Complete Reference Books - Storage Networks

Storage Networks: The Complete Reference
byRobert Spalding ISBN:0072224762

McGraw-Hill/Osborne 2003 (526 pages)
Written for storage professionals, this reference outlines the
various architectural considerations of storage area networks
(SAN) and network attached storage (NAS), the components
that satisfy each storage architecture, and various
connectivity options.
Table of Contents Back Cover
Table of Contents
Storage NetworksThe Complete Reference
Foreword
Introduction
Part I - Concepts of Storage Networking
Chapter 1 - The Data Storage and Data Access Problem
Chapter 2 - The Battle for Size and Access
Chapter 3 -
Decoupling the Storage Component: Putting Storage on the
Network
Chapter 4 -
Decoupling the Storage Component: Creating a Network for
Storage
Part II - Storage Fundamentals
Chapter 5 - Storage Architectures
Chapter 6 - Device Overviews
Chapter 7 - Connectivity Options
Chapter 8 - Data Organizational Methods
Part III - Network Attached Storage
Chapter 9 - Putting Storage on the Network: A Detailed Discussion
Chapter 10 - NAS Hardware Devices
Chapter 11 - NAS Software Components
Chapter 12 - NAS Connectivity Options
Part IV - Storage Area Networks
Chapter 13 - Architecture Overview
Chapter 14 - Hardware Devices
Chapter 15 - Software Components
Chapter 16 - Configuration Options for SANs
Part V - ApplicationPutting It Together
Chapter 17 - Defining the I/O Workload
Chapter 18 - Applying the SAN Solution
Chapter 19 - Applying the NAS Solution
Chapter 20 - Considerations When Integrating SAN and NAS
Part VI - ManagementKeeping It Running
Chapter 21 - Planning Business Continuity
Chapter 22 - Managing Availability
Chapter 23 - Maintaining Serviceability
Chapter 24 - Capacity Planning
Chapter 25 - Security Considerations
Part VII - Appendixes
Table of Contents
Foreword
Introduction
Part I - Concepts of Storage Networking
Chapter 1 - The Data Storage and Data Access Problem
Chapter 2 - The Battle for Size and Access
Chapter 3 -
Decoupling the Storage Component: Putting Storage on the
Network
Chapter 4 -
Decoupling the Storage Component: Creating a Network for
Storage
Part II - Storage Fundamentals
Chapter 5 - Storage Architectures
Chapter 6 - Device Overviews
Chapter 7 - Connectivity Options
Chapter 8 - Data Organizational Methods
Part III - Network Attached Storage
Chapter 9 - Putting Storage on the Network: A Detailed Discussion
Chapter 10 - NAS Hardware Devices
Chapter 11 - NAS Software Components
Chapter 12 - NAS Connectivity Options
Part IV - Storage Area Networks
Chapter 13 - Architecture Overview
Chapter 14 - Hardware Devices
Chapter 15 - Software Components
Chapter 16 - Configuration Options for SANs
Part V - ApplicationPutting It Together
Chapter 17 - Defining the I/O Workload
Chapter 18 - Applying the SAN Solution
Chapter 19 - Applying the NAS Solution
Chapter 20 - Considerations When Integrating SAN and NAS
Part VI - ManagementKeeping It Running
Chapter 21 - Planning Business Continuity
Chapter 22 - Managing Availability
Chapter 23 - Maintaining Serviceability
Chapter 24 - Capacity Planning
Chapter 25 - Security Considerations
Part VII - Appendixes
Appendix A - NAS Case Study: The International Image Processing Company
Appendix B - SAN Case Study: The Import Auto Industry
Appendix C -
SAN/NAS Management Case Study: The Southwestern CD
Company
Appendix D - Glossary
Index
List of Figures
List of Tables
List of Sidebars

Back Cover
This comprehensive guide helps you stay on top of storage networking innovations while understanding the realities
of storage networks in the data center. Use the essential concepts of storage networks including architecture,
devices, connectivity options, and data organization methods to design and apply comprehensive and successful
storage solutions.
Learn to work with the two major modelsNAS (Network Attached Storage) and SAN (Storage Area
Networking)and their integration with IP, RAID storage architectures, and virtualization basics. Whether youre a
network engineer, application designer, or IT executive, discover this useful reference to understand, plan, and
maintain a storage network.
Plan, install, and manage a successful storage network
Analyze network file systems, volume managers, and database storage challenges used in storage networks
Understand the realities of data protection and backup/recovery techniques using storage networks
Compare reliability and performance by reviewing the potential and limitations of storage component
technologies
Integrate storage networking solutions with structured and unstructured data using example data center
models
Consider the best options for using disk, tape, and optical media with storage networks
Overcome the challenges and limitations of storage networks using relational database systems and existing
LAN/WAN networks
Learn I/O workload planning and estimate your storage networking capacities and requirements
About the Author
Robert Spalding was one of the first to leverage dyadic processing within the mainframe environment, setting the
foundation for high-end symmetrical processing. He is currently working on advanced storage strategies for imaging
technologies, and distribution and archival technologies for unstructured data products such as audio and video.

Robert Spalding

McGraw-Hill/Osborne
2600 Tenth Street
Berkeley, California 94710
U.S.A.
To arrange bulk purchase discounts for sales promotions, premiums, or fund-raisers, please contact McGraw-
Hill/Osborne at the preceding address. For information on translations or book distributors outside the U.S.A.,
please see the International Contact Information page immediately following the index of this book.
Copyright 2003 by The McGraw-Hill Companies. All rights reserved. Printed in the United States of America.
Except as permitted under the Copyright Act of 1976, no part of this publication may be reproduced or
distributed in any form or by any means, or stored in a database or retrieval system, without the prior written
permission of the publisher, with the exception that the program listings may be entered, stored, and executed
in a computer system, but they may not be reproduced for publication.
1234567890 CUS CUS 019876543
ISBN 0-07-222476-2

Publisher
Brandon A. Nordin
Vice President & Associate Publisher
Scott Rogers
Acquisitions Editor
Franny Kelly
Project Editor
Jody McKenzie
Technical Editor
Giota Vavasis Then (Patty)
Copy Editors
Mike McGee
Lisa Theobald
Proofreader
Susie Elkind
Indexer
Karin Arrigoni
Computer Designers
Lucie Ericksen
John Patrus
Illustrators
Lyssa Wald
Michael Mueller
Alex Putney
Series Design
Peter F. Hancik
This book was composed with Corel VENTURA

Publisher.
Information has been obtained by McGraw-Hill/Osborne from sources believed to be reliable. However,
because of the possibility of human or mechanical error by our sources, McGraw-Hill/Osborne, or others,
McGraw-Hill/Osborne does not guarantee the accuracy, adequacy, or completeness of any information and is
not responsible for any errors or omissions or the results obtained from the use of such information.

Acknowledgments
Writing a technical book is very much like building a house. It springs from inspiration, necessity, aesthetics,
multiple talents, and hard work. Nothing would get published if the myriad of small tasks were not completed to
meet deadlines from a project plan. My appreciation goes to all of those talented people who worked and
contributed to this book.
If there is any dedication of this book it is to Donn Thornton for his inspiration. Donn was instrumental in
teaching me the value of understanding, listening, and more importantly thinking, not only outside the box, but
outside the universe. This book moved forward with the help of Kevin Ansell, whose experiences and
discussions helped me through many difficult times. My thanks and appreciation goes out to Kevin for his
sense of humor and encouragement.
Franny Kelly, acquisition editor at McGraw-Hill/Osborne, championed keeping this project alive and on
schedule. My thanks to Franny for trusting me with the completion of this large project. Regarding the
tremendous task of assembling all the writing into a real-live book, Id like to convey my sincere thanks to my
project editor, Jody McKenzie. Jody (along with her talented copyeditors) was critical to my technical
dissertations becoming a coherent set of text and artwork and, subsequently, a completed manuscript.
I admit it; technically, I had help. Patty Then, storage consultant extraordinaire proved a much-needed
sounding board in the review and validation of a wide set of technical concepts and specifications that make
up this book. Pattys knowledge of systems, storage infrastructures, and data centers was a key resource in
illustrating the technical aspects of the SAN. She is one of those rare individuals versed in performance,
capacity planning, and operational experiences that span both mainframe and open systems. I also need to
thank Himanshu Dwivedi for his work in Chapter 25 on the increasing interests and concerns regarding
security in storage infrastructures.
My thanks and appreciation also go to Ian Spalding, technical and creative writer, for his last-minute
assistance in writing some of the key chapters from my encrypted notes. In addition, Ian was enlisted again to
assist in proofing the book before final publication. We all wish you the best of success on your first novel.
Finally, all of this work would not have been possible without the support, assistance, inspiration, and help
from my partner and wife, Artie. Her help in moving this book toward completion was more than anyone could
ask. Thank you.
Robert SpaldingFebruary, 2003

About the Author
Robert Spalding, a technology consultant, leveraged his 30 years of advanced technology experiences to
develop Storage Networks: The Complete Reference for McGraw-Hill/Osborne. Being at the computer
technology forefront, he was one of the first to use dyadic processing within the mainframe environment,
setting the foundations for high-end symmetrical processing. His open systems management research and
study led the way in developing the first management tools for massive parallel processing configurations that
are still used today. As a research analyst covering the storage industry, he identified and published the early
research that evolved into complex storage network technologies of today.
Robert is currently working on advanced storage strategies for imaging technologies and distribution as well
as archival technologies for unstructured data products such as audio and video. Storage Networks: The
Complete Reference is the latest in a string of more than 50 publications that include research papers,
technical articles, and music compositions. He can be reached through his independent consultancy at
[email protected]

About the Contributing Writers
Ian Spalding has worked as a technical writer and editor for several start-up companies. He is currently at
work on his first novel.
Himanshu Dwivedi is a Managing Security Architect at @stake, Inc., where he leads the Storage Center of
Excellence (CoE), which focuses research and training around storage technology, including Network
Attached Storage (NAS) and Storage Area Networks (SAN). At @stake, Himanshu forms part of the San
Francisco-based Professional Services Organization (PSO) providing clients with network architecture
assessments, attack and penetration services, secure network design, and secure server analysis. He is also
part of the @stake Academy, where he is a lead instructor in several security classes.
Himanshu is considered an industry expert in the area of SAN security, specifically Fibre Channel Security. He
has given numerous presentations and workshops regarding the security in SANs, including the SNIA Security
Summit, Storage Networking World, TechTarget, the Fibre Channel Conference, SAN-West, and SAN-East.
He holds a wide spectrum of well-diversified security experiences, specializing in the telecommunications
industry. Additionally, Himanshu has worked with major organizations in the United States, Europe, South
America, and Asia, including some of the major software, manufacturing, and financial-based entities.

About the Technical Editor
Giota Vavasis Then (Patty) started her professional career in 1982 as an applications programmer. In 1986,
Patty began using her programming skills in the data center to write utilities for data-center automation as an
MVS Systems Programmer while supporting 24/7 Enterprise Data Center operations. While in this capacity,
Patty developed expertise in areas of Data Center Security, Storage Management, Capacity Planning,
Performance Measurement and Modeling, Business Continuance, and Disaster Recovery, to name a few. In
1992, Patty moved from working in the data center to working as a vendor. Past companies she has
represented include Hitachi Data Systems, EMC Corp, StorageTek, and, most recently, XIOtech/Seagate
Technologies.
Patty has held various positions over the last 21 years in the IT industry, such as Systems Support Engineer,
Sales and Consulting Engineer, National Accounts Manager, Customer Service District Manager, Product
Marketing Manager, and Sales and Education Consultant.
Awarded numerous honors over the years for her outstanding contributions, Patty currently heads the
Consortium Technologies Group, a consulting and education practice based in the Tampa Bay Area, that
brings together a team of /world-renowned experts with real-world customer-focused experience.

Introduction
Storage networking represents a fundamental paradigm shift in how data is stored and accessed within
computer systems. Consistent with any major technological change, storage networking brings with it the
challenge to understand new models of storage architectures and, more importantly, the realities of integrating
new solutions into the data center. Storage Networks: The Complete Reference was researched, designed,
and developed to be a general, yet pragmatic, reference for those in the IT community who are already
entrenched (or plan to be involved) in this fast moving, and at times, difficult-to-understand storage foundation
for the future.
I believe storage networks to be the first major enhancement in computer architecture since the client/server
revolution began in the late 1980s. Since that time, the need to store greater amounts of data has never
ceased. The success of the client/server phenomenon pushed the processing power of the server component
technologies to new levels that in turn exceeded the limits of storage devices and their related connectivity
components. This situation drove systems and storage vendors to develop an ever-increasing, yet short-lived,
set of disk and tape products that reflected the exponential appetite of end users for increasing capacities of
online data. Within the search for higher capacity devices with faster access to data, the combined interests of
the vendor and user communities lead to the development and formation of two influential standards, the
Peripheral Component Interface (PCI) and the Small Computer Systems Interconnect (SCSI). These
standards formed the initial foundations that focused attention on the I/O challenges within the client/server
architectures and enabled, more than ever before, peripheral components to be connected to servers allowing
additional devices to operate using those connections.
However, even these innovations could not effectively manage the torrent of storage demands that increased
exponentially during the 1990s. The storage industry and data centers of the time searched for new ways that
this overwhelming capacity of storage could be installed and accessed. This provided much of the emphasis
to shift the storage connectivity paradigm from a direct connect model to a network model. With storage
networking concepts emerging as a viable storage solution, this enabled two things. First, it uncovered an
existing model of storage networking from the shadows of a specialized UNIX orientation, the Network
Attached Storage (NAS) device. Secondly, it encouraged and spawned the integration of existing technologies
to proliferate storage on its own network, thus developing into the Storage Area Network (SAN) model.
Storage networking products both in the SAN and NAS industries proliferated during the late 1990s when the
technology investments were at their peak. The disparate innovation and technology du jour development this
often produced has leveled into an established industry with both established and entrepreneurial players.
Storage networking has crossed the product chasm and continues to evolve rapidly, although with more
emphasis on data-center value and integration. Advancements are moving forward with storage initiatives
such as iSCSI and connectivity through related advancements like InfiniBand and RapidIO.
Unfortunately, the value that storage networking brings to the IT community remains so diverse that it can
easily become compromised by the complexities of installation and management. The most important aspect
to storage networking is leveraging the value of a diverse set of solutions. I believe this will be the critical
motivator in the evolution of storage networking both now and in the foreseeable future. Establishing an
understanding of storage networking and identifying components of a storage networking solution and their
application within the data center became key design goals for this book.
Storage Networks: The Complete Reference was designed to be a general reference for the IT community.
The book was developed to address the pragmatic issues that relate to learning a new technology,
understanding it, and applying it from an IT perspective. As with all reference material, it can be used in a
number of ways. First, as a reference in understanding the concepts behind storage networks. Secondly, as a
tool to help plan and target applications of the technology. Finally, the book is meant as an extended reference
in helping your activities move beyond installation and into maintenance and management. However, it can
also be used as a curriculum for classroom studies, whether employing the complete book or its parts in a
formal or informal educational setting. Usage of the book depends on your existing knowledge of computers
and, specifically, storage systems.
The book is intended for IT professionals who have a need to understand storage networking and its related
issues from a data-center perspective. The examples in the book are composites of real-life situations and are
intended to reflect major points in planning, installing, and managing storage networking in the data center.
I believe storage networking will continue to evolve into its own infrastructure, and storage networking will form
the cornerstone of this migration and evolution, subsequently changing the landscape of data centers of the
future.

Foreword
Sean Derrington
Senior Program Director
Infrastructure Strategies
META Group, Inc.
IT organizations (ITOs) increasingly view storage as a strategic differentiator for existing and new applications.
Many ITOs recognize the critical role information plays in the business, and leveraging that information
requires an unassailable storage infrastructure foundation. I believe forward-looking ITOs have to continue to
decouple storage and server selection. This aids in ensuring storage solutions are viewed as much more than
an afterthought for new applications. For most organizations, I believe this is a best practice, and the trend will
continue as the storage market matures in technical capabilities and more IT professionals become well
versed in the arts and values of storage. Even though current capital expenditures are under extreme scrutiny
(and will continue to be through 2003), leading ITOs are acquiring storage solutions on a tactical basisyet,
they are planning and selecting the vendors on a strategic basis.
Overall, storage is moving to a services model, comparable to that of IP networking and (future) application
infrastructure. Storage services will be tiered in function, price, and complexity, and delivered to applications
and lines of business. For ITOs to deliver on this direction, storage (all aspects of itdisk, tape, disaster
recovery, capacity planning, and so on) must be perceived and treated as an infrastructure component.
Infrastructure components are common and reusable across as many consumers (for example, applications
and servers) as possible; storage should be no different. Consequently, this requires a storage infrastructure
to be an initial consideration (not an afterthought) to drive business priorities. Moreover, infrastructure design
and planning teams must have seats at the business project table and translate technology into business
value.
Typically, IT architects create architectural principles and guidelines addressing three- to five-year planning
cycles, while infrastructure teams design and plan component technologies (vendor- and product-specific),
targeting current through 36-month requirements. The challenge going forward will be how ITOs handle the
transition from storage infrastructure planning and design to storage operations. This is a critical success
factor for ITOs in general, but especially for storage management, because capabilities are rapidly maturing
and vendors are rapidly innovating. I believe that, through 2004, storage infrastructure and storage operations
may be organizationally in the same groupbut in the longer term (2005/06), I believe these responsibilities
will diverge and become distinct groups, mirroring current systems, network, and application infrastructure and
operations teams. The objective of infrastructure planning is to determine the scope, scale, and design of
infrastructure necessary to provide application service levels required by the business in the short, medium,
and long term. Furthermore, the primary design goal for information systems must be to enable rapid change
in business processes and in the applications and technical infrastructure that enable them!
The implementation of automated networked storage (Storage Area Network [SAN] or Network Attached
Storage [NAS]), which rationalizes (reduces the number of variants in) storage management applications and
optimally integrates these functions, will need to leverage tiered storage software, hardware, and professional
services to be able to measure (both internally and for the business) a successful storage infrastructure.
Automated networked storage will provide businesses with agility, dramatically increase productivity, and
enable ITOs to manage terabytes and petabytes successfully and cost-effectively. This book will address, in
detail, the various architectural considerations, the components that satisfy each storage architecture, and
various connectivity options for a networked storage environment.
Storage and Application Service-Level Agreements (SLAs)
Because all applications and servers are not created equal (as determined by business requirements),
prioritization for storage services (from premium enterprise-class to internal workgroup storage) must be
evaluated. By 2004/05, more than 60 percent of the servers in the data center will be connected to networked
storage, and I expect more than 70 percent of storage capacity to be networked in this time frame. Although
there can be modest increases in procurement costs to network storage, the resulting benefits in management
productivity, increased utilization, agility, infrastructure and operations personnel savings, and scalability are
substantial.
Servers sharing external storage resources can also aid in holistic capacity planning and increase overall
capacity utilization (such as in GB/TB, Fibre Channel [FC] fabrics, and external subsystemsimpacting data-
center floor space, and so on). For applications that are more volatile, storage can be reassigned to the
servers as the capacity threshold (6080 percent utilization) is reached. This is an area where enterprise
storage procurement and capacity planning are important (compared to internal storage). I generally
recommend a six- to nine-month buying cycle for enterprise storage and a competitive storage environment.
Overall, ITOs can plan on an approximate 35 percent per year decrease in hardware prices (810 percent per
quarter) and, given the state of the current global economy, should proceed cautiously on long-term
acquisitions (12+ months).

Storage Management
Enterprise storage management will likely surface as the most significant contributor to ITO operational
savings, because human resources remain the highest data-center cost. These costs will be increasingly
important as more servers participate in networked storage (the disk and tape SAN). ITOs must continue to
remove the dependencies and stove piped thinking of operating systems and storage management. Many
operational economies of scale and flexibility will be gained by employing such an approach (for example,
capacity managed per administrator). Certainly, storage management will still need to interface with the other
operational disciplines (such as systems management, network management, database administration), but a
center of storage management excellence is the design objective.
The goal is to streamline and automate many of the daily tasks that consume staff resources. For example,
during the past 24 months, FC fabric management has moved to a core function under the purview of many
enterprise storage vendors. And I believe additional tasks, such as storage provisioning/allocating (reducing
the number of individuals involved and the time required from hours or days to minutes), resource
management (which users/applications are using particular capacities), and topology mapping (which
resourcesphysical or logicalare allocated to applications) will be among the first tasks within multivendor
storage administration.
The key is to examine how and which operational tasks can be automated and leveraged (across server
operating systems and multivendor storage hardware), enabling storage capacity to increase while storage
infrastructure and storage operations staff remain constantor, hopefully, decrease over time.
For automation efficiencies, storage operations should identify solutions that are seamlessly integratedthat
is, a single application that can incorporate various functions (possibly multivendor) and not a loosely coupled
launching pad (or iconic launch) for disassociated islands of automation. Solutions leveraging a central
repository (persistent database) serving as the storage management data warehouse would enable ITO
efficiencies gained through the function of individual management components. Moreover, harnessing this
repository and feeding (analogous to the real-time data warehousing feedback loop) information to other
management elements or ISV applications (such as for chargeback/billing) will be of significant value. Without
this next level of integration/consolidation, users will duplicate efforts, not achieve the ability to scale
(performance, capacity, and so on) their environments, nor control storage staff costs.
I believe storage provisioning/allocation will be one of the initial fully automated areas, yet this will require
organizational (at minimum, functional) changes, because currently this task typically transcends individual
and possibly departmental responsibilities (for example, server, database, storage). Simplifying storage
allocation, which requires a multitude of individual tasks, is not a simple problem to solve. It typically involves
the following:
Assigning/configuring the appropriate LUNs from the physical storage subsystem
Identifying and allocating the number and location of host interconnects on the array
Masking the LUNs to the particular server(s)
Zoning the FC switched fabric
Incorporating and modifying any necessary FC security policies (if available)
Ensuring the volume manager/file system and the application and/or database on the host(s) are updated,
and incorporating the desired changes
Ensuring other storage management applications (for instance, backup/recovery, remote replication
services) comprehend the changes and the implications of such changes in storage policies
Center-of-excellence (COE) best practices dictate investigating the benefits/returns of consolidation and
integration tasks, the most significant of which are those that have the greatest, most immediate impact on the
costs that govern operations. Those gains will be seen as an annuity to the business, versus one-time
savings. This annuity could be significant not just in dollars, but also in business agility and infrastructure
flexibility.

Application Recoverability
Application/DBMS recoverability is one of the most compelling reasons for networked storage (predominately
SAN-based storage) participation. Based on the mean time to recovery (the time it takes from the recognition
of a failure to the time the application begins accepting transactions againnot simply restoring information
from tape to server) discussed with the business, numerous technology options can assist in meeting those
requirements. Snapshot or volume-based replication (either local or remote) can dramatically reduce the
recovery times to minutes, preventing the need for a tape restore (but not eliminating the need for tape
backup/recovery).
Although restoration is a significant and necessary capability, the state of the information/data can be much
more important than the existence of the information. Recovering data in an inconsistent stateor data from
one database that reflects a point in time that is significantly different from another reliant databaseis often
useless. Consistently managing replication and recovery options (possibly from multiple vendors) will prove to
be operationally beneficial, as will rationalizing backup/recovery software and hardware across as many
platforms as possible (often excluding the mainframe). Consistently managing various storage vendors
snapshot/volume replication can provide further operational savings, striving for policy-based automation.
Moreover, just as all applications are not created equal, storage replication software is no different. Integration
with operating systems, clustering software, applications, and databases, as well as quiescent time (the time
required for a consistent view of the application), will often differentiate offerings, given that replication
software has become much more prominent during the past 36 months.

Data and Media Center of Excellence
There are three high-level layers of the data and media COE: conceptual (policy), physical (infrastructure), and
operations (operational). Each of these three is a superset comprised of multiple functional and subfunctional
elements. Each of the functions within each category and associated job titles (suggested) are detailed as
follows and serve as a guideline for functional and organizational structure.
It is also important to note that, although many of these functional categories exist in the mainframe
environment, simply having mainframe personnel responsible for all systems is not necessarily appropriate.
There must be a balance of people, processes, and technology to create an effective COE, and the
technologies are often quite different between mainframe and non-mainframe environments. Nonetheless,
COEs should be viewed holistically and include the following.
Policy
Policies offer the highest-level view of storage management, dictate the required infrastructure to meet policy
mandates, and should be governed by a storage policy director (SPD). These policies, which are
interdependent and should be viewed holistically, include the following:
Service portfolios and service-level agreements SPDs should determine what service packages they
will offer to business units, as well as the associated cost/service tradeoffs, based on the needs of the
business units.
Data protection strategies Not all data is created equal, and the data protection mechanisms should be
aligned with the service portfolio.
Vendor selection strategies ITOs should determine, as a matter of policy, whether to pursue a best-of-
breed multivendor strategy (albeit rationalized and not a vendor du jour strategy), or a one throat to
choke single-vendor strategy. This will vary from one category to another, but should be done with an eye
toward strategic objectives, which are not always mutually exclusive. Either way, rigorous interoperability
mandates should be a prerequisite for consideration to enable maximum deployment flexibility.
Infrastructure
After the overall storage policies have been determined, a storage infrastructure architect should be
responsible for designing and implementing the components needed to achieve the service goals. This will
include device and network management (at the functional level) as well as a multitude of subfunctions,
including the following:
Physical topology design In most organizations, it will be necessary to use and integrate SAN, NAS,
and direct-attached storage either for optimum performance or for security requirements and data
protection. The ability of vendors to not only function/coexist, but also support (via customer and
professional services expertise) these advanced heterogeneous environments (multiprotocol, multivendor,
multifunction, cross-organizational, and cross- regional) will determine the most appropriate platforms
(both hardware and software) on which to base infrastructure for the flexibility to adapt, as subelements of
technology evolve over time.
Performance modeling and monitoring The storage infrastructure architect should be responsible for
not only designing sufficient resources, but also for ensuring that service levels are met. This includes
storage components (for example, disk arrays) as well as storage networks (such as FC fabrics) and
working with IP networking personnel, in the case of NAS.
Security Security is an underaddressed area in storage management, due mostly to the current lack of
capability, but should be made the responsibility of the storage infrastructure architect working in
conjunction with the enterprise security architect.
Other subfunctional tasks may include physical subsystem and tape library design, fault and performance
management, virtualization, SAN management, and enterprise planning/design/procurement.
Operations
Operational procedures will not vary greatly from current practices, but should be managed by dedicated
storage administrators. Functionally, traditional systems management should remain within the systems
management group, and the key is to begin separating the storage operations from systems, database, and
application management. To what degree will certainly be dependent on each organization, but the three main
functional disciplines are:
Performance management This consists of performance modeling and management (including FC-
based networks), asset management, chargeback, and billing by application of business unit.
Resource management This involves resource allocation or provisioningthe actual assignment of
storage (for example, LUNs, data zones, volumes) according to architectural/application requirements, as
well as quota management, usage management (for instance, by application or individual user), metering,
and capacity planning.
Availability management This deals with backup/recovery operations. Backup/recovery should be
included with storage management, not systems management. This includes backup/recovery scheduling,
media management, and problem resolution. High-availability server clustering management (in
conjunction with systems management), hierarchical storage management (where applicable), and
replication (for example, remote replication for business continuity) are also of consideration.
Similar to balancing the handoffs with enterprise architects, infrastructure, operations, and lines of business,
the data and media COE is no different. As can be seen, some of the functions and subfunctions transcend
the policy, infrastructure, and operations categories. Communication among groups is a key element for a
successful storage infrastructure that is adaptable to change and is measurable.

Actionable Items
ITOs should undertake the following initiatives:
Rationalizing storage hardware and storage software This should encompass all aspects of storage
across the server and application portfolio. Certainly, storage life cycles must be considered, yet a
rationalization strategy (reducing variations both vendor and configuration) will provide significant
strategic value, even in tactical times.
Creating a storage infrastructure ITOs should begin networking storage resources (SAN, NAS, and
backup/recovery architectures), leverage tiered storage offerings (for example, internal storage, midrange,
enterprise) and functional software (say, replication, server cluster integration, and backup/ recovery), and
look to common components (such as FC switches, host bus adapters, and so on) where possible. They
should also seek new elements that adhere to intensive interoperability standards and procedures to
ensure maximum configuration flexibility.
Optimizing storage operations This includes rationalizing and consolidating management tools and
personnel responsibilities; consistent storage management of multivendor environments is beginning to
emerge and is a significant strategic directive.
Creating a data and media center of excellence ITOs should employ a center of excellence using the
guidelines outlined in the previous section.

Outsourcing
From 2003 through 2006, as ITOs adopt a storage service delivery model resembling that of traditional
outsourcers, accurately measuring storage services and metrics will be a critical success factor in ITOs not
only gaining credibility, but also in justifying (and making an informed decision about) whether a storage and
storage management sourcing strategy should be employed. ITOs can use three categories when determining
measurable metrics. The initial set comprises alternatives to the traditional ROI/total cost of ownership senior
management mandate and is primarily an internal benchmark of (current and future) capabilities that quantify
how the strategy is providing value. The second and third sets, technical and operational, are more strategic
because some of these measurements are not possible with some storage vendors. However, ITOs should
keep the measurements simple, limiting the metrics to those that are the most important to the business.
Often, using network-based SLAs (for example, web hosting) as a framework and mapping them to storage
services provides a good starting point. The three categories available to ITOs are:
Internal benchmarking This includes storage managed per administrator, percentage of storage
utilization, days storage in inventory, data life-cycle multiplier, data availability, and mean time to recovery.
Technical measures These include storage availability, storage latency, data migration/exit clause,
diagnostic, capacity utilization, performance, resource utilization, and mean time and maximum time to
recover/resolve.
Operational measures These include maximum time to notify, moves/adds/changes, project-specific
SLAs, and vendor responsiveness.

Conclusions
ITOs are increasingly realizing the importance of storage, particularly among midsize companies that have
historically not had requirements for enterprise storage capabilities. Through 2005/06, most organizations will
have organized around, and created, a storage infrastructure and operations team (the data and media center
of excellence). Consequently, it is imperative that ITOs begin to view storage in regards to the strategic
importance it will haveeven in difficult economic environments that often, and many times incorrectly, result
in solely tactical decisions. ITOs should undertake a strategic evaluation examining how daily functions can be
leveraged across (and automated by) multivendor capabilities, and how tasks currently consuming significant
resources can be dramatically reduced (for example, recovery, provisioning, and procurement).
Moreover, as storage technologies continue to mature and more servers participate in networked storage
(SAN and NAS), ITOs will be forced into measuring delivered capabilities as storage becomes a service that is
delivered to the business in business speak (such as performance, availability, flexibility to change), and
associated cost tradeoffs based on their selection of tiered services is clearly understood.
Robert Spalding has been a respected authority on storage and storage management for the past 20 years.
Having worked with Robert in the past, he was the first person to begin describing the need, and preparations
necessary, for an adaptive storage infrastructure. In doing so in 1995, Bob began speaking about the need for
a Chief Storage Officer.
Much of this thought leadership has manifested itself in Roberts published works, and specifically in this book,
Storage Networks: The Complete Reference. As an industry analyst covering the entire storage and storage
management market, it has been my experience that simple yet technical reference books are extremely
useful. Im sure storage professionals will benefit greatly by having this as a staple manual on their desk.
Enjoy.

Sean Derrington is a leading authority on storage and storage management infrastructures. He specializes in
storage subsystems, Fibre Channel, Storage Area Networks, Network Attached Storage, backup/recovery,
and disaster recovery architectures, with a concentration on e-business, transactional, and analytic storage
infrastructures. Sean joined META Group in July, 1995. He has a B.S. degree in Material Science and
Engineering as well as a B.S. degree in Engineering and Public Policy, both from Carnegie Mellon University.
Sean can be reached at [email protected].

