8 column All
Convert an INT column to a DEC(p2,s2) column p2-s2 >= 10
Convert an INT column to a DEC(p2) column p2 >= 10 OR nf
Convert an INT column to a SMALLFLOAT column nf
Convert an INT column to a FLOAT column All
Convert an INT column to a CHAR(n) column n >= 11 AND nf
Convert a SERIAL column to an INT8 column All
Convert a SERIAL column to a DEC(p2,s2) column p2-s2 >= 10
Convert a SERIAL column to a DEC(p2) column p2 >= 10 OR nf
Convert a SERIAL column to a SMALLFLOAT column nf
Convert a SERIAL column to a FLOAT column All
Convert a SERIAL column to a CHAR(n) column n >= 11 AND nf
Convert a SERIAL column to a BIGSERIAL column All
Convert a SERIAL column to a SERIAL8 column All
Convert a SERIAL8 column to a BIGSERIAL column All
Convert a BIGSERIAL column to a SERIAL8 column All
Convert a DEC(p1,s1) column to a SMALLINT column p1-s1 < 5 AND (s1 == 0
OR nf)
Convert a DEC(p1,s1) column to an INTEGER column p1-s1 < 10 AND (s1 == 0
OR nf)

Chapter 6. Table performance considerations 6-37

 Table 6-2. MODIFY operations and conditions that use the in-place alter
algorithm (continued)
Operation on Column Condition
Convert a DEC(p1,s1) column to an INT8 column p1-s1 < 20 AND (s1 == 0
OR nf)
Convert a DEC(p1,s1) column to a SERIAL column p1-s1 < 10 AND (s1 == 0
OR nf)
Convert a DEC(p1,s1) column to a BIGSERIAL column p1-s1 < 20 AND (s1 == 0
OR nf)
Convert a DEC(p1,s1) column to a SERIAL8 column p1-s1 < 20 AND (s1 == 0
OR nf)
Convert a DEC(p1,s1) column to a DEC(p2,s2) column p2-s2 >= p1-s1 AND (s2
>= s1 OR nf)
Convert a DEC(p1,s1) column to a DEC(p2) column p2 >= p1 OR nf
Convert a DEC(p1,s1) column to a SMALLFLOAT column nf
Convert a DEC(p1,s1) column to a FLOAT column nf
Convert a DEC(p1,s1) column to a CHAR(n) column n >= 8 AND nf
Convert a DEC(p1) column to a DEC(p2) column p2 >= p1 OR nf
Convert a DEC(p1) column to a SMALLFLOAT column nf
Convert a DEC(p1) column to a FLOAT column nf
Convert a DEC(p1) column to a CHAR(n) column n >= 8 AND nf
Convert a SMALLFLOAT column to a DEC(p2) column nf
Convert a SMALLFLOAT column to a FLOAT column nf
Convert a SMALLFLOAT column to a CHAR(n) column n >= 8 AND nf
Convert a FLOAT column to a DEC(p2) column nf
Convert a FLOAT column to a SMALLFLOAT column nf
Convert a FLOAT column to a CHAR(n) column n >= 8 AND nf
Convert a CHAR(m) column to a CHAR(n) column n >= m OR (nf AND not
ANSI mode)
Increase the length of a character-type column Not in ANSI mode
databases
Increase the length of a DECIMAL or MONEY column All

Conditions that prevent in-place alter operations

When the table contains an opaque data type, a user-defined data type, an
LVARCHAR data type, a BOOLEAN data type, or a smart large object (BLOB or
CLOB), the database server does not use the in-place alter algorithm, even when
the column that is being altered is of a data type that can support in-place alter
operations.

The in-place alter algorithm is not used if the ALTER TABLE DROP statement
specifies BYTE or TEXT columns, or the ROWIDS keyword, or if the ALTER
TABLE ADD statement includes the ROWID keyword.

If any column data types in an ALTER TABLE MODIFY statement cannot be

converted by in-place alter operations, or if data movement is required for a

6-38 IBM Informix Performance Guide

fragmented table, the database server uses the slow alter algorithm for data type
conversion instead of using the in-place alter algorithm.

For example, the database server does not use the in-place alter algorithm in the
following situations:
v When more than one algorithm is needed
For example, assume that an ALTER TABLE MODIFY statement converts a
SMALLINT column to a DEC(8,2) column and converts an INTEGER column to
a CHAR(8) column. The conversion of the first column is an in-place alter
operation, but the conversion of the second column is a slow alter operation.
The database server uses the slow alter algorithm to execute this statement.
v When the ALTER TABLE operation moves data records to another fragment
For example, suppose you have a table with two integer columns and the
following fragment expression:
col1 < col2 IN dbspace1, REMAINDER IN dbspace2
If you issue an ALTER TABLE MODIFY statement to convert the integer values
to character values, the database server stores the row (4, 30) in dbspace1
before the alter operation, but stores it in dbspace2 after the alter operation, not
as integers, 4 < 30, but as characters, ’30’ < ’4’.
v When the database server cannot convert all possible values of the old data type
to the new data type.
For example, you cannot convert a BIGSERIAL column to a SERIAL column,
because the modified column cannot store BIGSERIAL values that are beyond
the range of SERIAL values. (However, you can change a column from SERIAL
to BIGSERIAL with an in-place alter operation, if other columns in the table do
not conflict with any of the other restrictions on in-place alter operations.)
Related concepts:
IBM Informix data types (Database Design Guide)
Related reference:
DECIMAL (SQL Reference)

Performance considerations for DML statements:

The database server performs additional actions if it detects any down-level

version page during the execution of data manipulation language (DML)
statements (INSERT, UPDATE, DELETE, SELECT). These actions can impact
performance.

Each time you execute an ALTER TABLE statement that uses the in-place alter
algorithm, the database server creates a new version of the table structure. The
database server keeps track of all versions of table definitions. The database server
resets the version status and all of the version structures and alter structures until
the entire table is converted to the final format, or until a slow alter is performed.

If the database server detects any down-level version page during the execution of
DML statements (INSERT, UPDATE, DELETE, and SELECT statements, and
MERGE statements that specify Insert, Update, or Delete clauses), it performs the
following actions:
v For UPDATE statements, the database server converts the entire data page or
pages to the final format.

Chapter 6. Table performance considerations 6-39

 v For INSERT statements, the database server converts the inserted row to the
final format and inserts it in the best-fit page. The database server converts the
existing rows on the best-fit page to the final format.
v For DELETE statements, the database server does not convert the data pages to
the final format.
v For SELECT statements, the database server does not convert the data pages to
the final format.
If your query accesses rows that are not yet converted to the new table
definition, you might notice a slight degradation in the performance of your
individual query, because the database server reformats each row before it is
returned.

Performance of in-place alters for DDL operations:

In-place alter operations on data definition language (DDL) statements can impact
performance. Therefore, you should monitor outstanding in-place alters, because
many outstanding alters affects subsequent ALTER TABLE statements.

The oncheck -pT tablename option displays data-page versions for outstanding
in-place alter operations. An in-place alter is outstanding when data pages still exist
with the old definition.

Figure 6-12 displays a portion of the output that the following oncheck command
produces after four in-place alter operations are run on the customer
demonstration table:

oncheck -pT stores_demo:customer

...
Home Data Page Version Summary

Version Count

0 (oldest) 2
1 0
2 0
3 0
4 (current) 0
...

Figure 6-12. Sample oncheck -pT output for the customer table

The Count field in Figure 6-12 displays the number of pages that currently use that
version of the table definition. This oncheck output shows that four versions are
outstanding:
v A value of 2 in the Count field for the oldest version indicates that two pages
use the oldest version.
v A value of 0 in the Count fields for the next four versions indicates that no
pages were to the latest table definition.

Important: As you perform more in-place alters on a table, each subsequent

ALTER statement or the SQL statements that execute against the tables with
outstanding alters take more time to execute than the previous statement.
Therefore, regularly identify outstanding in-place alters and then run dummy

6-40 IBM Informix Performance Guide

UPDATE statements to force the outstanding in-place alters to complete. You might
want to identify the outstanding alters weekly or monthly, depending on your
environment.

If your goal is saving runtime CPU, then plan to keep as few outstanding alters
operations on a table as possible (generally no more than 3 or 4). If your goal is to
save on disk space and your alter operations add or grow columns, then leaving in
place alters outstanding helps reduce disk space.

In a replication environment, you might need to experiment to determine how to

best manage in-place alters.

You can convert data pages to the latest definition with a dummy UPDATE
statement. For example, the following statement, which sets a column value to the
existing value, causes the database server to convert data pages to the latest
definition:
UPDATE tab1 SET col1 = col1;

This statement does not change any data values, but it converts the format of the
data pages to the latest definition.

After an update is executed on all pages of the table, the oncheck -pT command
displays the total number of data pages in the Count field for the current version
of the table.
Related concepts:
Run dummy UPDATE statements (Migration Guide)

Altering a column that is part of an index:

If the altered column is part of an index, the table is still altered in place, but in
this case the database server rebuilds the index or indexes implicitly. If you do not
need to rebuild the index, you should drop or disable it before you perform the
alter operation. Taking these steps improves performance.

However, if the column that you modify is a primary key or foreign key and you
want to keep this constraint, you must specify those keywords again in the ALTER
TABLE statement, and the database server rebuilds the index.

For example, suppose you create tables and alter the parent table with the
following SQL statements:
CREATE TABLE parent
(si SMALLINT PRIMARY KEY CONSTRAINT pkey);
CREATE TABLE child
(si SMALLINT REFERENCES parent ON DELETE CASCADE
CONSTRAINT ckey);
INSERT INTO parent (si) VALUES (1);
INSERT INTO parent (si) VALUES (2);
INSERT INTO child (si) VALUES (1);
INSERT INTO child (si) VALUES (2);
ALTER TABLE parent
MODIFY (si INT PRIMARY KEY CONSTRAINT pkey);

This ALTER TABLE example converts a SMALLINT column to an INT column.

The database server retains the primary key because the ALTER TABLE statement
specifies the PRIMARY KEY keywords and the pkey constraint. When you specify
a PRIMARY KEY constraint in the MODIFY clause, the database server also
silently creates a NOT NULL constraint on the same primary key column.
Chapter 6. Table performance considerations 6-41
 However, the database server drops any referential constraints to that primary key.
Therefore, you must also specify the following ALTER TABLE statement for the
child table:
ALTER TABLE child
MODIFY (si int references parent on delete cascade
constraint ckey);

Even though the ALTER TABLE operation on a primary key or foreign key column
rebuilds the index, the database server still takes advantage of the in-place alter
algorithm. The in-place alter algorithm can provide performance benefits, including
the following:
v It does not make a copy of the table in order to convert the table to the new
definition.
v It does not convert the data rows during the alter operation.
v It does not rebuild all indexes on the table.

Warning: If you alter a table that is part of a view, you must re-create the view to
obtain the latest definition of the table.

Fast alter
The database server uses the fast alter algorithm when the ALTER TABLE
statement changes attributes of the table but does not affect the data.

The database server uses the fast alter algorithm when you use the ALTER TABLE
statement to:
v Change the next-extent size.
v Add or drop a constraint.
v Change the lock mode of the table.
v Change the unique index attribute without modifying the column type.
v Add shadow columns for row versioning with the ADD VERCOLS keywords.

With the fast alter algorithm, the database server holds the lock on the table for
just a short time. In some cases, the database server locks the system catalog tables
only to change the attribute. In either case, the table is unavailable for queries for
only a short time.

Denormalize the data model to improve performance

You might need to denormalize the data model to reduce overhead and optimize
performance.

The entity-relationship data model, which the IBM Informix Guide to SQL: Tutorial
describes, produces tables that contain no redundant or derived data. According to
the tenets of relational database theory, these tables are well structured.

Sometimes, to meet extraordinary demands for high performance, you might need
to denormalize the data model by modifying it in ways that are undesirable from a
theoretical standpoint. This section describes some modifications and their
associated costs.

Shortening rows
Usually, tables with shorter rows yield better performance than those with longer
rows because disk I/O is performed in pages, not in rows. The shorter the rows of
a table, the more rows occur on a page. The more rows per page, the fewer I/O

6-42 IBM Informix Performance Guide

 operations it takes to read the table sequentially, and the more likely it is that a
nonsequential access can be performed from a buffer.

The entity-relationship data model puts all the attributes of one entity into a single
table for that entity. For some entities, this strategy can produce rows of awkward
lengths.

To shorten the rows, you can break columns into separate tables that are associated
by duplicate key values in each table. As the rows get shorter, query performance
should improve.

Expelling long strings

The most bulky attributes are often character strings. To make the rows shorter,
you can remove long strings from the entity table.

You can use the following methods to expel long strings:

v Use VARCHAR columns.
v Use TEXT data.
v Move strings to a companion table.
v Build a symbol table.

Convert CHAR columns into VARCHAR columns to shorten rows

(GLS)
A database might contain CHAR columns that you can convert to VARCHAR
columns. You can use a VARCHAR column to shorten the average row length
when the average length of the text string in the CHAR column is at least 2 bytes
shorter than the width of the column.

VARCHAR data is immediately compatible with most existing programs, forms,

and reports. You might need to recompile any forms produced by application
development tools to recognize VARCHAR columns. Always test forms and
reports on a sample database after you modify the table schema.

For information about other character data types, see the IBM Informix GLS User's
Guide.

Convert a long string to a TEXT data type column

When a string fills half a disk page or more, consider converting it to a TEXT data
type column in a separate blobspace.

The column within the row page is only 56 bytes long, which allows more rows on
a page than when you include a long string. However, the TEXT data type is not
automatically compatible with existing programs. The application needed to fetch a
TEXT value is a bit more complicated than the code for fetching a CHAR value
into a program.

Move strings to a companion table

Strings that are less than half a page waste disk space if you treat them as TEXT
data, but you can move them from the main table to a companion table.

If you split a table into two tables, the primary table and a companion table, repeat
the primary key in each table.

Chapter 6. Table performance considerations 6-43

 Build a symbol table
If a column contains strings that are not unique in each row, you can move those
strings to a table in which only unique copies are stored.

For example, the customer.city column contains city names. Some city names are
repeated in the column, and most rows have some trailing blanks in the field.
Using the VARCHAR data type eliminates the blanks but not the duplication.

You can create a table named cities, as the following example shows:
CREATE TABLE cities (
city_num SERIAL PRIMARY KEY,
city_name VARCHAR(40) UNIQUE
)

You can change the definition of the customer table so that its city column
becomes a foreign key that references the city_num column in the cities table.

To insert the city of the new customer into cities, you must change any program
that inserts a new row into customer. The database server return code in the
SQLCODE field of the SQL Communications Area (SQLCA) can indicate that the
insert failed because of a duplicate key. It is not a logical error; it simply means
that an existing customer is located in that city. For more information about the
SQLCA, see the IBM Informix Guide to SQL: Tutorial.

Besides changing programs that insert data, you must also change all programs
and stored queries that retrieve the city name. The programs and stored queries
must use a join to the new cities table in order to obtain their data. The extra
complexity in programs that insert rows and the extra complexity in some queries
is the result of giving up theoretical correctness in the data model. Before you
make the change, be sure that it returns a reasonable savings in disk space or
execution time.

Splitting wide tables

Consider all the attributes of an entity that has rows that are too wide for good
performance. Look for some theme or principle to divide them into two groups.
Then split the table into two tables, a primary table and a companion table,
repeating the primary key in each one.

The shorter rows allow you to query or update each table quickly.

Division by Bulk

One principle on which you can divide an entity table is bulk. Move the bulky
attributes, which are usually character strings, to the companion table. Keep the
numeric and other small attributes in the primary table. In the demonstration
database, you can split the ship_instruct column from the orders table. You can
call the companion table orders_ship. It has two columns, a primary key that is a
copy of orders.order_num and the original ship_instruct column.

Division by Frequency of Use

Another principle for division of an entity is frequency of use. If a few attributes

are rarely queried, move them to a companion table. In the demonstration
database, for example, perhaps only one program queries the ship_instruct,
ship_weight, and ship_charge columns. In that case, you can move them to a
companion table.

6-44 IBM Informix Performance Guide

 Division by Frequency of Update

Updates take longer than queries, and updating programs lock index pages and
rows of data during the update process, preventing querying programs from
accessing the tables. If you can separate one table into two companion tables, one
with the most-updated entities and the other with the most-queried entities, you
can often improve overall response time.

Performance Costs of Splitting Tables

Splitting a table uses extra disk space and adds complexity. Two copies of the
primary key occur for each row, one copy in each table. Two primary-key indexes
also exist. You can use the methods described in earlier sections to estimate the
number of added pages.

You must modify existing programs, reports, and forms that use SELECT * because
fewer columns are returned. Programs, reports, and forms that use attributes from
both tables must perform a join to bring the tables together.

In this case, when you insert or delete a row, two tables are altered instead of one.
If you do not coordinate the alteration of the two tables (by making them within a
single transaction, for example), you lose semantic integrity.

Redundant data
Normalized tables contain no redundant data. Every attribute appears in only one
table.

Normalized tables also contain no derived data. Instead, data that can be
computed from existing attributes is selected as an expression based on those
attributes.

Normalizing tables minimizes the amount of disk space used and makes updating
the tables as easy as possible. However, normalized tables can force you to use
joins and aggregate functions often, and those processes can be time consuming.

As an alternative, you can introduce new columns that contain redundant data,
provided you understand the trade-offs involved.

Adding redundant data

A correct data model avoids redundancy by keeping any attribute only in the table
for the entity that it describes. If the attribute data is needed in a different context,
you join tables to make the connection. But joining takes time. If a frequently used
join affects performance, you can eliminate it by duplicating the joined data in
another table.

In the stores_demo database, the manufact table contains the names of

manufacturers and their delivery times. An actual working database might contain
many other attributes of a supplier, such as address and sales representative name.

The contents of manufact are primarily a supplement to the stock table. Suppose
that a time-critical application frequently refers to the delivery lead time of a
particular product but to no other column of manufact. For each such reference,
the database server must read two or three pages of data to perform the lookup.

Chapter 6. Table performance considerations 6-45

 You can add a new column, lead_time, to the stock table and fill it with copies of
the lead_time column from the corresponding rows of manufact. That
arrangement eliminates the lookup and therefore speeds up the application.

Like derived data, redundant data takes space and poses an integrity risk. In the
example described in the previous paragraph, many extra copies of the lead time
for each manufacturer can exist. (Each manufacturer can appear in stock many
times.) The programs that insert or update a row of manufact must also update
multiple rows of stock.

The integrity risk is simply that the redundant copies of the data might not be
accurate. If a lead time is changed in manufact, the stock column is outdated until
it is also updated. As you do with derived data, define the conditions under which
redundant data might be wrong.

For more information about database design, see the IBM Informix Database Design
and Implementation Guide.

Reduce disk space in tables with variable length rows

You can enable the database server to insert more rows per page into tables with
variable-length rows, if you set the MAX_FILL_DATA_PAGES configuration
parameter to 1. Allowing more variable length rows per page has advantages and
disadvantages.

Potential advantages of allowing more variable length rows per page are:
v Reducing the disk space required to store data
v Enabling the server to use the buffer pool more efficiently
v Reducing table scan times

Possible disadvantages of using the MAX_FILL_DATA_PAGES allowing more

variable length rows per page are:
v The server might store rows in a different physical order.
v As the page fills, updates made to the variable-length columns in a row could
cause the row to expand so it no longer completely fits on the page. This causes
the server to split the row onto two pages, increasing the access time for the
row.

If the MAX_FILL_DATA_PAGES configuration parameter is enabled, the server

will add a new row to a recently modified page with existing rows if adding the
row leaves at least 10 percent of the page free for future expansion of all the rows
in the page. If the MAX_FILL_DATA_PAGES configuration parameter is not
enabled, the server will add the row only if there is sufficient room on the page to
allow the new row to grow to its maximum length.

If you enable the MAX_FILL_DATA_PAGES configuration parameter and you

want this to affect existing variable length rows, the existing tables must be
reloaded.

Reduce disk space by compressing tables and fragments

You can reduce disk space by compressing and uncompressing data in tables and
table fragments, consolidating free space in a table or fragment, and returning this
free space to the dbspace. Compression is advantageous for applications with a lot
of I/O activity and for applications in which the reduction of disk space usage is
6-46 IBM Informix Performance Guide
critical. However, if your applications run with high buffer cache hit ratios and
high performance is more important than space usage, you might not want to use
the compression feature, since compression might slightly decrease performance.

Compressing data, consolidating data, and returning free space have the following
benefits:
v Significant savings in disk storage space
v Reduced disk usage for compressed fragments
v Significant saving of logical log usage, which saves additional space and can
prevent bottlenecks for high-throughput OLTP after the compression operation is
completed.
v Fewer page reads, because more rows can fit on a page
v Smaller buffer pools, because more data fits in the same size pool
v Reduced I/O activity, because:
– More compressed rows than uncompressed rows fit on a page
– Log records for insert, update, and delete operations of compressed rows are
smaller
v Ability to compress older fragments of time-fragmented data that are not often
accessed, while leaving more recent data that is frequently accessed in
uncompressed form
v Ability to free space no longer needed for a table
v Faster backup and restore

Since compressed data covers fewer pages and has more rows per page than
uncompressed data, the query optimizer might choose different plans after
compression.

For more information, see the IBM Informix Administrator's Guide and the IBM
Informix Administrator's Reference.

Chapter 6. Table performance considerations 6-47

6-48 IBM Informix Performance Guide
Chapter 7. Indexes and index performance considerations
Informix provides several types of indexes. Some performance issues are associated
with indexes.

Types of indexes
Informix uses B-tree indexes, R-tree indexes, functional indexes, and indexes that
DataBlade modules provide for user-defined data. The server also uses forest of
trees (FOT) indexes, which are alternatives to B-tree indexes.
Related concepts:
“What is a functional index?” on page 7-26

B-tree indexes
Informix uses a B-tree index for columns that contain built-in data types (referred
to as a traditional B-tree index), columns that contain one-dimensional user-defined
data types (referred to as a generic B-tree index), and values that a user-defined data
type returns.

Built-in data types include character, datetime, integer, float, and so forth. For more
information about built-in data types, see IBM Informix Guide to SQL: Reference.

User-defined data types include opaque and distinct data types. For more
information about user-defined data types, see IBM Informix User-Defined Routines
and Data Types Developer's Guide.

The return value of a user-defined function can be a built-in or user-defined data

type, but not a simple large object (TEXT or BYTE data type) or a smart large
object (BLOB or CLOB data type). For more information about how to use
functional indexes, see “Using a functional index” on page 7-25.

For information about how to estimate B-tree index size, see “Estimating index
pages” on page 7-4.

Structure of conventional index pages

A conventional index is arranged as a hierarchy of pages (technically, a B-tree).

The following figure shows the B-tree structure of an index. The topmost level of
the hierarchy contains a single root page. Intermediate levels, when needed, contain
branch pages. Each branch page contains entries that see a subset of pages in the
next level of the index. The bottom level of the index contains a set of leaf pages.
Each leaf page contains a list of index entries that see rows in the table.

© Copyright IBM Corp. 1996, 2014 7-1

 Root page

Branch page Branch page

Leaf page Leaf page

Figure 7-1. B-tree structure of an index

The number of levels needed to hold an index depends on the number of unique
keys in the index and the number of index entries that each page can hold. The
number of entries per page depends, in turn, on the size of the columns being
indexed.

If the index page for a given table can hold 100 keys, a table of up to 100 rows
requires a single index level: the root page. When this table grows beyond 100
rows, to a size between 101 and 10,000 rows, it requires a two-level index: a root
page and between 2 and 100 leaf pages. When the table grows beyond 10,000 rows,
to a size between 10,001 and 1,000,000 rows, it requires a three-level index: the root
page, a set of 100 branch pages, and a set of up to 10,000 leaf pages.

Index entries contained within leaf pages are sorted in key-value order. An index
entry consists of a key and one or more row pointers. The key is a copy of the
indexed columns from one row of data. A row pointer provides an address used to
locate a row that contains the key. A unique index contains one index entry for
every row in the table.

For information about special indexes for Informix, see “Indexes on user-defined
data types” on page 7-21.
Related concepts:
“Forest of trees indexes”

Forest of trees indexes

A forest of trees index is like a B-tree index, but it has multiple root nodes and
potentially fewer levels. Multiple root nodes can alleviate root node contention,
because more concurrent users can access the index. A forest of trees index can also
improve the performance of a query by reducing the number of levels involved in
buffer read operations.

You can create a forest of trees index as an alternative to a B-Tree index, but not as
an alternative to an R-Tree index or other types of indexes.

Unlike a traditional B-tree index, which contains one root node, a forest of trees
index is a large B-Tree index that is divided into smaller subtrees (which you can
think of as buckets). These subtrees contain multiple root nodes and leaves. The
following figure shows the structure of a forest of trees index.

7-2 IBM Informix Performance Guide

 Subtree 1 Subtree 2 Subtree 3

Root nodes

Leaves

Figure 7-2. Structure of a forest of trees index

Informix stores and retrieves an item from a subtree by:

1. Computing a hash value from the columns that you selected when creating the
index.
2. Mapping the hash value to a subtree for storage or retrieval of the row.

Forest of trees indexes are detached indexes. The server does not support forest of
trees attached indexes.

You create a forest of trees index with the CREATE INDEX statement of SQL and
the HASH ON clause.

You enable or disable forest of trees indexes with the SET INDEXES statement of
SQL.

You can identify a forest of trees index by the FOT indicator in the Index Name field
in SET EXPLAIN output.

You can look up the number of hashed columns and subtrees in a forest of trees
index by viewing information in the sysindices table for the database containing
tables that have forest of trees indexes.

The server treats a forest of trees index the same way it treats a B-tree index.
Therefore, in a logged database, you can control how the B-tree scanner threads
remove deletions from both forest of trees and B-tree indexes.

Restrictions: You cannot:

v Create forest of trees indexes on columns with complex data types, UDTs, or
functional columns.
v Use the FILLFACTOR option of the CREATE INDEX statement when you create
forest of trees indexes, because the indexes are built from top to bottom.
v Create clustered forest of trees indexes.
v Run the ALTER INDEX statement on forest of trees indexes.
v Run the SET INDEXES statement on forest of trees indexes in a database of
secondary servers within a cluster environment.
v Use forest of trees indexes in queries that use aggregates, including minimum
and maximum range values.
v Perform range scans directly on the HASH ON columns of a forest of trees
index.
However, you can perform range scans on columns that are not listed in the
HASH ON column list. For range scans on columns listed in HASH ON column
list, you must create an additional B-tree index that contains the appropriate
column list for the range scan. This additional B-tree index might have the same
column list as the forest of trees index, plus or minus a column.

Chapter 7. Indexes and index performance considerations 7-3

 v Use a forest of trees index for an OR index path. The database server does not
use forest of trees indexes for queries that have an OR predicate on the indexed
columns.
Related concepts:
“Structure of conventional index pages” on page 7-1
HASH ON clause (SQL Syntax)
Related tasks:
“Improve query performance with a forest of trees index” on page 7-13
“Detecting root node contention” on page 7-13
“Creating a forest of trees index” on page 7-14
“Disabling and enabling a forest of trees index” on page 7-15
“Determining if you are using a forest of trees index” on page 7-16
Related reference:
CREATE INDEX statement (SQL Syntax)

R-tree indexes
Informix uses an R-tree index for spatial data (such as two-dimensional or
three-dimensional data).

For information about sizing an R-tree index, see the IBM Informix R-Tree Index
User's Guide.

Indexes that DataBlade modules provide

DataBlade modules can contain user-defined data types. A DataBlade module can
also provide a user-defined index for the new data type.

For example, the Excalibur Text Search DataBlade provides an index to search text
data. For more information, see the IBM Informix Excalibur Text Search DataBlade.

For more information about the types of data and functions that each DataBlade
module provides, see the user guide of each DataBlade module. For information
about how to determine the types of indexes available in your database, see
“Identifying the available access methods” on page 7-24.

Estimating index pages

The index pages associated with a table can add significantly to the size of a
dbspace.

By default, the database server creates the index in the same dbspace as the table,
but in a separate tblspace from the table. To place the index in a separate dbspace,
specify the IN keyword in the CREATE INDEX statement.

Although you cannot explicitly specify the extent size of an index, you can
estimate the number of pages that an index might occupy to determine if your
dbspace or dbspaces have enough space allocated.

Index extent sizes

The database server determines the extent size of an index based on the extent size
for the corresponding table, regardless of whether the index is fragmented or not
fragmented.

7-4 IBM Informix Performance Guide

 Formula for estimating the extent size of an attached index
For an attached index, the database server uses the ratio of the index key size to
the row size to assign an appropriate extent size for the index.

The following formula shows how the database server uses the ratio of the index
key size to the row size:
Index extent size = (index_key_size /
table_row_size) *
table_extent_size

In this formula:
v index_key_size is the total widths of the indexed column or columns plus 5 for
a key descriptor.
v table_row_size is the sum of all the columns in the row.
v table_extent_size is the value that you specify in the EXTENT SIZE keyword of
the CREATE TABLE statement.

If the index is not unique, then the extent size is reduced by 20 percent.

The database server also uses this same ratio for the next-extent size for the index:
Index next extent size =
(index_key_size/table_row_size)*
table_next_extent_size

Formula for estimating the extent size of a detached index

For a detached index, the database server uses the ratio of the index key size plus
some overhead bytes to the row size to assign an appropriate extent size for the
index.

The following formula shows how the database server uses the ratio of the index
key size plus some overhead bytes to the row size:
Detached Index extent size = ( (index_key_size +
9) / table_row_size) *
table_extent_size

For example, suppose you have the following values:

index_key_size = 8 bytes
table_row_size = 33 bytes
table_extent_size = 150 * 2-kilobyte page

The above formula calculates the extent size as follows:

Detached Index extent size = ( (8 + 9) /
33) * 150 * 2-kilobyte page
= (17/33) * 300 kilobytes
= 154 kilobytes

Important: For a non-unique index, the formula calculates an extent size that is
reduced by 20 percent.

Estimating conventional index pages

You can estimate the size of index pages, using a series of formulas.

To estimate the number of index pages:

1. Add up the total widths of the indexed column or columns.

Chapter 7. Indexes and index performance considerations 7-5

 This value is referred to as colsize. Add 4 to colsize to obtain keysize, the actual
size of a key in the index. For example, if colsize is 6, the value of keysize is 10.
2. Calculate the expected proportion of unique entries to the total number of
rows.
The formulas in subsequent steps see this value as propunique.
If the index is unique or has few duplicate values, use 1 for propunique.
If a significant proportion of entries are duplicates, divide the number of
unique index entries by the number of rows in the table to obtain a fractional
value for propunique. For example, if the number of rows in the table is
4,000,000 and the number of unique index entries is 1,000,000, the value of
propunique is .25.
If the resulting value for propunique is less than .01, use .01 in the calculations
that follow.
3. Estimate the size of a typical index entry with one of the following formulas,
depending on whether the table is fragmented or not:
a. For nonfragmented tables, use the following formula:
entrysize = (keysize * propunique) + 5 + 4
The value 5 represents the number of bytes for the row pointer in a
nonfragmented table.
For nonunique indexes, the database server stores the row pointer for each
row in the index node but stores the key value only once. The entrysize
value represents the average length of each index entry, even though some
entries consist of only the row pointer.
For example, if propunique is .25, the average number of rows for each
unique key value is 4. If keysize is 10, the value of entrysize is 11.5,
calculated as (10 * 0.25) + 5 + 4 = 2.5 + 9 = 11.5. The following calculation
shows the space required for all four rows:
space for four rows = 4 * 11.5 = 46
This space requirement is the same when you calculate it for the key value
and add the four row pointers, as the following formula shows:
space for four rows = 10 + (4 * 9) = 46
b. For fragmented tables, use the following formula:
entrysize = (keysize * propunique) + 9 + 4
The value 9 represents the number of bytes for the row pointer in a
fragmented table.
4. Estimate the number of entries per index page with the following formula:
pagents = trunc(pagefree/entrysize)
In this formula:
v pagefree is the page size minus the page header (2020 for a 2-kilobyte page
size).
v entrysize is the size of a typical index entry, which you estimated in the
previous step.
The trunc() function notation indicates that you should round down to the
nearest integer value.
5. Estimate the number of leaf pages with the following formula:
leaves = ceiling(rows/pagents)
In this formula:
v rows is the number of rows that you expect to be in the table.

7-6 IBM Informix Performance Guide

 v pagents is the number of entries per index page, which you estimated in
the previous step.
The ceiling() function notation indicates that you should round up to the
nearest integer value.
6. Estimate the number of branch pages at the second level of the index with the
following formula:
branches0 = ceiling(leaves/node_ents)
Calculate the value for node_ents with the following formula:
node_ents = trunc( pagefree / ( keysize + 4) + 4)
In this formula:
v pagefree is the page size minus the page header (2020 for a 2-kilobyte page
size).
v keysize is the colsize plus 4. You obtained this value in step 1.
In the formula, 4 represents the number of bytes for the leaf node pointer.
7. If the value of branches0 is greater than 1, more levels remain in the index.
To calculate the number of pages contained in the next level of the index, use
the following formula:
branchesn+1 = ceiling(branchesn/node_ents)
In this formula:
v branchesn is the number of branches for the last index level that you
calculated.
v branchesn+1 is the number of branches in the next level.
v node_ents is the value that you calculated in step 6.
8. Repeat the calculation in step 7 for each level of the index until the value of
branchesn+1 equals 1.
9. Add up the total number of pages for all branch levels calculated in steps 6
through 8. This sum is called branchtotal.
10. Use the following formula to calculate the number of pages in the compact
index:
compactpages = (leaves + branchtotal)
11. If your database server instance uses a fill factor for indexes, the size of the
index increases.
The default fill factor value is 90 percent. You can change the fill factor value
for all indexes with the FILLFACTOR configuration parameter. You can also
change the fill factor for an individual index with the FILLFACTOR clause of
the CREATE INDEX statement in SQL.
To incorporate the fill factor into your estimate for index pages, use the
following formula:
indexpages = 100 * compactpages / FILLFACTOR

The preceding estimate is a guideline only. As rows are deleted and new ones are
inserted, the number of index entries can vary within a page. This method for
estimating index pages yields a conservative (high) estimate for most indexes. For
a more precise value, build a large test index with real data and check its size with
the oncheck utility.

Tip: A forest of trees index can be larger than a B-Tree index. When you estimate
the size of a forest of trees index, the estimates apply to each subtree in the index.
Then, you must aggregate the buckets to calculate the total estimation.

Chapter 7. Indexes and index performance considerations 7-7

Managing indexes
An index on the appropriate column can save thousands, tens of thousands, or in
extreme cases, even millions of disk operations during a query. However, indexes
entail costs.

An index is necessary on any column or combination of columns that must be

unique. However, as discussed in Chapter 10, “Queries and the query optimizer,”
on page 10-1, the presence of an index can also allow the query optimizer to speed
up a query.

The optimizer can use an index in the following ways:

v To replace repeated sequential scans of a table with nonsequential access
v To avoid reading row data when processing expressions that name only indexed
columns
v To avoid a sort (including building a temporary table) when executing the
GROUP BY and ORDER BY clauses
Related concepts:
“Using a functional index” on page 7-25

Space costs of indexes

The first cost of an index is disk space. The presence of an index can add many
pages to a dbspace; it is easy to have as many index pages as row pages in an
indexed table. Additionally, in an environment where multiple languages are used,
indexes created for each language require additional disk space.

When you consider space costs, also consider whether increasing the page size of a
standard or temporary dbspace is beneficial in your environment. If you want a
longer key length than is available for the default page size, you can increase the
page size. If you increase the page size, the size must be an integral multiple of the
default page size, not greater than 16K bytes.

You might not want to increase the page size if your application contains small
sized rows. Increasing the page size for an application that randomly accesses
small rows might decrease performance. In addition, a page lock on a larger page
will lock more rows, reducing concurrency in some situations.
Related concepts:
B-tree index compression (Administrator's Guide)

Time costs of indexes

The second cost of an index is time whenever the table is modified.

The following descriptions assume that approximately two pages must be read to
locate an index entry. That is the case when the index consists of a root page, one
level of branch pages, and a set of leaf pages. The root page is assumed to be in a
buffer already. The index for a very large table has at least two intermediate levels,
so about three pages are read when the database server references such an index.

Presumably, one index is used to locate a row being altered. The pages for that
index might be found in page buffers in shared memory for the database server.
However, the pages for any other indexes that need altering must be read from
disk.

7-8 IBM Informix Performance Guide

 Under these assumptions, index maintenance adds time to different kinds of
modifications, as the following list shows:
v When you delete a row from a table, the database server must delete its entries
from all indexes.
The database server must look up the entry for the deleted row (two or three
pages in) and rewrite the leaf page. The write operation to update the index is
performed in memory, and the leaf page is flushed when the least recently used
(LRU) buffer that contains the modified page is cleaned. This operation requires
two or three page accesses to read the index pages if needed and one deferred
page access to write the modified page.
v When you insert a row, the database server must insert its entries in all indexes.
The database server must find a place in which to enter the inserted row within
each index (two or three pages in) and rewrite (one deferred page out), for a
total of three or four immediate page accesses per index.
v When you update a row, the database server must look up its entries in each
index that applies to an altered column (two or three pages in).
The database server must rewrite the leaf page to eliminate the old entry (one
deferred page out) and then locate the new column value in the same index
(two or three more pages in) and the row entered (one more deferred page out).

Insertions and deletions change the number of entries on a leaf page. Although
virtually every pagents operation requires some additional work to deal with a leaf
page that has either filled or been emptied, if pagents is greater than 100, this
additional work occurs less than 1 percent of the time. You can often disregard it
when you estimate the I/O impact.

In short, when a row is inserted or deleted at random, allow three to four added
page I/O operations per index. When a row is updated, allow six to eight page
I/O operations for each index that applies to an altered column. If a transaction is
rolled back, all this work must be undone. For this reason, rolling back a
transaction can take a long time.

Because the alteration of the row itself requires only two page I/O operations,
index maintenance is clearly the most time-consuming part of data modification.
For information about one way to reduce this cost, see “Clustering” on page 7-11.

Unclaimed index space

A background thread, the B-tree scanner, identifies an index with the most
unclaimed index space. Unclaimed index space degrades performance and causes
extra work for the server. When an index is chosen for scanning, the entire leaf of
the index is scanned for deleted (dirty) items that were committed, but not yet
removed from the index. The B-tree scanner removes these items when necessary.

The B-tree scanner allows multiple threads.

Use the BTSCANNER configuration parameter to specify the number of B-tree

scanner threads to start and the priority of the B-tree scanner threads when the
database server starts. For details, see the IBM Informix Administrator's Reference.

You can invoke the B-tree scanner from the command line.

Chapter 7. Indexes and index performance considerations 7-9

 Indexes on columns
You can create an index for one or more columns in a table. Indexes are required
on columns that must be unique and are not specified as primary keys.

In addition, you must add an index on columns that:

v Are used in joins that are not specified as foreign keys
v Are frequently used in filter expressions
v Are frequently used for ordering or grouping
v Do not involve duplicate keys
v Are amenable to clustered indexing

Filtered columns in large tables

If a column is often used to filter the rows of a large table, consider placing an
index on it. The optimizer can use the index to select the wanted columns and
avoid a sequential scan of the entire table.

Suppose you have a table that contains a large mailing list. If you find that a
postal-code column is often used to filter a subset of rows, consider putting an
index on that column.

This strategy yields a net savings of time only when the selectivity of the column
is high; that is, when only a small fraction of rows holds any one indexed value.
Nonsequential access through an index takes several more disk I/O operations
than sequential access does, so if a filter expression on the column passes more
than a fourth of the rows, the database server might as well read the table
sequentially.

As a rule, indexing a filter column saves time in the following cases:

v The column is used in filter expressions in many queries or in slow queries.
v The column contains at least 100 unique values.
v Most column values appear in fewer than 10 percent of the rows.

Order-by and group-by columns

You can place an index on the ordering column or columns of a table. The
database server then uses the index that to sort the query results in the most
efficient manner.

When a large quantity of rows must be ordered or grouped, the database server
must put the rows in order. One way that the database server performs this task is
to select all the rows into a temporary table and sort the table. But, as explained in
Chapter 10, “Queries and the query optimizer,” on page 10-1, if the ordering
columns are indexed, the optimizer sometimes reads the rows in sorted order
through the index, thus avoiding a final sort.

Because the keys in an index are in sorted sequence, the index really represents the
result of sorting the table. By placing an index on the ordering column or columns,
you can replace many sorts during queries with a single sort when the index is
created.

Avoiding columns with duplicate keys

Duplicate keys in indexes can cause performance problems. You can take steps to
avoid these problems.

7-10 IBM Informix Performance Guide

When duplicate keys are permitted in an index, entries that match a given key
value are grouped in lists. The database server uses these lists to locate rows that
match a requested key value. When the selectivity of the index column is high,
these lists are generally short. But when only a few unique values occur, the lists
become long and can cross multiple leaf pages.

Placing an index on a column that has low selectivity (that is, a small number of
distinct values relative to the number of rows) can reduce performance. In such
cases, the database server must not only search the entire set of rows that match
the key value, but it must also lock all the affected data and index pages. This
process can impede the performance of other update requests as well.

To correct this problem, replace the index on the low-selectivity column with a
composite index that has a higher selectivity. Use the low-selectivity column as the
leading column and a high-selectivity column as your second column in the index.
The composite index limits the number of rows that the database server must
search to locate and apply an update.

You can use any second column to disperse the key values as long as its value
does not change, or changes at the same time as the real key. The shorter the
second column the better, because its values are copied into the index and expand
its size.

Clustering
Clustering is a method for arranging the rows of a table so that their physical
order on disk closely corresponds to the sequence of entries in the index.

(Do not confuse the clustered index with an optical cluster, which is a method for
storing logically related TEXT or BYTE data together on an optical volume.)

When you know that a table is ordered by a certain index, you can avoid sorting.
You can also be sure that when the table is searched on that column, it is read
effectively in sequential order, instead of nonsequentially. These points are covered
in Chapter 10, “Queries and the query optimizer,” on page 10-1.

Tip: For information about eliminating interleaved extents by altering an index to

cluster, see “Creating or altering an index to cluster” on page 6-25.

In the stores_demo database, the orders table has an index, zip_ix, on the
postal-code column. The following statement causes the database server to put the
rows of the customer table in descending order by postal code:
ALTER INDEX zip_ix TO CLUSTER

To cluster a table on a nonindexed column, you must create an index. The

following statement reorders the orders table by order date:
CREATE CLUSTER INDEX o_date_ix ON orders (order_date ASC)

To reorder a table, the database server must copy the table. In the preceding
example, the database server reads all the rows in the table and constructs an
index. Then it reads the index entries in sequence. For each entry, it reads the
matching row of the table and copies it to a new table. The rows of the new table
are in the desired sequence. This new table replaces the old table.

Clustering is not preserved when you alter a table. When you insert new rows,
they are stored physically at the end of the table, regardless of their contents.

Chapter 7. Indexes and index performance considerations 7-11

 When you update rows and change the value of the clustering column, the rows
are written back into their original location in the table.

Clustering can be restored after the order of rows is disturbed by ongoing updates.
The following statement reorders the table to restore data rows to the index
sequence:
ALTER INDEX o_date_ix TO CLUSTER

Reclustering is usually quicker than the original clustering because reading out the
rows of a nearly clustered table is similar in I/O impact to a sequential scan.

Clustering and reclustering take a lot of space and time. To avoid some clustering,
build the table in the desired order initially.

Configuration parameters that affect the degree of clustering:

The clust field in the sysindexes or the sysindices table represents the degree of
clustering of the index. The values of several configuration parameters affect the
clust field.

The value of this field is affected by:

v The value in the buffers field of the BUFFERPOOL configuration parameter
v The DS_MAX_QUERIES configuration parameter, which specifies the maximum
number of PDQ queries that can run concurrently

Each of these configuration parameters affects the amount of buffer space available
for a single user session. Additional buffers can result in better clustering (a
smaller clust value in the sysindexes or sysindices tables).

You can create more buffers by performing one or both of the following tasks:
v Increasing the size of the buffer pool by updating the value of the
BUFFERPOOL configuration parameter
v Decreasing the value of the DS_MAX_QUERIES configuration parameter
Related reference:
BUFFERPOOL configuration parameter (Administrator's Reference)
DS_MAX_QUERIES configuration parameter (Administrator's Reference)

Nonunique indexes
In some applications, most table updates can be confined to a single time period.
You might be able to set up your system so that all updates are applied overnight
or on specified dates. Additionally, when updates are performed as a batch, you
can drop all nonunique indexes while you make updates and then create new
indexes afterward. This strategy can improve performance.

Dropping nonunique indexes can have the following positive effects:

v The updating program can run faster with fewer indexes to update. Often, the
total time to drop the indexes, update without them, and re-create them is less
than the time to update with the indexes in place. (For a discussion of the time
cost of updating indexes, see “Time costs of indexes” on page 7-8.)
v Newly made indexes are more efficient. Frequent updates tend to dilute the
index structure so that it contains many partly full leaf pages. This dilution
reduces the effectiveness of an index and wastes disk space.

7-12 IBM Informix Performance Guide

 As a time-saving measure, make sure that a batch-updating program calls for rows
in the sequence that the primary-key index defines. That sequence causes the pages
of the primary-key index to be read in order and only one time each.

The presence of indexes also slows down the population of tables when you use
the LOAD statement or the dbload utility. Loading a table that has no indexes is a
quick process (little more than a disk-to-disk sequential copy), but updating
indexes adds a great deal of overhead.

To avoid this overhead, you can:

1. Drop the table (if it exists).
2. Create the table without specifying any unique constraints.
3. Load all rows into the table.
4. Alter the table to apply the unique constraints.
5. Create the nonunique indexes.

If you cannot guarantee that the loaded data satisfies all unique constraints, you
must create unique indexes before you load the rows. It saves time if the rows are
presented in the correct sequence for at least one of the indexes. If you have a
choice, make it the row with the largest key. This strategy minimizes the number
of leaf pages that must be read and written.

Improve query performance with a forest of trees index

A forest of trees index is an alternate indexing method that alleviates the
performance bottlenecks and root node contention that can occur when many
concurrent users access a traditional B-tree index.

A forest of trees index differs from a B-tree index in that it has multiple root nodes
and fewer levels. Multiple root nodes can alleviate root node contention, because
more concurrent users can access the index.

If you know that a particular table has a deep tree, you can improve performance
by creating a forest of trees index with fewer levels in the tree. For example,
suppose you create an index where one of the columns is a 100 byte column
containing character data. If you have a large number of rows in that table, the tree
might contain six or seven levels. If you create a forest of trees index instead of a
B-tree index, you can create more than one tree with four levels, so that every
index traversal goes only four levels deep rather than seven levels deep.
Related concepts:
“Forest of trees indexes” on page 7-2

Detecting root node contention

You can analyze the output of the onstat -g spi command to identify the
performance bottlenecks that a forest of trees index can alleviate.

To detect root node contention and determine whether you need a forest of trees
index:
1. Run the onstat -g spi | sort -nr command to display information about spin
locks with long spins.
The output of the onstat -g spi command shows spin locks with waits, which
occur when threads are reading from or writing to an index concurrently and a
particular thread did not succeed in acquiring the lock on the first try.

Chapter 7. Indexes and index performance considerations 7-13

 2. Analyze the onstat -g spioutput. Look for loop and wait information in these
columns:
Num Waits: The Total number of times a thread waited for the spin lock.
Num Loops: The total number of attempts before a thread successfully
acquired the spin lock.
Avg Loop/Wait: The average number of attempts to acquire the spin lock,
computed as Num Loops / Num Waits.
For example, the following output snippet shows spin locks with large
numbers of waits and loops:
Spin locks with waits:
Num Waits Num Loops Avg Loop/Wait Name
332480 1568908 4.72 fast mutex, 3:bf[1234] 0x2d00008 0x1028a0d8000
39722 498769 12.56 mutex lock, name = log
20761 101831 4.90 fast mutex, 7:bf[62] 0x1300003 0x109da128000
14818 77680 5.24 mutex lock, name = MGM mutex
6523 34350 5.27 fast mutex, 3:bf[362] 0x20008e 0x10289a08000
3. Query sysmaster:systabnames with the hexadecimal representation of the part
number shown in the onstat -g spi output. If the tabname represents an index
name, the index is a forest of trees candidate.
For example, run this query:
echo "select tabname, hex(partnum) from systabnames
where hex(partnum) = ’0x02d00008’" | dbaccess sysmaster -

tabname daily_market_idx
(expression) 0x02d00008

$ echo ’select tabname, hex(partnum) from systabnames’

where hex(partnum) = 0x01300003 | dbaccess sysmaster -

tabname trade_history_idx
(expression) 0x01300003

$ echo ’select tabname, hex(partnum) from systabnames’

where hex(partnum) = 0x0020008E | dbaccess sysmaster -

tabname trade_request_idx2
(expression) 0x0020008E
Related concepts:
“Forest of trees indexes” on page 7-2
Related reference:
onstat -g spi command: Print spin locks with long spins (Administrator's
Reference)

Creating a forest of trees index

You use the CREATE INDEX statement with the HASH ON clause to create a
forest of trees index.

Prerequisite: Determine whether you need a forest of trees index to reduce

performance bottlenecks and contention or to reduce the number of levels in a
traditional B-Tree index.

To create a forest of trees index:

1. Choose the columns for the index and determine the number of subtrees to
create.

7-14 IBM Informix Performance Guide

 2. Create the index by using the CREATE INDEX statement with the HASH ON
clause:
For example, the following command creates a forest of trees index with 100
subtrees (buckets) on the C1 column:
CREATE INDEX fotidx ON tab(c1) hash on (c1) with 100 buckets
After you create a forest of trees index, it is enabled.

You can monitor onstat -g spi command output to verify that root node contention
no longer occurs. If you identify performance bottlenecks that are caused by highly
contended spin locks, you can rebuild the forest of trees index with more buckets.
Related concepts:
“Forest of trees indexes” on page 7-2
HASH ON clause (SQL Syntax)
Related reference:
CREATE INDEX statement (SQL Syntax)

Disabling and enabling a forest of trees index

You can use the INDEXES DISABLED option of the SET Database Object Mode
statement of SQL to disable a forest of trees index, if you want the server to stop
updating the index and to stop using it during queries. After you are ready to put
the index into production, you can use the INDEXES ENABLED option to
re-enable it.

To disable a forest of trees index:

Run the SET INDEXES DISABLED statement of SQL.

For example, for an index named fotidx, specify:

SET INDEXES fotidx DISABLED;

You can re-enable a disabled forest of trees index, for example, by specifying:
SET INDEXES fotidx ENABLED;
Related concepts:
“Forest of trees indexes” on page 7-2

Performing a range scan on a forest of trees index

While you cannot perform range scans directly on the HASH ON columns of a
forest of trees index, you can perform range scans on the columns that are not
listed in the HASH ON column list. To perform range scans on columns that are
listed in HASH ON column list, you must create an additional B-tree index that
contains the appropriate column list for the range scan.

To create indexes for range scans:

1. Create a forest of trees index with at least one column that is not hashed.
For example, specify:
CREATE INDEX idx1 on tab(c1,c2) HASH ON (c1) with 100 buckets;
You can perform a range scan directly on column c2, but not on column c1,
which is listed in HASH ON column list.

Chapter 7. Indexes and index performance considerations 7-15

 2. For range scans on the columns listed in HASH ON column list, create an
additional B-tree index that contains the appropriate column list for the range
scan. This additional B-tree index might have the same column list as the forest
of trees index, plus or minus a column.
For example, specify:
CREATE INDEX idx2 on tab(c1, c2, c3);
Related concepts:
HASH ON clause (SQL Syntax)
Related reference:
CREATE INDEX statement (SQL Syntax)

Determining if you are using a forest of trees index

You can determine whether an index is a forest of trees index by viewing SET
EXPLAIN output. A forest of trees index has FOT in the Index Name field of the
output.

In the following example of partial SET EXPLAIN output, informix.fot_idx is the

name of a forest of trees index.
Estimated Cost: 1
Estimated # of Rows Returned: 1

1) informix.t: INDEX PATH

(1) Index Name: informix.fot_idx (FOT)

Index Keys: c1 c2 (Serial, fragments: ALL)
Lower Index Filter: informix.t.c1 = 1
Related concepts:
“Forest of trees indexes” on page 7-2

Finding the number of hashed columns and subtrees in a

forest of trees index
You can look up the number of hashed columns and subtrees in a forest of trees
index by viewing information in the sysindices table for the database containing
tables that have forest of trees indexes.

To view information about a forest of trees index:

1. Query the sysindices table for the index.
2. Go to the row containing the forest of trees index and view information in the
nhashcols and nbuckets columns.

Creating and dropping an index in an online environment

You can use the CREATE INDEX ONLINE and DROP INDEX ONLINE statements
to create and drop an index in an online environment, when the database and its
associated tables are continuously available.

The CREATE INDEX ONLINE statement enables you to create an index without
having an exclusive lock placed over the table during the duration of the index
build. You can use the CREATE INDEX ONLINE statement even when reads or
updates are occurring on the table. This means index creation can begin
immediately.

7-16 IBM Informix Performance Guide

 When you create an index online, the database server logs the operation with a
flag, so data recovery and restore operations can recreate the index.

When you create an index online, you can use the ONLIDX_MAXMEM
configuration parameter to limit the amount of memory that is allocated to the
preimage log pool and to the updator log pool in shared memory. You might want to
do this if you plan to complete other operations on a table column while executing
the CREATE INDEX ONLINE statement on the column. For more information
about this parameter, see “Limiting memory allocation while creating indexes
online” on page 7-18.

The DROP INDEX ONLINE statement enables you to drop indexes even when
Dirty Read is the transaction isolation level.

The advantages of creating indexes using the CREATE INDEX ONLINE statement
are:
v If a new index is needed to improve the performance of queries on a table, you
can immediately create the index without a lock placed over the table.
v The database server can create an index while a table is being updated.
v The table is available for the duration of the index build.
v The query optimizer can establish better query plans, since the optimizer can
update statistics in unlocked tables.

The advantages of dropping indexes using the DROP INDEX ONLINE statement
are:
v You can drop an inefficient index without disturbing ongoing queries that are
using that index.
v After the index is flagged, the query optimizer will not use the index for new
SELECT operations on tables.

If you initiate a DROP INDEX ONLINE statement for a table that is being updated,
the operation does not occur until after the table update is completed. After you
issue the DROP INDEX ONLINE statement, no one can reference the index, but
concurrent operations can use the index until the operations terminate. The
database server waits to drop the index until all users have finished accessing the
index.

An example of creating an index in an online environment is:

CREATE INDEX idx_1 ON table1(col1) ONLINE

An example of dropping an index in an online environment is:

DROP INDEX idx_1 ONLINE

For more information about the CREATE INDEX ONLINE and DROP INDEX
ONLINE statements, see the IBM Informix Guide to SQL: Syntax.

When you cannot create or drop indexes online

You cannot use the CREATE INDEX ONLINE and the DROP INDEX ONLINE
statements under certain circumstances.

You cannot use the CREATE INDEX ONLINE statement:

v To create an index at the same time that a table is being altered

Chapter 7. Indexes and index performance considerations 7-17

 v To create a clustered index
v To create a Virtual-Index Interface (VII) /R-tree index
v To create a functional index
v To create an index that is partitioned by an interval fragmentation strategy
v To create an index on a table that is partitioned by an interval fragmentation
strategy

You cannot use the DROP INDEX ONLINE statement:

v To drop a Virtual-Index Interface (VII) /R-tree index
v To drop a clustered index

Creating attached indexes in an online environment

You can create attached indexes using the CREATE INDEX ONLINE statement, but
the statement only operates when Dirty Read is the transaction isolation level.

The index creation takes an exclusive lock on the table and waits for all other
concurrent processes scanning the table to quit using the index partitions before
creating the attached index. If the table is being read or updated, the CREATE
INDEX ONLINE statement waits for the exclusive lock for the duration of the lock
mode setting.

Limiting memory allocation while creating indexes online

The ONLIDX_MAXMEM configuration parameter limits the amount of memory
that is allocated to a single preimage pool and a single updator log pool.

The preimage and updator log pools, pimage_<partnum> and ulog_<partnum>,

are shared memory pools that are created when a CREATE INDEX ONLINE
statement is executed. The pools are freed when the execution of the statement is
completed.

The default value of the ONLIDX_MAXMEM configuration parameter is 5120

kilobytes. The minimum value that you can specify is 16 kilobytes; the maximum
value is 4294967295 kilobytes.

You can set the ONLIDX_MAXMEM configuration parameter before starting the
database server, or you can change it dynamically through the onmode -wf and
onmode -wm commands.

Improving performance for index builds

You can improve performance for index builds by adjusting the PDQ priority and
by allocating enough memory and temporary space for the entire index.

Whenever possible, the database server uses parallel processing to improve the
response time of index builds. The number of parallel processes is based on the
number of fragments in the index and the value of the PSORT_NPROCS
environment variable. The database server builds the index with parallel
processing even when the value of PDQ priority is 0.

You can often improve the performance of an index build by taking the following
steps:
1. Set PDQ priority to a value greater than 0 to obtain more memory than the
default 128 kilobytes.

7-18 IBM Informix Performance Guide

 When you set PDQ priority to greater than 0, the index build can take
advantage of the additional memory for parallel processing.
To set PDQ priority, use either the PDQPRIORITY environment variable or the
SET PDQPRIORITY statement in SQL.
2. Do not set the PSORT_NPROCS environment variable. If you have a computer
with multiple CPUs, the database server uses two threads per sort when it sorts
index keys and PSORT_NPROCS is not set. The number of sorts depends on
the number of fragments in the index, the number of keys, the key size, and
the values of the PDQ memory configuration parameters.
3. Allocate enough memory and temporary space to build the entire index.
a. Estimate the amount of virtual shared memory that the database server
might need for sorting.
For more information, see “Estimating memory needed for sorting.”
b. Specify more memory with the DS_TOTAL_MEMORY and
DS_MAX_QUERIES configuration parameters.
c. If not enough memory is available, estimate the amount of temporary space
needed for an entire index build.
For more information, see “Estimating temporary space for index builds” on
page 7-20.
d. Use the onspaces -t utility to create large temporary dbspaces and specify
them in the DBSPACETEMP configuration parameter or the
DBSPACETEMP environment variable.
For information about how to optimize temporary dbspaces, see “Configure
dbspaces for temporary tables and sort files” on page 5-8.

Estimating memory needed for sorting

To calculate the amount of virtual shared memory that the database server might
need for sorting, estimate the maximum number of sorts that might occur
concurrently and multiply that number by the average number of rows and the
average row size.

For example, if you estimate that 30 sorts could occur concurrently, the average
row size is 200 bytes, and the average number of rows in a table is 400, you can
estimate the amount of shared memory that the database server needs for sorting
as follows:
30 sorts * 200 bytes * 400 rows = 2,400,000 bytes

You can use the DS_NONPDQ_QUERY_MEM configuration parameter to

configure the amount sort memory available for non-PDQ queries.

Important: You can only use this parameter if the PDQ priority is set to zero. Its
setting has no effect if the PDQ priority is greater than zero.

The minimum and default value of DS_NONPDQ_QUERY_MEM is 128 kilobytes.

The maximum supported value is 25 percent of DS_TOTAL_MEMORY. For more
information, see “Configuring memory for queries with hash joins, aggregates, and
other memory-intensive elements” on page 13-34.

If the PDQ priority is greater than 0, the maximum amount of shared memory that
the database server allocates for a sort is controlled by the memory grant manager
(MGM). The MGM uses the settings of PDQ priority and the following
configuration parameters to determine how much memory to grant for the sort:
v DS_TOTAL_MEMORY
Chapter 7. Indexes and index performance considerations 7-19
 v DS_MAX_QUERIES
v MAX_PDQPRIORITY

For more information about allocating memory for parallel processing, see “The
allocation of resources for parallel database queries” on page 12-7.

Estimating temporary space for index builds

You can estimate the number of bytes of temporary space needed for an entire
index build.

To estimate the amount of temporary space needed for an index build, perform the
following steps:
1. Add the total widths of the indexed columns or returned values from
user-defined functions. This value is referred to as colsize.
2. Estimate the size of a typical item to sort with one of the following formulas,
depending on whether the index is attached or not:
a. For a nonfragmented table and a fragmented table with an index created
without an explicit fragmentation strategy, use the following formula:
sizeof_sort_item = keysize + 4
b. For fragmented tables with the index explicitly fragmented, use the
following formula:
sizeof_sort_item =
keysize + 8
3. Estimate the number of bytes needed to sort with the following formula:
temp_bytes = 2 * (rows *
sizeof_sort_item)
This formula uses the factor 2 because everything is stored twice when
intermediate sort runs use temporary space. Intermediate sort runs occur when
not enough memory exists to perform the entire sort in memory.
The value for rows is the total number of rows that you expect to be in the
table.

Storing multiple index fragments in a single dbspace

You can store multiple fragments of the same index in a single dbspace, reducing
the total number of dbspaces needed for a fragmented table. You must specify a
name for each fragment that you want to store in the same dbspace. Storing
multiple index fragments in a single dbspace simplifies the management of
dbspaces.

You can also use this feature to improve query performance over storing each
fragment in a different dbspace when a dbspace is located on a faster device.

For more information, see information about managing partitions in the IBM
Informix Administrator's Guide.

Improving performance for index checks

The oncheck utility provides better concurrency for tables that use row locking.
When a table uses page locking, oncheck places a shared lock on the table when it
performs index checks. Shared locks do not allow other users to perform updates,
inserts, or deletes on the table while oncheck checks or prints the index
information.

7-20 IBM Informix Performance Guide

 If the table uses page locking, the database server returns the following message if
you run oncheck without the -x option:
WARNING: index check requires a s-lock on stable whose
lock level is page.

For detailed information about oncheck locking, see the IBM Informix
Administrator's Reference.

The following summary describes locking performed during index checks:

v By default, the database server does not place a shared lock on the table when
you check an index with the oncheck -ci, -cI, -pk, -pK, -pl, or -pL options unless
the table uses page locking. When oncheck checks indexes for a table with page
locking, it places a shared lock on the table, so no other users can perform
updates, inserts, or deletes until the check has completed.
v By not placing a shared lock on tables using row locks during index checks, the
oncheck utility cannot be as accurate in the index check. For absolute assurance
of a complete index check, execute oncheck with the -x option. With the -x
option, oncheck places a shared lock on the table, and no other users can
perform updates, inserts, or deletes until the check completes.

You can query the systables system catalog table to see the current lock level of
the table, as the following sample SQL statement shows:
SELECT locklevel FROM systables
WHERE tabname = "customer"

If you do not see a value of R (for row) in the locklevel column, you can modify
the lock level, as the following sample SQL statement shows:
ALTER TABLE tab1 LOCK MODE (ROW);

Row locking might add other side effects, such as an overall increase in lock usage.
For more information about locking levels, see Chapter 8, “Locking,” on page 8-1.

Indexes on user-defined data types

You can define your own data types and the functions that operate on these data
types. You can define indexes on some kinds of user-defined data types.

DataBlade modules also provide extended data types and functions to the database
server.

You can define indexes on the following kinds of user-defined data types:
v Opaque data types
An opaque data type is a fundamental data type that you can use to define
columns in the same way you use built-in types. An opaque data type stores a
single value and cannot be divided into components by the database server. For
information about creating opaque data types, see the CREATE OPAQUE TYPE
statement in the IBM Informix Guide to SQL: Syntax and IBM Informix
User-Defined Routines and Data Types Developer's Guide. For more information
about the data types and functions that each DataBlade module provides, see the
user guide of each DataBlade module.
v Distinct data types
A distinct data type has the same representation as an existing opaque or built-in
data type but is different from these types. For information about distinct data

Chapter 7. Indexes and index performance considerations 7-21

 types, see the IBM Informix Guide to SQL: Reference and the CREATE DISTINCT
TYPE statement in the IBM Informix Guide to SQL: Syntax.

For more information about data types, see the IBM Informix Guide to SQL:
Reference.

Defining indexes for user-defined data types

As with built-in data types, you might improve the response time for a query
when you define indexes for new data types.

The response time for a query might improve when Informix uses an index for:
v Columns used to join two tables
v Columns that are filters for a query
v Columns in an ORDER BY or GROUP BY clause
v Results of functions that are filters for a query

For more information about when the query performance can improve with an
index on a built-in data type, see “Improve performance by adding or removing
indexes” on page 13-20.

Informix and DataBlade modules provide a variety of different types of indexes

(also referred to as secondary-access methods). A secondary-access method is a set of
database server functions that build, access, and manipulate an index structure.
These functions encapsulate index operations, such as how to scan, insert, delete,
or update nodes in an index.

To create an index on a user-defined data type, you can use any of the following
secondary-access methods:
v Generic B-tree index
A B-tree index is good for a query that retrieves a range of data values. For
more information, see “B-tree secondary-access method.”
v R-tree index
An R-tree index is good for searches on multidimensional data. For more
information, see the IBM Informix R-Tree Index User's Guide.
v Secondary-access methods that a DataBlade module provides for a new data
type
A DataBlade module that supports a certain type of data can also provide a new
index for that new data type. For more information, see “Using an index that a
DataBlade module provides” on page 7-27.

You can create a functional index on the resulting values of a user-defined function
on one or more columns. For more information, see “Using a functional index” on
page 7-25.

After you choose the desired index type, you might also need to extend an
operator class for the secondary-access method. For more information about how
to extend operator classes, see the IBM Informix User-Defined Routines and Data
Types Developer's Guide.

B-tree secondary-access method

Informix provides the generic B-tree index for columns in database tables. In
traditional relational database systems, the B-tree access method handles only
built-in data types and therefore it can only compare two keys of built-in data

7-22 IBM Informix Performance Guide

types. The generic B-tree index is an extended version of a B-tree that Informix
provides to support user-defined data types.

Tip: For more information about the structure of a B-tree index and how to
estimate the size of a B-tree index, see “Estimating index pages” on page 7-4.

Informix uses the generic B-tree as the built-in secondary-access method. This
built-in secondary-access method is registered in the sysams system catalog table
with the name btree. When you use the CREATE INDEX statement (without the
USING clause) to create an index, the database server creates a generic B-tree
index. For more information, see the CREATE INDEX statement in the IBM
Informix Guide to SQL: Syntax.

Tip: Informix also defines another secondary-access method, the R-tree index. For
more information about how to use an R-tree index, see the IBM Informix R-Tree
Index User's Guide.

Uses for a B-tree index:

A B-tree index is good for a query that retrieves a range of data values. If the data
to be indexed has a logical sequence to which the concepts of less than, greater than,
and equal apply, the generic B-tree index is a useful way to index your data.

Initially, the generic B-tree index supports the relational operators (<,<=,=,>=,>) on
all built-in data types and orders the data in lexicographical sequence.

The optimizer considers whether to use the B-tree index to execute a query if you
define a generic B-tree index on:
v Columns used to join two tables
v Columns that are filters for a query
v Columns in an ORDER BY or GROUP BY clause
v Results of functions that are filters for a query

Extending a generic B-tree index:

Initially, the generic B-tree can index data that is one of the built-in data types, and
it orders the data in lexicographical sequence. However, you can extend a generic
B-tree for some other data types.

You can extend a generic B-tree to support columns and functions on the following
data types:
v User-defined data types (opaque and distinct data types) that you want the B-tree
index to support
In this case, you must extend the default operator class of the generic B-tree
index.
v Built-in data types that you want to order in a different sequence from the
lexicographical sequence that the generic B-tree index uses
In this case, you must define a different operator class from the default generic
B-tree index.

An operator class is the set of functions (operators) that are associated with a
nontraditional B-tree index. For more details on operator classes, see “Choosing
operator classes for indexes” on page 7-27.

Chapter 7. Indexes and index performance considerations 7-23

 Identifying the available access methods
To supplement the built-in B-tree secondary-access method that Informix provides,
your enterprise might have installed DataBlade modules that implement additional
secondary-access methods. If additional access methods exist, they are defined in
the sysams system catalog table. You can query the sysams system catalog to
determine if additional access methods are available.

To identify the secondary-access methods that are available for your database,
query the sysams system catalog table with the following SELECT statement:
SELECT am_id, am_owner, am_name, am_type FROM sysams
WHERE am_type = ’S’;

An 'S' value in the am_type column identifies the access method as a

secondary-access method. This query returns the following information:
v The am_id and am_name columns identify the secondary-access method.
v The am_owner column identifies the owner of the access method.

In an ANSI-compliant database, the access-method name must be unique within

the name space of the user. The access-method name always begins with the owner
in the format am_owner.am_name.

By default, Informix provides the following definitions in the sysams system

catalog table for two secondary-access methods, btree and rtree.

Access Method am_id Column am_name Column am_owner Column

Generic B-tree 1 btree 'informix'
R-tree 2 rtree 'informix'

Important: The sysams system catalog table does not contain a row for the built-in
primary access method. This primary access method is internal to Informix and
does not require a definition in sysams. However, the built-in primary access
method is always available for use.

If you find additional rows in the sysams system catalog table (rows with am_id
values greater than 2), the database supports additional user-defined access
methods. Check the value in the am_type column to determine whether a
user-defined access method is a primary- or secondary-access method.

For more information about the columns of the sysams system catalog table, see
the IBM Informix Guide to SQL: Reference. For information about how to determine
the operator classes that are available in your database, see “Identifying the
available operator classes” on page 7-30.

User-defined secondary-access methods

If the concepts of less than, greater than, and equal do not apply to the data to be
indexed, you might consider using a user-defined secondary-access method instead of
the built-in secondary-access method, which is a B-tree index. You can use a
user-defined secondary-access method to access other indexing structures, such as
an R-tree index.

If your database supports a user-defined secondary-access method, you can specify

that the database server uses this access method when it accesses a particular
index. For information about how to determine the secondary-access methods that
your database defines, see “Identifying the available access methods.”

7-24 IBM Informix Performance Guide

To choose a user-defined secondary-access method, use the USING clause of the
CREATE INDEX statement. The USING clause specifies the name of the
secondary-access method to use for the index you create. This name must be listed
in the am_name column of the sysams system catalog table and must be a
secondary-access method (the am_type column of sysams is ’S’).

The secondary-access method that you specify in the USING clause of CREATE
INDEX must already be defined in the sysams system catalog. If the
secondary-access method has not yet been defined, the CREATE INDEX statement
fails.

When you omit the USING clause from the CREATE INDEX statement, the
database server uses B-tree indexes as the secondary-access method. For more
information, see the CREATE INDEX statement in the IBM Informix Guide to SQL:
Syntax.

R-tree indexes:

Informix supports the R-tree index for columns that contain spatial data such as
maps and diagrams. An R-tree index uses a tree structure whose nodes store
pointers to lower-level nodes.

At the leaves of the R-tree are a collection of data pages that store n-dimensional
shapes. For more information about the structure of an R-tree index and how to
estimate the size of an R-tree index, see the IBM Informix R-Tree Index User's Guide.

Using a functional index

You can create a column index on the actual values in one or more columns. You
can also create a functional index on the values of one or more columns that a
user-defined function returns from arguments.

Important: The database server imposes the following restrictions on the

user-defined routines (UDRs) on which a functional index is defined:
v The arguments cannot be column values of a collection data type.
v The function cannot return a large object (including built-in types BLOB, BYTE,
CLOB, and TEXT).
v The function cannot be a VARIANT function.
v The function cannot include any DML statement of SQL.
v The function must be a UDR, rather than a built-in function. However, you can
create an SPL wrapper that calls and returns the value from a built-in function
of SQL.
In addition, do not create functional indexes using any routine that calls the
built-in DECRYPT_BINARY( ) or DECRYPT_CHAR( ) functions, which can display
encrypted data values in plain text. (Do not attempt to use data values in any
encrypted column as an index key.)

To decide whether to use a column index or functional index, determine whether a

column index is the right choice for the data that you want to index. An index on
a column of some data types might not be useful for typical queries. For example,
the following query asks how many images are dark:
SELECT COUNT(*) FROM photos WHERE
darkness(picture) > 0.5

Chapter 7. Indexes and index performance considerations 7-25

 An index on the picture data itself does not improve the query performance. The
concepts of less than, greater than, and equal are not particularly meaningful when
applied to an image data type. Instead, a functional index that uses the darkness()
function can improve performance. You might also have a user-defined function
that runs frequently enough that performance improves when you create an index
on its values.
Related concepts:
“Managing indexes” on page 7-8

What is a functional index?:

A functional index can be a B-tree index, an R-tree index, or a user-defined index

type that a DataBlade module provides.