Part I: Concepts of Storage Networking
Chapter List
Chapter 1: The Data Storage and Data Access Problem
Chapter 2: The Battle for Size and Access
Chapter 3: Decoupling the Storage Component: Putting Storage on the Network
Chapter 4: Decoupling the Storage Component: Creating aNetwork for Storage

Overview
Computer storage, the ubiquitous component that has invaded both our personal and working space, has
taken on a life of its own with the new technology of storage networking. Like it or not, weve become a global
society with a currency of information thats exchanged within todays computer networks. Nearly every aspect
of our world, including the places where we live and work, are accessible through the Internetthe data from
which is stored on enormous online informational databases.
Our lives revolve around computer technology. Although we dont always realize it, accessing information on a
daily basis the way we do means there must be computers out there that store the data we need, making
certain its available when we need it, and ensuring the datas both accurate and up-to-date. Rapid changes
within the computer networking industry have had a dynamic effect on our ability to retrieve information, and
networking innovations have provided powerful tools that allow us to access data on a personal and global
scale.
Both the proliferation of data and the dynamics of storing it have given way to changes in how computer
storage is used and accessed. The networking innovations have driven global, local, and personal computing
to greater levels of sophistication in storage technology. Because of this, were now faced with the problems of
what data to keep, how to keep it, and where to keep it. And if I do keep it, just where do I put it? We store
information everywhere these days. We keep diskettes and CDs stuffed in our desks at home, collect disk and
tape files at work, while countries as a whole continuously add to the mountains of archived computer tapes
kept at cavernous hiding places over fear of nuclear fallout.
With so much data to store and with such global access to it, the collision between networking technology and
storage innovations was inevitable. The gridlock of too much data coupled with too many requests for access
has long challenged IT professionals. To storage and networking vendors, as well as computer researchers,
the problem is not new. And as long as personal computing devices and corporate data centers demand
greater storage capacity to offset our increasing appetite for access, the challenge will be with us. Chapter 1
will explore the data access challenges, driven by networking innovation, against data storage, exponentially
increasing with innovation in media, capacity, and deployment.

The Challenge of Designing Applications
Storing and accessing data starts with the requirements of a business application. What many application
designers fail to recognize are the multiple dependent data access points within an application design where
data storage strategies can be key. Recognizing the function of application design in todays component-
driven world is a challenging task, requiring understanding, analysis, and experience. These functions are
necessary to facilitate user data within applications. The most fundamental and dependent to the application is
the data storage strategy used. Be it a file orientation or relational model, this becomes a major factor in the
success of the application. (See Figure 1-1.)
Figure 1-1: Database choices for applications
In all fairness to the application designers and product developers, the choice of database is really very
limited. Most designs just note the type of database or databases required, be it relational or non-relational.
This decision in many cases is made from economic and existing infrastructure factors. For example, how
many times does an application come online using a database purely because thats the existing database of
choice for the enterprise? In other cases, applications may be implemented using file systems, when they
were actually designed to leverage the relational operations of an RDBMS.
Whether intelligently chosen or selected by default, the database choice leads directly to the overall interaction
the application will have with storage. The application utilizes, and is dependent upon, major subsystems,
such as memory management (RAM and system cache), file systems, and relational and non-relational
database systems. Each of these factors affects the storage environment in different ways. An application with
no regard for these factors results in application functionality that doesnt account for the availability,
limitations/advantages, and growth of storage resources. Most business applications simply dont take into
consideration the complexity of storage performance factors in their design activities. Consequently, a non-
linear performance-scaling factor is inherent in business applications, regardless of whether the application is
an expensive packaged product or an internally developed IT utility program. (See Figure 1-2.)
Figure 1-2: Components that effect application performance
Non-Linear Performance in Applications
The major factors influencing non-linear performance are twofold. First is the availability of sufficient online
storage capacity for application data coupled with adequate temporary storage resources, including RAM and
cache storage for processing application transactions. Second is the number of users who will interact with the
application and thus access the online storage for application data retrieval and storage of new data. With this
condition is the utilization of the temporary online storage resources (over and above the RAM and system
cache required), used by the application to process the number of planned transactions in a timely manner.
Well examine each of these factors in detail.
First, lets look at the availability of online storage. Certainly, if users are going to interact with an application,
the information associated with the application needs to be accessible in real time. Online storage is the
mechanism that allows this to happen. As seen in Figure 1-3, the amount of online storage required needs to
account for sufficient space for existing user data, data the application requires, and unused space for
expanding the user data with minimal disruption to the applications operation.
Figure 1-3: Online storage for applications
However, as users of the application submit transactions, there must be locations where the application can
temporarily store information prior to the complete execution of the transactionssuccessful or otherwise.
Given that online storage is the permanent record for the application, the use of RAM (the computers memory
resource) and system or disk cache (a temporary memory resource associated with the CPU component or
online storage component, also depicted in Figure 1-3) is used. Because of disk caches high speed and
capability to place significant amounts of application data temporarily until the complete execution of the
transaction, this storage strategy not only provides the fast execution of data, but also the necessary recovery
information in case of system outage, error, or early termination of the application transaction.
Second, among the factors influencing non-linear performance, is the number of users accessing the
application or planning data access. Indicated in Figure 1-4, the number of user transactions accessing
storage resources will correspond linearly to the amount of time each transaction takes to complete (provided
optimum resources are configured). Therefore, configurations can be developed so a fixed number of users
will be serviced within an acceptable response-time window (given optimum storage resources). Although
these optimum configurations depend on server computing capacity and adequate network resources, the
transactional demands of the application rely on the storage infrastructure. (See Figure 1-5.)
Figure 1-4: The optimum configuration of user transactions versus storage resources
Figure 1-5: The increasing diversity of application data
Its unlikely that application designers will ever fully develop systems that take into account the necessary
optimum storage configurations. The use of pre-existing components for database and file access will
continue, further removing the application from the storage infrastructure. These tendencies have placed
additional demands on storage strategies in order that the balance between application and storage resources
can be maintained. To address the data storage/data access problem, designers have been forced to think
outside of the box.
The Client/Server Storage Model
Our storage configurations today reflect the characteristics of the client/server model of distributed computing.
Evolving out of the centralized mainframe era, the first elementary configurations of networking computers
required a central computer for logging in to the network. This demanded that one computer within a network
be designated, and that the necessary information be stored and processed on that computer in order for
other users to access the network. Thus, the network server became the location where (on its hard drive)
network information was stored. When users logged in from their client computers, they first accessed the
information on the network server. Figure 1-6 shows the structure most storage configurations use today: the
client/server storage model.
Figure 1-6: The client/server storage model
The client/server storage model provides data storage capabilities for the server component as well as storage
for the network clients. Coupled with online storage, the servers became larger machines with increased RAM
and cache to handle the multiple requests for network access demanded by users. The online storage devices
also became more robust. With larger capacities, its own caching mechanisms, and greater external
connectivity for expansion, it enabled faster data access. However, servers quickly reached their maximum
performance capacities as the client user community demand grew faster than server capabilities. After a
while, the optimum storage configuration became almost impossible to achieve given server performance
limitations.
Along with the demand for client network connectivity, sharing information on the server increased the need
for capacity, requiring networks to use multiple servers. This was not due so much to server capacity but that
online storage needed to be large enough to handle the growing amount of information. So, as demand for
online information increased so did the need for storage resources, which meant the number of servers had to
increase as well.
This quickly grew into the world of server specialization. The network servers had to handle just the amount of
work necessary to log clients into and out of the network, keep their profiles, and manage network resources.
Client profiles and shared information were now being kept on their own file servers. The demand for online
storage space and multiple access by clients required that multiple servers be deployed to handle the load.
Finally, as the sophistication of the centralized mainframe computers was downsized, the capability to house
larger and larger databases demanded the deployment of the database server.
The database server continues to be one of the main drivers that push the client/server storage model into
new and expanded storage solutions. Why? Initially, this was due to size. The amount of information stored
within the databases, driven by user demands, quickly surpassed the capacities of the largest online storage
providers. In addition, the architecture of the relational model of databases (from products like Oracle, Sybase,
IBMs DB2) had a tremendous amount of overhead to provide the extremely attractive functionality.
Consequently, the user data only utilized half of the space needed, with the database occupying the remaining
half.
As the relational database model became pervasive, the amount of databases within the network grew
exponentially. This required that many databases become derivatives of other databases, and be replicated to
other database servers within the enterprise. These activities began to drive the storage growth rates to
extremely high levelson average, 75 percent, but greater than 100 percent occurred in some cases.
The database server model also required that servers become more robust and powerful. Servers supporting
databases evolved into multiple CPUs, larger amounts of RAM, levels of system cache, and the capability to
have several paths to the external storage resources. Essentially, these became configured exclusively as
storage servers, with RAM size, cache, and disk storage capacities being key configuration elements.
However, it also required system and storage planners to provide additional network paths to online storage
resources in order to keep up with user demand. Figure 1-7 depicts the level of sophistication required as
database servers became the dominant storage consumers.
Figure 1-7: An overview of database server configuration
As data grows, storage solutions with the client/server storage model continue to be problematic. Even the
multiple CPU servers and multigigabyte storage capacities could not ensure the servers capability to provide
optimum client access. Limits to the size of database systems, increasing content of database transactions,
and the advent of new datacentric applications required more sophistication within the storage systems.
However, it also demanded that more servers be used to provide access to application information. As a
result, growth rates began to show a varied and asymmetrical growth trend: some servers were deployed with
more storage than required, wasting space; while other servers continued to be split because their capacity
was exceeded, causing application outages due to space shortages.
An example is the exponential growth factors for storage supporting the data warehouse and data mart
application segments. Typical of this is a target marketing application that ensures customers with good credit
are selected and matched with offerings for financial investments. These matches are then stored and
targeted for mailings in which responses are tracked. Figure 1-8 illustrates the operational complexities of
storage exploitation with these environments.
Figure 1-8: Data moving through a data warehouse application
The actual data traverses from online storage A on server A, where the customer activity and tracking
database resides, to populate the data warehouse server online storage B on server DW. This selection of
good customers from server A populates the database on server DWs online storage B where it is matched
with financial investment product offerings also stored on server A. Subsequently, the aggregate data on C is
developed from combining good customers on server B to form a mailing and tracking list stored on server
DWs online storage C.
As we have described in our example, data moves from server to server in the client/server storage model.
The capability to store multiple types of data throughout the network provides the foundation for selecting,
comparing, and utilizing this data for business purposes throughout the enterprise. However, this comes at a
price since databases are costly in terms of large external storage and server resources, which they require to
both process data and communicate with the network. Moreover, the capability to maintain these
environmentsdatabase systems, more sophisticated hardware, and storage infrastructurerequires
additional technicians and programmers.
As data grows, so does user access. The problem of access has to do with understanding the challenges of
the client/server storage modelthe biggest challenge being the Internet. The amount of data stored from the
Internet has stressed the most comprehensive systems and pushed centralized mainframes back into the
limelight purely on their strength in supporting thousands of online users. However, largely due to the
economics of client/server systems, Internet infrastructures continue to be based on the client/server storage
model.
As web servers began to grow due to the ease of deploying web pages, not to mention the capability to
connect to an existing networking structure (the Internet, for instance), the amount of users able to connect to
a web server became limited only by addressing and linking facilities. Given the client/server connectivity,
servers on the Internet essentially had no limitations to access, other than their popularity and linking
capability. Because of this, an even more aggressive move was required to populate multiple servers and
specialize their uses. The example shown in Figure 1-9 shows how web servers specialize in user access,
user profiles and accounts, web pages, and mail. Web servers are also replicated to balance the demands of
users accessing particular sites.
Figure 1-9: An overview of web server storage
If we examine the performance and growth factors for storage supporting the web server applications, we find
data moving in a fairly predictable way. A typical example is the infrastructure of an Internet service provider
(ISP). Although they support multiple web-based applications, their main service is to provide personal users
with access to the Internet. Therefore, they store two fundamental items, user profiles and mail, as well as
links to other servers on the Internet. Figure 1-9 illustrates the operational simplicity of storage exploitation
within these environments.
The actual data traverses from online storage on web server A, where the customer authentication and access
takes place, then moves to online storage on WebUser servers where profiles, activity tracking, and home
pages are stored. Essentially, the user waits here after logging in, until they issue a request. If the request is to
check e-mail, the WebUser transfers the request to the WebMail server where the request is placed in an
index and transferred to the location of the users mail. In this example, were using WebMail TX.
However, if the user issues a request to access another URL in order to browse an outside web page, the
WebUser server will send the request to web server A or B. Either of these servers will transmit the request
through the Internet, locate the outside server, bring back the web page or pages, and notify the WebUser
server who in turn will transmit the web page(s) to the customers client PC.
Although Figure 1-6 is a simple example of an ISPs web server configuration, the process becomes
convoluted upon much interaction with the ISPs storage infrastructure. We can estimate the scope of this
infrastructure by multiplying the number of customers by one million. If each customer stores 8MB of data
within their profiles, e-mail, and activity tracking, this becomes 8 terabytes, or 8 million MBs, of necessary
online data storage. In estimating the scope of interaction (or data access), each customer issues, on
average, 20 requests for data with most being outside the ISP infrastructure. Each request requiring, on
average, 500KB of data to be transferred, therefore becomes 10MB of data transferred. Multiplying this by one
million customers equals 10 terabytes of data transferred through the ISP storage infrastructure every half
hour.
The capability to maintain these environments, like the database server environments with database systems,
multiple servers, and exponentially expanding storage infrastructure, requires an ever-increasing number of
skilled technicians and programmers.
Reaching the Client/Server Storage Limitations
The preceding examples of datacentric applications, data warehouses, and ISPs demonstrate the increasing
limitations of the client/server storage model. As the demand for data and online storage continues, capacities
and functionality must grow to meet these challenges. Although disk drives, are becoming physically smaller
and are holding increased amounts of data, and temporary storage devices such as caching have advanced,
they cannot overcome the I/O limitations of the client/server architecture.
We are left asking ourselves: Is it necessary to go through the server just to get to the data? Why cant we
maximize all the online storage we have? The amount of servers we have has far exceeded our capability to
manage the attached storage. Isnt there a way to share data among servers rather than duplicate it
throughout a company?
In addition, for every byte stored on disk there is the possibility that someone will access it. More and more
users are coming online through Internet-based capabilities. Increased functionality and access to a diversity
of data has created an informational workplace. Our own individual demands to store more and more data are
pushing the limits of computer networks. It will be an increasing challenge to accommodate a global online
community with client/server storage configurations.
The Boot Disk of Innovation
Data flow challenges have historically been the boot disk of innovation. The data storage/data access
problems have prompted the industry to bootstrap itself with creative solutions to respond to the limitations of
the client/server storage model. In 1995, in what began as efforts to decouple the storage component from the
computing and connectivity elements within a computer system, became a specialized architecture that
supported a significantly dense storage element (size, for example) with faster connectivity to the outside
world. Combining elements from the storage world with innovation from networking technologies provided the
first glimpse into how the industry would address the data storage/data access phenomenon. The
collaborative evolution of these two distinct entities, storage and networking, formed the genesis of what we
know today as storage networking.
As with most out of the box innovations, storage networking combines aspects of previous ideas,
technologies, and solutions and applies them in a totally different manner. In the case of storage networking, it
changes the traditional storage paradigm by moving storage from a direct connection to the server to a
Network connection. This design places storage directly on the network. This dynamic change to the I/O
capabilities of servers by decoupling the storage connectivity provides a basis for dealing with the non-linear
performance factors of applications. It also sets the foundations for highly scalable storage infrastructures that
can handle larger data access tasks, share data across servers, and facilitate the management of larger
online storage capacities. All of which are foundations for building a new storage infrastructure and
provisioning a comprehensive data storage/data access strategy.
However, this architectural change comes with costs. It requires rethinking how applications are deployed and
how the existing client/server storage model should evolve, and, how to develop new storage infrastructures.
This will require a revised view of IT storage infrastructure strategies. Characterized by struggles over size and
access, IT professionals and storage vendors have moved to the front lines of the battle.

Overview
Chapter 2 will begin to define the general concepts of the storage networking model and the differences
between it and the client/server storage model. Well explore the details of how application design and system
implementation influences storage infrastructures. With these discoveries, well begin to demonstrate the
benefits of storage networking and how it meets the challenges posed by traditional client/server storage
implementations as discussed in Chapter 1. By understanding the characteristics of business and support
applications, the inherent value in putting storage on the network begins to take shape. Using standard IT
planning activities and the upfront knowledge of application-oriented issues, integrating storage networking
design benefits prior to system implementations will greatly increase an IT organizations ability to develop a
successful storage infrastructure.
Client/server storage continues to use an I/O bus connectivity standard to communicate with storage devices,
both from an internal and external perspective to the server. The I/O bus standard has evolved from original
ISA bus configurations (standard to early PC servers) to todays Peripheral Component Interconnect or PCI.
The server uses the I/O bus to connect with a storage device; however, it then communicates with the device
through a device-specific language or protocol.
These range from ATA and ASA device protocol standards to SCSI, the Small Computer Storage Interface
protocol. SCSI is the most popular among device vendors and customers for use with externally connected
storage devices because of their capability to support multiple devices and more sophisticated data protection
strategies such as RAID (Redundant Array of Inexpensive Disks). It is important to distinguish between the
bus and device communications standards and methods because device communication protocols, like SCSI,
and data protection architectures, like RAID, will continue to be used within storage networking solutions.
However, as well discover, the I/O bus connectivity model will give way to network connectivity architecture.
(More detailed information on the I/O bus, bus standards, and device protocols can be found in Chapters 6
and 7.)
As indicated in Figure 2-1, the storage devices were directly connected to the server, regardless of the type of
protocol. This required all communications with the data stored on the storage devices to be handled by the
server. As we have seen from examples of data warehousing and web site data movement, this places a
tremendous amount of overhead on the server. Rendering it, as we have seen, to further specialization just to
handle the overhead of dealing with application data.
Figure 2-1: Directly connected storage in traditional client/server environments
Storage networks changed this entire model by creating unique network-based storage architectures that
enhance existing client/server computing topologies. By allowing storage devices to be connected directly to
an existing network, or through its own (storage) network, creates two important changes to traditional storage
configurations:
By providing more direct access paths between storage devices, servers, and the clients, the user
transactions can go more directly to the data, bypassing much of the overhead of I/O operations and
unnecessary access to and through the server.
Storage networking grants business applications greater access to data with increased efficiency. In other
words, storage networking makes it easier for applications to share data, providing servers the capability
to connect to larger amounts of data.
Storage networking technologies have evolved into two distinct modelsNetwork Attached Storage,
commonly referred to as NAS, and Storage Area Networks, referred to as SAN. NAS allows storage to be
placed onto an existing client/server network based upon Ethernet standards utilizing the standard TCP/IP
network protocols. SANs, on the other hand, create a unique network just for storage and are based on a
protocol called Fibre Channel. (More information can be found in Chapter 16.)
Both NAS and SAN are depicted in Figures 2-2 and 2-3, respectively, showing enhancements to the
client/server computing model.
Figure 2-2: NAS example showing enhancements to data access
Figure 2-3: SAN example showing enhancement to data size
Metro Data AreasA Look to the Future of Storage Infrastructures
In the future, distributed computer systems will increasingly take on characteristics of our urban and
transportation infrastructures. Our data is stored in large server farms that produce circumstances and
problems similar to those in a large metropolitan area, managing traffic flow in and out, policing and
monitoring performance and security, and handling overcrowding and sprawl and the occasional accident.
Fueled by powerful servers connected to larger and larger storage systems, the Internet will continue to
evolve into a complex array of data highways that literally crisscross the globe. This complex network of
highways allows users to access data regardless of location, application logic, or specific storage system.
These conditions will accelerate the growth of concentrated data areas that serve both public and private
data highways (e.g., Internet and intranet networks).
Metro Data Areas will be located in corporate data centers, Internet service provider infrastructures,
government and public institutions, and small business computer configurations. Research indicates that
Metro Data Area evolution is being driven by an almost 100 percent annual growth in online storage. The
Metro Data Area will be driven into the next generation of storage technology in an attempt to deal with
this congestion and scalability dilemma and to construct a data transportation system that will handle the
tremendous amount of data moving in and out of these areas.
Be prepared for the rush hour commute. How long will it eventually take to get your data?

Why Storage Networking Makes Sense
Many IT support organizations comment on their applications, saying, 'We have bigger data, more sources,
and need a single distribution strategy.' In the eyes of IT, this translates as:
Bigger data The data stored online and transferred between storage and server has increased in size,
and the amount of data transmitted between server and client PC is much larger, driven by structured
data (text and numeric data) combined with unstructured data (images, audio, and video).
More sources The application must work on several sources of data to satisfy the client transaction. This
means there are several online storage units that the server must connect to process the application.
Single distribution strategy The results of the application need to be placed in a central location for
access. Generally, this means one thing: Internet accessibility.
The Problem: Size
IT organizations also discuss ways in which they will address these problems. A typical snippet of
conversation might sound like this: 'We must have wider bandwidth for data transfers from storage. The
problem is size. The databases we work with are exceeding the limitations of the server's storage
connectivity.' This translates as:
Wider bandwidth is needed. The connection between the server and storage unit requires a faster data
transfer rate. The client/server storage model uses bus technology to connect and a device protocol to
communicate, limiting the data transfer to about 10MB per second (maybe 40MB per second, tops).
The problem is size. The database and supporting online storage currently installed has exceeded its
limitations, resulting in lagging requests for data and subsequent unresponsive applications. You may be
able to physically store 500GB on the storage devices; however, it's unlikely the single server will provide
sufficient connectivity to service application requests for data in a timely fashion-thereby bringing on the
non-linear performance window quite rapidly.
Solution Storage networking enables faster data transfers, as well as the capability for servers to access
larger data stores through applications and systems that share storage devices and data.
The Problem: Access
Others in IT may argue the point, saying, 'Actually, the real problem is access. We don't have sufficient
resources to access the application server. This will only get worse if we go to a single distribution strategy.'
This translates as:
The problem is access. There are too many users for the supported configuration. The network cannot
deliver the user transactions into the server nor respond in a timely manner. Given the server cannot
handle the number of transactions submitted, the storage and server components are grid locked in
attempting to satisfy requests for data to be read or written to storage.
The single distribution strategy needs revisiting. A single distribution strategy can create an information
bottleneck at the disembarkation point. We will explore this later in Parts III and IV of this book where
application of SAN and NAS solutions are discussed. It's important to note, however, that a single
distribution strategy is only a logical term for placing user data where it is most effectively accessed. It
doesn't necessarily mean they are placed in a single physical location.
Solution With storage networking, user transactions can access data more directly, bypassing the overhead
of I/O operations and unnecessary data movement operations to and through the server.
As we have demonstrated, albeit at a very high level, storage networking strategies can address each of these
issues and make application strategies like single distribution a successful reality. However, this book's
definition of storage infrastructures encompasses the entire IT experience, including both business and
overhead applications. To understand storage networking technology and its benefits, it's important we define
a few terms, especially those regarding applications that have evolved to drive the storage configurations.

Business Applications Defined for Storage
Business applications, or, more specifically, enterprise business applications, will be most effected by and find
value in storage networking. In order to completely understand this phenomenon, we must first define an
enterprise business application, as well as differentiate it from maintenance applications and support programs
processed within IT organizations.
Microsoft defines enterprise business applications as complex, scalable, distributed, component-based, and
mission-critical applications that may be deployed on a variety of platforms across corporate networks,
intranets, or the Internet.
[1]
Enterprise business applications are datacentric and user friendly, but must meet
stringent requirements for security, administration, and maintenance. Beyond these common qualities,
Microsoft further characterizes enterprise business applications by highlighting three specific attributes, which
include:
Large A long-lived, multiuser, multideveloper, multimachine, multicomponent application that can
manipulate massive data and utilize extensive parallel processing, network distributed resources, and
complex logic, as well as be deployed across multiple platforms and interoperate with many other
applications.
Business oriented To meet specific business requirements, the application encodes business policies,
processes, rules, and entities, is developed in a business organization, and is deployed in a manner
responsive to business needs.
Mission critical An enterprise business application must be robust enough to sustain continuous
operations, be extremely flexible for scalability and deployment, and allow for efficient maintenance,
monitoring, and administration.
Further noted is the difference in scope between enterprise applicationsfor example, multiuser, interactive,
highly available architectures, and personal applications intended for single-user interactions. Although many
PC applications function as the client in client/server architectures, and active code is commonly distributed to
our client through Java or Java-like applications, they lack the distributed nature and high availability
requirements previously stated.
Maintenance and Support Applications
Supporting large, highly available business applications and their infrastructures are numerous maintenance
and support applications. These are implemented within support infrastructure areas. IT management must
use these applications, often called tools and utilities, to maintain the mission-critical characteristics of
enterprise applications. (More information on management practices and processes can be found in Chapter
21.)
It is important to differentiate between these two distinct types of applications, especially given the often
confusing and misleading global use of the term application among software and hardware vendors. This
difference is important due to IT organizational boundaries that focus IT specialties between the programming
and design personnel, the people who develop and implement the business applications, and systems support
and administration personnelthe people who support the infrastructure.
The maintenance and support applications have evolved to demonstrate the same non-linear performance
that business applications do (see Chapter 1 for a refresher on this). Therefore, they can perform just as
poorly and be just as degrading to the operations of the data center as their business counterparts. The
support applications have the following distinct attributes:
Their results are focused upon supporting and maintaining server platforms and components that create a
continuous operational infrastructure.
They enhance ITs abilities to maintain levels of reliability, availability, and serviceability for the data center
and evolving Metro Data Areas.
The usage of support applications provides necessary information to maintain a scalable and manageable
data center and Metro Data Area infrastructure.
The tools and products within these areas defined as storage management applications can be some of the
most datacentric within the data center. They are defined by the following broad categories:
Archive/Backup/Recovery Provides IT with the tools to initiate data protection policies that copy data,
provide recovery mechanisms, and move data to offline storage media.
Storage Resource Management Enhances the usage of resources that make up the storage
infrastructure. These include utilities such as monitoring the activity of stored data, mechanisms, and the
enforcement of online storage space quotas, as well as the collection of historical I/O system information
and status.
Storage Data Management Complements the storage infrastructure with enhancements to file/systems,
volume management strategies, fault tolerant mechanisms, and utilities such as online space data
maintenance activities, defragmentation, optimization, and compression.
Storage Security A new category that has extended into the storage area from network user
authentication and resource access technologies.
Although storage management applications are necessary in providing an organized and manageable
environment, they are not business applications. The characteristics and base purposes of both are very
different. While business applications automate and ultimately provide the process whereby the business runs
(as well as directly impact overall business performance), storage management applications are considered
necessary evils and maintenance overhead to business application processing and availability. Make no
mistake, the evils of support tools and practices are a necessary cost of doing business, and when not
managed effectively, they will impact business performance.
[1]
Reference Microsoft Publication, Visual Studio - Developing for the Enterprise, Document No. DTX03-
55298-0598

Building the Business Application
In trying to bring more and more data into play, application designers typically place things closest to the user.
However, as we learned in Chapter 1, if this is done without regard for storage or data infrastructures, the
application can be quickly compromised once in production. The application design process (as shown in
Figure 2-4) will then give way to the realities of system implementation (as shown in Figure 2-5).
Figure 2-4: The application development lifecycle
Figure 2-5: Application implementation challenges
The application designers concentrate on the business requirements and produce a requirements document.
This determines what will be provided to end users at the completion of the development process. The
document, once finished and signed off by management, will also be used as a basis for an external design
document. This is generally comprised of an external view of end-user functions, access points, and outputs.
It's possible that sources, types, and quantities of data will be factored into external design discussions. This is
followed by internal design activities where the application programmers determine the logic and identify
external software development components to be used in building the application system.
It is during the internal design process that application designers and programmers work with database
administrators to provide a logical design for the application data. Unfortunately, in many cases the physical
attributes of the database requirements remain unaddressed. These design activities are followed by an
iterative process of actually building (for example, coding) the application by a team of application
programmers.
As components are completed, they are tested on their own-something known as unit testing-and then
integrated with other components to form a major function of the application system. These functions or
combined sets of components are integrated and tested in what evolves into a larger process of system
testing. As all the components come together, user testing and acceptance begins, where actual end users
test the system for accuracy and functionality. The components are modified in an iterative process as
functionality errors are encountered. Once the system is completely tested by a segment of end users and
signed off by management, the system moves into production mode. This is where the application first
encounters the storage infrastructure.
Moving into production mode is a tricky business. In large systems, other activities come into play such as
user education, migrating users from older systems into new systems, and ensuring that the application's
business data stays in sync from one system to the next.
However, as the new business applications progress to this point, implementation will produce some
interesting challenges, especially in the absence of storage infrastructure information and planning. Now it's a
matter of how the application works with an unknown server and storage infrastructure. If the storage
infrastructure has not been taken into account, the window of application non-linear performance (see Chapter
1) becomes very thin and quickly results in an unresponsive application as soon as users and data are applied
in production mode. (See Figure 2-6.)
Figure 2-6: The realities of application implementation
A third process in the application development cycle, albeit unplanned, is a post-implementation analysis and
understanding of the real storage and data access requirements of the application system. This generally
results in quickly provisioning additional resources into the storage configurations to handle the production
processing. Due to the immediacy of this action, the resources are provisioned at a macro level without
accounting for growth or scalability.
Obviously, the third process should be integrated into the application's development process, probably
somewhere between the external and internal design activities. However, this requires two major pieces of
information. First is the availability of storage solutions to offer to the application designers. Secondly, a more
detailed understanding of input and output processes is needed to enhance the application's design through
I/O processing and storage requirements beforehand. (More information on this can be found in Chapters 6, 7,
and 8.)
It is important to note that enterprise storage infrastructures deal with multiple applications and should be seen
in context of an enterprise view rather than addressed piecemeal for each application system. This becomes
problematic, however, during an implementation frenzy when the new applications are being deployed online.
This points out another level of difficulty: developing an effective storage strategy that works with the
traditional client/server storage model. The ability to look at the requirements from a cumulative perspective,
noting the data size, characteristics, and access requirements on an enterprise scale, requires storage
configurations that support multiple applications in a scalable manner. The application design-development-
implementation process will not change, it will only get faster and less resource-aware. Accounting for it
through an enterprise storage infrastructure strategy helps, but this must be integrated with storage solutions
that mediate the challenge. (This forms a whole topic outside the scope of this book. Additional information
can be found later in Chapter 17.)

The Benefits of Storage Networks on Business Applications
As stated previously, storage networking technology is being implemented to deal with extreme congestion
and scalability issues within enterprise data centers. (See the Metro Data Areas sidebar earlier in this
chapter.) IT management can expect the following benefits with the design and implementation of storage
networking configurations and solutions.
Scalability of Access to Business Application Data
Figure 2-7 demonstrates the shift from centralized data access to networked data access for both SAN and
NAS configurations, respectively. The platform benefits are clear, more servers can connect and utilize
storage devices concurrently, thereby eliminating congested access to single servers, as well as the
complexities and overhead involved in maintaining multiple copies of data for each server. From the business
application perspective, this allows the application logic to be distributed throughout the servers, maintains
centralized access to the data associated with the application, and eliminates a single point of failure for the
application.
Figure 2-7: Moving to networked data access with a SAN
NoteOne associated attribute of this configuration could be a storage protection strategy of data mirroring
or RAID functionality to reduce the single point of failure for data availability.
Consolidation of Business Application Data
Since maintaining multiple, synchronized copies of data is no longer necessary, storage devices previously
connected directly to servers can now be used more efficiently by serving multiple servers with multiple sets of
data. Eliminating the copying/synchronization process increases business application availability and reliability
by decreasing data maintenance down times (copying/synchronization) and enhancing data integrity by
eliminating state/timing differences from multiple copies of application data.
Faster Access to Data Through Improved Bandwidth and Centralized
Configurations
The increased bandwidth associated with moving to storage network configurations has clear advantages in
the amount of data transferred. Data now moves up to two times faster using Fibre Channel protocols than
directly connected storage devices using SCSI protocol. Within SAN environments, this requires a new
network using Fibre Channel (FC) protocol. NAS configurations offer even higher levels of Ethernet speed
when configured on dedicated subnetworks that do not conflict with client network traffic. Both technologies
allow for configurations that can sustain an increased number of business application transactions while
setting a foundation for maintaining consistent and reliable response times.
NoteNAS boxes are used as thin servers with large storage capacities that remain dedicated to
processing I/O requests through a distributed file system identified to subscribing application
servers. As a result, NAS configurations become dedicated I/O extensions to multiple servers.
Usage will continue to increase capability to run popular relational databases in scope as NAS
becomes a reality.
Increased Business Application Availability Through Efficiencies in
Storage Management
As noted previously, there are multiple maintenance/support applications necessary to maintain platform
environments. Figure 2-8 demonstrates the capability of devices within a SAN configuration to communicate
with each other, thereby allowing many of the sever-based, datacentric maintenance/support applications to
be optimized. Support applications can now be offloaded from the initiating server, which means data is copied
from storage network device to storage network device (for example, disk-to-disk, disk-to-optical, disk-to-tape,
and so on). Most business applications suffer during this maintenance process because data is unavailable
during the time it is being copied. Although necessary for maintaining storage backup procedures and policies,
this type of operation (in other words, copying data directly from device to device) can greatly improve the
availability of business applications by reducing the time data is unavailable. Unfortunately, NAS suffers from
its server-based architecture (albeit a thin server) and attachment to an Ethernet network. Backup and
recovery operations are generally handled by a dedicated sever within the subnetwork. Presently, NAS
configurations do not allow for optimization strategies such as device-to-device communication, as done in
SAN configurations.
Figure 2-8: Device-to-device communications within a SAN

The Effects of Storage Networks on IT Organizations
A quote for your consideration: Storage networking? Why yes, storage has always been networked. I know
every morning when I come in and log in to the network, I access the sales tracking files positioned on the
server. I also know that when the server goes down, I cant get to those files shared by the sales territory
managers. Networked storage being new? Its always been in the computer systems Ive used. Yep, couldnt
live without it.
This manager of computer applications is obviously commenting on the new concept of networking storage. In
his mind, storage has always been networked and is certainly not a new concept to him. In fact, as youve
probably deduced, hes developed his professional activities to be dependent on storage shared on a network,
across departments, and accessible to both headquarters and remote personnel. What hes talking about is
the concept of the client/server storage model that has existed for many years and is now taken for granted in
most global 2000 companies within the world.
And now, another quote from another IT professional: Storage networking? No, we havent gone in that
direction yet. Were still waiting for the technology to become more mature and stable. However, we have
plans to implement a SAN within the next years planning cycle. Were still relying on our direct attached
storage systems to get us through these next budget cycles. Now, to be fair, Im not sure what the web guys
are doing. They may have some other configuration theyre working with.
What this IT manager of technical services is commenting on is the companys direction in implementing
storage networking in the form of a SAN. Same company, same network, as described by his peer in the
applications area, although in his mind, storage networking is a completely different animal than the traditional
client/server storage model of storage networking.
If we next talk to the Internet/intranet infrastructure manager, we may yet find another story.
Storage networking? Well, not really. We have Network Attached Storage that supports our web servers.
Thats really different from storage networks, isnt it? Im not sure. In any event, our NAS storage boxes serve
us well in allowing multiple storage locations that the web servers can go to for web pages, storing customer
transaction data, customer files, and profiles. Of course, much of the transactional data gets downloaded to
the application servers for processing accounts and updating the real financial records. So, I guess, yes we
have networked storage across our enterprise. I think the server guys are looking at SANs though, so you may
want to talk with them.
The point is that storage networking can and will be implemented across a broad range of services within the
data center. Although it is increasingly available through two distinct architectural models, SAN and NAS, it is
not readily recognized or understood within the data center. In order to utilize this new technology effectively,
you must understand whats driving the storage infrastructures (which we have done in Chapters 1 and 2), as
well as the details of each storage networking model, which will allow you to design and implement effective
solutions.
Most managers responsible for application design and development think of storage as a necessary systems
infrastructureone that should always be thereso we must proceed with the knowledge that some things
can be changed, while others can only be influenced. Part I of this book was designed not only as an
introduction to storage networking but as an initial discussion regarding storage networks within the data
center. If your job involves the technical services of your data center, you may want your peers in application
development to read Part I (or perhaps purchase their own copy). It may spark an ongoing dialogue that finds
each of you moving toward activities that drive and build a storage networking infrastructure unique to your
environment.

Chapter 3: Decoupling the Storage Component: Putting
Storage on the Network
Overview
In Chapter 2, we found the increasing size of online data to be a driving factor in pushing the limitations of the
traditional client/server storage model. The processes that surround the design and development of business
applications, and which fuel data growth, are not equipped to supplement the planning and design of storage
infrastructures, no matter how critical storage becomes. Once applications reach production stage, the
realities regarding actual users and the limitations of current storage environments are quickly quantified in
application performance. The addition of support applications, generally addressed as an after thought, will
further increase demands on new or current storage devices. As the likelihood of major renovations to existing
storage infrastructures increases, post-implementation enhancements become very active, while the impact
on business driven by the risk to successful application implementation becomes an even greater concern.
A solution to these problems is the design and implementation of more scalable and responsive storage
infrastructures based upon storage networking concepts. As we discussed in Chapter 2, storage networking
requires separating the traditional storage components from their server counterparts. This is a major shift
from tightly coupled architectures of the traditional client/server era. However, once implemented, users can
greatly benefit from increased scalability in both size and access and lower the risk of application
implementation and maintenance failures.
This chapter will explore the concepts surrounding one of the current and increasingly popular models of
storage networking: Network Attached Storage or NAS. NAS facilitates the decoupling of storage from the
server by allowing storage to be directly connected to existing client/server networks. It provides a method of
addressing the data access and storage challenges, while creating a scalable storage infrastructure that will
support multiple applications. In addition, this model can compensate for the lack of planning that generally
arises in the implementation frenzy of application development, not to mention the ad hoc need to provide
ongoing storage capacity for business demands.

The NAS Idea
Prior to the days of Windows networking, Java downloads, and the commercialization of the ARPANET, early
users of computer networks worked within the UNIX operating system environments. ARPANET, incidentally,
was the global Internets progenitor; it stood for Advanced Research Projects Agency Network, an agency of
the U.S. Department of Defense. As their networks and applications grew, they found the problem of storage
an increasing frustration. Providing storage on ones own workstation that was shared within the network could
be hazardous to your work as well as to the users that shared that data. Accessing storage on someone elses
workstation was just as frustrating, especially if they had left and turned off their computer for the night (see
Figure 3-1).
Figure 3-1: Early UNIX networking storage issues
Then an idea occurred to them. If users are going to share files, why not simply dedicate a computer to house
large and shared files? For example, a computer connected to an existing network could have storage space
shared with users throughout the network. Thats exactly how early networked UNIX environments evolved to
share data and depend on a file Server, as depicted in Figure 3-2.
Figure 3-2: Early UNIX file servers
Given that working environment, the next logical step was the innovation of a server optimized for file I/O. The
reasoning was that additional overhead for the operating system wasnt needed; the server should have the
capacity for larger disk storage space, and be reliable most of the time. Any additional functionality that a user
might require, like running other applications, or displaying and using graphic functions, wouldnt be needed.
Thus, its necessary that a computer used in this fashion incorporate a large amount of disk storage to be
integrated with a thin OS that only does I/O, and that it can be attached to existing networks. In addition, it
should allow users to access their files on this special server as if the files were on their own workstation,
essentially allowing the files to be shared with other users throughout the network. As UNIX environments
became the platform for innovationdue to its unrestricted code base, use with academic communities, and
virtually free price tagthe setting for the development of an optimized network file server was put in motion.

NAS Evolution and Development
The built-in networking functionality of UNIX prompted the development of a system that not only allowed
users to share files across the network but which provided a method of organizing data that allowed the server
to share its files with users within the network. Thus, the importance of the UNIX file system became apparent.
The file system, as shown in Figure 3-3, forms the foundation for data sharing.
Figure 3-3: The file system
The file system is a method by which data is organized and accessed within a computer system. Generally a
component of the operating system, its employed by users and the OS itself. On behalf of application
programs and interactive users, the system organizes data into logically defined instances, where the
operating system can find information it needs to collect data on their behalf, and perform an I/O function to
satisfy their request. The OS and the users form logical associations between the segments of datawhich
are referred to as filesthrough a naming convention as defined by the operating system.
Emergence of the Network File System (NFS) drove the innovation behind the development of a special
computerwhose only job was to do I/O. Two pioneering innovators, Auspex and Network Appliance, became
the first to offer a downsized UNIX OS with larger amounts of disk storage and the capability to connect to a
TCP/IP network. It leveraged an NFS file system organization that allowed users and their client computers to
connect to their data, as shown in Figure 3-4. The result was the emergence of the first Network Attached
Storage product.
Figure 3-4: The network file system
The early NAS storage devices were UNIX-based and used largely in areas of academic research, scientific
applications, and computer science development. The early deployment provided a method of storing large
amounts of data used in research areas, such as simulating meteorological conditions, geological conditions,
and nuclear testingin other words, it aided any project that involved massive amounts of data which couldnt
be easily stored on individual workstations or file servers. Early NAS products also proved to be quite
productive in code development areas, where computer code had to be shared between developers and
development teams. Figures 3-5a and 3-5b depict these types, respectively.
Figure 3-5a: Early NAS scientific applications
Figure 3-5b: Early NAS code development applications
The emergence of networking required the file system to become shared throughout the computers on the
network. Consequently, the network file system allows a computer to make available its file system to other
users within the network. Other systems and users within the network can attach to the file system and access
the data. The remote files from the shared computer are thus seen as local files by the local operating system,
thereby providing a virtual set of files which dont actually exist on the local computer. Instead, their location is
Virtual, or Virtualized.
The next evolution was the integration into the traditional client/server environments supporting Windows
operating systems. This initially provided a much more hybrid approach, given the capacities of early windows
networks, proprietary protocols, and usage of database servers running UNIX. The first stage of Windows
networking was the formation of a type of network file server, still in use today, using proprietary networking
systems. Figure 3-6 depicts these early configurations that functioned as File Servers, not necessarily
network attached storage, but which served in a similar capacity to the early UNIX servers mentioned
previously.
Figure 3-6: Early Microsoft file servers
However, Windows did have the capability to perform Remote Procedure Calls (RPCs) and through this
function it was able to participate within a networked file system. The capability to provide and link to network
drives became the Windows NFS compatible architecture, with the capability to participate within TCP/IP
networks as well as proprietary networks based on Novell and Banyan proprietary packet-switched
architectures. (More detailed information on file systems, network drives, and virtualization can be found in
Part II of this book.)
These networking architectures gave way to the networking protocol standards of TCP/IP, based upon the
standard OSI model. (Open Systems Interconnection is a standard description or reference model for how
messages should be transmitted between any two points in a telecommunication network.) Although largely
driven by everyones attempt to integrate with the Internet and growing web application infrastructure, this was
something that the early UNIX users had long been doingfor example, already connecting to ARPANET.
These activities caused the development of a file system for use within the prolific web server software.
Although Internet servers evolved and consisted largely of UNIX-based servers, the web application software
phenomenon was founded on Windows systems. This formed the basis of the Common Internet File System
(CIFS), a file system that was compatible with Windows web Internet environments and which could
participate within a standard TCP/IP networked environmentand thus within UNIX-based environments.
Therefore those NAS vendors who already supported TCP/IP networks and network file systems, moved
quickly to support multiple operating environments and file systems. Todays NAS devices come in a wide
variety of flavors. However, most industry leaders support both UNIX-based environments based upon an NFS
file system architecture with support for Windows through CIFS file systems. As indicated in Figure 3-7 these
offerings often allow both systems access to a NAS device.
Figure 3-7: Schematic of the NAS configuration
As we discussed previously, early NAS devices supported UNIX environments only. These file-oriented
servers were used in academic, government, and computer science applications in which data requirements
surpassed standard internal and external storage devices where no other device could be used. These
environments furthered the black box characteristics of NAS as a storage device. The reason being that
users could not configure NAS servers to run other applications or even access the OS itself. The OS in the
NAS box is what is called a Real-Time Operating System (RTOS), a clever name for a specialized UNIX OS
with a special set of functions for a specific assignment. In the NAS case, this was file I/Ofor example, the
capability to optimize the execution of file access to the connected storage, and leverage the NFS system as a
shared catalogue for network users.

An Operational Overview of NAS
NAS boxes remain thin servers with large storage capacities that function as dedicated I/O file servers.
Requests from clients are routed to the NAS server through network file systems that are installed on
subscribing application servers. As a result, NAS configurations become dedicated I/O extensions to multiple
servers. Usage has given way to popular relational databases (for example, Microsoft SQL/Server and
Oracles relational database products), although these remain problematic given their file orientation (see the
section titled NAS Caveats later in this chapter).
NAS servers work today in a number of different environments and settings, the most diverse being in storage
networking. However, their value continues to focus on data access issues within high-growth environments,
as well as how they address particular size challenges found in todays diversity of data types.
Leveraging NAS for Data Access
The inherent value of NAS continues to be its capability to provide storage quickly and cost effectively by
using the resources that already exist in the data center. Todays solutions offer compatibility in both UNIX and
Windows environments and connect easily into users TCP/IP networks.
A typical solution for handling a large number of users who need access to data is depicted in Figure 3-8.
Here, an end-user constituency of 1500 is represented. In the scenario, the users access data files under
100KB in size for read-only purposes that are stored across a number of servers. If the user base increases
by another thousand over the next year, the workload could be handled by upgrading current servers as well
as adding new servers with larger storage sizes to support this increase in data and traffic. This solution,
however, would be costly and difficult to manage.
Figure 3-8: A NAS for data access
However, as Figure 3-8 shows, the NAS solution is used to combine the data spread across the current ten
servers, placing it on the NAS storage device. Users requests for data now come to one of the five application
servers attached to the network and are then redirected to the NAS box where their request for data from a
particular file is executed. Upon completion of the requested I/O operation within the NAS box, the data
request is redirected to the requesting user. Because the NAS devices are optimized to handle I/O, their
capacity for I/O operations exceeds typical general-purpose servers. The increased I/O capacity accounts for
the collapsing of servers from ten to five. Moreover, they can scale higher in storage capacity. As the users
and their respective data grows, increasing the NAS devices provides a more scalable solution for handling
increased data access requirements within the scenario.
This solution does two major things for the data center. First, it provides a much larger window in which the
applications can operate before they reach the non-linear performance curves. Given that both users and
storage resources can increase on a linear basis, this adds to the value and stability of the enterprise storage
infrastructure and can support multiple applications of this type through these configurations.
Secondly, it provides a cost-effective solution by applying optimized resources to the problem. By utilizing
resources that existed within the data centerfor example, existing server file systems and the TCP/IP
networkit eliminated the need for additional general-purpose servers and associated network hardware and
software. More importantly, it reduced the need for additional personnel to manage these systems.
Leveraging NAS for Data Size
The ability of a single server to handle large amounts of data is becoming a common problem. As Figure 3-5
illustrates, a configuration supporting storage of geophysical data can be tremendously large but with relatively
few users and less-sensitive availability requirements. However, storing gigabyte-sized files is a problem for
general-purpose servers. Not only is it difficult to access the data, but its also very costly to copy the source
data consisting of external storage devices such as tape to the processing servers.
By adding a NAS solution to this scenario (see Figure 3-9), the data, as in our first example, is consolidated on
the NAS devices and optimized for I/O access through the network. Although the sizes of the files accessed
are much larger, the number of users are low and the resulting user requests for data can be handled by the
existing network. Therefore, the current resources are effectively utilized within the required service levels of
the sophisticated applications.
Figure 3-9: A NAS for data size
In this case, the user requests for data are similarly received by the server and redirected to the NAS devices
on the existing network. The NAS completes the requested I/O, albeit one that is much larger in this case,
achieving levels of 300 to 500MB per file, and is then redirected by the server back to the requesting user.
However, an additional anomaly to this example is the sourcing of the data through tapes that are sent to the
data center. The tapes are read from the server and copied directly to the NAS devices. Conversely, data from
the NAS is archived back to tape as new simulations are developed. Given that multiple servers can reach the
NAS devices, multiple input streams can be processing simultaneously in order to perform I/O to and from the
offline media, which in this case are the tape units.
This solution, although different in perspective from our first example, provides the same benefits. The larger
window of operation for the application itself demonstrates that appropriate storage resources balanced with
data size and the number of users will stabilize the linear performance of the application. It also shows that
NAS provides a more cost-effective solution by consolidating the data within an optimized set of devices (for
instance, the NAS devices) and utilizing the existing resources of the data center which in this case were the
server and associated file systems and network resources.

NAS Caveats
The caveats for NAS are twofold. The architecture employed by NAS products does not lend itself to
applications that are dependent on sophisticated manipulation of the storage mediameaning any application
that either has its own embedded file system to map and store data to the online storage directly, or that works
with data management products, such as relational databases, which do the same. Secondly, applications that
deal with highly volatile data do not perform wellin other words, data that is subject to constant and
continuous updating by a large user base.
Writing to disk directly is a characteristic of many applications first developed within the UNIX operating
environment. Relational databases commonly have this architecture. The problem manifests itself in
performance as these systems have developed their own ways of manipulating data and in a sense have their
own internal file system. Having these run under an additional file system penalizes the application by forcing it
to do certain things twice.
However, any application that contains a database designed in a monolithic fashion will have performance
problems when operating within NAS storage environments. For example, e-mail applications have embedded
databases that are not scalable. The monolithic nature of their databases can therefore force all traffic into a
single path or device creating a performance bottleneck and recovery problem if the storage device or path is
encumbered in any way.
On the other hand, NAS provides the most cost-effective solution for applications that are file-oriented, contain
read-only files, or which have large user bases. The architecture behind NAS within these application
environments provides a scalable way for large numbers of users to obtain information quickly without
encumbering the servers. This also provides an effective way to upgrade storage quickly through existing
networking resources. This has proven itself within the ISP data centers where NAS storage devices are the
solution of choice given their capability to handle large amounts of data through the storage of web pages, and
users who require random read-only access.
We have discussed both the NAS architecture and issues in summary fashion. More detailed discussions of
the components that make up the NAS solution, how they operate, and how to apply the NAS solution in data
center settings will be discussed in Part V of this book.

Chapter 4: Decoupling the Storage Component:
Creating a Network for Storage
Overview
Chapter 3 summarized the architecture, configurations, and benefits as well as caveats of putting storage on
an existing network using NAS storage devices. NAS can use existing network resources and excels in
addressing large-scale data access read-only applications. However, it is limited by its file-oriented I/O
architecture that limits this storage networking solution when addressing heavy online transactional processing
(OLTP) applications.
An alternative to NAS is the other cornerstone of storage networking, the Storage Area Network (SAN). SANs,
like their counterparts in NAS, allow storage to be connected to a network and provide access to multiple
clients and servers. The fundamental differences are the methods in which they accomplish this. SANs require
their own network to operate, which provides a significant increase to throughput. SANs also provide direct I/O
access from the devices connected to the network, thus providing yet another fundamental shift in distributing
the I/O workload among applications.
This chapter will discuss the fundamental concepts of the Storage Area Network as well as providing an
insight into the evolution of distributing I/O operations within and throughout processing configurations.
Because SANs fundamentally shift the I/O architecture of processing configurations by moving direct storage
operations to a network, its important to understand a holistic picture of how these complex technologies
developed. Within this chapter is a brief discussion of the evolutionary ancestors of the SAN, most of which
have not been rendered obsolete or extinct. Many of the solutions may be configurations installed within your
data center.

The Data-centric World
We live in a data-centric world that consumes information at an amazingly fast pace. The information we
process as individuals all starts out as data stored somewhere and, more importantly, is a requirement driven
by an application that generates the information that we are presented with. Where did all these applications
come from and why do they require such vast amounts of data? Certainly these are questions that are beyond
the scope of this book; however, they remain the fundamental driving force behind the innovation and
advancement of I/O processing.
The support of the increasing data-centric nature of OLTP and data analysis application systems evolved
through experiences in large database applications using centralized mainframe configurations. Even as some
high-end processor configurations became distributed by connecting systems within a tightly coupled network,
the ability to handle high I/O transactional rates and large-scale databases exceeded their limitations. The
advancement and proliferation of relational database solutions exacerbated the problem with its exponential
storage resource appetite when compared to traditional processing of the time. As this relational database
phenomenon filtered into the client/server processing configurations, this increased not only the number of
server configurations, but their complexities and challenges as well. It also provided another target to support
the ever increasing population of data-centric applications.
Many of today's applications rely on some form of relational database products that, as discussed in Chapter
3, were designed to work directly with disk storage devices. In addition to this evolution of processing,
combining large user populations with Very Large Data Bases (VLDB) requires a greater sophistication of I/O
and storage functionality than what is available on traditional, albeit large-scale, mainframe, client/server
storage systems, or NAS.
The effects of the high end OLTP and data-centric data warehouse applications accelerated the advancement
of I/O architectures within high-end server systems and mainframes. These also spawned the development of
increasingly complex distributed processing configurations to support the growth of I/O operations driven by
business requirements that called for increasing amounts of data to be available online. These advancements
have taken the form of Symmetric Multiprocessing Systems (SMP) and Massive Parallel Processing (MPP)
systems. These architectures advanced the I/O capabilities and functionality of traditional computer systems,
and as a lasting effect set a foundation for the development of storage area networking.

Distributing Computer Processing
The enhancement to traditional systems architectures by computer designers in meeting the requirements of
data-centric applications was to distribute the I/O processing. The efforts in research and design moved I/O
further away from the computer elements, (CPU, RAM, and bus connectivity components), which in turn
provided an opportunity to increase the number of I/O paths to storage devices.
Symmetric Multiprocessing Systems (SMP)
Symmetric Multiprocessing Systems (SMP) became one of the first alternative designs to offer the use of more
than one CPU component, as indicated in Figure 4-1. This first became available with IBM mainframes to
increase the throughput scalability and processing capabilities by providing additional CPU components. The
machines that tightly integrated these features with the operating system (MVS) were known by their design
terms of dyadic, triadic, and quadratic, depending on the number of CPUs available on the machine. This
further enhanced the processing of IBM mainframe configurations with their available loosely clustered
systems of MVS-JES2 and little-known tightly coupled clustered systems of MVS_JES3 (see the Note in this
section).
Figure 4-1: SMP processing
This type of architecture has become widely available through all major system vendors as the SMP designs
provided an entry into increased processing power necessary to handle larger applications. However, as with
IBM mainframe configurations that are integrated with MVS features and functions, all SMP systems are
closely integrated with operating systems features. This is necessary to handle the additional activities of
processing with more than a single CPU, sharing memory resources and spaces, and coordination within a
shared I/O bus. Although they provide additional processing capacities for large-scale applications, they also
bring their own limitations. Two are very apparent in Figure 4-1: the sharing of system RAM and the scalability
of the I/O system.
NoteMultiple Virtual Systems (MVS) is a proprietary operating system offered by IBM and used
exclusively with its mainframe computer systems. Available since the middle 1970s, MVSs longevity
can partly be attributed to its modular architecture. This allows software subsystems to specialize in
particular areas of operation. MVS was early in its move to separate I/O functions into a separate
subsystem. In addition, it offers proprietary subsystems of Job Entry Subsystems (providing two
alternatives, JES2 and JES3) that provide, among many other functions, the inter-systems
communications necessary for a cluster of mainframes to share processing workloads. IBM
installations are moving to zOS, an enhanced version of MVS that supports POSIX compliance and
open systems standard functionality such as TCP/IP access and web software services.
Massive Parallel Processing Systems (MPP)
Another advancement from the traditional client/server model was the innovation of Massively Parallel
Processing Systems (MPP). These systems enabled multiple computer systems, each specializing in a
particular aspect of an application process communicating through a high-speed link. The ability to apply
parallel tasks to complex applications provided the processing throughput necessary to complete what was
once considered impossible. The MPP architecture evolved into two categories of machines: one for process
intensive applications and one for database intensive applications. Each machine category was differentiated
by its high-speed link architecture, its integrated database functionality, and configuration flexibility.
The links could be network or switched based (or in some cases a hybrid of both). This link provided a
communications network that enabled each node to work on individual processing tasks, while controlling its
own systems resources, as illustrated in Figure 4-2. This delineates the node structure within the MPP system
whereby each computing node does not share computer resources such as CPU, RAM, or local I/O, thus
acquiring the name Shared Nothing configurations.
Figure 4-2: MPP processing
Even with the potential that MPP architectures demonstrated, the complexities have proved far greater and
limited the proliferation of these machines as general-purpose servers. The complicated enhancements to
operating systems communications to enable the coordination of multiple single-image operating systems has
proved to be problematic and costly. In addition, the overhead required as increased computing nodes are
added increases in a non-linear fashion and consequently limits the effectiveness and throughput. The size,
operational complexities, and management challenges of MPP configurations have limited their usage to
specialized applications.

Distributing I/O Processing
As advancements were being made in SMP and MPP architectures, storage technologies remained tightly
coupled with their computer counterparts. They continued to lag behind processing advancements and
followed an I/O execution model of direct attachment to the server regardless of whether it was connected to
SMP configurations or MPP processing nodes. However, some advancement was made with shared I/O as
evidenced in particular MPP configurations. Sometimes considered the predecessor to the SAN, the 'shared
nothing' model can be viewed as a logical extension to the MPP parallel environments. (Additional discussions
of the SMP and MPP models of processing and I/O systems can be found in Part II.)
The processing advancements driven by SMP and MPP technologies began to set the foundation for
advanced storage architectures. Storage provided the focal point to integrate several innovations such as an
enhanced I/O channel protocol, a high-speed interconnection, and 'shared nothing' architecture to support the
ever increasing need for data access resulting from exponential increases in data size.
An Enhanced I/O Protocol-Fibre Channel
Fibre Channel is a layered connectivity standard, as illustrated in Figure 4-3. It demonstrates the channel
characteristics of an I/O bus, the flexibility of a network, and the scalability potential of MIMD computer
architectures.
Figure 4-3: Fibre Channel Protocol
NoteMIMD stands for Multi-Instruction Multi-Data. It defines a computer design taxonomy in which
multiple instructions can be executed against multiple data sources. Although synonymous with MPP
computer configurations, the MIMD operations can also be supported by SMP configurations where
application tasks are processing in parallel by multiple CPUs operating on multiple data sources.
Because MPP configurations are made up of discrete computers, their I/O scalability is much greater
than SMP configurations that must share I/O operations across a single bus configuration.
In implementation, fibre channel uses a serial connectivity scheme that allows for the highest-level bandwidth
of any connectivity solution available today, 10gigbit. This architecture allows for implementations to reach as
high as 200MB/sec burst rate for I/O operations, with aggregate rates depending on workload and network
latency considerations. Regardless of the latency issues, this is a tremendous enhancement to throughput
over traditional bus connectivity.
A High-Speed Interconnect-Switched Fabric Network
FC operates on a serial link design and uses a packet type of approach for encapsulation of the user data. FC
transmits and receives these data packets through the node participants within the fabric. Figure 4-4 shows a
simplified view of the FC fabric transmission from server node to storage node. The packets shipped by FC
are called frames and are made up of header information, including addresses, the user data at an incredible
2,048 bytes, 2k bytes per frame, and ERC information.
Figure 4-4: Switched Fabric Network
'Shared Nothing' Architecture
As we discussed earlier in Chapter 2, these systems form the foundation for Massively Parallel Processing
(MPP) systems. Each node is connected to a high-speed interconnect and communicates with all nodes within
the system (see Figure 4-5). These self-contained computer nodes work together, or in parallel, on a particular
workload. These systems generally have nodes that specialize in particular operations, such as database
query parsing and preprocessing for input services. Other nodes share the search for data within a database
that is distributed among nodes specializing in data access and ownership. The sophistication of these
machines sometimes outweighs their effectiveness, given that they require a multi-image operating system, for
example an OS on each node, sophisticated database, and storage functions to partition the data throughout
the configuration, and finally, the speed, latency, and throughput of the interconnect. In these systems, both
workload input processing and data acquisition can be performed in parallel, providing significant throughput
increases.
Figure 4-5: 'Shared Nothing' Architecture
Each of these seemingly disparate technologies evolved separately: fabrics coming from developments in the
network industry, Fibre Channel resulting from work on scientific device interconnects, and 'shared nothing'
architectures arising from parallel processing advancements and developments in the VLDB technologies.
Directing these technologies toward storage formed an entirely new network that provides all the benefits of a
fabric, the enhanced performance of frame-level protocols within Fibre Channel, and the capability for multiple
application and storage processing nodes to communicate and share resources.

The SAN Idea
Direct connect storage in the client/server storage model proved to have limitations in bandwidth, the
maximum number of devices available on a bus, and concurrency of traffic. Attaching application servers and
Fibre Channel storage devices to a central location creates a network of server and storage devices. This
configuration allows the application servers to access the storage devices in a network fashion. In other
words, the devices, both servers and storage, must have an address on the network, have the capability to log
in, and have methods available to communicate with other devices within the network.
As depicted in Figure 4-6, servers and storage devices participate in the network. All devices log in to the
network, or fabric OS, which is done by sending messages through Fibre Channel protocol frames.
Figure 4-6: A storage area network

SAN Evolution and Development
The first stage of SAN evolution was the development of Fibre Channel (FC)-based storage devices used with
direct connections, or a point-to-point configuration. These storage devices used FC addressing to process
requests to and from the connected devices, presumably a storage array attached to a server. Although this
enhanced the bandwidth of transferred data, FC devices had the capacity to handle 100 MBps, and were
limited to the two devices within the point-to-point configuration. Although this worked well and used the
existing device communications of SCSI commands, it proved restrictive in regards to the growth of large
storage infrastructures.
As shown in Figure 4-7, the second stage employed arbitrated loop architectures, letting more devices
participate through the use of FC Hubs. This allowed storage devices to continue operating within an
arbitration scheme to share bandwidth, but allowed additional devices to participate within a network fashion.
Although this worked better than point-to-point solutions, the overhead necessary to mediate the Hub, adding
additional devices within the loops required sharing the bandwidth and thus sacrificed much of the high-speed
efficiencies gained with FC devices.
Figure 4-7: Early SAN configurations using HUB architectures
The most current state of SAN configuration is the switched fabric network, as shown in Figure 4-8. This
configuration allows any-to-any device communications to take place within the network. Each device
participating within the network gets full use of 100 MBps bandwidth without sacrificing its performance as
other devices are added. Minimal overhead to moving FC frames through the network is accomplished
through an FC network switch.
Figure 4-8: Current SAN configurations using fabric architectures
Switched Network allowed a significant increase in the number of devices connected with the network. This
allowed for servers and storage devices to be connected in a network fashion and communicate bidirectionally
in an any-to-any manner.