When you create a functional index, the database server computes the values of
the user-defined function and stores them as key values in the index. When a
change in the table data causes a change in one of the values of an index key, the
database server automatically updates the functional index.

You can use a functional index for functions that return values of both
user-defined data types (opaque and distinct) and built-in data types. However,
you cannot define a functional index if the function returns a simple-large-object
data type (TEXT or BYTE).

For more information about the types of indexes, see “Defining indexes for
user-defined data types” on page 7-22. For information about space requirements
for functional indexes, see “Estimating index pages” on page 7-4.
Related concepts:
“Types of indexes” on page 7-1

When is a functional index used?:

The optimizer considers whether to use a functional index to access the results of
functions that are in a SELECT clause or are in the filters in the WHERE clause.

Creating a functional index:

You can build a functional index on a user-defined function. The user-defined

function can be either an external function or an SPL function.

To build a functional index on a user-defined function:

1. Write the code for the user-defined function if it is an external function.
2. Register the user-defined function in the database with the CREATE
FUNCTION statement.
3. Build the functional index with the CREATE INDEX statement.

For example, to create a functional index on the darkness() function:

1. Write the code for the user-defined darkness() function that operates on the
data type and returns a decimal value.
2. Register the user-defined function in the database with the CREATE
FUNCTION statement:

7-26 IBM Informix Performance Guide

 CREATE FUNCTION darkness(im image)
RETURNS decimal
EXTERNAL NAME ’/lib/image.so’
LANGUAGE C NOT VARIANT
In this example, you can use the default operator class for the functional index
because the return value of the darkness() function is a built-in data type,
DECIMAL.
3. Build the functional index with the CREATE INDEX statement.
CREATE TABLE photos
(
name char(20),
picture image
...
);
CREATE INDEX dark_ix ON photos (darkness(picture));
In this example, assume that the user-defined data type of image has already
been defined in the database.

The optimizer can now consider the functional index when you specify the
darkness() function as a filter in the query:
SELECT count(*) FROM photos WHERE
darkness(picture) > 0.5

You can also create a composite index with user-defined functions. For more
information, see “Use composite indexes” on page 13-20.

Warning: Do not create a functional index using either the DECRYPT_BINARY()

or the DECRYPT_CHAR() function. These functions store plain text data in the
database, defeating the purpose of encryption. For more information about
encryption, see the IBM Informix Administrator's Guide.

Using an index that a DataBlade module provides

DataBlade modules can provide new data types that users can access. A DataBlade
module can also provide a new index for the new data type.

For example, the Excalibur Text Search DataBlade module provides an index to
search text data. For more information, see the Excalibur Text Search DataBlade
Module User's Guide.

For more information about the types of data and functions that each DataBlade
module provides, see the user guide for the DataBlade module. For information
about how to determine the types of indexes available in your database, see
“Identifying the available access methods” on page 7-24.

Choosing operator classes for indexes

For most situations, use the default operators that are defined for a
secondary-access method. However, when you want to order the data in a different
sequence or provide index support for a user-defined data type, you must extend
an operator class.

For more information about how to extend an operator class, see IBM Informix
User-Defined Routines and Data Types Developer's Guide.

Chapter 7. Indexes and index performance considerations 7-27

 Operator classes
An operator class is a set of function names that is associated with a
secondary-access method. These functions allow the secondary-access method to
store and search for values of a particular data type.

The query optimizer for the database server uses an operator class to determine if
an index can process the query with the least cost. An operator class indicates two
things to the query optimizer:
v Which functions that appear in an SQL statement can be evaluated with a given
index
These functions are called the strategy functions for the operator class.
v Which functions the index uses to evaluate the strategy functions
These functions are called the support functions for the operator class.

With the information that the operator class provides, the query optimizer can
determine whether a given index is applicable to the query. The query optimizer
can consider whether to use the index for the given query when the following
conditions are true:
v An index exists on the particular column or columns in the query.
v For the index that exists, the operation on the column or columns in the query
matches one of the strategy functions in the operator class associated with the
index.

The query optimizer reviews the available indexes for the table or tables and
matches the index keys with the column specified in the query filter. If the column
in the filter matches an index key, and the function in the filter is one of the
strategy functions of the operator class, the optimizer includes the index when it
determines which query plan has the lowest execution cost. In this manner, the
optimizer can determine which index can process the query with the least cost.

Informix stores information about operator classes in the sysopclasses system

catalog table.

Strategy and support functions of a secondary access method:

Informix uses the strategy functions of a secondary-access method to help the query
optimizer determine whether a specific index is applicable to a specific operation
on a data type.

If an index exists and the operator in the filter matches one of the strategy
functions in the operator class, the optimizer considers whether to use the index
for the query.

Informix uses the support functions of a secondary-access method to build and

access the index. These functions are not called directly by end users. When an
operator in the query filter matches one of the strategy functions, the
secondary-access method uses the support functions to traverse the index and
obtain the results. Identification of the actual support functions is left to the
secondary-access method.

7-28 IBM Informix Performance Guide

Default operator classes:

Each secondary-access method has a default operator class associated with it. By
default, the CREATE INDEX statement associates the default operator class with an
index.

For example, the following CREATE INDEX statement creates a B-tree index on the
postalcode column and automatically associates the default B-tree operator class
with this column:
CREATE INDEX postal_ix ON customer(postalcode)

For more information about how to specify a new default operator class for an
index, see “User-defined operator classes” on page 7-31.

Built-in B-tree operator class

The built-in secondary-access method (the generic B-tree) has a default operator
class called btree_ops, which is defined in the sysopclasses system catalog table.

By default, the CREATE INDEX statement associates the btree_ops operator class
with it when you create a B-tree index. For example, the following CREATE
INDEX statement creates a generic B-tree index on the order_date column of the
orders table and associates with this index the default operator class for the B-tree
secondary-access method:
CREATE INDEX orddate_ix ON orders (order_date)

Informix uses the btree_ops operator class to specify:

v The strategy functions to tell the query optimizer which filters in a query can
use a B-tree index
v The support function to build and search the B-tree index

B-tree strategy functions:

The btree_ops operator class defines the names of strategy functions for the btree
access method.

The strategy functions that the btree_ops operator class defines are:
v lessthan (<)
v lessthanorequal (<=)
v equal (=)
v greaterthanorequal (>=)
v greaterthan (>)

These strategy functions are all operator functions. That is, each function is
associated with an operator symbol; in this case, with a relational-operator symbol.
For more information about relational-operator functions, see the IBM Informix
User-Defined Routines and Data Types Developer's Guide.

When the query optimizer examines a query that contains a column, it checks to
see if this column has a B-tree index defined on it. If such an index exists and if the
query contains one of the relational operators that the btree_ops operator class
supports, the optimizer can choose a B-tree index to execute the query.

Chapter 7. Indexes and index performance considerations 7-29

 B-tree support function:

The btree_ops operator class has one support function, a comparison function
called compare(). The btree_ops operator class has one support function, a
comparison function called compare().

The compare() function is a user-defined function that returns an integer value to

indicate whether its first argument is equal to, less than, or greater than its second
argument, as follows:
v A value of 0 when the first argument is equal to the second argument
v A value less than 0 when the first argument is less than the second argument
v A value greater than 0 when the first argument is greater than the second
argument

The B-tree secondary-access method uses the compare() function to traverse the
nodes of the generic B-tree index. To search for data values in a generic B-tree
index, the secondary-access method uses the compare() function to compare the
key value in the query to the key value in an index node. The result of the
comparison determines if the secondary-access method needs to search the
next-lower level of the index or if the key resides in the current node.

The generic B-tree access method also uses the compare() function to perform the
following tasks for generic B-tree indexes:
v Sort the keys before the index is built
v Determine the linear order of keys in a generic B-tree index
v Evaluate the relational operators
v Search for data values in an index

The database server uses the compare() function to evaluate comparisons in the
SELECT statement. To provide support for these comparisons for opaque data
types, you must write the compare() function. For more information, see the IBM
Informix User-Defined Routines and Data Types Developer's Guide.

The database server also uses the compare() function when it uses a B-tree index to
process an ORDER BY clause in a SELECT statement. However, the optimizer does
not use the index to perform an ORDER BY operation if the index does not use the
btree-ops operator class.

Identifying the available operator classes

You can identify the operator classes that are available for your database by
querying the sysopclasses system catalog table.

The database server provides the default operator class for the built-in
secondary-access method, the generic B-tree index. In addition, your environment
might have installed DataBlade modules that implement other operator classes. All
operator classes are defined in the sysopclasses system catalog table.

To identify the operator classes that are available for your database, query the
sysopclasses system catalog table with the following SELECT statement:
SELECT opclassid, opclassname, amid, am_name
FROM sysopclasses, sysams
WHERE sysopclasses.amid = sysams.am_id

This query returns the following information:

7-30 IBM Informix Performance Guide

v The opclassid and opclassname columns identify the operator class.
v The am_id and am_name columns identify the associated secondary-access
methods.

By default, the database server provides the following definitions in the

sysopclasses system catalog table for two operator classes, btree_ops and
rtree_ops.

opclassid opclassname am_name

Access Method Column Column amid Column Column
Generic B-tree 1 btree_ops 1 btree
R-tree 2 rtree_ops 2 rtree

If you find additional rows in the sysopclasses system catalog table (rows with
opclassid values greater than 2), your database supports user-defined operator
classes. Check the value in the amid column to determine the secondary-access
methods to which the operator class belongs.

The am_defopclass column in the sysams system catalog table stores the
operator-class identifier for the default operator class of a secondary-access
method. To determine the default operator class for a given secondary-access
method, you can run the following query:
SELECT am_id, am_name, am_defopclass, opclass_name
FROM sysams, sysopclasses
WHERE sysams.am_defopclass = sysopclasses.opclassid

By default, the database server provides the following default operator classes.

am_id am_name am_defopclass

Access Method Column Column Column opclass_name Column
Generic B-tree 1 btree 1 btree_ops
R-tree 2 rtree 2 rtree_ops

For more information about the columns of the sysopclasses and sysams system
catalog tables, see the IBM Informix Guide to SQL: Reference. For information about
how to determine the access methods that are available in your database, see
“Identifying the available access methods” on page 7-24.

User-defined operator classes

The CREATE INDEX statement specifies the operator class to use for each
component of an index. If you do not specify an operator class, the CREATE
INDEX statement uses the default operator class for the secondary-access method
that you create. You can use a user-defined operator class for components of an
index.

To specify a user-defined operator class for a particular component of an index,

you can:
v Use a user-defined operator class that your database already defines.
v Use an R-tree operator class, if your database defined the R-tree
secondary-access method. For more information about R-trees, see the IBM
Informix R-Tree Index User's Guide.

Chapter 7. Indexes and index performance considerations 7-31

 If your database supports multiple-operator classes for the secondary-access
method that you want to use, you can specify which operator classes the database
server is to use for a particular index. For information on how to determine the
operator classes that your database defines, see “Identifying the available operator
classes” on page 7-30.

Each part of a composite index can specify a different operator class. You choose
the operator classes when you create the index. In the CREATE INDEX statement,
you specify the name of the operator class to use after each column or function
name in the index-key specification. Each name must be listed in the opclassname
column of the sysopclasses system catalog table and must be associated with the
secondary-access method that the index uses.

For example, if your database defines the abs_btree_ops secondary-access method

to define a new sort order, the following CREATE INDEX statement specifies that
the table1 table associates the abs_btree_ops operator class with the col1_ix B-tree
index:
CREATE INDEX col1_ix ON table1(col1 abs_btree_ops)

The operator class that you specify in the CREATE INDEX statement must already
be defined in the sysopclasses system catalog with the CREATE OPCLASS
statement. If the operator class has not yet been defined, the CREATE INDEX
statement fails. For information about how to create an operator class, see IBM
Informix User-Defined Routines and Data Types Developer's Guide.

7-32 IBM Informix Performance Guide

Chapter 8. Locking
The database server uses locks, which can affect concurrency and performance. You
can monitor and administer locks.

Locks
A lock is a software mechanism that you can set to prevent others from using a
resource. You can place a lock on a single row or key, a page of data or index keys,
a whole table, or an entire database.

Additional types of locks are available for smart large objects. For more
information, see “Locks for smart large objects” on page 8-16.

The maximum number of rows or pages locked in a single transaction is controlled

by the total number of locks configured. The number of tables in which those rows
or pages are locked is not explicitly controlled.

Locking granularity
The level and type of information that the lock protects is called locking granularity.
Locking granularity affects performance.

When a user cannot access a row or key, the user can wait for another user to
unlock the row or key. If a user locks an entire page, a higher probability exists
that more users will wait for a row in the page.

The ability of more than one user to access a set of rows is called concurrency. The
goal of the database administrator is to increase concurrency to increase total
performance without sacrificing performance for an individual user.

Row and key locks

Row and key locks generally provide the best overall performance when you are
updating a relatively small number of rows, because they increase concurrency.
However, the database server incurs some overhead in obtaining a lock. For an
operation that changes a large number of rows, obtaining one lock per row might
not be cost effective.

For an operation that changes a large number of rows, consider “Page locks” on
page 8-2.

The default locking mode is page-locking. If you want row or key locks, you must
create the table with row locking on or alter the table.

The following example shows how to create a table with row locking on:
CREATE TABLE customer(customer_num serial, lname char(20)...)
LOCK MODE ROW;

The ALTER TABLE statement can also change the lock mode.

When the lock mode is ROW and you insert or update a row, the database server
creates a row lock. In some cases, you place a row lock by simply reading the row
with a SELECT statement.

© Copyright IBM Corp. 1996, 2014 8-1

 When the lock mode is ROW and you insert, update, or delete a key (performed
automatically when you insert, update, or delete a row), the database server
creates a lock on the key in the index.

Key-value locks
When a user deletes a row within a transaction, the row cannot be locked because
it does not exist. However, the database server must somehow record that a row
existed until the end of the transaction. The database server uses key-value locking
to lock the deleted row.

When the database server deletes a row, key values in the indexes for the table are
not removed immediately. Instead, each key value is marked as deleted, and a lock
is placed on the key value.

Other users might encounter key values that are marked as deleted. The database
server must determine whether a lock exists. If a lock exists, the delete has not
been committed, and the database server sends a lock error back to the application
(or it waits for the lock to be released if the user executed SET LOCK MODE TO
WAIT).

One of the most important uses for key-value locking is to assure that a unique
key remains unique through the end of the transaction that deleted it. Without this
protection mechanism, user A might delete a unique key within a transaction, and
user B might insert a row with the same key before the transaction commits. This
scenario makes rollback by user A impossible. Key-value locking prevents user B
from inserting the row until the end of user A's transaction.

Page locks
Page locking is the default mode when you create a table without the LOCK
MODE clause. With page locking, instead of locking only the row, the database
server locks the entire page that contains the row. If you update several rows on
the same page, the database server uses only one lock for the page.

When you insert or update a row, the database server creates a page lock on the
data page. In some cases, the database server creates a page lock when you simply
read the row with a SELECT statement.

When you insert, update, or delete a key (performed automatically when you
insert, update, or delete a row), the database server creates a lock on the page that
contains the key in the index.

Important: A page lock on an index page can decrease concurrency more

substantially than a page lock on a data page. Index pages are dense and hold a
large number of keys. By locking an index page, you make a potentially large
number of keys unavailable to other users until you release the lock. Tables that
use page locks cannot support the USELASTCOMMITTED concurrency feature,
which is described in the “Committed Read isolation” on page 8-6 section.

Page locks are useful for tables in which the normal user changes a large number
of rows at one time. For example, an orders table that holds orders that are
commonly inserted and queried individually is not a good candidate for page
locking. But a table that holds old orders and is updated nightly with all of the
orders placed during the day might be a good candidate. In this case, the type of
isolation level that you use to access the table is important. For more information,
see “Isolation level” on page 8-5.

8-2 IBM Informix Performance Guide

Table locks
In a data warehouse environment, it might be more appropriate for queries to
acquire locks of larger granularity. For example, if a query accesses most of the
rows in a table, its efficiency increases if it acquires a smaller number of table locks
instead of many page or row locks.

The database server can place two types of table locks:

v Shared lock
No other users can write to the table.
v Exclusive lock
No other users can read from or write to the table.

Another important distinction between these two types of table locks is the actual
number of locks placed:
v In shared mode, the database server places one shared lock on the table, which
informs other users that no updates can be performed. In addition, the database
server adds locks for every row updated, deleted, or inserted.
v In exclusive mode, the database server places only one exclusive lock on the
table, no matter how many rows it updates. If you update most of the rows in
the table, place an exclusive lock on the table.

Important: A table lock on a table can decrease update concurrency radically. Only
one update transaction can access that table at any given time, and that update
transaction locks out all other transactions. However, multiple read-only
transactions can simultaneously access the table. This behavior is useful in a data
warehouse environment where the data is loaded and then queried by multiple
users.

You can switch a table back and forth between table-level locking and the other
levels of locking. This ability to switch locking levels is useful when you use a
table in a data warehouse mode during certain time periods but not in others.

A transaction tells the database server to use table-level locking for a table with the
LOCK TABLE statement. The following example places an exclusive lock on the
table:
LOCK TABLE tab1 IN EXCLUSIVE MODE;

The following example places a shared lock on the table:

LOCK TABLE tab1 IN SHARE MODE:

In some cases, the database server places its own table locks. For example, if the
isolation level is Repeatable Read, and the database server must read a large
portion of the table, it places a table lock automatically instead of setting row or
page locks. The database server places a table lock on a table when it creates or
drops an index.

Database locks
You can place a lock on the entire database when you open the database with the
DATABASE statement. A database lock prevents read or update access by anyone
but the current user.

The following statement opens and locks the sales database:

DATABASE sales EXCLUSIVE

Chapter 8. Locking 8-3

Configuring the lock mode
When you create a table, the default lock mode is page. You can change the lock
mode (and thus increase or decrease concurrency) when you create or alter tables
or by setting the IFX_DEF_TABLE_LOCKMODE environment variable or the
DEF_TABLE_LOCKMODE configuration parameter.

If you know that most of your applications might benefit from a lock mode of row,
you can take one of the following actions:
v Use the LOCK MODE ROW clause in each CREATE TABLE statement or ALTER
TABLE statement.
v Set the IFX_DEF_TABLE_LOCKMODE environment variable to ROW so that all
tables you subsequently create within a session use ROW without the need to
specify the lock mode in the CREATE TABLE statement or ALTER TABLE
statement.
v Set the DEF_TABLE_LOCKMODE configuration parameter to ROW so that all
tables subsequently created within the database server use ROW without the
need to specify the lock mode in the CREATE TABLE statement or ALTER
TABLE statement.

If you change the lock mode with the IFX_DEF_TABLE_LOCKMODE

environment variable or DEF_TABLE_LOCKMODE configuration parameter, the
lock mode of existing tables are not affected. Existing tables continue to use the
lock mode with which they were defined at the time they were created.

In addition, if you previously changed the lock mode of a table to ROW, and
subsequently execute an ALTER TABLE statement to alter some other characteristic
of the table (such as add a column or change the extent size), you do not need to
specify the lock mode. The lock mode remains at ROW and is not set to the default
PAGE mode.

You can still override the lock mode of individual tables by specifying the LOCK
MODE clause in the CREATE TABLE statement or ALTER TABLE statement.

The following list shows the order of precedence for the lock mode on a table:
v The system default is page locks. The database server uses this system default if
you do not set the configuration parameter, do not set the environment variable,
or do not specify the LOCK MODE clause in the SQL statements.
v If you set the DEF_TABLE_LOCKMODE configuration parameter, the database
server uses this value when you do not set the environment variable, or do not
specify the LOCK MODE clause in the SQL statements.
v If you set the IFX_DEF_TABLE_LOCKMODE environment variable, this value
overrides the DEF_TABLE_LOCKMODE configuration parameter and system
default. The database server uses this value when you do not specify the LOCK
MODE clause in the SQL statements.
v If you specify the LOCK MODE clause in the CREATE TABLE statement or
ALTER TABLE statement, this value overrides the
IFX_DEF_TABLE_LOCKMODE, the DEF_TABLE_LOCKMODE configuration
parameter and system default.

8-4 IBM Informix Performance Guide

Setting the lock mode to wait
When an application process encounters a lock, the default behavior of the
database server is to return an error. Instead, you can run an SQL statement to set
the lock mode to wait. This specifies that an application process does not proceed
until the lock is removed.

To suspend the current process until the lock releases, run the following SQL
statement :
SET LOCK MODE TO WAIT;

You can also specify the maximum number of seconds that a process waits for a
lock to be released before issuing an error. In the following example, the database
server waits for 20 seconds before issuing an error:
SET LOCK MODE TO WAIT 20;

To return to the default behavior (no waiting for locks), execute the following
statement:
SET LOCK MODE TO NOT WAIT;

Locks with the SELECT statement

The type and duration of locks that the database server places depend on the
isolation level set in the application, the database mode (logging, nonlogging, or
ANSI,) and on whether the SELECT statement is within an update cursor. These
locks can affect overall performance because they affect concurrency.

Isolation level
The number and duration of locks placed on data during a SELECT statement
depend on the level of isolation that the user sets. The type of isolation can affect
overall performance because it affects concurrency.

Before you execute a SELECT statement, you can set the isolation level with the
SET ISOLATION statement, which is an Informix extension to the ANSI SQL-92
standard, or with the ANSI/ISO-compliant SET TRANSACTION. The main
differences between the two statements are that SET ISOLATION has an additional
isolation level, Cursor Stability, and SET TRANSACTION cannot be executed more
than once in a transaction as SET ISOLATION can. The SET ISOLATION statement
is an Informix extension to the ANSI SQL-92 standard. The SET ISOLATION
statement can change the enduring isolation level for the session

Dirty Read isolation

The Dirty Read isolation (or ANSI Read Uncommitted) level does not place any
locks on any rows fetched during a SELECT statement. Dirty Read isolation is
appropriate for static tables that are used for queries.

Use Dirty Read isolation with care if update activity occurs at the same time. With
Dirty Read, the reader can read a row that has not been committed to the database
and might be eliminated or changed during a rollback. For example, consider the
following scenario:
User 1 starts a transaction.
User 1 inserts row A.
User 2 reads row A.
User 1 rolls back row A.

Chapter 8. Locking 8-5

 User 2 reads row A, which user 1 rolls back seconds later. In effect, user 2 read a
row that was never committed to the database. Uncommitted data that is rolled
back can be a problem in applications.

Because the database server does not check or place any locks for queries, Dirty
Read isolation offers the best performance of all isolation levels. However, because
of potential problems with uncommitted data that is rolled back, use Dirty Read
isolation with care.

Because problems with uncommitted data that is rolled back are an issue only with
transactions, databases that do not have transaction (and hence do not allow
transactions) use Dirty Read as a default isolation level. In fact, Dirty Read is the
only isolation level allowed for databases that do not have transaction logging.

Committed Read isolation

A reader with the Committed Read isolation (or ANSI Read Committed) isolation
level checks for locks before returning a row. By checking for locks, the reader
cannot return any uncommitted rows.

The database server does not actually place any locks for rows read during
Committed Read. It simply checks for any existing rows in the internal lock table.

Committed Read is the default isolation level for databases with logging if the log
mode is not ANSI-compliant. For databases created with a logging mode that is
not ANSI-compliant, Committed Read is an appropriate isolation level for most
activities. For ANSI-compliant databases, Repeatable Read is the default isolation
level.

Ways to reduce the risk of Committed Read isolation level conflicts:

In the Committed Read isolation level, locks held by other sessions can cause SQL
operations to fail if the current session cannot acquire a lock or if the database
server detects a deadlock. (A deadlock occurs when two users hold locks, and each
user wants to acquire a lock that the other user owns.) The LAST COMMITTED
keyword option to the SET ISOLATION COMMITTED READ statement of SQL
reduces the risk of locking conflicts.

The LAST COMMITTED keyword option to the SET ISOLATION COMMITTED

READ statement of SQL instructs the server to return the most recently committed
version of the rows, even if another concurrent session holds an exclusive lock.
You can use the LAST COMMITTED keyword option for B-tree and functional
indexes, tables that support transaction logging, and tables that do not have
page-level locking or exclusive locks. For more information, see information about
the SET ISOLATION statement in the IBM Informix Guide to SQL: Syntax.

For databases created with transaction logging, you can set the
USELASTCOMMITTED configuration parameter to specify whether the database
server uses the last committed version of the data, rather than wait for the lock to
be released, when sessions using the Dirty Read or Committed Read isolation level
(or the ANSI/ISO level of Read Uncommitted or Read Committed) attempt to read
a row on which a concurrent session holds a shared lock. The last committed
version of the data is the version of the data that existed before any updates
occurred.

If no value or a value of NONE is set for the USELASTCOMMITTED configuration

parameter or for the USELASTCOMMITTED session environment variable,

8-6 IBM Informix Performance Guide

sessions in a COMMITTED READ or READ COMMITTED isolation level wait for
any exclusive locks to be released, unless the SET ISOLATION COMMITTED
READ LAST COMMITTED statement of SQL instructs the database server to read
the most recently committed version of the data.

Setting the USELASTCOMMITTED configuration parameter to operate with the

Committed Read isolation level can affect performance only if concurrent
conflicting updates occur. When concurrent conflicting updates occur, the
performance of queries depends on the dynamics of the transactions. For example,
a reader using the last committed version of the data, might need to undo the
updates made to a row by another concurrent transaction. This situation involves
reading one or more log records, thereby increasing the I/O traffic, which can
affect performance.
Related reference:
USELASTCOMMITTED configuration parameter (Administrator's Reference)

Cursor Stability isolation

A reader with Cursor Stability isolation acquires a shared lock on the row that is
currently fetched. This action assures that no other user can update the row until
the user fetches a new row.

In the example for a cursor in Figure 8-1, at fetch a row the database server releases
the lock on the previous row and places a lock on the row being fetched. At close
the cursor, the server releases the lock on the last row.

set isolation to cursor stability

declare cursor for SELECT * FROM customer
open the cursor
while there are more rows
fetch a row
do work
end while
close the cursor

Figure 8-1. Locks placed for cursor stability

If you do not use a cursor to fetch data, Cursor Stability isolation behaves in the
same way as Committed Read. No locks are actually placed.

Repeatable Read isolation

Repeatable Read isolation (ANSI Serializable and ANSI Repeatable Read) is the
strictest isolation level. With Repeatable Read, the database server locks all rows
examined (not just fetched) for the duration of the transaction.

The example in Figure 8-2 on page 8-8 shows when the database server places and
releases locks for a repeatable read. At fetch a row, the server places a lock on the
row being fetched and on every row it examines in order to retrieve this row. At
close the cursor, the server releases the lock on the last row.

Chapter 8. Locking 8-7

 set isolation to repeatable read
begin work
declare cursor for SELECT * FROM customer
open the cursor
while there are more rows
fetch a row
do work
end while
close the cursor
commit work

Figure 8-2. Locks placed for repeatable read

Repeatable Read is useful during any processing in which multiple rows are
examined, but none must change during the transaction. For example, suppose an
application must check the account balance of three accounts that belong to one
person. The application gets the balance of the first account and then the second.
But, at the same time, another application begins a transaction that debits the third
account and credits the first account. By the time that the original application
obtains the account balance of the third account, it has been debited. However, the
original application did not record the debit of the first account.

When you use Committed Read or Cursor Stability, the previous scenario can
occur. However, it cannot occur with Repeatable Read. The original application
holds a read lock on each account that it examines until the end of the transaction,
so the attempt by the second application to change the first account fails (or waits,
depending upon SET LOCK MODE).

Because even examined rows are locked, if the database server reads the table
sequentially, a large number of rows unrelated to the query result can be locked.
For this reason, use Repeatable Read isolation for tables when the database server
can use an index to access a table. If an index exists and the optimizer chooses a
sequential scan instead, you can use directives to force use of the index. However,
forcing a change in the query path might negatively affect query performance.

Locking nonlogging tables

The database server does not place page or row locks on a nonlogging table when
you use the table within a transaction. However, you can lock nonlogging tables to
prevent concurrency problems when other users are modifying a nonlogging table

Use one of the following methods to prevent concurrency problems when other
users are modifying a nonlogging table:
v Lock the table in exclusive mode for the whole transaction.
v Use Repeatable Read isolation level for the whole transaction.

Important: Nonlogging raw tables are intended for fast loading of data. You
should change the table to STANDARD before you use it in a transaction or
modify the data within it.

Update cursors
An update cursor is a special kind of cursor that applications can use when the
row might potentially be updated. Update cursors use promotable locks in which the
database server places an update lock on the row when the application fetches the
row. The lock is changed to an exclusive lock when the application uses an update
cursor and UPDATE...WHERE CURRENT OF to update the row.

8-8 IBM Informix Performance Guide

When the update lock is on the row as the application fetches it, other users can
still view the row.

In some cases, the database server might place locks on rows that the database
server has examined but not actually fetched. Whether this behavior occurs
depends on how the database server executes the SQL statement.

The advantage of an update cursor is that you can view the row with the
confidence that other users cannot change it or view it with an update cursor
while you are viewing it and before you update it.

If you do not update the row, the default behavior of the database server is to
release the update lock when you execute the next FETCH statement or close the
cursor. However, if you execute the SET ISOLATION statement with the RETAIN
UPDATE LOCKS clause, the database server does not release any currently
existing or subsequently placed update locks until the end of the transaction.

The code in Figure 8-3 shows when the database server places and releases update
locks with a cursor. At fetch row 1, the database server places an update lock on
row 1. At fetch row 2, the server releases the update lock on row 1 and places an
update lock on row 2. However, after the database server executes the SET
ISOLATION statement with the RETAIN UPDATE LOCKS clause, it does not
release any update locks until the end of the transaction. At fetch row 3, it places an
update lock on row 3. At fetch row 4, it places an update lock on row 4. At commit
work, the server releases the update locks for rows 2, 3, and 4.

declare update cursor

begin work
open the cursor
fetch row 1
fetch row 2
SET ISOLATION TO COMMITTED READ
RETAIN UPDATE LOCKS
fetch row 3
fetch row 4
commit work

Figure 8-3. When update locks are released

In an ANSI-compliant database, update cursors are usually not needed because

any select cursor behaves the same as an update cursor without the RETAIN
UPDATE LOCKS clause.

The code in Figure 8-4 on page 8-10 shows the database server promoting an
update lock to an exclusive lock. At fetch the row, the server places an update lock
on the row being fetched. At update the row, the server promotes the lock to
exclusive. At commit work, it releases the lock.

Chapter 8. Locking 8-9

 declare update cursor
begin work
open the cursor
fetch the row
do work
update the row (use WHERE CURRENT OF)
commit work

Figure 8-4. When update locks are promoted

To use an update cursor, run SELECT FOR UPDATE in your application.

Locks placed with INSERT, UPDATE, and DELETE statements

When you execute an INSERT, UPDATE, or DELETE statement, the database
server uses exclusive locks. An exclusive lock means that no other users can
update or delete the item until the database server removes the lock.

In addition, no other users can view the row unless they are using the Dirty Read
isolation level.

When the database server removes the exclusive lock depends on whether the
database supports transaction logging:
v If the database supports logging, the database server removes all exclusive locks
when the transaction completes (commits or rolls back).
v If the database does not support logging, the database server removes all
exclusive locks immediately after the INSERT, MERGE, UPDATE, or DELETE
statement completes, except when the lock is on the row that is currently being
fetched into an update cursor.
In this situation, the lock is retained during the fetch operation on the row, but
only until the server fetches the next row, or until the server updates the current
row by promoting the lock to an exclusive lock.

In a nonlogging database, the promotable update lock on a row fetched for update
can be released by a DDL operation on the database while the INSERT, MERGE,
UPDATE, or DELETE statement that originally created the lock is still running. To
reduce the risk of data corruption if a concurrent session modifies the unlocked
row, restrict operations that use promotable update locks to databases that support
transaction logging.

The internal lock table

The database server stores locks in an internal lock table. When the database server
reads a row, it checks if the row or its associated page, table, or database is listed
in the lock table. If it is in the lock table, the database server must also check the
lock type.

The following table shows the types of locks that the lock table can contain.

Lock Type Description Statement That Usually Places the Lock

S Shared lock SELECT
X Exclusive lock INSERT, UPDATE, DELETE
U Update lock SELECT in an update cursor
B Byte lock Any statement that updates VARCHAR columns

8-10 IBM Informix Performance Guide

 A byte lock is generated only if you shrink the size of a data value in a VARCHAR
column. The byte lock exists solely for roll forward and rollback execution, so a
byte lock is created only if you are working in a database that uses logging. Byte
locks appear in onstat -k output only if you are using row-level locking; otherwise,
they are merged with the page lock.

In addition, the lock table might store intent locks, with the same lock type as
previously shown. In some cases, a user might need to register his or her possible
intent to lock an item, so that other users cannot place a lock on the item.

Depending on the type of operation and the isolation level, the database server
might continue to read the row and place its own lock on the row, or it might wait
for the lock to be released (if the user executed SET LOCK MODE TO WAIT). The
following table shows the locks that a user can place if another user holds a certain
type of lock. For example, if one user holds an exclusive lock on an item, another
user requesting any kind of lock (exclusive, update, or shared) receives an error.

Hold X lock Hold U lock Hold S lock

Request X lock No No Yes
Request U lock No No Yes
Request S lock No Yes Yes

Monitoring locks
You can analyze information about locks and monitor locks by viewing
information in the internal lock table that contains stored locks.

View the lock table with onstat -k. Figure 8-5 shows sample output for onstat -k.

Locks
address wtlist owner lklist type tblsnum rowid key#/bsiz
300b77d0 0 40074140 0 HDR+S 10002 106 0
300b7828 0 40074140 300b77d0 HDR+S 10197 123 0
300b7854 0 40074140 300b7828 HDR+IX 101e4 0 0
300b78d8 0 40074140 300b7854 HDR+X 101e4 102 0
4 active, 5000 total, 8192 hash buckets

Figure 8-5. onstat -k output

In this example, a user is inserting one row in a table. The user holds the following
locks (described in the order shown):
v A shared lock on the database
v A shared lock on a row in the systables system catalog table
v An intent-exclusive lock on the table
v An exclusive lock on the row

To determine the table to which the lock applies, execute the following SQL
statement. For tblsnum, substitute the value shown in the tblsnum field in the
onstat -k output.
SELECT *
FROM SYSTABLES
WHERE HEX(PARTNUM) = "tblsnum";

Chapter 8. Locking 8-11

 Where tblsnum is the modified value that onstat -k returns. For example, if onstat
-k returns 10027f, tbslnum is 0x0010027F.

You can also query the syslocks table in the sysmaster database to obtain
information about each active lock. The syslocks table contains the following
columns.

Column Description
dbsname Database on which the lock is held
tabname Name of the table on which the lock is held
rowidlk ID of the row on which the lock is held (0
indicates a table lock.)
keynum The key number for the row
type Type of lock
owner Session ID of the lock owner
waiter Session ID of the first waiter on the lock

Configuring and managing lock usage

The LOCKS configuration parameter specifies the initial size of the internal lock
table. If the database server increases the size of the lock table, you should increase
the size of the LOCKS configuration parameter.

For information about how to determine an initial value for the LOCKS
configuration parameter, see “The LOCKS configuration parameter and memory
utilization” on page 4-15.

If the number of locks needed by sessions exceeds the value set in the LOCKS
configuration parameter, the database server attempts to increase the lock table by
doubling its size. Each time that the lock table overflows (when the number of
locks needed is greater than the current size of the lock table), the database server
increases the size of the lock table, up to 99 times. Each time that the database
server increases the size of the lock table, the server attempts to double its size.
However, the server will limit each actual increase to no more than the maximum
number of added locks shown in Table 8-1. After the 99th time that the database
server increases the lock table, the server no longer increases the size of the lock
table, and an application needing a lock receives an error.

Maximum number of locks allowed on 32-bit and 64-bit platforms

The following table shows the maximum number of allowed locks.

Table 8-1. Maximum number of locks on 32-bit and 64-bit platforms
Maximum
Maximum Number of
Maximum Number of Locks Added Maximum
Number of Dynamic Lock Per Lock Table Number of
Platform Initial Locks Table Extensions Extension Locks Allowed
32-bit 8,000,000 99 100,000 8,000,000 + (99 x
100,000)
64-bit 500,000,000 99 1,000,000 500,000,000 + (99
x 1,000,000)

8-12 IBM Informix Performance Guide

 View messages concerning increases to the size of the lock
table

Every time the database server increases the size of the lock table, the server places
a message in the message log file. You should monitor the message log file
periodically and increase the size of the LOCKS configuration parameter if you see
that the database server has increased the size of the lock table.

Monitor out-of-locks errors

To monitor the number of times that applications receive the out-of-locks error,
view the ovlock field in the output of onstat -p. You can also see similar
information from the sysprofile table in the sysmaster database. The following
rows contain the relevant statistics.

Row Description
ovlock Number of times that sessions attempted to
exceed the maximum number of locks
lockreqs Number of times that sessions requested a
lock
lockwts Number of times that sessions waited for a
lock

Examine how applications use locks

If the database server is using an unusually large number of locks, you can
examine how individual applications are using locks, as follows:
1. Monitor sessions with onstat -u to see if a particular user is using an especially
high number of locks (a high value in the locks column).
2. If a particular user uses a large number of locks, examine the SQL statements in
the application to determine whether you should lock the table or use
individual row or page locks.

A table lock is more efficient than individual row locks, but it reduces concurrency.

One way to reduce the number of locks placed on a table is to alter a table to use
page locks instead of row locks. However, page locks reduce overall concurrency
for the table, which can affect performance.

You can also reduce the number of locks placed on a table by locking the table in
exclusive mode.
Related concepts:
“The LOCKS configuration parameter and memory utilization” on page 4-15

Monitoring lock waits and lock errors

You can view information about sessions, lock usage, and lock waits.

If the application executes SET LOCK MODE TO WAIT, the database server waits
for a lock to be released instead of returning an error. An unusually long wait for a
lock can give users the impression that the application is hanging.

In Figure 8-6 on page 8-14, the onstat -u output shows that session ID 84 is waiting
for a lock (L in the first column of the Flags field). To find out the owner of the

Chapter 8. Locking 8-13

 lock, use the onstat -k command.

onstat -u

Userthreads
address flags sessid user tty wait tout locks nreads nwrites
40072010 ---P--D 7 informix - 0 0 0 35 75
400723c0 ---P--- 0 informix - 0 0 0 0 0
40072770 ---P--- 1 informix - 0 0 0 0 0
40072b20 ---P--- 2 informix - 0 0 0 0 0
40072ed0 ---P--F 0 informix - 0 0 0 0 0
40073280 ---P--B 8 informix - 0 0 0 0 0
40073630 ---P--- 9 informix - 0 0 0 0 0
400739e0 ---P--D 0 informix - 0 0 0 0 0
40073d90 ---P--- 0 informix - 0 0 0 0 0

40074140Y-BP---
81 lsuto 4 50205788 0 4 106 221
400744f0 --BP--- 83 jsmit - 0 0 4 0 0
400753b0 ---P--- 86 worth - 0 0 2 0 0
40075760 L--PR--
84 jones 3
300b78d8 -1 2 0 0
13 active, 128 total, 16 maximum concurrent

onstat -k

Locks
address wtlist owner lklist type tblsum rowid key#/bsiz
300b77d0 0 40074140 0 HDR+S 10002 106 0
300b7828 0 40074140 300b77d0 HDR+S 10197 122 0
300b7854 0 40074140 300b7828 HDR+IX 101e4 0 0

300b78d840075760
40074140300b7854 HDR+X 101e4 100 0
300b7904 0 40075760 0 S 10002 106 0
300b7930 0 40075760 300b7904 S 10197 122 0
6 active, 5000 total, 8192 hash buckets

Figure 8-6. onstat -u output that shows lock usage

To find out the owner of the lock for which session ID 84 is waiting:
1. Obtain the address of the lock in the wait field (300b78d8) of the onstat -u
output.
2. Find this address (300b78d8) in the Locks address field of the onstat -k output.
The owner field of this row in the onstat -k output contains the address of the
user thread (40074140).
3. Find this address (40074140) in the Userthreads field of the onstat -u output.
The sessid field of this row in the onstat -u output contains the session ID (81)
that owns the lock.

To eliminate the contention problem, you can have the user exit the application
gracefully. If this solution is not possible, you can stop the application process or
remove the session with onmode -z.

Monitoring the number of free locks

You can find the current number of free locks on a lock-free list by viewing the
output of the onstat -L command .
Related reference:
onstat -L command: Print the number of free locks (Administrator's Reference)

8-14 IBM Informix Performance Guide

Monitoring deadlocks
You can monitor deadlocks. A deadlock occurs when two users hold locks, and each
user wants to acquire a lock that the other user owns.

For example, user pradeep holds a lock on row 10. User jane holds a lock on row
20. Suppose that jane wants to place a lock on row 10, and pradeep wants to place
a lock on row 20. If both users execute SET LOCK MODE TO WAIT, they
potentially might wait for each other forever.

Informix uses the lock table to detect deadlocks automatically and stop them
before they occur. Before a lock is granted, the database server examines the lock
list for each user. If a user holds a lock on the resource that the requestor wants to
lock, the database server traverses the lock wait list for the user to see if the user is
waiting for any locks that the requestor holds. If so, the requestor receives a
deadlock error.

Deadlock errors can be unavoidable when applications update the same rows
frequently. However, certain applications might always be in contention with each
other. Examine applications that are producing a large number of deadlocks and
try to run them at different times.

To monitor the number of deadlocks, use the deadlks field in the output of onstat
-p.

In a distributed transaction, the database server does not examine lock tables from
other database server systems, so deadlocks cannot be detected before they occur.
Instead, you can set the DEADLOCK_TIMEOUT configuration parameter.
DEADLOCK_TIMEOUT specifies the number of seconds that the database server
waits for a remote database server response before it returns an error. Although
reasons other than a distributed deadlock might cause the delay, this mechanism
keeps a transaction from hanging indefinitely.

To monitor the number of distributed deadlock timeouts, use the dltouts field in
the onstat -p output.

Monitoring isolation levels that sessions use

The onstat -g ses and onstat -g sql output shows the isolation level that a session
is currently using.

The following table summarizes the values in the IsoLvl column in onstat -g ses
and onstat -g sql output.
Value Description
DR Dirty Read
CR Committed Read
CS Cursor Stability
CRU Committed Read with RETAIN UPDATE LOCKS
CSU Cursor Stability with RETAIN UPDATE LOCKS
DRU Dirty Read with RETAIN UPDATE LOCKS
LC Committed Read, Last Committed
RR Repeatable Read

Chapter 8. Locking 8-15

 If a great deal of lock contention occurs, check the isolation level of sessions to
make sure it is appropriate for the application.

Locks for smart large objects

Smart large objects have several unique locking behaviors because their columns
are typically much larger than other columns in a table.

This section discusses the following unique behaviors:

v Types of locks on smart large objects
v Byte-range locking
v Lock promotion
v Dirty Read isolation level with smart large objects

Types of locks on smart large objects

The database server uses one of the following granularity levels for locking smart
large objects:
v The sbspace chunk header partition
v The smart large object
v A byte range of the smart large object

The default locking granularity is at the level of the smart large object. In other
words, when you update a smart large object, by default the database server locks
the smart large object that is being updated.

Locks on the sbspace chunk header partition only occur when the database server
promotes locks on smart large objects. For more information, see “Lock promotion”
on page 8-19.

Byte-range locking
Rather than locking the entire smart large object, you can lock only a specific byte
range of a smart large object.

Byte-range locking is advantageous because it allows multiple users to update the

same smart large object simultaneously, as long as they are updating different parts
of it. Also, users can read a part of a smart large object while another user is
updating or reading a different part of the same smart large object.

Figure 8-7 on page 8-17 shows two locks placed on a single smart large object. The
first lock is on bytes 2, 3, and 4. The second lock is on byte 6 alone.

8-16 IBM Informix Performance Guide

 1
2
Lock on bytes 2 to 4 3
4
5
Lock on byte 6 6

Figure 8-7. Example of byte-range locking

How the database server manages byte-range locks

The database server manages byte-range locks in the lock table in a similar fashion
to other locks placed on rows, pages, and tables. However, the lock table must also
store the byte range.

If you place a second lock on a byte range adjacent to a byte range that is currently
locked, the database server consolidates the two locks into one lock on the entire
range.

If a user holds locks that the Figure 8-7 shows, and the user requests a lock on byte
five, the database server consolidates the locks placed on bytes two through six
into one lock.

Likewise, if a user unlocks only a portion of the bytes included within a byte-range
lock, the database server might be split into multiple byte-range locks. In the
Figure 8-7 the user could unlock byte three, which causes the database server to
change the one lock on bytes two through four to one lock on byte two and one
lock on byte four.

Using byte-range locks

By default, the database server places a lock on the smart large object. Instead, you
can enable byte-range locking.

To use byte-range locks, you must perform one of the following actions:
v To set byte-range locking for the sbspace that stores the smart large object, use
the onspaces utility. The following example sets byte-range locking for the new
sbspace:
onspaces -c -S slo -g 2 -p /ix/9.2/liz/slo -o 0 -s 1000
-Df LOCK_MODE=RANGE
When you set the default locking mode for the sbspace to byte-range locking,
the database server locks only the necessary bytes when it updates any smart
large objects stored in the sbspace.
v To set byte-range locking for the smart large object when you open it, use one of
the following methods:
– In DB-Access: Set the MI_LO_LOCKRANGE flag in the mi_lo_open()
DataBlade API function.
– In ESQL/C: Set the LO_LOCKRANGE flag in the ifx_lo_open() Informix
ESQL/C function. When you set byte-range locking for the individual smart
large object, the database server implicitly locks only the necessary bytes
when it selects or updates the smart large object.

Chapter 8. Locking 8-17

 v To lock a byte range explicitly, use one of the following functions:
– For DB-Access:mi_lo_lock()
– For ESQL/C:ifx_lo_lock()
These functions lock the range of bytes that you specify for the smart large
object. If you specify an exclusive lock with either function, UPDATE
statements do not place locks on the smart large object if they update the
locked bytes.
The database server releases exclusive byte-range locks placed with
mi_lo_lock() or ifx_lo_lock() at the end of the transaction. The database
server releases shared byte-range locks placed with mi_lo_lock() or
ifx_lo_lock() based on the same rules as locks placed with SELECT
statements, depending upon the isolation level. You can also release shared
byte-range locks with one of the following functions:
– For DB-Access:mi_lo_unlock(). For more information about the DataBlade API
functions, see the IBM Informix DataBlade API Programmer's Guide.
– For ESQL/C:ifx_lo_unlock(). For more information about Informix ESQL/C
functions, see the IBM Informix ESQL/C Programmer's Manual.

Monitoring byte-range locks

You can use onstat -k to list all byte-range locks. Use the onstat -K command to
list byte-range locks and all waiters for byte-range locks.

Figure 8-8 shows an excerpt from the output of onstat -k.

Byte-Range Locks
rowid/LOid tblsnum address status owner offset size type
104 200004 a020e90 HDR
[2, 2, 3] a020ee4 HOLD a1b46d0 50 10 S
202 200004 a021034 HDR
[2, 2, 5] a021088 HOLD a1b51e0 40 5 S
102 200004 a035608 HDR
[2, 2, 1] a0358fc HOLD a1b4148 0 500 S
a035758 HOLD a1b3638 300 100 S
21 active, 2000 total, 2048 hash buckets

Figure 8-8. Byte-range locks in onstat -k output

Byte-range locks produce the following information in the onstat -k output.

Column Description
rowid The rowid of the row that contains the
locked smart large object
LOid The three values: sbspace number, chunk
number, and sequence number (a value that
represents the position in the chunk)
tblsnum The number of the tblspace that holds the
smart large object
address The address of the lock
status The status of the lock

HDR is a placeholder. HOLD indicates the

user specified in the owner column owns
the lock. WAIT (shown only with onstat -K)
indicates that the user specified in the owner
column is waiting for the lock.

8-18 IBM Informix Performance Guide

 Column Description
owner The address of the owner (or waiter)

Cross reference this value with the address

in onstat -u.
offset The offset into the smart large object where
the bytes are locked
size The number of bytes locked, starting at the
value in the offset column
type S (shared lock) or X (exclusive)

Setting number of locks for byte-range locking

When you use byte-range locking, the database server can use more locks because
of the possibility of multiple locks on one smart large object. Even though the lock
table grows when it runs out of space, you might want to increase value of the
LOCKS configuration parameter to match lock usage so that the database server
does not need to allocate more space dynamically.

Be sure to monitor the number of locks used with onstat -k, so you can determine
if you need to increase the value of the LOCKS configuration parameter.

Lock promotion
The database server uses lock promotion to decrease the total number of locks held
on smart large objects. Too many locks can result in poorer performance because
the database server frequently searches the lock table to determine if a lock exists
on an object.

If the number of locks held by a transaction exceeds 33 percent of the current

number of allocated locks for the database server, the database server attempts to
promote any existing byte-range locks to a single lock on the smart large object.

If the number of locks that a user holds on a smart large object (not on byte ranges
of a smart large object) equals or exceeds 10 percent of the current capacity of the
lock table, the database server attempts to promote all of the smart-large-object
locks to one lock on the smart-large-object header partition. This kind of lock
promotion improves performance for applications that are updating, loading, or
deleting a large number of smart large objects. For example, a transaction that
deletes millions of smart large objects would consume the entire lock table if the
database server did not use lock promotion. The lock promotion algorithm has
deadlock avoidance built in.

You can identify a smart-large-object header partition in onstat -k by 0 in the

rowid column and a tablespace number with a high-order first byte-and-a-half that
corresponds to the dbspace number where the smart large object is stored. For
example, if the tblspace number is listed as 0x200004 (the high-order zeros are
truncated), the dbspace number 2 corresponds to the dbspace number listed in
onstat -d.

Even if the database server attempts to promote a lock, it might not be able to do
so. For example, the database server might not be able to promote byte-range locks
to one smart-large-object lock because other users have byte-range locks on the
same smart large object. If the database server cannot promote a byte-range lock, it
does not change the lock, and processing continues as normal.

Chapter 8. Locking 8-19

 Dirty Read isolation level and smart large objects
You can use the Dirty Read isolation level for smart large objects.

For information about how Dirty Reads affects consistency, see “Dirty Read
isolation” on page 8-5.

Set the Dirty Read isolation level for smart large objects in one of the following
ways:
v Use the SET TRANSACTION MODE or SET ISOLATION statement.
v Use the LO_DIRTY_READ flag in one of the following functions:
– For DB-Access:mi_lo_open()
– For ESQL/C:ifx_lo_open()

If consistency for smart large objects is not important, but consistency for other
columns in the row is important, you can set the isolation level to Committed
Read, Cursor Stability, or Repeatable Read and open the smart large object with the
LO_DIRTY_READ flag.

8-20 IBM Informix Performance Guide

Chapter 9. Fragmentation guidelines
One of the most frequent causes of poor performance in relational database
systems is contention for data that resides on a single I/O device. Proper
fragmentation of high-use tables can significantly reduce I/O contention. These
topics discuss the performance considerations that are involved when you use
table fragmentation.

The database server supports table fragmentation (also partitioning), which allows
you to store data from a single table on multiple disk devices. Fragmentation
enables you to define groups of rows or index keys within a table according to
some algorithm or scheme. You can store each group or fragment (also referred to
as a partition) in a separate dbspace associated with a specific physical disk.

For information about fragmentation and parallel execution, see Chapter 12,
“Parallel database query (PDQ),” on page 12-1.

For an introduction to fragmentation concepts and methods, see the IBM Informix
Database Design and Implementation Guide. For information about the SQL statements
that manage fragments, see the IBM Informix Guide to SQL: Syntax.

Planning a fragmentation strategy

You can decide on a fragmentation goal for your database and devise a strategy to
meet that goal.

A fragmentation strategy consists of two parts:

v A distribution scheme that specifies how to group rows into fragments
You specify the distribution scheme in the FRAGMENT BY clause of the
CREATE TABLE, CREATE INDEX, or ALTER FRAGMENT statements.
v The set of dbspaces in which you locate the fragments
You specify the set of dbspaces or in the IN clause (storage option) of these SQL
statements.

To formulate a fragmentation strategy:

1. Decide on your primary fragmentation goal, which should depend, to a large
extent, on the types of applications that access the table.
2. Make the following decisions based on your primary fragmentation goal:
v Whether to fragment the table data, the table index, or both
v What the ideal distribution of rows or index keys is for the table
3. Choose either an expression-based or round-robin distribution scheme:
v If you choose an expression-based distribution scheme, you must then design
suitable fragment expressions.
v If you choose a round-robin distribution scheme, the database server
determines which rows to put into a specific fragment.
For more information, see “Distribution schemes” on page 9-6.
4. To complete the fragmentation strategy, you must decide on the number and
location of the fragments:
v The number of fragments depends on your primary fragmentation goal.

© Copyright IBM Corp. 1996, 2014 9-1

 v Where you locate fragments depends on the number of disks available in
your configuration.

When you plan a fragmentation strategy, be aware of these space and page issues:
v Although a 4-terabyte chunk can be on a 2-kilobyte page, only 32 gigabytes can
be utilized in a dbspace because of a rowid format limitation.
v For a fragmented table, all fragments must use the same page size.
v For a fragmented index, all fragments must use the same page size.
v A table can be in one dbspace and the index for that table can be in another
dbspace. These dbspaces do not need to have the same page size.

Fragmentation goals
You can analyze your application and workload to identify fragmentation goals
and to determine the balance to strike among fragmentation goals.

Fragmentation goals can include:

v Improved performance for individual queries
To improve the performance of individual queries, fragment tables appropriately
and set resource-related parameters to specify system resource use (memory,
CPU virtual processors, and so forth).
v Reduced contention between queries and transactions
v If your database server is used primarily for online transaction processing
(OLTP) and only incidentally for decision-support queries, you can often use
fragmentation to reduce contention when simultaneous queries against a table
perform index scans to return a few rows.
v Increased data availability
Careful fragmentation of dbspaces can improve data availability if devices fail.
Table fragments on the failed device can be restored quickly, and other
fragments are still accessible.
v Improved data-load performance
When you use the High-Performance Loader (HPL) to load a table that is
fragmented across multiple disks, it allocates threads to light append the data
into the fragments in parallel. For more information about this load method, see
the IBM Informix High-Performance Loader User's Guide.
You can also use the alter fragment ON table statement with the attach clause to
add data quickly to a very large table. For more information, see “Improve the
performance of operations that attach and detach fragments” on page 9-19.

The performance of a fragmented table is primarily governed by the following

factors:
v The storage option that you use for allocating disk space to fragments (discussed
in “Considering physical fragmentation factors” on page 9-5)
v The distribution scheme used to assign rows to individual fragments (discussed
in “Distribution schemes” on page 9-6)

Improved query performance through fragmentation strategy

If the primary goal of fragmentation is improved performance for individual
queries, try to distribute all of the rows of the table evenly over the different disks.
Overall query-completion time is reduced when the database server does not need
to wait for data retrieval from a table fragment that has more rows than other
fragments.

9-2 IBM Informix Performance Guide

If queries access data by performing sequential scans against significant portions of
tables, fragment the table rows only. Do not fragment the index. If an index is
fragmented and a query has to cross a fragment boundary to access the data, the
performance of the query can be worse than if you do not fragment.

If queries access data by performing an index read, you can improve performance
by using the same distribution scheme for the index and the table.

If you use round-robin fragmentation, do not fragment your index. Consider

placing that index in a separate dbspace from other table fragments.

For more information about improving performance for queries, see “Query
expressions for fragment elimination” on page 9-15 and Chapter 13, “Improving
individual query performance,” on page 13-1.

Reduced contention between queries and transactions

Fragmentation can reduce contention for data in tables that multiple queries and
OLTP applications use. Fragmentation often reduces contention when many
simultaneous queries against a table perform index scans to return a few rows.

For tables subjected to this type of load, fragment both the index keys and data
rows with a distribution scheme that allows each query to eliminate unneeded
fragments from its scan. Use an expression-based distribution scheme. For more
information, see “Distribution schemes that eliminate fragments” on page 9-14.

To fragment a table for reduced contention, start by investigating which queries

access which parts of the table. Next, fragment your data so that some of the
queries are routed to one fragment while others access a different fragment. The
database server performs this routing when it evaluates the fragmentation rule for
the table. Finally, store the fragments on separate disks.

Your success in reducing contention depends on how much you know about the
distribution of data in the table and the scheduling of queries against the table. For
example, if the distribution of queries against the table is set up so that all rows
are accessed at roughly the same rate, try to distribute rows evenly across the
fragments. However, if certain values are accessed at a higher rate than others, you
can compensate for this difference by distributing the rows over the fragments to
balance the access rate. For more information, see “Designing an expression-based
distribution scheme” on page 9-8.

Increased data availability

When you distribute table and index fragments across different disks or devices,
you improve the availability of data during disk or device failures. The database
server continues to allow access to fragments stored on disks or devices that
remain operational.

This availability has important implications for the following types of applications:
v Applications that do not require access to unavailable fragments
A query that does not require the database server to access data in an
unavailable fragment can still successfully retrieve data from fragments that are
available. For example, if the distribution expression uses a single column, the
database server can determine if a row is contained in a fragment without
accessing the fragment. If the query accesses only rows that are contained in
available fragments, a query can succeed even when some of the data in the
table is unavailable. For more information, see “Designing an expression-based
distribution scheme” on page 9-8.

Chapter 9. Fragmentation guidelines 9-3

 v Applications that accept the unavailability of data
Some applications might be designed in such a way that they can accept the
unavailability of data in a fragment and require the ability to retrieve the data
that is available. To specify which fragments can be skipped, these applications
can execute the SET DATASKIP statement before they execute a query.
Alternatively, the database server administrator can use the onspaces -f option to
specify which fragments are unavailable.

If your fragmentation goal is increased availability of data, fragment both table

rows and index keys so that if a disk drive fails, some of the data is still available.
If applications must always be able to access a subset of your data, keep those
rows together in the same mirrored dbspace.

Increased granularity for backup and restore

You must consider backup and restore factors when you are deciding how to
distribute dbspaces across disk.

Backup and restore factors to consider are:

v Data availability. When you decide where to place your tables or fragments,
remember that if a device that contains a dbspace fails, all tables or table
fragments in that dbspace are inaccessible, even though tables and fragments in
other dbspaces are accessible. The need to limit data unavailability in the event
of a disk failure might influence which tables you group together in a particular
dbspace.
v Cold versus warm restores. Although you must perform a cold restore if a
dbspace that contains critical data fails, you need to perform only a warm
restore if a noncritical dbspace fails. The desire to minimize the impact of cold
restores might influence the dbspace that you use to store critical data.

For more information about backup and restore, see your IBM Informix Backup and
Restore Guide.

Examining your data and queries

To determine a fragmentation strategy, you must gather information about the
table that you might fragment. You must also know how the data in the table is
used.

To gather information about your table:

1. Identify the queries that are critical to performance to determine if the queries
are online transaction processing (OLTP) or decision-support system (DSS)
queries.
2. Use the SET EXPLAIN statement to determine how the data is being accessed.
For information about the output of the SET EXPLAIN statement, see “Report
that shows the query plan chosen by the optimizer” on page 10-9. To determine
how the data is accessed, you can sometimes simply review the SELECT
statements along with the table schema.
3. Determine what portion of the data each query examines.
For example, if certain rows in the table are read most of the time, you can
isolate them in a small fragment to reduce I/O contention for other fragments.
4. Determine which statements create temporary files.
Decision-support queries typically create and access large temporary files, and
placement of temporary dbspaces can be critical to performance.

9-4 IBM Informix Performance Guide

 5. If particular tables are always joined together in a decision-support query,
spread fragments for these tables across different disks.
6. Examine the columns in the table to determine which fragmentation scheme
would keep each scan thread equally busy for the decision-support queries.
To see how the column values are distributed, create a distribution on the
column with the UPDATE STATISTICS statement and examine the distribution
with dbschema.
dbschema -d database -hd table

Considering physical fragmentation factors

When you fragment a table, the physical placement issues that pertain to tables
apply to individual table fragments. Because each fragment resides in its own
dbspace on a disk, you must address these issues separately for the fragments on
each disk.

For details about placement issues that apply to tables, see Chapter 6, “Table
performance considerations,” on page 6-1.

Fragmented and nonfragmented tables differ in the following ways:

v For fragmented tables, each fragment is placed in a separate, designated dbspace
or multiple named fragments of the table are created within a single dbspace.
For nonfragmented tables, the table can be placed in the default dbspace of the
current database.
Regardless of whether the table is fragmented or not, you should create a single
chunk on each disk for each dbspace.
v Extent sizes for a fragmented table are usually smaller than the extent sizes for
an equivalent nonfragmented table because fragments do not grow in increments
as large as the entire table. For more information on how to estimate the space
to allocate, see “Estimating table size” on page 6-5.
v In a fragmented table, the row pointer is not a unique unchanging pointer to the
row on a disk. The database server uses the combination of fragment ID and
row pointer internally, inside an index, to point to the row. These two fields are
unique but can change over the life of the row. An application cannot access the
fragment ID; therefore, you should use primary keys to access a specific row in
a fragmented table. For more information, see the IBM Informix Database Design
and Implementation Guide.
v An attached index or an index on a nonfragmented table uses 4 bytes for the
row pointer. A detached index uses 8 bytes of disk space per key value for the
fragment ID and row pointer combination. For more information about how to
estimate space for an index, see “Estimating index pages” on page 7-4. For more
information on attached indexes and detached indexes, see “Strategy for
fragmenting indexes” on page 9-10.

Decision-support queries usually create and access large temporary files; placement
of temporary dbspaces is a critical factor for performance. For more information
about placement of temporary files, see “Spreading temporary tables and sort files
across multiple disks” on page 6-4.

Chapter 9. Fragmentation guidelines 9-5

Distribution schemes
After you decide whether to fragment table rows, index keys, or both, and you
decide how the rows and keys should be distributed over fragments, you can
decide on a scheme to implement this distribution. Informix supports random
distribution among fragments and value-based distribution among fragments.

Random distribution among fragments

Round-robin fragmentation
This type of fragmentation places rows one after another in fragments,
rotating through the series of fragments to distribute the rows evenly.
For smart large objects, you can specify multiple sbspaces in the PUT
clause of the CREATE TABLE or ALTER TABLE statement to distribute
smart large objects in a round-robin distribution scheme so that the
number of smart large objects in each space is approximately equal.
For INSERT statements, the database server uses a hash function on a
random number to determine the fragment in which to place the row. For
INSERT cursors, the database server places the first row in a random
fragment, the second in the next fragment sequentially, and so on. If one of
the fragments is full, it is skipped.

Value-based distribution among fragments:

Expression-based fragmentation
This type of fragmentation puts rows that contain specified values in the
same fragment. You specify a fragmentation expression that defines criteria
for assigning a set of rows to each fragment, either as a range rule or some
arbitrary rule.
You can specify a remainder fragment that holds all rows that do not match
the criteria for any other fragment, although a remainder fragment reduces
the efficiency of the expression-based distribution scheme.
List-based fragmentation
This type of fragmentation puts rows that contain specified values that
match one of the specified values in a list of discrete values in the same
fragment. For each fragment, you specify a list of one or more constant
expressions as fragment expressions that correspond to one or more columns
in the table. The column or set of columns from which the fragment
expressions are calculated is called the fragment key.
You can optionally specify a remainder fragment that holds all rows that do
not match the criteria for any other fragment. You can also optionally
specify a NULL fragment that stores rows with missing data in the
fragment key columns (because its fragment expression is NULL or IS
NULL).
The most important difference between fragmentation by list and
fragmentation by expression is that every value in the list for each
fragment must be unique among all the lists for fragments of the same
table or index.
Interval-based fragmentation
This type of fragmentation partitions data into fragments that are based on
quantified values within a specific interval within the range of fragment
key of a single numeric, DATE, or DATETIME column in the same
fragment. You specify at least one range expression as the fragment

9-6 IBM Informix Performance Guide

 expression that defines the upper limit of fragment key values for each
fragment, and an interval expression that specifies the size of the range of
system-defined fragments that the database server creates automatically.
You can optionally define a NULL fragment to store rows with missing
data in the fragment key column, but no remainder fragment is supported or
needed. The database server automatically creates a new fragment to store
rows with non-NULL fragment key values outside the range of any
existing fragment. The fragments that you define with range expressions
are called range fragments, and the system-defined fragments that the
database server creates at runtime are called interval fragments. This type of
distribution scheme is sometimes called a range interval distribution
strategy.
Related concepts:
“Specify temporary tables in the DBSPACETEMP configuration parameter” on
page 5-11
Fragmentation: Storage distribution strategies (Data Warehouse Guide)
Related reference:
Fragmenting by LIST (SQL Syntax)
List fragment clause (SQL Syntax)
Fragmenting by RANGE INTERVAL (SQL Syntax)
Interval Fragment clause (SQL Syntax)

Choosing a distribution scheme

When choosing a distribution scheme, you must consider the ease of data
balancing, whether you want fragments to be eliminated, and the effect of the data
skip feature.

Table 9-1 compares round-robin and expression-based distribution schemes.

Table 9-1. Distribution-Scheme Comparisons
Distribution
Scheme Ease of Data Balancing Fragment Elimination Data Skip
Round-robin Automatic. Data is The database server cannot You cannot determine if the integrity
balanced over time. eliminate fragments. of the transaction is compromised
when you use the data-skip feature.
However, you can insert into a table
fragmented by round-robin.
Expression-based Requires knowledge of the If expressions on one or two You can determine whether the
data distribution. columns are used, the database integrity of a transaction has been
server can eliminate fragments compromised when you use the
for queries that have either data-skip feature. You cannot insert
range or equality expressions. rows if the appropriate fragment for
those rows is down.

The distribution scheme that you choose depends on the following factors:
v The features in Table 9-1 of which you want to take advantage
v Whether or not your queries tend to scan the entire table
v Whether or not you know the distribution of data to be added
v Whether or not your applications tend to delete many rows

Chapter 9. Fragmentation guidelines 9-7

 v Whether or not you cycle your data through the table

Basically, the round-robin scheme provides the easiest and surest way of balancing
data. However, with round-robin distribution, you have no information about the
fragment in which a row is located, and the database server cannot eliminate
fragments.

In general, round-robin is the correct choice only when all the following conditions
apply:
v Your queries tend to scan the entire table.
v You do not know the distribution of data to be added.
v Your applications tend not to delete many rows. (If they do, load balancing can
be degraded.)

An expression-based scheme might be the best choice to fragment the data if any
of the following conditions apply:
v Your application calls for numerous decision-support queries that scan specific
portions of the table.
v You know what the data distribution is.
v You plan to cycle data through a database.

If you plan to add and delete large amounts of data periodically, based on the
value of a column such as date, you can use that column in the distribution
scheme. You can then use the alter fragment attach and alter fragment detach
statements to cycle the data through the table.

The ALTER FRAGMENT ATTACH and DETACH statements provide the following
advantages over bulk loads and deletes:
v The rest of the table fragments are available for other users to access. Only the
fragment that you attach or detach is not available to other users.
v With the performance enhancements, the execution of an ALTER FRAGMENT
ATTACH or DETACH statement is much faster than a bulk load or mass delete.

For more information, see “Improve the performance of operations that attach and
detach fragments” on page 9-19.

In some cases, an appropriate index scheme can circumvent the performance

problems of a particular distribution scheme. For more information, see “Strategy
for fragmenting indexes” on page 9-10.

Designing an expression-based distribution scheme

The first step in designing an expression-based distribution scheme is to determine
the distribution of data in the table, particularly the distribution of values for the
column on which you base the fragmentation expression.

To obtain this information, run the UPDATE STATISTICS statement for the table
and then use the dbschema utility to examine the distribution.

After you know the data distribution, you can design a fragmentation rule that
distributes data across fragments as required to meet your fragmentation goal. If
your primary goal is to improve performance, your fragment expression should
generate an even distribution of rows across fragments.

9-8 IBM Informix Performance Guide

 If your primary fragmentation goal is improved concurrency, analyze the queries
that access the table. If certain rows are accessed at a higher rate than others, you
can compensate by opting for an uneven distribution of data over the fragments
that you create.

Try not to use columns that are subject to frequent updates in the distribution
expression. Such updates can cause rows to move from one fragment to another
(that is, be deleted from one and added to another), and this activity increases
CPU and I/O overhead.

Try to create nonoverlapping regions based on a single column with no

REMAINDER fragment for the best fragment-elimination characteristics. The
database server eliminates fragments from query plans whenever the query
optimizer can determine that the values selected by the WHERE clause do not
reside on those fragments, based on the expression-based fragmentation rule by
which you assign rows to fragments. For more information, see “Distribution
schemes that eliminate fragments” on page 9-14.

Suggestions for improving fragmentation

You can improve fragmentation for optimal performance in decision-support and
OLTP queries.

The following suggestions are guidelines for fragmenting tables and indexes:
v For optimal performance in decision-support queries, fragment the table to
increase parallelism, but do not fragment the indexes. Detach the indexes, and
place them in a separate dbspace.
v For best performance in OLTP queries, use fragmented indexes to reduce
contention between sessions. You can often fragment an index by its key value,
which means the OLTP query only has to look at one fragment to find the
location of the row.
If the key value does not reduce contention, as when every user looks at the
same set of key values (for instance, a date range), consider fragmenting the
index on another value used in the WHERE clause. To cut down on fragment
administration, consider not fragmenting some indexes, especially if you cannot
find a good fragmentation expression to reduce contention.
v Use round-robin fragmentation on data when the table is read sequentially by
decision-support queries. Round-robin fragmentation is a good method for
spreading data evenly across disks when no column in the table can be used for
an expression-based fragmentation scheme. However, in most DSS queries, all
fragments are read.
v To reduce the total number of required dbspaces and decrease the time needed
for searches, you can store multiple named fragments within the same dbspace.
v If you are using expressions, create them so that I/O requests, rather than
quantities of data, are balanced across disks. For example, if the majority of your
queries access only a portion of data in the table, set up your fragmentation
expression to spread active portions of the table across disks, even if this
arrangement results in an uneven distribution of rows.
v Keep fragmentation expressions simple. Fragmentation expressions can be as
complex as you want. However, complex expressions take more time to evaluate
and might prevent fragments from being eliminated from queries.
v Arrange fragmentation expressions so that the most restrictive condition for each
dbspace is tested within the expression first. When the database server tests a
value against the criteria for a given fragment, evaluation stops when a

Chapter 9. Fragmentation guidelines 9-9

 condition for that fragment tests false. Thus, if the condition that is most likely
to be false is placed first, fewer conditions need to be evaluated before the
database server moves to the next fragment. For example, in the following
expression, the database server tests all six of the inequality conditions when it
attempts to insert a row with a value of 25:
x >= 1 and x <= 10 in dbspace1,
x > 10 and x <= 20 in dbspace2,
x > 20 and x <= 30 in dbspace3
By comparison, only four conditions in the following expression need to be
tested: the first inequality for dbspace1 (x <= 10), the first for dbspace2 (x <=
20), and both conditions for dbspace3:
x <= 10 and x >= 1 in dbspace1,
x <= 20 and x > 10 in dbspace2,
x <= 30 and x > 20 in dbspace3
v Avoid any expression that requires a data-type conversion. Type conversions
increase the time to evaluate the expression. For instance, a DATE data type is
implicitly converted to INTEGER for comparison purposes.
v Do not fragment on columns that change frequently unless you are willing to
incur the administration costs. For example, if you fragment on a date column
and older rows are deleted, the fragment with the oldest dates tends to empty,
and the fragment with the recent dates tends to fill up. Eventually you must
drop the old fragment and add a new fragment for newer orders.
v Do not fragment every table. Identify the critical tables that are accessed most
frequently. Put only one fragment for a table on a disk.
v Do not fragment small tables. Fragmenting a small table across many disks
might not be worth the overhead of starting all the scan threads to access the
fragments. Also, balance the number of fragments with the number of processors
on your system.
v When you define a fragmentation strategy on an unfragmented table, check the
next-extent size to ensure that you are not allocating large amounts of disk space
for each fragment.