An Operational Overview of SAN
SANs are constructed with many new components from the storage network. The foundation is the FC switch,
which provides the physical connection that allows 'any-to-any' communications within the fabric. SAN
switches provide the hardware and software foundations needed to facilitate the network-the hardware itself
being composed of ports that permit the connection of FC-based devices, such as storage arrays and servers.
To participate in the storage network servers requires a special host adapter, similar to a network adapter
known as an FC Host Bus Adapter (HBA). The HBAs supply the storage network drivers that allow the server
to communicate with the switch and ultimately log in and communicate with storage devices.
Finally, storage devices used with the fabric must be FC-compliant devices (that is, they must speak FC to
communicate with the network). As with Ethernet networks, early users of SANs required the use of bridges
and routers to allow traditional storage devices (such as tape drives) to participate within the fabric. These
devices translated FC protocols to SCSI bus level protocols, basically breaking frames down into bus
segments so a SCSI-level device (such as a tape drive) could be connected to the FC fabric network.
SANs for Data Access
As more paths to data become available, so does the theoretical limit of users accessing that data. SANs
have the capability to increase the number of paths to the data by way of the number of actual I/O paths, or
through the actual transfer of data to the application and, subsequently, the user. If we add up the FC's
capability to operate at gigabit speeds, the SANs' capacity to support a large user base becomes apparent. An
example is depicted in Figure 4-9.
Figure 4-9: SANs for an OLTP database environment
SANs for Data Size
The best example for dealing with size is data warehouses. As described in Chapter 1, these datacentric
applications have extreme data storage requirements in terms of the number of bytes of online storage. What
makes these applications particularly difficult to configure is the complexity involved in processing the data
and the movement of large subsets of data to source and process complex transactions. Figure 4-10 shows
the use of SANs to enhance a particular data warehouse application in terms of the size of databases
involved. This also pertains to SANs data movement activities, which are accomplished while maintaining a
service level.
Figure 4-10: SANs for a data warehouse environment

The Caveats for SAN
The caveats for SAN are several. The main issues of complexity and cost center around the heterogeneous
nature of the SAN itself. Because SANs are configured with several discrete components (see Figure 4-5), the
complexity of configuration and implementation becomes a challenge.
Cost becomes an issue as the SAN devices remain fairly new to storage infrastructure markets and therefore
have not reached commodity-pricing status. Its also important to note the rapid change due to the relative
immaturity of the products, and the reliance on other areas of technologies such as Operating Systems,
storage devices, and management software. The associated cost in terms of user knowledge and training in
new technologies that make up the SANsuch as switches, fabrics, Fibre Channel, and HBAscontribute to
this caveat.
However, an often-overlooked challenge is the esoteric nature of the logical extensions of a servers software
that deals with storage. File systems, volume managers, and physical I/O operations all play necessary roles
in performing storage operations. Within the SAN, these operations become more logical and have to coexist
with other servers that share the fabric network and devices connected. I/O functions must extend into the
fabric itself and operate through Fibre Channel software, storage software, and device micro-code, all being
remote to the server. These new software functions encapsulated within the FC fabric (some of them user
accessible and some micro-code enabled) drive the next logical caveat: management.
Management of the SAN remains problematic even though new products and solutions continue to appear.
However, this scenario of rapid change, due to the newness of the technology, largely renders only
management solutions, which in the end will be merely transient given that SANs will continue to outgrow their
simple configurations, add significant new functions from software vendors, and become integrated with other
network and server technologies.
On the other hand, SANs provides the next generation of scalable solutions for large-scale enterprise
applications. The SAN architecture supports environments that require scalable solutions for large user
populations that access datacentric (that is, database-driven) information that has read/write volatility. This
has proven itself within the enterprise data centers where SANs are the solution of choice given their capability
to handle large amounts of data, using applications such as OLTP and Data Warehousing. In addition,
customers who are dealing with large volumes of unstructured data (such as audio, video, and static images)
are migrating to SANs to support these applications.
In discussing both the NAS and SAN architectures in summary fashion, a more thorough picture of the
components that make them up will come to light. How these technologies operate, as well as how to apply
their value to solutions within data center settings, will be discussed in greater detail in Parts V and VI.

Part II: Storage Fundamentals
Chapter List:
Chapter 5: Storage Architectures
Chapter 6: Device Overviews
Chapter 7: Connectivity Options
Chapter 8: Data Organizational Methods

In Part II, we will review the basic elements and devices that make up computer storage. Although the level of
information will be summary in detail, these are the components you are likely to encounter in dealing with
changes to storage infrastructures brought on by storage networking. In many instances, it will be these
devices, their performance, and logical configurations that are overlooked as they extend beyond the bounds
of typical storage systems. As such, they are critical factors when addressing issues affecting storage
performance. More importantly are the effects these elements, devices, and components will have when
integrating storage networking into your infrastructure.
In Part I, the decoupling of storage devices caused us to reevaluate the necessary functions to develop an
optimum data highway to the server. This quickly becomes a complex task as some storage devices become
remotely connected, such as disks and disk controllers. With the introduction of networking and switching
technologies, their capabilities and functionality begins to change and their operation with other components
that transport data to and from the CPU can become compromised. However, understanding how data travels
in traditional direct connect configurations can be complex in and of itself. It is imperative that these situations
are understood both in context with data center practices and storage device principles of operations, prior to
an in-depth discussion of storage networking.
Storage Elements
Storage can be defined as the components and locations where data is staged before and after it is accessed
by the computer's central processing unit (CPU). The components that make up specific storage devices in
today's computers are multifold, however they still have one main objective. They move data in the most
efficient manner in and out of the CPU to process application program instructions.
The storage locations are multifunctional in usage and multidimensional in capacity. Figure 5-1 depicts the
locations where data is stored for action by the CPU. Although you wouldn't expect to classify some of these
locations as storage, their purpose is to provide a location for data, albeit temporary in many cases, before or
after operation by the CPU. A parochial view of hard disk or archival storage does not take into account all the
components involved in transporting data to and from processing objectives. By gaining a complete
understanding of all storage locations, we systemically address all the elements that affect the supply of data
to the CPU and therefore the system's potential, improving application program performance and storage
device optimization.
Figure 5-1: The diversity of storage locations
Storage locations are characterized by their volatility and temporary nature. This is demonstrated through their
effect on the data stored within each storage component. Data stored on hard magnetic disk, tape, or optical is
considered non-volatile and therefore stores data permanently in context with its logical form-for example, a
customer file, binary program, or an image. In contrast to these devices, there are the components such as
system level cache, controller cache, memory buffers, and disk read/write cache-buffers that are highly
volatile. These components are used for only temporary locations and are staged in physical locations in
increasing proximity to the CPU. The nearest component to the CPU is the fastest, and thus it is more volatile
with the data it transports.
Component speed and CPU proximity drive the balance between slower and faster elements that make up this
store and forward like architecture. As we saw in Figure 5-1, many of the faster components are necessary to
account for the slower performance of others. Moving data into position to be accessible by the CPU requires
staging areas such as system level cache so that CPU processing does not have to wait for a slow
mechanical device such as a disk drive to transport a copy of its data to the CPU. The same can be said for
the reverse operations; once the CPU has completed its instructions, it can use a staging area such as system
level cache so it does not have to wait for the write operation to be complete out to the slower disk drive or be
forced to wait the time it takes to flush and rewrite the buffer space in memory.
The corollary, or trade-off to this condition, is the economics of speed and capacity to price. Consequently, the
faster the component, the higher the price for parking data. An example of the dynamics of price difference
can be found in the comparison of solid-state disks (SSD) versus magnetic disks. Each can perform similar
functions, however the SSD made up of logic circuits from silicon wafers in the form of a chip is much more
expensive than its counterpart, the magnetic disk, which operates using mechanical devices and electronics
on magnetic media. Although each stores data, the SSD device, operating electronically within a chip, is
exponentially faster than the mechanical operation of the magnetic disk. Although there are many examples
within the storage hierarchy, the point is the following: it becomes economically prohibitive to store large
capacities on higher performing components.
Although it's possible to load small databases into the main memory of large machines, you still have the
problem of keeping the permanent copy up to date in case of hardware or software error or system
termination. Therefore, we find the additional corollary to speed and capacity is reliability. As system outages
occur, the temporary storage locations are flushed or cleared, at which point the potential for data loss is
imminent, and for most users, unacceptable. Therefore, recovery of the data that was contained within these
storage locations has to be taken into account. This becomes especially challenging as we discover that some
storage components are accessed remotely as networks are introduced into the connectivity scheme. As a
result, many components (such as controller caches) become shared. We will examine in more detail the
capability to recover various storage elements within these conditions later in Part IV, and how they will
become critical design factors in Part III, regarding both NAS and SANs.
Data Storage Hierarchy
The benefits of understanding all the elements of storage is that it begins to form a foundation for working with
data volatility and the effective use of the multifunctional components that data must traverse to satisfy the
application. As previously stated, the effect these components have on system and application performance,
such as controller cache and a disk's average access response time performance can be the difference
between effective utilization strategies and overcompensation for device inefficiencies. With this knowledge,
the traditional data storage hierarchy (as shown in Figure 5-2) takes on new and challenging characteristics.
Figure 5-2: The data storage hierarchy
The data storage hierarchy extends from the CPU to the storage media. The following highlights some of the
major components that make up this data highway:
CPU Registers The component of the central processing unit that stores program instructions and
intermediate, prestage, and post operations data.
First Level L1 Cache The component that provides a data staging area closest to the central processing
unit; employed for prestaging and post-staging data used by the CPU instructions.
Second Level L2 Cache The component that provides additional data staging areas to the CPU, but
which are further away and which act as supplements to the L1 cache component.
Memory Buffers Locations within the memory where data is staged for pre and post-processing by CPU
instructions.
Direct Memory Access (DMA) A component that allows peripherals to transfer data directly into and out
of computer memory, bypassing processor activities.
Host Adapter Cache/Buffer The component within a device adapter that provides staging for data that is
being read or written to disk storage components.
Hard Disk Controller Cache/Buffer An additional component that provides another staging area for read
data that is waiting to move into the memory or L2 or L1 cache, or waiting to be written into locations
within the disk storage media.
Hard Disk The component that provides a permanent location for system and application data slated to
be stored for access by system and applications programs.
Archival Storage The locations where data is stored for later use. This may be for archiving historical
data no longer needed for business or systems applications, or for copies of business or systems data
needed for recovery operations in the event of system outages or other computer disasters.

Storage Systems
Storage systems have been evolving to logical stand-alone extensions to the client/server processing
architecture (refer to Chapters 1 and 2). Consequently, we have storage systems that are larger, in terms of
floor space, than the servers they connect to, and costlier, in terms of dollars with software and IT
administration. Storage infrastructure costs make up nearly 75 percent of the cost of a client/server
configuration. Why? Because we require more and more data to be available onlinein other words, to be
available to the end users through some form of interactive application. The result is more disks being
attached to the client/server model. With the increase in the number of devices, the resulting complexity in
meeting these demands is enabled through the Data Storage Hierarchy.
So, what makes up storage systems for the data center? There are multiple components that make up the
storage infrastructure that has evolved into a highway for moving data. As discussed previously, the objective
for these components has not changed that is, they are used to stage data as close to the central
processing units as possible, and as we have demonstrated, extend the storage highway into the servers
themselves. This can drive the boundaries of the traditional views of storage into new perspectives and extend
the discussion of what makes up storage systems for the data center. For our discussion, we have simplified
our taxonomy to encompass the major functional configurations in which storage systems are deployed. To
accomplish this, we have included in our examples the components that continued to be housed in the server.
By way of definition and illustration, we have also added a new definition to our discussion, the workload.
Consider Figure 5-3 when viewing our examples.
Figure 5-3: The definition of workloads
Typical Storage System Configurations
Online storage systems define a variety of configurations used to support applications that require immediate
access to business and systems data, as shown in Figure 5-4. These configurations form the basis for support
of many of the additional workloads required to operate the data center.
Figure 5-4: An online storage system
OLTP Workload and Typical Configurations
Online Transaction Processing (OLTP) storage systems are configured with high- performance storage
hardware and software components to support business applications in processing transactions in a real-time
setting. These environments employ a predefined end-user response time that has a prescribed set of
business data operating within several conditional states.
Application performance is expressed as service levels within the data center. Service levels are guarantees
or targets defined between end-user and data center management. These generally consist of elements of
application availability (up time) and performance (response time). Additional information and discussion can
be found in Part VI.
Batch Workload and Typical Configurations
Batch processing storage systems are configured to support applications that are processed in a background
fashion and which dont support end-user access, but that nonetheless require immediate access to business
and systems data to meet a predefined elapsed time interval for processing. These environments are often
highly integrated with OLTP environments and process data captured as a result of the OLTP system, as
shown in Figure 5-5. However, unlike OLTP systems, they are usually optimized toward high throughput
without regard to transactional service levels or response times. Additional discussions in Part VI will address
the interaction between batch and OLTP workloads and the impact they can have on their respective service
levels.
Figure 5-5: A batch storage system
Archival Storage
Archival storage depicts a system used to support applications that archive business and systems data no
longer needed for online applications, or the copies of business and systems data needed for recovery
operations in the event of system outages, data corruption, or disaster planning.
Data Protection Configurations
Data protection storage systems are configured to support maintenance applications that copy data in both
logical and segmented forms for later reconstruction. Most are processed in a background fashion and
therefore dont support end-user access, but nonetheless require immediate access to business and systems
data to meet predefined elapsed time intervals for processing. These environments are often highly integrated
with both OLTP and batch environments and provide recovery of business and systems data in the event of
various system outages. These systems are usually optimized toward high throughput without regard to
transactional service levels or response times, as shown in Figure 5-6.
Figure 5-6: The data protection workload
Near-Line Storage Configurations
Hierarchical Storage Management (HSM) storage systems (illustrated in Figure 5-7) typically support OLTP
applications but are configured with lower performance, and higher capacity, storage hardware, and software
components. These systems support a business applications capability to process transactions in a near-real
time setting, meaning that some user transactions are asynchronous and can take longer to complete as
compared to real-time OLTP response times. However, these environments run with a predefined end-user
response time, that although longer, runs with a prescribed set of business data that operates within several
conditional states.
Figure 5-7: A typical configuration for HSM workloads
In all fairness to the complexities of large data centers, these systems, in practice, are integrated within the
existing production configurations. For example, in most cases the online storage configuration that supports
the OLTP application is the same one that supports the batch cycle during an off-shift cycle. Whether this is
the most effective for batch workloads is a discussion of our first corollary: speed and capacity versus price.
The archive system is often integrated into systems that support some level of HSM or near-line system.
Again, whether it interferes with backup jobs run at night is a discussion of our second corollary: speed and
capacity versus reliability.
Consequently, when evaluating storage systems, one must take into account the multifunction requirements
placed upon it. The likelihood that the most end-user sensitive configurations, such as OLTP- and HSM-like
workloads, will take precedence in resources and administration is quite high. These are the systems that
generally provide support for the end users day-to-day business operations and which bring visibility to its
customers.
With our global definitions and typical workloads, the online storage (OLTP) configurations will form the
primary foundation for the storage infrastructure. This will be enhanced and augmented with other, albeit less
time-sensitive applications, such as HSM supported applications, the application batch cycle, and last but not
least, the archival of data, backup, and recovery jobs necessary for data protection, disaster planning, and
legal purposes.

Overview
Magnetic disks form the foundation of online storage systems. These devices are augmented in storage
architectures with magnetic tape, enabling an economical and enduring media for storing historical
information. Additional storage technologies such as solid-state disks and optical devices play both
specialized and enabling roles that support these two fundamental storage devices. Although the concepts in
this chapter may seem elementary in nature, the mastery of these storage principles are an integral part of
understanding how to apply these basic storage components optimally when designing and implementing a
storage network.
Storage architectures must support various, and at times, disparate business applications. These applications
demonstrate various I/O requirements that place a demand on the storage devices. The specific I/O
requirements for the application are referred to as an I/O workload (more information about I/O workloads can
be found in Chapter 17). The understanding of storage at the device level will contribute to developing an
entire storage system that forms the critical basis for all application performance.
Application systems designed with the supporting storage devices in context with workloads begin to develop
into responsive infrastructures. Without device knowledge and consideration, storage systems often form
monolithic storage farms that dictate a one size fits all approach. Such conditions set the stage for
unresponsive and costly application performance. With today's Plug and Play implementation frenzy, the
thought and analysis given aligning compatible storage devices with storage workloads is generally an
afterthought (more information regarding workload planning for storage networks are contained in Part V and
Chapters 21 and 22).

Peripheral Connectivity Components and Concepts
Peripheral devices are conceptually defined as those computer elements that are not part of the computers
motherboard but must attach to the periphery of this self-contained system. Storage peripheral devices are
connected using two distinct functions, the host adapter and the controller. Although at times they are
integrated into one device, the distinctions between the two, though often fuzzy, are required for all storage
attachment. These devices are depicted in Figure 6-1, which illustrates their proximity to the main computer
system. The basic functionality of the host adapter allows peripheral devices to connect to a servers internal
bus system. The controllers basic function, although it has many, allows multiple devices to communicate
through one adapter path. Together they form the connectivity structure of a servers I/O system.
Figure 6-1: Peripheral connections to the main computer elements
This encompasses all devices that process I/O from the server to include storage, but also printers, plotters,
graphic displays, keyboards, and other devices that interact and are directly connected to the system. For our
discussion, we will focus on storage devices. However, we will devote a portion of our discussion to network
connectivity adapters given their integration and importance in developing storage networks.
A short discussion about network connectivity can provide an important set of principles that encompass the
functions of direct device connectivity and how networks interface with the server. As shown in Figure 6-2, the
Network Interface Card (NIC) connects the system bus to the external network, be it a local area network
(LAN) or a wide area network (WAN). The components are the same as with any host adapter; it consists of
an electronics card that connects to the server motherboard or a backplane.
Figure 6-2: Connecting to the network
This is supported by a set of related software that interacts with the low-level hardware protocols and the
operating system, sometimes referred to as firmware or microcode. Firmware refers to software instructions
that interact directly with microprocessor components. These instructions are specific to the type of
microprocessor being used and generally not accessible to users.
NoteA backplane is a type of connector circuit board where the board has no other circuitry than to
connect together bus slots or connectors. Thus, an entire computer system can reside on a
motherboard that can connect into a backplane for common I/O attachment. Refer to blade
computing, high-end SMP, and shared I/O configurations in Chapter 7 and Appendix D.
The functions of the Network Interface Card (NIC) are to transmit the identity of the server to the network,
receive messages addressed to the server, and translate the network protocols to pass on to the operating
system components. The user or application data and instructions encapsulated within the networking
protocols can now be used by the servers applications.
On the out-bound side, the reverse takes place. The application turns over its data and target location to the
operating system, which works with the NIC software drivers to encapsulate the data and ensure its
transmitted in the correct way to the network.
The network software that enables this communicates and works with the NIC in TCP (Transmission Control
Protocol) or IP (Internet Protocol). Together they form what is called the TCP/IP software layers, or
stacksthe open systems standard used by all computer manufacturers to communicate within networks. As
previously illustrated, we can view TCP/IP simply as the message format and envelope, where TCP allows the
application and operating system to encapsulate the user data while the IP portion contains the envelope that
transports the data to its network target locations.
As discussed previously, all peripheral devices have a host adapter and controller component. This concept,
although viewed somewhat more abstractly in networks, is the same. However, to place this in its proper
perspective we must view the other devices in the network as being externally connected, be they servers or
clients. That means the controller functionality in networks consists of switches and routers that interconnect
the multiple devicesin this case, other computers that are attached to the network.
Figure 6-3 extends the network diagram into the area of controllers, with switches, routers, telephone
exchange equipment, and other WAN-related communication devices. As indicated in this illustration, the
equipment and software that make up the network are numerous; however, none of it needs to be known by
the server or client, other than the network address. We have seen the high-level benefits and configurations
when this concept is extended into storage components. Unlike traditional transient network traffic, storage I/O
operations require a more defined protocol using the same host adapter and controller concepts.
Figure 6-3: Overview of connectivity inside the network

The Host Adapter
The host adapter connects the computers system bus to the external pathway where various devices are
attached. In many cases, this is another bus technology, one that specializes in communicating with specific
device types and which requires unique commands to perform I/O operations with the server. Host adapters
consist of two main components. First is a hardware component, generally in the form of electronic cards that
attach to the server motherboard or backplane (more about bus connectivity in Chapter 7). Just as important is
the related software component that allows the specific devices to drive the underlying hardware bus
protocols, the operating system, and, ultimately, the application.
The combined components, as shown in Figure 6-4, consist of microcode that functions within the electronic
card circuitry, microcode of the basic input/output (BIOS) that exists on the computers motherboard, and high-
level software that executes on the computer. The high-level software (commonly called drivers) works with
the hardware microcode to provide the device command set (native I/O calls), services for the device, the
operating system interface, and error recovery and checking routines. All of these functions communicate with
the operating system through the BIOS microcode embedded in the computers motherboard. Any device
utility software running on the computer will use the same native software.
Figure 6-4: Host adapter functions
Host adapters are provided as bundled solutions from storage vendors, or piecemeal through a host of
hardware vendors. This requires the configuration activities to be both challenging and creative. Regardless of
how the solution is acquired, the configurations are pretty standard, albeit limited. First, there are only so many
host adapters you can attach to a server. This is a function of the server itself and, more specifically, the bus
technology within the server. Second is the effect of multiple software drivers that can operate within a single
server systemwhich means that having multiple vendor software drivers can become risky, given the
different implementation that occurs within the division between the hardware (for example, the microcode)
and the high-level software driver executing within the server.
This requires a homogeneous approach to implementationin other words, a common vendor solution to host
adapters running on a single server. In addition, the number of adapters available for an individual
servermore specifically, the servers busdefines the high end of paths that can be physically configured for
directly connected I/O peripherals. This presents a theoretical limiting factor in storage sizes given the type of
workload.
The number of paths, as depicted by Figure 6-5, shows the limitations of single path configurations when
dealing with even medium-sized databases. Other components within the storage hierarchy will also be
affected.
Figure 6-5: Path limitations through single adapter strategies

The Controller
The controller allows multiple devices to be connected to the adapter. As the adapter forms the primary path
to and from the server, the controller provides the connectivity to the actual devices that communicate with the
server. As indicated in Figure 6-6, the controller functions can be manifold but primarily control the flow of data
to and from the devices. In doing so, it provides an addressing scheme for the devices, shields the server from
the operational complexities of the device, and provides another temporary location for data (prior to it being
written or cached) for faster access.
Figure 6-6: Controller functions
Like adapters, controllers consist of both hardware and software components. However, their primary
functionality is driven by hardware microcode, as indicated in Figure 6-6.
By keeping track of a disks position and being the initiator of the drives read/ write electronics (see the next
section for more details on magnetic head disk assemblies, referred to as HDA mechanisms), the controller
provides data buffering between the disk and the host interface or adapter with its error detection, correction,
retry, and recovery functions. Both of these functions are susceptible to changes in connectivity due to their
communications with the operating system.
The controller also offers the systems attached disks as an acceptable storage location even though the
actual physical mapping may be very different. This allows the controller to translate I/O system requests into
actual physical commands that can be executed on the real disk drive. Typical in most systems, the controllers
and operating system view the disk drive locations as a linear pool of logical locations. This level of abstraction
translation, though having existed in I/O configurations for many years, is necessary to translate logical blocks
of data into physical addresses on the storage media.
This is a key concept within the controller functionality. It demonstrates how the operating system
communicates within a logical set of storage locations that are considered virtual to the controller, which
translates those locations into the physical realities the controller must manage. Figure 6-7 represents an
overview of how logical views are mapped into physical views.
Figure 6-7: The controllers view of reality

Magnetic Disk Storage
Magnetic disk storage is the packaging of magnetic media disks, packaged read/write head electronics on a
spinning electro-mechanical device. Disks are made up of round platters coated with a magnetic material that
can be written and read electronically through the encoding of the magnetic media with binary data.
Figure 6-8 shows the platters that make up the magnetic media and the read/write electronics that surround
the disks, all of which resembles old analogue record players. The read/write electronics are recording heads
that both read and write data. As discussed previously, the head assembly can house data buffers that
compensate for delays in reading or writing data.
Figure 6-8: The disk drive anatomy
The platters are organized into tracks, concentric rings that form the writable surface of the platter. These are
further divided into sectors which form the smallest readable/writable units on the disk. As the disk rotates,
portions of the sectors become visible as the read/write heads are positioned to that track.
Drives are made up of more than one disk, with each storing data on both sides. Each surface must have its
own read/write head and a set of tracks which can be accessed through a single position (for instance, when
vertically aligned). This is called a cylinder. Given this organization, every sector on the drive can be uniquely
addressed by its cylinder, head, and sector numbers.
Sectors are important because they are the fundamental metric that determines the size of data stored within
the disk. The number of bytes in a sector corresponds to the disks formatted capacity. As indicated in Figure
6-8, the sector must contain information about addresses and synchronizing HDA error corrections in the
header fields. Given that the header and associated synchronization gaps are not usable for data, the
formatted capacity is always less. Formatting a drive writes the header information and arbitrary data pattern
to verify correct error checking. Although entire drives are normally formatted at one time, a technique called
soft sectoring allows a single track or hard sector to be formatted.
Technical specifications that surround disks include the capacity, transfer rate, and average access time.
Each metric is important when considering the given workload supported. While capacity may be an issue, the
number of transactions necessary to read data from the disk may compromise seek time. On the other hand,
large transfers of data (like that used by datacentric applications in data warehousing) may push capacity and
transfer rates to their optimum while sacrificing average seek time.
Capacity, as mentioned previously, comes in two forms: formatted and unformatted. As youve probably
guessed, the larger the disk, the greater the amount of addressing and error correction code (ECC) overhead,
thereby reducing space for actual user data.
Transfer rate and throughput refers to the speed at which the HDA processes read and write operations.
Throughput is the amount of data the drive can deliver or accept at the interface on a sustained basis. This is
important because the term interface can have many connotations and implementations.
Average access time is the time needed to position the HDA to a specific cylinder, as well as and the time it
takes for the requested sector to rotate under the heads. Thus, disk rotation speeds are related to capacity.

Disk Systems
By integrating controllers into the disk assemblies, we begin to have complete storage systems. These come
in an amazing amount of configurations and offerings, but the basic model is the attached controller that uses
multiple disks drives within the enclosure.
As disk drives are linked together, they form an array. This is used for basic capacity and performance
enhancements. Disks configured in this manner can be used individually as Disk1 through Disk4, as shown in
Figure 6-9, or they can be used in an integrated fashion by combing the capacities of the array and using them
as one large disk. For example, in Figure 6-9, Disk1 through 4 can be combined to form a virtual disk V which
can use the entire capacity of the array. In this scenario, the application I/O sees disk V as one large disk
drive even though its only a virtual representation of the four physical drives. The application and the
operating system understands that data can be stored on disk drive V and lets the controller take care of
where the data is actually written using the pool of disks 1 through 4.
Figure 6-9: Disk arrays
This forms the basis for advanced functionality in storage systems such as Redundant Arrays of Independent
Disks (RAID) and the concept of storage virtualization. The functionality of RAID provides disk redundancy,
fault tolerant configurations, and data protection resiliency. All this is facilitated through the controller
mechanisms discussed previously and illustrated in Figure 6-6. The RAID decoupling of the physical drives
from the application software I/O requests has driven the usage and advancement of storage virtualization
technologies and products.
Letting the servers operating system believe it has a pool of storage capacity frees up the I/O manager
services, allowing it to perform other tasks, and facilitates the development of applications without the low-
level coding of specific I/O calls. Remember, there is no such thing as a free lunch and the translation of calls
to the virtual pool of data has to be processed either in the storage controller or in the third- party storage
software running on the server. It may even be a combination of both. Either way, the overhead continues to
exist, although it may be offloaded and rely on RAID processed within the controller.
Disk Arrays
The first level of storage array configuration is JBOD (Just a Bunch Of Disks), depicted in Figure 6-9. JBOD
links a set of disk drives together to form an array where each drive becomes an addressable unit. These
provide additional capacity; however, they do not provide any fault resiliency in the event of an inoperable
drive. Partitioning data throughout the disks and providing a layer of virtualization services can be done
through software, which is generally part of the I/O management of the operating system. These functions are
also available through third-party software applications.
Although these functions offer a level of data redundancy (depending on the sophistication of the software),
they do not provide any method of fault tolerance for continuous operations in the event of a lost disk drive.
Consequently, while these guarantee some data loss protection, they do not guarantee that the data remains
available even if disk drives become inoperable within the array. Functions that provide continuous operations
and fault resiliency are provided by RAID.
RAID Storage Arrays
RAID is, as its name implies, a redundant array of independent disks that becomes an addressable unit made
up of separate and independent disk drives (shown in Figure 6-10). The main difference from JBOD arrays is
that RAID partitions data throughout the array and recovery functions. The recovery functions are developed
using disk parity information that is calculated to reassemble missing data from a failed drive to the remaining
drives within the array. The data is distributed throughout the array in a manner most effective to the recovery
and protection strategy. RAID has several levels in which these recovery and protection strategies can be
implemented.
Figure 6-10: The basic RAID architecture for a storage array
The various RAID levels of implementation provide recovery and data protection that is appropriate to the
applications and users continuity requirements. This allows for performance, data protection, and automatic
recovery from drive failures within the array. RAID has become the de facto standard for disk hardware fault
tolerance functions. As an accepted standard, all storage manufacturers offer it; however, all storage vendors
have also evolved into a diversity of proprietary functions that define the type of fault tolerant protection
offered. These are encapsulated within the storage firmware run on the controller and are referred to as both
RAID software functions and RAID firmware.
To make things more confusing, some RAID functions can be provided through software that runs on the
server. This differentiation becomes distinct as we look at the standard RAID levels as defined throughout the
industry. As such, RAID functionality can be selected either through hardware or software components, or a
combination of both.
RAID configuration levels range from numbers 0 through 5 for basic RAID services; however, extended RAID
levels, although proprietary to vendor implementations, have gone beyond basic configurations. Given the
numerous options for partitioning data and the diversity of ways that parity information can be calculated and
partitioned, the options for RAID can be numerous. Through the evolution of real experiences and exposure to
typical applications and I/O workloads, two of the five have proven the most valuable, and therefore are the
most popular. These are RAID levels 1 and 5. RAID levels 1 and 5 will most likely be the ones implemented
within the data center. However, we will pay particular attention to RAID level 4 as an example of how a
vendor can use a proprietary and bundled level.
RAID level 0 simply uses the storage array for partitioning of data without any disk parity information. This
provides for data protection without a recovery mechanism should any of the data become unavailable from a
failed drive. Many software-only data protection mechanisms use a RAID level 0 configuration, where the data
is written to primary and secondary files. Similar to RAID level 1 (mirroring) without any disk recovery
mechanism, this is sometimes called software mirroring. In other cases, the data is striped across the disk
array for performance reasons. This allows the disk write process to take advantage of several head and disk
assembly mechanisms for multiple files with high disk write requirements.
RAID level 1 offers the ability to provide a primary and secondary set of data. As one file is updated, its
secondary is also updated, keeping a safe copy for data protection just like RAID level 0. The major difference
is the calculation of parity information, as shown in Figure 6-11. This is used in the event a disk drive fails
within the RAID array. In that case, the available mirrored copy of the data continues processing. When the
drive failure is corrected, the unavailable copy of the mirror is reassembled through the parity information
contained within the RAID controller. Theoretically, the data is never unavailable and remains online for the
application use.
Figure 6-11: The RAID level 1 configuration
RAID level 5 is a partitioning of the data across the RAID array for performance purposes. This is depicted in
Figure 6-12 along with the complement of disk parity information also striped across the array. In the event of
a disk failure, the missing data is reassembled through the parity information processed by the RAID
controller. Once the failed disk drive is corrected, the data is reassembled back onto the failed drive and the
entire array is back in full operation. Theoretically, the data is never unavailable as in RAID level 1.
RAID levels 2 and 3 are derivatives of mirroring and level 5, and use different strategies for storing the parity
information. Although RAID level 4 is also a derivative of level 5, it provides striping across the array of user
data, and reserves a disk within the array for processing of parity data (shown in Figure 6-13). This is used in
some NAS solutions where the storage is bundled and the RAID array processing is hardcoded within the
packagein other words, you dont have the option to move to other RAID levels.