Strategy for fragmenting indexes

When you fragment a table, the indexes that are associated with that table are
fragmented implicitly, according to the fragmentation scheme that you use. You
can also use the FRAGMENT BY clause of the CREATE INDEX statement to
fragment the index on any table explicitly.

Each index of a fragmented table occupies its own tblspace with its own extents.

You can fragment the index with either of the following strategies:
v Same fragmentation strategy as the table
v Different fragmentation strategy from the table

Attached indexes
An attached index is an index that implicitly follows the table fragmentation
strategy (distribution scheme and set of dbspaces in which the fragments are
located). The database server automatically creates an attached index when you
first fragment a table.

To create an attached index, do not specify a fragmentation strategy or storage

option in the CREATE INDEX statement, as in the following sample SQL
statements:

9-10 IBM Informix Performance Guide

CREATE TABLE tb1(a int)
FRAGMENT BY EXPRESSION
(a >=0 AND a < 5) IN dbsbspace1,
(a >=5 AND a < 10) IN dbspace2
...
;

CREATE INDEX idx1 ON tb1(a);

For fragmented tables that use expression-based or round-robin distribution

schemes, you can also create multiple partitions of a table or index within a single
dbspace. This enables you to reduce the number of required dbspaces, thereby
simplifying the management of dbspaces.

To create an attached index with partitions, include the partition name in your SQL
statements, as shown in this example:
CREATE TABLE tb1(a int)
FRAGMENT BY EXPRESSION
PARTITION part1 (a >=0 AND a < 5) IN dbs1,
PARTITION part2 (a >=5 AND a < 10) IN dbs1
...
;

CREATE INDEX idx1 ON tb1(a);

You can use "PARTITION BY EXPRESSION" instead of "FRAGMENT BY

EXPRESSION" in CREATE TABLE, CREATE INDEX, and ALTER FRAGMENT ON
INDEX statements as shown in this example:
ALTER FRAGMENT ON INDEX idx1 INIT PARTITION BY EXPRESSION
PARTITION part1 (a <= 10) IN dbs1,
PARTITION part2 (a <= 20) IN dbs1,
PARTITION part3 (a <= 30) IN dbs1;

Use ALTER FRAGMENT syntax to change fragmented indexes that do not have
partitions into indexes that have partitions. The syntax below shows how you
might convert a fragmented index into an index that contains partitions:
CREATE TABLE t1 (c1 int) FRAGMENT BY EXPRESSION
(c1=10) IN dbs1, (c1=20) IN dbs2, (c1=30) IN dbs3
CREATE INDEX ind1 ON t1 (c1) FRAGMENT BY EXPRESSION
(c1=10) IN dbs1, (c1=20) IN dbs2, (c1=30) IN dbs3
ALTER FRAGMENT ON INDEX ind1 INIT FRAGMENT BY EXPRESSION
PARTITION part_1 (c1=10) IN dbs1, PARTITION part_2 (c1=20) IN dbs1,
PARTITION part_3 (c1=30) IN dbs1,

Creating a table or index containing partitions improves performance by enabling

the database server to search more quickly and by reducing the required number
of dbspaces.

The database server fragments the attached index according to the same
distribution scheme as the table by using the same rule for index keys as for table
data. As a result, attached indexes have the following physical characteristics:
v The number of index fragments is the same as the number of data fragments.
v Each attached index fragment resides in the same dbspace as the corresponding
table data, but in a separate tblspace.
v An attached index or an index on a nonfragmented table uses 4 bytes for the
row pointer for each index entry. For more information about how to estimate
space for an index, see “Estimating index pages” on page 7-4.

Chapter 9. Fragmentation guidelines 9-11

 Informix does not support forest of trees attached indexes.

Detached indexes
A detached index is an index with a separate fragmentation strategy that you set up
explicitly with the CREATE INDEX statement.

The following sample SQL statements create a detached index:

CREATE TABLE tb1 (a int)
FRAGMENT BY EXPRESSION
(a <= 10) IN tabdbspc1,
(a <= 20) IN tabdbspc2,
(a <= 30) IN tabdbspc3;

CREATE INDEX idx1 ON tb1 (a)

FRAGMENT BY EXPRESSION
(a <= 10) IN idxdbspc1,
(a <= 20) IN idxdbspc2,
(a <= 30) IN idxdbspc3;

This example illustrates a common fragmentation strategy, to fragment indexes in

the same way as the tables, but specify different dbspaces for the index fragments.
This fragmentation strategy of putting the index fragments in different dbspaces
from the table can improve the performance of operations such as backup,
recovery, and so forth.

By default, all new indexes that the CREATE INDEX statement creates are
detached and stored in separate tablespaces from the data unless the deprecated
IN TABLE syntax is specified.

To create a detached index with partitions, include the partition name in your SQL
statements, as shown in this example:
CREATE TABLE tb1 (a int)
FRAGMENT BY EXPRESSION
PARTITION part1 (a <= 10) IN dbs1,
PARTITION part2 (a <= 20) IN dbs2,
PARTITION part3 (a <= 30) IN dbs3;

CREATE INDEX idx1 ON tb1 (a)

FRAGMENT BY EXPRESSION
PARTITION part1 (a <= 10) IN dbs1,
PARTITION part2 (a <= 20) IN dbs2,
PARTITION part3 (a <= 30) IN dbs3;

You can use the PARTITION BY EXPRESSION keywords instead of the FRAGMENT BY
EXPRESSION keywords in the CREATE TABLE, CREATE INDEX, and ALTER
FRAGMENT ON INDEX statements.

If you do not want to fragment the index, you can put the entire index in a
separate dbspace.

You can fragment the index for any table by expression. However, you cannot
explicitly create a round-robin fragmentation scheme for an index. Whenever you
fragment a table using a round-robin fragmentation scheme, convert all indexes
that accompany the table to detached indexes for the best performance.

Detached indexes have the following physical characteristics:

9-12 IBM Informix Performance Guide

 v Each detached index fragment resides in a different tblspace from the
corresponding table data. Therefore, the data and index pages cannot be
interleaved within the tblspace.
v Detached index fragments have their own extents and tblspace IDs. The tblspace
ID is also known as the fragment ID and partition number. A detached index uses
8 bytes of disk space per index entry for the fragment ID and row pointer
combination. For more information on how to estimate space for an index, see
“Estimating index pages” on page 7-4.

Forest of trees indexes are detached indexes. They cannot be attached indexes.

The database server stores the location of each table and index fragment, along
with other related information, in the sysfragments system catalog table. You can
view the sysfragments system catalog table to access information about
fragmented tables and indexes, including the following :
v The value in the partn column is the partition number or fragment id of the
table or index fragment. The partition number for a detached index is different
from the partition number of the corresponding table fragment.
v The value in the strategy column is the distribution scheme used in the
fragmentation strategy.

For a complete description of column values that the sysfragments system catalog
table contains, see the IBM Informix Guide to SQL: Reference. For information about
how to use sysfragments to monitor your fragments, see “Monitoring fragment
use” on page 9-28.

Restrictions on indexes for fragmented tables

If the database server scans a fragmented index, multiple index fragments must be
scanned and the results merged together. (The exception is if the index is
fragmented according to some index-key range rule, and the scan does not cross a
fragment boundary.) Because of this requirement, performance on index scans
might suffer if the index is fragmented.

Because of these performance considerations, the database server places the

following restrictions on indexes:
v You cannot fragment indexes by round-robin.
v You cannot fragment unique indexes by an expression that contains columns
that are not in the index key.

For example, the following statement is not valid:

CREATE UNIQUE INDEX ia on tab1(col1)
FRAGMENT BY EXPRESSION
col2<10 in dbsp1,
col2>=10 AND col2<100 in dbsp2,
col2>100 in dbsp3;

Strategy for fragmenting temporary tables

You can fragment an explicit temporary table across dbspaces that reside on
different disks.

You can create a temporary, fragmented table with the TEMP TABLE clause of the
CREATE TABLE statement. However, you cannot alter the fragmentation strategy

Chapter 9. Fragmentation guidelines 9-13

 of fragmented temporary tables (as you can with permanent tables). The database
server deletes the fragments that are created for a temporary table at the same time
that it deletes the temporary table.

You can define your own fragmentation strategy for an explicit temporary table, or
you can let the database server dynamically determine the fragmentation strategy.

For more information about explicit and implicit temporary tables, see your IBM
Informix Administrator's Guide.

Distribution schemes that eliminate fragments

Fragment elimination is a database server feature that reduces the number of
fragments involved in a database operation. This capability can improve
performance significantly and reduce contention for the disks on which fragments
reside.

Fragment elimination improves both response time for a given query and
concurrency between queries. Because the database server does not need to read in
unnecessary fragments, I/O for a query is reduced. Activity in the LRU queues is
also reduced.

If you use an appropriate distribution scheme, the database server can eliminate
fragments from the following database operations:
v The fetch portion of the SELECT, INSERT, delete or update statements in SQL
The database server can eliminate fragments when these SQL statements are
optimized, before the actual search.
v Nested-loop joins
When the database server obtains the key value from the outer table, it can
eliminate fragments to search on the inner table.

Whether the database server can eliminate fragments from a search depends on
two factors:
v The distribution scheme in the fragmentation strategy of the table that is being
searched
v The form of the query expression (the expression in the WHERE clause of a
SELECT, INSERT, delete or update statement)

Fragmentation expressions for fragment elimination

Some operators in expressions result in automatic fragment elimination.

When the fragmentation strategy is defined with any of the following operators,
fragment elimination can occur for a query on the table.
IN
=
<
>
<=
>=
AND
OR
NOT
IS NULL (only when not combined with other expressions using AND or OR operators)

9-14 IBM Informix Performance Guide

 If the fragmentation expression uses any of the following operators, fragment
elimination does not occur for queries on the table.
!=
IS NOT NULL

For examples of fragmentation expressions that allow fragment elimination, see

“Effectiveness of fragment elimination” on page 9-16.

Query expressions for fragment elimination

A query expression (the expression in the WHERE clause) can consist of simple
expressions, not simple expressions, and multiple expressions.

The database server considers only simple expressions or multiple simple

expressions combined with certain operators for fragment elimination.

A simple expression consists of the following parts:

column operator value
Simple Expression Part
Description
column Is a single column name
The database server supports fragment elimination on all column types
except columns that are defined with the NCHAR, NVARCHAR, BYTE,
and TEXT data types.
operator
Must be an equality or range operator
value Must be a literal or a host variable

The following examples show simple expressions:

name = "Fred"
date < "08/25/2008"
value >= :my_val

The following examples are not simple expressions:

unitcost * count > 4500
price <= avg(price)
result + 3 > :limit

The database server considers two types of simple expressions for fragment
elimination, based on the operator:
v Range expressions
v Equality expressions

Range expressions in query

The database server can handle one or two column fragment elimination on
queries with any combination of five relational operators in the WHERE clause.

Range expressions use the following relational operators:

<
>
<=
>=
!=

Chapter 9. Fragmentation guidelines 9-15

 The database server can also eliminate fragments when these range expressions are
combined with the following operators:
AND, OR, NOT
IS NULL, IS NOT NULL
MATCH, LIKE

If the range expression contains MATCH or LIKE, the database server can also
eliminate fragments if the string does not begin with a wildcard character. The
following examples show query expressions that can take advantage of fragment
elimination:
columna MATCH "ab*"
columna LIKE "ab%" OR columnb LIKE "ab*"

Equality expressions in query

The database server can handle one or multiple column fragment elimination on
queries with a combination of equality operators in the WHERE clause.

Equality expressions use the following equality operators:

=, IN

The database server can also eliminate fragments when these equality expressions
are combined with the following operators:
AND, OR

Effectiveness of fragment elimination

The database server cannot eliminate fragments when you fragment a table with a
round-robin distribution scheme. Furthermore, not all expression-based distribution
schemes give you the same fragment-elimination behavior.

The following table summarizes the fragment-elimination behavior for different

combinations of expression-based distribution schemes and query expressions.
Partitions in fragmented tables do not affect the fragment-elimination behavior
shown in the following table.
Table 9-2. Fragment elimination for different types of expression-based distribution schemes
and query expressions
Overlapping or
Type of Query Nonoverlapping Non-contiguous Nonoverlapping
(WHERE clause) Fragments on a Fragments on a Single Fragments on
Expression Single Column Column Multiple Columns
Range expression Fragments can be Fragments cannot be Fragments cannot be
eliminated. eliminated. eliminated.
Equality Fragments can be Fragments can be Fragments can be
expression eliminated. eliminated. eliminated.

This table shows that the distribution schemes enable fragment elimination, but the
effectiveness of fragment elimination is determined by the WHERE clause of the
specified query.

For example, consider a table fragmented with the following expression:

FRAGMENT BY EXPRESSION
100 < column_a AND column_b < 0 IN dbsp1,
100 >= column_a AND column_b < 0 IN dbsp2,
column_b >= 0 IN dbsp3

9-16 IBM Informix Performance Guide

The database server cannot eliminate any fragments from the search if the WHERE
clause has the following expression:
column_a = 5 OR column_b = -50

However, the database server can eliminate the fragment in dbspace dbsp3 if the
WHERE clause has the following expression:
column_b = -50

Furthermore, the database server can eliminate the two fragments in dbspaces
dbsp2 and dbsp3 if the WHERE clause has the following expression:
column_a = 5 AND column_b = -50

Partitions in fragmented tables do not affect fragment-elimination behavior.

Nonoverlapping fragments on a single column

A fragmentation rule that creates nonoverlapping fragments on a single column is
the preferred fragmentation rule from a fragment-elimination standpoint.

The advantage of this type of distribution scheme is that the database server can
eliminate fragments for queries with range expressions as well as queries with
equality expressions. You should meet these conditions when you design your
fragmentation rule. Figure 9-1 gives an example of this type of fragmentation rule.

...
FRAGMENT BY EXPRESSION
a<=8 OR a IN (9,10) IN dbsp1,
10<a AND a<=20 IN dbsp2,
a IN (21,22,23) IN dbsp3,
a>23 IN dbsp4;

Figure 9-1. Example of nonoverlapping fragments on a single column

You can create nonoverlapping fragments using a range rule or an arbitrary rule
based on a single column. You can use relational operators, as well as AND, IN,
OR, or BETWEEN. Be careful when you use the BETWEEN operator. When the
database server parses the BETWEEN keyword, it includes the end points that you
specify in the range of values. Avoid using a REMAINDER clause in your
expression. If you use a REMAINDER clause, the database server cannot always
eliminate the remainder fragment.

Overlapping fragments on a single column

The fragments on a single column can be overlapping and noncontiguous. You can
use any range, MOD function, or arbitrary rule that is based on a single column.

The only restriction for this category of fragmentation rule is that you base the
fragmentation rule on a single column.

Figure 9-2 on page 9-18 shows an example of this type of fragmentation rule.

Chapter 9. Fragmentation guidelines 9-17

 ...
FRAGMENT BY EXPRESSION
a<=8 OR a IN (9,10,21,22,23) IN dbsp1,
a>10 IN dbsp2;

Figure 9-2. Example of overlapping fragments on a single column

If you use this type of distribution scheme, the database server can eliminate
fragments on an equality search but not a range search. This distribution scheme
can still be useful because all INSERT and many UPDATE operations perform
equality searches.

This alternative is acceptable if you cannot use an expression that creates

nonoverlapping fragments with contiguous values. For example, in cases where a
table is growing over time, you might want to use a MOD function rule to keep
the fragments of similar size. Expression-based distribution schemes that use MOD
function rules fall into this category because the values in each fragment are not
contiguous.

Nonoverlapping fragments, multiple columns

The database server uses an arbitrary rule to define nonoverlapping fragments
based on multiple columns.

The following figures show an example of nonoverlapping fragments on two

columns.

...
FRAGMENT BY EXPRESSION
0<a AND a<=10 AND b IN ('E’, 'F’, 'G’) IN dbsp1,
0<a AND a<=10 AND b IN ('H’, 'I’, 'J’) IN dbsp2,
10<a AND a<=20 AND b IN ('E’, 'F’, 'G’) IN dbsp3,
10<a AND a<=20 AND b IN ('H’, 'I’, 'J’) IN dbsp4,
20<a AND a<=30 AND b IN ('E’, 'F’, 'G’) IN dbsp5,
20<a AND a<=30 AND b IN ('H’, 'I’, 'J’) IN dbsp6;

Figure 9-3. Example of nonoverlapping fragments on two columns

9-18 IBM Informix Performance Guide

 column b

b IN ('H', 'I','J')

b IN ('E', 'F','G')

column a
0 < a <= 10 10 < a <= 20 20 < a <= 30

Figure 9-4. Schematic example of nonoverlapping fragments on two columns

If you use this type of distribution scheme, the database server can eliminate
fragments on an equality search but not a range search. This capability can still be
useful because all INSERT operations and many UPDATE operations perform
equality searches. Avoid using a REMAINDER clause in the expression. If you use
a REMAINDER clause, the database server cannot always eliminate the remainder
fragment.

This alternative is acceptable if you cannot obtain sufficient granularity using an

expression based on a single column.

Improve the performance of operations that attach and detach

fragments
When you use ALTER FRAGMENT ATTACH and DETACH statements to add or
remove a large amount of data in a very large table, you can take steps to improve
the performance of the attach and detach operations.

The ALTER FRAGMENT DETACH statement provides a way to delete a segment

of the table data rapidly. Similarly, the ALTER FRAGMENT ATTACH statement
provides a way to load large amounts of data incrementally into an existing table
by taking advantage of the fragmentation technology. However, the ALTER
FRAGMENT ATTACH and DETACH statements can take a long time to execute
when the database server rebuilds indexes on the surviving table.

The database server provides performance optimizations for the ALTER

FRAGMENT ATTACH and DETACH statements that cause the database server to
reuse the indexes on the surviving tables. Therefore, the database server can
eliminate the index build during the attach or detach operation, which:
v Reduces the time that it takes for the ALTER FRAGMENT ATTACH and ALTER
FRAGMENT DETACH statements to execute
v Improves the table availability

Chapter 9. Fragmentation guidelines 9-19

 The ALTER FRAGMENT operation requires exclusive access and exclusive locks on
all of the tables involved in the operation. If you enable the FORCE_DDL_EXEC
environment option for a session, you can force out other transactions that have
opened or have locks on the tables involved in an ALTER FRAGMENT ON TABLE
operation. If the server is unable to get exclusive access and exclusive locks on the
table, the server will start rolling back the transactions that are open or have locks
on the table until the value specified with the FORCE_DDL_EXEC option is
reached. You might want to enable the FORCE_DDL_EXEC environment option
this on a busy system, perhaps one that runs 24 hours a day, and you do not want
to wait for sessions to close before you alter a fragment.

Improving ALTER FRAGMENT ATTACH performance

You can take advantage of the performance optimizations for the ALTER
FRAGMENT ATTACH statement if your database meets certain requirements.

To take advantage of the performance optimization, you must meet all of the
following requirements:
v Formulate appropriate distribution schemes for your table and index fragments.
v Ensure that no data movement occurs between the resultant partitions due to
fragment expressions.
v Update statistics for all the participating tables.
v Make the indexes on the attached tables unique if the index on the surviving
table is unique.

Important: Only logging databases can benefit from the performance

improvements for the ALTER FRAGMENT ATTACH statement. Without logging,
the database server works with multiple copies of the same table to ensure
recoverability of the data when a failure occurs. This requirement prevents reuse of
the existing index fragments.

Distribution schemes for reusing indexes

You can use one of three distribution schemes that allow the attach operation of
the ALTER FRAGMENT statement to reuse existing indexes.

These distributions schemes are:

v Fragmenting the index in the same way as the table
v Fragmenting the index with the same set of fragment expressions as the table
v Attaching unfragmented tables to form a fragmented table

Fragmenting the index in the same way as the table:

You fragment an index in the same way as the table when you create an index
without specifying a fragmentation strategy.

A fragmentation strategy is the distribution scheme and set of dbspaces in which

the fragments are located. For details, see “Planning a fragmentation strategy” on
page 9-1.

Example of Fragmenting the Index in the Same Way as the Table

Suppose you create a fragmented table and index with the following SQL
statements:

9-20 IBM Informix Performance Guide

CREATE TABLE tb1(a int)
FRAGMENT BY EXPRESSION
(a >=0 AND a < 5) IN db1,
(a >=5 AND a <10) IN db2;

CREATE INDEX idx1 ON tb1(a);

Suppose you then create another table that is not fragmented, and you
subsequently decide to attach it to the fragmented table.
CREATE TABLE tb2 (a int, CHECK (a >=10 AND a<15))
IN db3;

CREATE INDEX idx2 ON tb2(a)

IN db3;

ALTER FRAGMENT ON TABLE tb1

ATTACH
tb2 AS (a >= 10 and a<15) AFTER db2;

This attach operation can take advantage of the existing index idx2 if no data
movement occurs between the existing and the new table fragments. If no data
movement occurs:
v The database server reuses index idx2 and converts it to a fragment of index
idx1.
v The index idx1 remains as an index with the same fragmentation strategy as the
table tb1.

If the database server discovers that one or more rows in the table tb2 belong to
preexisting fragments of the table tb1, the database server:
v Drops and rebuilds the index idx1 to include the rows that were originally in
tables tb1 and tb2
v Drops the index idx2

For more information about how to ensure no data movement between the existing
and the new table fragments, see “Ensuring no data movement when you attach a
fragment” on page 9-23.

Fragmenting the index with the same distribution scheme as the table:

You fragment an index with the same distribution scheme as the table when you
create an index that uses the same fragment expressions as the table.

The database server determines if the fragment expressions are identical, based on
the equivalency of the expression tree instead of the algebraic equivalence. For
example, consider the following two expressions:
(col1 >= 5)
(col1 = 5 OR col1 > 5)

Although these two expressions are algebraically equivalent, they are not identical
expressions.

Example of Fragmenting the Index with the Same Distribution Scheme as the
Table

Suppose you create two fragmented tables and indexes with the following SQL
statements:

Chapter 9. Fragmentation guidelines 9-21

 CREATE TABLE tb1 (a INT)
FRAGMENT BY EXPRESSION
(a <= 10) IN tabdbspc1,
(a <= 20) IN tabdbspc2,
(a <= 30) IN tabdbspc3;
CREATE INDEX idx1 ON tb1 (a)
FRAGMENT BY EXPRESSION
(a <= 10) IN idxdbspc1,
(a <= 20) IN idxdbspc2,
(a <= 30) IN idxdbspc3;

CREATE TABLE tb2 (a INT CHECK a> 30 AND a<= 40)

IN tabdbspc4;
CREATE INDEX idx2 ON tb2(a)
IN idxdbspc4;

Suppose you then attach table tb2 to table tb1 with the following sample SQL
statement:
ALTER FRAGMENT ON TABLE tb1
ATTACH tb2 AS (a <= 40);

The database server can eliminate the rebuild of index idx1 for this attach
operation for the following reasons:
v The fragmentation expression for index idx1 is identical to the fragmentation
expression for table tb1. The database server:
– Expands the fragmentation of the index idx1 to the dbspace idxdbspc4
– Converts index idx2 to a fragment of index idx1
v No rows move from one fragment to another because the CHECK constraint is
identical to the resulting fragmentation expression of the attached table.
For more information about how to ensure no data movement between the
existing and the new table fragments, see “Ensuring no data movement when
you attach a fragment” on page 9-23.

Attaching unfragmented tables together:

You can take advantage of the performance benefits of the ALTER FRAGMENT
ATTACH operation when you combine two unfragmented tables into one
fragmented table.

For example, suppose you create two unfragmented tables and indexes with the
following SQL statements:
CREATE TABLE tb1(a int) IN db1;
CREATE INDEX idx1 ON tb1(a) in db1;
CREATE TABLE tb2(a int) IN db2;
CREATE INDEX idx2 ON tb2(a) in db2;

You might want to combine these two unfragmented tables with the following
sample distribution scheme:
ALTER FRAGMENT ON TABLE tb1
ATTACH
tb1 AS (a <= 100),
tb2 AS (a > 100);

If no data migrates between the fragments of tb1 and tb2, the database server
redefines index idx1 with the following fragmentation strategy:

9-22 IBM Informix Performance Guide

CREATE INDEX idx1 ON tb1(a) F
FRAGMENT BY EXPRESSION
a <= 100 IN db1,
a > 100 IN db2;

Important: This behavior results in a different fragmentation strategy for the index
prior to Version 7.3 and Version 9.2 of the database server. In earlier versions, the
ALTER FRAGMENT ATTACH statement creates an unfragmented detached index
in the dbspace db1.

Ensuring no data movement when you attach a fragment

You can ensure there is no data movement when you attach a fragment by
establishing identical check constraint expressions and verifying that fragment
expressions are not overlapping.

To ensure that no data movement occurs when you attach a fragment:

1. Establish a check constraint on the attached table that is identical to the
fragment expression that it will assume after the ALTER FRAGMENT ATTACH
operation.
2. Define the fragments with nonoverlapping expressions.

For example, you might create a fragmented table and index with the following
SQL statements:
CREATE TABLE tb1(a int)
FRAGMENT BY EXPRESSION
(a >=0 AND a < 5) IN db1,
(a >=5 AND a <10) IN db2;

CREATE INDEX idx1 ON tb1(a);

Suppose you create another table that is not fragmented, and you subsequently
decide to attach it to the fragmented table.
CREATE TABLE tb2 (a int, check (a >=10 and a<15))
IN db3;

CREATE INDEX idx2 ON tb2(a)

IN db3;

ALTER FRAGMENT ON TABLE tb1

ATTACH
tb2 AS (a >= 10 AND a<15) AFTER db2;

This ALTER FRAGMENT ATTACH operation takes advantage of the existing index
idx2 because the following steps were performed in the example to prevent data
movement between the existing and the new table fragment:
v The check constraint expression in the CREATE TABLE tb2 statement is identical
to the fragment expression for table tb2 in the ALTER FRAGMENT ATTACH
statement.
v The fragment expressions specified in the CREATE TABLE tb1 and the ALTER
FRAGMENT ATTACH statements are not overlapping.

Therefore, the database server preserves index idx2 in dbspace db3 and converts it
into a fragment of index idx1. The index idx1 remains as an index with the same
fragmentation strategy as the table tb1.

Indexes on attached tables

The database server tries to reuse the indexes on the attached tables as fragments
of the resultant index. However, the corresponding index on the attached table
Chapter 9. Fragmentation guidelines 9-23
 might not exist or might not be usable due to disk-format mismatches. In these
cases, it might be faster to build an index on the attached tables rather than to
build the entire index on the resultant table.

Informix estimates the cost to create the whole index on the resultant table. The
server then compares this cost to the cost of building the individual index
fragments for the attached tables and chooses the index build with the least cost.

Automatically Gathered Statistics for New Indexes

When the CREATE INDEX statement runs successfully, with or without the
ONLINE keyword, Informix automatically gathers the following statistics for the
newly created index:
v Index-level statistics, equivalent to the statistics gathered in the UPDATE
STATISTICS operation in LOW mode, for all types of indexes, including B-tree,
Virtual Index Interface, and functional indexes.
v Column-distribution statistics, equivalent to the distribution generated in the
UPDATE STATISTICS operation in HIGH mode, for a non-opaque leading
indexed column of an ordinary B-tree index. The resolution of the HIGH mode
is 1.0 for a table size that is less than 1 million rows and 0.5 for higher table
sizes. Tables with more than 1 million rows have a better resolution because they
have more bins for statistics.

The automatically gathered distribution statistics are available to the query

optimizer when it designs query plans for the table on which the new index was
created.

Run UPDATE STATISTICS Before Attaching Tables

To ensure that cost estimates are correct, you should execute the UPDATE
STATISTICS statement on all of the participating tables before you attach the
tables. The LOW mode of the UPDATE STATISTICS statement is sufficient to
derive the appropriate statistics for the optimizer to determine cost estimates for
rebuilding indexes.

For more information about using the UPDATE STATISTICS statement, see the
IBM Informix Guide to SQL: Syntax.

Example for situation when corresponding index does not exist:

When a table does not have an index on a column that can serve as the fragment
of the resultant index, the database server estimates the cost of building the index
fragment for the column, compares this cost to rebuilding the entire index for all
fragments on the resultant table, and chooses the index build with the least cost.

Suppose you create a fragmented table and index with the following SQL
statements:
CREATE TABLE tb1(a int, b int)
FRAGMENT BY EXPRESSION
(a >=0 AND a < 5) IN db1,
(a >=5 AND a <10) IN db2;
CREATE INDEX idx1 ON tb1(a);

Suppose you then create two more tables that are not fragmented, and you
subsequently decide to attach them to the fragmented table.

9-24 IBM Informix Performance Guide

CREATE TABLE tb2 (a int, b int,
CHECK (a >=10 and a<15)) IN db3;
CREATE INDEX idx2 ON tb2(a) IN db3;
CREATE TABLE tb3 (a int, b int,
CHECK (a >= 15 and a<20)) IN db4;
CREATE INDEX idx3 ON tb3(b) IN db4;

ALTER FRAGMENT ON TABLE tb1

ATTACH tb2 AS (a >= 10 and a<15) tb3 AS (a >= 15 and a<20);

The three CREATE INDEX statements automatically calculate distribution statistics

for the leading column of each index in HIGH mode, as well as index statistics and
table statistics in LOW mode.

The only time the UPDATE STATISTICS LOW FOR TABLE statement is required is after
a CREATE INDEX statement in a situation in which the table has other preexisting
indexes, as shown in this example:
CREATE TABLE tb1(col1 int, col2 int);
CREATE INDEX index idx1 on tb1(col1);
(equivalent to update stats low on table tb1)
LOAD from tb1.unl insert into tb1; (load some data)
CREATE INDEX idx2 on tb1(col2);

The statement CREATE INDEX idx2 on tb1(col2) is NOT completely equivalent to

UPDATE STATISTICS LOW FOR TABLE tb1, because the CREATE INDEX statement
does not update index- level statistics for the preexisting index called idx1.

In the preceding example, table tb3 does not have an index on column a that can
serve as the fragment of the resultant index idx1. The database server estimates the
cost of building the index fragment for column a on the consumed table tb3 and
compares this cost to rebuilding the entire index for all fragments on the resultant
table. The database server chooses the index build with the least cost.

Example for situation when index on table is not usable:

When the index on a table is not usable, the database server estimates the cost of
building the index fragment, compares this cost to rebuilding the entire index for
all fragments on the resultant table, and chooses the index build with the least
cost.

Suppose you create tables and indexes as in the previous section, but the index on
the third table specifies a dbspace that the first table also uses. The following SQL
statements show this scenario:
CREATE TABLE tb1(a int, b int)
FRAGMENT BY EXPRESSION
(a >=0 AND a < 5) IN db1,
(a >=5 AND a <10) IN db2;
CREATE INDEX idx1 ON tb1(a);
CREATE TABLE tb2 (a int, b int, check (a >=10 and a<15))
IN db3;
CREATE INDEX idx2 ON tb2(a)
IN db3;

CREATE TABLE tb3 (a int, b int, check (a >= 15 and a<20))

IN db4;
CREATE INDEX idx3 ON tb3(a)
IN db2 ;

Chapter 9. Fragmentation guidelines 9-25

 This example creates the index idx3 on table tb3 in the dbspace db2. As a result,
index idx3 is not usable because index idx1 already has a fragment in the dbspace
db2, and the fragmentation strategy does not allow more than one fragment to be
specified in a given dbspace.

Again, the database server estimates the cost of building the index fragment for
column a on the consumed table tb3 and compares this cost to rebuilding the
entire index idx1 for all fragments on the resultant table. Then the database server
chooses the index build with the least cost.

Improving ALTER FRAGMENT DETACH performance

You can improve the performance of ALTER FRAGMENT DETACH statements by
formulating appropriate distribution schemes for your table and index fragments
and by eliminating the index build during the execution of ALTER FRAGMENT
DETACH statements.

To eliminate the index build during execution of the ALTER FRAGMENT

DETACH statement, use one of the following fragmentation strategies:
v Fragment the index in the same way as the table.
v Fragment the index with the same distribution scheme as the table.

Important: Only logging databases can benefit from the performance

improvements for the ALTER FRAGMENT DETACH statement. Without logging,
the database server works with multiple copies of the same table to ensure
recoverability of the data when a failure occurs. This requirement prevents reuse of
the existing index fragments.

Fragmenting the index in the same way as the table

You fragment an index in the same way that you fragment the table when you
create a fragmented table and subsequently create an index without specifying a
fragmentation strategy.

For example, suppose you create a fragmented table and index with the following
SQL statements:
CREATE TABLE tb1(a int)
FRAGMENT BY EXPRESSION
(a >=0 AND a < 5) IN db1,
(a >=5 AND a <10) IN db2,
(a >=10 AND a <15) IN db3;
CREATE INDEX idx1 ON tb1(a);

The database server fragments the index keys into dbspaces db1, db2, and db3
with the same column a value ranges as the table because the CREATE INDEX
statement does not specify a fragmentation strategy.

Suppose you then decide to detach the data in the third fragment with the
following SQL statement:
ALTER FRAGMENT ON TABLE tb1
DETACH db3 tb3;

Because the fragmentation strategy of the index is the same as the table, the
ALTER FRAGMENT DETACH statement does not rebuild the index after the
detach operation. The database server drops the fragment of the index in dbspace
db3, updates the system catalog tables, and eliminates the index build.

9-26 IBM Informix Performance Guide

 Fragmenting the index using same distribution scheme as the
table
You fragment an index with the same distribution scheme as the table when you
create the index that uses the same fragment expressions as the table.

A common fragmentation strategy is to fragment indexes in the same way as the

tables but to specify different dbspaces for the index fragments. This fragmentation
strategy of putting the index fragments into different dbspaces from the table can
improve the performance of operations such as backup, recovery, and so forth.

For example, suppose you create a fragmented table and index with the following
SQL statements:
CREATE TABLE tb1(a int, b int)
FRAGMENT BY EXPRESSION
(a >=0 AND a < 5) IN db1,
(a >=5 AND a <10) IN db2,
(a >=10 AND a <15) IN db3;

CREATE INDEX idx1 on tb1(a)

FRAGMENT BY EXPRESSION
(a >=0 AND a< 5) IN db4,
(a >=5 AND a< 10) IN db5,
(a >=10 AND a<15) IN db6;

Suppose you then decide to detach the data in the third fragment with the
following SQL statement:
ALTER FRAGMENT ON TABLE tb1
DETACH db3 tb3;

Because the distribution scheme of the index is the same as the table, the ALTER
FRAGMENT DETACH statement does not rebuild the index after the detach
operation. The database server drops the fragment of the index in dbspace db3,
updates the system catalog tables, and eliminates the index build.

Forcing out transactions when altering table fragments

You can enable the server to force out transactions that have opened or have locks
on the tables when you issue an ALTER FRAGMENT ON TABLE operation in a
logging database. You might want to do this on a busy system, perhaps one that
runs 24 hours a day, and you do not want to wait for sessions to close before you
alter a fragment.

Be aware that if you enable the server to force out the transactions, the server will
roll back other users' transactions. The server also closes the hold cursors during
rollback by the session that performs the ALTER FRAGMENT ON TABLE
operation.

Prerequisites:
v You must be user informix or have DBA privileges on the database.
v The table must be in a logging database.

To force out transactions when altering a table fragment:

1. Set the FORCE_DDL_EXEC environment option of the SET ENVIRONMENT
statement to one of the following values:

Chapter 9. Fragmentation guidelines 9-27

 v ’ON’, ’on’, or ’1’ to enable the server to force out transactions that are open
or have a lock on the table when an ALTER FRAGMENT ON TABLE
statement is issued until the server gets a lock and exclusive access on the
table.
v A numeric positive integer that represents an amount of time in seconds. The
numeric value enables the server to force out transactions until the server
gets exclusive access and exclusive locks on the table or until the specified
time limit occurs. If the server cannot force out transactions by the specified
amount of time, the server stops attempting to force out the transactions.
For example, to enable the FORCE_DDL_EXEC environment option to operate
for 100 seconds when an ALTER FRAGMENT ON TABLE statement is issued,
specify:
SET ENVIRONMENT FORCE_DDL_EXEC ’100’;
2. Set the lock mode to wait to ensure that the server will wait a specified amount
of time before forcing out any transactions.
For example, to set the lock mode to wait for 20 seconds, specify:
SET LOCK MODE TO WAIT 20;
For more information, see “Setting the lock mode to wait” on page 8-5.
3. Run an ALTER FRAGMENT ON TABLE statement, for example, to attach,
detach, modify, add, or drop the fragment.

After you complete an ALTER FRAGMENT ON TABLE operation with the

FORCE_DDL_EXEC environment option enabled, you can turn the
FORCE_DDL_EXEC environment option off. For example, specify:
SET ENVIRONMENT FORCE_DDL_EXEC ’OFF’
Related reference:
FORCE_DDL_EXEC Environment Option (SQL Syntax)

Monitoring fragment use

Once you determine a fragmentation strategy, you can monitor fragmentation.

You can monitor fragmentation in the following ways:

v Run individual onstat utility commands to capture information about specific
aspects of a running query.
v Run a SET EXPLAIN statement before you run a query to write the query plan
to an output file.

Monitoring fragmentation with the onstat -g ppf command

With the onstat -g ppf command, you can view partition information and monitor
the I/O activity to verify your strategy and determine whether the I/O is balanced
across fragments.

The onstat -g ppf output includes the number of read-and-write requests sent to
each fragment that is currently open. Because a request can trigger multiple I/O
operations, these requests do not indicate how many individual disk I/O
operations occur, but you can get a good idea of the I/O activity from the
displayed columns.

The brfd column in the output displays the number of buffer reads in pages. (Each
buffer can contain one page.) This information is useful if you need to monitor the
time a query takes to execute. Typically query execution time has a strong

9-28 IBM Informix Performance Guide

 dependency on the number of required buffer reads. If the size of client-server
buffering is small and your database contains TEXT data, query execution time can
involve significantly more buffer reads, because the server reads the prior TEXT
data.

The onstat -g ppf output by itself does not identify the table in which a fragment
is located. To determine the table for the fragment, join the partnum column in the
output to the partnum column in the sysfragments system catalog table. The
sysfragments table displays the associated table id. You can also find the table
name for the fragment by joining the table id column in sysfragments to the table
id column in systables.

To determine the table name in onstat -g ppf output:

1. Obtain the value in the partnum field of the onstat -g ppf output.
2. Join the tabid column in the sysfragments system catalog table with the tabid
column in the systables system catalog table to obtain the table name from
systables.
Use the partnum field value that you obtain in step 1 in the SELECT statement.
SELECT a.tabname FROM systables a, sysfragments b
WHERE a.tabid = b.tabid
AND partn = partnum_value;

Monitoring fragmentation with SET EXPLAIN output

When the table is fragmented, the output of the SET EXPLAIN ON statement
shows which table or index the database server scans to execute the query.

The SET EXPLAIN output identifies the fragments with a fragment number. The
fragment numbers are the same as those contained in the partn column in the
sysfragments system catalog table.

The following example of partial SET EXPLAIN output shows a query that takes
advantage of fragment elimination and scans two fragments in table t1:
QUERY:
------
SELECT * FROM t1 WHERE c1 > 12

Estimated Cost: 3
Estimated # of Rows Returned: 2

1) informix.t1: SEQUENTIAL SCAN (Serial, fragments: 1, 2)

Filters: informix.t1.c1 > 12

If the optimizer must scan all fragments (that is, if it is unable to eliminate any
fragment from consideration), the SET EXPLAIN output displays fragments: ALL. In
addition, if the optimizer eliminates all the fragments from consideration (that is,
none of the fragments contain the queried information), the SET EXPLAIN output
displays fragments: NONE.

For information about how the database server eliminates a fragment from
consideration, see “Distribution schemes that eliminate fragments” on page 9-14.

For more information about the SET EXPLAIN ON statement, see “Report that
shows the query plan chosen by the optimizer” on page 10-9.

Chapter 9. Fragmentation guidelines 9-29

9-30 IBM Informix Performance Guide
Chapter 10. Queries and the query optimizer
These topics describe query plans, explain how the database server manages query
optimization, and discuss factors that you can use to influence the query plan.
These topics also describe performance considerations for SPL routines, the UDR
cache, and triggers.

The parallel database query (PDQ) features in the database server provide the
largest potential performance improvements for a query. Chapter 12, “Parallel
database query (PDQ),” on page 12-1 describes PDQ and the Memory Grant
Manager (MGM) and explains how to control resource use by queries.

PDQ provides the most substantial performance gains if you fragment your tables
as described in Chapter 9, “Fragmentation guidelines,” on page 9-1.

Chapter 13, “Improving individual query performance,” on page 13-1 explains how
to improve the performance of specific queries.

Data warehouse queries and performance issues related to dimensional databases

are described in the IBM Informix Data Warehouse Guide.
Related concepts:
Performance tuning dimensional databases (Data Warehouse Guide)

The query plan

The query optimizer evaluates the different ways in which a query might be
performed and determines the best way to select the requested data. During this
evaluation, the optimizer formulates a query plan to fetch the data rows that are
required to process a query.

For example, when evaluating the different ways in which a query might be
performed, the optimizer must determine whether indexes should be used. If the
query includes a join, the optimizer must determine the join plan (hash or nested
loop) and the order in which tables are evaluated or joined.

The following topics describe the components of a query plan and show examples
of query plans.

The access plan

The way that the optimizer chooses to read a table is called an access plan. The
simplest method to access a table is to read it sequentially, which is called a table
scan. The optimizer chooses a table scan when most of the table must be read or
the table does not have an index that is useful for the query.

The optimizer can also choose to access the table by an index. If the column in the
index is the same as a column in a filter of the query, the optimizer can use the
index to retrieve only the rows that the query requires. The optimizer can use a
key-only index scan if the columns requested are within one index on the table. The
database server retrieves the needed data from the index and does not access the
associated table.

© Copyright IBM Corp. 1996, 2014 10-1

 Important: The optimizer does not choose a key-only scan for a VARCHAR
column. If you want to take advantage of key-only scans, use the ALTER TABLE
with the MODIFY clause to change the column to a CHAR data type.

The optimizer compares the cost of each plan to determine the best one. The
database server derives cost from estimates of the number of I/O operations
required, calculations to produce the results, rows accessed, sorting, and so forth.

The join plan

When a query contains more than one table, Informix joins the tables using filters
in the query. The way that the optimizer chooses to join the tables is the join plan.

In the following query, the customer and orders table are joined by the
customer.customer_num = orders.customer_num filter:
SELECT * from customer, orders
WHERE customer.customer_num = orders.customer_num
AND customer.lname = "Higgins";

The join method can be a nested-loop join or a hash join.

Because of the nature of hash joins, an application with isolation level set to
Repeatable Read might temporarily lock all the records in tables that are involved
in the join, including records that fail to qualify the join. This situation leads to
decreased concurrency among connections. Conversely, nested-loop joins lock
fewer records but provide reduced performance when a large number of rows are
accessed. Thus, each join method has advantages and disadvantages.

Nested-loop join
In a nested-loop join, the database server scans the first, or outer table, and then
joins each of the rows that pass table filters to the rows found in the second, or
inner table.

Figure 10-1 on page 10-3 shows tables and rows, and the order they are read, for
query:
SELECT * FROM customer, orders
WHERE customer.customer_num=orders.customer_num
AND order_date>"01/01/2007";

The database server accesses an outer table by an index or by a table scan. The
database server applies any table filters first. For each row that satisfies the filters
on the outer table, the database server reads the inner table to find a match.

The database server reads the inner table once for every row in the outer table that
fulfills the table filters. Because of the potentially large number of times that the
inner table can be read, the database server usually accesses the inner table by an
index.

10-2 IBM Informix Performance Guide

 customer orders
ordernum custno
custno custname 6692 1234
1234 XYZLTD 6693 1234
1235 XSPORTS 6695 1235

1. Scan outer table. 2. Read inner table once

for each row found in
outer table.

Figure 10-1. Nested-loop join

If the inner table does not have an index, the database server might construct an
autoindex at the time of query execution. The optimizer might determine that the
cost to construct an autoindex at the time of query execution is less than the cost to
scan the inner table for each qualifying row in the outer table.

If the optimizer changes a subquery to a nested-loop join, it might use a variation

of the nested-loop join, called a semi join. In a semi join, the database server reads
the inner table only until it finds a match. In other words, for each row in the
outer table, the inner table contributes at most one row. For more information on
how the optimizer handles subqueries, see “Query plans for subqueries” on page
10-15.

Hash join
The optimizer usually uses a hash join when at least one of the two join tables
does not have an index on the join column or when the database server must read
a large number of rows from both tables. No index and no sorting is required
when the database server performs a hash join.

A hash join consists of two activities: first building the hash table (build phase) and
then probing the hash table (probe phase). Figure 10-2 on page 10-4 shows the hash
join in detail.

In the build phase, the database server reads one table and, after it applies any
filters, creates a hash table. Think of a hash table conceptually as a series of buckets,
each with an address that is derived from the key value by applying a hash
function. The database server does not sort keys in a particular hash bucket.

Smaller hash tables can fit in the virtual portion of database server shared memory.
The database server stores larger hash files on disk in the dbspace specified by the
DBSPACETEMP configuration parameter or the DBSPACETEMP environment
variable.

In the probe phase, the database server reads the other table in the join and applies
any filters. For each row that satisfies the filters on the table, the database server
applies the hash function on the key and probes the hash table to find a match.

Chapter 10. Queries and the query optimizer 10-3

 orders customer
ordernum custno amount custno custname
6692 1234 $27.50 1234 XYZLTD
6693 1235 $38.90S 1235 XSPORTS

1. Create hash table (apply filters first). 2. Probe hash table.

Hash table
bucket rows
3978 6692
4588 6693

Figure 10-2. How a hash join is executed

Join order
The order that tables are joined in a query is extremely important. A poor join
order can cause query performance to decline noticeably.

The following SELECT statement calls for a three-way join:

SELECT C.customer_num, O.order_num
FROM customer C, orders O, items I
WHERE C.customer_num = O.customer_num
AND O.order_num = I.order_num

The optimizer can choose one of the following join orders:

v Join customer to orders. Join the result to items.
v Join orders to customer. Join the result to items.
v Join customer to items. Join the result to orders.
v Join items to customer. Join the result to orders.
v Join orders to items. Join the result to customer.
v Join items to orders. Join the result to customer.

For an example of how the database server executes a plan according to a specific
join order, see “Example of query-plan execution.”

Example of query-plan execution

This topic contains an example of a query with a SELECT statement that calls for a
three-way join and describes one possible query plan.

The following SELECT statement calls for a three-way join:

SELECT C.customer_num, O.order_num
FROM customer C, orders O, items I
WHERE C.customer_num = O.customer_num
AND O.order_num = I.order_num

Assume also that no indexes are on any of the three tables. Suppose that the
optimizer chooses the customer-orders-items path and the nested-loop join for
both joins (in reality, the optimizer usually chooses a hash join for two tables
without indexes on the join columns). Figure 10-3 on page 10-5 shows the query
plan, expressed in pseudocode. For information about interpreting query plan
information, see “Report that shows the query plan chosen by the optimizer” on
page 10-9.

10-4 IBM Informix Performance Guide

 for each row in the customer table do:
read the row into C
for each row in the orders table do:
read the row into O
if O.customer_num = C.customer_num then
for each row in the items table do:
read the row into I
if I.order_num = O.order_num then
accept the row and send to user
end if
end for
end if
end for
end for

Figure 10-3. A query plan in pseudocode

This procedure reads the following rows:

v All rows of the customer table once
v All rows of the orders table once for each row of the customer table
v All rows of the items table once for each row of the customer-orders pair

This example does not describe the only possible query plan. Another plan merely
reverses the roles of customer and orders: for each row of orders, it reads all rows
of customer, looking for a matching customer_num. It reads the same number of
rows in a different order and produces the same set of rows in a different order. In
this example, no difference exists in the amount of work that the two possible
query plans need to do.

Example of a join with column filters

The presence of a column filter can change the query plan. A column filter is a
WHERE expression that reduces the number of rows that a table contributes to a
join.

The following example shows the query described in “Example of query-plan

execution” on page 10-4 with a filter added:
SELECT C.customer_num, O.order_num
FROM customer C, orders O, items I
WHERE C.customer_num = O.customer_num
AND O.order_num = I.order_num
AND O.paid_date IS NULL

The expression O.paid_date IS NULL filters out some rows, reducing the number
of rows that are used from the orders table. Consider a plan that starts by reading
from orders. Figure 10-4 on page 10-6 displays this sample plan in pseudocode.

Chapter 10. Queries and the query optimizer 10-5

 for each row in the orders table do:
read the row into O
if O.paid_date is null then
for each row in the customer table do:
read the row into C
if O.customer_num = C.customer_num then
for each row in the items table do:
read the row into I
if I.order_num = O.order_num then
accept row and return to user
end if
end for
end if
end for
end if
end for

Figure 10-4. Query plan that uses a column filter

Let pdnull represent the number of rows in orders that pass the filter. It is the value
of COUNT(*) that results from the following query:
SELECT COUNT(*) FROM orders WHERE paid_date IS NULL

If one customer exists for every order, the plan in Figure 10-4 reads the following
rows:
v All rows of the orders table once
v All rows of the customer table, pdnull times
v All rows of the items table, pdnull times

Figure 10-5 shows an alternative execution plan that reads from the customer table
first.

for each row in the customer table do:

read the row into C
for each row in the orders table do:
read the row into O
if O.paid_date is null and
O.customer_num = C.customer_num then
for each row in the items table do:
read the row into I
if I.order_num = O.order_num then
accept row and return to user
end if
end for
end if
end for

Figure 10-5. The alternative query plan in pseudocode

Because the filter is not applied in the first step that Figure 10-5 shows, this plan
reads the following rows:
v All rows of the customer table once
v All rows of the orders table once for every row of customer
v All rows of the items table, pdnull times

The query plans in Figure 10-4 and Figure 10-5 produce the same output in a
different sequence. They differ in that one reads a table pdnull times, and the other

10-6 IBM Informix Performance Guide

reads a table SELECT COUNT(*) FROM customer times. By choosing the
appropriate plan, the optimizer can save thousands of disk accesses in a real
application.

Example of a join with indexes

The presence of indexes and constraints in query plans provides the optimizer
with options that can greatly improve query-execution time.

This topic shows the outline of a query plan that differs from query shown in
“Example of a join with column filters” on page 10-5, because it is constructed
using indexes.

for each row in the customer table do:

read the row into C
look up C.customer_num in index on orders.customer_num
for each matching row in the orders index do:
read the table row for O
if O.paid_date is null then
look up O.order_num in index on items.order_num
for each matching row in the items index do:
read the row for I
construct output row and return to user
end for
end if
end for
end for

Figure 10-6. Query plan with indexes

The keys in an index are sorted so that when the database server finds the first
matching entry, it can read any other rows with identical keys without further
searching, because they are located in physically adjacent positions. This query
plan reads only the following rows:
v All rows of the customer table once
v All rows of the orders table once (because each order is associated with only one
customer)
v Only rows in the items table that match pdnull rows from the customer-orders
pairs

This query plan achieves a great reduction in cost compared with plans that do not
use indexes. An inverse plan, reading orders first and looking up rows in the
customer table by its index, is also feasible by the same reasoning.

The physical order of rows in a table also affects the cost of index use. To the
degree that a table is ordered relative to an index, the overhead of accessing
multiple table rows in index order is reduced. For example, if the orders table
rows are physically ordered according to the customer number, multiple retrievals
of orders for a given customer would proceed more rapidly than if the table were
ordered randomly.

In some cases, using an index might incur additional costs. For more information,
see “Index lookup costs” on page 10-26.

Chapter 10. Queries and the query optimizer 10-7

 Query plans that include an index self-join path
An index self-join is a type of index scan that you can think of as a union of many
small index scans, each one with a single unique combination of lead-key columns
and filters on non-lead-key columns.

The union of small index scans results in an access path that uses only subsets of
the full range of a composite index. The table is logically joined to itself, and the
more selective non-leading index keys are applied as index-bound filters to each
unique combination of the leading key values.

An index self-join is beneficial for situations in which:

v The lead key of an index has many duplicates, and
v Predicates on the lead key are not selective, but predicates on the non-leading
index keys are selective.

The query in Figure 10-7 shows the SET EXPLAIN output for a query plan that
includes an index self-join path.

QUERY:
------
SELECT a.c1,a.c2,a.c3 FROM tab1 a WHERE (a.c3 >= 100103) AND
(a.c3 <= 100104) AND (a.c1 >= ’PICKED ’) AND
(a.c1 <= ’RGA2 ’) AND (a.c2 >= 1) AND (a.c2 <= 7)
ORDER BY 1, 2, 3

Estimated Cost: 155

Estimated # of Rows Returned: 1
1) informix.a: INDEX PATH
(1) Index Keys: c1 c2 c3 c4 c5 (Key-Only) (Serial, fragments: ALL)
Index Self Join Keys (c1 c2 )
Lower bound: informix.a.c1 >= ’PICKED ’ AND (informix.a.c2 >= 1 )
Upper bound: informix.a.c1 <= ’RGA2 ’ AND (informix.a.c2 <= 7 )
Lower Index Filter: (informix.a.c1 = informix.a.c1 AND
informix.a.c2 = informix.a.c2 ) AND informix.a.c3 >= 100103
Upper Index Filter: informix.a.c3 <= 100104
Index Key Filters: (informix.a.c2 <= 7 ) AND
(informix.a.c2 >= 1 )

Figure 10-7. SET EXPLAIN output for a query with an index self-join path

In Figure 10-7, an index exists on columns c1, c2, c3, c4, and c5. The optimizer
chooses c1 and c2 as lead keys, which implies that columns c1 and c2 have many
duplicates. Column c3 has few duplicates and thus the predicates on column c3 (c3
>= 100103 and c3 <= 100104) have good selectivity.

As Figure 10-7 shows, an index self-join path is a self-join of two index scans using
the same index. The first index scan retrieves each unique value for lead key
columns, which are c1 and c2. The unique value of c1 and c2 is then used to probe
the second index scan, which also uses predicates on column c3. Because
predicates on column c3 have good selectivity:
v The index scan on the inner side of the nested-loop join is very efficient,
retrieving only the few rows that satisfy the c3 predicates.
v The index scan does not retrieve extra rows.

Thus, for each unique value of c1 and c2, an efficient index scan on c1, c2 and c3
occurs.

10-8 IBM Informix Performance Guide

 The following lines in the example indicate that the optimizer has chosen an index
self join path for this table, with columns c1 and c2 as the lead keys for the index
self-join path:
Index Self Join Keys (c1 c2 )
Lower bound: informix.a.c1 >= ’PICKED ’ AND (informix.a.c2 >= 1 )
Upper bound: informix.a.c1 <= ’RGA2 ’ AND (informix.a.c2 <= 7 )

The example shows the bounds for columns c1 and c2, which you can conceive of
as the bounds for the index scan to retrieve the qualified leading keys of the index.

The following information in the example shows the self-join:

(informix.a.c1 = informix.a.c1 AND informix.a.c2 = informix.a.c2 )

This information represents the inner index scan. For lead key columns c1 and c2
the self- join predicate is used, indicating the value of c1 and c2 comes from the
outer index scan. The predicates on column c3 serve as an index filter that makes
the inner index scan efficient.

Regular index scans do not use filters on column c3 to position the index scan,
because the lead key columns c1 and c2 do not have equality predicates.

The INDEX_SJ directive forces an index self-join path using the specified index, or
choosing the least costly index in a list of indexes, even if data distribution
statistics are not available for the leading index key columns. The
AVOID_INDEX_SJ directive prevents a self-join path for the specified index or
indexes. Also see “Access-method directives” on page 11-4 and the IBM Informix
Guide to SQL: Syntax.

Query plan evaluation

The optimizer considers all query plans by analyzing factors such as disk I/O and
CPU costs.

The optimizer constructs all feasible plans simultaneously using a bottom-up,

breadth-first search strategy. That is, the optimizer first constructs all possible join
pairs. It eliminates the more expensive pair of any redundant pair. (Redundant pairs
are join pairs that contain the same tables and produce the same set of rows as
another join pair.)

For example, if neither join specifies an ordered set of rows by using the ORDER
BY or GROUP BY clauses of the SELECT statement, the join pair (A x B) is
redundant with respect to (B x A).

If the query uses additional tables, the optimizer joins each remaining pair to a
new table to form all possible join triplets, eliminating the more expensive of
redundant triplets and so on for each additional table to be joined. When a
non-redundant set of possible join combinations has been generated, the optimizer
selects the plan that appears to have the lowest execution cost.

Report that shows the query plan chosen by the optimizer

Any user who runs a query can use the SET EXPLAIN statement or the EXPLAIN
directive to display the query plan that the optimizer chooses.

Chapter 10. Queries and the query optimizer 10-9

 For information about how to specify the directives, see “EXPLAIN directives” on
page 11-8. The user enters the SET EXPLAIN ON statement or the SET EXPLAIN
ON AVOID_EXECUTE statement before the SQL statement for the query, as the
following example shows.
SET EXPLAIN ON AVOID_EXECUTE;
SELECT * FROM customer, orders
WHERE customer.customer_num = orders.customer_num
AND customer.lname = "Higgins";

If a user does not have any access to SQL code source, the Database Administrator
can set dynamically the SET EXPLAIN using the onmode -Y command.

After the database server executes the SET EXPLAIN ON statement or sets
dynamically the SET EXPLAIN with onmode -Y command, the server writes an
explanation of each query plan to a file for subsequent queries that the user enters.
Related concepts:
“The explain output file”
“Query statistics section provides performance debugging information” on page
10-11
Using the FILE TO option (SQL Syntax)
Default name and location of the explain output file on UNIX (SQL Syntax)
Default name and location of the output file on Windows (SQL Syntax)
“Report that shows the query plan chosen by the optimizer” on page 10-9
“Enabling external directives” on page 11-15
Related reference:
SET EXPLAIN statement (SQL Syntax)
onmode -Y: Dynamically change SET EXPLAIN (Administrator's Reference)
onmode and Y arguments: Change query plan measurements for a session
(SQL administration API) (Administrator's Reference)

The explain output file

The SET EXPLAIN statement enables or disables recording measurements of
queries in the current session, including the plan of the query optimizer, an
estimate of the number of rows returned, and the relative cost of the query. The
measurements appear in an output file.

When you run the onmode -Y command to turn on dynamic SET EXPLAIN, the
output is displayed in a new explain output file. If a file from the SET EXPLAIN
statement exists, the database server stops using it, and instead uses the file
created by onmode -Y until the administrator turns off dynamic SET EXPLAIN for
the session.

The output file specifies if external directives are in effect.

The following codes in the Query Statistics section of the explain output file
provide information about external tables:
v xlcnv identifies an operation that is loading data from an external table and
inserting the data into a base table. Here x = external table, l = loading, and cnv
= converter

10-10 IBM Informix Performance Guide

v xucnv identifies an operation that is unloading data from an external table and
inserting the data into a base table. Here x = external table, u = unloading, and
cnv = converter

The Query Statistics section of the explain output file is a useful resource for
debugging performance problems. See “Query statistics section provides
performance debugging information.”
Related concepts:
Using the FILE TO option (SQL Syntax)
Default name and location of the explain output file on UNIX (SQL Syntax)
Default name and location of the output file on Windows (SQL Syntax)
“Report that shows the query plan chosen by the optimizer” on page 10-9
“Query statistics section provides performance debugging information”
“Enabling external directives” on page 11-15
Related reference:
SET EXPLAIN statement (SQL Syntax)
onmode -Y: Dynamically change SET EXPLAIN (Administrator's Reference)
onmode and Y arguments: Change query plan measurements for a session
(SQL administration API) (Administrator's Reference)

Query statistics section provides performance debugging

information
If the EXPLAIN_STAT configuration parameter is enabled, a Query Statistics
section appears in the explain output file that the SET EXPLAIN statement of SQL
and the onmode -Y session_id command displays.

The Query Statistics section of the explain output file shows the estimated number
of rows that the query plan expects to return, the actual number of returned rows,
and other information about the query. You can use this information, which
provides an indication of the overall flow of the query plan and how many rows
flow through each stage of the query, to debug performance problems.

The following example shows query statistics in SET EXPLAIN output. If the
estimated and actual number of rows scanned or joined are quite different, the
statistics on those tables might be old and should be updated.

Chapter 10. Queries and the query optimizer 10-11

 select * from tab1, tab2 where tab1.c1 = tab2.c1 and tab1.c3 between 0 and 15

Estimated Cost: 104

Estimated # of Rows Returned: 69

1) zelaine.tab2: SEQUENTIAL SCAN

2) zelaine.tab1: INDEX PATH

(1) Index Keys: c1 c3 (Serial, fragments: ALL)

Lower Index Filter: (zelaine.tab1.c1 = zelaine.tab2.c1
AND zelaine.tab1.c3 >= 0 )
Upper Index Filter: zelaine.tab1.c3 <= 15
NESTED LOOP JOIN

Query statistics:
-----------------

Table map :
----------------------------
Internal name Table name
----------------------------
t1 tab2
t2 tab1

type table rows_prod est_rows rows_scan time est_cost

-------------------------------------------------------------------
scan t1 50 50 50 00:00:00 4

type table rows_prod est_rows rows_scan time est_cost

-------------------------------------------------------------------
scan t2 67 69 4 00:00:00 2

type rows_prod est_rows time est_cost

-------------------------------------------------
nljoin 67 70 00:00:00 104

Figure 10-8. Query statistics in SET EXPLAIN output

Related concepts:
“The explain output file” on page 10-10
Using the FILE TO option (SQL Syntax)
Default name and location of the explain output file on UNIX (SQL Syntax)
Default name and location of the output file on Windows (SQL Syntax)
“Report that shows the query plan chosen by the optimizer” on page 10-9
“Sample query plan reports”
“Enabling external directives” on page 11-15
Related reference:
SET EXPLAIN statement (SQL Syntax)
onmode -Y: Dynamically change SET EXPLAIN (Administrator's Reference)
onmode and Y arguments: Change query plan measurements for a session
(SQL administration API) (Administrator's Reference)

Sample query plan reports

The topics in this section describe sample query plans that you might want to
display when analyzing the performance of different kinds of queries.

10-12 IBM Informix Performance Guide

Related concepts:
“Query statistics section provides performance debugging information” on page
10-11

Single-table query
This topic shows sample SET EXPLAIN output for a simple query and a complex
query on a single table.

Figure 10-9 shows SET EXPLAIN output for a simple query.

QUERY:
------
SELECT fname, lname, company FROM customer

Estimated Cost: 2
Estimated # of Rows Returned: 28

1) virginia.customer: SEQUENTIAL SCAN

Figure 10-9. Partial SET EXPLAIN output for a simple query

Figure 10-10 shows SET EXPLAIN output for a complex query on the customer
table.

QUERY:
------
SELECT fname, lname, company FROM customer
WHERE company MATCHES ’Sport*’ AND
customer_num BETWEEN 110 AND 115
ORDER BY lname

Estimated Cost: 1
Estimated # of Rows Returned: 1
Temporary Files Required For: Order By

1) virginia.customer: INDEX PATH

Filters: virginia.customer.company MATCHES ’Sport*’

(1) Index Keys: customer_num (Serial, fragments: ALL)

Lower Index Filter: virginia.customer.customer_num >= 110
Upper Index Filter: virginia.customer.customer_num <= 115

Figure 10-10. Partial SET EXPLAIN output for a complex query

The following output lines in Figure 10-10 show the scope of the index scan for the
second query:
v Lower Index Filter: virginia.customer.customer_num >= 110
Start the index scan with the index key value of 110.
v Upper Index Filter: virginia.customer.customer_num <= 115
Stop the index scan with the index key value of 115.

Multitable query
This topic shows sample SET EXPLAIN output for a multiple-table query.

Chapter 10. Queries and the query optimizer 10-13

 QUERY:
------
SELECT C.customer_num, O.order_num, SUM (I.total_price)
FROM customer C, orders O, items I
WHERE C.customer_num = O.customer_num
AND O.order_num = I.order_num
GROUP BY C.customer_num, O.order_num

Estimated Cost: 78
Estimated # of Rows Returned: 1
Temporary Files Required For: Group By

1) virginia.o: SEQUENTIAL SCAN

2) virginia.c: INDEX PATH

(1) Index Keys: customer_num (Key-Only) (Serial, fragments: ALL)

Lower Index Filter:
virginia.c.customer_num = virginia.o.customer_num
NESTED LOOP JOIN

3) virginia.i: INDEX PATH

(1) Index Keys: order_num (Serial, fragments: ALL)

Lower Index Filter: virginia.o.order_num = virginia.i.order_num
NESTED LOOP JOIN

Figure 10-11. Partial SET EXPLAIN output for a multi-table query

The SET EXPLAIN output lists the order in which the database server accesses the
tables and the access plan to read each table. The plan in Figure 10-11 indicates
that the database server is to perform the following actions:
1. The database server is to read the orders table first.
Because no filter exists on the orders table, the database server must read all
rows. Reading the table in physical order is the least expensive approach.
2. For each row of orders, the database server is to search for matching rows in
the customer table.
The search uses the index on customer_num. The notation Key-Only means that
only the index need be read for the customer table because only the
c.customer_num column is used in the join and the output, and the column is
an index key.
3. For each row of orders that has a matching customer_num, the database server
is to search for a match in the items table using the index on order_num.

Key-first scan
This topic shows a sample query that uses a key-first scan, which is an index scan
that uses keys other than those listed as lower and upper index filters.

10-14 IBM Informix Performance Guide

 create index idx1 on tab1(c1, c2);

select * from tab1 where (c1 > 0) and ( (c2 = 1) or (c2 = 2))
Estimated Cost: 4
Estimated # of Rows Returned: 1

1) pubs.tab1: INDEX PATH

(1) Index Keys: c1 c2 (Key-First) (Serial, fragments: ALL)

Lower Index Filter: pubs.tab1.c1 > 0
Index Key Filters: (pubs.tab1.c2 = 1 OR pubs.tab1.c2 = 2)

Figure 10-12. Partial SET EXPLAIN output for a key-first scan

Even though in this example the database server must eventually read the row
data to return the query results, it attempts to reduce the number of possible rows
by applying additional key filters first. The database server uses the index to apply
the additional filter, c2 = 1 OR c2 = 2, before it reads the row data.

Query plans for subqueries

The optimizer can change a subquery to a join automatically if the join provides a
lower cost.

For example, Figure 10-13 sample output of the SET EXPLAIN ON statement
shows that the optimizer changes the table in the subquery to be the inner table in
a join.

QUERY:
------
SELECT company, fname, lname, phone
FROM customer c
WHERE EXISTS(
SELECT customer_num FROM cust_calls u
WHERE c.customer_num = u.customer_num)

Estimated Cost: 6
Estimated # of Rows Returned: 7

1) virginia.c: SEQUENTIAL SCAN

2) virginia.u: INDEX PATH (First Row)

(1) Index Keys: customer_num call_dtime (Key-Only)

(Serial, fragments: ALL)
Lower Index Filter: virginia.c.customer_num = virginia.u.customer_num
NESTED LOOP JOIN (Semi Join)

Figure 10-13. Partial SET EXPLAIN output for a flattened subquery

For more information about the SET EXPLAIN ON statement, see “Report that
shows the query plan chosen by the optimizer” on page 10-9.

When the optimizer changes a subquery to a join, it can use several variations of
the access plan and the join plan:
v First-row scan
A first-row scan is a variation of a table scan. When the database server finds
one match, the table scan halts.
v Skip-duplicate-index scan

Chapter 10. Queries and the query optimizer 10-15

 The skip-duplicate-index scan is a variation of an index scan. The database
server does not scan duplicates.
v Semi join
The semi join is a variation of a nested-loop join. The database server halts the
inner-table scan when the first match is found. For more information about a
semi join, see “Nested-loop join” on page 10-2.

Query plans for collection-derived tables

A collection-derived table is a special method that the database server uses to
process a query on a collection. To use a collection-derived table, a query must
contain the TABLE keyword in the FROM clause of an SQL statement.

For more information about how to use collection-derived tables in an SQL

statement, see the IBM Informix Guide to SQL: Syntax.

Although the database does not actually create a table for the collection, it
processes the data as if it were a table. Collection-derived tables allow developers
to use fewer cursors and host variables to access a collection, in some cases.

These SQL statements create a collection column called children:

CREATE ROW TYPE person(name CHAR(255), id INT);
CREATE TABLE parents(name CHAR(255),
id INT,
children LIST(person NOT NULL));

The following query creates a collection-derived table for the children column and
treats the elements of this collection as rows in a table:
SELECT name, id
FROM TABLE(MUTLISET(SELECT children
FROM parents
WHERE parents.id
= 1001)) c_table(name, id);

Alternatively, you can specify a collection-derived table in the FROM clause, as

shown in this example:
SELECT name, id
FROM (SELECT children
FROM parents
WHERE parents.id
= 1001) c_table(name, id);

Example showing how the database server completes the query:

Informix performs several steps when completing a query for collection-derived

tables.

When completing a query, the database server performs the steps shown in this
example:
1. Scans the parent table to find the row where parents.id = 1001
This operation is listed as a SEQUENTIAL SCAN in the SET EXPLAIN output
that Figure 10-14 on page 10-17 shows.
2. Reads the value of the collection column called children.
3. Scans the single collection and returns the value of name and id to the
application.
This operation is listed as a COLLECTION SCAN in the SET EXPLAIN output
that Figure 10-14 on page 10-17 shows.

10-16 IBM Informix Performance Guide

 QUERY:
------
SELECT name, id
FROM (SELECT children
FROM parents
WHERE parents.id
= 1001) c_table(name, id);

Estimated Cost: 2
Estimated # of Rows Returned: 1

1) lsuto.c_table: COLLECTION SCAN

Subquery:
---------
Estimated Cost: 1
Estimated # of Rows Returned: 1

1) lsuto.parents: SEQUENTIAL SCAN

Filters: lsuto.parents.id = 1001

Figure 10-14. Query plan that uses a collection-derived table

Derived tables folded into parent queries:

You can improve the performance of collection-derived tables by using SQL to fold
derived tables in simple queries into a parent query instead of into query results
that are put into a temporary table.

Use SQL like that in this example:

select * from ((select col1, col2 from tab1)) as vtab(c1,c2)

However, if the query is complex because it involves aggregates, ORDER BY

operations, or the UNION operation, the server creates a temporary table.

The database server folds derived tables in a manner that is similar to the way the
server folds views through the IFX_FOLDVIEW configuration parameter (described
in “Enable view folding to improve query performance” on page 13-31). When the
IFX_FOLDVIEW configuration parameter is enabled, views are folded into a parent
query. The views are not folded into query results that are put into a temporary
table.

The following examples show derived tables folded into the main query.

Chapter 10. Queries and the query optimizer 10-17

 select * from ((select vcol0, tab1.col1 from
table(multiset(select col2 from tab2 where col2 > 50 ))
vtab2(vcol0),tab1 )) vtab1(vcol1,vcol2)
where vcol1 = vcol2

Estimated Cost: 2
Estimated # of Rows Returned: 1

1) informix.tab2: SEQUENTIAL SCAN

Filters: informix.tab2.col2 > 50

2) informix.tab1: SEQUENTIAL SCAN

Filters:
Table Scan Filters: informix.tab1.col1 > 50

DYNAMIC HASH JOIN

Dynamic Hash Filters: informix.tab2.col2 = informix.tab1.col1

Figure 10-15. Query plan that uses a derived table folded into the parent query

select * from (select col1 from tab1 where col1 = 100) as vtab1(c1)
left join (select col1 from tab2 where col1 = 10) as vtab2(vc1)
on vtab1.c1 = vtab2.vc1

Estimated Cost: 5
Estimated # of Rows Returned: 1

1) informix.tab1: SEQUENTIAL SCAN

Filters: informix.tab1.col1 = 100

2) informix.tab2: AUTOINDEX PATH

(1) Index Keys: col1 (Key-Only)

Lower Index Filter: informix.tab1.col1 = informix.tab2.col1
Index Key Filters: (informix.tab2.col1 = 10 )

ON-Filters:(informix.tab1.col1 = informix.tab2.col1
AND informix.tab2.col1 = 10 )
NESTED LOOP JOIN(LEFT OUTER JOIN)

Figure 10-16. Second query plan that uses a derived table folded into the parent query

The following example shows a complex query involving the UNION operation.
Here, a temporary table has been created.

select * from (select col1 from tab1 union select col2 from tab2 )
as vtab(vcol1) where vcol1 < 50

Estimated Cost: 4
Estimated # of Rows Returned: 1

1) (Temp Table For Collection Subquery): SEQUENTIAL SCAN

Figure 10-17. Complex derived-table query that creates a temporary table

10-18 IBM Informix Performance Guide

 XML query plans in IBM Data Studio
IBM Data Studio consists of a set of tools to use for administration, data modeling,
and building queries from data that comes from data servers. The EXPLAIN_SQL
routine prepares a query and returns a query plan in XML. The IBM Data Studio
Administration Edition can use the EXPLAIN_SQL routine to obtain a query plan
in XML format, interpret the XML, and render the plan visually.

If you plan to use IBM Data Studio to obtain Visual Explain output, you must
create and specify a default sbspace name for the SBSPACENAME configuration
parameter in your onconfig file. The EXPLAIN_SQL routine creates BLOBs in this
sbspace.

For information about using IBM Data Studio, see IBM Data Studio
documentation.

Factors that affect the query plan

When the optimizer determines the query plan, it assigns a cost to each possible
plan and then chooses the plan with the lowest cost. The optimizer analyzes
several factors to determine the cost of each query plan.

Some of the factors that the optimizer uses to determine the cost of each query
plan are:
v The number of I/O requests that are associated with each file system access
v The CPU work that is required to determine which rows meet the query
predicate
v The resources that are required to sort or group the data
v The amount of memory available for the query (specified by the
DS_TOTAL_MEMORY and DS_MAX_QUERIES parameters)

To calculate the cost of each possible query plan, the optimizer:

v Uses a set of statistics that describes the nature and physical characteristics of
the table data and indexes
v Examines the query filters
v Examines the indexes that can be used in the plan
v Uses the cost of moving data to perform joins locally or remotely for distributed
queries

For queries that access remote tables in cross-server operations, certain

characteristics can significantly degrade performance relative to the corresponding
DML operations on tables and views in the local database. Query specifications
that can potentially limit performance with remote tables include the following
specifications:
v ANSI LEFT OUTER JOIN syntax
v Derived tables based on remote tables
v TEMP tables as materialized views that reference remote tables.
Limitations on remote views
Reoptimization can occur with multiple executions of queries involving
remote views. The optimizer does not pick up the query plans from
statement cache even if the statement cache is enabled.

Chapter 10. Queries and the query optimizer 10-19

 Statistics held for the table and index
The accuracy with which the query optimizer can assess the execution cost of a
query plan depends on the information available to the optimizer. Use the
UPDATE STATISTICS statement to maintain simple statistics about a table and its
associated indexes. Updated statistics provide the query optimizer with
information that can minimize the amount of time required to perform queries on
that table.

The database server starts a statistical profile of a table when the table is created,
and the profile is refreshed when you issue the UPDATE STATISTICS statement.
The query optimizer does not recalculate the profile for tables automatically. In
some cases, gathering the statistics might take longer than executing the query.

To ensure that the optimizer selects a query plan that best reflects the current state
of your tables, run UPDATE STATISTICS at regular intervals. For guidelines, see
“Update statistics when they are not generated automatically” on page 13-11.

The optimizer uses the following system catalog information as it creates a query
plan:
v The number of rows in a table, as of the most recent UPDATE STATISTICS
statement
v Whether a column is constrained to be unique
v The distribution of column values, when requested with the MEDIUM or HIGH
keyword in the UPDATE STATISTICS statement
For more information about data distributions, see “Creating data distributions”
on page 13-13.
v The number of disk pages that contain row data

The optimizer also uses the following system catalog information about indexes:
v The indexes that exist on a table, including the columns that they index, whether
they are ascending or descending, and whether they are clustered
v The depth of the index structure (a measure of the amount of work that is
needed to perform an index lookup)
v The number of disk pages that index entries occupy
v The number of unique entries in an index, which can be used to estimate the
number of rows that an equality filter returns
v Second-largest and second-smallest key values in an indexed column

Only the second-largest and second-smallest key values are noted, because the
extreme values might have a special meaning that is not related to the rest of the
data in the column. The database server assumes that key values are distributed
evenly between the second largest and second smallest. Only the initial 4 bytes of
these keys are stored. If you create a distribution for a column associated with an
index, the optimizer uses that distribution when it estimates the number of rows
that match a query.

For more information about system catalog tables, see the IBM Informix Guide to
SQL: Reference.

Filters in the query

The query optimizer bases query-cost estimates on the number of rows to be
retrieved from each table. In turn, the estimated number of rows is based on the

10-20 IBM Informix Performance Guide

 selectivity of each conditional expression that is used within the WHERE clause. A
conditional expression that is used to select rows is termed a filter.

The selectivity is a value between 0 and 1 that indicates the proportion of rows
within the table that the filter can pass. A selective filter, one that passes few rows,
has a selectivity near 0, and a filter that passes almost all rows has a selectivity
near 1. For guidelines on filters, see “Improve filter selectivity” on page 13-2.

The optimizer can use data distributions to calculate selectivity for the filters in a
query. However, in the absence of data distributions, the database server calculates
selectivity for filters of different types based on table indexes. The following table
lists some of the selectivity values that the optimizer assigns to filters of different
types. Selectivity that is calculated using data distributions is even more accurate
than the selectivity that this table shows.

In the table:
v indexed-col is the first or only column in an index.
v 2nd-max, 2nd-min are the second-largest and second-smallest key values in
indexed column.
v The plus sign ( + ) means logical union ( = the Boolean OR operator) and the
multiplication symbol ( x ) means logical intersection ( = the Boolean AND
operator).
Table 10-1. Selectivity values that the optimizer assigns to filters of different types
Filter Expression Selectivity (F)
indexed-col = literal-valueindexed-col = F = 1/(number of distinct keys in index)
host-variableindexed-col IS NULL
tab1.indexed-col = tab2.indexed-col F = 1/(number of distinct keys in the larger index)
indexed-col > literal-value F = (2nd-max - literal-value)/(2nd-max - 2nd-min)
indexed-col < literal-value F = (literal-value - 2nd-min)/(2nd-max - 2nd-min)
any-col IS NULLany-col = any-expression F = 1/10
any-col > any-expressionany-col < any-expression F = 1/3
any-col MATCHES any-expressionany-col LIKE F = 1/5
any-expression
EXISTS subquery F = 1 if subquery estimated to return >0 rows, else 0
NOT expression F = 1 - F(expression)
expr1 AND expr2 F = F(expr1) x F(expr2)
expr1 OR expr2 F = F(expr1) + F(expr2) - (F(expr1) x F(expr2))
any-col IN list Treated as any-col = item1 OR . . . OR any-col = itemn.
any-col relop ANY subquery Treated as any-col relop value1 OR . . . OR any-col relop
valuen for estimated size of subquery n.

Here relop is any relational operator, such as <, >, >=, <=.

Indexes for evaluating a filter

The query optimizer notes whether an index can be used to evaluate a filter. For
this purpose, an indexed column is any single column with an index or the first
column named in a composite index.

Chapter 10. Queries and the query optimizer 10-21

 If the values contained in the index are all that is required, the database server
does not read the rows. It is faster to omit the page lookups for data pages
whenever the database server can read values directly from the index.

The optimizer can choose an index for any one of the following cases:
v When the column is indexed and a value to be compared is a literal, a host
variable, or an uncorrelated subquery
The database server can locate relevant rows in the table by first finding the row
in an appropriate index. If an appropriate index is not available, the database
server must scan each table in its entirety.
v When the column is indexed and the value to be compared is a column in
another table (a join expression)
The database server can use the index to find matching values. The following
join expression shows such an example:
WHERE customer.customer_num = orders.customer_num
If rows of customer are read first, values of customer_num can be applied to an
index on orders.customer_num.
v When processing an ORDER BY clause
If all the columns in the clause appear in the required sequence within a single
index, the database server can use the index to read the rows in their ordered
sequence, thus avoiding a sort.
v When processing a GROUP BY clause
If all the columns in the clause appear in one index, the database server can read
groups with equal keys from the index without requiring additional processing
after the rows are retrieved from their tables.

Effect of PDQ on the query plan

When the parallel database query (PDQ) feature is turned on, the optimizer can
choose to execute a query in parallel. This can improve performance dramatically
when the database server processes queries that decision-support applications
initiate.

For more information, see Chapter 12, “Parallel database query (PDQ),” on page
12-1.

Effect of OPTCOMPIND on the query plan

The OPTCOMPIND setting influences the access plan that the optimizer chooses
for single and multiple-table queries. You can change the value of OPTCOMPIND
within a session for different kinds of queries.

To change the value of OPTCOMPIND within a session, use the SET

ENVIRONMENT OPTCOMPIND command, not the OPTCOMPIND configuration
parameter. For more information about using this command, see “Setting the value
of OPTCOMPIND within a session” on page 3-11.

Single-table query
For single-table scans, when OPTCOMPIND is set to 0 or 1 and the current
transaction isolation level is Repeatable Read, the optimizer considers two types of
access plans.

If:
v An index is available, the optimizer uses it to access the table.

10-22 IBM Informix Performance Guide

 v No index is available, the optimizer considers scanning the table in physical
order.

When OPTCOMPIND is not set in the database server configuration, its value
defaults to 2. When OPTCOMPIND is set to 2 or 1 and the current isolation level
is not Repeatable Read, the optimizer chooses the least expensive plan to access the
table.

Multitable query
For join plans, the OPTCOMPIND setting influences the access plan for a specific
ordered pair of tables.

Set OPTCOMPIND to 0 if you want the database server to select a join method
exactly as it did in previous versions of the database server. This option ensures
compatibility with previous versions.

If OPTCOMPIND is set to 0 or set to 1 and the current transaction isolation level is

Repeatable Read, the optimizer gives preference to the nested-loop join.

Important: When OPTCOMPIND is set to 0, the optimizer does not choose a hash
join.

If OPTCOMPIND is set to 2 or set to 1 and the transaction isolation level is not

Repeatable Read, the optimizer chooses the least expensive query plan from among
those previously listed and gives no preference to the nested-loop join.

Effect of available memory on the query plan

Informix constrains the amount of memory that a parallel query can use based on
the values of the DS_TOTAL_MEMORY and DS_MAX_QUERIES configuration
parameters. If the amount of memory available for the query is too low to execute
a hash join, the database server uses a nested-loop join instead.

For more information about parallel queries and the DS_TOTAL_MEMORY and
DS_MAX_QUERIES parameters, see Chapter 12, “Parallel database query (PDQ),”
on page 12-1.

Time costs of a query

You can adjust a few, but not all, of the response-time effects of actions that the
database server performs when processing a query.

The following costs can be reduced by optimal query construction and appropriate
indexes:
v Sort time
v Data mismatches
v In-place ALTER TABLE
v Index lookups

For information about how to optimize specific queries, see Chapter 13,
“Improving individual query performance,” on page 13-1.

Chapter 10. Queries and the query optimizer 10-23

 Memory-activity costs
The database server can process only data in memory. It must read rows into
memory to evaluate those rows against the filters of a query. After the server finds
rows that satisfy those filters, it prepares an output row in memory by assembling
the selected columns.

Most of these activities are performed quickly. Depending on the computer and its
workload, the database server can perform hundreds or even thousands of
comparisons each second. As a result, the time spent on in-memory work is
usually a small part of the execution time.

Although some in-memory activities, such as sorting, take a significant amount of

time, it takes much longer to read a row from disk than to examine a row that is
already in memory.

Sort-time costs
A sort requires in-memory work as well as disk work. The in-memory work
depends on the number of columns that are sorted, the width of the combined sort
key, and the number of row combinations that pass the query filter. You can reduce
the cost of sorting.

You can use the following formula to calculate the in-memory work that a sort
operation requires:
Wm = (c * Nfr) + (w * Nfrlog2(Nfr))
Wm is the in-memory work.
c is the number of columns to order and represents the costs to extract
column values from the row and concatenate them into a sort key.
w is proportional to the width of the combined sort key in bytes and stands
for the work to copy or compare one sort key. A numeric value for w
depends strongly on the computer hardware in use.
Nfr is the number of rows that pass the query filter.

Sorting can involve writing information temporarily to disk if the amount of data
to sort is large. You can direct the disk writes to occur in the operating-system file
space or in a dbspace that the database server manages. For details, see “Configure
dbspaces for temporary tables and sort files” on page 5-8.

The disk work depends on the number of disk pages where rows appear, the
number of rows that meet the conditions of the query predicate, the number of
rows that can be placed on a sorted page, and the number of merge operations
that must be performed. Use the following formula to calculate the disk work that
a sort operation requires:
Wd = p + (Nfr/Nrp) * 2 * (m - 1))
Wd is the disk work.
p is the number of disk pages.
Nfr is the number of rows that pass the filters.
Nrp is the number of rows that can be placed on a page.
m represents the number of levels of merge that the sort must use.

10-24 IBM Informix Performance Guide

 The factor m depends on the number of sort keys that can be held in memory. If
there are no filters, Nfr/Nrp is equivalent to p.

When all the keys can be held in memory, m=1 and the disk work is equivalent to
p. In other words, the rows are read and sorted in memory.

For moderate to large tables, rows are sorted in batches that fit in memory, and
then the batches are merged. When m=2, the rows are read, sorted, and written in
batches. Then the batches are read again and merged, resulting in disk work
proportional to the following value:
Wd = p + (2 * (Nfr/Nrp))

The more specific the filters, the fewer the rows that are sorted. As the number of
rows increases, and the amount of memory decreases, the amount of disk work
increases.

To reduce the cost of sorting, use the following methods:

v Make your filters as specific (selective) as possible.
v Limit the projection list to the columns that are relevant to your problem.

Row-reading costs
When the database server needs to examine a row that is not already in memory, it
must read that row from disk. The database server does not read only one row; it
reads the entire page that contains the row. If the row spans more than one page, it
reads all of the pages.

The actual cost of reading a page is variable and hard to predict. The actual cost is
a combination of the factors shown in the following table.

Factor Effect of Factor

Buffering If the needed page is in a page buffer
already, the cost to read is nearly zero.
Contention If two or more applications require access to
the disk hardware, I/O requests can be
delayed.
Seek time The slowest thing that a disk does is to seek;
that is, to move the access arm to the track
that holds the data. Seek time depends on
the speed of the disk and the location of the
disk arm when the operation starts. Seek
time varies from zero to nearly a second.
Latency The transfer cannot start until the beginning
of the page rotates under the access arm.
This latency, or rotational delay, depends on
the speed of the disk and on the position of
the disk when the operation starts. Latency
can vary from zero to a few milliseconds.

The time cost of reading a page can vary from microseconds for a page that is
already in a buffer, to a few milliseconds when contention is zero and the disk arm
is already in position, to hundreds of milliseconds when the page is in contention
and the disk arm is over a distant cylinder of the disk.

Chapter 10. Queries and the query optimizer 10-25

 Sequential access costs
Disk costs are lowest when the database server reads the rows of a table in
physical order.

When the first row on a page is requested, the disk page is read into a buffer page.
After the page is read in, it does not need not to be read again; requests for
subsequent rows on that page are filled from the buffer until all the rows on that
page are processed. When one page is exhausted, the page for the next set of rows
must be read in.

When you use unbuffered devices for dbspaces, and the table is organized
properly, the disk pages of consecutive rows are placed in consecutive locations on
the disk. This arrangement allows the access arm to move very little when it reads
sequentially. In addition, latency costs are usually lower when pages are read
sequentially.
Related reference:
Read-ahead operations (Administrator's Guide)

Nonsequential access costs

The disk-access time is much higher when a disk device reads table pages
nonsequentially than when it reads that same table sequentially.

Whenever a table is read in random order, additional disk accesses are required to
read the rows in the required order. Disk costs are higher when the rows of a table
are read in a sequence unrelated to physical order on disk. Because the pages are
not read sequentially from the disk, both seek and rotational delays occur before
each page can be read.

Nonsequential access often occurs when you use an index to locate rows. Although
index entries are sequential, there is no guarantee that rows with adjacent index
entries must reside on the same (or adjacent) data pages. In many cases, a separate
disk access must be made to fetch the page for each row located through an index.
If a table is larger than the page buffers, a page that contained a row previously
read might be cleaned (removed from the buffer and written back to the disk)
before a subsequent request for another row on that page can be processed. That
page might have to be read in again.

Depending on the relative ordering of the table with respect to the index, you can
sometimes retrieve pages that contain several needed rows. The degree to which
the physical ordering of rows on disk corresponds to the order of entries in the
index is called clustering. A highly clustered table is one in which the physical
ordering on disk corresponds closely to the index.

Index lookup costs

The database server incurs additional costs when it finds a row through an index.
The index is stored on disk, and its pages must be read into memory with the data
pages that contain the desired rows.

An index lookup works down from the root page to a leaf page. The root page,
because it is used so often, is almost always found in a page buffer. The odds of
finding a leaf page in a buffer depend on the size of the index, the form of the
query, and the frequency of column-value duplication. If each value occurs only

10-26 IBM Informix Performance Guide

 once in the index and the query is a join, each row to be joined requires a
nonsequential lookup into the index, followed by a nonsequential access to the
associated row in the table.

Reading duplicate values from an index

Reading an index with duplicate entries incurs an additional cost over reading the
table sequentially.

Each entry or set of entries with the same value must be located in the index.
Then, for each entry in the index, a random access must be made to the table to
read the associated row. However, if there are many duplicate rows per distinct
index value, and the associated table is highly clustered, the added cost of joining
through the index can be slight.

Searching for NCHAR or NVARCHAR columns in an index

A query using an index on an NCHAR or NVARCHAR scans the entire index,
resulting in additional time costs.
Global Language Support (GLS) Only
Indexes that are built on NCHAR or NVARCHAR columns are sorted
using a locale-specific comparison value. For example, the Spanish double-l
character (ll) might be treated as a single unique character instead of a pair
of ls.
In some locales, the comparison value is not based on the code-set order.
The index build uses the locale-specific comparison value to store the key
values in the index. As a result, a query using an index on an NCHAR or
NVARCHAR scans the entire index because the database server searches
the index in code-set order.

In-place ALTER TABLE costs

For certain conditions, the database server uses an in-place alter algorithm to
modify each row when you execute an ALTER TABLE statement. After the alter
table operation, the database server inserts rows using the latest definition. If your
query accesses rows that are not yet converted to the new table definition, you
might notice a slight degradation in the performance of your individual query,
because the database server reformats each row in memory before it is returned.

For more information about the conditions and performance advantages when an
in-place alter occurs, see “Altering a table definition” on page 6-35.

View costs
A complex view could run more slowly than expected.

You can create views of tables for a number of reasons:

v To limit the data that a user can access
v To reduce the time that it takes to write a complex query
v To hide the complexity of the query that a user needs to write

However, a query against a view might execute more slowly than expected when
the complexity of the view definition causes a temporary table to be created to
process the query. This temporary table is referred to as a materialized view. For
example, you can create a view with a union to combine results from several
SELECT statements.

Chapter 10. Queries and the query optimizer 10-27

 The following sample SQL statement creates a view that includes unions:
CREATE VIEW view1 (col1, col2, col3, col4)
AS
SELECT a, b, c, d
FROM tab1 WHERE
UNION
SELECT a2, b2, c2, d2
FROM tab2 WHERE
...
UNION
SELECT an, bn, cn, dn
FROM tabn WHERE
;

When you create a view that contains complex SELECT statements, the end user
does not need to handle the complexity. The end user can just write a simple
query, as the following example shows:
SELECT a, b, c, d
FROM view1
WHERE a < 10;

However, this query against view1 might execute more slowly than expected
because the database server creates a fragmented temporary table for the view
before it executes the query.

Another situation when the query might execute more slowly than expected is if
you use a view in an ANSI join. The complexity of the view definition might cause
a temporary table to be created.

To determine if you have a query that must build a temporary table to process the
view, execute the SET EXPLAIN statement. If you see Temp Table For View in the
SET EXPLAIN output file, your query requires a temporary table to process the
view.

Small-table costs
A table is small if it occupies so few pages that it can be retained entirely in the
page buffers. Operations on small tables are generally faster than operations on
large tables.

As an example, in the stores_demo database, the state table that relates

abbreviations to names of states has a total size of fewer than 1000 bytes; it fits in
no more than two pages. This table can be included in any query at little cost. No
matter how this table is used, it costs no more than two disk accesses to retrieve
this table from disk the first time that it is required.

Data-mismatch costs
An SQL statement can encounter additional costs when the data type of a column
that is used in a condition differs from the definition of the column in the CREATE
TABLE statement.

For example, the following query contains a condition that compares a column to a
data type value that differs from the table definition:
CREATE TABLE table1 (a integer, );
SELECT * FROM table1
WHERE a = ’123’;

10-28 IBM Informix Performance Guide

 The database server rewrites this query before execution to convert 123 to an
integer. The SET EXPLAIN output shows the query in its adjusted format. This
data conversion has no noticeable overhead.

The additional costs of a data mismatch are most severe when the query compares
a character column with a noncharacter value and the length of the number is not
equal to the length of the character column. For example, the following query
contains a condition in the WHERE clause that equates a character column to an
integer value because of missing quotation marks:
CREATE TABLE table2 (char_col char(3), );
SELECT * FROM table2
WHERE char_col = 1;

This query finds all of the following values that are valid for char_col:
’ 1’
’001’
’1’

These values are not necessarily clustered together in the index keys. Therefore, the
index does not provide a fast and correct way to obtain the data. The SET
EXPLAIN output shows a sequential scan for this situation.

Warning: The database server does not use an index when the SQL statement
compares a character column with a noncharacter value that is not equal in length
to the character column.

Encrypted-value costs
An encrypted value uses more storage space than the corresponding plain-text
value because all of the information needed to decrypt the value except the
encryption key is stored with the value.

Most encrypted data requires approximately 33 percent more storage space than
unencrypted data. Omitting the hint used with the password can reduce
encryption overhead by up to 50 bytes. If you are using encrypted values, you
must make sure that you have sufficient space available for the values.

GLS functionality costs

Sorting or indexing certain data sets can degrade performance.

For information about the performance degradation that occurs from indexing
some data sets, see “Searching for NCHAR or NVARCHAR columns in an index”
on page 10-27.

If you do not need a non-ASCII collation sequence, use the CHAR and VARCHAR
data types for character columns whenever possible. Because CHAR and
VARCHAR data require simple value-based comparison, sorting and indexing
these columns is less expensive than for non-ASCII data types (NCHAR or
NVARCHAR, for example).

For more information about other character data types, see the IBM Informix GLS
User's Guide.

Network-access costs
Moving data over a network imposes delays in addition to those you encounter
with direct disk access.

Chapter 10. Queries and the query optimizer 10-29

 Network delays can occur when the application sends a query or update request
across the network to a database server on another computer. Although the
database server performs the query on the remote host computer, that database
server returns the output to the application over the network.

Data sent over a network consists of command messages and buffer-sized blocks of
row data. Although the details can differ depending on the network and the
computers, the database server network activity follows a simple model in which
one computer, the client, sends a request to another computer, the server. The server
replies with a block of data from a table.

Whenever data is exchanged over a network, delays are inevitable in the following
situations:
v When the network is busy, the client must wait its turn to transmit. Such delays
are usually less than a millisecond. However, on a heavily loaded network, these
delays can increase exponentially to tenths of seconds and more.
v When the server is handling requests from more than one client, requests might
be queued for a time that can range from milliseconds to seconds.
v When the server acts on the request, it incurs the time costs of disk access and
in-memory operations that the preceding sections describe.

Transmission of the response is also subject to network delays.

Network access time is extremely variable. In the best case, when neither the
network nor the server is busy, transmission and queuing delays are insignificant,
and the server sends a row almost as quickly as a local database server might.
Furthermore, when the client asks for a second row, the page is likely to be in the
page buffers of the server.

Unfortunately, as network load increases, all these factors tend to worsen at the
same time. Transmission delays rise in both directions, which increases the queue
at the server. The delay between requests decreases the likelihood of a page
remaining in the page buffer of the responder. Thus, network-access costs can
change suddenly and quite dramatically.

If you use the SELECT FIRST n clause in a distributed query, you will still see only
the requested amount of data. However, the local database server does not send
the SELECT FIRST n clause to the remote site. Therefore, the remote site might
return more data.

The optimizer that the database server uses assumes that access to a row over the
network takes longer than access to a row in a local database. This estimate
includes the cost of retrieving the row from disk and transmitting it across the
network.

For information about actions that might improve performance across the network,
see the following sections:
v “Optimizer estimates of distributed queries” on page 13-29
v “MaxConnect for multiple connections UNIX” on page 3-26
v “Multiplexed connections and CPU utilization” on page 3-25
v “Network buffer pools” on page 3-17

10-30 IBM Informix Performance Guide

Optimization when SQL is within an SPL routine
If an SPL routine contains SQL statements, the database server optimizes and
executes the SQL within the SPL routine.

The topics in this section contain information about how and when the database
server optimizes and executes these routines.

SQL optimization
If an SPL routine contains SQL statements, at some point the query optimizer
evaluates the possible query plans for SQL in the SPL routine and selects the query
plan with the lowest cost. The database server puts the selected query plan for
each SQL statement in an execution plan for the SPL routine.

When you create an SPL routine with the CREATE PROCEDURE statement, the
database server attempts to optimize the SQL statements within the SPL routine at
that time. If the tables cannot be examined at compile time (because they do not
exist or are not available), the creation does not fail. In this case, the database
server optimizes the SQL statements the first time that the SPL routine executes.

The database server stores the optimized execution plan in the sysprocplan system
catalog table for use by other processes. In addition, the database server stores
information about the SPL routine (such as procedure name and owner) in the
sysprocedures system catalog table and an ASCII version of the SPL routine in the
sysprocbody system catalog table.

Figure 10-18 summarizes the information that the database server stores in system
catalog tables during the compilation process.

Procedure characteristics sysprocedures

procname procid
get_sum 1
SPL code get_avg 2

Query tree, dependency list

sysprocplan
Pcode procid data
1 Ny4yMkMx...
sysprocbody 2 Ny4yMkMx...
procid data
1 create dba get_sum(order_a...
2 create dba get_avg(order_a...

Figure 10-18. SPL information stored in system catalog tables

Displaying the execution plan

When you execute an SPL routine, it is already optimized. You can display the
query plan for each SQL statement contained in the SPL routine

To display the query plan, execute the SET EXPLAIN ON statement prior to one of
the following SQL statements that always tries to optimize the SPL routine:
v CREATE PROCEDURE
v UPDATE STATISTICS FOR PROCEDURE
For example, use the following statements to display the query plan for an SPL
routine:

Chapter 10. Queries and the query optimizer 10-31

 SET EXPLAIN ON;
UPDATE STATISTICS FOR PROCEDURE procname;

Automatic reoptimization
In some situations, the database server reoptimizes an SQL statement the next time
an SPL routine.

If the AUTO_REPREPARE configuration parameter or the IFX_AUTO_REPREPARE

session environment variable is disabled, the following error can result when
prepared objects or SPL routines are executed after the schema of a table
referenced by the prepared object or indirectly referenced by the SPL routine has
been modified:
-710 Table <table-name> has been dropped, altered, or renamed.

The database server uses a dependency list to keep track of changes that would
cause reoptimization the next time that an SPL routine executes.

The database server reoptimizes an SQL statement the next time an SPL routine
executes after one of the following situations:
v Execution of any data definition language (DDL) statement (such as ALTER
TABLE, DROP INDEX, and CREATE INDEX) that might alter the query plan
v Alteration of a table that is linked to another table with a referential constraint
(in either direction)
v Execution of UPDATE STATISTICS FOR TABLE for any table involved in the
query
The UPDATE STATISTICS FOR TABLE statement changes the version number of
the specified table in systables.
v Renaming a column, database, or index with the RENAME statement

Whenever the SPL routine is reoptimized, the database server updates the
sysprocplan system catalog table with the reoptimized execution plan.

Reoptimizing SPL routines

You can run an SQL statement that reoptimizes an SPL routine to prevent
automatic reoptimization.

If you do not want to incur the cost of automatic reoptimization when you first
execute an SPL routine after one of the situations that “Automatic reoptimization”
lists, execute the UPDATE STATISTICS statement with the FOR PROCEDURE
clause immediately after the situation occurs. In this way, the SPL routine is
reoptimized before any users execute it.

To prevent unnecessary reoptimization of all SPL routines, ensure that you specify
a specific procedure name in the FOR PROCEDURE clause.
UPDATE STATISTICS for procedure myroutine;

For guidelines to run UPDATE STATISTICS, see “Update statistics when they are
not generated automatically” on page 13-11.

Optimization levels for SQL in SPL routines

The current optimization level set in an SPL routine affects how the SPL routine is
optimized.

The algorithm that a SET OPTIMIZATION HIGH statement invokes is a

sophisticated, cost-based strategy that examines all reasonable query plans and

10-32 IBM Informix Performance Guide

 selects the best overall alternative. For large joins, this algorithm can incur more
overhead than desired. In extreme cases, you can run out of memory.

The alternative algorithm that a SET OPTIMIZATION LOW statement invokes

eliminates unlikely join strategies during the early stages, which reduces the time
and resources spent during optimization. However, when you specify a low level
of optimization, the optimal strategy might not be selected because it was
eliminated from consideration during early stages of the algorithm.

For SPL routines that remain unchanged or change only slightly and that contain
complex SELECT statements, you might want to set the SET OPTIMIZATION
statement to HIGH when you create the SPL routine. This optimization level stores
the best query plans for the SPL routine. Then set optimization to LOW before you
execute the SPL routine. The SPL routine then uses the optimal query plans and
runs at the more cost-effective rate if reoptimization occurs.

Execution of an SPL routine

When the database server executes an SPL routine with the EXECUTE
PROCEDURE statement, with the CALL statement, or within an SQL statement,
the server performs several activities.

The database server performs these activities:

v It reads the interpreter code from the system catalog tables and converts it from
a compressed format to an executable format. If the SPL routine is in the UDR
cache, the database server retrieves it from the cache and bypasses the
conversion step.
v It executes any SPL statements that it encounters.
v When the database server encounters an SQL statement, it retrieves the query
plan from the database and executes the statement. If the query plan has not
been created, the database server optimizes the SQL statement before it executes.
v When the database server reaches the end of the SPL routine or when it
encounters a RETURN statement, it returns any results to the client application.
Unless the RETURN statement has a WITH RESUME clause, the SPL routine
execution is complete.

SPL routine executable format stored in UDR cache

The first time that a user executes an SPL routine, the database server stores the
executable format and any query plans in the UDR cache in the virtual portion of
shared memory.

When another user executes an SPL routine, the database server first checks the
UDR cache. SPL execution performance improves when the database server can
execute the SPL routine from the UDR cache. The UDR cache also stores UDRs,
user-defined aggregates, and extended data types definitions.
Related reference:
“Configure and monitor memory caches” on page 4-21

Adjust the UDR cache

The default number of SPL routines, UDRs, and other user-defined definitions in
the UDR cache is 127. You can change the number of entries with the
PC_POOLSIZE configuration parameter.

The database server uses a hashing algorithm to store and locate SPL routines in
the UDR cache. You can modify the number of buckets in the UDR cache with the
Chapter 10. Queries and the query optimizer 10-33
 PC_HASHSIZE configuration parameter. For example, if the value of the
PC_POOLSIZE configuration parameter is 100 and the value of the PC_HASHSIZE
configuration parameter is 10, each bucket can have up to 10 SPL routines and
UDRs.

Too many buckets cause the database server to move out cached SPL routines
when the bucket fills. Too few buckets increase the number of SPL routines in a
bucket, and the database server must search though the SPL routines in a bucket to
determine if the SPL routine that it needs is there.

When the number of entries in a bucket reaches 75 percent, the database server
removes the least recently used SPL routines from the bucket (and hence from the
UDR cache) until the number of SPL routines in the bucket is 50 percent of the
maximum SPL routines in the bucket.

Monitor the UDR cache by running the onstat -g prc command.

Important: PC_POOLSIZE and PC_HASHSIZE also control other memory caches

for the database server (excluding the buffer pool, the SQL statement cache, the
data distribution cache, and the data-dictionary cache). When you modify the size
and number of hash buckets for SPL routines, you also modify the size and
number of hash buckets for the other caches (such as the aggregate cache, oplcass,
and typename cache).
Related reference:
onstat -g prc command: Print sessions using UDR or SPL routines
(Administrator's Reference)
PC_POOLSIZE configuration parameter (Administrator's Reference)
PC_HASHSIZE configuration parameter (Administrator's Reference)

Trigger execution
A trigger is a database object that automatically executes one or more SQL
statements (the triggered action) when a specified data manipulation language
operation (the triggering event) occurs. You can define one or more triggers on a
table to execute after a SELECT, INSERT, UPDATE or DELETE triggering event.

You can also define INSTEAD OF triggers on a view. These triggers specify the
SQL statements to be executed as triggered actions on the underlying table when a
triggering INSERT, UPDATE or DELETE statement attempts to modify the view.
These triggers are called INSTEAD OF triggers because only the triggered SQL
action is executed; the triggering event is not executed. For more information about
using triggers, see the IBM Informix Guide to SQL: Tutorial and information about
the CREATE TRIGGER statement in the IBM Informix Guide to SQL: Syntax.

10-34 IBM Informix Performance Guide

 systriggers
Trigger characteristics
trigname trigid
up_qty 1
get_avg 2
SPL code
Trigger executable systrigbody
trigid data
1 Ny4yMkMx...
2 Ny4yMkMx...

Figure 10-19. Trigger information stored in system catalog tables

When you use the CREATE TRIGGER statement to register a new trigger, the
database server:
v Stores information about the trigger in the systriggers system catalog table.
v Stores the text of the statements that the trigger executes in the systrigbody
system catalog table.

The sysprocedures system catalog table identifies trigger routines that can be
invoked only as triggered actions.

Memory-resident tables of the sysmaster database indicate whether the table or

view has triggers on it.

Whenever a SELECT, INSERT, UPDATE, or DELETE statement is issued, the

database server checks to see if the statement is a triggering event that activates a
trigger for the table and columns (or for the view) on which the DML statement
operates. If the statement requires activating triggers, the database server retrieves
the statement text of the triggered actions from the systrigbody table and runs the
triggered DML statements or SPL routine before, during, or after the triggering
events. For INSTEAD OF triggers on a view, the database server performs the
triggered actions instead of the triggering events.

Performance implications for triggers

In many situations, triggers can improve performance slightly because of the
reduction in the number of messages passed from the client to the database server.

For example, if the trigger fires five SQL statements, the client saves at least 10
messages passed between the client and database server (one to send the SQL
statement and one for the reply after the database server executes the SQL
statement). Triggers improve performance the most when they execute more SQL
statements and the network speed is comparatively slow.

When the database server executes an SQL statement, it must perform the
following actions:
v Determine if triggers must be fired
v Retrieve the triggers from systriggers and systrigbody

These operations cause only a slight performance impact that can be offset by the
decreased number of messages passed between the client and the server.

However, triggers executed on SELECT statements have additional performance

implications. The following sections explain these implications.

Chapter 10. Queries and the query optimizer 10-35

 SELECT triggers on tables in a table hierarchy
When the database server executes a SELECT statement that includes a table that is
involved in a table hierarchy, and the SELECT statement fires a SELECT trigger,
performance might be slower if the SELECT statement that invokes the trigger
involves a join, sort, or materialized view.

In this case, the database server does not know which columns are affected in the
table hierarchy, so it can execute the query differently. The following behaviors
might occur:
v Key-only index scans are disabled on the table that is involved in a table
hierarchy.
v If the database server needs to sort data selected from a table involved in a table
hierarchy, it copies all of the columns in the SELECT list to the temporary table,
not just the sort columns.
v If the database server uses the table included in the table hierarchy to build a
hash table for a hash join with another table, it bypasses the early projection,
meaning it uses all of the columns from the table to build the hash table, not just
the columns in the join.
v If the SELECT statement contains a materialized view (meaning a temporary
table must be built for the columns in a view) that contains columns from a
table involved in a table hierarchy, all columns from the table are included in the
temporary table, not just the columns actually contained in the view.

SELECT triggers and row buffering

The lack of buffering for SELECT statements that fire SELECT triggers might
reduce performance slightly compared to an identical SELECT statement that does
not fire a SELECT trigger.

In SELECT statements whose tables do not fire SELECT triggers, the database
server sends more than one row back to the client and stores the rows in a buffer
even though the client application requested only one row with a FETCH
statement. However, for SELECT statements that contain one or more tables that
fire a SELECT trigger, the database server sends only the requested row back to the
client instead of a buffer full. The database server cannot return other rows to the
client until the trigger action occurs.

10-36 IBM Informix Performance Guide

Chapter 11. Optimizer directives
Optimizer directives are comments that tell the query optimizer how to execute a
query. You can use optimizer directives to improve query performance.

What optimizer directives are

Optimizer directives are specifications formatted as comments that provide
information to the query optimizer about how to execute a query.

You can use two kinds of optimizer directives:

v Optimizer directives in the form of instructions that are embedded in queries
(For more information, see “Optimizer directives that are embedded in queries.”
v External optimizer directives that you create and save for use as temporary
workaround solutions to problems when you do not want to change SQL
statements in queries. (For more information, see “External optimizer
directives.”)

Optimizer directives that are embedded in queries

Optimizer directives embedded in queries are comments in a SELECT statement
that provide information to the query optimizer on how to execute a query. You
can also place directives in UPDATE and DELETE statements, instructing the
optimizer how to access the data.

Optimizer directives can either be explicit directions (for example, "use this index"
or "access this table first"), or they can eliminate possible query plans (for example,
"do not read this table sequentially" or "do not perform a nested-loop join").

External optimizer directives

External optimizer directives are optimizer directives that an administrator can
create and store in the sysdirectives catalog table. The administrator can then use
an ONCONFIG variable to make the directives available.

Client users also specify an environment variable and can choose to use these
optimizer directives in queries in situations when they do not want to insert
comments in SQL statements.

External directives are useful when it is not feasible to rewrite a query for a
short-term solution to a problem, for example, when a query starts to perform
poorly. Rewriting the query by changing the SQL statement is preferable for
long-term solutions to problems.

External directives are for occasional use only. The number of directives stored in
the sysdirectives catalog should not exceed 50. A typical enterprise only needs 0 to
9 directives.

© Copyright IBM Corp. 1996, 2014 11-1

Reasons to use optimizer directives
In most cases, the optimizer chooses the fastest query plan. You can use optimizer
directives when the optimizer does not choose the best query plan to perform a
query, because of the complexity of the query, or because the query does not have
enough information about the nature of the data. A poor query plan produces poor
performance.

Before you decide when to use optimizer directives, you should understand what
makes a good query plan.

The optimizer creates a query plan based on costs of using different table-access
paths, join orders, and join plans.

Some query plan guidelines are:

v Do not use an index when the database server must read a large portion of the
table. For example, the following query might read most of the customer table:
SELECT * FROM customer WHERE STATE <> "ALASKA";
Assuming the customers are evenly spread among all 50 states, you might
estimate that the database server must read 98 percent of the table. It is more
efficient to read the table sequentially than to traverse an index (and
subsequently the data pages) when the database server must read most of the
rows.
v When you choose between indexes to access a table, use an index that can rule
out the most rows. For example, consider the following query:
SELECT * FROM customer
WHERE state = "NEW YORK" AND order_date = "01/20/97"
Assuming that 200,000 customers live in New York and only 1000 customers
ordered on any one day, the optimizer most likely chooses an index on
order_date rather than an index on state to perform the query.
v Place small tables or tables with restrictive filters early in the query plan. For
example, consider the following query:
SELECT * FROM customer, orders
WHERE customer.customer_num = orders.customer_num
AND
customer.state = "NEVADA";
In this example, if you read the customer table first, you can rule out most of
the rows by applying the filter that chooses all rows in which state = "NEVADA".
By ruling out rows in the customer table, the database server does not read as
many rows in the orders table (which might be significantly larger than the
customer table).
v Choose a hash join when neither column in the join filter has an index.
In the previous example, if customer.customer_num and orders.customer_num
are not indexed, a hash join is probably the best join plan.
v Choose nested-loop joins if:
– The number of rows retrieved from the outer table after the database server
applies any table filters is small, and the inner table has an index that can be
used to perform the join.
– The index on the outermost table can be used to return rows in the order of
the ORDER BY clause, eliminating the need for a sort.

For information about query plans, see “The query plan” on page 10-1. For more
information about directives, see

11-2 IBM Informix Performance Guide

 v “Preparation for using directives”
v “Guidelines for using directives”
v “Types of optimizer directives that are supported in SQL statements”

Preparation for using directives

In most cases, the optimizer chooses the fastest query plan. However, you can take
steps to assist the optimizer and to prepare for using directives.

To prepare for using directives, make sure that you perform the following tasks:
v Run UPDATE STATISTICS.
Without accurate statistics, the optimizer cannot choose the appropriate query
plan. Run UPDATE STATISTICS any time that the data in the tables changes
significantly (many new rows are added, updated, or deleted). For more
information, see “Update the statistics for the number of rows” on page 13-12.
v Create distributions.
One of the first things that you should try when you attempt to improve a slow
query is to create distributions on columns involved in a query. Distributions
provide the most accurate information to the optimizer about the nature of the
data in the table. Run UPDATE STATISTICS HIGH on columns involved in the
query filters to see if performance improves. For more information, see
“Creating data distributions” on page 13-13.

In some cases, the query optimizer does not choose the best query plan because of
the complexity of the query or because (even with distributions) it does not have
enough information about the nature of the data. In these cases, you can attempt to
improve performance for a particular query by using directives.

Guidelines for using directives

Guidelines for directives include frequently analyzing the effectiveness of the
query and using negative directives.

Consider the following guidelines:

v Examine the effectiveness of a particular directive frequently to make sure it
continues to operate effectively. Imagine a query in a production program with
several directives that force an optimal query plan. Some days later, users add,
update, or delete a large number of rows, which changes the nature of the data
so much that the once optimal query plan is no longer effective. This example
illustrates how you must use directives with care.
v Use negative directives (such as AVOID_NL, AVOID_FULL, and so on)
whenever possible. When you exclude a behavior that degrades performance,
you rely on the optimizer to use the next-best choice rather than attempt to force
a path that might not be optimal.

Types of optimizer directives that are supported in SQL statements

Directives that are in SQL statements are embedded in queries. These directives
include access-method directives, join-order directives, join-plan directives, and
optimization-goal directives.

Include the directives in the SQL statement as a comment that occurs immediately
after the SELECT, UPDATE, or DELETE keyword. The first character in a directive
is always a plus (+) sign. In the following query, the ORDERED directive specifies

Chapter 11. Optimizer directives 11-3

 that the tables should be joined in the same order as they are listed in the FROM
clause. The AVOID_FULL directive specifies that the optimizer should discard any
plans that include a full table scan on the listed table (employee).
SELECT --+ORDERED, AVOID_FULL(e) * FROM employee e, department d
> 50000;

For a complete syntax description for directives, see the IBM Informix Guide to SQL:
Syntax.

To influence the choice of a query plan that the optimizer makes, you can alter the
following aspects of a query:
v Access method
v Join order
v Join method
v Optimization goal
v Star-join directives

You can also use EXPLAIN directives instead of the SET EXPLAIN statement to
show the query plan. The following sections describe these aspects in detail.

Access-method directives
The database server uses an access method to access a table. The server can either
read the table sequentially via a full table scan or use any one of the indexes on
the table. Access-method directives influence the access method.

The following table lists the directives that influence the access method:

Access-Method Directive Description

INDEX Tells the optimizer to use the index specified
to access the table. If the directive lists more
than one index, the optimizer chooses the
index that yields the least cost.
AVOID_INDEX Tells the optimizer not use any of the indexes
listed. You can use this directive with the
AVOID_FULL directive.
INDEX_SJ Forces an index self-join path using the
specified index, or choosing the least costly
index in a list of indexes, even if data
distribution statistics are not available for the
leading index key columns of the index.

For information about index self-join paths,

see “Query plans that include an index
self-join path” on page 10-8.
AVOID_INDEX_SJ Tells the optimizer not to use an index
self-join path for the specified index or
indexes.
FULL Tells the optimizer to perform a full table
scan.
AVOID_FULL Tells the optimizer not to perform a full table
scan on the listed table. You can use this
directive with the AVOID_INDEX directive.

11-4 IBM Informix Performance Guide

 Access-Method Directive Description
INDEX_ALL or MULTI_INDEX Access the table by using the specified
indexes for a multi-index scan.

The INDEX_ALL and MULTI_INDEX

keywords are synonyms.
AVOID_MULTI_INDEX Tells the optimizer not to consider a
multi-index scan path for the specified table.

In some cases, forcing an access method can change the join method that the
optimizer chooses. For example, if you exclude the use of an index with the
AVOID_INDEX directive, the optimizer might choose a hash join instead of a
nested-loop join.

The optimizer considers an index self-join path only if all of the following
conditions are met:
v The index does not have functional keys, user-defined types, built-in opaque
types, or non-B-tree indexes
v Data distribution statistics are available for the index key column under
consideration
v The number of rows in the table is at least 10 times the number of unique
combinations of all possible lead-key column values.

If all of these conditions are met, the optimizer estimates the cost of an index
self-join path and compares it with the costs of alternative access methods. The
optimizer then picks the best access method for the table. For more information
about the access-method directives and some examples of their usage, see the IBM
Informix Guide to SQL: Syntax.

Join-order directives
The join-order directive ORDERED tells the optimizer to join tables in the order
that the SELECT statement lists them.

Effect of join order on join plan

By specifying the join order, you might affect more than just how tables are joined.

For example, consider the following query:

SELECT --+ORDERED, AVOID_FULL(e)
* FROM employee e, department d
WHERE e.dept_no = d.dept_no AND e.salary > 5000

In this example, the optimizer chooses to join the tables with a hash join. However,
if you arrange the order so that the second table is employee (and must be
accessed by an index), the hash join is not feasible.
SELECT --+ORDERED, AVOID_FULL(e)
* FROM department d, employee e
WHERE e.dept_no = d.dept_no AND e.salary > 5000;

The optimizer chooses a nested-loop join in this case.

Join order when you use views

The ORDERED directive that is inside a view or is in a query that contains a view
affect the join order.

Chapter 11. Optimizer directives 11-5

 Two cases can affect join order when you use views:
v The ORDERED directive is inside the view.
The ORDERED directive inside a view affects the join order of only the tables
inside the view. The tables in the view must be joined contiguously. Consider
the following view and query:
CREATE VIEW emp_job_view as
SELECT {+ORDERED}
emp.job_num, job.job_name
FROM emp, job
WHERE emp.job_num = job.job_num;

SELECT * from dept, emp_job_view, project

WHERE dept.dept_no = project.dept_num
AND emp_job_view.job_num = project.job_num;
The ORDERED directive specifies that the emp table come before the job table.
The directive does not affect the order of the dept and project table. Therefore,
all possible join orders are as follows:
– emp, job, dept, project
– emp, job, project, dept
– project, emp, job, dept
– dept, emp, job, project
– dept, project, emp, job
– project, dept, emp, job
v The ORDERED directive is in a query that contains a view.
If an ORDERED directive appears in a query that contains a view, the join order
of the tables in the query are the same as they are listed in the SELECT
statement. The tables within the view are joined as they are listed within the
view.
In the following query, the join order is dept, project, emp, job:
CREATE VIEW emp_job_view AS
SELECT
emp.job_num, job.job_name
FROM emp, job
WHERE emp.job_num = job.job_num;
SELECT {+ORDERED}
* FROM dept, project, emp_job_view
WHERE dept.dept_no = project.dept_num
AND emp_job_view.job_num = project.job_num;
An exception to this rule is when the view cannot be folded into the query, as in
the following example:
CREATE VIEW emp_job_view2 AS
SELECT DISTINCT
emp.job_num, job.job_name
FROM emp,job
WHERE emp.job_num = job.job_num;
In this example, the database server executes the query and puts the result in a
temporary table. The order of tables in this query is dept, project, temp_table.

Join-method directives
The join-method directives influence how the database server joins two tables in a
query.

The following directives influence the join method between two tables:
v USE_NL

11-6 IBM Informix Performance Guide

 Use the listed tables as the inner table in a nested-loop join.
v USE_HASH
Access the listed tables with a hash join. You can also choose whether the table
is used to create the hash table or to probe the hash table.
v AVOID_NL
Do not use the listed tables as the inner table in a nested-loop join. A table listed
with this directive can still participate in a nested-loop join as an outer table.
v AVOID_HASH
Do not access the listed tables with a hash join. Optionally, you can allow a hash
join but restrict the table from being the one that is probed or the table from
which the hash table is built.

You can specify the keyword /BUILD after the name of a table in a USE_HASH or
AVOID_HASH optimizer directives:
v With USE_HASH directives, the /BUILD keyword tells the optimizer to use the
specified table to build the hash table.
v With AVOID_HASH, the /BUILD keyword tells the optimizer to avoid using the
specified table to build the hash table.

You can specify the keyword /PROBE after the name of a table in a USE_HASH or
AVOID_HASH optimizer directives:
v With USE_HASH directives, the /PROBE keyword tells the optimizer to use the
specified table to probe the hash table.
v With AVOID_HASH directives, the /PROBE keyword tells the optimizer to
avoid using the specified table to probe the hash table.

Optimization-goal directives
In some queries, you might want to find only the first few rows in the result of a
query. Or, you might know that all rows must be accessed and returned from the
query. You can use the optimization-goal directives to find the first row that
satisfies the query or all rows that satisfy the query.

For example, you might want to find only the first few rows in the result of a
query, because an Informix ESQL/C program opens a cursor for the query and
performs a FETCH to find only the first row.

Use the optimization-goal directives to optimize the query for either one of these
cases:
v FIRST_ROWS
Choose a plan that optimizes the process of finding only the first row that
satisfies the query.
v ALL_ROWS
Choose a plan that optimizes the process of finding all rows (the default
behavior) that satisfy the query.

If you use the FIRST_ROWS directive, the optimizer might abandon a query plan
that contains activities that are time-consuming up front. For example, a hash join
might take too much time to create the hash table. If only a few rows must be
returned, the optimizer might choose a nested-loop join instead.

Chapter 11. Optimizer directives 11-7

 In the following example, assume that the database has an index on
employee.dept_no but not on department.dept_no. Without directives, the
optimizer chooses a hash join.
SELECT *
FROM employee, department
WHERE employee.dept_no = department.dept_no

However, with the FIRST_ROWS directive, the optimizer chooses a nested-loop

join because of the high initial overhead required to create the hash table.
SELECT {+first_rows} *
FROM employee, department
WHERE employee.dept_no = department.dept_no

Star-join directives
Star-join directives can specify how the query optimizer joins two or more tables,
among which one or more dimension tables have foreign-key dependencies on one
or more fact tables.

The following directives can influence the join plan for tables that logically
participate in a star schema or in a snowflake schema:
v FACT
The optimizer considers a query plan in which the specified table is a fact table
in a star-join execution plan.
v AVOID_FACT
The optimizer does not consider a star-join execution plan that treats the
specified table (or any of the tables in the list of tables) as a fact table.
v STAR_JOIN
The optimizer favors a star-join execution plan, if available.
v AVOID_STAR_JOIN
The optimizer chooses a query execution plan that is not a star-join plan.

These star-join directives have no effect unless the parallel database query feature
(PDQ) is enabled.
Related concepts:
Concepts of dimensional data modeling (Data Warehouse Guide)
Keys to join the fact table with the dimension tables (Data Warehouse Guide)
Use the snowflake schema for hierarchical dimension tables (Data Warehouse
Guide)
Related reference:
Star-Join Directives (SQL Syntax)

EXPLAIN directives
You can use the EXPLAIN directives to display the query plan that the optimizer
chooses, and you can specify to display the query plan without running the query.

You can use these directives:

v EXPLAIN
Displays the query plan that the optimizer chooses.
v EXPLAIN AVOID_EXECUTE

11-8 IBM Informix Performance Guide

 Displays the query plan that the optimizer chooses, but does not run the query.

When you want to display the query plan for one SQL statement only, use these
EXPLAIN directives instead of the SET EXPLAIN ON or SET EXPLAIN ON
AVOID_EXECUTE statements.

When you use AVOID_EXECUTE (either the directive or in the SET EXPLAIN
statement), the query does not execute but displays the following message:
No rows returned.

Figure 11-1 shows sample output for a query that uses the EXPLAIN
AVOID_EXECUTE directive.

QUERY:
------
select --+ explain avoid_execute
l.customer_num, l.lname, l.company,
l.phone, r.call_dtime, r.call_descr
from customer l, cust_calls r
where l.customer_num = r.customer_num

DIRECTIVES FOLLOWED:
EXPLAIN
AVOID_EXECUTE
DIRECTIVES NOT FOLLOWED:

Estimated Cost: 7
Estimated # of Rows Returned: 7

1) informix.r: SEQUENTIAL SCAN

2) informix.l: INDEX PATH

(1) Index Keys: customer_num (Serial, fragments: ALL)

Lower Index Filter: informix.l.customer_num = informix.r.customer_num
NESTED LOOP JOIN

Figure 11-1. Result of EXPLAIN AVOID_EXECUTE directives

The following table describes the pertinent output lines in Figure 11-1 that describe
the chosen query plan.

Output Line in Figure 11-1 Chosen Query Plan Description

DIRECTIVES FOLLOWED: EXPLAIN AVOID_EXECUTE Use the directives EXPLAIN and
AVOID_EXECUTE to display the query
plan and do not execute the query.
Estimated # of Rows Returned: 7 Estimate that this query returns seven
rows.
Estimated Cost: 7 This estimated cost of 7 is a value that the
optimizer uses to compare different query
plans and select the one with the lowest
cost.
1) informix.r: SEQUENTIAL SCAN Use the cust_calls r table as the outer table
and scan it to obtain each row.
2) informix.l: INDEX PATH For each row in the outer table, use an
index to obtain the matching row(s) in the
inner table customer l.

Chapter 11. Optimizer directives 11-9

 Output Line in Figure 11-1 on page 11-9 Chosen Query Plan Description
(1) Index Keys: customer_num (Serial, Use the index on the customer_num
fragments: ALL) column, scan it serially, and scan all
fragments (the customer l table consists of
only one fragment).
Lower Index Filter: Start the index scan at the customer_num
informix.l.customer_num = value from the outer table.
informix.r.customer_num

Example of directives that can alter a query plan

Directives can alter the query plan. You can use particular directives to force the
optimizer to choose a particular type of query plan, for example one that uses hash
joins and the order of tables as they appear in the query.

The following example shows how directives can alter the query plan.

Suppose you have the following query:

SELECT * FROM emp,job,dept
WHERE emp.location = 10
AND emp.jobno = job.jobno
AND emp.deptno = dept.deptno
AND dept.location = "DENVER";

Assume that the following indexes exist:

ix1: emp(empno,jobno,deptno,location)
ix2: job(jobno)
ix3: dept(location)

You run the query with SET EXPLAIN ON to display the query path that the
optimizer uses.
QUERY:
------
SELECT * FROM emp,job,dept
WHERE emp.location = "DENVER"
AND emp.jobno = job.jobno
AND emp.deptno = dept.deptno
AND dept.location = "DENVER"

Estimated Cost: 5
Estimated # of Rows Returned: 1

1) informix.emp: INDEX PATH

Filters: informix.emp.location = ’DENVER’

(1) Index Keys: empno jobno deptno location (Key-Only)

2) informix.dept: INDEX PATH

Filters: informix.dept.deptno = informix.emp.deptno

(1) Index Keys: location

Lower Index Filter: informix.dept.location = ’DENVER’
NESTED LOOP JOIN

3) informix.job: INDEX PATH

11-10 IBM Informix Performance Guide

 (1) Index Keys: jobno (Key-Only)
Lower Index Filter: informix.job.jobno = informix.emp.jobno
NESTED LOOP JOIN

The diagram in Figure 11-2 shows a possible query plan for this query. The query
plan has three levels of information: (1) a nested-loop join, (2) an index scan on
one table and a nested-loop join, and (3) index scans on two other tables.

Nested-loop join

Nested-loop join
Index scan
job
with ix2

Index scan Index scan

emp dept
with ix1 with ix3

Figure 11-2. Possible query plan without directives

Perhaps you are concerned that using a nested-loop join might not be the fastest
method to execute this query. You also think that the join order is not optimal. You
can force the optimizer to choose a hash join and order the tables in the query plan
according to their order in the query, so the optimizer uses the query plan that
Figure 11-3 shows. This query plan that has three levels of information: (1) a hash
join, (2) an index scan and a hash join, and (3) an index scan on two other tables.

Hash join (build on dept)

Hash join Index scan

(build on job) dept
with ix2

Index scan
emp job Full table scan
with ix1

Figure 11-3. Possible query plan with directives

To force the optimizer to choose the query plan that uses hash joins and the order
of tables shown in the query, use the directives that the following partial SET
EXPLAIN output shows:
QUERY:
------
SELECT {+ORDERED,
INDEX(emp ix1),
FULL(job),

Chapter 11. Optimizer directives 11-11

 USE_HASH(job /BUILD),
USE_HASH(dept /BUILD),
INDEX(dept ix3)}
* FROM emp,job,dept
WHERE emp.location = 1
AND emp.jobno = job.jobno
AND emp.deptno = dept.deptno
AND dept.location = "DENVER"

DIRECTIVES FOLLOWED:
ORDERED
INDEX ( emp ix1 )
FULL ( job )
USE_HASH ( job/BUILD )
USE_HASH ( dept/BUILD )
INDEX ( dept ix3 )
DIRECTIVES NOT FOLLOWED:

Estimated Cost: 7
Estimated # of Rows Returned: 1

1) informix.emp: INDEX PATH

Filters: informix.emp.location = ’DENVER’

(1) Index Keys: empno jobno deptno location (Key-Only)

2) informix.job: SEQUENTIAL SCAN

DYNAMIC HASH JOIN

Dynamic Hash Filters: informix.emp.jobno = informix.job.jobno

3) informix.dept: INDEX PATH

(1) Index Keys: location

Lower Index Filter: informix.dept.location = ’DENVER’

DYNAMIC HASH JOIN

Dynamic Hash Filters: informix.emp.deptno = informix.dept.deptno

Configuration parameters and environment variables for optimizer

directives
You can use the DIRECTIVES configuration parameter to turn on or off all
directives that the database server encounters, and you can use the
IFX_DIRECTIVES environment variable to override the setting of the
DIRECTIVES configuration parameter.

If the DIRECTIVES configuration parameter is set to 1 (the default), the optimizer

follows all directives. If the DIRECTIVES configuration parameter is set to 0, the
optimizer ignores all directives.

You can override the setting of DIRECTIVES. If the IFX_DIRECTIVES

environment variable is set to 1 or ON, the optimizer follows directives for any SQL
the client session executes. If IFX_DIRECTIVES is 0 or OFF, the optimizer ignores
directives for any SQL in the client session.

Any directives in an SQL statement take precedence over the join plan that the
OPTCOMPIND configuration parameter forces. For example, if a query includes
the USE_HASH directive and OPTCOMPIND is set to 0 (nested-loop joins
preferred over hash joins), the optimizer uses a hash join.

11-12 IBM Informix Performance Guide

Optimizer directives and SPL routines
Directives operate differently for a query in an SPL routine because a SELECT
statement in an SPL routine is not necessarily optimized immediately before the
database server executes it.

The optimizer creates a query plan for a SELECT statement in an SPL routine
when the database server creates an SPL routine or during the execution of some
versions of the UPDATE STATISTICS statement.

The optimizer reads and applies directives at the time that it creates the query
plan. Because it stores the query plan in a system catalog table, the SELECT
statement is not reoptimized when it is executed. Therefore, settings of
IFX_DIRECTIVES and DIRECTIVES affect SELECT statements inside an SPL
routine when they are set at any of the following times:
v Before the CREATE PROCEDURE statement
v Before the UPDATE STATISTICS statements that cause SQL in SPL to be
optimized
v During certain circumstances when SELECT statements have variables supplied
at runtime

Forcing reoptimization to avoid an index and previously prepared

statement problem
If the AUTO_REPREPARE configuration parameter and the
IFX_AUTO_REPREPARE session environment variable are enabled, Informix
automatically recompiles prepared statements and SPL routines after the schema of
a referenced table is modified by a DDL statement. If the AUTO_REPREPARE
configuration parameter or the IFX_AUTO_REPREPARE session environment
variable is disabled, you can take steps to prevent errors.

If the AUTO_REPREPARE configuration parameter or the IFX_AUTO_REPREPARE

session environment variable is disabled, the following error can result when
prepared objects or SPL routines are executed after the schema of a table
referenced by the prepared object or indirectly referenced by the SPL routine has
been modified.
-710 Table <table-name> has been dropped, altered, or renamed.

This error can occur with explicitly prepared statements. These statements have the
following form:
PREPARE statement id FROM quoted string

After a statement has been prepared in the database server and before execution of
the statement, a table to which the statement refers might have been renamed or
altered, possibly changing the structure of the table. Problems can occur as a result.

Adding an index to the table after preparing the statement can also invalidate the
statement. A subsequent OPEN command for a cursor fails if the cursor refers to
the invalid prepared statement; the failure occurs even if the OPEN command has
the WITH REOPTIMIZATION clause.

If an index was added after the statement was prepared, you must prepare the
statement again and declare the cursor again. You cannot simply reopen the cursor
if it was based on a prepared statement that is no longer valid.

Chapter 11. Optimizer directives 11-13

 This error can also occur with SPL routines. Before the database server executes a
new SPL routine the first time, it optimizes the code (statements) in the SPL
routine. Optimization makes the code depend on the structure of the tables that
the procedure references. If the table structure changes after the procedure is
optimized, but before it is executed, this error can occur.

Each SPL routine is optimized the first time that it is run (not when it is created).
This behavior means that an SPL routine might succeed the first time it is run but
fail later under virtually identical circumstances. The failure of an SPL routine can
also be intermittent, because failure during one execution forces an internal
warning to reoptimize the procedure before the next execution.

The database server keeps a list of tables that the SPL routine references explicitly.
Whenever any one of these explicitly referenced tables is modified, the database
server reoptimizes the procedure the next time the procedure is executed.

However, if the SPL routine depends on a table that is referenced only indirectly,
the database server cannot detect the need to reoptimize the procedure after that
table is changed. For example, a table can be referenced indirectly if the SPL
routine invokes a trigger. If a table that is referenced by the trigger (but not
directly by the SPL routine) is changed, the database server does not know that it
should reoptimize the SPL routine before running it. When the procedure is run
after the table has been changed, this error can occur.

Use one of two methods to recover from this error:

v Issue the UPDATE STATISTICS statement to force reoptimization of the
procedure.
v Rerun the procedure.

To prevent this error, you can force reoptimization of the SPL routine. To force
reoptimization, execute the following statement:
UPDATE STATISTICS FOR PROCEDURE procedure name

You can add this statement to your program in either of the following ways:
v Place the UPDATE STATISTICS statement after each statement that changes the
mode of an object.
v Place the UPDATE STATISTICS statement before each execution of the SPL
routine.
For efficiency, you can put the UPDATE STATISTICS statement with the action that
occurs less frequently in the program (change of object mode or execution of the
procedure). In most cases, the action that occurs less frequently in the program is
the change of object mode.

When you follow this method of recovering from this error, you must execute the
UPDATE STATISTICS statement for each procedure that references the changed
tables indirectly unless the procedure also references the tables explicitly.

You can also recover from this error by simply rerunning the SPL routine. The first
time that the stored procedure fails, the database server marks the procedure as
needing reoptimization. The next time that you run the procedure, the database
server reoptimizes the procedure before running it. However, running the SPL
routine twice might not be practical or safe. A safer choice is to use the UPDATE
STATISTICS statement to force reoptimization of the procedure.

11-14 IBM Informix Performance Guide

External optimizer directives
If you are user informix, you can create, save, and delete external directives.

Creating and saving external directives

You can define external directives by creating association records that include query
optimizer directives, and saving those records in the sysdirectives system catalog
table. Association records associate a list of one or more optimizer directives with a
specific query text. The database server can apply those optimizer directives to
subsequent instances of the same query text.

Use the SAVE EXTERNAL DIRECTIVES statement to create the association record
to use for the list of one or more query directives These directives are applied
automatically to subsequent instances of the same query.

The following example shows a SAVE EXTERNAL TABLE statement that registers
an association-record in the system catalog as a new row in the sysdirectives table
that can be used as a query optimizer directive.
SAVE EXTERNAL DIRECTIVES {+INDEX(t1,i11)} ACTIVE FOR
SELECT {+INDEX(t1, i2) } c1 FROM t1 WHERE c1=1;

The following data is stored in the association record that the SQL statement above
defined:
id 16
query select {+INDEX(t1, i2) } c1 from t1 where c1=1
directive INDEX(t1,i11)
directivecode BYTE value
active 1
hashcode -589336273

Here {+INDEX(t1,i11)}, the external directive that followed the DIRECTIVES

keyword, will be applied to future instances of the specified query, but the inline
{+INDEX(t1,i2)} directive will be ignored.

The information in the external directives that immediately follow the

DIRECTIVES keyword must be within comment indicators, just as the same
directives would appear in SELECT, UPDATE, MERGE, and DELETE statements,
except that blank characters, rather than comma ( , ) symbols, are the required
separators if the list of external directives includes more than one directive.

Enabling external directives

After you create and save external directives, you must set the configuration
parameter and environmental variable that enable the directives. The database
server searches for a directive for a query only if the external directives are set on
both the database server and the client.

Enable the directive by using a combination of the EXT_DIRECTIVES configuration

parameter, which is in the ONCONFIG file, and the IFX_EXTDIRECTIVES
client-side environment variable.

The EXT_DIRECTIVE values that you can use are:

Value Explanation
0 (default) Off. The directive cannot be enabled, even if IFX_EXTDIRECTIVES is
enabled.

Chapter 11. Optimizer directives 11-15

 Value Explanation
1 On. The directive can be enabled for a session if
IFX_EXTDIRECTIVES is enabled.
2 On. The directive can be used even if IFX_EXTDIRECTIVES is not
enabled.

You can also use the EXTDIRECTIVES option of the SET ENVIRONMENT
statement to enable or disable external directives during a session. What you
specify with the EXTDIRECTIVES option overwrites the external directive setting
that is specified in the EXT_DIRECTIVES configuration parameter in the
ONCONFIG file.

To overwrite the value for enabling or disabling the external directive in the
ONCONFIG file:
v To enable the external directives during a session, specify 1, on, or ON as the
value for SET ENVIRONMENT EXTDIRECTIVES.
v To disable the external directives during a session, specify 0, off, or OFF as the
value for SET ENVIRONMENT EXTDIRECTIVES.

To enable the default values specified in the EXT_DIRECTIVES configuration

parameter and in the client-side IFX_EXTDIRECTIVES environment variable
during a session, specify DEFAULT as the value for the EXTDIRECTIVES option of
the SET ENVIRONMENT statement.

The explain output file specifies whether external directives are in effect.
Related concepts:
“The explain output file” on page 10-10
“Query statistics section provides performance debugging information” on page
10-11
“Report that shows the query plan chosen by the optimizer” on page 10-9
Using the FILE TO option (SQL Syntax)
Default name and location of the explain output file on UNIX (SQL Syntax)
Default name and location of the output file on Windows (SQL Syntax)
Related reference:
SET EXPLAIN statement (SQL Syntax)
onmode -Y: Dynamically change SET EXPLAIN (Administrator's Reference)
onmode and Y arguments: Change query plan measurements for a session
(SQL administration API) (Administrator's Reference)

Deleting external directives

When you no longer need an external directive, the DBA or user informix can use
the DELETE statement of SQL to remove it from the sysdirectives system catalog
table.

When external directives are enabled and the sysdirectives system catalog table is
not empty,
v the database server compares every query with the query text of every ACTIVE
external directive,

11-16 IBM Informix Performance Guide

v and for queries executed by the DBA (or by user informix) with every TEST
ONLY external directive.

The purpose of external directives is to improve the performance of queries that

match the query string, but the use of such directives can potentially slow other
queries, if the query execution optimizer must compare the query strings of a large
number of active external directives with the text of every SELECT statement. For
this reason, IBM recommends that the DBA not allow the sysdirectives table to
accumulate more than a few ACTIVE rows. (An alternative way to avoid
unintended performance impact on other queries is to disable support for external
directives by setting the EXT_DIRECTIVES configuration parameter to 0. Setting
the IFX_EXTDIRECTIVES client environment variable to 0 has the same effect.)

Chapter 11. Optimizer directives 11-17

11-18 IBM Informix Performance Guide
Chapter 12. Parallel database query (PDQ)
You can manage how the database server performs PDQ and you can monitor the
resources that the database server uses for PDQ.

What PDQ is
Parallel database query (PDQ) is a database server feature that can improve
performance dramatically when the server processes queries that decision-support
applications initiate. PDQ enables Informix to distribute the work for one aspect of
a query among several processors. For example, if a query requires an aggregation,
Informix can distribute the work for the aggregation among several processors.

PDQ also includes tools for resource management.

Another database server feature, table fragmentation, allows you to store the parts of
a table on different disks. PDQ delivers maximum performance benefits when the
data that you query is in fragmented tables. For information about how to use
fragmentation for maximum performance, see “Planning a fragmentation strategy”
on page 9-1.
Related concepts:
“Database server operations that use PDQ” on page 12-2
“The allocation of resources for parallel database queries” on page 12-7
“Managing PDQ queries” on page 12-12
“Monitoring resources used for PDQ and DSS queries” on page 12-16

Structure of a PDQ query

Each decision-support query has a primary thread. The database server can start
additional threads to perform tasks for the query (for example, scans and sorts).
Depending on the number of tables or fragments that a query must search and the
resources that are available for a decision-support query, the database server
assigns different components of a query to different threads.

The database server initiates these PDQ threads, which are listed as secondary
threads in the SET EXPLAIN output.

Secondary threads are further classified as either producers or consumers, depending

on their function. A producer thread supplies data to another thread. For example,
a scan thread might read data from shared memory that corresponds to a given
table and pass it along to a join thread. In this case, the scan thread is considered a
producer, and the join thread is considered a consumer. The join thread, in turn,
might pass data along to a sort thread. When doing so, the join thread is
considered a producer, and the sort thread is considered a consumer.

Several producers can supply data to a single consumer. When this situation
occurs, the database server sets up an internal mechanism, called an exchange, that
synchronizes the transfer of data from those producers to the consumer. For
instance, if a fragmented table is to be sorted, the optimizer typically calls for a
separate scan thread for each fragment. Because of different I/O characteristics, the
scan threads can be expected to complete at different times. An exchange is used to

© Copyright IBM Corp. 1996, 2014 12-1

 funnel the data produced by the various scan threads into one or more sort threads
with a minimum of buffering. Depending on the complexity of the query, the
optimizer might call for a multilayered hierarchy of producers, exchanges, and
consumers. Generally speaking, consumer threads work in parallel with producer
threads so that the amount of intermediate buffering that the exchanges perform
remains negligible.

The database server creates these threads and exchanges automatically and
transparently. They terminate automatically as they complete processing for a
given query. The database server creates new threads and exchanges as needed for
subsequent queries.

Database server operations that use PDQ

Informix processes some types of SQL operations that the database server
processes in parallel. However some situations limit the degree of parallelism that
Informix can use.

In the topics on database server operations that use PDQ in this section, a query is
any SELECT statement.
Related concepts:
“What PDQ is” on page 12-1

Parallel update and delete operations

Informix performs some types of update and delete operations in parallel.

The database server takes the following two steps to process UPDATE and
DELETE statements:
1. Fetches the qualifying rows.
2. Applies the action of updating or deleting.

The database server performs the first step of an UPDATE or DELETE statement in
parallel, with the following exceptions:
v The targeted table in a DELETE statement has a referential constraint that can
cascade to a child table.
v The UPDATE or DELETE statement contains an OR clause and the optimizer
chooses an OR index to process the OR filter.
v The UPDATE statement contains a subquery that the optimizer converts into a
join.

Parallel insert operations

Informix performs some types of insert operations in parallel.