Magnetic Tape Storage
Tape is the most ubiquitous form of storage today, with more data being stored on tape than any other media.
Although tape storage started as the earliest form of mass storage, it survives today due to its cost resiliency
and continued capability to store large amounts of systems and application data for safekeeping.
Tape Drives
The most important aspect of tape systems is their sequential nature. Even though a good deal of innovation
has been applied to tape technology, write mechanisms, and optimization, the information remains
sequentially bound. This is important because of throughput of write processes and speed of access. The
ability to find a specific location with tape media requires the transport system to sequentially move throughout
the tape to a specific sector.
Once the location is found, it can read the information and reposition itself for additional operations. Within the
writing operation, the tape must be positioned in the proper location and written sequentially. Although this has
been enhanced through software utilities that provide tape catalogue information that speeds location search
for data that has been written to tape, the random nature of access that is inherent with magnetic disks does
not exist within tape.
Another important operational note that should not be overlooked is the effect of tape controller functionality.
The tape controller, as shown in Figure 6-14, is necessary for many control functions. Among the most
important is the capability to address multiple drives within the tape system. Just as with disk controllers, tape
controllers provide a level of transparency to the operating system and applications. In addition, and most
importantly, the majority of controllers will have some form of error recovery and correction when dealing with
data as it comes from the system bus, generally from the RAM buffers. If the tape system has error recovery
code (ERC) in its controller, the system RAM will dump its contents to the tape controller buffers and refresh
its buffer locations for another load. If not, there is a long wait as the RAM buffers slowly transfer their
locations to the read/write head and wait for confirmation.
Figure 6-14: An overview of the SCSI tape drive
The effects of tape devices and operations within a storage network environment cannot be overlooked. We
will discuss tape systems associated with SAN and NAS in Parts III, IV, and V. However, its important to point
out that fast system resources, such as RAM, become further decoupled from slower devices like tape media
through the separation of network devices. As a result, the timing and throughput differences can become
magnified and problematic.
Formats
Like disks, a distinguishing characteristic of tape drives is the format used to write and read data to and from
the media. Popular formats in todays data center environments range from Digital Linear Tape (DLT), the
most popular for commercial enterprises, to new formats such as Linear Tape Open (LTO). The general
format specifies the type of media and the technique the tape heads use to write the data, most of which are
supported by an open standard so that multiple vendors can provide an approximation of an interoperable
solution.
The tape media is divided into parallel tracks. The number of tracks, the configuration of the read/write heads,
and the mechanisms surrounding tape transport all make up the uniqueness of each format. Regardless of
format, the tape media is formatted into blocks. As discussed, the format of the tape depends on the geometry
of the media, the layout of the drive heads, and the transport control mechanisms, which all go together to
define the size of the block in bytes that can be written. Blocks make up segments that can be referenced
within the location of the media for faster access. Each segment has certain blocks that contain error
correction codes and space for directories used by software utilities reserved within track 0 or a special
directory track.
Read/Write Mechanisms
The read/write head mechanisms determine not only the format of the tape (tracks, parity, density, and so on),
but more importantly the throughput and density of the write process and the speed of the read process. In
most cases, todays tape drives contain read/write heads that perform dual roles. Surrounded by a read head,
a write can take place as the read function verifies the operation, and vice versa as the tape transport
changes direction. As mentioned earlier in our discussion of controllers, todays tape systems support
increased levels of data buffering and, in some cases, caching, file systems, and fault-tolerant configurations
similar to RAID technologies with disk systems.
Tape Systems
The utilization of tape systems predates all other mass storage strategies used for commercial and scientific
computing. The importance of tape in data centers, both present and future, cannot be overlooked. In
understanding this legacy form of storage, there are two important points to consider. First is the natural
extension to tape systems (which is unlike online disk media): the Tape Library. As data centers create sets of
information for recovery, archival, and legal purposes, the tapes must be stored in environments that are
secure, environmentally protected, and accessible through some form of catalogue structure. Second, is the
level of sophistication vendors have developed to support the offline, nearline, and online strategies that
encompass tape usage.
The Tape Library
Lets first examine the tape library. Tape libraries provide both logical and physical archival processes to
support the data center. The most important aspect of the library is rapid access to archived information. This
requires some type of index, catalogue, and structure whereby the tape hardware working with the catalogue
and archival software can locate a tape, mount in a drive, verify (yes, its very important to verify its the right
tape), and read.
Provisioning and preparation follows this to support all the direct copying that is done. Generally performed
during an off-shift, production information is written to tapes for a multitude of purposes. Although in many
cases this will be the backup/recovery process, the sophistication of application jobs will utilize several levels
of backup to ensure user data integrity is maintained. Also important is the type and form of copying that is
performed in moving data to and from an online media. This means that some processes copy all data from
the disk, regardless of the application, while others copy data to tape using specific file information, and yet
others provide sophisticated update processes to archived data only changed if the online data has changed.
Some of the more sophisticated processes take periodic snapshots of production data and copy these to
other online or offline media.
Auto-loader libraries are highly sophisticated mechanical hardware that automate much of the provisioning and
preparation in administrating tape media. Auto-loader development was driven largely by the need to provide a
cost-effective way to eliminate the mundane and manual job of mounting tapes. Another major value is the
ability to provide a larger number of pre-stage tapes for tasks like nightly backup and archival processing.
Auto-loader libraries are integrated with tape drives and controller mechanisms and provide various additional
levels of functionality in terms of buffering, error recovery, compression, and catalogue structure. Enterprise-
level media libraries use mega-libraries, called silos, to provide an almost fully automated set of tape activities.
As you would suspect, many enterprise-level tape library systems are integrated into storage area networks.
These SANs enable tape library systems to translate the SCSI-based tape drive operations with the Fibre
Channelbased SAN network. NAS, on the other hand, has only recently integrated tape systems into its
solutions. These can become more problematic given the direct attached internal storage characteristics of the
NAS device.
How these processes and resources are managed requires a good deal of planning and execution within a
generally tight time schedule. In order for this to happen, the correct amount of blank media should be
available, formatted, and mounted in a timely fashion, the scheduling and processing of jobs needs to proceed
according to a predefined schedule, and any exceptions must be handled in a timely manner.
Optical Storage
Optical storage serves as a media where laser light technologies write information in a fashion similar to disk
architectures. Although logically they share similar characteristics, such as random access of the stored data
and have head disk assemblies (HDAs) to read the media, these similarities disappear with the media used
and operation of the read/write assembly. Optical storage compact discs (CD-ROMs) and DVDs (digital video
disks) are driven by laser technology. As we all know, CDs and DVDs are removable storage with much higher
densities than traditional diskette technology. As such, they have become a very successful product
distribution media, replacing tapes.
Optical has achieved notoriety as a new distribution format in the fields of software, entertainment, and
education. However, only recently have they been able to function as a reusable media with digital read/write
capabilities. Unfortunately, their speed and throughput have not met requirements for use within the data
center for commercial backup/recovery, or other archival media.
CDs and DVD optical media have never been favored as a replacement for magnetic disks or tapes in larger
enterprise data centers. However, they has been slowly gaining a niche in the support of special applications
that deal with unstructured data, the distribution of data, and delivery of information. Examples are datacentric
objects like videos, integrated audio/video application data, and entertainment and educational integrated
video, image, audio, and text information. The basic configurations for optical media are very much like tape
drives and libraries. In commercial applications, they are supported through optical libraries which include
multiple drives and auto-loader slots for selecting media.
The major difference between magnetic disks and tape is the use of lasers to write information on the optical
media. This requires modifying a type of plastic media through the use of the laser beam, either by bleaching
it, distorting it, or creating a bubble. The data is written in tracks but uses a different geometry that employs a
spiraling technique to format the tracks from the center of the disk to its outer areas. Regardless of how it
forms the modification, this is the method of writing data to the optical disk. When accessed for read
operations, the optical device turns at a variable rate so that access throughout the disk is close to linear.
However, even with optical attributes and increased data capacities, the optical media has yet to achieve the
throughput necessary for online transactional multiuser systems, or the necessary controller sophistication for
data protection and recovery operations. Consequently, optical libraries will not proliferate within storage
network installations even though they are heavily used in applications that favor imaging, videos, and audio.
Systems that are SCSI-based can participate with SANs through a bridge/router device (see Chapter 14) and
are available as NAS devices using a NAS microkernel front end with optical drives and autoloader hardware
attached.

Overview
In previous chapters, we discussed how effective storage systems are developed from the integration of
multiple storage devices. Understanding the structure of these devices can help effectively utilize these
technologies to support a single complex workload or a combination of disparate workloads. Underlying these
concepts of storage architecture, device operation, and anatomy are the connectivity strategies that complete
the data highway system. All these elements will begin to come together in this chapter as we discuss the
connectivity options available to transport and deliver data from the storage system components to the
computer.
Traditional system connectivity strategies have long been based upon bus architectures. A computer bus is
defined as the lines and controls that connect the diverse components of a computer, ultimately making it a
complete system. As we discussed in Chapter 5, this connectivity is required for the internal server
components, CPU, RAM, and cache. Obviously, we know that external I/O components, such as storage
devices, disks and tapes, and other peripherals like displays, keyboards, and mouse devices also require this
connectivity, but in a different manner. This chapter will begin to explain the components and operation of
typical bus- and network-oriented technologies, and the creative uses of various buses and network
connectivity components that form the foundation for large multiple user systems.
The ability to connect storage systems through networks has greatly impacted the storage I/O industry. All of it
surrounds the connectivity strategy, and the movement of storage systems from discrete bus technologies to
network-oriented architectures. Understanding the bus is paramount to leveraging the movement of storage
I/O into this new network model.
Another important reason to cover traditional bus technologies contrasted against the new network
interconnects is that existing buses do not go away. They remain, and will remain, fundamental to getting data
to and from the CPU (refer to Chapter 5 for more information). Moving data from the storage device through
the storage system and on to the server within a network infrastructure still requires the effective operation of
an internal bus. In order to be effective, traditional bus components must be optimized since they become
parts of a new storage connectivity infrastructure. Otherwise, the most advanced and feature-rich storage
network falls short at the server.

Connections: Getting on the Bus
Internal traffic within a computers motherboard is handled through a set of fast connections etched into the
circuitry, enabling transfer of data to and from the CPU and internal temporary data locationsfor example,
RAM, cache, and CPU registers. The connectivity for external (external to the motherboard) hardware is
handled through the edge (of the board) as an expansion to the board. Take a look at Figure 7-1, which shows
a universal computer bus.
Figure 7-1: A universal computer bus
These connections to the outside world (external to the motherboard) are referred to as expansion buses and
I/O channels. They provide hardware attachment mechanisms that extend the system capabilities beyond
what is on the motherboards. This is where I/O devices are connected and, as we discussed earlier, where the
host adapter components are attached. It is here where we require our first transfer ticket provided by the host
adapters, to get from one bus to the other. As the internal components require a bus to connect to each other,
the I/O devices also have that same need. As Figure 7-1 illustrates, the internal bus, I-bus, is adapted to the
I/O bus, each transferring and controlling data for the connected components. Each bus controls devices with
different mechanisms, different timings, and capacities.
The buses provide the following fundamental functions. As we can see, these are largely electronic
mechanisms that do their work internally through a host of components. As we will explore later, the bus
mechanisms must be able to work with particular devices that speak their language and have common
electronic traits to move data.
Underlying all these sophisticated and complex electronics are several fundamental elements that make up
the characteristics of the I/O bus, as shown in Figure 7-2. These include: throughput (commonly referred to as
bandwidth), address space (how wide the bus is), interrupt processing (or can I react to external events and
change processing modes?), and electro-mechanical attributes, such as how long can I be before my signal is
degraded, and am I a circuit board or a backplanein other words, will I connect to the motherboard or will the
motherboards connect to me.
Figure 7-2: The internal functions of a typical I/O bus
Bandwidth
Bandwidth is the amount of information the bus can transport during a given time. This is usually measured in
MB per second. This is a derivative of clock speed and data width. In other words, a bus using a 5MHz clock
speed and a 4 byte data width has a throughput, or bandwidth, of 5 4 = 20MB per second. Keep in mind this
is what is called aggregate throughput; it is reduced by protocol overhead. We will discuss real speeds in more
detail later in the chapter.
Addressing
The address space provides the capability to determine unique identities for source and destination of
communications within the bus itself. This is determined by the width of the address component; the larger the
address space, the more device operations can be supported. Generally, most buses operate today with 32-bit
addressing architectures, providing an address space of 4GB. (This is important if you are working with older
bus technology or an advanced technology such as InfiniBand or Fibre Channel; see Part V.)
Interrupt Functions
An external event may occur at any time and the bus must be able to take action in order to continue
processing. This interrupt processing is the fundamental concept behind the interactive systems processing
that must occur with multiuser, multitasking workloads. The I/O bus must be capable of interruption of a data
transfer from disk to tape for an event with a higher priority. Be it an error processing routine or a call to read a
block of data from an application, without this capability the bus could not be used for interactive applications,
instead it could only be used for a single device.
Electro-Mechanical
Internal buses on the motherboard may only be inches long; external I/O connections must extend several feet
or meters. These connections are generally some type of shielded cable and will be placed into areas of
electromagnetic interference. I/O buses must have attributes that protect the signals integrity while providing
an adequate physical extensibility. The cable for an I/O bus is defined with regard to its maximum length,
resistance, whether shielded or unshielded.
The motherboard of a computer will have a number of connections or slots that are nothing more than bus
connections. Host adapters are connected at these edge points. The other type of connectivity is the
backplane (refer to Chapter 6). The backplane can have no other electronics or circuitry than to provide
connection points for host adapters. However, a backplane can also be used to connect complete
motherboard systems together along a shared bus. These will be used in tightly coupled SMP and MPP
systems. They are also becoming popular for use with integrating thin computing nodes (for example, a
system-on-board or a motherboard with no peripherals), which entails sharing a common external bus that
may be a network or another type of system interconnectused explicitly in something referred to as blade
computing. (Refer to Chapter 20 regarding future protocols such as InfiniBand, an external example, and
Rapid IO, an internal example.)

Bus Evolution
The importance of bus evolution is visible in terms of the clocking signal increases, numbers of data lines,
interrupt sharing, and something called bus mastering. Bus mastering allowed additional processors and
devices (for example, IDE and SCSI adapters) to contend for bus processing. By contrast, previous
architectures where the system processor had complete control of the bus and scalability became problematic.
The decoupling of the bus through bus mastering from a single point of control came with IBMs Micro Channel
Architecture (MCA); however, it was enhanced with EISA to include auto and software configuration functions.
This first laid the groundwork for plug n play functionality, while the second aided in reconfiguring the bus
(changing device priorities, addressing, and so on).
NoteThis standard bus architecture (Extended Industry Standard Architecture or EISA) extends the ISA
standard from a 16-bit to a 32-bit interface.
The physical decoupling and enhanced scalability of the bus was achieved with the Peripheral Component
Interconnect Standard (PCI) and the resulting PCI mezzanine bus architecture. This was the first bus
implementation not tightly coupled with the system processor. This allowed the PCI bus to expand its
connectivity by using a re-configurable structure and bridging, as necessary, to add additional hardware
devices.

Bus Operations
Buses operate by requesting the use of paths to transfer data from a source element to a target element. In its
simplest form, a controller cache, defined as a source, transfers data to a disk, defined as a target. The
controller signals the bus processor that he needs the bus. When processing, control is given to the controller
cache operation. It owns the bus for that operation and no one else connected to the bus can use it. However,
as indicated earlier, the bus processor can interrupt the process for a higher priority process and give control
to that device.
This is called a bus arbitration scheme, as shown in Figure 7-3. The devices connected to the bus arbitrate for
bus control based on a predefined set of commands and priorities. The devices can be both source and target
for data transfers. This type of internal communication makes up the bus protocol and must be available to all
sources and targets-for example, connected devices that participate within the I/O bus.
Figure 7-3: Bus operations: arbitration for control
Parallel vs. Serial
Bus operations will perform data transfer operations using parallel or serial physical connections. These two
connections perform the same task, but have distinct architectures, each with their own pros and cons. It's
likely you'll be confronted with each as storage systems begin to integrate both for their strengths. Over and
above this option is the secondary choice of the type used within each category, such as point-to-point and
differential for parallel SCSI and various serial implementations. It's important to understand the limitations, as
well as the strengths, when applying these to particular application requirements.
Parallel connections for storage utilize multiple lines to transport data simultaneously. This requires a complex
set of wires making parallel cables quite large and thus subjecting them to an increased overhead of ERC
(error, recovery, and correction), length limitations, and sensitivity to external noise. Of the parallel
connections, there are two types: the single ended bus and the differential bus. The single ended bus is just
what it describes, a connection point where devices are chained together with a termination point at the end.
On the other hand, a differential bus is configured more like a circuit where each signal has a corresponding
signal. This allows for increased distances and decreasing sensitivity to external noise and interference
among the multiple wires that make up the cable. Of these two parallel connections, differential will be the
more effective for high-end storage systems; however, they are also more costly given the increases in
function and reliability.
Serial connections are much simpler, consisting basically of two wires. Although they pass data in serial
fashion, they can do this more efficiently with longer distances, and with increased reliability. High-speed
network communications transmission media such as ATM and fiber optics already use serial connections.
Wide buses that use parallel connections have the potential for cross-talk interference problems, resulting in
intermittent signaling problems, which ultimately show up in reliability and performance issues for storage
operation. The use of two wires allows substantially improved shielding. Although it may seem counter-
intuitive, serial connections can provide greater bandwidth than wide parallel connections. The simpler
characteristics of a serial connection's physical make-up allow a clock frequency to be increased 50 to 100
times, improving the throughput by a multiple of 200.

Differences in Bus and Network Architectures
Computer networks connect various devices of individual characteristics so they may communicate with one
another. Some networks are peer-oriented, where all the devices are deemed equal and vie for network
transport resources, transmission, and reception. Other networks (the majority of networks in use today) are
hierarchical, where some devices have control and authority over other devices regarding who and how they
can communicate, when they can transmit, and when and what they can receive. However, both architectures
all share similar functions in that they transmit and receive data according to some type of predefined
standard.
This demonstrates two distinct differences from bus architectures. First, networks have a defined transmit and
receive function. This allows networks to continually pass data into the network path and receive data without
regard for control and operation of the connection. This is different than a bus where strict flow control and
arbitration takes place with devices assuming control over the path. Second, networks enable data
transmissions over a long distance. Networks provide computers a method of transmitting and receiving data
over great distances without regard to the underlying communications infrastructure. This requires network
architectures to encapsulate the data to be sent over various types of media and transmission circuitry.
For the majority of networks today this is accomplished through a defined standard called TCP/IP. As we
mentioned previously, this is a layered approach to communication functions that address all the requirements
of transferring data from one device to another while making the communications transparent to the
computers application. Figure 7-4 shows the standard layers of TCP/IP. However, underlying the complexities
of these functions is the simple concept of encapsulation and transmission. TCP provides the functions for
taking the data from the application and OS and wrapping them in a form that will fit into the IP envelope. IP
then acts as the envelope by putting the TCP wrapper into the IP packet (nee envelope) and sending it into
the network with the appropriate addressing information.
Figure 7-4: The standard TCP/IP layers and functions
The reverse happens on the receiving end. The IP functions receive packets addressed to its host device and
pass these to TCP. TCP unwraps the data and presents it to the receiving application.
IP communications take place via multiple types of physical media. The media affects the integrity of the
communications, the speed of the transmission, and the distances supported. Regardless of the media, the
electronic standard that guides transmissions at the wire level is Ethernet. Working from packet encapsulation
architecture, data packets once they hit the network vie for transmission time, resources, and priority.
Obviously, the most prominent performance characteristic is the size of the packet. Packet size is determined
by the implementation of Ethernet standards. These standards, as they are implemented in vendor network
devices, such as NICs, switches, and routers, support this physical level of interface to the network.
Capacities differ from standard Ethernet connections to the multi-gigabit capacity for Gigabit Ethernet (Gbe).
Most application transactions that access the network (this includes the application data, overhead of TCP/IP,
and network error, recovery, and correction) will exceed the packet size, even for Gbe networks. So, moving
data from a server to a client or from a server to a backup device can utilize larger and larger quantities of
network bandwidth. Given this condition, nearly all remote application transactions will be accomplished
through multiple packet transmissions.
Similar in fashion to the bus, its bandwidth, addressing, and interrupt functions also drive the network.
However, major differences show up in interrupt functions (in other words, how do I get access to the network
and utilize its resources?). Additional differences show up in the capacities of bandwidth, given the diversity of
physical media that networks have to traverse. Physical addressing, an integral part of a networks topology,
also becomes diverse, driven by the particular implementation of the networking standard a vendors devices
are supporting (for example, fast Ethernet and gigabit Ethernet).
The PCI Bus
As mentioned previously, todays standard for connecting peripherals into the motherboard is the Peripheral
Component Interconnect (PCI). The operation is straightforward with the PCI bus. PCI determines, through the
use of the PCI controller, the destination of the data. The destination could be local or to an expansion slot. If
it is destined to an expansion slot address, the host adapter takes over and translates the protocol from PCI to
host adapter protocol. This could be IDE, SCSI, USB, or Firewire.
The importance of the PCI bus is that it decouples control of the data path to the PCI bus itself. Therefore, it
puts more performance responsibility on the PCI bus components, and relies on its bandwidth and speed
characteristics. Most PCI buses today have a 32-bit bandwidth with various clock speeds ranging from 33MHz
to 1GHz. The other importance of the PCI bus is its mezzanine architecture, whereby expansion of the bus
itself can take place, extending the scalability of the number of peripherals connected.
The SCSI Bus
The most popular storage bus in use today is the SCSI bus. The Small Computer System Interface (SCSI) is a
standard that allows the connection of various devices through a parallel interface. Although the standard calls
for support of multiple devices, general usage has been to externally connect disk and tape systems to the
server. Given the bandwidth, addressability, and speed, it has become the defacto standard for the external
connection of storage devices into open computing systems encompassing both UNIX and Windows.
The SCSI bus allows connections from eight to sixteen devices in a single bus. The operation is through
initiator and target architecture. Only two devices can use the bus at one time. The initiator gains control of the
bus and transfers the command to a device on the bus. The receiving device is the target and processes the
command. It then sends a response back to the initiator. SCSI devices connected to the bus are identified by
their SCSI ID, which also serves as its address. The ID identifies the devices priority for arbitration0 being
lowest and 7 or 15 being the highest.
It is possible for one server to have multiple host SCSI adapters with access to multiple SCSI buses. Although
these, by convention, will have different SCSI IDs, they are completely separate as they communicate through
the host adapter to the PCI bus. As previously discussed, the software drivers assign the SCSI IDs. However,
the SCSI controller provides the actual access to the physical device. The controller does this by assigning
each device a Logical Unit Number (LUN). This, in effect, virtualizes the devices that are connected behind the
controller.
Figure 7-5 depicts the most common SCSI configurations where servers with a single initiator support multiple
targets (such as controllers). In this configuration, controllers support multiple LUNs or disks attached. Figure
7-6 also depicts a configuration with multiple servers supporting multiple targets. Although each target or
controller is unique, multiple LUNs are addressed with similar LUN addresses. This works because the
controllers own and direct each of their LUNs.
Figure 7-5: SCSI configuration single host
Figure 7-6: SCSI configuration multiple hosts

Fibre Channel
It's a network it's a bus no, it's Fibre Channel. Fibre Channel is a layered connectivity standard that has
the characteristics of a bus, the flexibility of a network, and the scalability potential of MIMD (multi-instruction
multidata) configurations. In implementation, it uses a serial connectivity scheme that allows for the highest
level bandwidth of any connectivity solution available today. This architecture allows for implementations to
reach a burst rate as high as 200 MBps for I/O operations, with aggregate rates depending on workload and
network latency considerations. Regardless of the latency issues, this is a tremendous enhancement to
throughput over traditional bus connectivity.
NoteMulti-instruction multidata computer taxonomy (known as MIMD) occurs where multiple instructions
are executed with operations on multiple data sources. This has become synonymous with MPP
computer configurations where parallel processing at the CPU level and I/O levels exemplify this
theoretical term.
To say that Fibre Channel is a network architecture or that Fibre Channel is a serial bus connectivity standard
is an exercise for the esoteric debating society. Discussion here will result in nothing more than a last word at
the science club meeting. In all fairness to the innovations surrounding Fibre Channel (FC), it was designed
incorporating the characteristics of both to meet the growing scalability requirements of computer connections.
It has accomplished this task in particular with storage systems.
The FC standard is implemented using a layered protocol consisting of five levels. As depicted in Figure 7-7,
the first three levels deal with the physical attributes. As shown here, these are FC-0 through FC-2. These
layers control the lowest level operations, from media types supported, to signal encoding for transmission, to
the encapsulation and transmission control functions. FC-3 through FC-4 provide attributes for common
services and mapping to other protocols. The most popular so far being the SCSI and IP mappings that
support storage system operation and remote linking to IP networks.
Figure 7-7: Fibre Channel layered architecture
FC can be implemented through three different topologies, as indicated in Figures 7-8, 7-9, and 7-10. These
range from the most simple, point-to-point configurations, to the more complex, albeit more powerful, fabric
implementation. The Fibre Channel Arbitrated Loop (FC-AL) topologies have served as a transitional move as
FC storage systems moved from FC Hub configurations, where the FC-AL provided access to the bandwidth
benefits of the protocol; however, the loop arbitration overhead sacrificed most of these enhancements.
Figure 7-8: The FC point-to-point topology
Figure 7-9: The FC arbitrated loop topology
Figure 7-10: The FC fabric topology
Fibre Channel Operation
FC operates on a serial link design and uses a packet type of approach for encapsulation of the user data. FC
transmits and receives these data packets through the node participants within the fabric. Figure 7-11 shows a
simplified view of the FC fabric transmission from server node to storage node. The packets shipped by FC
are called frames and are made up of header information that include addresses, user data (encapsulating an
incredible 2,192 bytes per frame), and error recovery code (ERC) information.
Figure 7-11: FC operations
Each device that uses the FC Fabric is a node. FC provides both flow control and addressing for nodes when
they contact and log in to the fabric, as illustrated in Figure 7-11 where Node 01 is assigned the port address
of N01-fabric1. This is the N-Port address for the external device, in this case a Host Adapter for a server.
This is connected to Fabric port F-01 of the switch's fabric (in other words, the fabric switch with an address of
F01-fabric1 is shown as a fabric port). The storage system is connected using the same pattern.
The devices are controlled using calculations of available buffer space within each device as well as
knowledge of time sequences, thereby keeping frame overflows to a minimum. Finally, the fabric itself
executes under a predefined and supported class of service. Of these classes, only two are appropriate for
typical storage systems: Class 2 and Class 3. These define a network connectionless service this way:
acknowledgement of transmission is designated as Class 2; no notification of transmission is displayed as
Class 3.

Other I/O Buses
As peripheral devices increased in both number and diversity, the need to connect these to the computer
system quickly and efficiently also grew. With devices such as digital cameras, larger external and removable
storage devices, scanners, printers, and network options, the diversity and flexibility of the traditional computer
bus was expanded to include support for all these devices. The systems and storage industry responded with
innovative solutions that enhanced and implemented the traditional computer bus with serial technologies and
higher speed protocols such as Firewire. In this section, we discuss two of the most popular bus evolutions:
the Universal Serial Bus (USB) and Firewire. USB and Firewire have been successful solutions in the
workstation and PC industry. Other evolutions to the traditional bus architecture are evident in the opposite
side of the spectrum in the super computer and parallel processing industries. These technologies, driven by
complex computational problems, have resulted in creative architecture that bind multiple computers together
applying their combined power. These architectures are characterized by their creative bus implementation in
connecting hundreds of processors and peripheral devices together. We summarize some general
configurations in our discussion of these creative bus strategies.
USB and Firewire
Like the SCSI standard, the Universal Serial Bus (USB) and IEEE1394 standard, Firewire, was developed into
a product to serve the multiple and diverse types of external devices that can interface with a single PC or
workstation user. Used almost entirely for single user systems, the bus concept is similar to SCSI in terms of a
host adapter translating PCI communications to USB or Firewire adapters whose protocols communicate
within serial bus architectures. One of the differences is the support for the diverse and often disparate
number of peripheral devices requiring connection to single user systems. These devices range from mice to
joysticks to optical scanning equipment. With the exception of CDs, storage has come late to this area and
remains lacking in providing levels of functionality needed for server level implementation.
Both of these solutions are ways of expanding the I/O capability of computers; however, the use of USB or
Firewire disks for large-scale commercial use has not taken hold. Although this came about for many reasons,
it is primarily due to the direction the USB standard took to support asynchronous and isochronous data
transfers within a half duplex architecture. This limited performance for any bandwidth gains provided by a
serial connection. In addition, disk manufacturers have not supported the serial interface and command
structure used with disk drive functions. Finally, the limiting number of devices that can be addressed and the
inability to deal with advanced storage features such as controller/LUN functions and RAID partitioning will
continue to orient USB toward the single-user PC market.
Creative Connection Strategies
There are many ways of connecting peripherals to the server. Using the bus, network, and hybrid technologies
that we discussed earlier, these strategies/implementations have evolved from the exotic innovations of
supercomputing, alternatives to symmetrical multiprocessing (SMP), and developments within distributed
computing architectures. All of these provide some type of high-speed interconnect between computing nodes
that enable increased parallel computing tasks, allow support for larger workloads and increased data sizes,
and which give way to increased overall performance scalability. These interconnects can be characterized as
implementations of various bus and network technologies that produce non-standard configurations.
Generally, these implementations demonstrate characteristics of a network with sophisticated data distribution
functions that leverage specific CPU architectures (for example, IBM RISC-based systems, SUN Sparc
systems, and Intel CISC systems).
There are three general types of high-speed interconnect topologies that have been commercialized. These
are the shared nothing, shared I/O, and shared memory models. These models supplement the basic
components of computer systems (for instance, RAM, CPU, and internal Bus), and provide an extension to a
tightly coupled or loosely coupled set of distributed computer systems.
Shared Nothing
Figure 7-12 illustrates the configuration of a shared nothing system. These systems form the foundation for
Massively Parallel Processing (MPP) systems, where each computer node is connected to a high-speed
interconnect and communicates with nodes within the system to work together (or in parallel) on a workload.
These systems generally have nodes that specialize in particular functions, such as database query parsing
and preprocessing for input services. Other nodes share the search for data within a database by distributing it
among nodes that specialize in data access and ownership. The sophistication of these machines sometimes
outweighs their effectiveness, given that it requires a multi-image operating system (for instance, an OS on
each node), sophisticated database and storage functions to partition the data throughout the configuration,
and the speed, latency, and throughput of the interconnect. In these systems, both workload input processing
and data acquisition can be performed in parallel, providing significant throughput increases.
Figure 7-12: A shared nothing high-speed interconnect
Shared I/O
Figure 7-13 indicates the configuration of a system that has shared I/O. Although more common than the MPP
machines, the shared I/O requires the computer systems to share a common I/O bus and therefore extend the
capability to provide additional access to larger amounts of data. Operation of these machines requires that a
single image operating system control the operation of I/O and application processing across the computing
nodes and shared I/O. These systems offer enhanced access to large amounts of data where I/O content can
be large and multiple computing nodes must operate in parallel to process the workload.
Figure 7-13: A shared I/O high-speed interconnect
Shared Memory
Figure 7-14 shows the most common of alternative interconnects: the shared memory model. This makes up
most of the high-end SMP machines where multiple CPUs share a large RAM, allowing CPUs to process
workloads in parallel with enhanced performance from a common RAM address space thats both physical and
virtual. Also operating under a single image operating system, these systems enhance the capability to
process workloads that are high-end OLTP with small I/O content.
Figure 7-14: A shared memory high-speed interconnect

Without the ability to view, manipulate, and update data stored on the computer, the most sophisticated
storage infrastructure is meaningless. Our last storage fundamental covers the methods and types of
organizational models used to facilitate the productive use of stored data. Data organizational models and
methods are part of the entire storage infrastructure. Their proximity to other elements within the infrastructure
indicates that structures, such as file systems and databases, are the critical connectivity between application
and end user. Data is stored for many different reasons and accessed in as many ways to support a rapidly
changing business environment. The knowledge, awareness, and insight to the diversity of data organizational
models is as important and critical as that of storage systems and connectivity options.
Organizing Data
Data in computer systems is organized by creating databases. Databases are defined as a collection of
interrelated material stored in an environment that is both convenient and efficient to use in retrieving
information. This is done through a combination of file systems, database management systems (DBMS), and
applications. Combined with an effective storage system, the database can offer an environment where data
can not only be stored and updated on a reliable basis but also be accessed in multiple ways.
Underlying the DBMS is the file system that works as a component of the operating system. The file system
functions as the operating systems ability to organize and retrieve the physical data that exists on attached
storage media and within memory devices. File systems deal with the physical aspects of storage media, the
data locations, as they are stored within the media segments, and the status of the data within. They also
provide the first level of transparency to the user or owner of the data.
Database management systems (DBMS) provide methods for defining the data, enhanced access to data
generally through some type of enhanced indexing system, and functions for maintaining the data structures
as well as the data itself. Defining and accessing the data through a DBMS, users can additionally perform
calculations and logical functions on the stored data, and present the data in enhanced views (usually tabular
reports that include graphics and charts).
Business applications take advantage of the file system and DBMS functions by using their transparent data
retrieval and dealing directly with the data without having to understand or know the underlying operations of
the DBMS or a native file system. As we indicated in Chapter 1, most application designers and programmers
can use these high-level functions without regard to the underlying hardware and software infrastructure.
Others can be locked into specific DBMSs and file systems that have become standardized within their data
centers.

Interrelationships of Performance
File systems, database systems, and storage systems are interrelated and dictate levels of performance for
the storage infrastructure. File systems, for their part, form the basis for efficient and effective access to an
applications data. Problems occur, however, when the file system is doing its job with unoptimized disks,
insufficient disk space, or bandwidth constraints in the bus. As application performance becomes problematic,
the file system attributes should be considered for better alignment with the available storage infrastructure.
Many relational databases dont use an operating systems file system at all. Given thats the case in many
database instances, then what substitutes the functionality of the file system within the database? Actually,
many database systems interact closely with installed storage systems through the use of their own file system
and virtual I/O manager.
Finally, many applications rely on the interaction of files that are networked. This provides yet another
dimension to the interrelationship of the file system to storage and the challenges to the integrity and reliability
of the data stored within a network.
Virtualization Times Three
Data organizational methods begin the first slide into the confusion of virtualization. This is because File
Systems provide a layer of abstraction to the physical location of data, the location of disks, and the storage
media itself. Database systems further abstract the physical data, physical location, and the actual data.
Finally, the application using both organizational methods will further hide (for example, virtualize) the
complexities of finding and managing the data. In many cases, we are left with the question of where the
actual data is.
The operating system component that encompasses the allocation of space, data structures, and which keeps
track of these organizational storage criteria is the File System. File Systems operate within the I/O manager
component of the OS. In most operating systems, such as UNIX and Windows, the file system operates as a
driver or an additional service that will interact with the OSs I/O management component. This allows multiple
file systems to simultaneously operate on a single server.

Requirements for Enterprise-Class File Systems
Enterprise-class file systems need far more than convenience and efficiency to be considered adequate
support for enterprise OLTP, data warehouse, Internet, and complex batch operations. The following are some
of the global requirements that should be addressed by any enterprise-class file system and the issues they
raise:
Security Limiting user access to files and exercising control over sensitive physical resourcessuch as
volumes and system filescontinue to be the two major requirements for security within file systems. This
has become cumbersome as multiple and often duplicate security functions are offered by both file and
database systems. From the storage perspective, there exists a lack of effective and integrated resource
security and control functions that are available within the storage infrastructure components.
Data Recovery Disasters, both large and small, are going to occur. The ability to recover in a timely
fashion has long been a requirement, as databases become systems of record for both large and small
enterprises. This, however, becomes more sophisticated and complex as recovery functions for
databases differ from recovery functions for file systems which, in turn, differ from the recovery system of
an entire application.
Fault Resiliency/Tolerance This is the timely part of the recovery requirement. Most notably, the
requirement that says you have to be up for 24/7. Although a noble thought, the reality of keeping an
enterprise system available in this manner does not exist yet. Having said that, however, the functionality
to repair and recover from most hardware failures without disrupting the application can be accomplished
at various levels. The problem continues to be that enterprise application systems are likened to living
entities that consume resources, and many of these are just temporary and need to be refreshed from
time to time. Secondly, the ability to design and place card-holders (for lack of a better term) in the
processing time frames has failed to mature with the sophisticated hardware/recovery systems available
today.
Note24/7 systems have existed for some time in mainframe configurations, with full redundancy at
various levels. However, as data centers migrate their operations from the mainframe to open
systems using commodity-based platforms, additional levels of maturity must evolve before
open system hardware and software provide the same levels of desired fault tolerance.
Support for Advanced Storage Systems and Application Requirements Last, but certainly not least, is
the support for future enhancements to disk drives, disk formats, tape, and so on. Theres no question
that this is an important factor given that databases and file systems are closely tied to their storage
relatives and must keep up. The typical IT oversight in this requirement concerns the effects that support
of back level software (for instance, file systems and databases) has on hardware manufacturers.
Although it looks altruistic on the outside, inefficiencies develop when this continues for any length of time.

File Systems
From the most academic definition, the file system is a logical sequence of blocks on a storage media. Files
are a basic construct of the operating system. With the exception of certain data (refer to relational databases
further in this chapter), all data is kept in some type of file format. As such, it requires a system to maintain
and manage the relationship with storage on behalf of the operating system. These file system functions are
shown in Figure 8-1 and can be categorized in the following manner: allocation, management, and operations.
These functions are described as follows:
Allocation File systems provide the capability to organize I/O devices into functioning units of storage.
Management File systems provide the activities necessary to track, protect, and manipulate data stored
within the I/O devices.
Operations File systems locate logical sequences of data (for instance, a file) through various search
methods depending on data recoverability and the sophistication of the system.
Figure 8-1: The components of a file system
Allocation
The file system allocates the two most important elements of the storage media, the volume, which defines the
physical device, and its related attributes. Secondly, the files, which are collections of data, are accessible
through some type of naming convention to the operating system and, subsequently, the applications are
stored physically on the storage media. In todays environments, multiple products can make up these
components. Accessible as add-ons to the basic components of the operating system, these products support
two distinct but interrelated functions: file management and volume management. These products are
available as recoverable, journal, or high-performance file systems and as enhanced or extended volume
managers.
Volume Definition
Allocation of space within storage, especially in disks, begins with the initialization or formatting of the physical
device itself. This is done by creating a volume that corresponds to a logical partition of space on a disk.
Given the esoteric nature of volume initialization, volumes can be assigned to part of a disk or can span
physical disks.
The activity of formatting defines the smallest unit of addressable space within the device and the creation of
special files that will act as directories or catalogues for file allocation and access. In Microsoft operating
systems, these are referred to as clusters and are mapped to the physical sectors of the physical disk. In
UNIX, this is generally defined, given there are multiple UNIX variants, as a super block. Consequently, there
exists a relationship between the file system allocation and the actual physical attributes of the disk, such as
capacity and density. Both file system unitsclusters and blocks have an integral number of physical
sectors. Therefore, large physical disks storing small files can become very inefficient when the lowest unit of
storage is a cluster that utilizes multiple large sectors.
In addition, the overhead necessary to put the disk into activity (for instance, its formatting) explains the
difference between formatted capacities of a drive versus unformatted capacity.
File Allocation
Once a volume is initialized, it is ready to store a file or logical sets of blocks. Files are allocated according to
the attributes set with volume formatting. Given that, a file is allocated space through its smallest unit of
measurement, which is the cluster or block. Files are tracked through special files that have been created with
volume allocation. These master file tables (MFT), as they exist in Microsoft OSs (or inodes, as theyre
referred to in UNIX systems), contain data about the volume, space attributes, and files stored within the
volume.
Access to files starts here with the master file table or inode. Within these special files are indexing
information, security access lists, file attribute lists, and any extended attributes that the file may have. Once
the file has been created, these attributes are managed from these locations stored on the volume.
Management
The MFT, file allocation table, or super block-inode structures are increasingly being referred to as metadata.
Metadata simply means information about data. As we discussed previously, the operating system requires a
function to store and access data. One of these is the fundamental boot structure needed by the OS to
initialize itself during a power up or restart sequence. In addition to the metadata files allocation during
formatting, the process identifies and creates a boot file that becomes instrumental in locating the special
system information needed during the OS boot processes.
Another important allocation during this process is the creation of a log file, thereby providing a recoverable
file system, one that is required for enterprise-level operations. A recoverable file system ensures volume
consistency by using logging functions similar to transaction processing models. If the system crashes, the file
system restores the volume by executing a recovery procedure that utilized activity information stored within
the log file.
In file systems, all data stored on a volume is contained in a file. This includes the file allocation tables and
volume table of contents structures used to located and retrieve files, boot data, and the allocation state of the
entire volume. Storing everything in files allows the data to be easily located and maintained by the file system,
and a security attribute or descriptor can protect each separated file. If a particular part of the physical disk
becomes corrupt, the file system can relocate the metadata files to prevent the disk from becoming
inaccessible.
NoteThere is an important exception to the preceding statement that all data stored on a volume is
contained in a file. The exception is the allocation of disks using raw partitions where an application
will bypass the file system and manage the raw storage segments. This is common for many
relational database products.

Types of File Systems
Windows utilizes several types of file systems that support both client and server operations. Most client OSs
only need to support fundamental volume and file allocation functions. This is handled through its ability to
initialize and partition volumes with formatting and file allocation through a simple file allocation table.
However, within server products, a more robust file system is required, as shown in Figure 8-2. Windows
server products provide a NT File System (NTFS), which provides a robust file system that supports enhanced
volume and file allocation functions. These products include volume fault resiliency features for partitioning
volumes with redundancy features, thus providing a subset of RAID features through software. Also supported
is recoverability through logging functions. Enhanced security attributes are provided to support multiuser
environments with levels of data protection for both user access and file and volume protection.
Figure 8-2: An enterprise-class file system
UNIX has several variants; however, most offer a POSIX-compliant file system that is a bare-bones
component of the operating system. Enhanced and recoverable file systems are available through third-party
products. These add-ons provide both enhanced performance and journaling. In UNIX, recoverable file
systems are referred to as journal file systems, e.g., logging being synonymous with journaling. More
importantly, the UNIX markets have the availability and legacy of network file systems, where files are shared
across a network through a common file system and access protocol.
NoteThe POSIX Standard, also known as the IEEE 1003.x POSIX standard, is the result of working
groups under the auspices of IEEE. It describes a set of standard operating system interfaces.
Access to information via the Internet and through World Wide Web interfaces has produced yet another
variant of the file system, the Hyper Text Transport Protocol, or HTTP. Although more of a common file access
protocol, it does define a file structure within the web-based environments depicted in Figure 8-3. Because it is
supported across both UNIX and Windows, it provides a common protocol for accessing files through a
network.
Figure 8-3: Hypertext Transport Protocol (HTTP)
Another advancement in common file protocols is the Common Internet File System (CIFS), as shown in
Figure 8-4. Starting as a Microsoft initiative, it has now become an open standard. Based on the need to
provide file sharing among a diverse set of operating system environments within a network setting, it is a
model derivative of NFS that allows Microsoft Operating Systems to access shared files on Microsoft servers,
UNIX servers, the Web, and most recently, Apple servers. CIFSs are based on a Microsoft networking
protocol of Server Message Block (SMB), and have both advantages and limitations compared to NFS.
Figure 8-4: The Common Internet File System (CIFS)
Other file systems are generally derivatives of UNIX variants and are used for specialized applications, such
as scientific and academic/research applications. However, one alternative system (as shown in Figure 8-5)
has produced a new type of access, especially within networked environments. This is the Direct Access File
System, or DAFS. DAFS can be especially important given its use of Direct Memory Access (DMA) functions
for transferring data. Future DAFS products will have extremely fast access times given their capability to
transfer data between memory locations.
Figure 8-5: Direct Access File System (DAFS)
The importance of file systems as they relate to storage systems and infrastructures is straightforward and
should not be overlooked. As you encounter more detailed discussions in Parts III and IV regarding NAS and
SAN, we will discover another dimension to the importance of file systems and their capability to operate within
storage networked environments. As you may have already discovered, NFS, CIFs, and HTTP all play a big
role in these environments.

Databases
A database system is made up of individual software components that interact with each other and with other
system components, enabling applications access to various abstractions of physical data. Some of the
external components include the functionality of file systems, storage cache, and resource manager functions.
Other database system components work as an interrelated system of functions to process data requests on
behalf of applications. The databases sole purpose is to provide consistent, efficient, and reliable data to end
users.
I suppose its politically correct to substitute information for data here. However, end users today
manipulate more data than the average application programmer did ten years ago. The point is, I dont think
end users are intimidated any more regarding the handling, manipulation, and programmed access to
computer data. The sophistication of computer usage among smart users, the available tools to extract,
manipulate, and process data, have placed many users into the category of Super-User, Query Monster, or
Departmental Programmer. The innovation and continued evolution of the database system has almost
singularly provided the impetus for these changes. Fueled by the standard query language, SQL, and its
derivatives, where programming has been taken to more accessible levels forall computer users to understand
and manipulate on their own, the world has seen an explosion in database implementation, population, and
usage.
Looking for Data
A major purpose of a database system is to provide many users with an abstract view of the data. Depending
on who you are, you need different levels of abstraction when accessing existing data. IT professionals
obviously require different types of access than end users who assume enterprise OLTP and batch
applications will provide them with the right information (sorry, I mean data) at the right time. Underlying both
IT professional and end-user activities, designing applications, data models, and required maintenance
activities is a diversity of data access procedures. The various levels of access and abstraction are described
here:
Physical Abstraction Used primarily by system database administrators (DBAs) and system
administrators (storage administrators are slowly coming on board with these concepts), physical access
is the lowest level of data abstraction. This allows complex low-level data types and structures to be
described. It also provides a means to describe how the actual data is stored.
Conceptual Abstraction Used by programmers, application designers, and application DBAs, this is a
level of abstraction up from the physical level that describes what user data is actually stored in the
database. This also describes relationships that exist within and among the data elements. The
conceptual level allows IT professionals to describe the entire database in terms of simple structures or
data models. The physical level uses data models, structures, and descriptions to invoke complex
physical structures to develop the database.
View Abstraction Driven by those who derive value from its usage (usually the end user), views are the
primary level of abstraction for development of graphical user interfaces (GUIs). End users are not
concerned about the entire set of information within the database, or the complexities of its internal
structures. As such, view levels become the highest level of abstraction of the data and describe only one
part of a database, simplifying the interaction of end users with the database. Depending on the levels of
abstraction required, most users require multiple views of the data.
From a storage point of view, it is important to understand not only the physical abstractions of the data, but
also the database components that interact with native operating system functions and storage drivers. This
becomes increasingly important as database systems push the limits of storage infrastructures unlike any
other system or application. However, it is also important to study the translations of abstractions to
understand storage-related problems as System and Application DBAs, Administrators, and end users
encounter them.
The following are the typical components of a database system:
Database Manager Functions that provide the interface between the physical data and the application
programs, integrated application functions such as search and mathematical algorithms, and physical
access to the data.
File Manager Functions that allocate space on storage systems and manage the data structures used to
represent the conceptual abstraction of data. This component, in many cases, is integrated into the
database manager, bypassing the operating systems file system for these functions.
Query Processor A component that translates statements from a query language, such as standard SQL,
into low-level instructions that the database manager components will execute on behalf of the end-user
query or application program.
Preprocessor Compiler A component that translates database language queries, such as SQL that are
embedded in application programs. The compiler turns these into instructions for the query processor,
which, in turn, processes it and passes it on to the database manager.
Data Language Compiler Functions that convert data language definitions into data structures called
metadata . The Metadata table entries are then stored in a metadata database, sometimes called a data
dictionary or repository.
Additional data structures are necessary as part of the physical implementation of the database system. These
are the data dictionary or data repositories, as we just discussed. Finally, index structures provide enhanced
search access to data when searching for particular values or data files.
Databases change as data is added and deleted. The collection of data in the database at any given time is
called an instance, while the overall design of a database is called the database scheme or schema. The latter
is expressed as a conceptual abstraction, and is specified through a data definition language. The results of a
data definition schema are placed into a data dictionary or metadata table. A data manipulation language is a
set of procedures to access and manipulate the data such as the standard query language or SQL.
Database Systems: Different Animals
There are only a few popular database models that have survived the mainframe-to-client/ server-to-Internet
evolution. Interestingly, some of the older models continue to process workloads today, albeit mostly on
mainframe processing complexes. Databases can be developed from a simple collection of sequential files,
sometimes called flat files because of their non-dimensional access restrictions, to the most complex
mainframe hierarchical databases. Regardless of implementation, major database management systems have
evolved from three conceptual models.
These are the networked databases based on a network data model. Characterized by its capability to provide
links within its fields of database records. The hierarchical data model is used by many file systems in their
functions to provide faster access to directories and file indices. These are recognizable through their use of
tree structures, such as the B-Tree and B+Tree data structures. The third and most popular model is the
relational model, which forms the structure for all popular relational databases systems, such as Oracle,
Sybase, IBMs DB2, Microsofts SQL/Server and others. The database model that provides the most
challenges, problems, and headaches to storage infrastructures today is the same one that provides the most
challenges, problems, and headaches to storage networking infrastructures: the relational database.
Relational database systems consist of a collection of tables, each of which has a unique name. Each field in
the table is considered a distinct entity with attributesand has relationships to other entities within its own table
as well as to others in the collection of tables that make up the database. User data is populated throughout
the tables by individual records. This forms the basis for the tremendous power of a relational database, the
ability to compare data elements to perform set mathematical functions, or What if questions. From an
overall systems perspective, this provides two challenges:
First is the compute-intensive nature of relational processing. Because relational processing utilizes set
mathematics to find the answer to a database query, the data necessary to construct the sets and
processing to compute the answer is one of the most resource-intensive activities within data processing.
Second is the I/O-intensive nature of relational processing. I/O content and activity becomes intensive as
the rows of data from tables are loaded into memory to build set constructs to compare data elements
until all relational operations are complete. With a database of any size, the compute requirements
coupled with the I/O requirements render relational database processing one of the most resource-
intensive activities for commercial data processing.
Consequently, the relationship with storage for databases becomes very special. Although there are several
considerations when maintaining a storage infrastructure for RDBMS, here are some particularly important
points to consider as we move into the area of storage networking.
Relational databases have a legacy from UNIX and mainframe environments. This has resulted in a design
whereby most relational database systems (RDBMS) contain their own I/O processing and file management
functions. This means that when installed, they bypass the operating systems file system and utilize their own.
This is referred to as an RDBMS that uses a raw disk partition. In effect, they are performing their own I/O and
substituting their own file system, essentially mapping the raw disk to their own to enable low-level access.
Historically, they had to do this to provide acceptable levels of performance and reliability in the days when
UNIX did not have an optimized file system, much less a journaling file system. Given the resource-intensive
nature of these relational operations, early RDBMS vendors developed their own I/O systems.
Another important factor to keep in mind is the nature of the relational data structure. It is a table, not a file, a
table. The physical data is stored in native block formats for those RDBMSs that use their own I/O and file
system, or for those whose physical data is masked from the system when using a standard file system. This
means that without special utilities and knowledge of the database, its state, and its metadata, that the ability
to reconstruct the data structure from a disruption is impossible. Therefore, databases are difficult, at best, to
maintain from a data maintenance aspect (for instance, backup/recovery operations, archival, and volume
management).
Databases, especially the relational model, are a challenge to maintain and manage. They have a close
relationship with storage systems given their own processing characteristics and development legacy in the
area of I/O technologies. Storage networking adds new dimensions to the existing challenges in data
organization models and methods. We will explore the role of databases in the world of NAS and SAN in the
upcoming sections.

Part III: Network Attached Storage
Chapter List:
Chapter 9 : Putting Storage on the Network: A Detailed Discussion
Chapter 10: NAS Hardware Devices
Chapter 11: NAS Software Components
Chapter 12: NAS Connectivity Options

Chapter 9: Putting Storage on the Network: A Detailed
Discussion
Network Attached Storage (NAS) is the evolution of network servers. It was developed from efforts to
configure, optimize, and operate a server directed only toward one thing: providing shared storage for users
on a network. NAS is one of two pillars that make up storage networking infrastructures in todays data
centers. It has evolved into a class of storage devices that address storage needs from departmental storage
appliances to enterprise-class storage servers managing multiterabytes of data. This section will explain,
discuss, and evaluate the NAS architecture, from the hardware and software required, to the necessary
connectivity components. Developed using concepts from Part II, and expanding the NAS overview from Part
I, this information will encompass a detailed view of the NAS internal structures.
The NAS Architecture
NAS is a specialized computer that provides file access and storage for a client/server network. Because it is
a specialized solution, its major components are proprietary and optimized for one activity: shared file I/O
within a network. NAS is considered a bundled solution consisting of prepackaged hardware and installed
software regardless of vendor selection. The bundled solution encompasses a server part, a computer system
with RAM, a CPU, buses, network and storage adapters, storage controllers, and disk storage. The software
portion contains the operating system, file system, and network drivers. The last element is the network
segment, which consists of the network interfaces and related connectivity components.
Plug and Play is an important characteristic of the NAS packaged solution. Given the bundling and
configuration that comes with the product (which is dependent on the class of NAS solution you are
implementing), its out of the box and attached to the network in less than five minutes. Although this may be a
fairly accurate portrayal for installation and identification to the network, the Plug and Play value must be
considered in context with the size and scope of the NAS device. There are many implementation caveats to
consider and ample thought should be given to existing server clusters and storage systems already in place.
However, for most small networks, providing a Plug and Play NAS solution is a productive solution. Though
more sophisticated departmental solutions will take longer, the bundled characteristics of the NAS devices
save considerable time in planning and implementation. Of course, enterprise-level solutions should have
much more thought put into their planning and implementation. Obviously, with this class of storage, care
should be taken not to fall prey to the five-minute Plug and Play temptation which could result in much larger
infrastructure problems.
NAS can be a productive solution. Its simplicity is one of its greatest strengths. The entire perspective of the
NAS phenomenon, however, should be considered prior to any implementation. The following points
summarize the issues that are important to know and understand to facilitate a long-term productive
configuration:
NAS is a specialized server solution with proprietary hardware and software.
NAS is optimized for file I/O; support of any other types of workload I/O will prove problematic.
NAS is one of he easiest ways to accommodate increases to storage capacity for LAN-based shared files.
NAS can be one of the most difficult to administer and manage when numbers of boxes exceed
reasonable complexities.
NAS uses existing network resources by attaching to Ethernet-based TCP/IP network topologies.
Although NAS uses existing network resources, care should be taken to design expansions to networks
so increases in NAS storage traffic do not impact end-user transactional traffic.
NAS uses a Real Time Operating System (RTOS) that is optimized through proprietary enhancements to
the kernel. Unfortunately, you cant get to it, see it, or change it. Its a closed box.

The NAS Hardware Architecture
NAS is a specialized server solution with proprietary hardware and software.
NAS products provide a black box style of computer externally configured as a server with extensible internal
components configurable to the class of storage requirements you need. The black box computer style is
available in both RISC and CISC or Intel processing chip sets. The computer systems hardware is proprietary
to the manufacturer in terms of the internal component standardization, assembly, and the lack of
interchangeable components.
The Server Part
Figure 9-1 is a generic layout of NAS hardware components. As you can see, they are just like any other
server system. They have CPUs built on motherboards that are attached to bus systems. RAM and system
cache, meanwhile, are the same with preconfigured capacities, given the file I/O specialization. What we dont
see in the underlying system functionality are the levels of storage optimization that have evolved. NAS
internal hardware configurations are optimized for I/O, and in particular file I/O. The results of this optimization
turns the entire system into one large I/O manager for processing I/O requests on a network.
Figure 9-1: The NAS hardware components
Given the black box orientation of the internal server components means you dont have to worry about further
optimization and tuning for file I/O processing for the NAS solution. However, time spent on evaluating the
hardware components of the vendors solution is important, because once you install it, further changes will be
only up to the manufacturers extensibility of the model. Additionally, all file I/O workloads are not alike, as
much as we would like to think so (see Part VI). Given the various and diverse products that exist, the
evaluation of the speeds and capacities of the NAS hardware configuration are critical factors contributing to
the effective storage and application performance.
Figure 9-1 shows the components that are optimized for file I/O, but does not speak to how they are optimized.
This is done in the combination of the internal hardware and the proprietary operating system software as
illustrated and discussed later in the software part of this chapter. However, it is important to note that
extensive efforts have gone into the design and layout of the NAS internal hardware that provides the
optimization for file I/O. It is unlikely we can find out how this is done unless we work for the particular NAS
vendor. Needless to say, this work is proprietary to each vendor and is a significant part of the value
proposition of their products.
The Storage Part
NAS is a storage solution, and as such, perhaps its most important aspect is the storage infrastructure it
provides. NAS supports multiple storage types, capacities, and configurations, including select levels of RAID
at the top end (for example, enterprise class NAS devices). However, support for RAID levels 1 (mirroring) and
5 (striping with parity) is beginning to become available already preconfigured in low-end NAS appliance
devices. Support for incremental storage capacity upgrades depends on the vendor and flexibility of their
models for upgrading.
Figure 9-1 is a generic representation of the storage part of the NAS configuration. Note that the proprietary
nature of the hardware continues through to the storage connectivity and storage devices. Again, this part of
the value proposition of the NAS bundled solution is optimized for I/O operations. There is an important design
distinction to point out. Although largely assumed given the traditional I/O operations of the NAS device as a
server, it becomes lost in the focus on file I/O operations. The NAS device processes file I/O requests from
clients within a network. As with any computer, these are processed into read/write requests through the OS
and file system. Ultimately, this results in low-level disk operations performed at a block level, which means
the NAS device still performs normal block level I/O operations within its internal processing configuration.
The sophistication of the storage system itself, and how it performs block I/O processing, depends on the type
and class of NAS system. However, the storage infrastructure (the storage system, temporary locations,
connectivity, and so on) is tightly coupled with the server part as well as the reconfiguration and capacity
upgrades that depend on the flexibility of the vendor.
NAS devices are managed differently both internally and from an administrator perspective. The formatting of
NAS disk devices is performed under the control of the proprietary operating system and, more importantly, its
file system component. Consideration for cluster or block size may be out of the hands of the administrator,
and thus limits the flexibility to the types of files the storage system will handle. Monitoring device activity is
dependent on the tools provided by the vendor. Although additional third-party tools support limited levels of
NAS activities, tools for the internal performance of system, storage, and network parts are lacking.
The Network Part
NAS supports Ethernet-based network connections, with all released standards supported (for example, 100-
baseT through Gigabit Ethernet). Connectivity to the server bus is through standard NIC-type connections,
however the bus connectivity from the adapter to the system bus is proprietary to the vendor. Most will find
PCI connectivity standard even though processing chip sets vary from RISC-based systems to CISC-based
Intel systems.
The NIC network components are available from single adapter configurations to extensible configurations that
support multiple NIC connections such as those depicted in our generic NAS component diagram, shown in
Figure 9-1. The adapters will support standard TCP/IP stacks with most connections being offered by NAS
vendors through OEM products from various suppliers. Because of this situation, network performance can
vary given the capacities and functionality of the NIC.
Departmental solutions are offered with only single NIC adapters, while mid-range NAS solutions offer multiple
connections. However, even network connections for extensible mid-range solutions generally dont exceed a
maximum of four NIC adapters. Enterprise NAS solutions will be extensible according to bus structures that
are available from the vendor. This NIC connectivity and performance can be a limiting factor in deploying
NAS in high growth areas. However, this is generally addressed by adding multiple NAS solutions. (More
about these considerations is discussed in Parts V and VI.)

The NAS Software Architecture
NAS is optimized for file I/O; support of any other types of workload I/O is problematic.
NAS software consists of three major components: the micro-kernel OS part, a proprietary file system, and the
network software drivers. These are depicted in Figure 9-2.
Figure 9-2: The NAS software components
The Micro-Kernel OS Part
The operating system for NAS is a UNIX kernel derivative, as indicated in the server discussion. The system is
developed with only OS components that enable effective file I/O operations. Components such as memory
management, resource management, and process management are all optimized to service file I/O requests.
This implementation as a black box architecture renders extreme limitations to initiating other processes, jobs,
or activities through the OS. Besides the management tools that the NAS vendor or third-party independent
software vendor provides, other processing activities within the server are not available.
Process management is programmed to handle the file I/O processes as the highest priority. Interrupt
processing queues and handles other requests secondary to the I/O manager. Various scheduling methods
and algorithms are used to optimize these tasks, such as first-in, first-out and random selection by queue. In
this manner, all file I/O processes are completed prior to starting any new overhead tasks that the system may
require. Balanced among these is the need to manage network processes, as well as handling the overhead
of the TCP stack. Given its reentrant functions, several network requests should be processing at any given
time. This becomes a balancing act for the OS to handle sufficient network requests with I/O manager
activities in order to complete block I/O operation to the storage.
Memory management will be optimized to handle as many file requests as possiblethat means sufficient
buffer space and cache management to facilitate the process management schedule. Capacity will play an
important role in the capability of the OS to handle interactive requests. Most will be set for read-only
operations to cache as many requests on the up-side (network input), and read-ahead buffer on the down-side
(storage reads). Interchange will take place as the network process is completed in encapsulating the read
request and passing control to the IP process and transmission. As we can see, a significant amount of time is
used with TCP and IP process management.
Given the focused job the OS has to do, resource management is fairly static. I/O operations take over all the
resources necessary to complete its tasks. Therefore, resource availability takes on a greater level of
importance. In addition, the level of controller processing not only plays a significant role in completing the
operations but is also fundamental in making resources available to the greatest number of processes. The
operations of disk controllers and the further enhancements to controller processing, such as RAID levels and
cache management, play significant roles in completing the I/O tasks.
NAS File Systems
The standard file systems included with NAS devices will support one of two types: the Common Internet File
System (CIFs), the Network File System (NFS), or both. In many cases (as shown in Figure 9-3), the standard
network file system interfaces with the NAS proprietary file system. Most NAS devices need to do this to
manage their own storage. According to the vendors, the implementation of other types of connectivity
protocols is supported inherently. As an example, the heterogeneous use of NAS devices to support
multiclient access is evolving beyond CIFs through protocols such as SAMBA. Other proprietary network
protocols such as Direct Access File System (DAFS) are beginning to emerge as well (see Part II).
Figure 9-3: NAS file system configurations
NoteSamba is a freeware program that provides end users access to commonly shared resources, and is
used primarily to share files and devices between Windows and UNIX systems.