The types of insert operations that the server performs in parallel are:
v SELECT...INTO TEMP inserts using explicit temporary tables.
v INSERT INTO...SELECT inserts using implicit temporary tables.

Explicit inserts with SELECT...INTO TEMP statements

The database server can insert rows in parallel into explicit temporary tables that
you specify in SQL statements of the form SELECT....INTO TEMP.

For example, the database server can perform the inserts in parallel into the
temporary table, temp_table, as the following example shows:

12-2 IBM Informix Performance Guide

 SELECT * FROM table1 INTO TEMP temp_table

To perform parallel inserts into a temporary table:

1. Set PDQ priority > 0.
You must meet this requirement for any query that you want the database
server to perform in parallel.
2. Set DBSPACETEMP to a list of two or more dbspaces.
This step is required because of the way that the database server performs the
insert. To perform the insert in parallel, the database server first creates a
fragmented temporary table. So that the database server knows where to store
the fragments of the temporary table, you must specify a list of two or more
dbspaces in the DBSPACETEMP configuration parameter or the
DBSPACETEMP environment variable. In addition, you must set
DBSPACETEMP to indicate storage space for the fragments before you execute
the SELECT...INTO statement.

The database server performs the parallel insert by writing in parallel to each of
the fragments in a round-robin fashion. Performance improves as you increase the
number of fragments.

Implicit inserts with INSERT INTO...SELECT statements

The database server can also insert rows in parallel into implicit tables that it
creates when it processes SQL statements of the form INSERT INTO...SELECT.

For example, the database server processes the following INSERT statement in
parallel:
INSERT INTO target_table SELECT * FROM source_table

The target table can be either a permanent table or a temporary table.

The database server processes this type of INSERT statement in parallel only when
the target tables meet the following criteria:
v The value of PDQ priority is greater than 0.
v The target table is fragmented into two or more dbspaces.
v The target table has no enabled referential constraints or triggers.
v The target table is not a remote table.
v In a database with logging, the target table does not contain filtering constraints.
v The target table does not contain columns of TEXT or BYTE data type.

The database server does not process parallel inserts that reference an SPL routine.
For example, the database server never processes the following statement in
parallel:
INSERT INTO table1 EXECUTE PROCEDURE ins_proc

Parallel index builds

Index builds can take advantage of PDQ and can be parallelized. The database
server performs both scans and sorts in parallel for index builds.

The following operations initiate index builds:

v Create an index.
v Add a unique, primary key.
v Add a referential constraint.

Chapter 12. Parallel database query (PDQ) 12-3

 v Enable a referential constraint.

When PDQ is in effect, the scans for index builds are controlled by the PDQ
configuration parameters described in “The allocation of resources for parallel
database queries” on page 12-7.

If you have a computer with multiple CPUs, the database server uses two sort
threads to sort the index keys. The database server uses two sort threads during
index builds without the user setting the PSORT_NPROCS environment variable.

Parallel user-defined routines

If a query contains a user-defined routine (UDR) in an expression, the database
server can execute a query in parallel when you turn on PDQ.

The database server can perform the following parallel operations if the UDR is
written and registered appropriately:
v Parallel scans
v Parallel comparisons with the UDR

For more information about how to enable parallel execution of UDRs, see
“Parallel UDRs” on page 13-38.

Hold cursors that use PDQ

When hold cursors that are created by declaring the WITH HOLD qualifier have
no locks, PDQ is enabled.

PDQ will be set for hold cursors in the following cases:

v Queries with Dirty Read or Committed Read isolation level, ANSI, and
read-only cursor
v Queries with Dirty Read or Committed Read isolation level, NON-ANSI,
non-updateable cursor

SQL operations that do not use PDQ

The database server does not process some types of queries in parallel.

For example, the server does not process the following types of queries in parallel:
v Queries started with an isolation level of Cursor Stability
Subsequent changes to the isolation level do not affect the parallelism of queries
already prepared. This situation results from the inherent nature of parallel
scans, which scan several rows simultaneously.
v Queries that use a cursor declared as FOR UPDATE
v Queries in the FOR EACH ROW section of the Action clause of a Select trigger
v A DELETE or MERGE statement in the FOR EACH ROW section of the Action clause
of a Delete trigger
v An INSERT or MERGE statement in the FOR EACH ROW section of the Action
clause of an Insert trigger
v An UPDATE or MERGE statement in the FOR EACH ROW section of the Action
clause of an Update trigger
v Data definition language (DDL) statements.
For a complete list of the DDL statements of SQL that Informix supports, see the
IBM Informix Guide to SQL: Syntax.

12-4 IBM Informix Performance Guide

 In addition, the database server does not process sequence objects in PDQ
operations. If your SQL statement includes sequencing operations, such as
expressions that include the NEXTVAL or CURRVAL operators, PDQ processing is
unavailable for that statement.

Update statistics operations affected by PDQ

An SQL UPDATE STATISTICS statement that is not processed in parallel, is
affected by PDQ because it must allocate the memory used for sorting. Thus the
behavior of the UPDATE STATISTICS statement is affected by the memory
management associated with PDQ.

The database server must allocate the memory that the UPDATE STATISTICS
statement uses for sorting.

If you have an extremely large database and indexes are fragmented, UPDATE
STATISTICS LOW can automatically run statements in parallel. For more
information, see “Update statistics in parallel on very large databases” on page
13-17.

SPL routines and triggers and PDQ

Statements that involve SPL routines are not executed in parallel. However,
statements within procedures are executed in parallel.

When the database server executes an SPL routine, it does not use PDQ to process
non-related SQL statements contained in the procedure. Each SQL statement can be
executed independently in parallel, however, using intraquery parallelism when
possible. As a consequence, you should limit the use of procedure calls from
within data manipulation language (DML) statements if you want to use the
parallel-processing abilities of the database server. For a complete list of DML
statements, see the IBM Informix Guide to SQL: Syntax.

The database server uses intraquery parallelism to process the statements in the
body of an SQL trigger in the same way that it processes statements in SPL
routines. For restrictions on using PDQ for queries in some triggered actions of
Select, Insert, and Update triggers, see “SQL operations that do not use PDQ” on
page 12-4.

Correlated and uncorrelated subqueries

The database server does not use PDQ to process correlated subqueries. Only one
thread at a time can execute a correlated subquery. While one thread executes a
correlated subquery, other threads that request to execute the subquery are blocked
until the first one completes.

For uncorrelated subqueries, only the first thread that makes the request actually
executes the subquery. Other threads simply use the results of the subquery and
can do so in parallel.

As a consequence, it is strongly recommended that, whenever possible, you use

joins rather than subqueries to build queries so that the queries can take advantage
of PDQ.

OUTER index joins and PDQ

The database server reduces the PDQ priority of queries that contain OUTER index
joins to LOW (if the priority is set to a higher value) for the duration of the query.

Chapter 12. Parallel database query (PDQ) 12-5

 If a subquery or a view contains OUTER index joins, the database server lowers
the PDQ priority of only that subquery or view, not of the parent query or any
other subquery.

Remote tables used with PDQ

Although the database server can process the data stored in a remote table in
parallel, the data is communicated serially because the database server allocates a
single thread to submit and receive the data from the remote table. The database
server lowers the PDQ priority of queries that require access to a remote database
to LOW.

In this case, all local scans are parallel, but all local joins and remote access are
nonparallel.

The Memory Grant Manager

The Memory Grant Manager (MGM) is a database server component that
coordinates the use of memory, CPU virtual processors (VPs), disk I/O, and scan
threads among decision-support queries. The MGM uses the DS_MAX_QUERIES,
DS_TOTAL_MEMORY, DS_MAX_SCANS, and MAX_PDQPRIORITY configuration
parameters to determine the quantity of these PDQ resources that can be granted
to a decision-support query.

The MGM dynamically allocates the following resources for decision-support

queries:
v The number of scan threads that are started for each decision-support query
v The number of threads that can be started for each query
v The amount of memory in the virtual portion of database server shared memory
that the query can reserve

When your database server system has heavy OLTP use, and you find performance
is degrading, you can use the MGM facilities to limit the resources that are
committed to decision-support queries. During off-peak hours, you can designate a
larger proportion of the resources to parallel processing, which achieves higher
throughput for decision-support queries.

The MGM grants memory to a query for such activities as sorts, hash joins, and
processing of GROUP BY clauses. The amount of memory that decision-support
queries use cannot exceed DS_TOTAL_MEMORY.

The MGM grants memory to queries in quantum increments. To calculate the

approximate size of the quantum, use the following formula:
memory quantum = DS_TOTAL_MEMORY / DS_MAX_QUERIES

For example, if DS_TOTAL_MEMORY is 12 MB and DS_MAX_QUERIES is 4, the

quantum is 3 MB (12/4). Thus, with these values in effect, a quantum of memory
equals 3 MB. The database server can adjust the size of the quantum dynamically
when it grants memory. In general, memory is allocated more efficiently when
quanta are smaller. You can often improve performance of concurrent queries by
increasing DS_MAX_QUERIES to reduce the size of a quantum of memory.

To monitor resources that the MGM allocates, run the onstat -g mgm command.
This command shows only the amount of memory that is used; it does not show
the amount of memory that is granted.

12-6 IBM Informix Performance Guide

 The MGM also grants a maximum number of scan threads per query that is based
on the values of the DS_MAX_SCANS and the DS_MAX_QUERIES parameters.

The following formula yields the maximum number of scan threads per query:
scan_threads = min (nfrags, DS_MAX_SCANS * (pdqpriority / 100)
* (MAX_PDQPRIORITY / 100))
nfrags Is the number of fragments in the table with the largest number of
fragments.
pdqpriority
Is the value for PDQ priority that is set by either the PDQPRIORITY
environment variable or the SQL statement SET PDQPRIORITY.

The PDQPRIORITY environment variable and the SQL statement SET PDQPRIORITY
request a percentage of PDQ resources for a query. You can use the
MAX_PDQPRIORITY configuration parameter to limit the percentage of the
requested resources that a query can obtain and to limit the impact of
decision-support queries on OLTP processing.
Related concepts:
Chapter 4, “Effect of configuration on memory utilization,” on page 4-1
“Limiting the priority of decision-support queries” on page 12-8
“The DS_TOTAL_MEMORY configuration parameter and memory utilization” on
page 4-12
Related reference:
onstat -g mgm command: Print MGM resource information (Administrator's
Reference)

The allocation of resources for parallel database queries

When you configure the database server, consider how the use of PDQ affects
users of OLTP, decision-support (DSS) applications, and other applications. Then
you can plan how to allocate resources for PDQ.

When the database server uses PDQ to perform a query in parallel, it puts a heavy
load on the operating system. In particular, PDQ exploits the following resources:
v Memory
v CPU VPs
v Disk I/O (to fragmented tables and temporary table space)
v Scan threads

You can control how the database server uses resources in the following ways:
v Limit the priority of parallel database queries.
v Adjust the amount of memory.
v Limit the number of scan threads.
v Limit the number of concurrent queries.
Related concepts:
“What PDQ is” on page 12-1

Chapter 12. Parallel database query (PDQ) 12-7

 Limiting the priority of decision-support queries
You can limit the parallel processing resources that decision-support (DSS) queries
consume by adjusting the values of PDQPRIORITY environment variable, the
MAX_PDQPRIORITY configuration parameter, and other configuration parameters.

The default value for the PDQ priority of individual applications is 0, which means
that PDQ processing is not used. The database server uses this value unless one of
the following actions overrides it:
v You set the PDQPRIORITY environment variable.
v The application uses the SET PDQPRIORITY statement.

The PDQPRIORITY environment variable and the MAX_PDQPRIORITY

configuration parameter work together to control the amount of resources to
allocate for parallel processing. Setting these correctly is critical for the effective
operation of PDQ.

The MAX_PDQPRIORITY configuration parameter allows you to limit the parallel

processing resources that DSS queries consume. Thus, the PDQPRIORITY
environment variable sets a reasonable or recommended priority value, and
MAX_PDQPRIORITY limits the resources that an application can claim.

The MAX_PDQPRIORITY configuration parameter specifies the maximum

percentage of the requested resources that a query can obtain. For instance, if
PDQPRIORITY is 80 and MAX_PDQPRIORITY is 50, each active query receives
an amount of memory equal to 40 percent of DS_TOTAL_MEMORY, rounded
down to the nearest quantum. In this example, MAX_PDQPRIORITY effectively
limits the number of concurrent decision-support queries to two. Subsequent
queries must wait for earlier queries to finish before they acquire the resources that
they need to run.

An application or user can use the DEFAULT tag of the SET PDQPRIORITY
statement to use the value for PDQ priority if the value has been set by the
PDQPRIORITY environment variable. DEFAULT is the symbolic equivalent of a
-1 value for PDQ priority.

You can use the onmode command-line utility to change the values of the
following configuration parameters temporarily:
v Use onmode -M to change the value of DS_TOTAL_MEMORY.
v Use onmode -Q to change the value of DS_MAX_QUERIES.
v Use onmode -D to change the value of MAX_PDQPRIORITY.
v Use onmode -S to change the value of DS_MAX_SCANS.

These changes remain in effect only as long as the database server remains up and
running. When the database server starts, it uses the values listed in the
ONCONFIG file.

For more information about the preceding parameters, see Chapter 4, “Effect of
configuration on memory utilization,” on page 4-1. For more information about
onmode, see your IBM Informix Administrator's Reference.

If you must change the values of the decision-support parameters on a regular

basis (for example, to set MAX_PDQPRIORITY to 100 each night for processing
reports), you can use a scheduled operating-system job to set the values. For
information about creating scheduled jobs, see your operating-system manuals.

12-8 IBM Informix Performance Guide

To obtain the best performance from the database server, choose values for the
PDQPRIORITY environment variable and MAX_PDQPRIORITY parameter,
observe the resulting behavior, and then adjust the values for these parameters. No
well-defined rules exist for choosing these environment variable and parameter
values. The following sections discuss strategies for setting PDQPRIORITY and
MAX_PDQPRIORITY for specific needs.
Related concepts:
“The Memory Grant Manager” on page 12-6

Limiting the value of the PDQ priority

You can adjust the value of the MAX_PDQPRIORITY configuration parameter to
adjust the PDQ priority and allocate more resources to either OLTP or
decision-support processing.

The MAX_PDQPRIORITY configuration parameter limits the PDQ priority that the
database server grants when users either set the PDQPRIORITY environment
variable or issue the SET PDQPRIORITY statement before they issue a query.
When an application or an end user attempts to set a PDQ priority, the priority
that is granted is multiplied by the value that MAX_PDQPRIORITY specifies.

Set the value of MAX_PDQPRIORITY lower when you want to allocate more
resources to OLTP processing.

Set the value of MAX_PDQPRIORITY higher when you want to allocate more
resources to decision-support processing.

The possible range of values is 0 to 100. This range represents the percent of
resources that you can allocate to decision-support processing.

Maximizing OLTP throughput for queries

At times, you might want to allocate resources to maximize the throughput for
individual OLTP queries rather than for decision-support queries.

In this case, set MAX_ PDQPRIORITY to 0, which limits the value of PDQ priority
to OFF. A PDQ priority value of OFF does not prevent decision-support queries
from running. Instead, it causes the queries to run without parallelization. In this
configuration, response times for decision-support queries might be slow.

Conserving resources when using PDQ

If applications make little use of queries that require parallel sorts and parallel
joins, consider using the LOW setting for PDQ priority.

If the database server is operating in a multiuser environment, you might set

MAX_PDQPRIORITY to 1 to increase interquery performance at the cost of some
intraquery parallelism. A trade-off exists between these two different types of
parallelism because they compete for the same resources. As a compromise, you
might set MAX_PDQPRIORITY to some intermediate value (perhaps 20 or 30) and
set PDQPRIORITY to LOW. The environment variable sets the default behavior to
LOW, but the MAX_PDQPRIORITY configuration parameter allows individual
applications to request more resources with the SET PDQPRIORITY statement.

Allowing maximum use of parallel processing

Set PDQPRIORITY and MAX_PDQPRIORITY to 100 if you want the database
server to assign as many resources as possible to parallel processing.

Chapter 12. Parallel database query (PDQ) 12-9

 This setting is appropriate for times when parallel processing does not interfere
with OLTP processing.

Determining the level of parallel processing

You can use different numeric settings for PDQPRIORITY to experiment with the
effects of parallelism on a single application.

For information about how to monitor parallel execution, see “Monitoring

resources used for PDQ and DSS queries” on page 12-16.

Limits on parallel operations associated with PDQ priority

The database server reduces the PDQ priority of queries that contain outer joins to
LOW (if set to a higher value) for the duration of the query. If a subquery or a view
contains outer joins, the database server lowers the PDQ priority only of that
subquery or view, not of the parent query or of any other subquery.

The database server lowers the PDQ priority of queries that require access to a
remote database (same or different database server instance) to LOW if you set it to
a higher value. In that case, all local scans are parallel, but all local joins and
remote accesses are nonparallel.

Using SPL routines with PDQ queries

The database server freezes the PDQ priority that is used to optimize SQL
statements within SPL routines at the time of procedure creation or the last manual
recompilation with the UPDATE STATISTICS statement. You can change the client
value of PDQPRIORITY.

To change the client value of PDQPRIORITY, embed the SET PDQPRIORITY

statement within the body of your SPL routine.

The PDQ priority value that the database server uses to optimize or reoptimize an
SQL statement is the value that was set by a SET PDQPRIORITY statement, which
must have been executed within the same procedure. If no such statement has
been executed, the value that was in effect when the procedure was last compiled
or created is used.

The PDQ priority value currently in effect outside a procedure is ignored within a
procedure when it is executing.

It is suggested that you turn PDQ priority off when you enter a procedure and
then turn it on again for specific statements. You can avoid tying up large amounts
of memory for the procedure, and you can make sure that the crucial parts of the
procedure use the appropriate PDQ priority, as the following example illustrates:
CREATE PROCEDURE my_proc (a INT, b INT, c INT)
Returning INT, INT, INT;
SET PDQPRIORITY 0;
...
SET PDQPRIORITY 85;
SELECT ... (big complicated SELECT statement)
SET PDQPRIORITY 0;
...
;

12-10 IBM Informix Performance Guide

Adjusting the amount of memory for DSS and PDQ queries
You can estimate the amount of shared memory to allocate to decision-support
(DSS) queries. Then, if necessary, you can adjust the value of the
DS_TOTAL_MEMORY configuration parameter, which specifies the amount of
memory available for PDQ queries.

Use the following formula as a starting point for estimating the amount of shared
memory to allocate to DSS queries:
DS_TOTAL_MEMORY = p_mem - os_mem - rsdnt_mem -
(128 kilobytes * users) - other_mem
p_mem represents the total physical memory that is available on the host
computer.
os_mem
represents the size of the operating system, including the buffer cache.
resdnt_mem
represents the size of Informix resident shared memory.
users is the number of expected users (connections) specified in the NETTYPE
configuration parameter.
other_mem
is the size of memory used for other (non-IBM Informix) applications.

The value for DS_TOTAL_MEMORY that is derived from this formula serves only
as a starting point. To arrive at a value that makes sense for your configuration,
you must monitor paging and swapping. (Use the tools provided with your
operating system to monitor paging and swapping.) When paging increases,
decrease the value of DS_TOTAL_MEMORY so that processing the OLTP workload
can proceed.

The amount of memory that is granted to a single parallel database query depends
on many system factors, but in general, the amount of memory granted to a single
parallel database query is proportional to the following formula:
memory_grant_basis = (DS_TOTAL_MEMORY/DS_MAX_QUERIES) *
(PDQPRIORITY / 100) *
(MAX_PDQPRIORITY / 100)

However, if the currently executing queries on all databases of the server instance
require more memory than this estimate of the average allocation, another query
might overflow to disk or might wait until concurrent queries completed execution
and released sufficient memory resources for running the query. The following
alternative formula estimates the PDQ memory available for a single query
directly:
memory_for_single_query = DS_TOTAL_MEMORY *
(PDQPRIOIRTY / 100) *
(MAX_PDQPRIORITY / 100)

Limiting the number of concurrent scans

The database server apportions some number of scans to a query according to its
PDQ priority (among other factors). You can adjust the value of the
DS_MAX_SCANS configuration parameter to limit the number of concurrent scans.

The DS_MAX_SCANS and MAX_PDQPRIORITY configuration parameters allow

you to limit the resources that users can assign to a query, according to the
following formula:

Chapter 12. Parallel database query (PDQ) 12-11

 scan_threads = min (nfrags, (DS_MAX_SCANS * (pdqpriority / 100)
* (MAX_PDQPRIORITY / 100) )
nfrags is the number of fragments in the table with the largest number of
fragments.
pdqpriority
is the PDQ priority value set by either the PDQPRIORITY environment
variable or the SET PDQPRIORITY statement.

For example, suppose a large table contains 100 fragments. With no limit on the
number of concurrent scans allowed, the database server would concurrently
execute 100 scan threads to read this table. In addition, many users could initiate
this query.

As the database server administrator, you set the DS_MAX_SCANS configuration

parameter to a value lower than the number of fragments in this table to prevent
the database server from being flooded with scan threads by multiple
decision-support queries. You can set DS_MAX_SCANS to 20 to ensure that the
database server concurrently executes a maximum of 20 scan threads for parallel
scans. Furthermore, if multiple users initiate parallel database queries, each query
receives only a percentage of the 20 scan threads, according to the PDQ priority
assigned to the query and the value for the MAX_PDQPRIORITY configuration
parameter that the database server administrator sets.

Limiting the maximum number of PDQ queries

You can adjust the maximum number of PDQ queries that can run concurrently by
changing the value of the DS_MAX_QUERIES configuration parameter.

The DS_MAX_QUERIES configuration parameter limits the number of concurrent

decision-support queries that can run.

To estimate the number of decision-support (DSS) queries that the database server
can run concurrently, count each query that runs with PDQ priority set to 1 or
greater as one full query.

The database server allocates less memory to queries that run with a lower
priority, so you can assign lower-priority queries a PDQ priority value that is
between 1 and 30, depending on the resource impact of the query. The total
number of queries with PDQ priority values greater than 0 cannot exceed the
value of DS_MAX_QUERIES.

Managing PDQ queries

The database server administrator, the writer of an application, and the users all
have a certain amount of control over the amount of resources that Informix
allocates to processing a query. The database server administrator exerts control
through the use of configuration parameters. The application developer or the user
can exert control through an environment variable or SQL statement.
Related concepts:
“What PDQ is” on page 12-1

Analyzing query plans with SET EXPLAIN output

You can use SET EXPLAIN output to study the query plans of an application. The
output of the SET EXPLAIN statement shows decisions that the query optimizer

12-12 IBM Informix Performance Guide

 makes. It shows whether parallel scans are used, the maximum number of threads
required to answer the query, and the type of join used for the query.

You can restructure a query or use OPTCOMPIND to change how the optimizer
treats the query.

Influencing the choice of a query plan

The OPTCOMPIND environment variable and the OPTCOMPIND configuration
parameter indicate the preferred join plan, thus assisting the optimizer in selecting
the appropriate join method for parallel database queries. To influence the
optimizer in its choice of a join plan, you can set the OPTCOMPIND configuration
parameter.

The value that you assign to the OPTCOMPIND configuration parameter is

referenced only when applications do not set the OPTCOMPIND environment
variable.

Set OPTCOMPIND to 0 if you want the database server to select a join plan
exactly as it did in versions of the database server prior to version 6.0. This option
ensures compatibility with previous versions of the database server.

An application with an isolation mode of Repeatable Read can lock all records in a
table when it performs a hash join. For this reason, you should set OPTCOMPIND
to 1.

If you want the optimizer to make the determination for you based on cost,
regardless of the isolation level of applications, set OPTCOMPIND to 2.

You can use the SET ENVIRONMENT OPTCOMPIND command to change the
value of OPTCOMPIND within a session. For more information about using this
command, see “Setting the value of OPTCOMPIND within a session” on page 3-11.

For more information about OPTCOMPIND and the different join plans, see “The
query plan” on page 10-1.

Setting the PDQ priority dynamically

You can use the SET PDQPRIORITY statement to set PDQ priority dynamically
within an application. The PDQ priority value can be any integer from -1 through
100.

The PDQ priority set with the SET PDQPRIORITY statement supersedes the
PDQPRIORITY environment variable.

The DEFAULT tag for the SET PDQPRIORITY statement allows an application to
revert to the value for PDQ priority as set by the environment variable, if any. For
more information about the SET PDQPRIORITY statement, see the IBM Informix
Guide to SQL: Syntax.

Enabling the database server to allocate PDQ memory

You can use the IMPLICIT_PDQ session environment option of the SET
ENVIRONMENT statement to enable the database server to determine the amount
of PDQ memory to allocate to queries during the current session. This potentially
overrides the current PDQPRIORITY setting.

Chapter 12. Parallel database query (PDQ) 12-13

 The maximum amount of memory that the database server can allocate, however,
is limited by the physical memory available to your system, and by the settings of
these parameters:
v The PDQPRIORITY environment variable
v The most recent SET PDQPRIORITY statement of SQL
v The MAX_PDQPRIORITY configuration parameter
v The DS_TOTAL_MEMORY configuration parameter
v The BOUND_IMPL_PDQ session environment variable.
The IMPLICIT_PDQ session environment option is available only on systems that
support PDQPRIORITY.

By default, the IMPLICIT_PDQ session environment variable is set to OFF. When

IMPLICIT_PDQ is set to OFF, the server does not override the current
PDQPRIORITY setting.

To enable the database server to determine memory allocations for queries and to
distribute memory among query operators according to their needs, enter the
following statement before you issue the query:
SET ENVIRONMENT IMPLICIT_PDQ ON;

If you instead set the IMPLICIT_PDQ value to an integer in the range from 1 to
100, the database server scales its estimate by the specified value. If you set a low
value, the amount of memory assigned to the query is reduced, which might
increase the amount of query-operator overflow.

The IMPLICIT_PDQ functionality for a query requires at least LOW distribution

statistics on all tables that the query accesses. If distribution statistics are missing
for one or more tables that the query references, the IMPLICIT_PDQ setting has no
effect on the resources available for query execution. This restriction also applies to
star join queries, which are not supported in the case of missing statistics.

Limiting PDQ resource allocation by setting BOUND_IMPL_PDQ

If IMPLICIT_PDQ is set to ON or to a numeric value, you can also use the

BOUND_IMPL_PDQ session environment option of the SET ENVIRONMENT
statement of SQL to specify that the allocated PDQ memory should be no greater
than the current explicit PDQPRIORITY value or range. If the IMPLICIT_PDQ
session environment setting is OFF, whether explicitly off by default, then the
BOUND_IMPL_PDQ setting has no effect.

For example, the following statement forces the database server to use explicit
PDQPRIORITY values as guidelines in allocating memory, if the IMPLICIT_PDQ
session environment option has already been set:
SET ENVIRONMENT BOUND_IMPL_PDQ ON;

If you instead specify a positive integer in the range from 1 to 100, the explicit
PDQPRIORITY value is scaled by that setting during the current session.

When the BOUND_IMPL_PDQ session environment option is set to ON (or to

one), you require the database server to use the explicit PDQPRIORITY setting as
the upper bound for memory that can be allocated to a query. If you set both
IMPLICIT_PDQ and BOUND_IMPL_PDQ, then the explicit PDQPRIORITY value
determines the upper limit of memory that can be allocated to a query.

12-14 IBM Informix Performance Guide

 If you include an integer value in the SET ENVIRONMENT statement, you must
put quote marks around the value. However, do not put quote marks around the
ON and OFF keywords.

The following examples are statements with integer values:

SET ENVIRONMENT IMPLICIT_PDQ "50";
SET ENVIRONMENT BOUND_IMPL_PDQ "80";

User control of PDQ resources

To indicate the PDQ priority of a query, you can set the PDQPRIORITY
environment variable or execute the SET PDQPRIORITY statement prior to issuing
a query. These PDQ priority options allow you to request a certain amount of
parallel-processing resources for the query.

The resources that you request and the amount of resources that the database
server allocates for the query can differ. This difference occurs when the database
server administrator uses the MAX_PDQPRIORITY configuration parameter to put
a ceiling on user-requested resources, as the following topic explains.

DBA control of resources for PDQ and DSS queries

To manage the total amount of resources that the database server allocates to
parallel database and decision-support (DSS) queries, the database server
administrator can set the environment variable and configuration parameters.

Controlling resources allocated to PDQ

To control resources allocated to PDQ, you can set the PDQPRIORITY
environment variable. The queries that do not set the PDQPRIORITY environment
variable before they issue a query do not use PDQ. In addition, to place a ceiling
on user-specified PDQ priority levels, you can set the MAX_PDQPRIORITY
configuration parameter.

When you set the PDQPRIORITY environment variable and MAX_PDQPRIORITY

parameter, you exert control over the resources that the database server allocates
between OLTP and DSS applications. For example, if OLTP processing is
particularly heavy during a certain period of the day, you might want to set
MAX_PDQPRIORITY to 0. This configuration parameter puts a ceiling on the
resources requested by users who use the PDQPRIORITY environment variable,
so PDQ is turned off until you reset MAX_PDQPRIORITY to a nonzero value.

DBA control of resources allocated to decision-support queries

A DBA can control the resources that the database server allocates to
decision-support queries by setting the DS_TOTAL_MEMORY, DS_MAX_SCANS,
and DS_MAX_QUERIES configuration parameters.

In addition to setting limits for decision-support memory and the number of

decision-support queries that can run concurrently, the database server uses these
parameters to determine the amount of memory to allocate to individual
decision-support queries as users submit them. To do so, the database server first
calculates a unit of memory called a quantum by dividing DS_TOTAL_MEMORY
by DS_MAX_QUERIES. When a user issues a query, the database server allocates a
percent of the available quanta equal to the PDQ priority of the query.

You can also limit the number of concurrent decision-support scans that the
database server allows by setting the DS_MAX_SCANS configuration parameter.

Chapter 12. Parallel database query (PDQ) 12-15

 Previous versions of the database server allowed you to set a PDQ priority
configuration parameter in the ONCONFIG file. If your applications depend on a
global setting for PDQ priority, you can use one of the following methods:
v For UNIX: Define PDQPRIORITY as a shared environment variable in the
informix.rc file. For more information about the informix.rc file, see the IBM
Informix Guide to SQL: Reference.
v For Windows: Set the PDQPRIORITY environment variable for a particular
group through a logon profile. For more information about the logon profile, see
your operating-system manual.

Monitoring resources used for PDQ and DSS queries

You can monitor the resources (shared memory and threads) that the Memory
Grant Manager (MGM) has allocated for PDQ queries and the resources that PDQ
and decision-support (DSS) queries currently use.

You monitor PDQ resource use in the following ways:

v Run individual onstat utility commands to capture information about specific
aspects of a running query.
v Execute a SET EXPLAIN statement before you run a query to write the query
plan to an output file.
Related concepts:
“What PDQ is” on page 12-1

Monitoring PDQ resources by using the onstat Utility

You can use various onstat utility commands to determine how many threads are
active and the shared-memory resources that those threads use.

You can use the onstat -g mgm command to monitor how the Memory Grant
Manager (MGM) coordinates memory use and to scan threads.
Related reference:
onstat -g mgm command: Print MGM resource information (Administrator's
Reference)

Monitoring PDQ threads with onstat utility commands

You can obtain information about all of the threads that are running for a
decision-support query by running the onstat -u and onstat -g ath commands.

The onstat -u option lists all the threads for a session. If a session is running a
decision-support query, the output lists the primary thread and any additional
threads. For example, session 10 in Figure 12-1 on page 12-17 has a total of five
threads running.

12-16 IBM Informix Performance Guide

 Userthreads
address flags sessid user tty wait tout locks nreads nwrites
80eb8c ---P--D 0 informix - 0 0 0 33 19
80ef18 ---P--F 0 informix - 0 0 0 0 0
80f2a4 ---P--B 3 informix - 0 0 0 0 0
80f630 ---P--D 0 informix - 0 0 0 0 0
80fd48 ---P--- 45 chrisw ttyp3 0 0 1 573 237
810460 ------- 10 chrisw ttyp2 0 0 1 1 0
810b78 ---PR-- 42 chrisw ttyp3 0 0 1 595 243
810f04 Y------ 10 chrisw ttyp2 beacf8 0 1 1 0
811290 ---P--- 47 chrisw ttyp3 0 0 2 585 235
81161c ---PR-- 46 chrisw ttyp3 0 0 1 571 239
8119a8 Y------ 10 chrisw ttyp2 a8a944 0 1 1 0
81244c ---P--- 43 chrisw ttyp3 0 0 2 588 230
8127d8 ----R-- 10 chrisw ttyp2 0 0 1 1 0
812b64 ---P--- 10 chrisw ttyp2 0 0 1 20 0
812ef0 ---PR-- 44 chrisw ttyp3 0 0 1 587 227
15 active, 20 total, 17 maximum concurrent

Figure 12-1. onstat -u output

The onstat -g ath output also lists these threads and includes a name column that
indicates the role of the thread. Threads that a primary decision-support thread
started have a name that indicates their role in the decision-support query. For
example, Figure 12-2 lists four scan threads, started by a primary thread (sqlexec).

Threads:
tid tcb rstcb prty status vp-class name
...
11 994060 0 4 sleeping(Forever) 1cpu kaio
12 994394 80f2a4 2 sleeping(secs: 51) 1cpu btclean
26 99b11c 80f630 4 ready 1cpu onmode_mon
32 a9a294 812b64 2 ready 1cpu sqlexec
113 b72a7c 810b78 2 ready 1cpu sqlexec
114 b86c8c 81244c 2 cond wait(netnorm) 1cpu sqlexec
115 b98a7c 812ef0 2 cond wait(netnorm) 1cpu sqlexec
116 bb4a24 80fd48 2 cond wait(netnorm) 1cpu sqlexec
117 bc6a24 81161c 2 cond wait(netnorm) 1cpu sqlexec
118 bd8a24 811290 2 ready 1cpu sqlexec
119 beae88 810f04 2 cond wait(await_MC1) 1cpu scan_1.0
120 a8ab48 8127d8 2 ready 1cpu scan_2.0
121 a96850 810460 2 ready 1cpu scan_2.1
122 ab6f30 8119a8 2 running 1cpu scan_2.2

Figure 12-2. onstat -g ath Output

Monitoring resources allocated for a session running a DSS

query
You can monitor the resources allocated for, and used by, a session that is running
a decision-support (DSS) query by running the onstat -g ses command.

The onstat -g ses option displays the following information:

v The shared memory allocated for a session that is running a decision-support
query
v The shared memory used by a session that is running a decision-support query
v The number of threads that the database server started for a session

Chapter 12. Parallel database query (PDQ) 12-17

 For example, in Figure 12-3, session number 49 is running five threads for a
decision-support query.

session #RSAM total used

id user tty pid hostname threads memory memory
57 informix - 0 - 0 8192 5908
56 user_3 ttyp3 2318 host_10 1 65536 62404
55 user_3 ttyp3 2316 host_10 1 65536 62416
54 user_3 ttyp3 2320 host_10 1 65536 62416
53 user_3 ttyp3 2317 host_10 1 65536 62416
52 user_3 ttyp3 2319 host_10 1 65536 62416
51 user_3 ttyp3 2321 host_10 1 65536 62416
49 user_1 ttyp2 2308 host_10 5 188416 178936
2 informix - 0 - 0 8192 6780
1 informix - 0 - 0 8192 4796

Figure 12-3. onstat -g ses output

Identifying parallel scans in SET EXPLAIN output

When PDQ is turned on, the SET EXPLAIN output shows whether the optimizer
chose parallel scans. If the optimizer chose parallel scans, the output lists Parallel.
(If PDQ is turned off, the output lists Serial.)

If PDQ is turned on, the optimizer also indicates the maximum number of threads
that are required to answer the query. The # of Secondary Threads field in the SET
EXPLAIN output indicates the number of threads that are required in addition to
your user session thread. The total number of threads necessary is the number of
secondary threads plus 1.

The following example shows SET EXPLAIN output for a table with fragmentation
and PDQ priority set to LOW:
SELECT * FROM t1 WHERE c1 > 20

Estimated Cost: 2
Estimated # of Rows Returned: 2

1) informix.t1: SEQUENTIAL SCAN (Parallel, fragments: 2)

Filters: informix.t1.c1 > 20

# of Secondary Threads = 1

The following example of partial SET EXPLAIN output shows a query with a hash
join between two fragmented tables and PDQ priority set to ON. The query is
marked with DYNAMIC HASH JOIN, and the table on which the hash is built is
marked with Build Outer.
QUERY:
------
SELECT h1.c1, h2.c1 FROM h1, h2 WHERE h1.c1 = h2.c1

Estimated Cost: 2
Estimated # of Rows Returned: 5

1) informix.h1: SEQUENTIAL SCAN (Parallel, fragments: ALL)

2) informix.h2: SEQUENTIAL SCAN (Parallel, fragments: ALL)

12-18 IBM Informix Performance Guide

DYNAMIC HASH JOIN (Build Outer)
Dynamic Hash Filters: informix.h1.c1 = informix.h2.c1

# of Secondary Threads = 6

The following example of partial SET EXPLAIN output shows a table with
fragmentation, PDQ priority set to LOW, and an index that was selected as the
access plan:
SELECT * FROM t1 WHERE c1 < 13

Estimated Cost: 2
Estimated # of Rows Returned: 1

1) informix.t1: INDEX PATH

(1) Index Keys: c1 (Parallel, fragments: ALL)

Upper Index Filter: informix.t1.c1 < 13

# of Secondary Threads = 3

Chapter 12. Parallel database query (PDQ) 12-19

12-20 IBM Informix Performance Guide
Chapter 13. Improving individual query performance
You can test, monitor, and improve queries.
Related concepts:
Tune the new version for performance and adjust queries (Migration Guide)

Test queries using a dedicated test system

You can test a query on a system that does not interfere with production database
servers. However, you must be careful, because testing queries on a separate
system might distort your tuning decisions.

Even if you use your database server as a data warehouse, you might sometimes
test queries on a separate system until you understand the tuning issues that are
relevant to the query.

If you are trying to improve performance of a large query, one that might take
several minutes or hours to complete, you can prepare a scaled-down database in
which your tests can complete more quickly. However, be aware of these potential
problems:
v The optimizer can make different choices in a small database than in a large one,
even when the relative sizes of tables are the same. Verify that the query plan is
the same in the real and the model databases.
v Execution time is rarely a linear function of table size. For example, sorting time
increases faster than table size, as does the cost of indexed access when an index
goes from two to three levels. What appears to be a big improvement in the
scaled-down environment can be insignificant when applied to the full database.

Therefore, any conclusion that you reach as a result of tests in the model database
must be tentative until you verify them in the production database.

You can often improve performance by adjusting your query or data model with
the following goals in mind:
v If you are using a multiuser system or a network, where system load varies
widely from hour to hour, try to perform your experiments at the same time
each day to obtain repeatable results. Start tests when the system load is
consistently light so that you are truly measuring the impact of your query only.
v If the query is embedded in a complicated program, you can extract the SELECT
statement and embed it in a DB-Access script.
Related concepts:
Tune the new version for performance and adjust queries (Migration Guide)

Display the query plan

Before you change a query, display its query plan to determine the kind and
amount of resources that the query requires. The query plan shows what parallel
scans are used, the maximum number of threads required, and the indexes used.

After you study the query plan, examine your data model to see if the changes this
chapter suggests will improve the query.

© Copyright IBM Corp. 1996, 2014 13-1

 You can display the query plan with one of the following methods:
v Execute one of the following SET EXPLAIN statements just before the query:
– SET EXPLAIN ON
This SQL statement displays the chosen query plan. For a description of the
SET EXPLAIN ON output, see “Report that shows the query plan chosen by
the optimizer” on page 10-9.
– SET EXPLAIN ON AVOID_EXECUTE
This SQL statement displays the chosen query plan and does not execute the
query.
v Use one of the following EXPLAIN directives in the query:
– EXPLAIN
– EXPLAIN AVOID_EXECUTE

For more information about these EXPLAIN directives, see “EXPLAIN directives”
on page 11-8.

Improve filter selectivity

You can control the amount of information that a query evaluates. The greater the
precision with which you specify the desired rows, the greater the likelihood that
your queries will complete quickly.

To control the amount of information that the query evaluates, use the WHERE
clause of the SELECT statement. The conditional expression in the WHERE clause
is commonly called a filter.

For information about how filter selectivity affects the query plan that the
optimizer chooses, see “Filters in the query” on page 10-20. The following sections
provide some guidelines to improve filter selectivity.

Filters with user-defined routines

You can improve the selectivity of query filters that include user-defined routines
(UDRs).

You can improve the selectivity if the UDRs have the following features:
v Functional indexes
You can create a functional index on the resulting values of a user-defined routine
or a built-in function that operates on one or more columns. When you create a
functional index, the database server computes the return values of the function
and stores them in the index. The database server can locate the return value of
the function in an appropriate index without executing the function for each
qualifying column.
For more information about indexing user-defined functions, see “Using a
functional index” on page 7-25.
v User-defined selectivity functions
You can write a function that calculates the expected fraction of rows that
qualify for the function. For a brief description of user-defined selectivity
functions, see “Selectivity and cost functions” on page 13-39. For more
information about how to write and register user-defined selectivity functions,
see IBM Informix User-Defined Routines and Data Types Developer's Guide.

13-2 IBM Informix Performance Guide

Avoid some filters
For best performance, avoid filters for certain difficult regular expressions and
filters for noninitial strings.

Avoid difficult regular expressions

The MATCHES and LIKE keywords support wildcard matches, which are
technically known as regular expressions. Some regular expressions are more
difficult than others for the database server to process.

A wildcard in the initial position, as in the following example (find customers

whose first names do not end in y), forces the database server to examine every
value in the column:
SELECT * FROM customer WHERE fname NOT LIKE ’%y’

You cannot use an index with such a filter, so the table in this example must be
accessed sequentially.

If a difficult test for a regular expression is essential, avoid combining it with a

join. If necessary, process the single table and apply the test for a regular
expression to select the desired rows. Save the result in a temporary table and join
that table to the others.

Regular-expression tests with wildcards in the middle or at the end of the operand
do not prevent the use of an index when one exists.

Avoid noninitial substrings

For best performance, avoid filters for noninitial strings. A filter based on a
noninitial substring of a column requires the database server to test every value in
the column.

For example, in the following code, a noninitial substring requires the database
server to test every value in the column:
SELECT * FROM customer
WHERE zipcode[4,5] > ’50’

The database server cannot use an index to evaluate such a filter.

The optimizer uses an index to process a filter that tests an initial substring of an
indexed column. However, the presence of the substring test can interfere with the
use of a composite index to test both the substring column and another column.

Use join filters and post-join filters

The database server provides support for using a subset of the ANSI join syntax.

This syntax that includes the following keywords:

v ON keyword to specify the join condition and any optional join filters
v LEFT OUTER JOIN keywords to specify which table is the dominant table (also
referred to as outer table)

For more information about this ANSI join syntax, see the IBM Informix Guide to
SQL: Syntax.

In an ANSI outer join, the database server takes the following actions to process
the filters:

Chapter 13. Improving individual query performance 13-3

 v Applies the join condition in the ON clause to determine which rows of the
subordinate table (also referred to as inner table) to join to the outer table
v Applies optional join filters in the ON clause before and during the join
If you specify a join filter on a base inner table in the ON clause, the database
server can apply it prior to the join, during the scan of the data from the inner
table. Filters on a base subordinate table in the ON clause can provide the
following additional performance benefits:
– Fewer rows to scan from the inner table prior to the join
– Use of index to retrieve rows from the inner table prior to the join
– Fewer rows to join
– Fewer rows to evaluate for filters in the WHERE clause
For information about what occurs when you specify a join filter on an outer
table in the ON clause, see the IBM Informix Guide to SQL: Syntax.
v Applies filters in the WHERE clause after the join
Filters in the WHERE clause can reduce the number of rows that the database
server needs to scan and reduce the number of rows returned to the user.
The term post-join filters refers to these WHERE clause filters.

When distributed queries that use ANSI-compliant LEFT OUTER syntax for
specifying joined tables and nested loop joins are executed, the query is sent to
each participating database server for operations on local tables of those servers.

For example, the demonstration database has the customer table and the cust_calls
table, which tracks customer calls to the service department. Suppose a certain call
code had many occurrences in the past, and you want to see if calls of this kind
have decreased. To see if customers no longer have this call code, use an outer join
to list all customers.

Figure 13-1 shows a sample SQL statement to accomplish this ANSI join query and
the SET EXPLAIN ON output for it.

QUERY:
------
SELECT c.customer_num, c.lname, c.company,
c.phone, u.call_dtime, u.call_code, u.call_descr
FROM customer c
LEFT JOIN cust_calls u ON c.customer_num = u.customer_num
ORDER BY u.call_dtime

Estimated Cost: 14
Estimated # of Rows Returned: 29
Temporary Files Required For: Order By

1) virginia.c: SEQUENTIAL SCAN

2) virginia.u: INDEX PATH

(1) Index Keys: customer_num call_dtime (Serial, fragments: ALL)

Lower Index Filter: virginia.c.customer_num = virginia.u.customer_num

ON-Filters:virginia.c.customer_num = virginia.u.customer_num
NESTED LOOP JOIN(LEFT OUTER JOIN)

Figure 13-1. SET EXPLAIN ON output for an ANSI join

Look at the following lines in the SET EXPLAIN ON output in Figure 13-1:

13-4 IBM Informix Performance Guide

v The ON-Filters: line lists the join condition that was specified in the ON clause.
v The last line of the SET EXPLAIN ON output shows all three keywords (LEFT
OUTER JOIN) for the ANSI join even though this query specifies only the LEFT
JOIN keywords in the FROM clause. The OUTER keyword is optional.

Figure 13-2 shows the SET EXPLAIN ON output for an ANSI join with a join filter
that checks for calls with the I call_code.

QUERY:
------
SELECT c.customer_num, c.lname, c.company,
c.phone, u.call_dtime, u.call_code, u.call_descr
FROM customer c LEFT JOIN cust_calls u
ON c.customer_num = u.customer_num
AND u.call_code = ’I’
ORDER BY u.call_dtime

Estimated Cost: 13
Estimated # of Rows Returned: 25
Temporary Files Required For: Order By

1) virginia.c: SEQUENTIAL SCAN

2) virginia.u: INDEX PATH

Filters: virginia.u.call_code = ’I’

(1) Index Keys: customer_num call_dtime (Serial, fragments: ALL)

Lower Index Filter: virginia.c.customer_num = virginia.u.customer_num

ON-Filters:(virginia.c.customer_num = virginia.u.customer_num
AND virginia.u.call_code = ’I’ )
NESTED LOOP JOIN(LEFT OUTER JOIN)

Figure 13-2. SET EXPLAIN ON output for a join filter in an ANSI join

The main differences between the output in Figure 13-1 on page 13-4 and
Figure 13-2 are as follows:
v The optimizer chooses a different index to scan the inner table.
This new index exploits more filters and retrieves a smaller number of rows.
Consequently, the join operates on fewer rows.
v The ON clause join filter contains an additional filter.

The value in the Estimated # of Rows Returned line is only an estimate and does
not always reflect the actual number of rows returned. The sample query in
Figure 13-2 returns fewer rows than the query in Figure 13-1 on page 13-4 because
of the additional filter.

Figure 13-3 on page 13-6 shows the SET EXPLAIN ON output for an ANSI join
query that has a filter in the WHERE clause.

Chapter 13. Improving individual query performance 13-5

 QUERY:
------
SELECT c.customer_num, c.lname, c.company,
c.phone, u.call_dtime, u.call_code, u.call_descr
FROM customer c LEFT JOIN cust_calls u
ON c.customer_num = u.customer_num
AND u.call_code = ’I’
WHERE c.zipcode = "94040"
ORDER BY u.call_dtime

Estimated Cost: 3
Estimated # of Rows Returned: 1
Temporary Files Required For: Order By

1) virginia.c: INDEX PATH

(1) Index Keys: zipcode (Serial, fragments: ALL)

Lower Index Filter: virginia.c.zipcode = ’94040’

2) virginia.u: INDEX PATH

Filters: virginia.u.call_code = ’I’

(1) Index Keys: customer_num call_dtime (Serial, fragments: ALL)

Lower Index Filter: virginia.c.customer_num = virginia.u.customer_num

ON-Filters:(virginia.c.customer_num = virginia.u.customer_num
AND virginia.u.call_code = ’I’ )
NESTED LOOP JOIN(LEFT OUTER JOIN)

PostJoin-Filters:virginia.c.zipcode = ’94040’

Figure 13-3. SET EXPLAIN ON output for the WHERE clause filter in an ANSI join

The main differences between the output in Figure 13-2 on page 13-5 and
Figure 13-3 are as follows:
v The index on the zipcode column in the post-join filter is chosen for the
dominant table.
v The PostJoin-Filters line shows the filter in the WHERE clause.

Automatic statistics updating

The database server updates statistics automatically according to a predefined
schedule and a set of expiration policies. The Auto Update Statistics (AUS)
maintenance system identifies tables and indexes that require new optimizer
statistics and runs the appropriate UPDATE STATISTICS statements to optimize
query performance.

The AUS maintenance system updates the statistics for tables that are in logged
databases, regardless of the database locale. By making current table statistics
available to the query optimizer, the AUS maintenance system can reduce the risk
of performance degradation from inefficient query plans.

Depending on your system, you might need to adjust the AUS expiration policies
or schedule. The AUS maintenance system resides in the sysadmin database.

You can also view and adjust the AUS maintenance system for table statistics in
the IBM OpenAdmin Tool (OAT) for Informix.
Related concepts:

13-6 IBM Informix Performance Guide

 “Update statistics when they are not generated automatically” on page 13-11

How AUS works

The Auto Update Statistics (AUS) maintenance system uses a combination of
Scheduler sensors, tasks, thresholds, and tables to evaluate and update statistics.

The Scheduler tasks, sensors, thresholds, and tables reside in the sysadmin
database. By default, only user informix is granted access to the sysadmin
database.

The following sequence of events describes how statistics are automatically

updated:
1. The mon_table_profile sensor of the Scheduler runs every day to read data
from the systables table in the sysmaster database. The sensor updates the
mon_table_profile table in the sysadmin database with information about how
much each table has changed.
2. The Auto Update Statistics Evaluation task gathers information every day from
the mon_table_profile table and the systable, sysdistrib, syscolumns, and
sysindices tables in the sysmaster database.
3. The Auto Update Statistics Evaluation task determines which tables need
updates based on the expiration policies.
4. The Auto Update Statistics Evaluation task generates UPDATE STATISTICS
statements and inserts them into the aus_command table in the sysadmin
database.
5. The Auto Update Statistics Refresh task runs the UPDATE STATISTICS
statements from the aus_command table on Saturday and Sunday mornings
between 1:00 AM and 5:00 AM and inserts the results back into the
aus_command table. Any UPDATE STATISTICS statements that did not
complete before 5:00 AM remain in the aus_command table.

The following table describes the tasks, sensors, thresholds, tables, and views in
the sysadmin database that comprise the AUS maintenance system.
Table 13-1. AUS components
Component Type Description
mon_table_profile sensor Compiles table profile information, including
the total number of updates, inserts, and deletes
that occurred on each table.

Defined in the ph_task table.

mon_table_profile table Stores table profile information gathered by its
sensor. Many other Scheduler tasks use
information from this table.
Auto Update Statistics task Identifies tables with stale statistics, based on
Evaluation expiration policies, and generates UPDATE
STATISTICS statements for those tables.

Defined in the ph_task table.

Chapter 13. Improving individual query performance 13-7

 Table 13-1. AUS components (continued)
Component Type Description
aus_command table Stores a list of prioritized UPDATE STATISTICS
statements that are scheduled to be run, and the
results of those statements after they are run.

The aus_cmd_state column indicates the status

of each UPDATE STATISTICS statement:
v P = Pending
v I = In progress
v E = Error
v C = Complete without errors

If the command status is E, the associated SQL

error code is listed in the aus_cmd_err_sql
column and the associated ISAM error code is
listed in the aus_cmd_err_isam column.
Auto Update Statistics task Runs the prepared UPDATE STATISTICS
Refresh statements on Saturdays and Sundays between
1:00 AM and 5:00 AM.

Defined in the ph_task table.

expiration policies thresholds Define the criteria for when to update statistics.

Defined in the ph_threshold table.

aus_cmd_comp view Shows information from the aus_command
table about UPDATE STATISTICS statements
that were run successfully.
aus_cmd_list view Shows information from the aus_command
table about UPDATE STATISTICS statements
that are scheduled to be run.

For information about other features of the Scheduler, see its description in the
IBM Informix Administrator's Guide. For information about the sysadmin database,
see the IBM Informix Administrator's Reference.

AUS expiration policies

The Auto Update Statistics (AUS) maintenance system uses expiration policies as
criteria for identifying user tables that have changed to the extent that their
statistics need to be recalculated.

The ph_threshold table of the sysadmin database stores the following thresholds
for defining AUS expiration policies:
Table 13-2. AUS expiration policy thresholds
Threshold Name Default Value Description
AUS_AGE 30 (days) A time-based expiration policy. Statistics
or distributions are updated for a table
after this amount of time regardless of
how much data has changed.

13-8 IBM Informix Performance Guide

Table 13-2. AUS expiration policy thresholds (continued)
Threshold Name Default Value Description
AUS_AUTO_RULES 1 (enabled) If enabled, statistics are updated using
the higher of the following default
minimum guidelines or user-created
distribution options:
v All tables are updated in LOW mode.
v All the leading index keys are
updated in HIGH mode.
v All non-leading index keys are
updated in MEDIUM mode.
v The minimum resolution for MEDIUM
mode is 2.0.
v The minimum confidence for
MEDIUM mode is 0.95.
v The minimum resolution for HIGH
mode is 0.5.

If the UPDATE STATISTICS statement

was run manually for a table, the
UPDATE STATISTICS statements
generated by the AUS maintenance
system do not reduce the level,
resolution, confidence, or sampling size
options.

If disabled, user-created distribution

options are used.
AUS_CHANGE 10 (percent) A modification-based expiration policy.
Statistics or distributions are updated for
a table after this percentage of data is
changed.
AUS_PDQ 10 (priority) The PDQ priority for UPDATE
STATISTICS statements run by the AUS
maintenance system. By default, all
fragments for each table are analyzed in
parallel. For more information about
PDQ priority, see “Update statistics in
parallel on very large databases” on
page 13-17.
AUS_SMALL_TABLES 100 (rows) Statistics or distributions are updated
every time for a table that has fewer
than this number of rows.

Changing AUS expiration policies

You can change AUS expiration policies to customize how often statistics are
updated based on how old the statistics are, how much data has changed, or how
large the table is.

You must be connected to the sysadmin database as user informix or another

authorized user.

To change the value of an expiration policy, update the value column in the
ph_threshold table in the sysadmin database.

Chapter 13. Improving individual query performance 13-9

 For example, if you find that queries against small tables with 1000 rows or fewer
run faster if their statistics are updated more frequently, you can change the
expiration policy to ensure that their statistics are updated every week. The
following example changes the value of the AUS_SMALL_TABLES threshold to
1000:
UPDATE ph_threshold
SET value = 1000
WHERE name = "AUS_SMALL_TABLES";

The new threshold takes effect the next time the Auto Update Statistics Evaluator
task runs.

Viewing AUS statements

You can view the UPDATE STATISTICS statements generated by the AUS
maintenance system in the aus_cmd_list view before they are run and in the
aus_cmd_comp view after they are run successfully. Both tables are in the
sysadmin database.

You must be connected to the sysadmin database as user informix or another

authorized user.

To view all scheduled UPDATE STATISTICS statements, run this statement:

SELECT * FROM aus_cmd_list;

To see all UPDATE STATISTICS statements that were run successfully in the
previous 30 days, run this statement:
SELECT * FROM aus_cmd_comp;

To view all UPDATE STATISTICS statements that failed, run this statement:
SELECT aus_cmd_exe, aus_cmd_err_sql, aus_cmd_err_isam
FROM aus_command
WHERE aus_cmd_state = "E";

You can also see this information in the IBM OpenAdmin Tool (OAT) for Informix.

Rescheduling AUS
You can change when and for how long the Auto Update Statistics Refresh task
runs.

Updating statistics is resource-intensive. Therefore, by default, statistics are

automatically updated on Saturdays and Sundays between 1:00 AM and 5:00 AM.
If you find that not all pending UPDATE STATISTICS statements can be run in this
time period, or you want statistics to be refreshed more often, you can change the
start time, the end time, and the days of the week to perform this task.

You must be connected to the sysadmin database as user informix or another

authorized user.

To change the schedule of the Auto Update Statistics Refresh task, update the
ph_task table where the value of the tk_name column is Auto Update Statistics
Refresh.

The following example changes the ending time of the task to 6:00 AM:

13-10 IBM Informix Performance Guide

 UPDATE ph_task
SET tk_stop_time = "06:00:00"
WHERE tk_name = "Auto Update Statistics Refresh";

The following example changes the days that the task is run to every day of the
week (Saturday and Sunday are enabled by default):
UPDATE ph_task
SET tk_monday = "T",
SET tk_tuesday = "T",
SET tk_wednesday = "T",
SET tk_thursday = "T",
SET tk_friday = "T",
WHERE tk_name = "Auto Update Statistics Refresh";

Disabling AUS
You can prevent statistics from being updated automatically by disabling the AUS
maintenance system.

You must be connected to the sysadmin database as user informix or another

authorized user.

To disable AUS, you must disable both the Auto Update Statistics Evaluation task
and the Auto Update Statistics Refresh task:
1. Update the value of the tk_enable column of the ph_task table to F where the
value of the tk_name column is Auto Update Statistics Evaluation.
2. Update the value of the tk_enable column of the ph_task table to F where the
value of the tk_name column is Auto Update Statistics Refresh.

The following example disables both tasks:

UPDATE ph_task
SET tk_enable = "F"
WHERE tk_name = "Auto Update Statistics Evaluation";

UPDATE ph_task
SET tk_enable = "F"
WHERE tk_name = "Auto Update Statistics Refresh";

Update statistics when they are not generated automatically

The UPDATE STATISTICS statement updates the statistics in the system catalog
tables that the optimizer uses to determine the lowest-cost query plan.

Important: You do not need to run UPDATE STATISTICS operations when the
statistics are generated automatically.

The following statistics are generated automatically by the CREATE INDEX

statement, with or without the ONLINE keyword:
v Index-level statistics, equivalent to the statistics gathered in the UPDATE
STATISTICS operation in LOW mode, for B-tree indexes.
v Column-distribution statistics, equivalent to the distribution generated in the
UPDATE STATISTICS operation in HIGH mode, for a non-opaque leading
indexed column of an ordinary B-tree index.

To ensure that the optimizer selects a query plan that best reflects the current state
of your tables, run UPDATE STATISTICS at regular intervals when the statistics are
not generated automatically.

Chapter 13. Improving individual query performance 13-11

 Tip: If you run UPDATE STATISTICS LOW on the sysutils database before you
use ON-Bar, the time ON-BAR needs for processing is reduced.

The following table summarizes when to run different UPDATE STATISTICS

statements if the statistics are not generated automatically. If you need to run
UPDATE STATISTICS statements and you have many tables, you can write a script
to generate these UPDATE STATISTICS statements.

Reference for Details and

When to Execute UPDATE STATISTICS Statement Examples
Number of rows has changed UPDATE STATISTICS LOW “Update the statistics for the
significantly number of rows” or “Drop data
DROP DISTRIBUTIONS distributions if necessary when
upgrading” on page 13-13
For all columns that are not the UPDATE STATISTICS LOW “Creating data distributions”
leading column of any index on page 13-13
Queries have non-indexed join UPDATE STATISTICS MEDIUM “Creating data distributions”
columns or filter columns on page 13-13
DISTRIBUTIONS ONLY
Queries have an indexed join UPDATE STATISTICS HIGH table (leading “Creating data distributions”
columns or filter columns column in index) on page 13-13
Queries have a multicolumn UPDATE STATISTICS HIGH table (first “Creating data distributions”
indexed defined on join columns or differing column in multicolumn index) on page 13-13
filter columns
Queries have a multicolumn UPDATE STATISTICS LOW table (all columns “Creating data distributions”
indexed defined on join columns or in multicolumn index) on page 13-13
filter columns
Queries have many small tables (fit UPDATE STATISTICS HIGH on small tables “Creating data distributions”
into one extent) on page 13-13
Queries use SPL routines UPDATE STATISTICS for procedure “Reoptimizing SPL routines”
on page 10-32

For information about the specific statistics that the database server keeps in the
system catalog tables, see “Statistics held for the table and index” on page 10-20.
Related concepts:
“Automatic statistics updating” on page 13-6
Related reference:
UPDATE STATISTICS statement (SQL Syntax)

Update the statistics for the number of rows

When you run UPDATE STATISTICS LOW, the database server updates the
statistics in the table, row, and page counts in the system catalog tables. You
should run UPDATE STATISTICS LOW as often as necessary to ensure that the
statistic for the number of rows is as current as possible.

If the cardinality of a table changes often, run the statement more often for that
table.

LOW is the default mode for UPDATE STATISTICS.

13-12 IBM Informix Performance Guide

 The following sample SQL statement updates the statistics in the systables,
syscolumns, and sysindexes system catalog tables but does not update the data
distributions:
UPDATE STATISTICS FOR TABLE tab1;

Drop data distributions if necessary when upgrading

When you upgrade to a new version of the database server, you might need to
drop distributions to remove the old distribution structure in the sysdistrib system
catalog table.

To drop the old distribution structure in the sysdistrib system catalog table, run
this statement:
UPDATE STATISTICS DROP DISTRIBUTIONS;

Drop distributions in LOW mode without gathering statistics

You can remove distribution information from the sysdistrib table and update the
systables.version column in the system catalog for those tables whose distributions
were dropped, without gathering any LOW mode table and index statistics.

You do this using the DROP DISTRIBUTIONS ONLY option in the UPDATE
STATISTICS statement. Using the DROP DISTRIBUTIONS ONLY option can result
in faster performance because the database server does not gather the table and
index statistics that the LOW mode option generates when the ONLY keyword
does not follow the DROP DISTRIBUTIONS keywords.

For detailed information about how to use the DROP DISTRIBUTIONS ONLY
option, see the IBM Informix Guide to SQL: Syntax.

Creating data distributions

You can generate statistics for a table and you can build data distributions for each
table that your query accesses.

(You do not need to run UPDATE STATISTICS operations when the statistics are
generated automatically.)

The database server creates data distributions, which provide information to the
optimizer, any time the UPDATE STATISTICS MEDIUM or UPDATE STATISTICS
HIGH command is executed.

Important:

The database server creates data distributions by sampling a column's data, sorting
the data, building distributions bins, and inserting the results into the sysdistrib
system catalog table.

You can control the sample size for the scan through the keyword HIGH or
MEDIUM. The difference between UPDATE STATISTICS HIGH and UPDATE
STATISTICS MEDIUM is the number of rows sampled. UPDATE STATISTICS
HIGH scans the entire table, while UPDATE STATISTICS MEDIUM samples only a
subset of rows, based on the confidence and resolution used by the UPDATE
STATISTICS statement.

You can use the LOW keyword with the UPDATE STATISTICS statement only for
fully qualified index keys.

Chapter 13. Improving individual query performance 13-13

 If a distribution has been generated for a column, the optimizer uses that
information to estimate the number of rows that match a query against a column.
Data distributions in sysdistrib supersede values in the colmin and colmax
column of the syscolumns system catalog table when the optimizer estimates the
number of rows returned.

When you use data-distribution statistics for the first time, try to update statistics
in MEDIUM mode for all your tables and then update statistics in HIGH mode for
all columns that head indexes. This strategy produces statistical query estimates for
the columns that you specify. These estimates, on average, have a margin of error
less than percent of the total number of rows in the table, where percent is the value
that you specify in the RESOLUTION clause in the MEDIUM mode. The default
percent value for MEDIUM mode is 2.5 percent. (For columns with HIGH mode
distributions, the default resolution is 0.5 percent.)

With the DISTRIBUTIONS ONLY option, you can execute UPDATE STATISTICS
MEDIUM at the table level or for the entire system because the overhead of the
extra columns is not large.

The database server uses the storage locations that the DBSPACETEMP
environment variable specifies only when you use the HIGH option of UPDATE
STATISTICS.

You can prevent UPDATE STATISTICS operations from using indexes when sorting
rows by setting the third parameter of the DBUPSPACE environment variable to a
value of 1.

For each table that your query accesses, build data distributions according to the
following guidelines. Also see the examples below the guidelines.

To generate statistics on a table:

1. Identify the set of all columns that appear in any single-column or
multi-column index on the table.
2. Identify the subset that includes all columns that are not the leading column of
any index.
3. Run UPDATE STATISTICS LOW on each column in that subset.

To build data distributions for each table that your query accesses:
1. Run a single UPDATE STATISTICS MEDIUM for all columns in a table that do
not head an index.
Use the default parameters unless the table is very large, in which case you
should use a resolution of 1.0 and confidence of 0.99.
2. Run the following UPDATE STATISTICS statement to create distributions for
non-index join columns and non-index filter columns:
UPDATE STATISTICS MEDIUM DISTRIBUTIONS ONLY;
3. Run UPDATE STATISTICS HIGH for all columns that head an index. For the
fastest execution time of the UPDATE STATISTICS statement, you must execute
one UPDATE STATISTICS HIGH statement for each column, as shown in the
example below this procedure.
4. If you have indexes that begin with the same subset of columns, run UPDATE
STATISTICS HIGH for the first column in each index that differs, as shown in
the second example below this procedure.
5. For each single-column or multi-column index on the table:

13-14 IBM Informix Performance Guide

 a. Identify the set of all columns that appear in the index.
b. Identify the subset that includes all columns that are not the leading column
of any index.
c. Run UPDATE STATISTICS LOW on each column in that subset. (LOW is
the default.)
6. For the tables on which indexes were created in Step 3, run an UPDATE
STATISTICS statement to update the sysindexes and syscolumns system
catalog tables, as shown in the following example:
UPDATE STATISTICS FOR TABLE t1(a,b,e,f);
7. For small tables, run UPDATE STATISTICS HIGH, for example:
UPDATE STATISTICS HIGH FOR TABLE t2;

Because the statement constructs the statistics only once for each index, these steps
ensure that UPDATE STATISTICS executes rapidly.

Examples
Example of UPDATE STATISTICS HIGH statements for all columns that head
an index
Suppose you have a table t1 with columns a, b, c, d, e, and f with the
following indexes:
CREATE INDEX ix_1 ON t1 (a, b, c, d) ...
CREATE INDEX ix_3 ON t1 (f) ...
Run the following UPDATE STATISTICS statements for the columns that
head an index:
UPDATE STATISTICS HIGH FOR TABLE t1(a);
UPDATE STATISTICS HIGH FOR TABLE t1(f);
These UPDATE STATISTICS HIGH statements replace the distributions
created with the UPDATE STATISTICS MEDIUM statements with high
distributions for index columns.
Example of UPDATE STATISTICS HIGH statements for the first column in each
index that differs:
For example, suppose you have the following indexes on table t1:
CREATE INDEX ix_1 ON t1 (a, b, c, d) ...
CREATE INDEX ix_2 ON t1 (a, b, e, f) ...
CREATE INDEX ix_3 ON t1 (f) ...
Step 3 on page 13-14 executes UPDATE STATISTICS HIGH on column a
and column f. Then run UPDATE STATISTICS HIGH on columns c and e.
UPDATE STATISTICS HIGH FOR TABLE t1(c);
UPDATE STATISTICS HIGH FOR TABLE t1(e);

In addition, you can run UPDATE STATISTICS HIGH on column b, although this
is usually not necessary.
Related concepts:
“Virtual portion of shared memory” on page 4-2
Related reference:
UPDATE STATISTICS statement (SQL Syntax)

Chapter 13. Improving individual query performance 13-15

 Updating statistics for join columns
In some situations, you might want to run the UPDATE STATISTICS statement
with the HIGH keyword for specific join columns.

Because of improvements and adjusted cost estimates to establish better query

plans, the optimizer depends greatly on an accurate understanding of the
underlying data distributions in certain cases. You might still think that a complex
query does not execute quickly enough, even though you followed the guidelines
in “Creating data distributions” on page 13-13. If your query involves equality
predicates, take one of the following actions:
v Run the UPDATE STATISTICS statement with the HIGH keyword for specific
join columns that appear in the WHERE clause of the query. If you followed the
guidelines in “Creating data distributions” on page 13-13, columns that head
indexes already have HIGH mode distributions.
v Determine whether HIGH mode distribution information about columns that do
not head indexes can provide a better execution path, take the following steps:

To determine if UPDATE STATISTICS HIGH on join columns might make a

difference:
1. Issue the SET EXPLAIN ON statement and rerun the query.
2. Note the estimated number of rows in the SET EXPLAIN output and the actual
number of rows that the query returns.
3. If these two numbers are significantly different, run UPDATE STATISTICS
HIGH on the columns that participate in joins, unless you have already done
so.

Important: If your table is very large, UPDATE STATISTICS with the HIGH mode
can take a long time to execute.

The following example shows a query that involves join columns:

SELECT employee.name, address.city
FROM employee, address
WHERE employee.ssn = address.ssn
AND employee.name = ’James’

In this example, the join columns are the ssn fields in the employee and address
tables. The data distributions for both of these columns must accurately reflect the
actual data so that the optimizer can correctly determine the best join plan and
execution order.

You cannot use the UPDATE STATISTICS statement to create data distributions for
a table that is external to the current database. For additional information about
data distributions and the UPDATE STATISTICS statement, see the IBM Informix
Guide to SQL: Syntax.

Updating statistics for columns with user-defined data types

Programmers can write functions that gather statistics for columns with
user-defined data types. You can store the data distributions for user-defined data
types in an sbspace.

Because information about the nature and use of a user-defined data type (UDT) is
not available to the database server, it cannot collect the colmin and colmax
column of the syscolumns system catalog table for user-defined data types. To
gather statistics for columns with user-defined data types, programmers must write

13-16 IBM Informix Performance Guide

 functions that extend the UPDATE STATISTICS statement. For more information,
see the performance chapter in IBM Informix User-Defined Routines and Data Types
Developer's Guide.

Because the data distributions for user-defined data types can be large, you can
optionally store them in an sbspace instead of the sysdistrib system catalog table.

To store data distributions for user-defined data types in an sbspace:

1. Use the onspaces -c -S command to create an sbspace.
To ensure recoverability of the data distributions, specify LOGGING=ON in the -Df
option, as the following sample shows:
% onspaces -c -S distrib_sbsp -p /dev/raw_dev1 -o 500 -s
20000
-m /dev/raw_dev2 500 -Ms 150 -Mo 200 -Df
"AVG_LO_SIZE=32,LOGGING=ON"
For information about sizing an sbspace, see “Estimating pages that smart large
objects occupy” on page 6-10.
For more information about specifying storage characteristics for sbspaces, see
“Configuration parameters that affect sbspace I/O” on page 5-20.
2. Specify the sbspace that you created in step 1 in the configuration parameter
SYSSBSPACENAME.
3. Specify the column with the user-defined data type when you run the UPDATE
STATISTICS statement with the MEDIUM or HIGH keywords to generate data
distributions.

To print the data distributions for a column with a user-defined data type, use the
dbschema -hd option.

Update statistics in parallel on very large databases

If you have an extremely large database and indexes are fragmented, UPDATE
STATISTICS LOW can automatically run statements in parallel.

To enable statements to automatically run in parallel, you must run UPDATE

STATISTICS LOW with the PDQ priority set to a value that is between 1 and 10. If
the PDQ priority is set to 1, 10 percent of the index fragments are analyzed at one
time for the current table. If the PDQ priority is set to 5, 50 percent of the index
fragments are analyzed at one time for the current table. If the PDQ priority is set
to 10, all fragments are analyzed at one time for the current table. (If the PDQ
priority is set to a value that is higher than 10, Informix operates as if the PDQ
priority is set to 10, analyzing all fragments at one time for the current table.)

If you run UPDATE STATISTICS MEDIUM or HIGH, you can set the PDQ priority
to a value that is higher than 10. Because the UPDATE STATISTICS MEDIUM and
HIGH statements perform a large amount of sorting operations, increasing the
PDQ priority to a value that is higher than 10 provides additional memory than
can improve the speed of the sorting operations.