Network Connectivity
The third of the three major NAS components supports standard TCP/IP communications within a diversity of
transport media. The most popular topology remains Ethernet. Support here is evolving with increasing
bandwidth in light of Gigabit Ethernet standards, sophisticated participation within enterprise networks with
quality of server (Qos) and routing participation, and advancements in coexistence with other storage models,
such as FC SANs.
A limiting factor of file I/O processing with NAS is the TCP/IP network environment and the processing of TCP
layers. As we discussed, the communication between client applications and server I/O is processed within the
encapsulation layers. Within the NAS architecture, these layers (which are normally processed within the
server) are now processed within the NAS box. To address these challenges, the TCP/IP overhead should be
given great thought when implementing a NAS solution. However, TCP/IP performance issues are difficult to
address, given the black box nature of the NAS software and the OEM supplier relationship of the NIC
components.
An advancement that may help this condition is the development of TCP off-load engines (TOE). These are
special NIC cards that off-load much of the TCP encapsulation processing onto the NIC card itself. Although
this solution is just emerging, it reflects evolving technology directions that optimize latency within IP networks.
NoteTCP off-loads derive their value by processing TCP layers within the NIC card. Anytime software
instructions are downloaded into lower-level machine instructions, it becomes highly optimized,
operating at machine speeds. However, this is done at the sacrifice of software flexibility, given its
much harder to change micro-code or ASIC instructions than software thats operating under a high-
level language within the OS.

NAS as a Storage System
Putting all the parts together allows a highly optimized storage device to be placed on the network with a
minimum of network disruption, not to mention significantly reduced server OS latency and costs, where
configuration flexibility and management are kept to a minimum. The NAS solutions are available for a
diversity of configurations and workloads. They range from departmental solutions where NAS devices can be
deployed quickly and in departmental environment settings to mid-range and enterprise-class products that
are generally deployed in data center settings.
The Departmental NAS Architecture
NAS is most often used to support the high growth of file servers that populate Windows client-based
networks. An example is shown in Figure 9-4. Its an almost perfect solution for consolidating unoptimized file
servers (for instance, Windows NT or Windows 2000 servers that operate only as network file servers for
clients on the network).
Figure 9-4: The departmental NAS configuration
The Internet NAS Architecture
An optimized solution is a more scalable and cost-effective answer as multiple servers can be consolidated
into a single NAS device. NAS is also heavily used and is a good solution for interoperable access to files
within web environments, as illustrated in Figure 9-5. NAS is the perfect storage solution for Internet
connections to web-based files accessed through HTTP.
Figure 9-5: The Internet NAS architecture
The Enterprise NAS Architecture
At the high end are solutions that support large capacities of data, usually having reduced user transaction
rates yet high I/O content. The storage and usage of unstructured data characterizes these workloads,
although other structured data exhibiting these same characteristics can be NAS candidates. Figure 9-6
shows a specialized application being supported though high-end NAS devices.
Figure 9-6: Enterprise NAS configurations

Overview
NAS devices form an effective Network I/O manager for remote applications. From the perspective of the NAS
hardware, functioning as an I/O manager, careful consideration should be given to the components and their
corresponding characteristics that provide this complex functionality. This chapter will discuss NAS hardware
components and their characteristics along with the issues that will confront these elements. Within these
discussions will be definitions of key hardware components of the NAS device, configuration and capacities,
and typical NAS I/O workloads as they relate to NAS hardware.
NAS hardware working in conjunction with related components function as an I/O manager. The NAS
hardware components described in this chapter are the processor elements, RAM and related buffer
management, and, of course, the storage system elements. We cannot overlook the communications and
input-processing components largely made up of NIC interface elements and related bus connectivity options.
These elements will be described and discussed in Chapter 12, which deals with NAS connectivity options, but
it is important to recognize them in relation to the NAS hardware.
Closely related to NAS hardware are the effects of NAS processing models. Three models of NAS processing
exist which drive NAS configurations and related component capacities:
Simple extensions of a clients file system, as a shared storage resource, make up simple file processing
(Sfp).
Remote file acquisition with added reliability from a heterogeneous set of clients constitute quality file
processing (Qfp).
Enterprise-level processing of files using NAS within a transactional type of system with heterogeneous
access and multitier applications form a complex file processing (Cfp) configuration.
With the NAS hardware functioning as an I/O manager, the performance of its components will be
characterized by the workloads it processes. NAS workloads are made up of file-oriented processing activities
performed on behalf of various applications. These workloads consist of I/O operations with disparate I/O
content. Client network files transferring office application files offer a dramatically different I/O content
contrasted against the unstructured data such as image, video, and audio. On the other hand, web-based and
other specialized application file structures exhibit everything from variable length files to compressed and
encryption-encapsulated, structures.

NAS Hardware as the I/O Manager
As we discussed in the previous chapter, the NAS devices perform a singular directive with a duality of
function, network file I/O processing and file storage. A more concise term would be that it functions as an I/O
manager. So, what is an I/O manager?
The I/O manager is a major component of computers and is made up from a complex set of hardware and
software elements. Their combined operations enable the necessary functions to access, store, and manage
data. A discussion of many of these components can be found in Part II. A comprehensive view of the
internals of I/O management is beyond the scope of this book, however an overview of a theoretical I/O
manager will help you understand the hardware operations of NAS.
Figure 10-1 shows the typical components that make up a theoretical I/O manager.
Figure 10-1: The I/O manager components
The I/O subsystem is made up of hardware, the processor instruction sets, low-level micro-code, and software
subroutines that communicate and are called by user applications. The I/O manager is, in effect, the
supervisor of all I/O within the system. Some of the important aspects of the I/O manager are the following:
Hardware Interrupt Handler Routines that handle priority interrupts for hardware devices and traps
associated with error and correction operations.
System and User Time A system clock that provides timing for all system operations. I/O operations
called by the user and executed in kernel mode take place in system time and are performed by the clock
handler routines.
I/O Device Addressing A system whereby hardware devices are addressed by a major and minor
numbering scheme. Major device numbers are indexed to a device table. The minor device number
identifies the device unit number. For disks, the minor number refers to the device number and the
partition number. For tapes, the major number refers to the controller slot number, and the minor number
refers to the drive number.
I/O Buffer Pool Retaining data blocks in memory and writing them back only if they are modified
optimizes disk access. Certain data blocks are used very frequently and can be optimized through
buffering in RAM.
Device I/O Drivers Block I/O devices such as disk and tape units require instructions on how they must
work and operate within the processing environment. Called device drivers, they consist of two parts. The
first part provides kernel service to system calls, while the second part, the device interrupt handler, calls
the device priority routines, manages and dispatches device errors, and resets devices and the status of a
device.
Read/Write Routines System calls that derive the appropriate device driver routine, which actually
performs the read or write operation to the device.
I/O Priority Interface System routine that handles requests for the transfer of data blocks. This routine
manages the location of the block, making sure it is handled properly if in the buffer, or initiates an
operation to locate disk address and block location within a device.
Raw I/O on Block Devices A system call and method used for very fast access to block devices,
effectively bypassing the I/O buffer pool. These types of low-level operations of the I/O manager are very
important in order to better understand the architecture and usage of database implementations using I/O
manager functions.
NAS Hardware I/O Manager Components
If we view the NAS hardware functioning as an I/O manager, the critical components will be the processor
element, the memory/cache elements, and the storage element. Certainly, the connectivity to the outside world
is important and as such will be handled through its own chapter (Chapter 12 in this book). However, the
internal hardware aspects of the NAS devices determine how fast it can process I/O transactions, how much
data it can store, the types of workloads it can handle, and how it can recover from failing components.
The NAS Processor Element
The NAS processor component is critical to the power required to handle specific types of workloads. Although
most vendors have moved to a CISC model (for example, using Intel chips to keep costs down), many of the
leading vendors offer RISC-based processor components. The difference, although seemingly slight, can be
substantial when evaluating the performance required for complex workloads.
RISC (Reduced Instruction Set Computing) versus CISC (Complex Instruction Set Computing) continues to be
an on-going discussion within chip designers and system vendors. The differences are based upon different
architectural designs in how the CPU component processes its instructions. The differences are depicted in
Figure 10-2, which shows that RISC uses simple software primitives, or instructions, which are stored in
memory as regular programs.
Figure 10-2: RISC versus CISC
CISC provides the instruction sets using micro-code in the processor ROM (read-only memory). The RISC
processors have fewer and simpler instructions and consequently require less processing logic to
interpretthat is, execute. This results in a high execution rate for RISC-based processors. Consequently,
more instructions will produce a gross higher processing rate with everything being balanced at a higher
throughput. Another important characteristic of RISC-based processor architecture is that they have a large
number of registers compared to CISC processors. More instructions that are closer to the ALU unit means
faster gross processing and processor throughput.
Given the architectural differences, the capability to provide high system performance, with repetitive
instructions, against a focused workload puts the RISC-based system on top. However, this is not achieved
without cost. RISC-based systems are more costly than CISC-based systems, specifically Intel commodity
processors. Therefore, a balance must be achieved between the performance sophistication of the NAS
device versus the processing complexities of the I/O workload.
This is important as decisions made on cost alone result in non-scalable CISC-based NAS devices, or devices
that may have a short-term usefulness in supporting a complex I/O workload. On the other hand, making a
purchase based on performance alone may lead to an overallocation of processor performance which could
be handled just as effectively with CISC-based systems performing in multiples (for example, deploying many
NAS devices).
This is also important because it defines the type of Micro-kernel operating system the vendor supplies. All
RISC-based systems are UNIX-based micro-kernel systems, while the CISC-based systems are a combination
of both UNIX and Windows micro-kernel systems.
When evaluating NAS devices, you should at least know the type of processor architecture thats being used.
Looking at technical specification sheets beyond the capacities of the disks and storage RAID capabilities, its
important to understand the scalability of instruction set computing for the workload being supported. You will
find that all CISC-based processing will use Intel-based processors with the most recognizable RISC-based
processors being offered through Sun, HP, and IBM.
The NAS Memory Element
RAM is an important factor in processing I/O requests in conjunction with available cache and cache
management. Caching has become a standard enhancement to I/O processing. Cache management is a
highly complex technology and subject to workload and specific vendor implementations. However, most
caching solutions will be external to the processor and utilize cache functionality within the storage array
controllers. As NAS solutions are bundled with storage arrays, any caching will be integrated into the NAS
device using both RAM and storage controller cache. Caching solutions within the NAS hardware will remain
vendor specific and available through the configurations. Note that some vendors have included a NVRAM
(Non-Volatile Random Access Memory) solution that will provide a level of performance enhancement but a
significant safeguard in data integrity to writes that occur within the workload.
The functions of the I/O hardware manager enable buffer management, which is a key element in the effective
handling of I/O requests. As requests are passed from the TCP stack to the application, in this case the
software components in the I/O manager, they are queued for processing through buffer management. Given
the slower performance of the disk elements compared to RAM, the RAM I/O buffers provide an important
element to quick and orderly processing of requests that may come in faster than the disk I/O can handle.
Therefore, the more buffer capacity you have, the more requests can be processed in a timely fashion.
For data operations inside the processor element, the functions of system cache can also provide an
additional level of efficiency. Our discussion regarding RISC versus CISC showed the efficiencies of having
software instructions loaded as close to the CPU as possible. Having a combination of RISC-based
instructions and user data blocks can provide additional levels of efficiency and speed. This can also prove
effective in CISC architectures given the cache usage for user data blocks and the redundant nature of the I/O
operations performed by the processor element within the NAS devices.
A popular solution to the esoteric I/O bottleneck problem is to add memory. Although this is an overly simple
solution to a complex problem, in the case of NAS simple file processing (Sfp), it can generally be effective.
The caveats (refer to Chapter 17) are many and lie within the read/write nature of the workload. Given that
many NAS implementations are for read-only high-volume I/O transactions, the RAM element is an important
part of the solution for increasing capacities for I/O buffer management.
The other effect RAM has on NAS devices is on the input side (for example, TCP/IP processing). Given that
similar requests may develop faster than the processor element can execute them, the capability to queue
these requests provides a more efficient environment in processing the TCP stack. In addition, the same will
be true on the outbound side as requests for data are queued and wait their turn to be transmitted to the
network.
NAS devices are configured with static RAM/Cache options in the low-end appliance level devices. Sizes of
256MB are common and are balanced to the small workloads these devices are configured for. Sizes and
options increase dramatically in the mid-range and high-end models, with RAM options able to reach limits of
one to two gigabyte levels. The configuration of system cache is highly dependent on the vendor and their
proprietary implementation of the hardware processor elements.
The NAS Storage Element
Being that NAS is a storage device, one of its most important elements is the storage system element. As
stated earlier, NAS comes with a diverse set of configurations for storage. These range from direct attached
disk configurations with minimal RAID and fault resiliency support to disk subsystems complete with FC
connectivity and a full complement of RAID options. Again, choice of NAS type is dependent on your needs
and the workloads being supported.
At the appliance level there are two types of entry-level configurations. First is the minimum Plug and Play
device that provides anywhere from 100 to 300GB of raw storage capacity. These devices provide a PC-like
storage system connected to an Ethernet network using IDE connected devices with no RAID support. These
configurations provide a single path to the disks, which range anywhere from 2 to 4 large capacity disks. As
shown in Figure 10-3, the system provides support for a workgroup or departmental workload as Windows-
based network file systems. However, these limited devices also do well with applications that support
unstructured data (for example, image and video) within a departmental or engineering workstation group.
Given their price/performance metrics, their acquisition and usage within an enterprise can be very productive.
Figure 10-3: NAS appliance-level storage configurations
The other ends of the appliance spectrum are the entry-level enterprise and mid-range NAS devices, as
shown in Figure 10-4. These devices are also Plug and Play. However, capacities start to take on proportions
that support a larger network of users and workload types. Storage systems capacities range from 300GB to 1
terabyte in size with RAID functionality at levels 0, 1, and 5. At these levels, the storage system is SCSI-based
with options for the type of SCSI bus configuration required. With this, SCSI controller functionality becomes
increasingly sophisticated. Although most use a single path into the SCSI controller, the SCSI bus
functionality, because of its architecture, has multiple LUNs connecting the disk arrays.
Figure 10-4: NAS mid-range and enterprise storage configurations
The enterprise NAS devices extend the previous configurations with storage capacities, SCSI controller
functionality, SCSI device options, and extensibility. Capacities extend into the terabyte range with multiple
path connections into the disk arrays. The number and density of disks will depend on the array configuration.
However, the top NAS configurations can support over 1,000 disk drives within a single device. At this level,
RAID functionality becomes common place with pre-bundled configurations for levels 1 and 5, depending on
your workloads.
Across the mid-range and enterprise offerings of NAS are three important hardware factors worth considering.
These are fault resilience, FC integration, and tape connectivity. The NAS architecture renders these devices
as functional units operating in a singular and unattended capacity. Therefore, the capability to provide a level
of hardware fault tolerance particularly for the density and operation of the storage system has become a
characteristic of the NAS solution. While Storage Area Networks provide another pillar to the storage
networking set of solutions, the requirement for NAS to interoperate with SANs is increasing as related user
data becomes interspersed among these devices. Finally, one of the key drawbacks to keeping critical user
data on NAS was its inability to participate in an enterprise (or departmental) backup and recovery solution.
This deficiency has slowly abated through the increased offerings of direct access to tape systems bundled
with the NAS devices.
The ability to recover from a disk hardware failure has become sophisticated and standard among enterprise
storage systems. These features have become integrated into the higher mid-range solutions and enterprise
NAS devices. Look for full power supply and fan redundancy at the high end, and elements of both as these
recovery options evolve from appliance-level solutions. This can be beneficial, as the environmental factors
will differ greatly from closet installations to crowded data center raised-floor implementations. As one failure
perpetuates another, this is an important place to consider fault-tolerant elements, because the inoperative
power supply will bring the entire disk array down and potentially damage disk drives in the process. On the
other hand, an inoperative or poorly operating fan will allow drives to overheat with potentially disastrous
effects.
Another important aspect to the fault-tolerant elements is the capability to recover from a single inoperative
drive within one of the disk arrays. NAS devices at the mid-range and enterprise levels will offer hot-
swappable drive units. Used in conjunction with RAID functionality, either (software- or hardware-based) will
allow the device to be repaired without the data being unavailable. Performance, however, is another
consideration during a disk crash recovery period and is dependent on the workload and I/O content
(additional discussion on this condition can be found in Part VI).
As with any recovery strategy, the importance of backing up the data on the storage system cant be
underestimated. As stated previously, this is one area that has been lacking in the configurations and
operation of NAS devices: how to back up the data within existing enterprise backup/recovery operations.
Look for NAS devices to provide attachments to tape units. These will be offered as SCSI bus attachments
where enterprise-level tape devices can be attached directly to the NAS device. Although a step in the right
direction, there are two important considerations to make before launching into that area. First is the
transparent implementation of yet another tape library segment. Copying data from the NAS device to an
attached tape unit requires some level of supervision, ownership, and operation. In addition, the software
required to provide the backup, and more importantly, the recovery operation has to run under the micro-
kernel proprietary operating system of the NAS device. So, be prepared to support new tapes and new
software to leverage this type of backup and recovery. (Additional discussion of this condition can be found in
Part VI.)
User data is now stored on disks directly attached to servers, using Storage Area Network and NAS devices.
The applications using the storage infrastructure will largely ignore the specifics of the device infrastructure
being used. That means that more interplay will happen with data that is stored between these different
architectures. This condition is accelerating the need for NAS and SAN to communicate and provide some
level of data access of each others device. Look for high-end NAS devices to provide additional
enhancements to participate within an FC-based storage area network. This will be offered as a host bus
adapter from the NAS device with the appropriate software drivers running within the NAS OS. Using this type
of configuration, the NAS device can service a file I/O and a block level I/O into the SAN. (Additional
discussion on FC and Network connectivity can be found in Chapter 12.)
The sophistication and operational usage is just evolving for these types of integrated I/O activities. Data
integrity, security, and protocol considerations (to name a few) are being developed as of this writing. In
Chapter 17, well discuss some of the factors driving these requirements and various considerations to take
note of when implementing applications within these configurations.

The NAS Processing Models
NAS uses three modes of processing. Understanding these modes helps determine and identify the growth of
NAS workloads from the appliance level into the enterprise solutions. If we continue to view NAS as the I/O
Manager for file-level processing, the modes in the following sections can be expressed.
Simple File Processing (Sfp) This mode of processing simply extends the clients F/S and storage capacity
into the NAS device. As depicted by Figure 10-5, this providesa simple transactional level of communication
through a single type of protocol. Generally supporting Windows-based systems that require additional storage
capacity for shared files or multiuser access to workgroup files, Sfp is handled by appliance-level type NAS
devices with capacities that range up to 300GB at the high-end with minimal RAID support. Due to the limited
transactional level (for example, access to shared files versus number of application transactional users),
levels of RAM at or above (but not below) 512MB will generally support and balance the I/O operations for
these configurations. Said configuration cross over to entry level and mid-range NAS devices as two things
evolve: first, the number of users and consequently the size of user data; and second, the number of protocols
supported. This can manifest itself in the heterogeneous support of client computers (for example, the addition
of UNIX and Mac clients).
Figure 10-5: The NAS Sfp model
Quality File Processing (Qfp) As user requirements move into supporting multiple protocols and increasing
data capacities, the capability to provide quality processing increases. Quality processing, as shown in Figure
10-6, is defined as the capability to provide a level of data reliability to withstand component hardware failures,
as well as support transactional level processing and heterogeneous sets of users. Qfp is supported from
entry level and mid-range level NAS devices with components such as fault- tolerant hardware components
(power supply, fan redundancy, and hot-swappable drives), RAID functionality, and multipath disk array
extensibility. Qfp is depicted in Figure 10-6 where a larger user base is supported along with transactional-like
support through protocols such as HTTP. Storage capacities range from 300GB to 1TB, with RAM moving into
the gigabyte range. These configurations cross into the area of enterprise level NAS solutions as the work
begins to scale into multitier client/server applications with a heterogeneous user base and multiprotocol
support.
Figure 10-6: The NAS Qfp model
Complex File Processing (Cfp): Figure 10-7 illustrates the complexity of requirements that NAS can grow
into. The multitier processing of application servers and the capability to service high volume Internet traffic
forms the basis for Cfp. Within these configurations, the NAS devices must support quality processing and
handle the transparency of transactional redirection. User requests coming into an application server that
provides web services are redirected to the NAS devices where the files are physically stored. A more
complex processing configuration is the integration of storage area networks. In these configurations, user
transactions must be redirected to NAS file systems and further into the level of database block I/O contained
in the SAN storage network. Cfp will clearly stress the high end of NAS devices and push the limits of what
vendors offer. Clearly, this requires the features of fault-tolerant hardware given its participation within a highly
available transactional system. The terabyte levels of storage capacities and RAM gigabyte levels
notwithstanding, the uniform and reliable performance becomes paramount in Cfp. Consequently, additional
importance will be placed on recovery options (for example, tape connectivity and backup/recovery software,
management tools, and overall enterprise data center integration).
Figure 10-7: The NAS Cfp model
NAS Workloads and NAS Hardware
The NAS workload is characterized as file-oriented processing for various applications that end users require.
Typical among these are Windows client files using personal applications such as office applications, shared
forms, and templates. Other typical applications can manage unstructured data such as image, video, and
audio files. Larger scale enterprise applications are centered on relational database tables while enterprise
Internet applications are grounded in new web-based file formats. The application notwithstanding, the
workload that NAS has to contend with centers on the characteristics of the files they are storing, or to put it
another way, the I/O content they have to satisfy.
Within the context of NAS, the I/O content is determined by the type, format, and characteristics of the file. For
example, the content of a word document is very different than the content of a JPEG image. The reason is
not only self-evident but also simple: the size and structure of the data. If we look at the access of these
contrasting file types, we see the effect each has on the NAS hardware components.
If we take the next step and observe the I/O operations necessary to access these diverse file types from the
NAS storage system, we find that the data content transferred within each I/O operation can differ greatly.
Contrasted against the operating system, we find that the I/O operation is dependent on many things, such as
the number of paths to the controller, the number of LUNs, and disk density. Consequently, the number of I/O
operations, commonly referred to as IOPS, is a number without a corresponding level of efficiency.
NoteIOPS is defined as the number of I/O operations performed per second by a computer system. Its
used to measure system performance and throughput.
Its possible, as we illustrated in Figure 10-5, that the transfer of a large text document can theoretically take
more I/O operations, resulting in an inefficient transfer of data with each I/O. Conversely, the I/O operations for
the JPEG could be done with less I/O operations, resulting in an efficient transfer of bytes of data with each
I/Overy efficient! The difference is not only the density of the disk, the cluster size as set with the operating
system, but also the capacities of the internal components of the NAS hardware elements themselves. The
point is that the workload determines the capacities for individual, as well as groups of, NAS hardware
specifications. Ensure that workloads and relative I/O content are given due consideration prior to setting the
record for the fastest NAS install.
Given the black box orientation of the NAS device, look for efficiencies in the I/O manager that will provide
levels of flexibility in these contrasting workloads. Operational evaluations of NAS devices should also cover
the effects of a black box architecture on other data hierarchy locations, including system-level cache, RAM,
and types of processor architecture.
Additionally, careful evaluation is necessary in maintaining the NAS devices for data recovery, disaster
recovery, and data archiving. Operational discussions should also cover the challenges involved in
participating in larger storage infrastructures, where NAS must interface with various storage devices such as
tape, optical, and other server disk storage systems.
NAS hardware constitutes commodity components bundled into a Plug and Play storage solution. However,
the evaluation of NAS devices requires some level of diligence in understanding the processing model used,
the need for particular NAS components, such as processor types, device extensibilities, and capacities. More
important is the type and extensibility of the storage system used with the bundled solutions. Given the rapid
growth of end-user storage, configurations should ultimately be scalable to both the enterprise and workload.

Overview
As discussed in Chapter 10, NAS hardware works like a Network I/O manager for remote applications. From
the perspective of the NAS software, which functions as the I/O manager, it essentially becomes a remote file
system with attached storage. As such, similar consideration should be given to the software components that
utilize the corresponding NAS hardware features. Working together, they form the complex functionality that
facilitates Network Attached Storage. This chapter discusses the software components and their sometimes-
eclectic characteristics. Given the bundled architecture of the NAS devices and the black box orientation of
NAS, nee storage appliance, each component will confront different issues as they perform their primary
functions as a remote file system and attached storage.
NAS software can be categorized into three major components, as shown Figure 11-1. The first is the NAS
micro-kernel, an optimized operating system. Working with the micro-kernel is file system software and file
communication protocols that provide the NAS with its remote file capabilities. The foundation for the NAS
solution is the storage software itself, made up of device drivers, RAID software, and other feature
enhancements provided by particular vendors.
Figure 11-1: Overview of the NAS components
Although a fourth category exists, defined as the connectivity component, we defer discussion of this until the
next chapter. The importance of networking and connectivity issues demand enough attention that in-depth
discussions need to be dedicated to these subjects alone. Consequently, this chapter will deal with the internal
software components of the NAS devices, with network communication factors being addressed only in a high
level.