Adjust the amount of memory and disk space for UPDATE

STATISTICS
When you execute the UPDATE STATISTICS statement, the database server uses
memory and disk to sort and construct data distributions. You can affect the
amount of memory and disk space available for UPDATE STATISTICS operations.

Chapter 13. Improving individual query performance 13-17

 You can affect the amount of memory and disk space available for UPDATE
STATISTICS with the following methods:
v PDQ priority
You can obtain more memory for sorting when you set PDQ priority greater
than 0. The default value for PDQ priority is 0. To set PDQ priority, use either
the PDQPRIORITY environment variable or the SQL statement SET
PDQPRIORITY.
For more information about PDQ priority, see “The allocation of resources for
parallel database queries” on page 12-7.
v DBUPSPACE environment variable
You can use the DBUPSPACE environment variable to specify the amount of
system disk space (and the amount of memory for sorting values) that UPDATE
STATISTICS MEDIUM or UPDATE STATISTICS HIGH statements can use in
each pass to construct column distributions. If you specify too small a value, the
database server instead uses enough space to write the largest column to disk.
For more information about this environment variable, see the IBM Informix
Guide to SQL: Reference.

Data sampling during update statistics operations

If you have a large b-tree index with more than 100 K leaf pages, you can generate
index statistics based on sampling when you run UPDATE STATISTICS statements
in LOW mode. The gathering of statistics through sampling can increase the speed
of the update statistics operation.

By default, when UPDATE STATISTICS statements run, the database server reads
all index leaf pages in sequence to gather statistics such as the number of leaf
pages, the number of unique lead key values, and cluster information. For a large
index this can take a long time. With sampling, the database server reads a fraction
of the index leaf pages (the sample) and then deduces index statistics based on
statistics gathered from the sample.

A possible trade-off for less time in gathering statistics is the accuracy of the
statistics gathered. If there are significant skews in the data distribution for the
lead index key, the sampling approach can result in a large error margin for the
statistics gathered, which in turn might affect optimizer decisions in query plan
generation.

You cannot control how much data is in the sample.

To enable or disable sampling, use the USTLOW_SAMPLE configuration parameter

or the USTLOW_SAMPLE environment option of the SET ENVIRONMENT
statement.
Related concepts:
USTLOW_SAMPLE environment option (SQL Syntax)
Related reference:
USTLOW_SAMPLE configuration parameter (Administrator's Reference)

Display data distributions

You can use the dbschema utility to display data distributions.

13-18 IBM Informix Performance Guide

Unless column values change considerably, you do not need to regenerate the data
distributions. To verify the accuracy of the distribution, compare dbschema -hd
output with the results of appropriately constructed SELECT statements.

For example, the following dbschema command produces a list of distributions for
each column of table customer in database vjp_stores with the number of values
in each bin, and the number of distinct values:
dbschema -hd customer -d vjp_stores

Figure 13-4 shows the data distributions for the column zipcode that this
dbschema -hd command produces. Because this column heads the zip_ix index,
UPDATE STATISTICS HIGH was run on it, as the following output line indicates:
High Mode, 0.500000 Resolution

Figure 13-4 shows 17 bins with one distinct zipcode value in each bin.

dbschema -hd customer -d vjp_stores

...
Distribution for virginia.customer.zipcode

Constructed on 09/18/2000

High Mode, 0.500000 Resolution

--- DISTRIBUTION ---

( 02135 )
1: ( 1, 1, 02135 )
2: ( 1, 1, 08002 )
3: ( 1, 1, 08540 )
4: ( 1, 1, 19898 )
5: ( 1, 1, 32256 )
6: ( 1, 1, 60406 )
7: ( 1, 1, 74006 )
8: ( 1, 1, 80219 )
9: ( 1, 1, 85008 )
10: ( 1, 1, 85016 )
11: ( 1, 1, 94026 )
12: ( 1, 1, 94040 )
13: ( 1, 1, 94085 )
14: ( 1, 1, 94117 )
15: ( 1, 1, 94303 )
16: ( 1, 1, 94304 )
17: ( 1, 1, 94609 )

--- OVERFLOW ---

1: ( 2, 94022 )
2: ( 2, 94025 )
3: ( 2, 94062 )
4: ( 3, 94063 )
5: ( 2, 94086 )

Figure 13-4. Displaying Data Distributions with dbschema -hd

The OVERFLOW portion of the output shows the duplicate values that might
skew the distribution data, so dbschema moves them from the distribution to a

Chapter 13. Improving individual query performance 13-19

 separate list. The number of duplicates in this overflow list must be greater than a
critical amount that the following formula determines. Figure 13-4 on page 13-19
shows a resolution value of .0050. Therefore, this formula determines that any
value that is duplicated more than one time is listed in the overflow section.
Overflow = .25 * resolution * number_rows
= .25 * .0050 * 28
= .035

For more information about the dbschema utility, see the IBM Informix Migration
Guide.

Improve performance by adding or removing indexes

You can often improve the performance of a query by adding or, in some cases,
removing indexes. You can also enable the optimizer to automatically fetch a set of
keys from an index buffer.

To improve the performance of a query, consider using some of the methods that
the following topics describe.

In addition:
v Consider using the CREATE INDEX ONLINE and DROP INDEX ONLINE
statements to create and drop an index in an online environment, when the
database and its associated tables are continuously available. These SQL
statements enable you to create and drop indexes without having an access lock
placed over the table during the duration of the index builds or drops. For more
information, see “Creating and dropping an index in an online environment” on
page 7-16.
v Set the BATCHEDREAD_INDEX configuration parameter to enable the
optimizer to automatically fetch a set of keys from an index buffer. This reduces
the number of times a buffer is read.
Related reference:
BATCHEDREAD_INDEX configuration parameter (Administrator's Reference)

Replace autoindexes with permanent indexes

If the query plan includes an autoindex path to a large table, you can generally
improve performance by adding an index on that column. If you perform a query
regularly, you can save time by creating a permanent index.

If you perform the query occasionally, you can reasonably let the database server
build and discard an index.

Use composite indexes

The optimizer can use a composite index (one that covers more than one column)
in several ways.

The database server can use an index on columns a, b, and c (in that order) in the
following ways:
v To locate a particular row
The database server can use a composite index when the first filter is an equality
filter and subsequent columns have range (<, <=, >, >=) expressions. The
following examples of filters use the columns in a composite index:

13-20 IBM Informix Performance Guide

 WHERE a=1
WHERE a>=12 AND a<15
WHERE a=1 AND b < 5
WHERE a=1 AND b = 17 AND c >= 40
The following examples of filters cannot use that composite index:
WHERE b=10
WHERE c=221
WHERE a>=12 AND b=15
v To replace a table scan when all of the desired columns are contained within the
index
A scan that uses the index but does not reference the table is called a key-only
search.
v To join column a, columns ab, or columns abc to another table
v To implement ORDER BY or GROUP BY on columns a, ab, or abc but not on b,
c, ac, or bc

Execution is most efficient when you create a composite index with the columns in
order from most to least distinct. In other words, the column that returns the
highest count of distinct rows when queried with the DISTINCT keyword in the
SELECT statement should come first in the composite index.

If your application performs several long queries, each of which contains ORDER
BY or GROUP BY clauses, you can sometimes improve performance by adding
indexes that produce these orderings without requiring a sort. For example, the
following query sorts each column in the ORDER BY clause in a different direction:
SELECT * FROM t1 ORDER BY a, b DESC;

To avoid using temporary tables to sort column a in ascending order and column b
in descending order, you must create a composite index on (a, b DESC) or on (a
DESC, b). You need to create only one of these indexes because of the
bidirectional-traverse capability of the database server. For more information about
bidirectional traverse, see the IBM Informix Guide to SQL: Syntax.

On the other hand, it can be less expensive to perform a table scan and sort the
results instead of using the composite index when the following criteria are met:
v Your table is well ordered relative to your index.
v The number of rows that the query retrieves represents a large percentage of the
available data.

Indexes for data warehouse applications

Many data warehouse databases use a star schema, which consists of a fact table
and a number of dimensional tables. Queries that use tables in a star schema or
snowflake schema can benefit from the proper index on the fact table.

The fact table is generally large and contains the quantitative or factual information
about the subject. A dimensional table describes an attribute in the fact table.

When a dimension needs lower-level information, the dimension is modeled by a

hierarchy of tables, called a snowflake schema.

Consider the example of a star schema with one fact table named orders and four
dimensional tables named customers, suppliers, products, and clerks. The orders
table describes the details of each sale order, which includes the customer ID,

Chapter 13. Improving individual query performance 13-21

 supplier ID, product ID, and sales clerk ID. Each dimensional table describes an ID
in detail. The orders table is large, and the four dimensional tables are small.

The following query finds the total direct sales revenue in the Menlo Park region
(postal code 94025) for hard drives supplied by the Johnson supplier:
SELECT sum(orders.price)
FROM orders, customers, suppliers,product,clerks
WHERE orders.custid = customers.custid
AND customers.zipcode = 94025
AND orders.suppid = suppliers.suppid
AND suppliers.name = ’Johnson’
AND orders.prodid = product.prodid
AND product.type = ’hard drive’
AND orders.clerkid = clerks.clerkid
AND clerks.dept = ’Direct Sales’

This query uses a typical star join, in which the fact table joins with all
dimensional tables on a foreign key. Each dimensional table has a selective table
filter.

An optimal plan for the star join is to perform a cartesian product on the four
dimensional tables and then join the result with the fact table. The following index
on the fact table allows the optimizer to choose the optimal query plan:
CREATE INDEX ON orders(custid,suppid,prodid,clerkid)

Without this index, the optimizer might choose to first join the fact table with a
single dimensional table and then join the result with the remaining dimensional
tables. The optimal plan provides better performance.

For more information about star schemas and snowflake schemas, see the IBM
Informix Database Design and Implementation Guide.

Configure B-tree scanner information to improve transaction

processing
You can improve the performance of transaction processing in logged databases by
controlling how the B-tree scanner threads remove deletions from indexes.

The B-tree scanner improves transaction processing for logged databases when
rows are deleted from a table with indexes. The B-tree scanner automatically
determines which index partitions will be cleaned, based on a priority list. B-tree
scanner threads remove deleted index entries and rebalance the index nodes. The
B-tree scanner automatically determines which index items are to be deleted.

In a logged database, when a delete or an update operation is performed on a row,

any corresponding index entry is not immediately deleted. Instead, the
corresponding index entry is flagged as deleted until a B-tree scanner thread scans
the index and removes the deleted items. An index containing many deleted items
can cause a significant performance problem, because index searches need to scan
a larger number of items before finding the first valid item.

The default setting for B-tree scanning provides the following type of scanning,
depending on your indexes:
v If the table has more than one attached index, the B-tree scanner uses the leaf
scan mode. Leaf scan mode is the only type of scanning possible with multiple
attached indexes.

13-22 IBM Informix Performance Guide

v If the table contains a single attached index or if the indexes are detached, the
B-tree scanner uses alice (adaptive linear index cleaning) mode. The initial alice
scan mode is optimized for small- to medium-sized systems with few or no
indexes above 1 GB. However, if the database server detects that the alice mode
is inefficient, the alice scan mode setting is adjusted automatically to
accommodate larger indexes.

Depending on your application and the order in which the system adds and
deletes keys from the index, the structure of an index can become inefficient.

You use the BTSCANNER configuration parameter to specify the following

information, which defines the scan mode:
v The number of B-tree scanner threads to start when the database server starts
The number of B-tree scanner threads is configurable to any positive number.
One B-tree scanner thread will always clean an individual index partition, so if
you occasionally or consistently have a higher number of index partitions
requiring cleaning, you might want to use more than one B-tree scanner thread.
At runtime, you can turn off any B-tree scanner activity by issuing an onmode
-C command. This command stops all B-tree scanner threads.
v The threshold, which is the minimum number of deleted items an index must
encounter before an index is placed on the priority list for eligibility for scanning
and cleaning by a B-tree scanner thread
For example, if you increase the threshold beyond 5000, you might be able to
avoid frequent B-tree scanner activity on the indexes that receive the most
updates and deletes.
v The range size, in kilobytes, that an index or index fragment must exceed before
the index is cleaned with range scanning
v An alice mode value
v The level at which B-tree scanner threads compress indexes by merging two
partially used index pages

The server treats a forest of trees index the same way it treats a B-tree index.
Therefore, in a logged database, you can control how the B-tree scanner threads
remove deletions from both forest of trees and B-tree indexes.

The following table summarizes the differences between the scan modes.
Table 13-3. Scan modes for B-tree scanner threads
Performance
Scan Mode Description Advantages or Issues More Information
Leaf scan mode In this mode, the leaf This mode is only “Leaf and range scan
level of an attached possible on attached mode settings” on
index is completely indexes and is the page 13-27
scanned for deleted only mode the server
items. can use if more than
one attached index
exists in a partition.

Chapter 13. Improving individual query performance 13-23

 Table 13-3. Scan modes for B-tree scanner threads (continued)
Performance
Scan Mode Description Advantages or Issues More Information
Alice (adaptive linear If the BTSCANNER You can greatly “Alice scan mode
index cleaning) scan alice option is improve performance values” on page 13-25
mode enabled, every index and reduce I/O when
partition receives a using the alice mode.
bitmap that tracks Generally, alice mode
where a deleted item is 64 times more
was found in the efficient than range
index. The scan that scanning and can
occurs excludes all automatically tune
parts of the index itself for
where no delete unsatisfactory
operations are found. indexes, which range
scanning cannot do.
The initial size and
granularity of these
bitmaps depend on
the size of the
partitions they
represent and the
current system-wide
alice level. The server
periodically checks
each bitmap for its
efficiency by checking
the ratio of pages to
be cleaned to pages
read, adjusting
scanning if necessary
to get better
information. This
mode allocates
additional resources
(memory) to the
index that is
consuming excess
I/O.

13-24 IBM Informix Performance Guide

Table 13-3. Scan modes for B-tree scanner threads (continued)
Performance
Scan Mode Description Advantages or Issues More Information
Range scan mode Range scanning, Not recommended “Leaf and range scan
which is enabled with for Informix Version mode settings” on
the rangesize option, 11.10 or higher. Alice page 13-27
is performed in the scanning is exactly
range between the the same as range
low and high page scanning, but is 64
address. The leaf times more efficient,
level of the index uses the same
partition is only resources, and has
scanned within this 128 equal ranges.
range. The server
performs light scans, When you set alice
which do not mode scanning, range
immediately use and scanning does not
strain the buffer pool, have an effect.
even though cleaning
occurs through the If you decide to use
buffer pool. range scanning for
systems with only a
lot of large indexes,
set the rangesize
option to the
minimum partition
size for range
scanning.

For more information about the BTSCANNER configuration parameter and for
more information about how the database server maintains an index tree, see the
chapter on configuration parameters and the chapter on disk structure and storage
in the IBM Informix Administrator's Reference.

Use the onstat -C option to monitor the B-tree scanner activities.

Use the onmode -C option to change the configuration of B-tree scanners during
runtime.

For more information about onstat -C and onmode -C, see the IBM Informix
Administrator's Reference.

Alice scan mode values

You enable alice (adaptive linear index cleaning) mode by setting the alice option
to any value between 1 and 12 (finest initial granularity). For small- to
medium-sized systems with few or no indexes above 1 gigabyte, set the alice
option to 6 or 7. For systems with large indexes, set alice to a higher mode.

When you set alice mode, the higher the mode, the more memory is used per
index partition. However, the memory used is not a huge amount. The advantage
is less I/O, as shown in the following table.
Table 13-4. Alice mode settings
Alice Mode Setting Memory or Block I/O
0 Turns off alice scanning.
1 Uses exactly 8 bytes of memory (no adjusting).

Chapter 13. Improving individual query performance 13-25

 Table 13-4. Alice mode settings (continued)
Alice Mode Setting Memory or Block I/O
2 Uses exactly 16 bytes of memory (no adjusting).
3 Each block of pages will need 512 I/O operations for
cleaning.
4 Each block of pages will need 256 I/O operations for
cleaning.
5 Each block of pages will need 128 I/O operations for
cleaning.
6 (default) Each block of pages will need 64 I/O operations for
cleaning.
7 Each block of pages will need 32 I/O operations for
cleaning.
8 Each block of pages will need 16 I/O operations for
cleaning.
9 Each block of pages will need 8 I/O operations for cleaning.
10 Each block of pages will need 4 I/O operations for cleaning.
11 Each block of pages will need 2 I/O operations for cleaning.
12 Each block of pages will need 1 I/O operations for cleaning.

When you set the alice mode, you need to consider memory usage versus I/O. The
lower the alice mode setting, the less memory the index will use. The higher the
alice mode setting, the more memory the index will use. 12 is the highest mode
value, because it is a direct mapping of a single bit of memory to each instance of
I/O.

Suppose you have an online page size of 2 KB and the default B-Tree Scanner I/O
size of 256 pages. If you set the alice mode to 6, each byte of memory can
represent 131,072 index pages (256 MB). If you set the mode to 10, each byte of
memory can represent 8,192 index pages (16 MB). Thus, changing the mode setting
from 6 to 10 requests 16 times the memory, but requires 16 times less I/O.

If you have an index partition that uses 1 GB, then an alice mode setting of 6
would take 4 bytes of memory, while an alice mode setting of 10 would consume
64 bytes of memory, as shown in this formula:
( {mode block size} io per bit * 8 bits per byte * 256 page per io )

Setting the alice mode to a value between 3 and 12 sets the initial amount of
memory that is used for index cleaning. Subsequently, the B-tree scanners
automatically adjust the mode based on the efficiency of past cleaning operations.

For example, if after five scans (by default), the I/O efficiency is below 75 percent,
the server automatically adjusts to the next alice mode if you set the mode to a
value above 2. For example, if an index is currently operating in alice mode 6, a
B-tree scanner has cleaned the index at least 5 times, and the I/O efficiency is
below 75 percent, the server automatically adjusts to mode 7, the next higher
mode. This doubles the memory required, but reduces the I/O by a factor of 2.

The server will re-evaluate the index after five more scans to determine the I/O
efficiency again, and will continue to do this until mode 12. The server stops
making adjustments at mode 12.

13-26 IBM Informix Performance Guide

The following example sets the alice mode to 6:
BTSCANNER num=2,threshold=10000,alice=6,compression=default

Leaf and range scan mode settings

If a table has more than one attached index, the B-tree scanner uses the leaf scan
mode. If you want small indexes to be scanned by the leaf scan method, set the
rangesize option of the BTSCANNER configuration parameter to 100.

If you decide to enable range scan mode when a single index exists in the
partition, set rangesize option of the BTSCANNER configuration parameter to the
minimum size that a partition must have to be scanned using this mode. Specify
the size in kilobytes.

The following example specifies that:

v The server will start two B-tree scanner threads.
v The server will consider cleaning indexes in the hot list (a list of indexes that
caused the server to do extra work) when 50000 deleted items are found in the
index.
v Indexes with a partition size that is equal to or larger than 100 KB will be
cleaned using the range scan mode.
v Indexes with a partition size of less than 100 KB will be cleaned using the leaf
scan mode.
v Index compression is set at the medium (default) level
BTSCANNER num=2,threshold=50000,rangesize=100,compression=default

B-tree scanner index compression levels and transaction

processing performance
B-tree scanner threads compress indexes by merging two partially used index
pages if the amount of data on those pages is below the level that is specified by
the compression option. You can set the compression level to control the amount of
I/O required to find and load data.

B-tree scanner threads look for index pages that can be compressed because they
are below the specified level. The B-tree scanner can compress index pages with
deleted items and pages that do not have deleted items.

By default, a B-tree scanner compresses at the medium level. The following table
provides information about the performance benefits and trade-offs if you change
the compression level to high or low.
Table 13-5. B-Tree Scanner Compression Level Benefits and Trade-offs
Compression
Level Performance Benefits and Trade-offs When to Use
Low The low compression level is You might want to change the
beneficial for an index that is compression level to low if you
expected to grow quickly, with expect an index to grow quickly with
frequent B-tree node splits. When the frequent splits.
compression level is set to low, the
B-tree index will not require as many
splits as indexes with medium or
high compression levels, because
more free space remains in the B-tree
nodes.

Chapter 13. Improving individual query performance 13-27

 Table 13-5. B-Tree Scanner Compression Level Benefits and Trade-offs (continued)
Compression
Level Performance Benefits and Trade-offs When to Use
High In general, if an index is read-only or You might want to change the
90 percent of it is read-only, the high compression level to high under
compression level is beneficial these circumstances:
because searching for data will v If an index is read most of the
require fewer pages (and less I/O) to time, and delete and insert
traverse. Examples might be indexes operations occur a small
that do not have frequent changes or percentage of the time.
indexes undergoing batch (block)
v If tables are loaded and updated in
delete operations.
a batch and are kept for a period
Using high level of compression also of time as read-only tables.
means a performance trade-off,
because it takes more I/O to
compress the index more
aggressively. Select operations will
have less I/O when the compression
level is high.

If you do not need to change the compression level to high or low, set the
compression option of the BTSCANNER configuration parameter to med or
default.

Index Compression and the Index Fill Factor

In addition to the compression option that specifies when to attempt to join two
partially used pages, you can use the FILL FACTOR configuration parameter to
control when to add new index pages. The index fill factor, which you define with
the FILLFACTOR configuration parameter or the FILLFACTOR option of the
CREATE INDEX statement, is a percentage of each index page that will be filled
during the index build.

Setting the level for B-tree scanner compression of indexes

Informix provides several ways to specify the level at which B-tree scanner threads
will compress indexes pages. To optimize space and transaction processing, you
can lower the compression level if your indexes grow quickly. You can increase the
level if your indexes have few delete and insert operations or if batch updates are
performed.

Prerequisites:
v Determine if adjusting the level for index compression will improve
performance.
v Get statistics on the number of rows read, deleted, and inserted by running the
onstat -g ppf command. You can also view information in the sysptprof table.
v Analyze the statistics to determine if you want to change the threshold.
For information about compression levels and the circumstances under which you
might want to change the level, see “B-tree scanner index compression levels and
transaction processing performance” on page 13-27.

Specify the compression level for an instance with any of the following options:
v Set the compression field of the BTSCANNER configuration parameter to low,
med (medium), high, or default. (The system default value is med.)

13-28 IBM Informix Performance Guide

 v Run the onmode -C compression value command, where value is low, med
(medium), high, and default. The system default value is med.
v Run an SQL administration API function with this command:
SET INDEX COMPRESSION, partition number, compression level

Examples

Set the compression option of the BTSCANNER configuration parameter to

default as follows:
BTSCANNER num=4,threshold=10000,rangesize=-1,alice=6,compression=default

Set the compression option of the BTSCANNER configuration parameter to high as

follows:
BTSCANNER num=4,threshold=5000,compression=high

Specify the compression level using onmode -C, as follows:

onmode –C compression high

Run either of the following SQL administration API functions to set the
compression level for a single fragment of the index that has the partition number
1048960:
EXECUTE FUNCTION TASK("SET INDEX COMPRESSION", 1048960, "DEFAULT");
EXECUTE FUNCTION ADMIN("SET INDEX COMPRESSION", 1048960, "LOW");

Run the following SELECT statement to execute the task function over all index
fragments. This command sets the compression level for all fragments of an index
named idx1 in a database named db1.
SELECT sysadmin:TASK("SET INDEX COMPRESSION", partnum, "MED")
FROM sysmaster:systabnames
WHERE dbsname = ’dbs1’ AND tabname = ’idx1’;

You can also run the following SELECT TASK statement to execute the task
function over all index fragments and set the compression level for all fragments.
SELECT TASK("SET INDEX COMPRESSION", partn, "MED")
FROM dbs1:systables t, dbs1:sysfragments f
WHERE f.tabid = t.tabid AND f.fragtype = ’I’ AND indexname =’idx1’;

Determine the amount of free space in an index page

You can use the oncheck -pT command to determine the amount of free space in
each index page.

If your table has relatively low update activity and a large amount of free space
exists, you might want to drop and re-create the index with a larger value for
FILLFACTOR to make the unused disk space available.

Optimizer estimates of distributed queries

The optimizer assumes that access to a row from a remote database takes longer
than access to a row in a local database. The optimizer estimates include the cost
of retrieving the row from disk and transmitting it across the network.

For an example of this higher estimated cost, see “The query plan of a distributed
query” on page 13-30.

Chapter 13. Improving individual query performance 13-29

 Buffer data transfers for a distributed query
Informix uses several factors to determine the size of the buffer that sends and
receives data to and from a remote server.

The server uses the following factors to determine the buffer size:
v The row size
The database server calculates the row size by summing the average move size
(if available) or the length (from the syscolumns system catalog table) of the
columns.
v The setting of the FET_BUF_SIZE environment variable on the client
You might be able to reduce the size and number of data transfers by using the
FET_BUF_SIZE environment variable to increase the size of the buffer that the
database server uses to send and receive rows to and from the remote database
server.
The minimum buffer size is 1024 or 2048 bytes, depending on the row size. If
the row size is larger than either 1024 or 2048 bytes, the database server uses the
FET_BUF_SIZE value.
For more information about the FET_BUF_SIZE environment variable, see the
IBM Informix Guide to SQL: Reference.

The query plan of a distributed query

You can display the chosen query plan of a distributed query. The information
displayed for a distributed join differs from information displayed for a local join.

The following figure shows the chosen query plan for the distributed query.

QUERY:
------
select l.customer_num, l.lname, l.company,
l.phone, r.call_dtime, r.call_descr
from customer l,
vjp_stores@gilroy:cust_calls r
where l.customer_num = r.customer_num

Estimated Cost: 9
Estimated # of Rows Returned: 7

1) informix.r:
REMOTE PATH

2) informix.l: INDEX PATH

(1) Index Keys: customer_num (Serial, fragments: ALL)

Lower Index Filter: informix.l.customer_num = informix.r.customer_num
NESTED LOOP JOIN

Figure 13-5. Selected Output of SET EXPLAIN ALL for Distributed Query, Part 3

The following table shows the main differences between the chosen query plans for
the distributed join and the local join.

Output Line in Figure 13-5 for Output Line in Figure 11-1 on page
Distributed Query 11-9 for Local-Only Query Description of Difference
vjp_stores@gilroy: virginia.cust_calls informix.cust_calls The remote table name is prefaced
with the database and server
names.

13-30 IBM Informix Performance Guide

Output Line in Figure 13-5 on page Output Line in Figure 11-1 on page
13-30 for Distributed Query 11-9 for Local-Only Query Description of Difference
Estimated Cost: 9 Estimated Cost: 7 The optimizer estimates a higher
cost for the distributed query.
informix.r: REMOTE PATH informix.r: SEQUENTIAL SCAN The optimizer chose to keep the
outer, remote cust_calls table at
the remote site.
select x0.call_dtime,x0.call_descr,x0. The SQL statement that the local
customer_num from database server sends to the
vjp_stores:”virginia”.cust_ calls x0 remote site. The remote site
reoptimizes this statement to
choose the actual plan.

Improve sequential scans

You can improve the performance of sequential read operations on large tables by
eliminating repeated sequential scans.

Sequential access to a table other than the first table in the plan is ominous because
it threatens to read every row of the table once for every row selected from the
preceding tables.

If the table is small, it is harmless to read it repeatedly because the table resides
completely in memory. Sequential search of an in-memory table can be faster than
searching the same table through an index, especially if maintaining those index
pages in memory pushes other useful pages out of the buffers.

When the table is larger than a few pages, however, repeated sequential access
produces poor performance. One way to prevent this problem is to provide an
index to the column that is used to join the table.

Any user with the Resource privilege can build additional indexes. Use the
CREATE INDEX statement to make an index.

An index consumes disk space proportional to the width of the key values and the
number of rows. (See “Estimating index pages” on page 7-4.) Also, the database
server must update the index whenever rows are inserted, deleted, or updated; the
index update slows these operations. If necessary, you can use the DROP INDEX
statement to release the index after a series of queries, which frees space and
makes table updates easier.

Enable view folding to improve query performance

You can significantly improve the performance of a query that involves a view by
enabling view folding.

You do this by setting the IFX_FOLDVIEW configuration parameter to 1.

When view folding is enabled, views are folded into a parent query. Because the
views are folded into the parent query, the query results are not placed in a
temporary table.

You can use view folding in the following types of queries:

Chapter 13. Improving individual query performance 13-31

 v Views that contain a UNION ALL clause and the parent query has a regular
join, an Informix join, an ANSI join, or an ORDER BY clause

View folding does not occur for the following types of queries that perform a
UNION ALL operation involving a view:
v The view has one of the following clauses: AGGREGATE, GROUP BY, ORDER
BY, UNION, DISTINCT, or OUTER JOIN (either Informix or ANSI type).
v The parent query has a UNION or UNION ALL clause.
In these situations, a temporary table is created to hold query results.

Reduce the join and sort operations

After you understand what the query is doing, you can look for ways to obtain the
same output with less effort.

The following suggestions can help you rewrite your query more efficiently:
v Avoid or simplify sort operations.
v Use parallel sorts.
v Use temporary tables to reduce sorting scope.

Avoid or simplify sort operations

In many situations you can determine how to avoid or reduce frequent or complex
sort operations.

The sort algorithm is highly tuned and extremely efficient. It is as fast as any
external sort program that you might apply to the same data. You do not need to
avoid infrequent sorts or sorts of relatively small numbers of output rows.

However, you should try to avoid or reduce the scope of repeated sorts of large
tables. The optimizer avoids a sort step whenever it can use an index to produce
the output in its proper order automatically. The following factors prevent the
optimizer from using an index:
v One or more of the ordered columns is not included in the index.
v The columns are named in a different sequence in the index and the ORDER BY
or GROUP BY clause.
v The ordered columns are taken from different tables.

For another way to avoid sorts, see “Use temporary tables to reduce sorting scope”
on page 13-33.

If a sort is necessary, look for ways to simplify it. As discussed in “Sort-time costs”
on page 10-24, the sort is quicker if you can sort on fewer or narrower columns.
Related concepts:
“Ordering with fragmented indexes” on page 13-37

Use parallel sorts

When you cannot avoid sorting, the database server takes advantage of multiple
CPU resources to perform the required sort-and-merge operations in parallel. The
database server can use parallel sorts for any query, not just PDQ queries. You can
control the number of threads that the database server uses to sort rows.

13-32 IBM Informix Performance Guide

 To control the number of threads that the database server uses to sort rows, use the
PSORT_NPROCS environment variable.

When PDQ priority is greater than 0 and PSORT_NPROCS is greater than 1, the
query benefits both from parallel sorts and from PDQ features such as parallel
scans and additional memory. Users can use the PDQPRIORITY environment
variable to request a specific proportion of PDQ resources for a query. You can use
the MAX_PDQPRIORITY configuration parameter to limit the number of such user
requests. For more information, see “Limiting PDQ resources in queries” on page
3-11.

In some cases, the amount of data being sorted can overflow the memory resources
allocated to the query, resulting in I/O to a dbspace or sort file. For more
information, see “Configure dbspaces for temporary tables and sort files” on page
5-8.

Use temporary tables to reduce sorting scope

You can use a temporary, ordered subset of a table to increase the speed of a query.
The temporary table can also simplify the work of the query optimizer, cause the
optimizer to avoid multiple-sort operations, and simplify the work of the optimizer
in other ways.

For example, suppose your application produces a series of reports on customers

who have outstanding balances, one report for each major postal area, ordered by
customer name. In other words, a series of queries occurs, each of the following
form (using hypothetical table and column names):
SELECT cust.name, rcvbles.balance, ...other columns...
FROM cust, rcvbles
WHERE cust.customer_id = rcvbles.customer_id
AND rcvbls.balance > 0
AND cust.postcode LIKE ’98_ _ _’
ORDER BY cust.name

This query reads the entire cust table. For every row with the specified postal
code, the database server searches the index on rcvbles.customer_id and performs
a nonsequential disk access for every match. The rows are written to a temporary
file and sorted. For more information about temporary files, see “Configure
dbspaces for temporary tables and sort files” on page 5-8.

This procedure is acceptable if the query is performed only once, but this example
includes a series of queries, each incurring the same amount of work.

An alternative is to select all customers with outstanding balances into a temporary

table, ordered by customer name, as the following example shows:
SELECT cust.name, rcvbles.balance, ...other columns...
FROM cust, rcvbles
WHERE cust.customer_id = rcvbles.customer_id
AND cvbls.balance > 0
INTO TEMP cust_with_balance

You can then execute queries against the temporary table, as the following example
shows:
SELECT *
FROM cust_with_balance
WHERE postcode LIKE ’98_ _ _’
ORDER BY cust.name

Chapter 13. Improving individual query performance 13-33

 Each query reads the temporary table sequentially, but the table has fewer rows
than the primary table.

Configuring memory for queries with hash joins, aggregates,

and other memory-intensive elements
Certain configuration parameters can be set to provide more memory for queries
that require sorting, hash joins, aggregates, and other memory-intensive elements.

How you configure the amount of memory that is available for a query depends
on whether or not the query is a Parallel Database Query (PDQ).

Configuring memory for non-PDQ queries

If the PDQ priority is set to 0 (zero), you can change the amount of memory that is
available for a query that is not a PDQ query by changing the setting of the
DS_NONPDQ_QUERY_MEM configuration parameter. You can only use this
parameter if the PDQ priority is set to zero. Its setting has no effect if the PDQ
priority is greater than zero.

You can also change the value of DS_NONPDQ_QUERY_MEM with an onmode

-wm or onmode -wf command.

For example, if you use the onmode utility, specify a value as shown in the
following example:
onmode -wf DS_NONPDQ_QUERY_MEM=500

The minimum value for DS_NONPDQ_QUERY_MEM is 128 kilobytes. The

maximum supported value is 25 percent of DS_TOTAL_MEMORY. 128 kilobytes is
the default value of DS_NONPDQ_QUERY_MEM. If you specify a value for the
DS_NONPDQ_QUERY_MEM parameter, determine and adjust the value based on
the number and size of table rows involved in the query.

Informix might recalculate the value of DS_NONPDQ_QUERY_MEM initialization

if the value is more than 25 percent of the DS_TOTAL_MEMORY value.

If Informix changes the value that you set, the server sends a message in this
format:

DS_NONPDQ_QUERY_MEM recalculated and changed from old_value Kb to new_value

Kb.

In the message, old_value represents the value that you assigned to

DS_NONPDQ_QUERY_MEM in the user configuration file, and new_value
represents the value determined by Informix.

For formulas for estimating the amount of additional space to allocate for hash
joins, see “Estimating temporary space for dbspaces and hash joins” on page 5-12.

Configuring memory for PDQ queries

The Memory Grant Manager (MGM) component of Informix coordinates the use of
memory, CPU virtual processors (VPs), disk I/O, and scan threads among
decision-support queries. The MGM uses the DS_MAX_QUERIES,
DS_TOTAL_MEMORY, DS_MAX_SCANS, and MAX_PDQPRIORITY configuration
parameter settings to determine the quantity of these PDQ resources that can be

13-34 IBM Informix Performance Guide

 granted to a decision-support query. The MGM also grants memory to a query for
such activities as hash joins. For more information about the MGM, see “The
Memory Grant Manager” on page 12-6.

Optimize user-response time for queries

You can influence the amount of time that Informix takes to optimize a query and
to return rows to a user.

Optimization level
You normally obtain optimum overall performance with the default optimization
level, HIGH. The time that it takes to optimize the statement is usually
unimportant. However, if experimentation with your application reveals that your
query is still taking too long, you can set the optimization level to LOW.

If you change the optimization level to LOW, check the SET EXPLAIN output to
see if the optimizer chose the same query plan as before.

To specify a HIGH or LOW level of database server optimization, use the SET
OPTIMIZATION statement.
Related reference:
SET OPTIMIZATION statement (SQL Syntax)

Optimization goals
Optimizing total query time and optimizing user-response time are two
optimization goals for improving query performance.

Total query time is the time it takes to return all rows to the application. Total
query time is most important for batch processing or for queries that require all
rows be processed before returning a result to the user, as in the following query:
SELECT count(*) FROM orders
WHERE order_amount > 2000;

User-response time is the time that it takes for the database server to return a
screen full of rows back to an interactive application. In interactive applications,
only a screen full of data can be requested at one time. For example, the user
application can display only 10 rows at one time for the following query:
SELECT * FROM orders
WHERE order_amount > 2000;

Which optimization goal is more important can have an effect on the query path
that the optimizer chooses. For example, the optimizer might choose a nested-loop
join instead of a hash join to execute a query if user-response time is most
important, even though a hash join might result in a reduction in total query time.

Specifying the query performance goal

You can optimize user response time for your entire database server system, within
a session, or for individual queries.

The default behavior is for the optimizer to choose query plans that optimize the
total query time. You can specify optimization of user-response time at several
different levels:
v For the database server system

Chapter 13. Improving individual query performance 13-35

 To optimize user-response time, set the OPT_GOAL configuration parameter to
0, as in the following example:
OPT_GOAL 0
Set OPT_GOAL to -1 to optimize total query time.
v For the user environment
The OPT_GOAL environment variable can be set before the user application
starts.
UNIX Only
To optimize user-response time, set the OPT_GOAL environment
variable to 0, as in the following sample commands:
Bourne shell OPT_GOAL = 0
export OPT_GOAL

C shell setenv OPT_GOAL 0

For total-query-time optimization, set the OPT_GOAL environment variable to
-1.
v Within the session
You can control the optimization goal with the SET OPTIMIZATION statement
in SQL. The optimization goal set with this statement stays in effect until the
session ends or until another SET OPTIMIZATION statement changes the goal.
The following statement causes the optimizer to choose query plans that favor
total-query-time optimization:
SET OPTIMIZATION ALL_ROWS
The following statement causes the optimizer to choose query plans that favor
user-response-time optimization:
SET OPTIMIZATION FIRST_ROWS
v For individual queries
You can use FIRST_ROWS and ALL_ROWS optimizer directives to instruct the
optimizer which query goal to use. For more information about these directives,
see “Optimization-goal directives” on page 11-7.

The precedence for these levels is as follows:

v Optimizer directives
v SET OPTIMIZATION statement
v OPT_GOAL environment variable
v OPT_GOAL configuration parameter

For example, optimizer directives take precedence over the goal that the SET
OPTIMIZATION statement specifies.

Preferred query plans for user-response-time optimization

When the optimizer chooses query plans to optimize user-response time, it
computes the cost for retrieving the first row in the query for each plan and
chooses the plan with the lowest cost. In some cases, the query plan with the
lowest cost for retrieving the first row might not be the optimal path to retrieve all
rows in the query.

The following sections explain some of the possible differences in query plans.

13-36 IBM Informix Performance Guide

Nested-loop joins versus hash joins:

Hash joins generally have a higher cost to retrieve the first row than nested-loop
joins do. The database server must build the hash table before it retrieves any
rows. However, in some cases, total query time is faster if the database server uses
a hash join.

In the following example, tab2 has an index on col1, but tab1 does not have an
index on col1. When you execute SET OPTIMIZATION ALL_ROWS before you
run the query, the database server uses a hash join and ignores the existing index,
as the following portion of SET EXPLAIN output shows:
QUERY:
------
SELECT * FROM tab1,tab2
WHERE tab1.col1 = tab2.col1
Estimated Cost: 125
Estimated # of Rows Returned: 510
1) lsuto.tab2: SEQUENTIAL SCAN
2) lsuto.tab1: SEQUENTIAL SCAN
DYNAMIC HASH JOIN
Dynamic Hash Filters: lsuto.tab2.col1 = lsuto.tab1.col1

However, when you execute SET OPTIMIZATION FIRST_ROWS before you run
the query, the database server uses a nested-loop join. The clause (FIRST_ROWS
OPTIMIZATION) in the following partial SET EXPLAIN output shows that the
optimizer used user-response-time optimization for the query:
QUERY: (FIRST_ROWS OPTIMIZATION)
------
SELECT * FROM tab1,tab2
WHERE tab1.col1 = tab2.col1
Estimated Cost: 145
Estimated # of Rows Returned: 510
1) lsuto.tab1: SEQUENTIAL SCAN
2) lsuto.tab2: INDEX PATH
(1) Index Keys: col1
Lower Index Filter: lsuto.tab2.col1 = lsuto.tab1.col1
NESTED LOOP JOIN

Table scans versus index scans:

In cases where the database server returns a large number of rows from a table,
the lower-cost option for the total-query-time goal might be to scan the table
instead of using an index. However, to retrieve the first row, the lower-cost option
for the user-response-time goal might be to use the index to access the table.

Ordering with fragmented indexes:

When an index is not fragmented, the database server can use the index to avoid a
sort. However, when an index is fragmented, the ordering can be guaranteed only
within the fragment, not between fragments.

Usually, the least expensive option for the total-query-time goal is to scan the
fragments in parallel and then use the parallel sort to produce the proper ordering.
However, this option does not favor the user-response-time goal.

Instead, if the user-response time is more important, the database server reads the
index fragments in parallel and merges the data from all of the fragments. No
additional sort is generally needed.
Related concepts:

Chapter 13. Improving individual query performance 13-37

 “Avoid or simplify sort operations” on page 13-32

Optimize queries for user-defined data types

Queries that access user-defined data types (UDTs) can take advantage of the same
performance features that built-in data types use.

These features are:

v Indexes
If a query accesses a small number of rows, an index speeds retrieval because
the database server does not need to read every page in a table to find the rows.
For more information, see “Indexes on user-defined data types” on page 7-21.
v Parallel database query (PDQ)
Queries that access user-defined data can take advantage of parallel scans and
parallel execution.
To turn on parallel execution for a query, set the PDQPRIORITY environment
variable or use the SQL statement SET PDQPRIORITY. For more information
about how to set PDQ priority and configuration parameters that affect PDQ,
see “The allocation of resources for parallel database queries” on page 12-7.
v Optimizer directives

In addition, programmers can write the following functions or UDRs to help the
optimizer create an efficient query plan for your queries:
v Parallel UDRs that can take advantage of parallel database queries
v User-defined selectivity functions that calculate the expected fraction of rows
that qualify for the function
v User-defined cost functions that calculate the expected relative cost to execute a
user-defined routine
v User-defined statistical functions that the UPDATE STATISTICS statement can
use to generate statistics and data distributions
v User-defined negator functions to allow more choices for the optimizer

The following sections summarize these techniques. For a more complete

description of how to write and register user-defined selectivity functions and
user-defined cost functions, see IBM Informix User-Defined Routines and Data Types
Developer's Guide.

Parallel UDRs
One way to execute UDRs is in an expression in a query. You can take advantage
of parallel execution if the UDR is in an expression in the query.

For parallel execution, the UDR must be in one of the following parts of a query:
v WHERE clause
v SELECT list
v GROUP by list
v Overloaded comparison operator
v User-defined aggregate
v HAVING clause
v Select list for a parallel insertion statement
v Generic B-tree index scan on multiple index fragments if the compare function
used in the B-tree index scan is parallelizable

13-38 IBM Informix Performance Guide

 For example, suppose that you create an opaque data type circle, a table cir_t that
defines a column of type circle, a user-defined routine area, and then run the
following sample query:
SELECT circle, area(circle)
FROM cir_t
WHERE circle > "(6,2,4)";

In this sample query, the following operations can run in parallel:

v The UDR area(circle) in the SELECT list
If the table cir_t is fragmented, multiple area UDRs can run in parallel, one UDR
on each fragment.
v The expression circle > "(6,2,4)" in the WHERE clause
If the table cir_t is fragmented, multiple scans of the table can run in parallel,
one scan on each fragment. Then multiple “>” comparison operators can run in
parallel, one operator per fragment.

By default, a UDR does not run in parallel. To enable parallel execution of UDRs,
you must take the following actions:
v Specify the PARALLELIZABLE modifier in the CREATE FUNCTION or ALTER
FUNCTION statement.
v Ensure that the UDR does not call functions that are not PDQ thread-safe.
v Turn on PDQ priority.
v Use the UDR in a parallel database query.

Selectivity and cost functions

You can use the CREATE FUNCTION statement to create a UDR. Then, you can
use routine modifiers to change the cost or selectivity that is specified in the
statement.

After you create a UDR, you can place it in an SQL statement.

The following example shows how you can place a UDR in an SQL statement:
SELECT * FROM image
WHERE get_x1(image.im2) and get_x2(image.im1)

The optimizer cannot accurately evaluate the cost of executing a UDR without
additional information. You can provide the cost and selectivity of the function to
the optimizer. The database server uses cost and selectivity together to determine
the best path. For more information about selectivity, see “Filters with user-defined
routines” on page 13-2.

In the previous example, the optimizer cannot determine which function to execute
first, the get_x1 function or the get_x2 function. If a function is expensive to
execute, the DBA can assign the function a larger cost or selectivity, which can
influence the optimizer to change the query plan for better performance. In the
previous example, if get_x1 costs more to execute, the DBA can assign a higher
cost to the function, which can cause the optimizer to execute the get_x2 function
first.

You can add the following routine modifiers to the CREATE FUNCTION statement
to change the cost or selectivity that the optimizer assigns to the function:
v selfunc=function_name
v selconst=integer

Chapter 13. Improving individual query performance 13-39

 v costfunc=function_name
v percall_cost=integer

For more information about cost or selectivity modifiers, see the IBM Informix
User-Defined Routines and Data Types Developer's Guide.

User-defined statistics for UDTs

Because information about the nature and use of a user-defined data type (UDT) is
not available to the database server, it cannot collect distributions or the colmin
and colmax information (found in the syscolumns system catalog table) for a UDT.
Instead, you can create a special function that populates these statistics.

The database server runs the statistics collection function when you execute
UPDATE STATISTICS.

For more information about the importance of updating statistics, see “Statistics
held for the table and index” on page 10-20. For information about improving
performance, see “Updating statistics for columns with user-defined data types” on
page 13-16.

Negator functions
A negator function takes the same arguments as its companion function, in the same
order, but returns the Boolean complement. That is, if a function returns TRUE for a
given set of arguments, its negator function returns FALSE when passed the same
arguments, in the same order.

In certain cases, the database server can process a query more efficiently if the
sense of the query is reversed. That is, “Is x greater than y?” changes to “Is y less
than or equal to x?”

Optimize queries with the SQL statement cache

Before the database server runs an SQL statement, it must first parse and optimize
the statement. Optimizing statements can be time consuming, depending on the
size of the SQL statement.

The database server can store the optimized SQL statement in the virtual portion
of shared memory, in an area that is called the SQL statement cache. The SQL
statement cache (SSC) can be accessed by all users, and it allows users to bypass
the optimize step before they run the query. This capability can result in the
following significant performance improvements:
v Reduced response times when users are running the same SQL statements.
SQL statements that take longer to optimize (usually because they include many
tables and many filters in the WHERE clause) run faster from the SQL statement
cache because the database server does not optimize the statement.
v Reduced memory usage because the database server shares query data
structures among users.
Memory reduction with the SQL statement cache is greater when a statement
has many column names in the select list.

For more information about the effect of the SQL statement cache on the
performance of the overall system, see “Monitor and tune the SQL statement
cache” on page 4-25.

13-40 IBM Informix Performance Guide

When to use the SQL statement cache
Applications might benefit from use of the SQL statement cache if multiple users
execute the same SQL statements. The database server considers statements to be
the same if all characters match exactly.

For example, if 50 sales representatives execute the add_order application

throughout the day, they all execute the same SQL statements if the application
contains SQL statements that use host variables, such as the following example:
SELECT * FROM ORDERS WHERE order_num = :hostvar

This kind of application benefits from use of the SQL statement cache because
users are likely to find the SQL statements in the SQL statement cache.

The database server does not consider the following SQL statements exact matches
because they contain different literal values in the WHERE clause:
SELECT * FROM customer, orders
WHERE customer.customer_num = orders.customer_num
AND order_date > "01/01/07"
SELECT * FROM customer, orders
WHERE customer.customer_num = orders.customer_num
AND order_date > "01/01/2007"

Performance does not improve with the SQL statement cache in the following
situations:
v If a report application is run once nightly, and it executes SQL statements that
no other application uses, it does not benefit from use of the statement cache.
v If an application prepares a statement and then executes it many times,
performance does not improve with the SQL statement cache because the
statement is optimized just once during the PREPARE statement.

When a statement contains host variables, the database server replaces the host
variables with placeholders when it stores the statement in the SQL statement
cache. Therefore, the statement is optimized without the database server having
access to the values of the host variables. In some cases, if the database server had
access to the values of the host variables, the statement might be optimized
differently, usually because the distributions stored for a column inform the
optimizer exactly how many rows pass the filter.

If an SQL statement that contains host variables performs poorly with the SQL
statement cache turned on, try flushing the SQL statement cache with the onmode
-e flush command and running the query with values that are more frequently
used across multiple executions of the query. When you flush the cache, the
database server reoptimizes the query and generates a query plan that is optimized
for these frequently used values.

Important: The database server flushes an entry from the SQL statement cache
only if it is not in use. If an application prepares the statement and keeps it, the
entry is still in use. In this case, the application needs to close the statement before
the flush is beneficial.

Using the SQL statement cache

The DBA usually makes the decision to enable the SQL statement cache. If the SQL
statement cache is enabled, individual users can decide whether or not to use the
SQL statement cache for their specific environment or application.

Chapter 13. Improving individual query performance 13-41

 The database server incurs some processing overhead in managing the SQL
statement cache, so you should use the SQL statement cache only when multiple
users want to share the SQL statements.

To enable the SQL statement cache, set the STMT_CACHE configuration parameter
to a value that defines either of the following modes:
v Always use the SQL statement cache unless a user explicitly specifies do not use
the cache.
v Use the SQL statement cache only when a user explicitly specifies use it.

For more information, see “Enabling the SQL statement cache.” For more
information about the STMT_CACHE configuration parameter, see the IBM
Informix Administrator's Reference.

Enabling the SQL statement cache

The database server does not use the SQL statement cache when the
STMT_CACHE configuration parameter is 0 (the default value). You can change
this value to enable the SQL statement cache in one of two modes.

Use one of the following methods to change this STMT_CACHE default value:
v Update the ONCONFIG file to specify the STMT_CACHE configuration
parameter and restart the database server.
If you set the STMT_CACHE configuration parameter to 1, the database server
uses the SQL statement cache for an individual user when the user sets the
STMT_CACHE environment variable to 1 or executes the SET STATEMENT
CACHE ON statement within an application.
STMT_CACHE 1
If the STMT_CACHE configuration parameter is 2, the database server stores
SQL statements for all users in the SQL statement cache except when individual
users turn off the feature with the STMT_CACHE environment variable or the
SET STATEMENT CACHE OFF statement.
STMT_CACHE 2
v Use the onmode -e command to override the STMT_CACHE configuration
parameter dynamically.
If you use the enable keyword, the database server uses the SQL statement
cache for an individual user when the user sets the STMT_CACHE environment
variable to 1 or executes the SET STATEMENT CACHE ON statement within an
application.
onmode -e enable
If you use the on keyword, the database server stores SQL statements for all
users in the SQL statement cache except when individual users turn off the
feature with the STMT_CACHE environment variable or the SET STATEMENT
CACHE OFF statement.
onmode -e on

The following table summarizes the use of the SQL statement cache, which
depends on the setting of the STMT_CACHE configuration parameter (or the
execution of onmode -e) and the use in an application of the STMT_CACHE
environment variable and the SET STATEMENT CACHE statement.

13-42 IBM Informix Performance Guide

 STMT_ CACHE SET
Configuration STMT_CACHE STATEMENT
Parameter or Environment CACHE
onmode -e Variable Statement Resulting Behavior
0 (default) Not applicable Not applicable Statement cache not used
1 0 (or not set) OFF Statement cache not used
1 1 OFF Statement cache not used
1 0 (or not set) ON Statement cache used
1 1 ON Statement cache used
1 1 Not executed Statement cache used
1 0 Not executed Statement cache not used
2 1 (or not set) ON Statement cache used
2 1 (or not set) OFF Statement cache not used
2 0 ON Statement cache used
2 0 OFF Statement cache not used by user
2 0 Not executed Statement cache not used by user
2 1 (or not set) Not executed Statement cache used by user

Placing statements in the cache

SELECT, UPDATE, INSERT and DELETE statements can be placed in the SQL
statement cache, with some exceptions. When the database server checks if an SQL
statement is in the cache, it must find an exact match.

For a complete list of the exceptions and a list of requirements for an exact match,
see SET STATEMENT CACHE in the IBM Informix Guide to SQL: Syntax.

Monitoring memory usage for each session

You can use the onstat -u command and several onstat -g command options to
obtain memory information for each session.

You obtain memory information by identifying the SQL statements that use a large
amount of memory.

To identify SQL statements using large amount of memory:

1. Run the onstat -u command to display all user threads.
2. Run the onstat -g ses command to display memory of all sessions and see
which session has the highest memory usage.
3. Run the onstat -g ses session-id command to display more details on the session
with the highest memory usage.
4. Run the onstat -g stm session-id command to display the memory used by the
SQL statements.

Display all user threads and session memory usage

Use the onstat -u command to display all user threads and memory usage by
session ID.

When the session shares the memory structures in the SSC, the value in the used
memory column should be lower than when the cache is turned off. For example,
Figure 13-6 on page 13-44 shows sample onstat -u output when the SQL statement

Chapter 13. Improving individual query performance 13-43

 cache is not enabled, and Figure 13-7 shows output after it is enabled and the
queries in Session 4 are run again. Figure 13-6 shows that Session 4 has 45656 bytes
of used memory. Figure 13-7 shows that Session 4 has less used bytes (36920) when
the SSC is enabled.

session #RSAM total used

id user tty pid hostname threads memory memory
12 informix - 0 - 0 12288 7632
4 informix 11 5158 smoke 1 53248 45656
3 informix - 0 - 0 12288 8872
2 informix - 0 - 0 12288 7632

Figure 13-6. onstat -u Output when the SQL statement cache is not enabled

session #RSAM total used

id user tty pid hostname threads memory memory
17 informix - 0 - 0 12288 7632

16  informix 12 5258 smoke 1
40960
38784

4  informix 11 5158 smoke 1
53248
36920
3 informix - 0 - 0 12288 8872
2 informix - 0 - 0 12288 7632

Figure 13-7. onstat -u Output when the SQL statement cache is enabled

Figure 13-7 also shows the memory allocated and used for Session 16, which
executes the same SQL statements as Session 4. Session 16 allocates less total
memory (40960) and uses less memory (38784) than Session 4 (Figure 13-6 shows
53248 and 45656, respectively) because it uses the existing memory structures in
the SQL statement cache.

Display detailed session information and memory usage

Use the onstat -g ses session-id command to display detailed information for a
session, including memory usage.

The following onstat -g ses session-id output columns display memory usage:
v The Memory pools portion of the output
– The totalsize column shows the number of bytes currently allocated
– The freesize column shows the number of unallocated bytes
v The last line of the output shows the number of bytes allocated from the
sscpool.

Figure 13-8 on page 13-45 shows that Session 16 has currently allocated 69632
bytes, of which 11600 bytes are allocated from the sscpool.

13-44 IBM Informix Performance Guide

 onstat -g ses 14

session #RSAM total used

id user tty pid hostname threads memory memory
14 virginia 7 28734 lyceum 1 69632 67384

tid name rstcb flags curstk status

38 sqlexec a3974d8 Y--P--- 1656 cond wait(netnorm)

Memory pools count 1

name class addr
totalsize
freesize #allocfrag #freefrag
14 V a974020
69632 
2248  156 2

...
Sess SQL Current Iso Lock SQL ISAM F.E.
Id Stmt type Database Lvl Mode ERR ERR Vers
14 SELECT vjp_stores CR Not Wait 0 0 9.03

Current statement name : slctcur

Current SQL statement :

SELECT C.customer_num, O.order_num FROM customer C, orders O, items I
WHERE C.customer_num = O.customer_num AND O.order_num = I.order_num

Last parsed SQL statement :

SELECT C.customer_num, O.order_num FROM customer C, orders O, items I
WHERE C.customer_num = O.customer_num AND O.order_num = I.order_num

11600 byte(s) of memory is allocated from the sscpool

Figure 13-8. onstat -g ses session-id output

Display information about session SQL statements

Use the onstat -g sql session-id command to display information about the SQL
statements executed by a session.

The following figure shows that onstat -g sql session-id displays the same
information as the bottom portion of the onstat -g ses session-id command in
Figure 13-8, which includes the number of bytes allocated from the sscpool.

Chapter 13. Improving individual query performance 13-45

 onstat -g sql 14

Sess SQL Current Iso Lock SQL ISAM F.E.

Id Stmt type Database Lvl Mode ERR ERR Vers
14 SELECT vjp_stores CR Not Wait 0 0 9.03

Current statement name : slctcur

Current SQL statement :

SELECT C.customer_num, O.order_num FROM customer C, orders O, items I
WHERE C.customer_num = O.customer_num AND O.order_num = I.order_num

Last parsed SQL statement :

SELECT C.customer_num, O.order_num FROM customer C, orders O, items I
WHERE C.customer_num = O.customer_num AND O.order_num = I.order_num

11600 byte(s) of memory is allocated from the sscpool

Figure 13-9. onstat -g sql session-id output

Display information about the memory that SQL statements use

in a session
Use the onstat -g stm session-id to display information about the memory each SQL
statement uses in a session.

The following figure displays the output of onstat -g stm session-id for the same
session (14) as in onstat -g ses session-id in Figure 13-8 on page 13-45 and onstat -g
sql session-id in Figure 13-9.

When the SQL statement cache (SSC) is on, the database server creates the heaps in
the SSC pool. Therefore, the heapsz output field in Figure 13-10 shows that this
SQL statement uses 10056 bytes, which is contained within the 11600 bytes in the
SSC pool that the onstat -g sql 14 shows.

onstat -g stm 14

session 14 ---------------------------------------------------------------
sdblock
heapsz statement (’*’ = Open cursor)
aa11018
10056 *SELECT C.customer_num, O.order_num
FROM customer C, orders O, items I
WHERE C.customer_num = O.customer_num
AND O.order_num = I.order_num

Figure 13-10. onstat -g stm session-id output

Monitoring usage of the SQL statement cache

If you notice a sudden increase in response time for a query that had been using
the SQL statement cache, the entry might have been dropped or deleted. You can
monitor the usage of the SQL statement cache and check for a dropped or deleted
entry by displaying onstat -g ssc command output.

The database server drops an entry from the cache when one of the objects that the
query depends on is altered so that it invalidates the data dictionary cache entry
for the query. The following operations cause a dependency check failure:

13-46 IBM Informix Performance Guide

 v Execution of any data definition language (DDL) statement (such as ALTER
TABLE, DROP INDEX, or CREATE INDEX) that might alter the query plan
v Alteration of a table that is linked to another table with a referential constraint
(in either direction)
v Execution of UPDATE STATISTICS FOR TABLE for any table or column
involved in the query
v Renaming a column, database, or index with the RENAME statement

When an entry is marked as dropped or deleted, the database server must reparse
and reoptimize the SQL statement the next time it executes. For example,
Figure 13-11 shows the entries that the onstat -g ssc command displays after
UPDATE STATISTICS was executed on the items and orders table between the
execution of the first and second SQL statements.

The Statement Cache Entries portion of the onstat -g ssc output in Figure 13-11
displays a flag field that indicates whether or not an entry has been dropped or
deleted from the SQL statement cache.
v The first entry has a flag column with the value DF, which indicates that the
entry is fully cached, but is now dropped because its entry was invalidated.
v The second entry has the same statement text as the third entry, which indicates
that it was reparsed and reoptimized when it was executed after the UPDATE
STATISTICS statement.

onstat -g ssc

...
Statement Cache Entries:

lru hash ref_cnt hits
flag heap_ptr database user

----------------------
-------------------------------------------------------
...
2 232 1 1
DF aa3d020 vjp_stores virginia
SELECT C.customer_num, O.order_num
FROM customer C, orders O, items I
WHERE C.customer_num = O.customer_num
AND O.order_num = I.order_num

3232 1 0 -F aa8b020 vjp_stores virginia

SELECT C.customer_num, O.order_num
FROM customer C, orders O, items I
WHERE C.customer_num = O.customer_num
AND O.order_num = I.order_num
...

Figure 13-11. Sample onstat -g ssc command output for a dropped entry

Monitor sessions and threads

You can monitor the number of active sessions and threads and the amount of
resources that they are using. Monitoring sessions and threads is important for
sessions that perform queries as well as sessions that perform inserts, updates, and
deletes.

Some of the information that you can monitor for sessions and threads allows you
to determine if an application is using a disproportionate amount of the resources.

Chapter 13. Improving individual query performance 13-47

 Note: Session threads for a stored procedure with a PDQ priority setting and a
GROUP BY clause are not released until a session is completed.

Monitor sessions and threads with onstat commands

You can use several onstat utility commands to monitor active sessions and
threads.

Use the following onstat utility commands to monitor sessions and threads:
v onstat -u
v onstat -g ath
v onstat -a act
v onstat -a cpu
v onstat -a ses
v onstat -g mem
v onstat -g stm

Monitor threads with onstat –u output

Use the onstat –u command to display information about active threads that
require a database server task-control block.

Active threads include threads that belong to user sessions, as well as some that
correspond to database server daemons (for example, page cleaners). Figure 13-12
on page 13-49 shows an example of onstat -u output.

Also use the onstat -u command to determine if a user is waiting for a resource or
holding too many locks, or to get an idea of how much I/O the user has
performed.

The utility output displays the following information:

v The address of each thread
v Flags that indicate the present state of the thread (for example, waiting for a
buffer or waiting for a checkpoint), whether the thread is the primary thread for
a session, and what type of thread it is (for example, user thread, daemon
thread, and so on)
For information on these flags, see the IBM Informix Administrator's Reference.
v The session ID and user login ID for the session to which the thread belongs
A session ID of 0 indicates a daemon thread.
v Whether the thread is waiting for a specific resource and the address of that
resource
v The number of locks that the thread is holding
v The number of read calls and the number of write calls that the thread has
executed
v The maximum number of current, active user threads

If you execute onstat -u while the database server is performing fast recovery,
several database server threads might appear in the display.

13-48 IBM Informix Performance Guide

 Userthreads
address flags sessid user tty wait tout locks nreads nwrites
80eb8c ---P--D 0 informix - 0 0 0 33 19
80ef18 ---P--F 0 informix - 0 0 0 0 0
80f2a4 ---P--B 3 informix - 0 0 0 0 0
80f630 ---P--D 0 informix - 0 0 0 0 0
80fd48 ---P--- 45 chrisw ttyp3 0 0 1 573 237
810460 ------- 10 chrisw ttyp2 0 0 1 1 0
810b78 ---PR-- 42 chrisw ttyp3 0 0 1 595 243
810f04 Y------ 10 chrisw ttyp2 beacf8 0 1 1 0
811290 ---P--- 47 chrisw ttyp3 0 0 2 585 235
81161c ---PR-- 46 chrisw ttyp3 0 0 1 571 239
8119a8 Y------ 10 chrisw ttyp2 a8a944 0 1 1 0
81244c ---P--- 43 chrisw ttyp3 0 0 2 588 230
8127d8 ----R-- 10 chrisw ttyp2 0 0 1 1 0
812b64 ---P--- 10 chrisw ttyp2 0 0 1 20 0
812ef0 ---PR-- 44 chrisw ttyp3 0 0 1 587 227
15 active, 20 total, 17 maximum concurrent

Figure 13-12. onstat -u output

Related reference:
onstat -u command: Print user activity profile (Administrator's Reference)

Monitor threads with onstat -g ath output

Use the onstat -g ath command to view a list of all threads. Unlike the onstat –u
command, this list includes internal daemon threads that do not have a database
server task-control block.

The onstat -g ath command display does not include the session ID (because not
all threads belong to sessions).

The status field contains information on the status of thread, such as running,
cond wait, IO Idle, IO Idle, sleeping secs: number_of_seconds, or sleeping
forever. The following output example identifies many threads as sleeping
forever. To improve performance, you can remove or reduce the number of
threads that are identified as sleeping forever.

Chapter 13. Improving individual query performance 13-49

 Threads:
tid tcb rstcb prty status vp-class name
2 10bbf36a8 0 2 sleeping forever 3lio lio vp 0
3 10bc12218 0 2 sleeping forever 4pio pio vp 0
4 10bc31218 0 2 sleeping forever 5aio aio vp 0
5 10bc50218 0 2 sleeping forever 6msc msc vp 0
6 10bc7f218 0 2 sleeping forever 7aio aio vp 1
7 10bc9e540 10b231028 4 sleeping secs: 1 1cpu main_loop()
8 10bc12548 0 2 running 1cpu tlitcppoll
9 10bc317f0 0 3 sleeping forever 1cpu tlitcplst
10 10bc50438 10b231780 2 sleeping forever 1cpu flush_sub(0)
11 10bc7f740 0 2 sleeping forever 8aio aio vp 2
12 10bc7fa00 0 2 sleeping forever 9aio aio vp 3
13 10bd56218 0 2 sleeping forever 10aio aio vp 4
14 10bd75218 0 2 sleeping forever 11aio aio vp 5
15 10bd94548 10b231ed8 3 sleeping forever 1cpu aslogflush
16 10bc7fd00 10b232630 1 sleeping secs: 26 1cpu btscanner 0
32 10c738ad8 10b233c38 4 sleeping secs: 1 1cpu onmode_mon
50 10c0db710 10b232d88 2 cond wait netnorm 1cpu sqlexec

Figure 13-13. onstat -g ath output

Threads that a primary decision-support thread started have a name that indicates
their role in the decision-support query. The following figure shows four scan
threads that belong to a decision-support thread.

Threads:
tid tcb rstcb prty status vp-class name
11 994060 0 4 sleeping(Forever) 1cpu kaio
12 994394 80f2a4 2 sleeping(secs: 51) 1cpu btclean
26 99b11c 80f630 4 ready 1cpu onmode_mon
32 a9a294 812b64 2 ready 1cpu sqlexec
113 b72a7c 810b78 2 ready 1cpu sqlexec
114 b86c8c 81244c 2 cond wait(netnorm) 1cpu sqlexec
115 b98a7c 812ef0 2 cond wait(netnorm) 1cpu sqlexec
116 bb4a24 80fd48 2 cond wait(netnorm) 1cpu sqlexec
117 bc6a24 81161c 2 cond wait(netnorm) 1cpu sqlexec
118 bd8a24 811290 2 ready 1cpu sqlexec
119 beae88 810f04 2 cond wait(await_MC1) 1cpu scan_1.0
120 a8ab48 8127d8 2 ready 1cpu scan_2.0
121 a96850 810460 2 ready 1cpu scan_2.1
122 ab6f30 8119a8 2 running 1cpu scan_2.2

Figure 13-14. onstat -g ath output showing scan threads belonging to a decision-support
thread

Related concepts:
“Improve connection performance and scalability” on page 3-16
Related reference:
onstat -g ath command: Print information about all threads (Administrator's
Reference)

Monitor threads with onstat -g act output

Use the onstat -g act command to obtain a list of active threads. The onstat -g act
output shows a subset of the threads that are also listed in onstat -g ath output.

For sample output, see the IBM Informix Administrator's Reference.

Related reference:

13-50 IBM Informix Performance Guide

 onstat -g act command: Print active threads (Administrator's Reference)

Monitor threads with onstat -g cpu output

Use the onstat –g cpu command to display the last time the thread ran, how much
CPU time the thread used, the number of times the thread ran, and other statistics
about all the threads running in the server

The following output example shows the ID and name of each thread that is
running, the ID of the virtual processor in which each thread is running, the day
and time when each thread last ran, how much CPU time each thread used, the
number of times each thread was scheduled to run, and the status of each thread.

Thread CPU Info:

tid name vp Last Run CPU Time #scheds status
2 lio vp 0 3lio* 07/18 08:35:35 0.0000 1 IO Idle
3 pio vp 0 4pio* 07/18 08:35:36 0.0102 2 IO Idle
4 aio vp 0 5aio* 07/18 08:35:47 0.6876 68 IO Idle
5 msc vp 0 6msc* 07/18 11:47:24 0.0935 14 IO Idle
6 main_loop() 1cpu* 07/18 15:02:43 2.9365 23350 sleeping secs: 1
7 soctcppoll 7soc* 07/18 08:35:40 0.1150 1 running
8 soctcpio 8soc* 07/18 08:35:40 0.0037 1 running
9 soctcplst 1cpu* 07/18 11:47:24 0.1106 10 sleeping forever
10 soctcplst 1cpu* 07/18 08:35:40 0.0103 6 sleeping forever
11 flush_sub(0) 1cpu* 07/18 15:02:43 0.0403 23252 sleeping secs: 1
12 flush_sub(1) 1cpu* 07/18 15:02:43 0.0423 23169 sleeping secs: 1
13 flush_sub(2) 1cpu* 07/18 15:02:43 0.0470 23169 sleeping secs: 1
14 flush_sub(3) 1cpu* 07/18 15:02:43 0.0407 23169 sleeping secs: 1
15 flush_sub(4) 1cpu* 07/18 15:02:43 0.0307 23169 sleeping secs: 1
16 flush_sub(5) 1cpu* 07/18 15:02:43 0.0323 23169 sleeping secs: 1
17 flush_sub(6) 1cpu* 07/18 15:02:43 0.0299 23169 sleeping secs: 1
18 flush_sub(7) 1cpu* 07/18 15:02:43 0.0314 23169 sleeping secs: 1
19 kaio 1cpu* 07/18 14:56:42 1.4560 2375587 IO Idle
20 aslogflush 1cpu* 07/18 15:02:43 0.0657 23166 sleeping secs: 1
21 btscanner_0 1cpu* 07/18 15:00:53 0.0484 784 sleeping secs: 61
37 onmode_mon 1cpu* 07/18 15:02:43 0.3467 23165 sleeping secs: 1
43 dbScheduler 1cpu* 07/18 14:58:14 1.6613 320 sleeping secs: 31
44 dbWorker1 1cpu* 07/18 13:48:10 0.4264 399 sleeping forever
45 dbWorker2 1cpu* 07/18 14:48:11 1.9346 2936 sleeping forever
94 bf_priosweep() 1cpu* 07/18 15:01:42 0.0431 77 cond wait bp_cond

Figure 13-15. onstat -g cpu command output

Related reference:
onstat -g cpu: Print runtime statistics (Administrator's Reference)

Monitor session resources with onstat -g ses output

Use the onstat -g ses command to monitor the resources allocated for and used by
a session, in particular, a session that is running a decision-support query.

For example, in Figure 13-16 on page 13-52, session number 49 is running five
threads for a decision-support query.

Chapter 13. Improving individual query performance 13-51

 session #RSAM total used
id user tty pid hostname threads memory memory
57 informix - 0 - 0 8192 5908
56 user_3 ttyp3 2318 host_10 1 65536 62404
55 user_3 ttyp3 2316 host_10 1 65536 62416
54 user_3 ttyp3 2320 host_10 1 65536 62416
53 user_3 ttyp3 2317 host_10 1 65536 62416
52 user_3 ttyp3 2319 host_10 1 65536 62416
51 user_3 ttyp3 2321 host_10 1 65536 62416
49 user_1 ttyp2 2308 host_10 5 188416 178936
2 informix - 0 - 0 8192 6780
1 informix - 0 - 0 8192 4796

Figure 13-16. onstat -g ses output

Related reference:
onstat -g ses command: Print session-related information (Administrator's
Reference)

Monitor session memory with onstat -g mem and onstat -g stm

output
Use the onstat -g mem and onstat -g stm commands to obtain information about
the memory used for each session.

You can determine which session to focus on by the used memory column of the
onstat -g ses output.

Figure 13-17 on page 13-53 shows sample onstat -g ses output and some of the
onstat -g mem and onstat -g stm output for Session 16.
v The output of the onstat -g mem command shows the total amount of memory
used by each session.
The totalsize column of the onstat -g mem 16 output shows the total amount of
memory allocated to the session.
v The output of the onstat -g stm command shows the portion of the total
memory allocated to the current prepared SQL statement.
The heapsz column of the onstat -g stm 16 output in the following figure shows
the amount of memory allocated for the current prepared SQL statement.

13-52 IBM Informix Performance Guide

 onstat -g ses

session #RSAM total used

id user tty pid hostname threads memory memory
18 informix - 0 - 0 12288 8928
17 informix 12 28826 lyceum 1 45056 33752
16 virginia 6 28743 lyceum 1 90112 79504
14 virginia 7 28734 lyceum 1 45056 33096
3 informix - 0 - 0 12288 10168
2 informix - 0 - 0 12288 8928

onstat -g mem 16

Pool Summary:
name class addr totalsize freesize #allocfrag #freefrag
16 V a9ea020 90112 10608 159 5
...

onstat -g stm 16

session 16 ---------------------------------------------------------------
sdblock heapsz statement (’*’ = Open cursor)
aa0d018 10056 *SELECT C.customer_num, O.order_num
FROM customer C, orders O, items I
WHERE C.customer_num = O.customer_num
AND O.order_num = I.order_num

Figure 13-17. onstat -g mem and onstat -g stm to determine session memory

Related reference:
onstat -g lap command: Print light appends status information (Administrator's
Reference)
onstat -g mem command: Print pool memory statistics (Administrator's
Reference)

Monitor sessions and threads with ON-Monitor (UNIX)

On (UNIX), you can use ON-Monitor to view information about sessions and user
threads. This information is a subset of the information that the onstat -u
command displays.

Choose User from the Status menu. The following information appears:
v The session ID
v The user ID
v The number of locks that the thread is holding
v The number of read calls and write calls that the thread has executed
v Flags that indicate the present state of the thread (for example, waiting for a
buffer or waiting for a checkpoint), whether the thread is the primary thread for
a session, and what type of thread it is (for example, user thread, daemon
thread, and so on)

The following figure shows sample output.

Chapter 13. Improving individual query performance 13-53

 USER THREAD INFORMATION

Locks Disk Disk User thread

Session User Held Reads Writes Status

0 informix 0 96 2 ------D
0 informix 0 0 0 ------F
0 informix 0 0 0 -------
15 informix 0 0 0 Y-----M
0 informix 0 0 0 ------D
17 chrisw 1 3 34 Y------

Figure 13-18. Output from the User option of the ON-Monitor Status menu

Monitor sessions and threads with SMI tables

You can use the syssessions and the syssesprof system-monitoring interface (SMI)
tables to obtain information about sessions and threads.

Query the syssessions table to obtain the following information.

Column
Description
sid Session ID
username
Name (login ID) of the user
uid User ID
pid Process ID
connected
Time that the session started
feprogram
Absolute path of the executable program or application

In addition, some columns contain flags that show the following information;
v Whether the primary thread of the session is waiting for a latch, lock, log buffer,
or transaction
v If the thread is in a critical section.

Important: The information in the syssessions table is organized by session, and

the information in the onstat -u output is organized by thread. Also, unlike the
onstat -u output, the syssessions table does not include information about daemon
threads, only user threads.

Query the syssesprof table to obtain a profile of the activity of a session. This table
contains a row for each session with columns that store statistics on session activity
(for example, number of locks held, number of row writes, number of commits,
number of deletes, and so on).

For a complete list of the syssessions columns and descriptions of syssesprof

columns, see the chapter on the sysmaster database in the IBM Informix
Administrator's Reference.

13-54 IBM Informix Performance Guide

Monitor transactions
You can monitor transactions to track open transactions and the locks that those
transactions hold. You can use several onstat utility options to view transaction,
lock, and session statistics.

ISA uses information that the following onstat command-line options generate to
display session information, as the following table shows. Click the Refresh button
to rerun the onstat command and display fresh information.

The following onstat command-line options display session information.

To monitor Displays the output of See

Transaction statistics onstat -x “Display information about transactions”
User session statistics onstat -u “Display statistics on user sessions” on page
13-57
Lock statistics onstat -k “Display information about transaction locks”
on page 13-56
Sessions running SQL statements onstat -g sql session-id “Display statistics on sessions executing SQL
statements” on page 13-58

Display information about transactions

The output of the onstat -x command contains information that you can use to
monitor transactions.

The onstat -x output contains the following information for each open transaction:
v The address of the transaction structure in shared memory
v Flags that indicate the following information:
– The present state of the transaction (user thread attached, suspended, waiting
for a rollback)
– The mode in which the transaction is running (loosely coupled or tight
coupled)
– The stage that the transaction is in (BEGIN WORK, prepared to commit,
committing or committed, rolling back)
– The nature of the transaction (global transaction, coordinator, subordinate,
both coordinator and subordinate)
v The thread that owns the transaction
v The number of locks that the transaction holds
v The logical-log file in which the BEGIN WORK record was logged
v The current logical-log id and position
v The isolation level
v The number of attempts to start a recovery thread
v The coordinator for the transaction (if the subordinate is executing the
transaction)
v The maximum number of concurrent transactions since you last started the
database server

Chapter 13. Improving individual query performance 13-55

 The onstat utility is especially useful for monitoring global transactions. For
example, you can determine whether a transaction is executing in loosely coupled
or tightly coupled mode. These transaction modes have the following
characteristics:
v Loosely coupled mode
Each branch in a global transaction has a separate transaction ID (XID). This
mode is the default.
– The different database servers coordinate transactions, but do not share
resources. No two transaction branches, even if they access the same
database, can share locks.
– The records from all branches of a global transaction display as separate
transactions in the logical log.
v Tightly coupled mode
In a single global transaction, all branches that access the same database share
the same transaction ID (XID). This mode only occurs with the Microsoft
Transaction Server (MTS) transaction manager.
– The different database servers coordinate transactions and share resources
such as locks and log records. The branches with the same XID share locks
and can never wait on another branch with the same XID because only one
branch is active at one time.
– Log records for branches with the same XID appear under the same
transaction in the logical log.

Figure 13-19 shows sample output from onstat -x. The last transaction listed is a
global transaction, as the G value in the fifth position of the flags column indicates.
The T value in the second position of the flags column indicates that the
transaction is running in tightly coupled mode.

Transactions
address flags userthread locks beginlg curlog logposit isol retrys coord
ca0a018 A---- c9da018 0 0 5 0x18484c COMMIT 0
ca0a1e4 A---- c9da614 0 0 0 0x0 COMMIT 0
ca0a3b0 A---- c9dac10 0 0 0 0x0 COMMIT 0
ca0a57c A---- c9db20c 0 0 0 0x0 COMMIT 0
ca0a748 A---- c9db808 0 0 0 0x0 COMMIT 0
ca0a914 A---- c9dbe04 0 0 0 0x0 COMMIT 0
ca0aae0 A---- c9dcff8 1 0 0 0x0 COMMIT 0
ca0acac A---- c9dc9fc 1 0 0 0x0 COMMIT 0
ca0ae78 A---- c9dc400 1 0 0 0x0 COMMIT 0
ca0b044 AT--G c9dc9fc 0 0 0 0x0 COMMIT 0
10 active, 128 total, 10 maximum concurrent

Figure 13-19. onstat -x output

The output in Figure 13-19 shows that this transaction branch is holding 13 locks.
When a transaction runs in tightly coupled mode, the branches of this transaction
share locks.

Display information about transaction locks

The output of the onstat -k command contains details on the locks that a
transaction holds.

To find the relevant locks, match the address in the userthread column in onstat -x
output to the address in the owner column of onstat -k output.

13-56 IBM Informix Performance Guide

 Figure 13-20 shows sample output from onstat -x and the corresponding onstat -k
command. The a335898 value in the userthread column in the onstat -x output
matches the value in the owner column of the two lines of onstat -k output.

onstat -x

Transactions
address flags userthread locks beginlg curlog logposit isol retrys coord
a366018 A---- a334018 0 0 1 0x22b048 COMMIT 0
a3661f8 A---- a334638 0 0 0 0x0 COMMIT 0
a3663d8 A---- a334c58 0 0 0 0x0 COMMIT 0
a3665b8 A---- a335278 0 0 0 0x0 COMMIT 0
a366798 A----
a335898 2 0 0 0x0 COMMIT 0
a366d38 A---- a336af8 0 0 0 0x0 COMMIT 0
6 active, 128 total, 9 maximum concurrent

onstat -k

Locks
address wtlist
owner  lklist type tblsnum rowid key#/bsiz
a09185c 0
a335898 0 HDR+S 100002 20a 0
a0918b0 0
a335898 a09185c HDR+S 100002 204 0
2 active, 2000 total, 2048 hash buckets, 0 lock table overflows

Figure 13-20. onstat -k and onstat -x output

In the example in Figure 13-20, a user is selecting a row from two tables. The user
holds the following locks:
v A shared lock on one database
v A shared lock on another database

Display statistics on user sessions

The output of the onstat -u command contains statistics on user sessions.

You can find the session-id of the transaction by matching the address in the
userthread column of the onstat -x output with the address column in the onstat
-u output. The sessid column of the same line in the onstat -u output provides the
session id.

For example, Figure 13-21 on page 13-58 shows the address a335898 in the
userthread column of the onstat -x output. The output line in onstat -u with the
same address shows the session id 15 in the sessid column.

Chapter 13. Improving individual query performance 13-57

 onstat -x

Transactions
address flags userthread locks beginlg curlog logposit isol retrys coord
a366018 A---- a334018 0 0 1 0x22b048 COMMIT 0
a3661f8 A---- a334638 0 0 0 0x0 COMMIT 0
a3663d8 A---- a334c58 0 0 0 0x0 COMMIT 0
a3665b8 A---- a335278 0 0 0 0x0 COMMIT 0
a366798 A----
a335898 2 0 0 0x0 COMMIT 0
a366d38 A---- a336af8 0 0 0 0x0 COMMIT 0
6 active, 128 total, 9 maximum concurrent

onstat -u

address flags sessid user tty wait tout locks nreads nwrites
a334018 ---P--D 1 informix - 0 0 0 20 6
a334638 ---P--F 0 informix - 0 0 0 0 1
a334c58 ---P--- 5 informix - 0 0 0 0 0
a335278 ---P--B 6 informix - 0 0 0 0 0

a335898 Y--P---
15 informix 1 a843d70 0 2 64 0
a336af8 ---P--D 11 informix - 0 0 0 0 0
6 active, 128 total, 17 maximum concurrent

Figure 13-21. Obtaining session-id for userthread in onstat -x

For a transaction executing in loosely coupled mode, the first position of the flags
column in theonstat -u output might display a value of T. This T value indicates
that one branch within a global transaction is waiting for another branch to
complete. This situation could occur if two different branches in a global
transaction, both using the same database, tried to work on the same global
transaction simultaneously.

For a transaction executing in tightly coupled mode, this T value does not occur
because the database server shares one transaction structure for all branches that
access the same database in the global transaction. Only one branch is attached
and active at one time and does not wait for locks because the transaction owns all
the locks held by the different branches.

Display statistics on sessions executing SQL statements

The output of the onstat -g sql command contains statistics on the SQL statements
executed by the session

To obtain information about the last SQL statement that each session executed,
issue the onstat -g sql command with the appropriate session ID.

Figure 13-22 on page 14-1 shows sample output for this option using the same
session ID obtained from the onstat -u sample in Figure 13-21.

13-58 IBM Informix Performance Guide

 onstat -g sql 15

Sess SQL Current Iso Lock SQL ISAM F.E.

Id Stmt type Database Lvl Mode ERR ERR Vers
15 SELECT vjp_stores CR Not Wait 0 0 9.03

Current statement name : slctcur

Current SQL statement :

select l.customer_num, l.lname, l.company, l.phone, r.call_dtime,
r.call_descr from customer l, vjp_stores@gilroy:cust_calls r where
l.customer_num = r.customer_num

Last parsed SQL statement :

select l.customer_num, l.lname, l.company, l.phone, r.call_dtime,
r.call_descr from customer l, vjp_stores@gilroy:cust_calls r where
l.customer_num = r.customer_num

Figure 13-22. onstat -g sql output

Chapter 14. The onperf utility on UNIX

The onperf utility is a windowing environment that you can use to monitor the
database server performance. The onperf utility monitors the database server
running on the UNIX operating system.
Related reference:
“Database server tools” on page 2-3

Overview of the onperf utility

The onperf utility is a graphical tool that you can use for displaying most of the
same database server metrics that you can view on onstat utility reports.

The onperf utility provides the following advantages over the onstat utility:
v Displays metric values graphically in real time
v Allows you to choose which metrics to monitor
v Allows you to scroll back to previous metric values to analyze a trend
v Allows you to save performance data to a file for review at a later time

You cannot use the onperf utility on High-Availability Data Replication (HDR)
secondary servers, remote standalone (RS) secondary servers, or shared disk (SD)
secondary servers.

Basic onperf utility functions

The onperf utility displays the values of the database server metrics in a tool
window and saves the database server metric values to a file. You can review the
contents of this file.

Display metric values

The onperf utility displays database server metrics obtained from shared memory.

When onperf starts, it activates the following processes:

v The onperf process. This process controls the display of onperf tools.

© Copyright IBM Corp. 1996, 2014 14-1

 v The data-collector process. This process attaches to shared memory and passes
performance information to the onperf process for display in an onperf tool.

An onperf tool is a Motif window that an onperf process manages, as Figure 14-1
shows.

Shared memory
onperf tool
Data-collector process onperf process

onperf onperf

Figure 14-1. Data flow from shared memory to an onperf tool window

Save metric values to a file

The onperf utility saves collected metrics in a history file.

The onperf utility allows designated metrics to be continually buffered. The data
collector writes these metrics to a circular buffer called the data-collector buffer.
When the buffer becomes full, the oldest values are overwritten as the data
collector continues to add data. The current contents of the data-collector buffer are
saved to a history file, as Figure 14-2 illustrates.

Shared memory Data-collector process Data-collector buffer

1 onperf

Data-collector buffer Data-collector process History file

2 onperf

Figure 14-2. How onperf saves performance data

The onperf utility uses either a binary format or an ASCII representation for data
in the history file. The binary format is host-dependent and allows data to be
written quickly. The ASCII format is portable across platforms.

You have control over the set of metrics stored in the data-collector buffer and the
number of samples. You could buffer all metrics; however, this action might
consume more memory than is feasible. A single metric measurement requires 8
bytes of memory. For example, if the sampling frequency is one sample per second,
then to buffer 200 metrics for 3,600 samples requires approximately 5.5 megabytes
of memory. If this process represents too much memory, you must reduce the
depth of the data-collector buffer, the sampling frequency, or the number of
buffered metrics.

14-2 IBM Informix Performance Guide

 To configure the buffer depth or the sampling frequency, you can use the
Configuration dialog box. For more information about the Configuration dialog
box, see “The graph-tool Configure menu and the Configuration dialog box” on
page 14-8.

Review metric measurements

You can review the contents of a history file in a tool window. When you request a
tool to display a history file, the onperf utility starts a playback process that reads
the data from disk and sends the data to the tool for display.

The playback process is similar to the data-collector process mentioned under

“Save metric values to a file” on page 14-2. However, instead of reading data from
shared memory, the playback process reads measurements from a history file.
Figure 14-3 shows the playback process.

onperf tool
History file Playback process onperf process
onperf onperf

Figure 14-3. Flow of data from a history file to an onperf tool window

onperf utility tools

The onperf utility provides Motif windows, called tools, which display metric
values.
Table 14-1. onperf utility tools
Tool Description
Graph tool This tool allows you to monitor general performance activity.
You can use this tool to display any combination of metrics
that onperf supports and to display the contents of a history
file. For more information, see “Graph tool” on page 14-5.
Query-tree tool This tool displays the progress of individual queries. For more
information, see “Query-tree tool” on page 14-11.
Status tool This tool displays status information about the database server
and allows you to save the data that is currently held in the
data-collector buffer to a file. For more information, see “Status
tool” on page 14-11.
Activity tools These tools display specific database server activities. Activity
tools include disk, session, disk-capacity, physical-processor,
and virtual-processor tools. The physical-processor and
virtual-processor tools, respectively, display information about
all CPUs and VPs. The other activity tools each display the top
10 instances of a resource ranked by a suitable activity
measurement. For more information, see “Activity tools” on
page 14-12.

Requirements for running the onperf utility

The computer that is running the onperf utility must support the X terminal and
the mwm window manager.

Chapter 14. The onperf utility on UNIX 14-3

 When you install the database server, the following executable files are written to
the $INFORMIXDIR/bin directory:
v onperf
v onedcu
v onedpu
v xtree

In addition, the onperf.hlp online help file is placed in the directory

$INFORMIXDIR/hhelp.

When the database server is installed and running in online mode, you can bring
up onperf tools either on the computer that is running the database server or on a
remote computer or terminal that can communicate with your database server
instance. Figure 14-4 illustrates both possibilities. In either case, the computer that
is running the onperf tools must support the X terminal and the mwm window
manager.

UNIX platform running database server and X display server

with mwm window manager

onperf tool
onperf process
Database
server onperf

UNIX platform running database server Client platform running X and mwm
onperf tool
onperf process
Database
server onperf

Figure 14-4. Two options for running onperf

Starting the onperf utility and exiting from it

Before you start the onperf utility, set the DISPLAY and LD_LIBRARY_PATH
environment variables.

Prerequisite: Set the DISPLAY environment variable as follows:

C shell setenv DISPLAY displayname0:0 #

Bourne shell DISPLAY=displayname0:0 #

In these commands, displayname is the name of the computer or X terminal where

the onperf window should appear.

Set the LD_LIBRARY_PATH environment variable to the appropriate value for the
Motif libraries on the computer that is running onperf.

14-4 IBM Informix Performance Guide

 With the environment properly set up, you can enter onperf to bring up a
graph-tool window, as described in “The onperf user interface.”

You can monitor multiple database server instances from the same Motif client by
invoking onperf for each database server, as the following example shows:
INFORMIXSERVER=instance1 ; export INFORMIXSERVER; onperf
INFORMIXSERVER=instance2 ; export INFORMIXSERVER; onperf
...

Exiting from the onperf Utility

To exit from the onperf utility, use the Close option to close each tool window, use
the Exit option of a tool, or choose Window Manager > Close.

The onperf user interface

When you invoke the onperf utility, it displays an initial graph-tool window. From
this graph-tool window, you can display additional graph-tool windows as well as
the query-tree, data-collector, and activity tools.

The graph-tool windows have no hierarchy; you can create and close these
windows in any order.

Graph tool
The graph tool is the principal onperf interface. Use the graph tool to display any
set of database server metrics that the onperf data collector obtains from shared
memory.

The Figure 14-5 shows a diagram of a graph tool that displays a graph of metrics
for ISAM calls.

Graph Tool #1 -- server2

Graph Metrics View Configure Tools Help

100
80
60
40
20
0
09:55:30 09:56:00 09:56:30 09:57:00

ISAM Calls (Reduced x10)

Figure 14-5. Graph-Tool window

You cannot bring up a graph-tool window from a query-tree tool, a status tool, or
one of the activity tools.

Chapter 14. The onperf utility on UNIX 14-5

 Graph-tool title bar
When you invoke onperf, the initial graph-tool window displays serverName, the
database server that the INFORMIXSERVER environment variable specifies, in the
title bar. The data comes from the shared memory of the indicated database server
instance.

If the configuration of an initial graph-tool has not yet been saved or loaded from
disk, onperf does not display the name of a configuration file in the title bar.

If you open a historical data file, for example named caselog.23April.2PM, in this
graph-tool window, the title bar displays caselog.23.April.23.April.2PM.

Graph-tool graph menu

The Graph menu contains options for creating, opening, saving the contents of,
printing the contents of, annotating, and closing a graph tool.

The Graph menu provides the following options.

Option Use
New Creates a new graph tool. All graph tools that you create using this option
share the same data-collector and onperf processes. To create new graph
tools, use this option rather than invoke onperf multiple times.
Open History File
Loads a previously saved file of historical data into the graph tool for
viewing. If the file does not exist, onperf prompts you for one. When you
select a file, onperf starts a playback process to view the file.
Save History File
Saves the contents of the data-collector buffer to either an ASCII or a
binary file, as specified in the Configuration dialog box.
Save History File As
Specifies the filename in which to save the contents of the data-collector
buffer.
Annotate
Brings up a dialog box in which you can enter a header label and a footer
label. Each label is optional. The labels are displayed on the graph. When
you save the graph configuration, onperf includes these labels in the saved
configuration file.
Print Brings up a dialog box that allows you to select a destination file. You
cannot send the contents of the graph tool directly to a printer; you must
use this option to specify a file and subsequently send the file to a printer.
Close Closes the tool. When a tool is the last remaining tool of the onperf
session, this menu item behaves in the same way as the Exit option.
Exit Exits onperf.

Important: To save your current configuration before you load a new

configuration from a file, you must choose Configure > Save
Configuration or Configure > Save Configuration As.

Graph-tool metrics menu

The Metrics menu contains options for choosing the class of metrics to display in
the graph tool.

14-6 IBM Informix Performance Guide

 Metrics are organized by class and scope. When you select a metric for the graph
tool to display, you must specify the metric class, the metric scope, and the name
of the metric.

The metric class is the generic database server component or activity that the metric
monitors. The metric scope depends on the metric class. In some cases, the metric
scope indicates a particular component or activity. In other cases, the scope
indicates all activities of a given type across an instance of the database server.

The Metrics menu has a separate option for each class of metrics. For more
information about metrics, see “Why you might want to use onperf” on page
14-12.

When you choose a class, such as Server, you see a dialog box like the one in
Figure 14-6.

Select Metrics for Graph Tool #1

Server Metrics Available Selected Metrics

CPU System Time /Server/dbservername/CPU System Time
dbservername CPU User Time
Percent cached (Read)
Percent cached (Write)
Disk Reads Add
Disk Writes
Page Reads
Page Writes
Buffer Reads Remove
Buffer Writes
Isam Calls

OK Filter Cancel Help

Figure 14-6. The Select Metrics dialog box

The Select Metrics dialog box contains three list boxes. The list box on the left
displays the valid scope levels for the selected metrics class. For example, when
the scope is set to Server, the list box displays the dbservername of the database
server instance that is being monitored. When you select a scope from this list,
onperf displays the individual metrics that are available within that scope in the
middle list box. You can select one or more individual metrics from this list and
add them to the display by clicking Add. To remove them from the display, click
Remove.

Tip: You can display metrics from more than one class in a single graph-tool
window. For example, you might first select ISAM Calls, Opens, and Starts from
the Server class. When you choose the Option menu in the same dialog box, you
can select another metric class without exiting the dialog box. For example, you
might select the Chunks metric class and add the Operations, Reads, and Writes
metrics to the display.

The Filter button in the dialog box brings up an additional dialog box in which
you can filter long text strings shown in the Metrics dialog box. The Filter dialog
box also lets you select tables or fragments for which metrics are not currently
displayed.

Chapter 14. The onperf utility on UNIX 14-7

 After you make your selections, you can click OK to proceed, or Cancel if you
choose not to proceed.

Graph-tool view menu

The View menu contains options for changing how the graph tool appears.

The View menu provides the following options.

Line Changes the graph tool to the line format. Line format includes horizontal
and vertical scroll bars. The vertical scroll bar adjusts the scale of the
horizontal time axis. When you raise this bar, onperf reduces the scale and
vice versa. The horizontal scroll bar allows you to adjust your view along
the horizontal time axis.
To change the color and width of the lines in the line format, click the
legend in the graph tool. When you do, onperf generates a Customize
Metric dialog box that provides a choice of line color and width.
Horizontal Bar Graph
Changes the graph tool to the horizontal bar format.
Vertical Bar Graph
Changes the graph tool to the vertical bar format.
Pie Changes the graph tool to the pie-chart format. To display a pie chart, you
must select at least two metrics.
Quick Rescale Axis
Rescales the axis to the largest point that is currently visible on the graph.
This button turns off automatic rescaling.
Configure Axis
Displays the Axis Configuration dialog box. Use this dialog box to select a
fixed value for the y-axis on the graph or select automatic axis scaling.

The graph-tool Configure menu and the Configuration dialog box

The Configure menu contains options of opening, editing, and saving onperf
configuration information.

The Configure menu provides the following options.

Edit Configuration
Brings up the Configuration dialog box, which allows you to change the
settings for the current data-collector buffer, graph-tool display options,
and data-collector options. The Configuration dialog box appears in
Figure 14-7 on page 14-9.
Open Configuration
Restarts onperf with the settings that are stored in the configuration file.
Unsaved data in the data-collector buffer is lost. If no configuration file is
specified and the default does not exist, the following error message
appears:
Open file filename failed.

If the specified configuration file does not exist, onperf prompts for one.
Save Configuration
Saves the current configuration to a file. If no configuration file is currently
specified, onperf prompts for one.

14-8 IBM Informix Performance Guide

 Save Configuration As
Saves a configuration file. This option always prompts for a filename.

To configure data-collector options, graph-display options, and metrics about

which to collect data, choose the Edit Configuration option to bring up the
Configuration dialog box.

Configuration

History Buffer Configuration Selected Metric Groups

Server
dbservername
Add

Remove

Graph Display Options Data Collector Options

Graph Scroll: 10% Sample Interval: 1 second

Tool Interval: 3 Sample Intervals History Depth: 3600

Graph Width: 2 Min Save Mode: Binary

OK Filter Cancel Help

Figure 14-7. The Configuration dialog box

The Configuration dialog box provides the following options for configuring
display.
Option Use
History Buffer Configuration
Allows you to select a metric class and metric scope to include in the
data-collector buffer. The data collector gathers information about all
metrics that belong to the indicated class and scope.
Graph Display Options
Allows you to adjust the size of the graph portion that scrolls off to the left
when the display reaches the right edge, the initial time interval that the
graph is to span, and the frequency with which the display is updated.
Data Collector Options
Controls the collection of data. The sample interval indicates the amount of
time to wait between recorded samples. The history depth indicates the
number of samples to retain in the data-collector buffer. The save mode
indicates the data-collector data should be saved in binary or ASCII
format.

Graph-tool Tools menu

The Tools menu contains options that start additional onperf tools.

This menu provides the following options.

Chapter 14. The onperf utility on UNIX 14-9

 Query Tree
Starts a query-tree tool. For more information, see “Query-tree tool” on
page 14-11.
Status Starts a status tool. For more information, see “Status tool” on page 14-11.
Disk Activity
Starts a disk-activity tool. For more information, see “Activity tools” on
page 14-12.
Session Activity
Starts a session-activity tool. For more information, see “Activity tools” on
page 14-12.
Disk Capacity
Starts a disk-capacity tool. For more information, see “Activity tools” on
page 14-12.
Physical Processor Activity
Starts a physical-processor tool. For more information, see “Activity tools”
on page 14-12.
Virtual Processor Activity
Starts a virtual-processor tool. For more information, see “Activity tools”
on page 14-12.

Changing the scale of metrics

When onperf displays metrics, it automatically adjusts the scale of the y-axis to
accommodate the largest value. You can use the Customize Metric dialog box to
establish one for the current display.

For more information, see “Graph-tool view menu” on page 14-8.

Displaying recent-history values

When you use the onperf utility, you can scroll back over previous metric values
that are displayed in a line graph. This is useful for analyzing recent trends.

The time interval to which you can scroll back is the lesser of the following
intervals:
v The time interval over which the metric has been displayed
v The history interval that the graph-tool Configuration dialog box specifies
The length of time you can scroll back through cannot exceed the depth of the
data-collector buffer.
For more information, see “The graph-tool Configure menu and the
Configuration dialog box” on page 14-8.

Figure 14-8 on page 14-11 illustrates the maximum scrollable intervals for metrics
that span different time periods.

14-10 IBM Informix Performance Guide

 Data-collector buffer

Buffer depth

Metric 1

Scrollable interval for metric 1

Metric 2

Scrollable interval for metric 2

Time that metric 1 Time that metric 2 Current time

is first displayed is first displayed

Figure 14-8. Maximum scrollable intervals for metrics that span different time periods

Query-tree tool
The query-tree tool contains options for monitoring the performance of individual
queries.

The query-tree tool is a separate executable tool that does not use the data-collector
process. You cannot save query-tree tool data to a file.

This tool includes a Select Session button and a Quit button. When you select a
session that is running a query, the large detail window displays the SQL operators
that constitute the execution plan for the query. The query-tree tool represents each
SQL operator with a box. Each box includes a dial that indicates rows per second
and a number that indicates input rows. In some cases, not all the SQL operators
can be represented in the detail window. The smaller window shows the SQL
operators as small icons.

The Quit button allows you to exit from the query-tree tool.

Status tool
The status tool enables you to select metrics to store in the data-collector buffer. In
addition, you can use this tool to save the data currently held in the data-collector
buffer to a file.

Figure 14-9 on page 14-12 shows a status tool.

The status tool displays:

Chapter 14. The onperf utility on UNIX 14-11

 v The length of time that the data collector has been running
v The size of the data-collector process area, called the collector virtual memory size
When you select different metrics to store in the data-collector buffer, you see
different values for the collector virtual memory size.

Status Tool

File Tools Help

Server: Dynamic Server, Running 0:52:25

Shared memory size: 1.45 MB

Data Collector: Running 0:03:38

Collector virtual memory size: 0.63 MB

Figure 14-9. Status Tool window

The status tool File menu provides the following options.

Option Use
Close This option closes the tool. When it is the last remaining tool of the onperf
session, Close behaves in the same way as Exit.
Exit This option exits onperf.

Activity tools
Activity tools are specialized forms of the graph tool that display instances of the
specific activity, based on a ranking of the activity by some suitable metric.

You can choose from among the following activity tools:

v The disk-activity tool, which displays the top 10 activities ranked by total
operations
v The session-activity tool, which displays the top 10 activities ranked by ISAM
calls plus PDQ calls per second
v The disk-capacity tool, which displays the top 10 activities ranked by free space
in megabytes
v The physical-processor-activity tool, which displays all processors ranked by
nonidle CPU time
v The virtual-processor-activity tool, which displays all VPs ranked by VP user
time plus VP system time

The activity tools use the bar-graph format. You cannot change the scale of an
activity tool manually; onperf always sets this value automatically.

The Graph menu provides you with options for closing, printing, and exiting the
activity tool.

Why you might want to use onperf

You can use the onperf utility for routine monitoring, diagnosing sudden
performance loss, and diagnosing performance degradation.

14-12 IBM Informix Performance Guide

 The following sections provide suggestions for different ways to use the onperf
utility.

Routine monitoring with onperf

You can use the onperf utility to facilitate routine monitoring. For example, you
can display several metrics in a graph-tool window and run this tool throughout
the day.

Displaying these metrics allows you to monitor database server performance

visually at any time.

Diagnosing sudden performance loss

When you detect a sudden performance dip, you can use the onperf utility to
examine the recent history of the database server metrics values to identify any
trend.

The onperf utility allows you to scroll back over a time interval, as explained in
“Displaying recent-history values” on page 14-10.

Diagnosing performance degradation

You can save the metrics that the onperf utility displays, so you can analyze it and
compare it to other saved information. This can be useful when analyzing
performance problems that gradually develop and might be difficult to diagnose.

For example, if you detect a degradation in database server response time, it might
not be obvious from looking at the current metrics which value is responsible for
the slowdown. The performance degradation might also be sufficiently gradual
that you cannot detect a change by observing the recent history of metric values.
To allow for comparisons over longer intervals, onperf allows you to save metric
values to a file, as explained in “Status tool” on page 14-11.

onperf utility metrics

When you use the onperf utility, you can view various metric classes.

The following sections describe these metric classes. Each section indicates the
scope levels available and describes the metrics within each class.

Database server performance depends on many factors, including the

operating-system configuration, the database server configuration, and the
workload. It is difficult to describe relationships between onperf metrics and
specific performance characteristics.

The approach taken here is to describe each metric without speculating on what
specific performance problems it might indicate. Through experimentation, you can
determine which metrics best monitor performance for a specific database server
instance.

Database server metrics

The onperf utility displays metrics for the named database server, rather than a
component of the database server or disk space.

The onperf utility displays the following database server metrics.

Chapter 14. The onperf utility on UNIX 14-13

 Metric Name Description
CPU System Time System time, as defined by the platform vendor
CPU User Time User time, as defined by the platform vendor
Percent Cached (Read) Percentage of all read operations that are read from the buffer
cache without requiring a disk read, calculated as follows:
100 * ((buffer_reads — disk_reads) /
(buffer_reads))
Percent Cached (Write) Percentage of all write operations that are buffer writes,
calculated as follows:
100 * ((buffer_writes — disk_writes) /
(buffer_writes))
Disk Reads Total number of read operations from disk
Disk Writes Total number of write operations to disk
Page Reads Number of pages read from disk
Page Writes Number of pages transferred to disk
Buffer Reads Number of reads from the buffer cache
Buffer Writes Number of writes to the buffer cache
Calls Number of calls received at the database server
Reads Number of read calls received at the database server
Writes Number of write calls received at the database server
Rewrites Number of rewrite calls received at the database server
Deletes Number of delete calls received at the database server
Commits Number of commit calls received at the database server
Rollbacks Number of rollback calls received at the database server
Table Overflows Number of times that the tblspace table was unavailable
(overflowed)
Lock Overflows Number of times that the lock table was unavailable
(overflowed)
User Overflows Number of times that the user table was unavailable
(overflowed)
Checkpoints Number of checkpoints written since database server shared
memory began
Buffer Waits Number of times that a thread waited to access a buffer
Lock Waits Number of times that a thread waited for a lock
Lock Requests Number of times that a lock was requested
Deadlocks Number of deadlocks detected
Deadlock Timeouts Number of deadlock timeouts that occurred (Deadlock timeouts
involve distributed transactions.)
Checkpoint Waits Number of checkpoint waits; in other words, the number of
times that threads have waited for a checkpoint to complete
Index to Data Pages Number of read-ahead operations for index keys
Read-aheads
Index Leaves Read-aheads Number of read-ahead operations for index leaf nodes
Data-path-only Number of read-ahead operations for data pages
Read-aheads
Latch Requests Number of latch requests

14-14 IBM Informix Performance Guide

 Metric Name Description
Network Reads Number of ASF messages read
Network Writes Number of ASF messages written
Memory Allocated Amount of database server virtual-address space in kilobytes
Memory Used Amount of database server shared memory in kilobytes
Temp Space Used Amount of shared memory allocated for temporary tables in
kilobytes
PDQ Calls The total number of parallel-processing actions that the database
server performed
DSS Memory Amount of memory currently in use for decision-support
queries

Disk-chunk metrics
The onperf utility can display metrics for a specific disk chunk.

The disk-chunk metrics take the path name of a chunk as the metric scope.

Metric Name Description

Disk Operations Total number of I/O operations to or from the indicated chunk
Disk Reads Total number of reads from the chunk
Disk Writes Total number of writes to the chunk
Free Space (MB) The amount of free space available in megabytes

Disk-spindle metrics
The onperf utility can display metrics for a disk spindle.

The disk-spindle metrics take the path name of a disk device or operation-system
file as the metric scope.

Metric Name Description

Disk Operations Total number of I/O operations to or from the indicated disk or
buffered operating-system file
Disk Reads Total number of reads from the disk or operating-system file
Disk Writes Total number of writes to the disk or operating-system file
Free Space The amount of free space available in megabytes

Physical-processor metrics
The onperf utility can display CPU metrics.

The physical-processor metrics take either a physical-processor identifier (for

example, 0 or 1) or Total as the metric scope.

Metric Name Description

Percent CPU System Time CPU system time for the physical processors
Percent CPU User Time CPU user time for the physical processors
Percent CPU Idle Time CPU idle time for the physical processors

Chapter 14. The onperf utility on UNIX 14-15

 Metric Name Description
Percent CPU Time The sum of CPU system time and CPU user time for the
physical processors

Virtual-processor metrics
The onperf utility can display metrics for a virtual-processor class.

These metrics take a virtual-processor class as a metric scope (cpu, aio, kaio, and
so on). Each metric value represents a sum across all instances of this
virtual-processor class.

Metric Name Description

User Time Accumulated user time for a class
System Time Accumulated system time for a class
Semaphore Operations Total count of semaphore operations
Busy Waits Number of times that virtual processors in class avoided a
context switch by spinning in a loop before going to sleep
Spins Number of times through the loop
I/O Operations Number of I/O operations per second
I/O Reads Number of read operations per second
I/O Writes Number of write operations per second

Session metrics
The onperf utility can display metrics for an active session.

These metrics take the active session as the metric scope.

Metric Name Description

Page Reads Number of pages read from disk on behalf of a session
Page Writes Number of pages written to disk on behalf of a session
Number of Threads Number of threads currently running for the session
Lock Requests Number of lock requests issued by the session
Lock Waits Number of lock waits for session threads
Deadlocks Number of deadlocks involving threads that belong to the
session
Deadlock timeouts Number of deadlock timeouts involving threads that belong to
the session
Log Records Number of log records written by the session
ISAM Calls Number of ISAM calls by session
ISAM Reads Number of ISAM read calls by session
ISAM Writes Number of ISAM write calls by session
ISAM Rewrites Number of ISAM rewrite calls by session
ISAM Deletes Number of ISAM delete calls by session
ISAM Commits Number of ISAM commit calls by session
ISAM Rollbacks Number of ISAM rollback calls by session

14-16 IBM Informix Performance Guide

 Metric Name Description
Long Transactions Number of long transactions owned by session
Buffer Reads Number of buffer reads performed by session
Buffer Writes Number of buffer writes performed by session
Log Space Used Amount of logical-log space used
Maximum Log Space High-watermark of logical-log space used for this session
Used
Sequential Scans Number of sequential scans initiated by session
PDQ Calls Number of parallel-processing actions performed for queries
initiated by the session
Memory Allocated Memory allocated for the session in kilobytes
Memory Used Memory used by the session in kilobytes

Tblspace metrics
The onperf utility can display metrics for a particular tblspace.

A tblspace name is composed of the database name, a colon, and the table name
(database:table).

For fragmented tables, the tblspace represents the sum of all fragments in a table.
To obtain measurements for an individual fragment in a fragmented table, use the
Fragment Metric class.

Metric Name Description

Lock Requests Total requests to lock tblspace
Lock Waits Number of times that threads waited to obtain a lock for the
tblspace
Deadlocks Number of times that a deadlock involved the tblspace
Deadlock Timeouts Number of times that a deadlock timeout involved the tblspace
Reads Number of read calls that involve the tblspace
Writes Number of write calls that involve the tblspace
Rewrites Number of rewrite calls that involve the tblspace
Deletes Number of delete calls that involve the tblspace
Calls Total calls that involve the tblspace
Buffer Reads Number of buffer reads that involve tblspace data
Buffer Writes Number of buffer writes that involve tblspace data
Sequential Scans Number of sequential scans of the tblspace
Percent Free Space Percent of the tblspace that is free
Pages Allocated Number of pages allocated to the tblspace
Pages Used Number of pages allocated to the tblspace that have been
written
Data Pages Number of pages allocated to the tblspace that are used as data
pages

Chapter 14. The onperf utility on UNIX 14-17

 Fragment metrics
The onperf utility can display metrics for an individual table fragment.

These metrics take the dbspace of an individual table fragment as the metric scope.

Metric Name Description

Lock Requests Total requests to lock fragment
Lock Waits Number of times that threads have waited to obtain a lock for
the fragment
Deadlocks Number of times that a deadlock involved the fragment
Deadlock Timeouts Number of times that a deadlock timeout involved the fragment
Reads Number of read calls that involve the fragment
Writes Number of write calls that involve the fragment
Rewrites Number of rewrite calls that involve the fragment
Deletes Number of delete calls that involve the fragment
Calls Total calls that involve the fragment
Buffer Reads Number of buffer reads that involve fragment data
Buffer Writes Number of buffer writes that involve fragment data
Sequential Scans Number of sequential scans of the fragment
Percent Free Space Percent of the fragment that is free
Pages Allocated Number of pages allocated to the fragment
Pages Used Number of pages allocated to the fragment that have been
written to
Data Pages Number of pages allocated to the fragment that are used as data
pages

14-18 IBM Informix Performance Guide

Appendix A. Case studies and examples
This appendix contains a case study with examples of performance-tuning
methods that this publication describes.

Case study of a situation in which disks are overloaded

You can identify overloaded disks and the dbspaces that reside on those disks.
After you identify the overloaded disks, you can correct the problem.

The following case study illustrates a situation in which the disks are overloaded.
This study shows the steps taken to isolate the symptoms and identify the problem
based on an initial report from a user, and it describes the needed correction.

A database application that does not have the wanted throughput is being
examined to see how performance can be improved. The operating-system
monitoring tools reveal that a high proportion of process time was spent idle,
waiting for I/O. The database server administrator increases the number of CPU
VPs to make more processors available to handle concurrent I/O. However,
throughput does not increase, which indicates that one or more disks are
overloaded.

To verify the I/O bottleneck, the database server administrator must identify the
overloaded disks and the dbspaces that reside on those disks.

To identify overloaded disks and the dbspaces that reside on those disks:
1. To check the asynchronous I/O (AIO) queues, use onstat -g ioq. Figure A-1
shows the output.

AIO I/O queues:

q name/id len maxlen totalops dskread dskwrite dskcopy
opt 0 0 0 0 0 0 0
msc 0 0 0 0 0 0 0
aio 0 0 0 0 0 0 0
pio 0 0 1 1 0 1 0
lio 0 0 1 341 0 341 0
gfd 3 0 1 225 2 223 0
gfd 4 0 1 225 2 223 0
gfd 5 0 1 225 2 223 0
gfd 6 0 1 225 2 223 0
gfd 7 0 0 0 0 0 0
gfd 8 0 0 0 0 0 0
gfd 9 0 0 0 0 0 0
gfd 10 0 0 0 0 0 0
gfd 11 0
28
32693 29603 3090 0
gfd 12 0
18
32557 29373 3184 0
gfd 13 0
22
20446 18496 1950 0

Figure A-1. Output from the onstat -g ioq option

In Figure A-1, the maxlen and totalops columns show significant results:
v The maxlen column shows the largest backlog of I/O requests to accumulate
within the queue. The last three queues are much longer than any other
queue in this column listing.

© Copyright IBM Corp. 1996, 2014 A-1

 v The totalops column shows 100 times more I/O operations completed
through the last three queues than for any other queue in the column listing.
The maxlen and totalops columns indicate that the I/O load is severely
unbalanced.
Another way to check I/O activity is to use onstat -g iov. This option provides
a slightly less detailed display for all VPs.
2. To check the AIO activity for each disk device associated with each queue, use
onstat -g iof, as Figure A-2 shows.

gfd pathname bytes read page reads bytes write page writes io/s
3
/dev/infx5 85456896 41727 207394816 101267 572.9
op type count avg. time
seeks 0 N/A
reads 13975 0.0015
writes 51815 0.0018
kaio_reads 0 N/A
kaio_writes 0 N/A

Figure A-2. Partial output from the onstat -g iof option

Depending on how your chunks are arranged, several queues can be associated
with the same device.
3. To determine the dbspaces that account for the I/O load, use onstat -d, as
Figure A-3 shows.

Dbspaces
address number flags fchunk nchunks flags owner name
c009ad00 1 1 1 1 N informix rootdbs
c009ad44 2 2001 2 1 N T informix tmp1dbs
c009ad88 3 1 3 1 N informix oltpdbs
c009adcc 4 1 4 1 N informix histdbs
c009ae10 5 2001 5 1 N T informix tmp2dbs
c009ae54 6 1 6 1 N informix physdbs
c009ae98 7 1 7 1 N informix logidbs
c009aedc 8 1 8 1 N informix runsdbs
c009af20
9 1 9 3 N informix
acctdbs
9 active, 32 total

Chunks
address chk/dbs offset size free bpages flags pathname
c0099574 1 1 500000 10000 9100 PO- /dev/infx2
c009960c 2 2 510000 10000 9947 PO- /dev/infx2
c00996a4 3 3 520000 10000 9472 PO- /dev/infx2
c009973c 4 4 530000 250000 242492 PO- /dev/infx2
c00997d4 5 5 500000 10000 9947 PO- /dev/infx4
c009986c 6 6 510000 10000 2792 PO- /dev/infx4
c0099904 7 7 520000 25000 11992 PO- /dev/infx4
c009999c 8 8 545000 10000 9536 PO- /dev/infx4
c0099a34 9
9 250000 450000 4947 PO-
/dev/infx5
c0099acc 10
9 250000 450000 4997 PO-
/dev/infx6
c0099b64 11
9 250000 450000 169997 PO-
/dev/infx7
11 active, 32 total

Figure A-3. Output from the onstat -d option

In the Chunks output, the pathname column indicates the disk device. The
chk/dbs column indicates the numbers of the chunk and dbspace that reside on

A-2 IBM Informix Performance Guide

each disk. In this case, only one chunk is defined on each of the overloaded disks.
Each chunk is associated with dbspace number 9.

The Dbspaces output shows the name of the dbspace that is associated with each
dbspace number. In this case, all three of the overloaded disks are part of the
acctdbs dbspace.

Although the original disk configuration allocated three entire disks to the acctdbs
dbspace, the activity within this dbspace suggests that three disks are not enough.
Because the load is about equal across the three disks, it does not appear that the
tables are necessarily laid out badly or improperly fragmented. However, you
might get better performance by adding fragments on other disks to one or more
large tables in this dbspace or by moving some tables to other disks with lighter
loads.
Related reference:
onstat -g iof command: Print asynchronous I/O statistics (Administrator's
Reference)
onstat -g ioa command: Print combined onstat -g information (Administrator's
Reference)
onstat -g ioq command: Print I/O queue information (Administrator's
Reference)
onstat -g iov command: Print AIO VP statistics (Administrator's Reference)
onstat -d command: Print chunk information (Administrator's Reference)

Appendix A. Case studies and examples A-3

A-4 IBM Informix Performance Guide
Appendix B. Accessibility
IBM strives to provide products with usable access for everyone, regardless of age
or ability.

Accessibility features for IBM Informix products

Accessibility features help a user who has a physical disability, such as restricted
mobility or limited vision, to use information technology products successfully.

Accessibility features
The following list includes the major accessibility features in IBM Informix
products. These features support:
v Keyboard-only operation.
v Interfaces that are commonly used by screen readers.
v The attachment of alternative input and output devices.

Keyboard navigation
This product uses standard Microsoft Windows navigation keys.

Related accessibility information

IBM is committed to making our documentation accessible to persons with
disabilities. Our publications are available in HTML format so that they can be
accessed with assistive technology such as screen reader software.

IBM and accessibility

For more information about the IBM commitment to accessibility, see the IBM
Accessibility Center at http://www.ibm.com/able.

Dotted decimal syntax diagrams

The syntax diagrams in our publications are available in dotted decimal format,
which is an accessible format that is available only if you are using a screen reader.

In dotted decimal format, each syntax element is written on a separate line. If two
or more syntax elements are always present together (or always absent together),
the elements can appear on the same line, because they can be considered as a
single compound syntax element.

Each line starts with a dotted decimal number; for example, 3 or 3.1 or 3.1.1. To
hear these numbers correctly, make sure that your screen reader is set to read
punctuation. All syntax elements that have the same dotted decimal number (for
example, all syntax elements that have the number 3.1) are mutually exclusive
alternatives. If you hear the lines 3.1 USERID and 3.1 SYSTEMID, your syntax can
include either USERID or SYSTEMID, but not both.

The dotted decimal numbering level denotes the level of nesting. For example, if a
syntax element with dotted decimal number 3 is followed by a series of syntax
elements with dotted decimal number 3.1, all the syntax elements numbered 3.1
are subordinate to the syntax element numbered 3.

© Copyright IBM Corp. 1996, 2014 B-1

 Certain words and symbols are used next to the dotted decimal numbers to add
information about the syntax elements. Occasionally, these words and symbols
might occur at the beginning of the element itself. For ease of identification, if the
word or symbol is a part of the syntax element, the word or symbol is preceded by
the backslash (\) character. The * symbol can be used next to a dotted decimal
number to indicate that the syntax element repeats. For example, syntax element
*FILE with dotted decimal number 3 is read as 3 \* FILE. Format 3* FILE
indicates that syntax element FILE repeats. Format 3* \* FILE indicates that
syntax element * FILE repeats.

Characters such as commas, which are used to separate a string of syntax

elements, are shown in the syntax just before the items they separate. These
characters can appear on the same line as each item, or on a separate line with the
same dotted decimal number as the relevant items. The line can also show another
symbol that provides information about the syntax elements. For example, the lines
5.1*, 5.1 LASTRUN, and 5.1 DELETE mean that if you use more than one of the
LASTRUN and DELETE syntax elements, the elements must be separated by a comma.
If no separator is given, assume that you use a blank to separate each syntax
element.

If a syntax element is preceded by the % symbol, that element is defined elsewhere.

The string that follows the % symbol is the name of a syntax fragment rather than a
literal. For example, the line 2.1 %OP1 refers to a separate syntax fragment OP1.

The following words and symbols are used next to the dotted decimal numbers:
? Specifies an optional syntax element. A dotted decimal number followed
by the ? symbol indicates that all the syntax elements with a
corresponding dotted decimal number, and any subordinate syntax
elements, are optional. If there is only one syntax element with a dotted
decimal number, the ? symbol is displayed on the same line as the syntax
element (for example, 5? NOTIFY). If there is more than one syntax element
with a dotted decimal number, the ? symbol is displayed on a line by
itself, followed by the syntax elements that are optional. For example, if
you hear the lines 5 ?, 5 NOTIFY, and 5 UPDATE, you know that syntax
elements NOTIFY and UPDATE are optional; that is, you can choose one or
none of them. The ? symbol is equivalent to a bypass line in a railroad
diagram.
! Specifies a default syntax element. A dotted decimal number followed by
the ! symbol and a syntax element indicates that the syntax element is the
default option for all syntax elements that share the same dotted decimal
number. Only one of the syntax elements that share the same dotted
decimal number can specify a ! symbol. For example, if you hear the lines
2? FILE, 2.1! (KEEP), and 2.1 (DELETE), you know that (KEEP) is the
default option for the FILE keyword. In this example, if you include the
FILE keyword but do not specify an option, default option KEEP is applied.
A default option also applies to the next higher dotted decimal number. In
this example, if the FILE keyword is omitted, default FILE(KEEP) is used.
However, if you hear the lines 2? FILE, 2.1, 2.1.1! (KEEP), and 2.1.1
(DELETE), the default option KEEP only applies to the next higher dotted
decimal number, 2.1 (which does not have an associated keyword), and
does not apply to 2? FILE. Nothing is used if the keyword FILE is omitted.
* Specifies a syntax element that can be repeated zero or more times. A
dotted decimal number followed by the * symbol indicates that this syntax
element can be used zero or more times; that is, it is optional and can be

B-2 IBM Informix Performance Guide

 repeated. For example, if you hear the line 5.1* data-area, you know that
you can include more than one data area or you can include none. If you
hear the lines 3*, 3 HOST, and 3 STATE, you know that you can include
HOST, STATE, both together, or nothing.

Notes:
1. If a dotted decimal number has an asterisk (*) next to it and there is
only one item with that dotted decimal number, you can repeat that
same item more than once.
2. If a dotted decimal number has an asterisk next to it and several items
have that dotted decimal number, you can use more than one item
from the list, but you cannot use the items more than once each. In the
previous example, you can write HOST STATE, but you cannot write HOST
HOST.
3. The * symbol is equivalent to a loop-back line in a railroad syntax
diagram.
+ Specifies a syntax element that must be included one or more times. A
dotted decimal number followed by the + symbol indicates that this syntax
element must be included one or more times. For example, if you hear the
line 6.1+ data-area, you must include at least one data area. If you hear
the lines 2+, 2 HOST, and 2 STATE, you know that you must include HOST,
STATE, or both. As for the * symbol, you can repeat a particular item if it is
the only item with that dotted decimal number. The + symbol, like the *
symbol, is equivalent to a loop-back line in a railroad syntax diagram.

Appendix B. Accessibility B-3

B-4 IBM Informix Performance Guide
Notices
This information was developed for products and services offered in the U.S.A.

IBM may not offer the products, services, or features discussed in this document in
other countries. Consult your local IBM representative for information on the
products and services currently available in your area. Any reference to an IBM
product, program, or service is not intended to state or imply that only that IBM
product, program, or service may be used. Any functionally equivalent product,
program, or service that does not infringe any IBM intellectual property right may
be used instead. However, it is the user's responsibility to evaluate and verify the
operation of any non-IBM product, program, or service.

IBM may have patents or pending patent applications covering subject matter
described in this document. The furnishing of this document does not grant you
any license to these patents. You can send license inquiries, in writing, to:

IBM Director of Licensing

IBM Corporation
North Castle Drive
Armonk, NY 10504-1785
U.S.A.

For license inquiries regarding double-byte (DBCS) information, contact the IBM
Intellectual Property Department in your country or send inquiries, in writing, to:

Intellectual Property Licensing

Legal and Intellectual Property Law
IBM Japan, Ltd.
19-21, Nihonbashi-Hakozakicho, Chuo-ku
Tokyo 103-8510, Japan

The following paragraph does not apply to the United Kingdom or any other
country where such provisions are inconsistent with local law: INTERNATIONAL
BUSINESS MACHINES CORPORATION PROVIDES THIS PUBLICATION "AS IS"
WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESS OR IMPLIED,
INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
NON-INFRINGEMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR
PURPOSE. Some states do not allow disclaimer of express or implied warranties in
certain transactions, therefore, this statement may not apply to you.

This information could include technical inaccuracies or typographical errors.

Changes are periodically made to the information herein; these changes will be
incorporated in new editions of the publication. IBM may make improvements
and/or changes in the product(s) and/or the program(s) described in this
publication at any time without notice.

Any references in this information to non-IBM websites are provided for

convenience only and do not in any manner serve as an endorsement of those
websites. The materials at those websites are not part of the materials for this IBM
product and use of those websites is at your own risk.

© Copyright IBM Corp. 1996, 2014 C-1

 IBM may use or distribute any of the information you supply in any way it
believes appropriate without incurring any obligation to you.

Licensees of this program who wish to have information about it for the purpose
of enabling: (i) the exchange of information between independently created
programs and other programs (including this one) and (ii) the mutual use of the
information which has been exchanged, should contact:

IBM Corporation
J46A/G4
555 Bailey Avenue
San Jose, CA 95141-1003
U.S.A.

Such information may be available, subject to appropriate terms and conditions,

including in some cases, payment of a fee.

The licensed program described in this document and all licensed material
available for it are provided by IBM under terms of the IBM Customer Agreement,
IBM International Program License Agreement or any equivalent agreement
between us.

Any performance data contained herein was determined in a controlled

environment. Therefore, the results obtained in other operating environments may
vary significantly. Some measurements may have been made on development-level
systems and there is no guarantee that these measurements will be the same on
generally available systems. Furthermore, some measurements may have been
estimated through extrapolation. Actual results may vary. Users of this document
should verify the applicable data for their specific environment.

Information concerning non-IBM products was obtained from the suppliers of

those products, their published announcements or other publicly available sources.
IBM has not tested those products and cannot confirm the accuracy of
performance, compatibility or any other claims related to non-IBM products.
Questions on the capabilities of non-IBM products should be addressed to the
suppliers of those products.

All statements regarding IBM's future direction or intent are subject to change or
withdrawal without notice, and represent goals and objectives only.

All IBM prices shown are IBM's suggested retail prices, are current and are subject
to change without notice. Dealer prices may vary.

This information is for planning purposes only. The information herein is subject to
change before the products described become available.

This information contains examples of data and reports used in daily business
operations. To illustrate them as completely as possible, the examples include the
names of individuals, companies, brands, and products. All of these names are
fictitious and any similarity to the names and addresses used by an actual business
enterprise is entirely coincidental.

COPYRIGHT LICENSE:

This information contains sample application programs in source language, which

illustrate programming techniques on various operating platforms. You may copy,

C-2 IBM Informix Performance Guide

 modify, and distribute these sample programs in any form without payment to
IBM, for the purposes of developing, using, marketing or distributing application
programs conforming to the application programming interface for the operating
platform for which the sample programs are written. These examples have not
been thoroughly tested under all conditions. IBM, therefore, cannot guarantee or
imply reliability, serviceability, or function of these programs. The sample
programs are provided "AS IS", without warranty of any kind. IBM shall not be
liable for any damages arising out of your use of the sample programs.

Each copy or any portion of these sample programs or any derivative work, must
include a copyright notice as follows:

© (your company name) (year). Portions of this code are derived from IBM Corp.
Sample Programs.

© Copyright IBM Corp. _enter the year or years_. All rights reserved.

If you are viewing this information softcopy, the photographs and color
illustrations may not appear.

Privacy policy considerations

IBM Software products, including software as a service solutions, (“Software
Offerings”) may use cookies or other technologies to collect product usage
information, to help improve the end user experience, to tailor interactions with
the end user or for other purposes. In many cases no personally identifiable
information is collected by the Software Offerings. Some of our Software Offerings
can help enable you to collect personally identifiable information. If this Software
Offering uses cookies to collect personally identifiable information, specific
information about this offering’s use of cookies is set forth below.

This Software Offering does not use cookies or other technologies to collect
personally identifiable information.

If the configurations deployed for this Software Offering provide you as customer
the ability to collect personally identifiable information from end users via cookies
and other technologies, you should seek your own legal advice about any laws
applicable to such data collection, including any requirements for notice and
consent.

For more information about the use of various technologies, including cookies, for
these purposes, see IBM’s Privacy Policy at http://www.ibm.com/privacy and
IBM’s Online Privacy Statement at http://www.ibm.com/privacy/details the
section entitled “Cookies, Web Beacons and Other Technologies” and the “IBM
Software Products and Software-as-a-Service Privacy Statement” at
http://www.ibm.com/software/info/product-privacy.

Trademarks
IBM, the IBM logo, and ibm.com are trademarks or registered trademarks of
International Business Machines Corp., registered in many jurisdictions worldwide.
Other product and service names might be trademarks of IBM or other companies.
A current list of IBM trademarks is available on the web at "Copyright and
trademark information" at http://www.ibm.com/legal/copytrade.shtml.

Notices C-3
 Adobe, the Adobe logo, and PostScript are either registered trademarks or
trademarks of Adobe Systems Incorporated in the United States, and/or other
countries.

Intel, Itanium, and Pentium are trademarks or registered trademarks of Intel

Corporation or its subsidiaries in the United States and other countries.

Linux is a registered trademark of Linus Torvalds in the United States, other

countries, or both.

Microsoft, Windows, and Windows NT are trademarks of Microsoft Corporation in

the United States, other countries, or both.

Java and all Java-based trademarks and logos are trademarks or registered
trademarks of Oracle and/or its affiliates.

UNIX is a registered trademark of The Open Group in the United States and other
countries.

Other company, product, or service names may be trademarks or service marks of

others.

C-4 IBM Informix Performance Guide

Index
Special characters ALTER TABLE statement (continued)
in-place 6-26, 6-35, 10-27
$INFORMIXDIR/bin directory 14-4 in-place alter 6-36
$INFORMIXDIR/help directory 14-4 sbspace fragmentation 9-6
slow alter 6-36
slow alter algorithm 6-35
Numerics smart large objects 9-6
64-bit addressing Alters
buffers 4-10 in-place 6-42
tuning RESIDENT configuration parameter 4-17 slow 6-35
ANSI
Repeatable Read isolation level 8-7
A Serializable isolation level 8-7
ANSI-compliant database
Access method access-method name 7-24
ANSI-compliant name 7-24 Application developer
directives 11-4 general responsibility 1-14
list 7-22 setting PDQ priority 12-9
secondary 7-22, 7-25 SQLWARN array 5-29
Access plan Assigning table to a dbspace 6-1
defined 10-1 Association records 11-15
directives 11-4 Attached indexes
effects of OPTCOMPIND 10-22 creating 9-10
SET EXPLAIN output 10-13, 12-18 defined 9-10
subquery 10-15 extent size 7-5
Accessibility B-1 fragmentation 9-10
dotted decimal format of syntax diagrams B-1 physical characteristics 9-10
keyboard B-1 Auditing
shortcut keys B-1 facility 1-4
syntax diagrams, reading in a screen reader B-1 performance, and 5-45
Activity tools (onperf) AUDITPATH configuration parameter 5-45
defined 14-3 AUDITSIZE configuration parameter 5-45
onperf, using 14-12 AUS
ADTERR configuration parameter 5-45 expiration policies 13-8
ADTMODE configuration parameter 5-45 expiration policies, changing 13-9
Affinity viewing UPDATE STATISTICS statements 13-10
setting for processor 3-7 aus_cmd_comp 13-7
VPCLASS configuration parameter 3-7 aus_cmd_info 13-7
AIO aus_cmd_list 13-7
queues A-1 Auto Update Statistics Evaluation 13-7
virtual processors Auto Update Statistics Refresh 13-7
monitoring 3-22 AUTO_AIOVPS configuration parameter 3-8, 3-21, 5-30
VPs 3-8 AUTO_CKPTS configuration parameter 5-30
Algorithm, in-place alter 6-26, 10-27 AUTO_REPREPARE configuration parameter 11-13
Alice scan mode 13-22, 13-25 auto_tune_cpu_vps task 3-20
ALTER FRAGMENT statement Automated UPDATE STATISTICS 13-6
eliminating index build during DETACH 9-26, 9-27 disabling 13-11
least-cost index build during ATTACH 9-20, 9-21, 9-23, expiration policies 13-8
9-24 expiration policies, changing 13-9
moving table 6-1 ph_task table 13-10, 13-11
releasing space 6-27 rescheduling 13-10
when FORCE_DDL_EXEC is enabled 9-27 sequence of events 13-7
ALTER INDEX statement 6-25, 6-27, 7-11 viewing generated statements 13-10
TO CLUSTER clause 6-25
ALTER TABLE statement
adding or dropping a column 6-26
changing data type 6-26 B
changing extent sizes 6-20, 6-22 B-tree
changing lock mode 8-4 defined 7-1
changing sbspace characteristics 6-19 estimating index pages 7-4, 7-5
columns part of an index 6-41 generic 7-23, 13-38
fast alter algorithm 6-42 index usage 7-22

© Copyright IBM Corp. 1996, 2014 X-1

B-tree scanner BUFFERPOOL configuration parameter 4-2, 4-10, 5-20, 5-30,
alice mode 13-22, 13-25 5-40, 5-45
compression level 13-27 Buffers
configuring to improve transaction processing 13-22 data replication 4-39
index compression level 13-28 free network 3-18
leaf mode 13-27 lightweight I/O 5-23, 5-24
leaf scan mode 13-22 logical log 4-15, 5-20
range mode 13-27 network 3-18
scan modes 13-22, 13-25, 13-27 network, monitoring 3-18
Background I/O activities 5-29 physical log 4-15
Backup and restore smart large objects 5-24
fragmentation strategy for 9-4 TCP/IP connections 3-17
Backups Built-in data types
and restore B-tree index 7-1
table placement 6-5, 9-6 B-tree index, generic 7-23
BAR_MAX_BACKUP configuration parameter 5-41 functional index 7-1
BAR_NB_XPORT_COUNT configuration parameter 5-41 BYTE data type
BAR_PROGRESS_FREQ configuration parameter 5-41 blobspace 5-15
BAR_XFER_BUF_SIZE configuration parameter 5-41 estimating table size 6-5
BATCHEDREAD_TABLE configuration parameter 5-27 locating 6-9
Benchmarks, for throughput 1-4 memory cache 5-24
BLOB data type on disk 6-9
defined 5-6 parallel access 5-20
Blobpage staging area 5-24
estimating number in tblspace 6-8 storing 6-9
fullness explained 5-19 Byte-range locking
fullness, determining 5-18 byte lock 8-10
fullness, interpreting average 5-19 defined 8-16
logical log size 5-37 monitoring 8-18
oncheck -pB display 5-18 setting 8-17
oncheck utility
blobpage information 5-18
size 5-16
size and storage efficiency 5-17
C
Cache
sizing in blobspace 5-16
aggregate 10-33
storage statistics 5-17
data dictionary 4-21, 4-22, 4-24
when to store in blobspace 6-9
data distribution 4-21, 4-23, 4-24
Blobspaces
defined 4-21
advantages over dbspace 5-16
opclass 10-33
configuration effects 5-15
SQL statements 4-25
determining fullness 5-17
typename 10-33
Parallel
UDRs 10-33
access to table and simple large objects 5-16
Caches
simple large objects 6-9
private memory 3-24
Simple large objects
Cardinality
parallel access 5-16
changes, and UPDATE STATISTICS 13-12
specifying in CREATE TABLE 5-15
Case studies, extended A-1
storage statistics 5-17
Central processing unit
when to use 6-9
configuration parameters that affect 3-3
BOUND_IMPL_PDQ session environment variable 12-14
connections and 3-25, 3-26
Branch index pages 7-1
environment variables that affect 3-3
btree_ops operator class 7-29, 7-30
utilization and 1-9
BTSCANNER configuration parameter 13-28
VPs and 3-20
Buffer pools
CHAR data type
64-bit addressing 4-10, 4-17
converting to VARCHAR 6-43
BUFFERPOOL configuration parameter 4-10
GLS recommendations 10-29
bypass with light scans 5-27
key-only scans 10-1
bypass with lightweight I/O 5-23
Checking indexes 7-21
for non-default page sizes 4-10
Checkpoints
LRU queues 5-40
automatic 5-30
network 3-17, 3-18
Checkpoints
read cache rate 4-10
when occur 5-31
size, smart large objects 5-20
configuration parameters affecting 5-30
smart large objects 4-10, 5-20, 5-23
defined 5-31
Buffered
flushing of regular buffers 5-45
logging 5-7
logging and performance 5-34
monitoring 5-31

X-2 IBM Informix Performance Guide

Checkpoints (continued) Configuration parameters (continued)
physical log, effect on 5-32 affecting
specifying interval 5-31 auditing 5-45
when occur 5-29 backup and restore 5-41
Chunks checkpoints 5-30
critical data 5-5, 6-24 connections 3-18
dbspace configuration, and 5-1 CPU 3-3
disk partitions, and 5-2 critical data 5-8
CKPTINTVL configuration parameter 5-31 data dictionary 4-22, 4-24
Class name, virtual processors 3-4 data distributions 4-24
CLEANERS configuration parameter 5-40 ipcshm connection 3-14, 4-6
CLOB data type 5-6 logging I/O 5-34
Clustered index 6-25, 7-17 logical log 5-8
Clustering memory 4-8
configuration parameters that affect it 7-12 network free buffer 3-18
defined 7-11 ON-Bar utility 5-41
index for sequential access 10-26 page cleaning 5-39
Collection-derived table physical log 5-8
defined 10-16 poll threads 3-1, 3-13, 3-20
folded into parent query 10-17 recovery 5-42
improving performance 10-17 rollback 5-42
query plan for 10-16 root dbspace 5-8
Collections SQL statement cache 4-31, 13-42
scan 10-16 SQL statement cache cleaning 4-31
Columns SQL statement cache hits 4-21, 4-27, 4-28, 4-29, 4-32,
filter expression, with join 10-5 4-33, 4-34, 4-36
filtered 7-10 SQL statement cache memory 4-21, 4-27
with duplicate keys 7-11 SQL statement cache pools 4-34
Commands SQL statement cache size 4-21, 4-33
UNIX SQL statement memory limit 4-33
cron 4-7 UDR cache buckets 10-33
iostat 2-3 UDR cache entries 10-33
ps 2-3 AUDITPATH 5-45
sar 2-3, 4-10 AUDITSIZE 5-45
time 1-6 AUTO_AIOVPS 3-8, 3-21, 5-30
vmstat 2-3, 4-10 AUTO_CKPTS 5-30
COMMIT WORK statement 1-4 AUTO_REPREPARE 11-13
Committed Read isolation level 5-27, 8-6 BAR_MAX_BACKUP 5-41
Committed Read Last Committed isolation level 5-27 BAR_NB_XPORT_COUNT 5-41
Complex query, example of 10-13 BAR_PROGRESS_FREQ 5-41
compliance with standards xxii BAR_XFER_BUF_SIZE 5-41
Composite index 13-20 BTSCANNER 13-28
order of columns 13-20 BUFFERPOOL 4-2, 4-10, 5-20, 5-30, 5-45
use of 13-20 CKPTINTVL 5-31
Compressing CLEANERS 5-40
fragments 6-47 controlling PDQ resources 12-6
tables 6-47 CPU, and 3-1
Compression DATASKIP 5-29
benefits 6-47 DBSPACETEMP 5-8, 5-11, 5-13, 6-4, 7-18
Concurrency DD_HASHMAX 4-21, 4-22
defined 8-1 DD_HASHSIZE 4-21, 4-22
effects of isolation level 8-5 DEADLOCK_TIMEOUT 8-15
fragmentation 9-2 DEF_TABLE_LOCKMODE 8-4
isolation level, effects of 8-5, 10-2 DIRECT_IO 5-3, 5-4
locks, page 8-12 DIRECTIVES 11-12, 11-13
locks, row and key 8-1 DRAUTO 5-44
locks, table 8-3, 8-12 DRINTERVAL 5-44
page lock on index 8-2 DRLOSTFOUND 5-44
Concurrent I/O DRTIMEOUT 5-44
confirming use of 5-4 DS_HASHSIZE 4-21, 4-24
enabling 5-4 DS_MAX_QUERIES 3-12
overview 5-2, 5-4 DS_MAX_SCANS 3-12, 12-6, 12-11
Configuration DS_POOLSIZE 4-21, 4-24
evaluating 2-1 DS_TOTAL_MEMORY 4-12, 7-18, 12-6
Configuration parameters FASTPOLL 3-17
ADTERR 5-45 FILLFACTOR 7-5
ADTMODE 5-45 HDR_TXN_SCOPE 5-44

Index X-3
Configuration parameters (continued) Configuration parameters (continued)
INFORMIXOPCACHE 5-26 VP_MEMORY_CACHE_KB 3-24
LOCKBUFF 4-2 VPCLASS 3-4, 3-5, 3-6, 3-7, 3-8
LOCKS 4-2, 4-15, 8-12 CONNECT statement 5-2
LOGBUFF 4-15, 5-8, 5-20, 5-34 Connections
LOGFILES 5-32 CPU 3-25, 3-26
LOGSIZE 5-32, 5-35, 5-36 improving performance 3-16
LOW_MEMORY_RESERVE 4-15, 5-43 improving performance with MaxConnect 3-26
LTAPEBLK 5-42 multiplexed 3-25
LTAPEDEV 5-42 specifying number of 3-14
LTAPESIZE 5-42 type, ipcshm 3-1, 3-14
LTXEHWM 5-38 type, specifying 3-13, 3-14
LTXHWM 5-38 Constraints
MAX_FILL_DATA_PAGES 6-46 foreign-key 6-32
MAX_PDQPRIORITY 3-3, 3-11, 12-9, 12-11, 12-15, 13-33 referential 6-32
MIRROR 5-8 Contention
MIRROROFFSET 5-8 cost of reading a page 10-25
MIRRORPATH 5-8 reducing with fragmentation 9-3
MULTIPROCESSOR 3-10 Contiguous
NETTYPE 3-1, 3-14, 3-16, 3-18, 3-20, 4-6 disk space, allocation 6-23
NS_CACHE 3-16 extents
NUMFDSERVERS 3-16 advantage of performance 5-21, 6-13, 6-19, 6-24, 6-25
OFF_RECVRY_THREADS 5-42 space, eliminating interleaved extents 6-25
ON_RECVRY_THREADS 5-42 Cooked file space 5-2, 5-3
ONDBSPACEDOWN 5-33 performance using concurrent I/O 5-2, 5-4
ONLIDX_MAXMEM 7-18 performance using direct I/O 5-2, 5-3
OPCACHEMAX 5-26 Correlated subquery
OPT_GOAL 13-35 effect of PDQ 12-5
OPTCOMPIND 3-3, 3-10, 11-12, 12-13 Cost of user-defined routine 13-38, 13-39
PC_HASHSIZE 4-21, 10-33 Cost per transaction 1-7
PC_POOLSIZE 4-21, 10-33 CPU
PHYSBUFF 4-2, 4-15, 5-34 utilization, improving with MaxConnect 3-26
PHYSFILE 5-32 VP class and NETTYPE 3-14
PLCY_HASHSIZE 4-21 VPs
PLCY_POOLSIZE 4-21 configuration parameters affecting 3-5
PLOG_OVERFLOW_PATH 5-43 effect on CPU utilization 3-20
RESIDENT 4-17 limited by MAX_PDQPRIORITY 3-11
ROOTNAME 5-8 limited by PDQ priority 3-3
ROOTOFFSET 5-8 optimal number 3-10
ROOTPATH 5-8 used by PDQ 12-7
ROOTSIZE 5-8 VPs and fragmentation goals 9-2
RTO_SERVER_RESTART 5-30, 5-31, 5-41, 5-43 CPU VPs
SBSPACENAME 5-14, 5-20 adding automatically 3-20
SBSPACETEMP 5-14, 5-15 CREATE CLUSTER INDEX statement 7-11
SHMADD 4-2 CREATE CLUSTERED INDEX statement 3-3
SHMBASE 4-8 CREATE FUNCTION statement
SHMMAX 4-18, 4-19 selectivity and cost 13-39
SHMTOTAL 4-2, 4-18 virtual-processor class 3-4
SHMVIRT_ALLOCSEG 4-19 CREATE INDEX ONLINE statement 7-16, 7-17
SHMVIRTSIZE 4-2, 4-4, 4-19 CREATE INDEX statement
SINGLE_CPU_VP 3-10 attached index 9-10
STACKSIZE 4-20 detached index 9-12
STAGEBLOB 5-25 FILLFACTOR clause 7-5
STMT_CACHE 13-42 generic B-tree index 7-23
STMT_CACHE_HITS 4-21, 4-27, 4-28, 4-29, 4-30, 4-32, parallel build 12-3
4-34, 4-36 TO CLUSTER clause 6-25
STMT_CACHE_NOLIMIT 4-21, 4-27 USING clause 7-24
STMT_CACHE_NUMPOOL 4-34 CREATE PROCEDURE statement
STMT_CACHE_SIZE 4-21, 4-31, 4-33 SPL routines, optimizing 10-31
TAPEBLK 5-42 SQL, optimizing 10-31
TAPEDEV 5-42 CREATE TABLE statement
TAPESIZE 5-42 blobspace assignment 5-15
TBLTBLFIRST 6-10 creating system catalog table 5-2
TBLTBLNEXT 6-10 extent sizes 6-20
USELASTCOMMITTED 8-6 fragmenting 9-10, 9-12
USRC_HASHSIZE 4-21 with partitions 9-10, 9-12
USRC_POOLSIZE 4-21 PUT clause 6-19

X-4 IBM Informix Performance Guide

CREATE TABLE statement (continued) Data-dictionary (continued)
sbspace characteristics 6-19 advantages 4-22
sbspace fragmentation 9-6 configuring 4-22
simple large objects 6-9 Data-dictionary cache
smart large objects 9-6 configuring 4-24
TEMP TABLE clause 5-8, 5-14 Data-distribution cache
USING clause 7-31 defined 4-23
CREATE TEMP TABLE statement 9-13 monitoring 4-24
Critical data Database server administrator
configuration parameters that affect 5-8 allocating DSS memory 12-11
defined 5-33 controlling DSS resources 3-11, 12-15
introduced 5-5 creating staging-area blobspace 5-24
mirroring 5-6 halting database server 5-33
Critical media limiting number of DSS queries 12-12
mirroring 5-6 limiting PDQ priority 12-15
separating 5-5 marking dbspace down 5-33
Critical resource 1-7 placing system catalog tables 5-2
cron responsibility of 1-14, 5-1
UNIX scheduling facility 2-3, 2-4, 4-7 specifying unavailable fragments 9-3
Cursor Stability isolation level 8-7 using MAX_PDQPRIORITY 12-15
DATABASE statement 5-2
DataBlade API functions, smart large objects 5-20, 5-21, 6-17,
D 6-23, 8-20
DataBlade modules
Data
functional index 7-26
migration between fragments 9-22
new index 7-27
transfers per second 1-11
secondary access method 7-22
Data conversion 10-29
user-defined index 7-4, 7-21
Data dictionary
DATASKIP configuration parameter 5-29
DD_HASHMAX 4-22
DB-Access utility 2-4, 6-25
DD_HASHSIZE 4-22
dbload utility 6-25, 7-12
parameters affecting cache for 4-24
dbschema utility
Data distributions
data distributions 9-8
creating 10-20
distribution output 13-16, 13-19
creating on filtered columns 11-3
examining value distribution 9-4
dropping 13-13
dbspaces
effect on memory 4-2
chunk configuration 5-1
filter selectivity 10-21
configuration parameters affecting root 5-8
guidelines to create 13-13
mirroring root 5-6
how optimizer uses 10-20
page size, specifying 7-8
join columns 13-16
reorganizing to prevent extent interleaving 6-25
multiple columns 13-18
specifying page size when creating 4-10
parameters affect cache for 4-24
temporary tables and sort files 5-8, 6-4
sbspaces 13-16
DBSPACETEMP
syscolumns 13-13, 13-16
parallel inserts 12-2
sysdistrib 13-13
DBSPACETEMP configuration parameter 5-8, 5-11, 6-4, 7-18
user-defined data type 13-16
overriding 5-12
user-defined statistics 13-16, 13-40
DBSPACETEMP environment variable 5-8, 6-4, 7-18
Data replication
advantages over PSORT_DBTEMP 5-12
buffers 4-39
DBUPSPACE environment variable 13-18
performance 5-44
DD_HASHMAX configuration parameter 4-21, 4-22
Data types
DD_HASHSIZE configuration parameter 4-21, 4-22
BLOB 5-6
Deadlock 8-15
built-in, distinct, and opaque 7-22
DEADLOCK_TIMEOUT configuration parameter 8-15
BYTE 5-15, 6-5, 6-9
Decision-support queries 1-2
CHAR 6-43, 10-1, 10-29
balanced with transaction throughput 1-5
CLOB 5-6
controlling resources 12-15
effect of mismatch 10-28
effects of DS_TOTAL_MEMORY 4-12
NCHAR 6-43, 10-27
monitoring resources allocated 12-16, 12-17, 13-51
NVARCHAR 6-7, 10-27
monitoring threads 12-16, 13-49
simple large object, for 6-9
performance impact 1-7
TEXT 5-15, 6-5, 6-9, 6-43
use of temporary files 9-5
VARCHAR 6-7, 6-43, 10-1, 10-29
DEF_TABLE_LOCKMODE configuration parameter 8-4
Data-collector
defragment
buffer 14-2
partitions 6-28
process 14-1
DELETE
Data-dictionary
run in parallel 12-2
cache 4-21, 4-22

Index X-5
Denormalizing Disk I/O (continued)
data model 6-42 TPC-A benchmark 1-4
tables 6-43 unbuffered devices 5-12
Detached index Disks
defined 9-12 identifying overloaded ones A-1
extent size 7-5 Distinct data types 7-21
Dimensional tables, defined 13-21 DISTINCT keyword 13-20
Direct I/O Distributed queries
confirming use of 5-4 improving performance 13-29
enabling 5-4 used with PDQ 12-6
overview 5-2, 5-3 Distribution scheme
DIRECT_IO configuration parameter 5-3, 5-4 defined 9-1
DIRECTIVES configuration parameter 11-12, 11-13 designing 9-7, 9-8
Dirty Read isolation level 5-27, 8-10 methods described 9-6, 9-7
Disabilities, visual Dotted decimal format of syntax diagrams B-1
reading syntax diagrams B-1 DRAUTO configuration parameter 5-44
Disability B-1 DRINTERVAL configuration parameter 5-44
Disk DRLOSTFOUND configuration parameter 5-44
and saturation 5-1 DROP DISTRIBUTIONS keywords, in UPDATE STATISTICS
compression 6-47 statement 13-13
critical data 5-5 DROP INDEX ONLINE statement 7-16, 7-17
layout Dropping indexes 7-12
and table isolation 6-2 DRTIMEOUT configuration parameter 5-44
layout, and backup 6-5, 9-4 DS_HASHSIZE configuration parameter 4-21, 4-24
partitions and chunks 5-2 DS_MAX_QUERIES configuration parameter 3-12, 7-19, 13-34
space, storing TEXT and BYTE data 5-18 changing value 12-8
utilization 1-11 index build performance 7-18
Disk access limit query number 12-12
cost of reading row 10-25 MGM 12-6
performance 13-31 DS_MAX_SCANS configuration parameter 3-12, 12-6, 12-11,
performance effect of 10-25 13-34
sequential 13-31 changing value 12-8
sequential forced by query 13-3 MGM 12-6
Disk extent scan threads 12-6
for dbspaces 6-19 DS_NONPDQ_QUERY_MEM configuration parameter 4-8,
for sbspaces 5-21 5-12, 7-19, 13-34
Disk I/O DS_POOLSIZE configuration parameter 4-21, 4-24
allocating AIO VPs 3-8 DS_TOTAL_MEMORY configuration parameter 4-12, 7-18,
background database server activities 1-1 7-19, 13-34
balancing 5-10, 5-14 changing value 12-8
binding AIO VPs 3-8 DS_MAX QUERIES 3-12
blobspace data and 5-16 estimating value 4-12, 12-11
BUFFERPOOL configuration parameter 4-10 MAX_PDQPRIORITY 12-8
contention 10-25 MGM 12-6
effect of UNIX configuration 3-3 setting for DSS applications 12-15
effect of Windows configuration 3-3 setting for OLTP 12-11
effect on performance 5-1 DSS applications
for temporary tables and sort files 5-8 configuration parameter settings 4-9
hot spots, definition of 5-1 DSS resources
in query plan cost 10-1, 10-9, 10-19 limiting 12-8
isolating critical data 5-5 dtcurrent() function, ESQL/C, to get current date and
KAIO 3-8 time 1-7
light scans 5-27 Duplicate index keys, performance effects of 7-11
lightweight I/O 5-23 Dynamic lock allocation 4-2, 4-15, 13-49
log buffer size, effect of 5-7 Dynamic log
logical log 5-24 file allocation
mirroring, effect of 5-6 benefits 5-37
monitoring preventing hangs from rollback of long
AIO VPs 3-8 transaction 5-37
nonsequential access, effect of 7-10 size of new log 5-37
query response time 1-5
reducing 4-10, 6-43
sbspace data and 5-20
sequential scans 5-26
E
Environment variables
simple large objects 5-16
affecting
smart large objects 5-21, 5-23
CPU 3-3
to physical log 5-7
I/O 5-12

X-6 IBM Informix Performance Guide

Environment variables (continued) Extents (continued)
affecting (continued) reorganizing dbspace to prevent interleaving 6-25
multiplexed connections 3-25 size 6-20
network buffer pool 3-17, 3-18 size for attached index 7-5
network buffer size 3-17, 3-19 size for detached index 6-24, 7-5
parallel sorts 5-13 size for tblspace tblspace 6-10
sort files 5-12 size limit 6-24
sorting 5-5, 5-8 size, initial 6-10
SQL statement cache 13-42 size, next-extent 6-10
temporary tables 5-5, 5-8, 5-12 sizes for fragmented table 9-5
DBSPACETEMP 5-5, 5-8, 5-12, 6-4, 7-18 upper limit on number 6-24
DBUPSPACE 13-18 External optimizer directives 11-1, 11-15
FET_BUF_SIZE 13-30
IFX_AUTO_REPREPARE 11-13
IFX_DEF_TABLE_LOCKMODE 8-4
IFX_DIRECTIVES 11-12
F
Fact table
IFX_LARGE_PAGES 4-2
star schema 13-21
IFX_SESSION_MUX 3-25
Fast polling 3-17
INFORMIXOPCACHE 5-24, 5-26
Fast recovery
OPT_GOAL 13-35
configuration effects 5-42
OPTCOMPIND 3-3, 3-10, 12-13
physical log overflow 5-43
PDQPRIORITY
FASTPOLL configuration parameter 3-17
adjusting the value 12-8
FET_BUF_SIZE environment variable 13-30
for UPDATE STATISTICS 13-18
File descriptors 3-2
limiting resources 3-3
Files
parallel sorts 13-33
$INFORMIXDIR/bin 14-4
requesting PDQ resources 12-6
dbspaces for sort 6-4
setting PDQ priority 7-18
executables for onperf 14-4
PSORT_DBTEMP 5-12
saving performance metrics 14-2
PSORT_NPROCS 3-3, 5-13, 7-18, 13-33
TEMP or TMP user environment variable 5-8
STMT_CACHE 13-42
FILLFACTOR
equal() function 7-29
CREATE INDEX 13-29
Equality expression, definition of 9-16
FILLFACTOR clause
ESQL/C
CREATE INDEX statement 7-5
functions, for smart large objects 5-20, 5-21, 6-17, 6-23,
FILLFACTOR configuration parameter 7-5
8-20
Filter
Estimating space
columns 10-5
index extent size 7-5
columns in large tables 7-10
sbspaces 6-11
defined 10-21, 13-2
smart large objects 6-11
effect on performance 13-3
EXECUTE PROCEDURE statement 10-33
effect on sorting 10-24
Expiration policies, AUS 13-8
evaluated from index 13-20
changing 13-9
index used to evaluate 10-22
ph_threshold table 13-9
memory used to evaluate 10-24
explain output file 10-10
query plan 11-2
EXPLAIN_SQL routine 10-19
selectivity defined 10-21
EXPLAIN_STAT configuration parameter 10-11
selectivity estimates 10-21
Explicit temporary table 9-13
user-defined routines 13-2
Expression-based distribution scheme
Flattened subquery 10-15
defined 9-6
FORCE_DDL_EXEC environment Option 9-27
designing 9-8
Forced residency 4-17
fragment elimination 9-16
Foreground write 5-39
type to use 9-14
Foreign-key constraints 6-32
EXTENT SIZE clause 6-20
Forest of trees indexes
extents
creating 7-14
merging 6-28
determining if needed 7-13
Extents
disabling 7-15
allocating 6-20
enabling 7-15
attached index 9-12
identifying in SET EXPLAIN output 7-16
eliminating interleaved 6-25
implementing 7-14
index of fragmented table 9-10
in sysindices table 7-16
interleaved 6-24
overview 7-2, 7-13
managing 6-19
performing range scans 7-15
managing deallocation with TRUNCATE 6-27
viewing information 7-16
next-extent size 6-20
why use 7-13
performance 5-21, 6-19, 6-24
Formula
reclaiming empty space 6-24, 6-26
blobpage size 6-8

Index X-7
Formula (continued) Fragmentation (continued)
buffer pool size 4-10 strategy
connections per poll thread 3-14 ALTER FRAGMENT ATTACH clause 9-20, 9-25
CPU utilization 1-9 ALTER FRAGMENT DETACH clause 9-26, 9-27
data buffer size, estimate of 4-4 distribution schemes for fragment elimination 9-14
decision-support queries 12-11 finer granularity of backup and restore 9-4
disk utilization 1-11 how data used 9-4
DS total memory 4-14 improved performance 9-3
extends, upper limit 6-24 improving 9-9
file descriptors 3-2 increased availability of data 9-3
index extent size 7-5 indexes 9-10
index pages 6-5, 7-5 planning 9-1
initial stack size 4-20 reduced contention 9-3
LOGSIZE 5-36 space issues 9-1
memory grant basis 12-11 temporary tables 9-13
minimum DS memory 4-13 sysfragments system catalog 9-28
number of remainder pages 6-5 TEMP TABLE clause 9-13
operating-system shared memory 4-6 temporary tables 9-13
paging delay 1-10 Freeing shared memory 4-7
partial remainder pages 6-5 Functional index
PDQ resources allocated 3-11 creating 7-25, 7-26
quantum of memory 4-12, 12-6 DataBlade modules 7-26
rows per page 6-5 user-defined function 7-1
scan threads 12-6 using 7-25, 13-2
per query 3-12, 12-11 Functions, ESQL/C, dtcurrent() 1-7
semaphores 3-1
service time 1-8
shared memory
message portion size 4-6
G
Generic B-tree
resident portion size 4-4
index
virtual portion size 4-4
extending 7-23
shared-memory estimate 12-11
parallel UDRs 13-38
shared-memory increment size 4-18
user-defined data 7-1
sort operation, costs 10-24
when to use 7-23
threshold for free network buffers 3-18
Global file descriptor queues 3-22
Fragment
Graph tool (onperf)
elimination
bar graph 14-8
defined 9-14
Configure menu 14-8
equality expressions 9-16
defined 14-3, 14-5
fragmentation expressions 9-14
Graph menu 14-6
range expressions 9-15
Graph tool (onperf)
ID
View menu 14-8
and index entry 7-5
metric
defined 9-12
changing line color and width 14-8
fragmented table 9-5
changing scale 14-10
space estimates 9-5
Metrics menu 14-7
nonoverlapping
pie chart 14-8
multiple columns 9-18
Tools menu 14-9
single column 9-17
greaterthan() function 7-29
overlapping
greaterthanorequal() function 7-29
single column 9-17
GROUP BY
FRAGMENT BY clause 9-10
clause, composite index used 13-20
Fragmentation
clause, indexes 10-22, 13-32
altering fragments 9-27
clause, MGM memory 12-6
FRAGMENT BY EXPRESSION clause 9-10, 9-12
goals 9-1
improving ATTACH operation 9-20, 9-24
improving DETACH operation 9-26, 9-27 H
index restrictions 9-13 Hash join
indexes, attached 9-10 in directives 11-2, 11-5
indexes, detached 9-12 more memory for 5-12, 13-34
monitoring I/O requests 9-28 plan example 10-2
monitoring with onstat 9-28 temporary space 5-12
next-extent size 9-9 when used 10-3
no data migration during ATTACH 9-22 HDR_TXN_SCOPE configuration parameter 5-44
reducing contention 9-3 High-Performance Loader 6-1
smart large objects 9-6 Home pages in indexes 6-5

X-8 IBM Informix Performance Guide

Host variable Index (continued)
SQL statement cache 13-41 ordering columns in composite 13-20
Hot spots, defined 5-1 placement on disk 7-4
size estimate 7-5
snowflake or star schemas 13-21
I structure of entries 7-1
time cost 7-8
I/O utilization
User-defined data types 7-21, 7-31
options for monitoring 2-10
when not used by optimizer 10-29, 13-3
IBM Data Studio 10-19
when replaced by join plans 10-7
IBM Informix MaxConnect
when to rebuild 13-29
defined 3-26
Index self-join 10-8
IBM Informix Server Administrator
Index self-join path 10-8
capabilities 2-5
Indexes
defined 2-5
clustered 6-25, 7-17
Identifying overloaded disks A-1
improving performance 13-20
IFX_AUTO_REPREPARE session environment variable 11-13
industry standards xxii
IFX_BATCHEDREAD_TABLE session environment
INFORMIXOPCACHE environment variable 5-24, 5-26
variable 5-27
Inner table
IFX_DEF_TABLE_LOCKMODE environment variable 8-4
directives 11-6
IFX_DIRECTIVES environment variable 11-12
index 10-2
IFX_EXTDIRECTIVES environment variable 11-15
Input-output (I/O)
IFX_LARGE_PAGES environment variable 4-2
background activities 5-29
IFX_NETBUF_PVTPOOL_SIZE environment variable 3-17,
contention and high-use tables 6-2
3-18
disk saturation 5-1
IFX_NETBUF_SIZE environment variable 3-17, 3-19
tables, configuring 5-26
IFX_SESSION_MUX environment variable 3-25
INSERT cursor 9-6
IMPLICIT_PDQ session environment variable 12-14
INTO TEMP clause of the SELECT statement 5-8, 5-10, 5-15,
In-place alter algorithm
6-24
Alters
iostat command 2-3
in-place 6-35
ipcshm
performance advantages 6-35
connection 3-14
restrictions 6-36
ipcshm connection 4-6
Index
Isolating tables 6-2
adding for performance 7-10
Isolation level
and previously prepared statement problem 11-13
ANSI Repeatable Read 8-7
attached index extent size 7-5
ANSI Serializable 8-7
autoindex
Committed Read 5-27, 8-6
for inner table 10-2
Committed Read Last Committed 5-27
replacing with permanent 13-20
Cursor Stability 8-7
checking 7-21
Dirty Read 5-27, 8-5, 8-10
choosing columns 7-10
effect on concurrency 10-2
composite 13-20
effect on joins 10-2
cost of on NCHAR 10-27
Last Committed 8-6
cost of on NVARCHAR 10-27
light scans 5-27
cost of on VARCHAR 10-1
monitoring 2-5, 8-15
creating in online environment 7-16, 7-17
Repeatable Read 5-27, 8-7
DataBlade modules 7-27
Repeatable Read and OPTCOMPIND 10-22, 10-23, 12-13
detached index extent size 7-5
SET ISOLATION statement 8-5
disk space used by 7-8, 13-31
distinct types 7-21
dropping 6-32, 7-12
dropping in online environment 7-16, 7-17 J
duplicate entries 7-11 Join
duplicate keys, avoiding 7-11 avoiding 13-3
effect of physical order of table rows 10-7 column for composite index 13-20
estimating pages 7-5 directives 11-5
estimating space 7-4, 7-5 effect of large join on optimization 13-37
extent size 7-5 hash join 10-2
filtered columns 7-10 hash join, when used 10-3
functional 7-25, 13-2 method
impact on delete, insert, and update operations 7-8 directives 11-6
key-only scan 10-1 methods 10-2, 10-23
managing 7-8 nested-loop join 10-2
on CHAR column 10-1 order 10-4, 11-2, 11-3, 11-10
on fact table in star schema 13-21 outer 12-10
opaque data types 7-21 parallel execution 12-9
order-by and group-by columns 7-10 plan 11-3

Index X-9
Join (continued) Lock (continued)
defined 10-2 determining owner 8-13
directive precedence 11-12 dynamic allocation 4-2, 4-15
effects of OPTCOMPIND 10-23 isolation levels and join 10-2
hash 11-10, 12-13, 12-18 promotable 8-9
hash, in directives 11-2, 11-5 retaining update locks 8-9
isolation level effect 10-2 specifying mode 8-4
nested-loop 11-5, 11-7, 11-10 Lock table
OPTCOMPIND 12-13 specifying initial size 4-2, 4-15
optimizer choosing 11-2 LOCKBUFF configuration parameter 4-2
replacing index use 10-4 Locking
selected by optimizer 10-1 byte-range 8-16
star 13-21 Locks
subquery 10-15 byte 8-10
running UPDATE STATISTICS on columns 13-16 byte-range 8-16
semi join 10-15 changing lock mode 8-4
SET EXPLAIN output 12-13 concurrency 8-1
star configuring 8-12
directives 11-8 database 8-3
subquery 12-10 defined 8-1
subquery flattening 10-15 duration 8-5
thread 12-1 exclusive 8-10
three-way 10-4 granularity 8-1
view 12-10 initial number 8-12
with column filters 10-5 intent 8-10
Join and sort, reducing impact 13-32 internal lock table 8-10
isolation level 8-5
key-value 8-2
K maximum number of 4-15, 8-12
maximum number of rows or pages 8-1
Kernel asynchronous I/O (KAIO) 3-8
monitoring 8-11, 8-12, 8-14, 8-18
Key-first scan 10-14
not waiting for 8-5
Key-only index scan 10-1, 10-14, 10-36
page 8-2
row and key 8-1
shared 8-10
L specifying a mode 8-4
Last committed isolation level 8-6 table 8-3
Latch types 8-10
defined 4-39 update 8-10
monitoring 4-39, 4-40 waiting for 8-5
Latency, disk I/O 10-25 LOCKS configuration parameter 4-2, 4-15, 8-12
Leaf index pages, defined 7-1 LOGBUFF configuration parameter 4-15, 5-8, 5-20, 5-34
Leaf scan mode 13-22, 13-27 LOGFILES configuration parameter
Least recently used effect on checkpoints 5-32
flushing 5-45 use in logical-log size determination 5-34
memory management algorithm 1-10 Logging
queues 5-40 checkpoints 5-34
thresholds for I/O to physical log 5-7 configuration effects 5-34
lessthan() function 7-29 dbspaces 5-36
lessthanorequal() function 7-29 disabling on temporary tables 5-39
Light scans I/O activity 5-20
advantages 5-27 LOGSIZE configuration parameter 5-35, 5-36
defined 5-27 none with SBSPACETEMP configuration parameter 5-14
isolation level 5-27 simple large objects 5-16, 5-37
Lightweight I/O smart large objects 5-37
specifying in onspaces 5-24 with SBSPACENAME configuration parameter 5-14
specifying with LO_NOBUFFER flag 5-24 Logical log
when to use 4-10, 5-23, 5-24 assigning files to a dbspace 5-5
LIKE test 13-3 buffer size 4-15
LO_DIRTY_READ flag 8-20 buffered 5-7
LO_NOBUFFER flag, specifying lightweight I/O 5-24 configuration parameters that affect 5-8
LO_TEMP flag data replication buffers 4-39
temporary smart large object 5-14 determining disk space allocated 5-35
LOAD and UNLOAD statements 6-1, 6-25, 6-27, 7-12 logging mode 5-7
Locating simple large objects 6-9 mirroring 5-7
Lock simple large objects 5-37
blobpage 5-16 size guidelines 5-35

X-10 IBM Informix Performance Guide

Logical log (continued) Memory (continued)
smart large objects 5-37 SPL routines 12-10
space 5-35 SQL statement cache 13-40
unbuffered 5-7 typename 10-33
viewing records 1-4 UDR cache 10-33
LOGSSIZE configuration parameter 5-32 UNIX configuration parameters 3-3
Long transaction utilizing 1-10
ALTER TABLE operation 6-35 Windows parameters 3-3
configuration effects 5-37, 5-38 Memory Grant Manager
dynamic log effects 5-38 defined 12-6
LTXHWM configuration parameter 6-35 DSS queries 12-6
preventing hangs from rollback 5-37 memory allocated 4-12
Loosely-coupled mode 13-55 monitoring resources 12-6, 12-16
LOW_MEMORY_RESERVE configuration parameter 4-15, scan threads 12-6
5-43 sort memory 7-19
LRU tuning 5-45 Memory Grant Manager (MGM) 7-19, 13-34
lru_max_dirty value 5-30, 5-40, 5-45 Memory-management system 1-10
lru_min_dirty value 5-30, 5-40, 5-45 Messages
LRU. 1-10 portion of shared memory 4-3, 4-6
lrus value 5-40 Metadata
LTAPEBLK configuration parameter 5-42 area in sbspace
LTAPEDEV configuration parameter 5-42 contents 6-11
LTAPESIZE configuration parameter 5-42 estimating size 6-11, 6-12
LTXEHWM configuration parameter 5-38 logging 5-37
LTXHWM configuration parameter 5-38 mirroring 5-6
reserved space 6-11
improving I/O for smart large objects 6-12
M Metric classes, onperf
database server 14-13
Managing extents 6-19
disk chunk 14-15
Materialized view
disk spindle 14-15
defined 10-27
fragment 14-18
involving table hierarchy 10-36
physical processor 14-15
MAX_FILL_DATA_PAGES configuration parameter 6-46
session 14-16
MAX_PDQPRIORITY configuration parameter 3-11, 7-19,
tblspace 14-17
13-34
virtual processor 14-16
and PDQPRIORITY 3-3
Microsoft Transaction Server
changing value 12-8
tightly coupled mode 13-55
for DSS query limits 12-8, 12-9
MIRROR configuration parameter 5-8
increasing OLTP resources 12-9
Mirroring
limiting concurrent scans 12-11
critical media 5-6
limiting PDQ resources 5-13, 13-33
root dbspace 5-6, 5-8
limiting user-requested resources 12-15
sbspaces 5-6
MGM 12-6
MIRROROFFSET configuration parameter 5-8
PDQPRIORITY, and 12-9, 12-15
MIRRORPATH configuration parameter 5-8
Memory
MODIFY EXTENT SIZE keyword
activity costs 10-24
in ALTER TABLE statement 6-20
cache 4-21
MODIFY NEXT SIZE clause 6-20, 6-22
aggregate 10-33
mon_table_profile 13-7
data-dictionary 4-22
Monitoring
configuration parameters 4-8
aggregate cache 10-33
data-replication buffers 4-39
AIO virtual processors 3-22
estimate for sorting 7-19
buffer pool 4-10
hash join 5-12
buffers 4-10
hash joins 13-34
data-distribution cache 4-24
increase by logging 5-24
deadlocks 8-15
limited by
foreground writes 5-39
MAX_PDQPRIORITY 3-11
fragments 9-28
PDQ priority 3-3
global transactions 13-55, 13-57
STMT_CACHE_NOLIMIT 4-27
I/O queues for AIO VPs 3-8
monitoring by session 12-17
latch waits 4-39, 4-40
monitoring MGM allocation 12-6
light scans 5-27
network buffer pool 3-17, 3-18, 3-19
locks 8-11, 8-12, 8-13, 8-14, 13-55, 13-56
opclass cache 10-33
locks used by sessions 8-12
PDQ priority effect 7-19, 12-8
logical-log files 2-12
private caches 3-24
LRU queues 5-40
private network free-buffer pool 3-17, 3-18
memory per thread 4-4
quantum allocated by MGM 12-6, 12-15

Index X-11
Monitoring (continued) NETTYPE configuration parameter (continued)
memory usage 4-4 specifying connections 3-13, 3-14
memory utilization 2-9 Network
MGM resources 12-16 buffer pools 3-17, 3-18
network buffer size 3-19 buffer size 3-19
network buffers 3-18 common buffer pool 3-17, 3-19
OPCACHEMAX 5-24 communication delays 5-1
PDQ threads 12-16 connections 3-13
resources for a session 12-17 free-buffer threshold 3-18
sbspace metadata size 6-11, 6-12 monitoring buffers 3-18
sbspaces 6-13, 6-16 multiplexed connections 3-25
session memory 2-13, 4-4, 4-38, 13-43, 13-44, 13-45, 13-46, performance bottleneck 2-1
13-48, 13-52 performance issues 10-30
sessions 13-48, 13-51 private free-buffer pool 3-17, 3-18
smart large objects 6-13 NEXT SIZE clause 6-20
SPL routine cache 10-33 NFILE configuration parameters 3-2
SQL statement cache 4-30, 4-36, 13-46 NFILES configuration parameters 3-2
entries 13-46 NOFILE configuration parameters 3-2
pool 4-34, 4-35 NOFILES configuration parameters 3-2
size 4-31, 4-32, 4-35 NOVALIDATE keyword
STAGEBLOB blobspace 5-24 in ALTER TABLE statement 6-32
statement cache 4-29 in SET CONSTRAINTS statement 6-32
statement memory 2-13, 13-43, 13-46 in SET ENVIRONMENT statement 6-32
threads 13-48, 13-49, 13-50, 13-51 NS_CACHE configuration parameter 3-16
concurrent users 4-4 NUMFDSERVERS configuration parameter 3-16
per CPU VP 3-10 NVARCHAR data type 6-7
session 3-10, 12-16 table-size estimates 6-7
throughput 1-4
transaction 13-55
UDR cache 10-33
user sessions 13-57
O
Obtaining 6-5
user threads 13-55
OFF_RECVRY_THREADS configuration parameter 5-42
virtual processors 3-21, 3-22
OLTP applications
Monitoring database server
configuration parameter settings 4-9
active tblspaces 6-23
effects of MAX_PDQPRIORITY 3-11
blobspace storage 5-17
effects of PDQ 12-7
buffers 4-10
maximizing throughput with MAX_PDQPRIORITY 12-6,
sessions 2-13, 13-47
12-9
threads 2-7, 13-47
reducing DS_TOTAL_MEMORY 12-11
transactions 13-55
using MGM to limit DSS resources 12-6
virtual processors 3-21
OLTP query 1-2
Monitoring tools
ON_RECVRY_THREADS configuration parameter 5-42
database server utilities 2-4
ON-Bar utility
UNIX 2-3
configuration parameters 5-41
Windows 2-3
onaudit utility 5-45
Motif window manager 14-1, 14-3, 14-4
oncheck utility
Multiple residency
-pB option 2-11, 5-17
avoiding 3-1
-pe option 2-11, 6-14, 6-24, 6-25
Multiplexed connection
-pk option 2-11
defined 3-25
-pK option 2-11
how to use 3-25
-pl option 2-11
performance improvement 3-25
-pL option 2-11
MULTIPROCESSOR configuration parameter 3-10
-pp option 2-11
mwm window manager, required for onperf 14-4
-pP option 2-11
-pr option 2-11, 6-40
-ps option 2-11
N -pS option 2-11, 6-15
NCHAR data type 6-43 -pt option 2-11, 6-5
Negator functions 13-40 -pT option 2-11, 6-40
Nested-loop join 10-2, 11-5 checking index pages 7-21
NET VP class and NETTYPE 3-14 defined 2-11
NETTYPE configuration parameter 3-16, 4-4 displaying
connections 3-18 data-page versions 6-40
estimating LOGSIZE 5-36 free space 6-25
ipcshm connection 3-14, 4-6 free space in index 13-29
network free buffer 3-18 page size 6-40
poll threads 3-1, 3-20 size of table 6-5

X-12 IBM Informix Performance Guide

oncheck utility (continued) onstat utility (continued)
index sizing 7-5 -g ath option 3-10, 12-16, 13-48, 13-49, 13-51
monitoring -g cac stmt option 4-29
table growth 6-20 -g cpu option 13-48
obtaining information -g dic option 4-24
blobspaces 5-18, 5-19 -g dsc option 4-24
sbspaces 6-14 -g glo option 3-21, 3-25
outstanding in-place alters 6-40 -g ioq option 3-8, 3-22
physical layout of chunk 6-24 -g mem option 2-9, 13-48, 13-52
ONDBSPACEDOWN configuration parameter 5-33 -g mgm option 2-9, 12-6, 12-16
ONLIDX_MAXMEM configuration parameter 7-16, 7-18 -g ntm option 3-18
onload and onunload utilities 5-42, 6-1, 6-25, 6-27 -g ntu option 3-18
onlog utility 1-4, 2-12 -g opn option 6-29
onmode -Y 10-10 -g option 2-6
onmode utility -g osi option 2-9
-e option 13-41, 13-42 -g ppf option 9-28
-p option 3-20 -g prc option 10-33
-P option 3-8 -g rea option 3-22
-W option 4-30 -g scn to monitor light scans 5-27
changing STMT_CACHE_NOLIMIT 4-33 -g seg option 2-9, 4-4, 4-18
–F option 4-7 -g ses option 2-9, 2-13, 3-10, 4-4, 12-17, 13-43, 13-44, 13-45,
flushing SQL statement cache 13-41 13-46, 13-48, 13-51
forced residency 4-17 -g smb option 6-13
shared-memory connections 3-1 -g smb s option 6-16
onparams utility 5-5, 5-7 -g spi option 4-34
onperf utility -g sql option 2-13, 13-45, 13-46
activity tools 14-12 -g sql session-id option 13-55
data flow 14-1 -g ssc all option 4-30
defined 14-1 -g ssc option 4-29, 4-30, 4-31, 4-35, 4-36, 13-46
displaying recent history 14-10 -g ssc output description 4-36
graph tool 14-5 -g ssc pool option 4-35
metric classes -g stm option 2-9, 2-13, 4-4, 4-38, 13-43, 13-46, 13-48, 13-52
database server 14-13 -g sts option 4-4
disk chunk 14-15 -k option 8-11, 8-13, 8-19, 13-55, 13-56
disk spindle 14-15 -l option 2-6
fragment 14-18 -L option 8-14
physical processor 14-15 -m option 5-31
session 14-16 -O option 5-24
tblspace 14-17 -p option 1-4, 2-6, 4-10, 4-39, 8-12, 8-15
virtual processor 14-16 -R option 2-6
metrics 14-13 -s option 4-40
monitoring tool 2-4 -t option 6-23
query-tree tool 14-11 -u option 2-6, 4-4, 8-12, 8-13, 12-16, 13-48, 13-55, 13-57
replaying metrics 14-3 -x option 2-6, 13-55
requirements 14-3 monitoring
saving metrics 14-2 AIO virtual processors 3-22
starting 14-4 buffer use 4-10
status tool 14-11 byte locks 8-10
tools 14-3 locks 13-56
user interface 14-5 PDQ 12-16
onspaces utility sessions 13-48
-Df BUFFERING tag 5-24 tblspaces 6-23
-Df option 5-21, 6-19 transactions 13-55
-S option 6-19 user sessions 13-57
-t option 5-10, 5-14, 6-4, 7-18 virtual processors 3-21, 3-22, 3-25
EXTENT_SIZE flag for sbspaces 5-21 options for monitoring disk I/O utilization 2-10
sbspaces 5-20, 5-24 options for monitoring threads 2-7
smart large objects 6-17 options for monitoring transactions 2-13
specifying lightweight I/O 5-24 overview for performance monitoring 2-6
onstat utility ontape utility 5-42
-- option 2-6 Opaque data types 7-21
-a option 2-6 OPCACHEMAX configuration parameter
-b option 2-6, 4-4, 6-5, 6-8 defined 5-26
-d option 3-22, 6-11, 6-12 OpenAdmin Tool 2-7
-F option 2-6, 5-39 Operating system
-g act option 13-48, 13-50 configuration parameters 3-1
-g afr option 3-19 file descriptors 3-2

Index X-13
Operating system (continued) Optimizer directives (continued)
NOFILE, NOFILES, NFILE, or NFILES configuration ORDERED 11-3, 11-5, 11-6
parameters 3-2 purpose 11-1
semaphores 3-1 SPL routines 11-13
SHMMAX configuration parameter 4-6 star-join 11-3, 11-8
SHMMNI configuration parameter 4-6 types 11-3
SHMSEG configuration parameter 4-6 USE_HASH 11-6
SHMSIZE configuration parameter 4-6 USE_NL 11-6
timing commands 1-6 using DIRECTIVES 11-12
Operator class using IFX_DIRECTIVES 11-12
defined 7-23, 7-28 ORDER BY clause 10-22, 13-32
OPT_GOAL configuration parameter 13-35 Ordered merge 13-37
OPT_GOAL environment variable 13-35 Outer join
OPTCOMPIND effect on PDQ 12-6
directives 11-12 Outer table 10-2
effects on query plan 10-22 Output description
preferred join plan 12-13 onstat -g ssc 4-36
OPTCOMPIND configuration parameter 3-3, 3-10, 12-13 Outstanding in-place alters
OPTCOMPIND environment variable 3-3, 3-10, 12-13 defined 6-40
Optical Subsystem 5-24 displaying 6-40
Optimization goal performance impact 6-40
default total query time 13-35 Overloaded disks A-1
precedence of settings 13-36
setting with directives 11-7, 13-36
total query time 13-35, 13-37
user-response and fragmented indexes 13-37
P
Page
user-response time 13-35, 13-37
cleaning 5-39
Optimization level
memory 1-10
default 13-35
obtaining size 6-5
setting to low 13-35
specifying size for a standard dbspace 4-10, 7-8
table scan versus index scan 13-37
Page size 6-5
Optimizer
obtaining 4-4
autoindex path 13-20
Paging
choosing query plan 11-2, 11-3
defined 1-10
composite index use 13-20
DS_TOTAL_MEMORY 12-11
data distributions used by 13-13
expected delay 1-10
hash join 10-3
monitoring 2-2, 4-10
index not used by 13-3
RESIDENT configuration parameter 4-2
optimization goal 11-7, 13-35
Parallel
SET OPTIMIZATION statement 13-35
access to table and simple large objects 5-20
specifying high or low level of optimization 13-35
backup and restore 5-41
Optimizer directives
executing UDRs 13-38
access method 11-4
index builds 12-3
ALL_ROWS 11-7
inserts and DBSPACETEMP 12-2
altering query plan 11-10
joins 12-9
AVOID_EXECUTE 13-1
scans 12-18, 13-38
AVOID_FULL 11-3, 11-4
sorts
AVOID_HASH 11-6
PDQ priority 13-33
AVOID_INDEX 11-4
when used 5-13
AVOID_INDEX_SJ 10-9, 11-4
Parallel database queries
AVOID_NL 11-3, 11-6
allocating resources 12-7
effect on views 11-6
controlling resources 12-15
embedded in queries 11-1
effect of table fragmentation 12-1
EXPLAIN 11-8, 13-1
fragmentation 9-1
EXPLAIN AVOID_EXECUTE 11-8
how used 12-2
external 11-15
monitoring resources allocated 12-16
external directives 11-1
priority
FIRST_ROWS 11-7
effect of remote database 12-6
FULL 11-4
queries that do not use PDQ 12-4
guidelines 11-3
remote tables 12-6
INDEX 11-4
scans 3-12
INDEX_SJ 11-4
SET PDQPRIORITY statement 12-13
join method 11-6
SPL routines 12-5
join order 11-3, 11-5
SQL 9-1
OPTCOMPIND 11-12
statements affected by PDQ 12-5
Optimizer directives
triggers 12-3, 12-4, 12-5
INDEX_SJ 10-9
user-defined routines 13-38

X-14 IBM Informix Performance Guide

Parallel database queries (continued) Performance (continued)
using 12-1 effect of (continued)
Parallel processing disk I/O 5-1
fragmentation 9-9, 12-1 duplicate keys 7-11
MGM control of resources 12-6 filter expression 13-3
ON-Bar utility 5-41 filter selectivity 10-21
PDQ threads 12-1 indexes 7-10
user-defined routines 13-38 redundant data 6-45
Parallel UDRs regular expressions 13-3
defined 12-4, 13-38 sequential access 13-31
enabling 13-38 simple-large-object location 6-9
sample query 13-38 table size 13-31
when to use 13-38 goals 1-2
Partitioning improved by
defined 9-1 contiguous extents 5-21, 6-19
partitions specifying optimization level 13-35
defragmenting 6-28 temporary table 13-33
Partitions index time during modification 7-8
creating in a detached index 9-12 measurements 1-3
creating in a fragmented index 9-10 slowed by data mismatch 10-29
creating in an attached index 9-10 slowed by duplicate keys 7-11
for storing multiple fragments of the same index 7-20 tips 1-1
for storing multiple fragments of the same table 6-29 tips for a small database 1-2
PC_HASHSIZE configuration parameter 4-21, 10-33 Performance problems
PC_POOLSIZE configuration parameter 4-21, 10-33 early indications 1-1
PDQ sudden performance loss 14-13
DELETE operations 12-2 PHYSBUFF configuration parameter 4-2, 4-15, 5-34
UPDATE operations 12-2 PHYSFILE configuration parameter 5-32
PDQ priority Physical log
BOUND_IMPL_PDQ session environment variable 12-14 buffer size 4-15
DEFAULT tag 12-8 configuration parameters that affect 5-8
determining parallelism 12-10 effects of
effect of remote database 12-10 checkpoints on sizing 5-32
effect on parallel execution 12-8 frequent updating 5-32
effect on sorting memory 7-18 increasing size 4-10, 5-32
IMPLICIT_PDQ session environment variable 12-14 mirroring 5-7
maximum parallel processing 12-10 overflow during fast recovery 5-43
outer joins 12-10 when you have non-default page sizes 4-10
parallel execution limits 12-10 Playback process 14-3
SET PDQPRIORITY statement 12-13 PLCY_HASHSIZE configuration parameter 4-21
SPL routines 12-10 PLCY_POOLSIZE configuration parameter 4-21
PDQPRIORITY PLOG_OVERFLOW_PATH configuration parameter 5-43
environment variable Poll threads
requesting PDQ resources 12-6 added with network VP 3-20
limiting PDQ priority 12-9 configuring with NETTYPE configuration parameter 3-1,
PDQPRIORITY configuration parameter 3-13
effect of outer joins 12-6 connections per 3-14
PDQPRIORITY environment variable for connection 3-14, 3-20
adjusting the value 12-8 NETTYPE configuration parameter 3-14
for UPDATE STATISTICS 13-18 Priority
limiting PDQ priority 12-8, 12-9 setting on Windows 3-3
limiting resources 3-3 Probe table, directives 11-6
parallel sorts 13-33 PSORT_DBTEMP environment variable 5-12
setting PDQ priority 7-18 PSORT_DTEMP environment variable 5-8
Peak loads 1-7 PSORT_NPROCS environment variable 3-3, 5-13, 7-18, 13-33
Performance
basic approach to measurement and tuning 1-1
capturing data 2-3
contiguous extents 5-21, 6-19
Q
Quantum, of memory 3-12, 4-12, 12-6, 12-15
create a history 2-2, 2-3
Queries
dropping indexes for updates 7-12
improving performance 13-20
dropping indexes to speed modifications 6-32
resources allocated 12-15
effect of
response time and throughput 1-5
contiguous disk space 5-21, 6-13, 6-19
temporary files 9-5, 13-33
contiguous extents 6-24
Query plan
data mismatch 10-29
with index self-join 10-8
disk access 10-26, 13-31
Query plans 10-1

Index X-15
Query plans (continued) Response time (continued)
all rows 11-7 improving with multiplexed connections 3-25
altering with directives 11-3, 11-10 measuring 1-6
autoindex path 13-20 Response times
avoid query execution 11-8 SQL statement cache 13-40
chosen by optimizer 11-3 Root dbspace
collection-derived table 10-16 mirroring 5-6
disk accesses 10-5 Root index page 7-1
displaying 10-10, 13-30 ROOTNAME configuration parameter 5-8
first-row 11-7 ROOTOFFSET configuration parameter 5-8
fragment elimination 9-29, 12-18 ROOTPATH configuration parameter 5-8
how the optimizer chooses one 11-2 ROOTSIZE configuration parameter 5-8
indexes 10-7 Round-robin distribution scheme 9-7
join order 11-10 Round-robin fragmentation, smart large objects 9-6
pseudocode 10-4, 10-6 Row access cost 10-25
restrictive filters 11-2 Row pointer
row access cost 10-25 attached index 9-10
time costs 10-4, 10-23, 10-24, 10-25 detached index 9-12
Query statistics 10-11 in fragmented table 9-5
Query-tree tool (onperf) 14-3, 14-11 space estimates 7-5, 9-5
RTO_SERVER_RESTART configuration parameter 5-27, 5-30,
5-31, 5-41, 5-43
R RTO_SERVER_RESTART policy 5-30, 5-32, 5-35
R-tree index
defined 7-4, 7-25
using 7-22 S
Range expression, defined 9-15 Sampling
Range scan mode 13-27 in UPDATE STATISTICS LOW operations 13-18
Raw disk space 5-2, 5-3 sar command 2-3, 4-10
Read cache rate 4-10 Saturated disks 5-1
Read-ahead sbspace extents
configuring 5-26 performance 5-21, 6-12
defined 5-26 SBSPACENAME
Reclaiming empty extent space 6-26 configuration parameter 5-14
Recovery time objective logging 5-14
Recovery point objective 5-35 SBSPACENAME configuration parameter 5-20
Redundant data, introduced for performance 6-45 sbspaces
Redundant pairs, defined 10-9 configuration impacts 5-20
Referential constraints 6-32 creating 5-21
Regular expression, effect on performance 13-3 defined 5-6
Relational model estimating space 6-11
denormalizing 6-42 extent 5-21, 5-23
Remainder pages metadata requirements 6-11
tables 6-5 metadata size 6-11, 6-12
Remote database monitoring 6-13
effect on PDQPRIORITY 12-6 monitoring extents 6-14, 6-15
RENAME statement 10-32 SBSPACETEMP
Repeatable Read isolation level 5-27, 8-7, 10-22 no logging 5-14
Residency 4-17 SBSPACETEMP configuration parameter 5-14, 5-15
RESIDENT configuration parameter 4-17 Scans
Resident portion of shared memory 4-2, 4-4 bufferpool 5-27
Resizing table to reclaim empty space 6-26 DS_MAX_QUERIES 12-6
Resource utilization DS_MAX_SCANS 12-6
capturing data 2-3 first-row 10-15
CPU 1-9 key-only 10-1
defined 1-8 light 5-27
disk 1-11 lightweight I/O 5-23
factors that affect 1-12 limited by MAX_PDQPRIORITY 3-11
memory 1-10 limiting number 12-11
operating-system resources 1-7 limiting number of threads 12-11
performance 1-7 limiting PDQ priority 12-11
Resources memory-management system 1-10
critical 1-7 parallel 12-18
Response time parallel database query 3-12
actions that determine 1-5 read-ahead I/O 5-26
contrasted with throughput 1-5 sequential 5-26
improving with MaxConnect 3-26 skip-duplicate-index 10-15

X-16 IBM Informix Performance Guide

Scans (continued) SET EXPLAIN (continued)
table 10-1, 10-2, 13-20 optimizing 13-37
threads 3-10, 3-11, 3-12 order of tables accessed 10-13
Scheduler output
automated UPDATE STATISTICS tasks 13-7 statistics 10-11
Scheduler tasks parallel scans 12-18
auto_tune_cpu_vps 3-20 PDQ priority levels 12-18
Scheduling facility, cron 2-4, 4-7 query plan 10-10, 12-13
Screen reader resources required by query 13-1
reading syntax diagrams B-1 secondary threads 12-18
Secondary-access methods serial scans 12-18
DataBlade modules 7-22 simple query 10-13
defined 7-22, 7-25 SPL routines 10-31
defined by database server 7-23 subquery 10-15
generic B-tree 7-23 using 10-10, 10-15
R-Tree 7-25 SET EXPLAIN statement 10-10
SELECT statements SET INDEX COMPRESSION command 13-28
accessing data 9-4 SET ISOLATION statement 8-5, 8-9
collection-derived table 10-16 SET LOCK MODE statement 8-2, 8-5, 8-7, 8-10, 8-13, 8-15,
column filter 10-5 9-27
join order 10-4 SET LOG statement 1-4
materialized view 10-36 SET OPTIMIZATION statement
redundant join pair 10-9 setting ALL_ROWS 13-35
row size 6-6 setting FIRST_ROWS 13-35
SPL routines and directives 11-13 setting HIGH or LOW 13-35
three-way join 10-4 SPL routines 10-32
trigger performance 10-36 SET PDQPRIORITY statement
triggers 10-36 application 12-8, 12-13
using directives 11-1, 11-3 DEFAULT tag 12-8, 12-13
Selective filter in SPL routine 12-10
dimensional table 13-21 limiting CPU VP utilization 3-3
Selectivity sort memory 13-18
column, and filters 7-10 SET STATEMENT CACHE statement 4-27, 13-42
defined 10-21 SET TRANSACTION statement 8-5
estimates for filters 10-21 Shared memory
indexed columns 7-11 allowed per query 4-12
user-defined data 13-38, 13-39 amount for sorting 7-18, 7-19
Semaphores connection 3-14, 3-20
allocated for UNIX 3-1 freeing 4-7
Semi-join, defined 10-2 message portion 4-1, 4-3, 4-6
SEMMNI UNIX configuration parameter 3-1 resident portion 4-1, 4-2, 4-4
SEMMNS UNIX configuration parameter 3-1 size limit 4-18
SEMMSL UNIX configuration parameter 3-1 size of segments 4-18
Sequential virtual portion 4-1, 4-2, 4-4
access costs 10-26 SHMADD configuration parameter 4-2
scans 5-26, 13-31 SHMBASE configuration parameter 4-8
Service time formula 1-8 SHMMAX configuration parameter 4-6, 4-18, 4-19
Session SHMMNI operating-system configuration parameter 4-6
monitoring 2-13, 13-47, 13-48, 13-51 SHMSEG operating-system configuration parameter 4-6
monitoring memory 4-4, 4-38, 13-48, 13-52 SHMSIZE operating-system configuration parameter 4-6
setting optimization goal 13-35 SHMTOTAL configuration parameter 4-2, 4-18
SET DATASKIP statement 9-3 SHMVIRT_ALLOCSEG configuration parameter 4-19
SET ENVIRONMENT OPTCOMPIND 10-22, 12-13 SHMVIRTSIZE configuration parameter 4-2, 4-4, 4-19
SET ENVIRONMENT OPTCOMPIND command 3-11 Short rows, reducing disk I/O 6-43
SET EXPLAIN Shortcut keys
collection scan 10-16 keyboard B-1
complex query 10-13 Simple large objects
converted data 10-29 blobpage size 5-16
data mismatch 10-29 blobspace 5-15
decisions of query optimizer 12-13 configuration effects 5-15
determine UPDATE STATISTICS 13-16 disk I/O 5-16
directives 11-10 estimating number of blobpages 6-8
fragments scanned 9-29 estimating tblspace pages 6-9
how data accessed 9-4 how stored 6-9
join rows returned 13-16 in blobspace 5-16
key-first scan 10-14 in dbspace 6-5
optimizer access paths 10-13 locating 6-9

Index X-17
Simple large objects (continued) SPL routines (continued)
logging 5-16 when executed 10-33
logical-log size 5-37 when optimized 10-31
Optical Subsystem 5-24 SQL statement cache
SINGLE_CPU_VP configuration parameter 3-10 changing size 4-32
slow alter algorithm cleaning 4-31
restrictions 6-36 defined 13-40
Smart large objects effect on prepared statements 13-41
ALTER TABLE 6-19 enabling 13-42
buffer pool 4-10, 5-20, 5-23 exact match 13-43
buffer pool usage 6-17 flushing 13-41
buffering recommendation 5-24 hits 4-21, 4-27, 4-28, 4-29, 4-30, 4-32, 4-33, 4-34, 4-36
changing characteristics 6-19 host variables 13-41
CREATE TABLE statement 6-19 memory 4-21
data integrity 6-17 memory limit 4-27, 4-33
DataBlade API functions 5-20, 5-21, 6-17, 6-23, 8-20 monitoring 4-29, 4-30, 4-36
disk I/O 5-20 monitoring dropped entries 13-46
ESQL/C functions 5-20, 5-21, 6-17, 6-23, 8-20 monitoring pools 4-34, 4-35
estimating space 6-11 monitoring session memory 13-43, 13-44, 13-45, 13-46
extent size 5-21, 5-22, 6-17, 6-23 monitoring size 4-31, 4-32, 4-35
fragmentation 6-17, 9-6 monitoring statement memory 2-13, 13-43, 13-46
I/O operations 5-23, 6-12 nonshared entries 4-30
I/O performance 4-10, 5-20, 5-23, 6-12 number of pools 4-34
last-access time 6-17 performance benefits 4-25, 13-40
lightweight I/O 4-10, 5-23, 5-24 response times 13-40
lock mode 6-17 size 4-21, 4-31, 4-33
logging status 6-17 specifying 13-42
logical-log size 5-37 STMT_CACHE configuration parameter 4-27, 13-42
mirroring chunks 5-6 STMT_CACHE environment variable 13-42
monitoring 6-13 STMT_CACHE_SIZE configuration parameter 4-32
sbspace name 6-17 when to enable 13-42
sbspaces 5-20 when to use 13-41
setting isolation levels 8-20 SQLCODE field of SQL Communications Area 6-44
size 6-17 sqlhosts file
specifying characteristics 6-19 client buffer size 3-19
specifying size 5-21, 6-23 multiplexed option 3-25
storage characteristics 6-17 sqlhosts information
SMI tables connection type 3-13, 3-14, 3-16
monitoring latches 4-40 connections 4-19
monitoring sessions 13-54 number of connections 5-36
monitoring virtual processors 3-23 SQLWARN array 5-29
Snowflake schema 13-21 Stack
Sort memory 7-19 specifying size 4-20
Sorting STACKSIZE configuration parameter 4-20
avoiding with temporary table 13-33 STAGEBLOB configuration parameter 5-24
costs 10-24 defined 5-25
DBSPACETEMP configuration parameter 5-8 Staging area
DBSPACETEMP environment variable 5-8 optimal size for blobspace 5-25
effect of PDQ priority 13-18 standards xxii
effect on performance 13-32 Star join, defined 13-21
estimating temporary space 7-20 Star schema 13-21
memory estimate 7-19 Star-join directives 11-8
PDQ priority for 7-19 Statistics
query-plan cost 10-1 automatically generated 13-11
sort files 5-8 Status tool (onperf) 14-3, 14-11
triggers in a table hierarchy 10-36 STMT_CACHE environment variable 13-42
Space STMT_CACHE_HITS configuration parameter 4-21, 4-27,
reducing on disk 6-46, 6-47 4-28, 4-29, 4-30, 4-32, 4-34, 4-36
SPL 10-33 STMT_CACHE_NOLIMIT configuration parameter 4-21, 4-27
SPL routines STMT_CACHE_NUMPOOL configuration parameter 4-34
automatic reoptimization 10-32 STMT_CACHE_SIZE configuration parameter 4-21, 4-31, 4-33
display query plan 10-31 Storage characteristics
effect Smart large objects
of PDQ 12-5 last-access time 6-17
of PDQ priority 12-10 system default 6-17
optimization level 10-32 Storage spaces
query response time 1-5 for encrypted values 4-40, 10-29

X-18 IBM Informix Performance Guide

Storage statistics Structured Query Language (continued)
blobpages 5-17 SET EXPLAIN statement (continued)
blobspaces 5-17 collection scan 10-16
Stored Procedure Languages 10-33 complex query 10-13
Strategy functions directives 11-10
secondary-access methods 7-28 flattened subquery 10-15
Strings optimizer decisions 12-13
expelling long 6-43 order of tables 10-13
Structured Query Language show query plan 10-10
ALTER FRAGMENT statement 6-27 simple query 10-13
ALTER INDEX statement 6-25, 6-27, 7-11 SET EXPLAIN statement directives 11-10
TO CLUSTER clause 6-25 SET ISOLATION statement 8-5
ALTER TABLE statement 6-20, 6-26 SET LOCK MODE statement 8-2, 8-5, 8-7, 8-10, 8-13, 8-15
changing extent sizes 6-22 SET OPTIMIZATION statement 13-35
changing sbspace characteristics 6-19 SET PDQPRIORITY statement 3-3
sbspace fragmentation 9-6 DEFAULT tag 12-8, 12-13
COMMIT WORK statement 1-4 in application 12-8, 12-13
CONNECT statement 5-2 in SPL routine 12-10
CREATE CLUSTER INDEX statement 7-11 sort memory 13-18
CREATE FUNCTION statement 3-4 SET STATEMENT CACHE 4-27, 13-42
selectivity and cost 13-39 SET TRANSACTION statement 8-5
CREATE INDEX statement TO CLUSTER clause 6-25, 6-27
attached index 9-10 UPDATE STATISTICS statement 4-2, 10-20, 11-3
detached index 9-12 and directives 11-3, 11-13
generic B-tree index 7-23 creating data distributions 13-13
TO CLUSTER clause 6-25 data distributions 10-20
CREATE PROCEDURE statement, SQL optimization 10-31 effect of PDQ 12-5
CREATE TABLE statement guidelines to run 13-11, 13-18
blobspace assignment 5-15 HIGH mode 13-11, 13-13, 13-16, 13-19
extent sizes 6-20 LOW mode 13-11, 13-12, 13-18, 13-38
fragmentation 9-10, 9-12 MEDIUM mode 13-13, 13-16
lock mode 8-4 multiple column distributions 13-18
PUT clause 6-19 on join columns 13-16
sbspace characteristics 6-19 on user-defined data columns 13-16
sbspace fragmentation 9-6 optimizing SPL routines 12-10
simple large objects 6-9 query optimization 13-11
system catalog table 5-2 reoptimizing SPL routines 10-32
TEMP TABLE clause 5-8, 5-14 updating system catalog 10-20, 13-11
CREATE TEMP TABLE statement 9-13 user-defined data 13-38
DATABASE statement 5-2 WHERE clause 10-22, 13-2, 13-3
EXECUTE PROCEDURE statement 10-33 Subquery 12-10
EXTENT SIZE clause 6-20 flattening 10-15
FRAGMENT BY clause 9-10 rewriting 10-15
GROUP BY clause 10-22 Support functions
MGM memory 12-6 description for secondary access method 7-28
INSERT statements 9-6 Swap device 1-10
LOAD and UNLOAD statements 6-1, 6-25, 6-27, 7-12 Swap space 1-10, 4-6
MODIFY EXTENT SIZE clause 6-20 Swapping, memory 1-10, 12-11
MODIFY NEXT SIZE clause 6-20, 6-22 Symbol table
NEXT SIZE clause 6-20 building 6-44
optimizer directives 11-3 Syntax diagrams
ORDER BY clause 10-22 reading in a screen reader B-1
RENAME statement 10-32 sysdirectives system catalog table 11-1
SELECT statements sysmaster database 2-4
collection-derived tables 10-16 sysprofile table 8-12
column filter 10-5 System catalog tables
join order 10-4 data distributions 10-20
materialized view 10-36 optimizer use of 10-20
redundant join pair 10-9 sysams 7-24, 7-30
row size 6-6 syscolumns 13-13, 13-16
SPL routines and directives 11-13 sysdistrib 13-13, 13-16
three-way join 10-4 sysfragments 9-12, 9-29
triggers 10-36 sysopclasses 7-30
using directives 11-1, 11-3 sysprocbody 10-31
SET DATASKIP statement 9-3 sysprocedure 10-31
SET EXPLAIN statement 9-29 sysprocplan 10-31, 10-32
accessing data 9-4 systables 7-21, 10-32

Index X-19
System catalog tables (continued) TCP/IP buffers 3-17
systrigbody 10-34 TEMP or TMP user environment variable 5-8
systriggers 10-34 TEMP TABLE clause of the CREATE TABLE statement 5-8,
updated by UPDATE STATISTICS 10-20 5-14, 9-13
System resources, measuring utilization 1-7 Temporary dbspace
System-monitoring interface 2-4 creating 7-18
DBSPACETEMP configuration parameter 5-11
DBSPACETEMP environment variable 5-12
T for index builds 7-18, 7-20
onspaces -t 5-10
Table
optimizing 5-10
adding redundant data 6-45
root dbspace 5-8
assigning to dbspace 6-1
Temporary sbspace
companion, for long strings 6-43
configuring 5-14
configuring I/O for 5-26
onspaces -t 5-14
cost of access 13-31
optimizing 5-14
denormalizing 6-43
SBSPACETEMP configuration parameter 5-15
division by bulk 6-44
Temporary smart large object
estimating
LO_TEMP flag 5-14
blobpages in tblspace 6-8
Temporary tables
data page size 6-5
configuring 5-8
size with fixed-length rows 6-5
DBSPACETEMP configuration parameter 5-8, 5-11
size with variable-length rows 6-7
DBSPACETEMP environment variable 5-12
expelling long strings 6-43
decision-support queries 9-4
fact 13-21
Decision-support queries
frequently updated attributes 6-44
use of temporary files 9-4
infrequently accessed attributes 6-44
explicit 9-13
isolating high-use 6-2
fragmentation 9-13
locks 8-3
in root dbspace 5-5
managing
speeding up a query 13-33
extents 6-19
Temporary dbspace
managing indexes for 7-8
decision-support queries 9-4
nonfragmented 6-5
TEMPTAB_NOLOG configuration parameter 5-39
partitioning, defined 9-1
TEXT data type 6-43
placement on disk 6-1
in blobspace 5-15
reducing contention between 6-2
in table-size estimate 6-5
redundant and derived data 6-45
locating 6-9
remote, used with PDQ 12-6
memory cache 5-24
rows too wide 6-44
on disk 6-9
shorter rows 6-43
parallel access 5-20
size estimates 6-5
staging area 5-24
Table
Thrashing, defined 1-10
splitting if too wide 6-44
Thread-safe
temporary 6-4
UDRs 13-38
Table distributions
Threads
automated UPDATE STATISTICS 13-6
DS_MAX_SCANS configuration parameter 12-6
Table hierarchy
MAX_PDQPRIORITY 3-11
SELECT triggers 10-36
monitoring 2-7, 4-4, 13-47, 13-48, 13-49, 13-50, 13-51
Table scan
page-cleaner 5-7
defined 10-1
primary 12-1, 12-16
nested-loop join 10-2
secondary 12-1, 12-18
OPTCOMPIND 3-10
sqlexec 5-39, 12-16
replaced with composite index 13-20
Throughput
tables
benchmarks 1-4
defragmenting 6-28
capturing data 1-4
TAPEBLK configuration parameter 5-42
contrasted with response time 1-5
TAPEDEV configuration parameter 5-42
measure of performance 1-3
TAPESIZE configuration parameter 5-42
measured by logged COMMIT WORK statements 1-4
Tblspace
Tightly coupled 13-55, 13-57
attached index 9-12
Time
defined 6-5
getting current in ESQL/C 1-7
extent size for tblspace tblspace 6-10
getting user, processor and elapsed 1-6
monitoring
getting user, system, and elapsed 1-6
active tblspaces 6-23
time command 1-6
simple large objects 6-9
Timing
TBLTBLFIRST configuration parameter 6-10
commands 1-6
TBLTBLNEXT configuration parameter 6-10
functions 1-7
TCP connections 3-14, 3-20

X-20 IBM Informix Performance Guide

Timing (continued) UPDATE STATISTICS statement (continued)
monitoring 1-6 MEDIUM mode 13-13, 13-16
TO CLUSTER multiple column distributions 13-18
clause 6-25, 6-27 not needed when statistics are generated
TPC-A, TPC-B, TPC-C, and TPC-D benchmarks 1-4 automatically 13-11
Transaction processing on join columns 13-16
improving using B-tree scanner 13-22 on user-defined data columns 13-16
Transaction Processing Performance Council 1-4 optimizing SPL routines 11-13, 12-10
Transaction throughput, effects of MAX_PDQPRIORITY 3-11 providing information for query optimization 10-20
Transactions query optimization 13-11
cost 1-7 reoptimizing SPL routines 10-32
loosely coupled 13-55 updating system catalog 10-20, 13-11
monitoring 13-48, 13-55 user-defined data 13-38, 13-40
monitoring global transactions 13-55, 13-57 using on very large databases 13-17
rate 1-3 USCR_HASHSIZE configuration parameter 4-21
rollback 7-8 USELASTCOMMITTED configuration parameter 8-6
tightly-coupled mode 13-55, 13-57 User-defined aggregates
Triggers parallel execution 13-38
and PDQ 12-3, 12-4, 12-5 User-defined data types
behavior in table hierarchy 10-36 B-tree index 7-1
defined 10-34 cost of routine 13-38, 13-39
effect of PDQ 12-5 data distributions 13-16
performance 10-35 generic B-tree index 7-23
row buffering 10-36 opaque 7-21
Troubleshooting optimizing queries on 13-38
example of identifying overloaded disks A-1 selectivity 13-38, 13-39
performance degradation 14-13 UPDATE STATISTICS 13-16
sudden performance loss 14-13 User-defined index
TRUNCATE STATEMENT 6-27 DataBlade modules 7-4, 7-21
Truncating tables 6-27 User-defined routine cache
changing size 10-33
contents 10-33
U User-defined routines
negator function 13-40
UDR cache
parallel execution 12-4, 13-38
buckets 10-33
query filters 13-2
number of entries 10-33
query response time 1-5
Unbuffered devices 10-26
statistics 13-40
Unbuffered logging 5-7
thread-safe 13-38
UNIX
User-defined selectivity function 13-2
cron scheduling facility 2-4
User-defined statistics 13-40
iostat command 2-3
Users, types of xiv
network protocols 3-13, 3-16
USING clause, CREATE INDEX statement 7-24
ps command 2-3
USRC_POOLSIZE configuration parameter 4-21
sar command 2-3
USTLOW_SAMPLE configuration parameter 13-18
SEMMNI configuration parameter 3-1
USTLOW_SAMPLE keyword
SEMMNS configuration parameter 3-1
in SET ENVIRONMENT statement 13-18
SEMMSL configuration parameter 3-1
Utilities
time command 1-6
Contiguous
vmstat command 2-3
extents, allocation 6-20
UPDATE
DB-Access 6-25
run in parallel 12-2
dbload 6-25, 7-12
Update cursor 8-9
dbschema 9-4, 9-8, 13-16, 13-19
UPDATE STATISTICS statement
ISA 4-40
and PDQ priority 13-17
capabilities 2-5
automatically generated, viewing 13-10
defined 2-5
automatically running 13-6
monitoring performance 2-4
creating data distributions 13-13
onaudit 5-45
data distributions 10-20
oncheck
directives 11-3, 11-13
-pB option 2-11
effect of PDQ 12-5
-pe option 2-11, 6-14, 6-24, 6-25
effect on virtual portion of memory 4-2
-pk option 2-11
equivalent automatic operation 13-11
-pK option 2-11
guidelines to run 13-11, 13-18
-pl option 2-11
HIGH mode 11-3, 13-11, 13-13, 13-16, 13-19
-pL option 2-11
improving ALTER FRAGMENT ATTACH
-pp option 2-11
performance 9-24
-pP option 2-11
LOW mode 13-11, 13-12, 13-18, 13-38

Index X-21
Utilities (continued) Utilities (continued)
oncheck (continued) onstat utility (continued)
-pr option 2-11, 6-40 -g smb option 6-13
-ps option 2-11 -g smb s option 6-16
-pS option 2-11, 6-15 -g sql option 2-13
-pt option 2-11, 6-5 -g ssc option 13-46
-pT option 2-11, 6-40 -g stm option 2-9, 4-4, 4-38, 13-48, 13-52
and index sizing 7-5 -g sts option 4-4
introduced 2-11 -k option 8-11, 8-13
monitoring table growth 6-20 -l option 2-6
onload and onunload 5-42, 6-1, 6-25, 6-27 -m option 5-31
onlog 1-4, 2-12 -O option 5-24
onmode -p option 1-4, 2-6, 4-10, 4-39, 8-12, 8-15
-F option 4-7 -P option 2-6
-p option 3-20 -R option 2-6
-P option 3-8 -s option 4-40
-W option to change STMT_CACHE_NOLIMIT 4-33 -u option 2-6, 4-4, 8-12, 8-13, 12-16, 13-48
forced residency 4-17 -x option 2-6
shared-memory connections 3-1 monitoring buffer pool 4-10
onparams 5-5, 5-7 monitoring threads per session 3-10
onperf ontape utility 5-42
activity tools 14-12 Utilization
data flow 14-1 capturing data 2-3
defined 14-1 CPU 1-9, 3-1, 3-25
graph tool 14-5 defined 1-7
metrics 14-13 disk 1-11
query-tree tool 14-11 factors that affect 1-12
replaying metrics 14-3 memory 1-10
requirements 14-3 service time 1-8
saving metrics 14-2
starting 14-4
status tool 14-11
tools 14-3
V
VARCHAR data type
user interface 14-5
access plan 10-1
onspaces
byte locks 8-10
-Df BUFFERING tag 5-24
costs 10-29
-Df option 5-21, 6-19
expelling long strings 6-43
-S option 6-19
in table-size estimates 6-7
-t option 5-10, 5-14, 6-4, 7-18
when to use 6-43
EXTENT_SIZE flag for sbspaces 5-21
Variable-length rows 6-7
sbspaces 5-20
View
onstat utility
effect of directives 11-6
-- option 2-6
Virtual memory, size 4-6
-a option 2-6
Virtual portion 4-2, 4-4, 4-19
-b option 2-6, 4-4, 6-5, 6-8
Virtual processors
-d option 3-22, 6-11, 6-12
adding 3-20
-F option 5-39
class name 3-4
-g act option 13-48, 13-50
CPU 3-20
-g afr option 3-19
monitoring 3-21, 3-22
-g ath option 3-10, 12-16, 13-48, 13-49, 13-51
multicore processors 3-5
-g cac option 4-29
NETTYPE 3-14
-g cac stmt option 4-29
network, SOC or TLI 3-20
-g dsc option 4-24
poll threads for 3-14, 3-20
-g glo option 3-21
processor affinity 3-5
-g ioq option 3-8, 3-22
semaphores required 3-1
-g mem option 2-9, 13-48, 13-52
setting number of CPU VPs 3-5
-g mgm option 2-9, 12-6, 12-16
setting number of NET VPs 3-14
-g ntm option 3-18
starting additional 3-20
-g ntu option 3-18
user-defined 3-4
-g option 2-6
Visual disabilities
-g osi option 2-9
reading syntax diagrams B-1
-g ppf option 9-28
vmstat command 2-3, 4-10
-g prc option 10-33
VP_MEMORY_CACHE_KB configuration parameter 3-24
-g rea option 3-22
VPCLASS configuration parameter
-g scn option 5-27
process priority aging 3-6
-g seg option 2-9, 4-18
processor affinity 3-5
-g ses option 2-9, 2-13, 3-10, 4-4, 12-17, 13-48, 13-51
setting number of AIO VPs 3-8

X-22 IBM Informix Performance Guide

VPCLASS configuration parameter (continued)
setting number of CPU VPs 3-5
setting processor affinity 3-7
specifying class of virtual processors 3-4

W
WHERE clause 10-22, 13-2, 13-3
Windows
NETTYPE configuration parameter 3-14
network protocols 3-13, 3-16
parameters that affect CPU utilization 3-3
Performance Logs and Alerts 1-6, 2-3
TEMP or TMP user environment variable 5-8
Write once read many
optical subsystem 5-24

X
X display server 14-4

Index X-23
X-24 IBM Informix Performance Guide



Printed in USA

SC27-3544-06
Spine information:

Informix Product Family Informix Version 11.70 IBM Informix Performance Guide