A Remote File System with Attached Storage
As stated in the previous chapter, the NAS devices perform a singular directive with a duality of function,
network file I/O processing, and storage. Their directive as a remote file system or network file system has a
colorful legacy enhanced and developed though multiple projects that span many academic and commercial
computer research environments.
UNIX, having long been the mainstay of scientific, academic, and research computing environments, evolved
as an open system with enhancements coming from these communities on a continual, although discretionary,
basis. The need to share data in some organized yet distributed fashion became a popular and increasingly
important development effort. Projects at AT&T labs and several major universities began to research this
need in the late 1970s. UNIX became a viable offering to both the scientific and business community during
the distributed systems revolution in the early 1980s, at which time data sharing really became predominant
due to the exponential growth of users and open networks.
The problem of data sharing was addressed in two ways: First, by AT&T labs through their development of the
RFS file system; second, and more importantly, a development effort by SUN using the inherent components
of UNIX developed a Network File System (NFS) as the standard we know today. NFS was based upon the
use of the UNIX remote procedure call (RPC). An RPC is a programming term for a 'remote procedure call,'
where one system can execute a program on another system and return the results. Implemented on all open
UNIX code bases, this formed the basis for allowing a user on one system to access files on another (see
Figure 11-2, which shows how the UNIX RPC function allows a user to execute UNIX commands on a second
system).
Figure 11-2: What is an RPC?
Although many of the academic projects provided alternative ways of using the RPC function, the basis was
the same in terms of providing access between UNIX file systems. The implementation differences were
characterized by the diverse levels of file systems, functions performed by the file system versus activities on
the client, and the sophistication of administrative and housekeeping functions. It's important to point out that
in some sense parallel development took place in providing solutions for Windows-based systems as Novell
took a similar path in providing a file server solution with its Novell Netware Network Operating System.
Although similar to what was happening in the UNIX environments, it went further as it developed a proprietary
communications protocol creating a local network of computers based upon the Windows operating system.
Other functions were also developed around the open characteristics of the UNIX operating system. These
were concepts of file and record locking, security, and extensions to the UNIX I/O manager. Unfortunately,
these functions were open to compromise given the lack of enforcement of operating system standards,
processing activities, and core UNIX functions. Operating within these environments, users were confronted
with unexpected file deletions, records modification and corruption, and security violations. In many cases, this
was predicated by the system security level perspective of UNIX academics that security was the responsibility
of the user.
Sun's Network File System became an enabling component for specific vendors looking for ways to provide a
specialized solution for file sharing with networked environments. The idea grew out of customer NFS
configurations in which general purpose UNIX workstations were used only for storing common data within a
network and making it accessible through the server's UNIX file system, something which had been set up to
export portions of the file system to authorized clients. (Figure 11-3 shows the initial configurations of UNIX
workstations and the designations of storage workstations-for example, the server.)
Figure 11-3: An initial NFS configuration
These vendors developed small footprint servers running a specialized UNIX kernel optimized for I/O and
network connectivity. Its only job was to provide shared access to files within a network, where any client
computer attached to the network could store and access a file from this specialized UNIX server. Given that
the sole component visible to the network was its storage, the only appropriate acronym was Network
Attached Storage, vis--vis NAS. (Figure 11-4 shows the initial NAS configurations with UNIX workstations,
with a close-up of an early NAS system.)
Figure 11-4: An early NFS implementation
The NAS Micro-Kernel OS
The NAS legacy relies heavily on UNIX operating system foundations and functionality as it evolved to support
distributed computing and networking. The NAS device operates from a micro-kernel operating system and
although most NAS vendors offer their products with UNIX-based micro-kernels, it's not offered by everyone.
As we'll discover, the positive aspect of standards with UNIX and Windows plays an important role in the
heterogeneous functionality of the NAS devices. For example, the POSIX standard has provided a level of
commonality to key operating system components and has become important as NAS devices expand beyond
their native UNIX families to support other operating systems, communications, and file protocols.
NotePOSIX (Portable Operating System Interface) standards were developed by IEEE for the purpose of
cross-platform interoperability.
Core systems functionality within an operating system is contained in the kernel. This consists of a collection
of software that performs the basic functions of the computer and interfaces with the lower-level micro-code
and system ROM facilities provided by the hardware vendor. Although this core-level software of the operating
system will be called different names in different implementations, such as the nucleus in IBM's MVS OS, the
term 'kernel' has become synonymous with the core functionality of the operating system used by both UNIX
variants and Microsoft Windows.
Within the kernel are the routines that manage the various aspects of computer processing, such as process
management, scheduling, memory management, and I/O supervision by the I/O manager (refer to Chapter 10
for more information). The kernel also operates in a supervisory manner and services the requests of
applications that operate in user-mode, thereby forming the fundamental cycle of processing depicted in
Figure 11-5.
Figure 11-5: Overview of the NAS micro-kernel components
An interesting aspect of the UNIX kernel is the open nature of the UNIX operating system and the user's ability
to modify the kernel code and rebuild it to suit their particular needs. This created a chaotic systems
administration problem. However, outside the bounds of a stable environment, it did provide the basis for
enhancements that solved particular problems. This flexibility provided the best platform for those individuals
and companies that had to provide a special operating system for embedded or specialized applications.
These applications generally required operating a computer as a transparent entity where the OS supported a
singular application such as mechanical automation, monitoring, and process control. This required the OS to
be small, reliable, and transparent to the end user. The development of the OS for these particular
applications was built from UNIX kernel code that had been modified to suit the specific needs of each
application. Given that these applications had to run in real time, controlling equipment and monitoring
machines, they became known as Real Time Operating Systems or RTOS.
All NAS devices in today's world are based on some type of RTOS. Given their highly customized ability to
facilitate the OS for a single application, the RTOS offerings were highly compartmentalized, allowing vendors
that used the OS to customize the components they needed. Most special applications did not require the
entire functionality of the UNIX kernel and needed both performance optimization and limited overhead
customized for the particular application. This resulted in the more common term of 'micro-kernel.' Even
though I have used this term previously and many reading this section may have encountered the term before,
the importance of accurately describing the micro-kernel cannot be overlooked given its relation to NAS
devices.
A typical micro-kernel is shown in Figure 11-6. We can see that the selection of components depends on the
application; however, there are multiple components to choose from that provide vendors with complete
flexibility in how they customize the OS. Important NAS components include the following:
Figure 11-6: A typical micro-kernel
Task management, multitasking, scheduling, context switching, priority levels
Intertask communications, message queues, interrupt and exception handling, shared memory
Dynamic Memory Management
The system clock and timing facilities
A flexible I/O and local file system
POSIX support
OS hardware targets, PowerPC, MIPS, SPARC, NEC, 80x86, Pentium, i960
The customization of the micro-kernel in NAS devices centers on effective and efficient storage I/O, its
compatibilities to remote file systems, and its effective task processing for file-level operations. Given these
tasks, the NAS micro-kernel will look like Figure 11-7 and contain the necessary components. On top of the
micro-kernel are the embedded applications each vendor supplies in addition to the unique I/O processing
through customization of the kernel. Key areas within these applications are the storage and network device
drivers, file systems, and buffer and cache management.
Figure 11-7: An example of an NAS micro-kernel
Another key area to introduce is the implementation of non-volatile RAM used in many NAS devices (for
example, NV-RAM). Implemented through the physical RAM capacities, it is driven through specific memory
management routines that allow these buffer areas to be used for checkpoints in the status of the file system
and related I/O operations. This is memory that is protected through some type of battery backup so that when
power is off, the contents of memory are not flushed but are available when the system is restored. This
makes temporary storage less sensitive to system outages and permits recovery to take place quickly using
routines built into the micro-kernel proprietary applications.
NV-RAM becomes an important factor when working with both device drivers and RAID software.
Not all micro-kernels are based on UNIX. The Windows NT and Windows2000 I/O managers form the basis
for NAS micro-kernels used by an assortment of large and small vendors, through the Storage Appliance Kit
(SAK) offering. Although the importance of POSIX compliance pays off, with Microsoft providing levels of
compatibility, the characteristics of NT and subsequent Windows 2000 I/O manager derivatives have different
architectures and the following additional components.
The Microsoft OS model is an object-oriented architecture. This is characterized by its use of layered
architectures and I/O synchronicity (for example, the employment of file handles as compared to virtual files
used in UNIX). The encapsulation of instructions, data, and system functions used in the object models
provides programming power and flexibility; however, it comes with a higher cost of processing and resource
overhead.
Micro-kernel communications with device drivers operate with a layered architecture. This condition allows
operating subsystems such as a file system to call services to read a specific amount of data within a file. This
is accomplished by the operating system service passing this request to the I/O manager, who calls a disk
driver. The device driver actually positions the disk to the right cylinder, track, and cluster and reads the
specific data and transfers control and data back to the I/O manager. The I/O manager then returns the
requested data to the file system process that initiated the request on behalf of the application.
The Microsoft model uses mapped I/O in conjunction with its virtual memory and cache manager. This allows
access to files in memory and manages their currency through normal paging activities of the virtual memory
manager.
Although both UNIX and Microsoft architectures support synchronous and asynchronous I/O processing, the
level of processing affinity is consistent within the Microsoft model. The asynchronous model used for many
years in mainframe multitasking systems allows applications to continue to process while slower I/O devices
complete their operations. This becomes important in processing scalability as more CPUs are added to the
processor component.
Given its multilayered architecture and its ability to shield the developer from many of the OS internals, the
Microsoft model comes with additional overhead. Considering the increased levels of abstraction of object
orientation resulting in multilayered functions (for example, device drivers calling device drivers and file
handles as virtual files or representations of the files in memory), it remains difficult to minimize the extra
processing inherent with any of its OS RTOS derivatives.
The NAS File System
NAS uses a network file system or NFS. The implementation of NFS works by the use of the RPC, or remote
procedure calls, to manipulate a UNIX file system existing on a server. The system encapsulates these
functions to allow remote users on a network to issue commands to the server's file system. However, in order
to understand the internal happenings of the system, we must look briefly at UNIX file systems.
The UNIX File System
The file system is one of the major components of an operating system. It is responsible for storing information
on disk drives and retrieving and updating this information as directed by the user or application program.
Users and application programs access file systems through system calls such as open, create, read, write,
link, and unlink commands. The UNIX system consists of several different types of files. These are Ordinary
files, Directory files, Special Files, and FIFO files.
For our discussion, the most important are the Directory files that manage the cataloging of the file system.
Directories associate names with groups of files. Ordinary files are generally reflected through directory names
created by users and applications. Special files are used to access peripheral devices. FIFO files are used by
UNIX functions called PIPES that pass information and data from program to program.
The UNIX system contains a multilevel hierarchical directory structure set up as an inverted tree with a single
root node (see Figure 11-8). When the system is booted, the root node is visible to the system; other file
systems are attached by mounting them under the root or subsequently under the current directory known to
the system. While directories cannot be split across file systems (they're restricted to a single reference of a
physical disk or logical disk partition), files can appear in several directories and may be referenced by the
same or different names.
Figure 11-8: The UNIX hierarchical directory system
The directory entry for a file consists of the filename and a pointer to a data structure called an inode. The
inode describes the disk addresses for the file data segments. The association between a file and a directory
is called a link. Consequently a file may be linked to a number of directories. An important feature of UNIX
systems is that all links have equal status. Access control and write privileges are a function of the file itself
and have no direct connection with the directory. However, a directory has its own access control and the user
has to satisfy the directory access control first, followed by the file access control in order to read, write, or
execute the file.
Figure 11-9 shows an example of how the mount command works as users mount file systems for access. A
formatted block device may be mounted at any branch directory level of the current hierarchy.
Figure 11-9: Mounting a file system
Since file systems can be mounted, they can also be unmounted. The unmount system call handles all the
checking to ensure that it is unmounting the right device number, that it updates the system tables, and that
the inode mount flags are cleared.
Internal Operations of NFS
The implementation of NFS is supported on a number of UNIX implementations. These installations support
transparent network wide read/write access to files and directories. Workstations or file servers that allow their
file systems to be accessible for remote access accomplish this through the export command. The export
command allows these file systems to be sharable resources to the network. Workstations, or clients, wishing
to use these files import the file systems to access the files.
Let's follow an example of a user requesting access to data within an NFS configuration. User 'A' requires
engineering drawings and a specifications document in the /Eng folder that is supposed to be shared by all in
his department. Here is what has to happen for him to access this in a transparent mode.
First, as indicated in the following illustration, the system administrator has to set the /Eng directory as
exportable to the network. They do this by using the export command, which allows users who have access to
the network to share this directory and the files associated with it. Secondly, the user has to initiate a mount
command that allows the server side to issue a virtual file system, consisting of directories as indicated in
Figure 11-10. Although simple in an external representation, the internal activities provide us with a better
indication of the complexities involved in using NFS.
Figure 11-10: Mounting a file directory
The functions associated with this illustration are broken down as activities on the server file system. This is
both in the form of local commands to the file system and remote commands via RPC. The server RPC sends
out the acknowledgment any time a client wishes to mount the NFS file system (sometimes referred to as
mounting the NFS device), given that the UNIX file system refers to a logical disk. Consequently, the client
sends initial requests to the server in the form of RPC to initiate the import, linking, and access to the NFS file
system. Once this connection is made, the client then communicates via RPCs to gain read/write access to
the associated files.
Although transparent to most users, all this takes place under the cover of UNIX and the application, the
configuration and processing sequence provides an affordable storage network using general purpose
hardware, operating systems, and networks.
The Storage Software
The third major component is the storage software consisting of device drivers and RAID support. Most
prominent for consideration are RAID support and the relative inflexibility outside of pre-configured arrays, as
well as the use of different levels such as RAID 4. As discussed in Part II, RAID storage systems are offered
in various levels depending on the workload supported and fault resiliency required. This is important when
considering NAS storage software options and their relative RAID support because NAS is a bundled solution
and the level of RAID support may not be easily changed or upgraded.
Device driver support is also important, as the flexibility to upgrade and support new types of devices is driven
from this element of NAS software. As stated previously, NAS supports enhanced disk hardware and is
starting to offer support for attached tape drives for backup/recovery and archival operations. Within disk
support are drive connectivity options that are now being offered in both SCSI and FC disk arrays. FC support
should not be confused with storage area network connectivity or with functionality as a switched fabric.
Additional information about NAS-SAN integration can be found in Chapter 20.
Support for RAID is determined by the NAS solution. Entry-level appliance devices come with little to no RAID
support (for example, disk mirroring, both protected and unprotected). On the other end, enterprise
multiterabyte NAS solutions have a pre-configured RAID selection ranging from level 1, disk mirroring, to level
5, disk striping with protection. Other vendors offer levels of RAID that best support their file system and
processing architecture.
RAID level 1, disk mirroring with protection, is configured to provide a redundant file system. Figure 11-11
illustrates this point as the file system data is effectively duplicated within the storage arrays. Actually, the file
system processing knows that two systems exist and each will be updated as files are updated and changed.
Protection is provided by way of the storage software's ability to synchronize its processing so that in the event
of a failure on any of the arrays, the other set of files can continue processing. This poses a problem when it
comes to the resynchronization of both file systems after a failure. The functions of synchronization become a
feature of the NAS solution. As stated previously, the entry level NAS solutions provide just straight mirroring
with no synchronization. This is considered a RAID 0 or disk-mirroring solution without the benefit of
continuous operations. The data would be protected but the system would become inoperative while the
duplicate mirror is switched to become the primary file system.
Figure 11-11: NAS using RAID level 1
As the NAS solution assumes a larger role in transactional types of processing (for example, multiusers and
OLTP-type transaction support), the need for continuous processing becomes more defined. In these cases,
the NAS solution offers RAID level 5 support. This provides the file system with the ability to strip its files
throughout a storage array. To provide data and continuous operation protection, parity data is created on
each device across the array.
The storage software uses the parity information, which is also striped and distributed across each drive,
allowing it to recover from disk failures within the array. Recalculating the parity information and rebuilding the
lost data and information onto the remaining disks accomplishes this. Figure 11-12 illustrates the same file
system functioning within a RAID 5 environment.
RAID 5 works well in processing models where transactions are small and write operations are limited. As we
can conclude, the RAID 5 software and related disk operations in providing parity, file, and data currency
across the storage array can be intensive when performing transactional work with heavy updates. However,
this should be balanced with the read/write ratio of the workload and the ability to provide NV-RAM for
operational efficiencies.
A relatively unknown RAID level you will encounter with NAS is RAID level 4. Similar to level 5, RAID 4
provides data striping across the array; however, it reserves one disk within the array for its parity, as shown in
Figure 11-13. This offers a level of efficiency for write-intensive operations, but sacrifices operational
protection in the event the parity disk becomes inoperative. Again, this level of RAID should be balanced
against the workload and the quality of the NAS solution. Given the reliability of disk drives in today's market,
the likelihood of the parity disk becoming inoperative within a multidisk array becomes statistically non-
existent. However, for the workload that demands a fully redundant 99.999 percent uptime, this may become
an issue.
Regardless of the RAID level, the workload and subsequent service level dictates the type of storage
protection required. NAS offers a complement of RAID protection strategies such as popular levels of disk
mirroring with protection or RAID level 1, and data striping with protection offered though levels 4 and 5. It's
important to keep in mind that many of these storage protection offerings are bundled and pre-configured with
the NAS solution. Flexibility to change or upgrade to other levels of RAID will be at the discretion of the NAS
vendor.

NAS Software as an I/O Manager
If we view the NAS software components-micro-kernel, file systems, and storage software-as a composite I/O
manager, similar to how we viewed the NAS hardware, we uncover a tremendous value bundled into a single
device. Along with a composite of I/O related functions is the value of providing a distributed solution as
complex as file I/O through existing networks. This is an important distinction to identify as the difference
between a NAS product is compared to a Do It Yourself (DYI) commodity server being retrofitted to function as
a shared file server. Consider the following major functions that the NAS device provides, which the user
consequently doesn't have to worry about.
Remote File Processing The ability for an application's file I/O requests to be handled through the
network and serviced by a multipurpose file system.
Operating System Optimization The functions of an I/O optimized OS that is pre-configured, bundled,
and RAID ready. Includes I/O optimization for task, memory, and scheduling management.
File System Ubiquity Provides the levels of support necessary to process multiple file protocols and is
optimized within a single file system.
File Read/Write Processing Provides the ability to effectively, reliably, and securely store and retrieve
application data from a remote network location.
NAS Software and NAS Processing Models
In the previous chapter, we defined three NAS processing models. These were contrasted with NAS hardware
considerations which provided some insight into the general hardware requirements and capacities for each.
The following discussion can be used in conjunction with this information to balance the software
considerations and requirements for each of the NAS processing models. Support for each of these models
stresses particular elements of the NAS software and, as such, should be evaluated carefully as workloads
increase with each model and as software requirements grow from Sfp processing into enterprise-level Cfp
models.
The following represents some key considerations as simple file processing grows into more complex
processing requirements. Using these categorizations as guidelines, they can provide valuable insight into the
types of NAS devices required for departmental and data center file processing workloads.
Simple File Processing (Sfp) essentially extends the client's F/S, which should be accomplished with as
little overhead as possible. This requires a simple micro-kernel implementation with the minimum
enhancements. Processing is characterized by read-only synchronous processing with minimum queuing;
however, with this processing model effective buffer and cache management is noticeable.
Quality File Processing (Qfp), which processes files with added value for more reliability and integrity,
adds extra overhead as activity builds the importance of effective interrupt processing, device drivers, and
interaction among user- mode, kernel mode, and device-driver processing. Also, the sophistication and
usage of Authorized Control Lists (ACL) for security and the ability to provide significant reliability with
asynchronous processing and RAID levels is an important consideration of the NAS software for this
processing model. Increased transactional workloads now point to additional TCP processing, which will
leverage effective redirector functions.
Complex File Processing (Cfp) is the processing of files with the added value of user/application-defined
prioritization as well as access to other file and storage systems. As with hardware components, this will
be the greatest challenge as all elements of an I/O manager are stressed to their limits. Consequently, the
ability to provide effective and efficient storage protection, device driver communications, queue
management, and complex redirector processing is paramount.
I/O Workloads and NAS Software
The workloads for NAS devices have been characterized as file-oriented I/O operations performed on behalf
of remote applications executing somewhere within a network. An important distinction in considering these
workloads is the evolution of NAS devices as they take on a wider scope of I/O operations in supporting
applications that require more than just file access. These workloads have been traditionally difficult for
appliance- or black box-oriented RTOS machines, given their limited flexibility in performing operations outside
of file access and I/O. Two major areas that NAS is beginning to support are database products and native
data protection products.
Database processing, both in the context of the relational model and static model, typically exists where the
database engine executes on a local application server and the user data exists on the same server using
directly connected storage systems. When the user data in the form of tables are moved to a remote server or
storage system, this becomes problematic given the integrated processing that occurs within the database
engine and its associated tables (see Chapter 8). Making things more complicated in a remote setting are the
required components to support transactional type processing in the form of log files and lock management.
NoteThe Static Database Model defines a database system that is distributed with a specific application.
These bundled databases operate only with the application they are distributed with and
consequently are closed to modification and use as a general purpose database. They are often
referred to as embedded databases, or runtime databases, and are used with business applications
that range from e-mail, Enterprise Resource Planning (ERP), to Customer Resource Management
(CRM).
Activities have been underway to ease the restrictions of running databases on NAS devices. First are the
combined efforts of the NAS vendors and database companies. This has resulted in configurations where the
database engine continues to reside on a commodity server; however, the data tables are remotely stored on
the NAS device. This comes with many caveats, such as the support of these configurations by the database
and NAS vendors, given the complexity of the solution and limited user expertise. Another major consideration
is the limitations of database processing that can be sustained in these configurations. Additional discussions
in Chapter 19 will provide more detail on these implementations.
Data protection products such as backup and recovery software, mirroring, and data archiving are all used to
support integrated data center maintenance functions. These functions are integrated into data protection
activities and policies supported through software that function throughout a wide segment of operational
environments-for example, mainframe configurations, UNIX configurations, and Windows network server
environments. If a device doesn't easily fit into one of these categories, the likelihood that existing software
products will effectively support it is low. Such has been the case for NAS devices. The reasons are
straightforward given the RTOS nature of the NAS device and the additional processing that has to take place
when executing data protection software on any server. Optimized NAS devices do not have the capacity and
functions to run general-purpose backup and recovery as well as other data-protection solutions.
To overcome the limitations of data protection within NAS environments, NAS vendors and third-party storage
software vendors have been working to provide native NAS solutions-that is, solutions that are integrated into
the NAS devices. From the NAS vendors themselves, they have been enhancing their enterprise-level devices
to provide physical connectivity to tape devices, allowing the ability to copy and archive data to tape directly
from the NAS device. Although this is an effective way to handle the protection of data stored on the device,
the consideration is the integration or new development of an IT tape library. Will the new tape units integrate
into an existing tape library and how will they be protected and accessed are two questions for consideration
regarding this NAS data protection solution.
There are two other enhancements to the integration of data protection activities for NAS that increase its level
of enterprise reliability and availability support. These are mirroring and snapshot. Mirroring, a level of RAID
storage functionality previously discussed, allows NAS devices to participate in a high availability processing
strategy by providing the necessary support for real-time backup of data. This can be done locally or
configured remotely to function as a disaster recovery solution. The snapshot function allows similar
operations, however it provides a level of data redundancy, usually locally, that permits the processing of
applications to continue even as the data is copied for backup purposes. The snapshot function essentially
takes a 'snapshot' of an existing file system and maintains this copy according to the timing of the snapshot
itself. This can work in conjunction with other data protection methods, such as mirroring, to ensure multiple
data protection strategies are implemented.

Overview
NAS connectivity characterizes the real power of network storage. Connection directly to existing networks
allows the NAS devices to function as network I/O managers. However, the ease of implementation within a
network is a double-edged sword, providing administrators with an easy mode of implementation that masks
the serious implications of the network storage traffic driven by NAS workloads. The use of Ethernet network
topologies provides an effective and proven infrastructure when it comes to planning and implementing
appliance, mid-range, and enterprise NAS solutions.
From a software perspective, the use of TCP/IP allows configuration and support to be easily integrated into
existing IT knowledge and support structures. With the number of tools available to monitor networks and
network infrastructures, the ability to manage the connectivity of NAS devices is far more advanced than any
other storage networking methodology. However, the inherent nature of TCP/IP stack processing provides a
challenging mix of performance and configuration issues once connectivity transfers from the network to the
NAS server. These issues are still evolving as NAS workloads become more complex and network
connectivity extends beyond the Ethernet and TCP/IP infrastructures.
Connectivity plays a very visible role as NAS devices become integrated into other storage systems. The
evolution of storage networking is eliminating the boundaries between direct attach, Fibre Channel networks,
and many sophisticated WAN configurations. Although many of these external connectivity issues will be
addressed in Parts IV and V regarding integration with other storage models, we will discuss the effect and
conditions surrounding LAN and WAN connectivity issues.

Putting Storage on the Network
A major benefit of NAS devices is that they utilize existing Ethernet resources and additional high-speed
connectivity hardware. NAS vendors support all levels of Ethernet including 10 to 100 based T media to
Gigabit standards. The connections are enabled through industry standard NIC cards that offer both
concurrent Ethernet support and high-speed connections. Most vendors will offer various configurations
regarding the type of connections, number of NICs supported, and related TCP/IP drivers. However, most
devices will remain limited in the appliance and entry-level NAS solutions, with additional enhancements and
offerings through mid-range and enterprise level NAS servers.

NAS Connectivity Architecture: Hardware
Bus connectivity for the NICs become proprietary as they attach through the I/O adapters of the NAS server
bus. Most vendors adhere to the PCI standard; however, this is difficult to determine without specific inquiries
to the vendor. This becomes an important aspect given that the number of transactions supported depends on
the performance of the NICs and their respective bus interfaces.
NAS and LAN Topologies
The following discussion illustrates the various LAN topologies where NAS configurations are installed. Each
offers a set of challenges and considerations, and in order to effectively understand these, we offer a
summary overview of LAN fundamentals.
Understanding the fundamentals of network hardware devices is important to how the NAS solutions will
perform in any network topology. Network connectivity devices such as bridges, hubs, routers, and switches
operate at specific levels of the OSI reference network model while performing their duties. As you may recall,
the OSI network model forms the layers for TCP/IP processing. The most important layers to consider for
network devices are the physical access layer (layer 1), the datalink layer, (layer 2), the network layer (layer 3),
and the transport layer (layer 4). They are depicted in Figure 12-1 and described below through their
connection proximity and processing responsibilities.
Figure 12-1: OSI Layersfrom the physical to the transport
The Physical Layer (Layer 1)The physical layer provides the direct mechanical and electrical
connection to the network media (the wires and cabling that sends the data in bits throughout its physical
interconnections). Layer 1 interfaces directly with the data link layer through the media access control
(MAC) sublayer.
The Datalink Layer (Layer 2)The datalink layer separates the transmitted data into frames for
transmission over the physical media, such as 10 base T5, 10 base T2 cable, or UTP wiring. This layer
performs error-checking and is responsible for retransmitting frames. It provides a direct channel to the
network layer through a logical link control (LLC) sublayer and a lower sublayer which connects to the
transmission media called the media access control layer (MAC).
The Network Layer (Layer 3)The network layer provides information on the routing of data transmitted
from sender to receiver. This layer ecncapsulates the user data through routable network protocols such
as, IP, IPX, SNA, or AppleTalk.
The Transport Layer (Layer 4)The transport layer provides information on how a point-to-point
connection is developed. Layer 4 utilizes the network layer to determine the type of connection between
computers. This is based upon TCP or UDP protocols. These two connections set up sessions between
computers, and determine the protocol under which they will operate. However, it should be noted that
TCP is a connection-oriented protocol while UDP is connectionless, which characterizes their
performance and error recovery. This facilitates the use of Remote Procedure Calls between operating
systems (refer to RPCs in Chapter 11).
Using the OSI model, network devices operate using layers 1 through 3 to develop a local area network (LAN).
The network devices used to leverage each layer of the OSI model in building a campus network or LAN is
diverse, sometimes interchangeable, and can be confusing if you are inexperienced in network administration
and management. However, they are illustrated in Figure 12-2 and categorized into the following segments.
Figure 12-2: Network devices and their corresponding OSI layer
Repeaters, Concentrators, and HubsAlthough most are becoming obsolete given the evolution of LAN
switching devices, many still exist, and all operate at the physical level. Repeaters provide a
retransmission of signals in order to increase the range of physical segments of wiring. They continue to
be used to extend the physical distance capabilities of LAN wiring segments. Concentrators and hubs
provide the same servicesthat is, the connection of clients and servers through one wiring segment.
This results in all servers, PCs, and workstations on the segment sharing the bandwidth and forming a
single collision domain, as it relates to the machine access layer (MAC) algorithms that use the Ethernet
Carrier Sense Access/Collision Detection (CSMA/CD) protocol.
Bridges and Layer 2 SwitchesBridges interconnect different types of physical network topologies such
as FDDI, Ethernet, and token ring, and operate at layer 2: the data link layer. They also can be
responsible for filtering network traffic using the layer 2 MAC addresses. Layer 2 switches connect
different port connections and enable the transfer of data by destination address. They are commonly
used to provide a backbone network connection where clients can reach servers outside their domain and
interconnect to outside WAN services.
Routers and Layer 3 SwitchesRouters or layer 3 switches forward data packets based upon IP
address, link, or network availability and performance. Layer 3 switches perform similar functions as
routers. They provide this service from two distinct architectures, both operating under layer 3 information.
Port switches operate at the hardware level of the switch by directing network traffic from port to port
according to layer 3 information. They also perform at chip speeds. The follow-on from port switches are
frame switches that operate from layer-3 information; however, they perform port traffic direction by
examining the data packet and providing the most effective routing to the destination.
Layer 4 SwitchThese switches make connection decisions based on information in layer 4 regarding
session and application layer information. This allows them to provide such services as load balancing
across servers and broadcast services.
We can view the placement of NAS devices within two major topologies: appliance placement and enterprise
placement. Appliance placement facilitates the placement of data as close to the user as possible. This
requires the placement of an NAS device within a hub or single-network segment. The enterprise placement
provides greater access to data through multiple segments of the network. Given the variability of the network
designs and topologies, there are many ways of implementing these configurations. We will discuss these two
major approaches and their related considerations.
NAS Network Deployment
To discuss the NAS deployment in as much context as possible, we offer a typical network infrastructure, as
illustrated in Figure 12-3. This conceptual network supports a theoretical business that is populated with a
headquarters staff within a three-story building, and a data center which is housed in a separate building. The
business is externally connected to the Internet and outside leased lines. External sales and marketing
personnel utilize the network through Internet connectivity and servers connected through outside dedicated
leased lines. This conceptual network does not illustrate particular topologies, but is designed to represent
major components that will both influence and pose deployment challenges to NAS devices.
Figure 12-3: A typical network infrastructure
As mentioned previously, there are two categories of deployment: appliance level and enterprise level.
Although derived by various requirements, including the network infrastructure, initial deployment decisions
center on data ownership and access. This means the user community may feel they own the data, such as
an accounting and finance department, and have characteristics of homogeneous access (in other words, only
persons within the accounting department can access the information).
Another perspective is data that becomes developed from multiple sources and accessed by multiple user
communities. Often characterized by marketing and sales departments that utilize multiple sources to build
data bases, data access profiles become very complex. User access can be on multiple levels (say, for sales
people, executives, and order-entry administrators), and therefore requests for data may traverse multiple
network segments. More importantly, answer sets tend to move through multiple network segments.
Yet another view of external access within the data center comes from using combinations of homogeneous
and heterogeneous access. The most obvious are facilities that store web-based data for access to
companies outside sales personnel, partners, and customers. An important aspect of this data placement is,
of course, security; however, data integrity and availability is paramount as well.
NAS Appliance Network DeploymentGiven our typical company, lets analyze the homogeneous access of
data. This type of access and data placement easily justifies the appliance level of NAS solution. A subset of
our large network infrastructure is illustrated in Figure 12-4, where the Accounting department has installed an
NAS appliance, and access is contained within the network segment handling the first floor. Note that this
segment is configured with a hub that provides a connection between the PCs and workstations used in the
finance department, as well as the main server handling the finance client/server financial applications. The
NAS device handles the I/O load of the financial application server and performs very well as long as the
traffic is contained with the network segment.
Figure 12-4: The network segment supporting Accounting
The challenges and performance issues begin to show up as users from other network segments begin to
access the NAS device. Outside users degrade service by adding additional traffic that is serviced by a hub
device which must share its bandwidth within the network segment. Secondly, as users outside the network
segment access the NAS accounting data, they must traverse several jumps from one network segment to
another. These are referred to as hops. The more hops the more network latency is built up, adding to
aggregate traffic on the network backbone.
Finally, the type of I/O workload also determines the placement factors and, more importantly, the required
bandwidth for network segments and the capacities of network devices. Consider our example of the
accounting data. One reason this works well is the financial planning data thats used within this area. These
worksheets can be extremely large and I/O-intensive as they are processed by the financial application server.
The answer sets and results of these financial planning programs are extremely I/O-intensive, however, when
handled within the limits of the hub and configured connections (for instance, multiple ports for the server and
multiple ports for the NAS device).
NAS Enterprise Network DeploymentCertainly, not all data access is contained within a network segment.
As a matter of fact, many corporate applications must be accessed through the data center. Figure 12-5
shows the network segments within the data center. This portrays a more complex set of segments than the
local user networks. Considering that the data center must support multiple user access, applications, and
data, its requirements for network segmentation become comprehensive. If we add external access to the
Internet as well as dedicated leased lines, the network infrastructure becomes the focal point for corporate
data access and makes data placement a more daunting task.
Figure 12-5: Network segments within the data center
Lets take our example of sales and marketing data that must be accessed by multiple users, and which is
then developed from multiple data sources, both internal and external. Although this is also a good candidate
for NAS solution, placement and capacities become key to a successful implementation. Without restating
discussions that will be covered in Part V, lets assume that the sales and marketing data have the following
characteristics.
Sales DataSales orders, tracking, and customer information are contained in the ERP system, which is
processed and accessed through the UNIX servers within the data center. Data access is handled by
sales administration through the third-floor network segment, outside sales people from the web sites, and
finally by limited users that work in branch offices and access the ERP system through the dedicated
leased lines.
Marketing DataMarketing information, product information, brochures, market analysis, and data mart
analysis are contained in every system in the data center, the mainframe, the UNIX servers, and the web
servers. The information used to build market and customer analysis data marts are copied from sales
orders, customers, customer tracking, and products.
Two solutions are productive in these situations. First, is the use of NAS devices for the collection and storage
of web data. Second, is a large NAS device to hold the sales data that is accessed by sales personnel in the
field. An additional important installation would be the use of NAS devices to hold large data base tables that
make up the marketing datamart (for example, market and customer analysis data). Each of these solutions
can be placed within the data center for general access. Given that the data center connects to the network
backbone and contains a secondary network backbone within the data center (as shown in Figure 12-3), the
bandwidth to handle multiple user access can be configured with high-speed network resources and devices.
Although on the surface it may appear that the installation of these large NAS devices should be on the data
center backbone, when we look closer at the networks makeup, we find that each segment within the data
center is handled by a layer 3 layer 4 switch. Each of these devices provides additional flexibility in the
placement of NAS devices given their I/O responsibility. For example, the NAS sales data device could be
placed within the UNIX segment given the network access from both internal and external users. The NAS
device could be placed within the web segment given that access, security, and sourcing can be managed
with existing web workloads. (We placed a layer 4 switch in our example shown in Figure 12-5; this could
leverage workload balancing effectively, however, from a newer layer 5 switching device.)
NAS over WAN Topologies
Part of our discussion has touched on network challenges and various issues with LAN topologies. However,
as Figure 12-5 shows, there are outside network connections that influence various NAS configurations, along
with their placement and opportunities. Each configuration discussed earlier offers a set of challenges and
considerations which effectively utilize WAN connections.
To facilitate this understanding, we offer a summary overview of WAN and LAN-WAN fundamentals. Figure
12-6 shows the external connections from the data center into the outside world. There are two major types
depicted, Internet connections and dedicated leased-line connections. The Internet connections are leased
lines that attach to an Internet service provider (ISP) where additional connections provide external Internet
accesses. These lines are dedicated and depend on the grade and capacity of the line, the capacity and
resources of the ISP, and the type of service the company has contracted for. In any event, it is outside the
control of the NAS device and internal network infrastructure. Basically, you have to deal with what you have
available, in Internet terms.
Figure 12-6: WAN connectivity options
The second external connection is through dedicated lease lines that are available through the long distance
carrier, IXC, and local exchange LXC carrier. They provide dedicated lines and services through a wide variety
of telecommunications transport technologies that include frame relays, ATMs, and other communications
protocols. Actually, once it leaves the confines of the data center, the data being transferred is within the
control and management of the IXC or LXC, which could mean, and usually does mean, multiple
communications protocols, circuits, and equipment. Basically, you have to deal with what you have available,
in long distance and local carrier terms, same as the Internet.
The point is that when you leave the data center and transmit data either through a myriad of Internet
connections and equipment or the IXC and LXC set of circuits and equipment, it is out of your control. This
requires that you compensate for each of these conditions and deploy the NAS devices with due consideration
for the issues of each. Essentially, these are areas of security and recoverability. Given that you are
dependent on the availability of specific communications vendors as well as bandwidth and service levels from
these local and long distance vendors, your confidence and history with vendor services should drive your
deployment choices.
From the NAS perspective, there are two considerations for deployment that facilitate access and data
placement. These are access to data from the remote branch locations that require application access and
specific services of the data center. Secondly, the ability to provide levels of application and data recoverability
in a disaster recovery scenario is a major consideration for remote offices. Each of these deployments utilizes
dedicated leased lines. These solutions are depicted in Figure 12-7, where a NAS device is deployed within
the data center attached to the data center backbone that the remote offices utilize to back up their data. This
is followed by a more complex deployment where production NAS devices hold sales and order data which are
mirrored into a disaster recovery site where a similar NAS device holds the synchronized data.
Figure 12-7: NAS deployment using WAN connections
The second D/R deployment employs the ability to utilize RAID level 1 functions of the NAS devices to provide
an asynchronous link between the data center NAS device and the D/R site NAS device (see Figure 12-7).
This solution utilizes an asynchronous update process that does not require the production NAS device to wait
or hold I/O processing before the D/R site commits to writing the same data. The setup compensates for the
availability and reliability factors within the WAN connection to minimize disruption with production processing.
The backup deployment for the remote sites is less complex and, although performed in an asynchronous
manner, is generally done at night on an automated basis. Backup processes are run at the remote sites that
transfer copies or backups to the NAS backup device attached to the data center network backbone in order
to facilitate the greatest level of bandwidth and throughput as possible (see Figure 12-5). This compensates
for the WAN connection being consumed during the daytime with user transaction traffic and minimizes the
disruption of production processing within the data center during prime usage.
Internet access is the other major category for NAS placement that has WAN effectsalthough in many cases
it is secondary given the typical placement of the NAS device. NAS provides an excellent solution for the
storage of web files that are dynamically created through user access to web servers. In Figure 12-7, our
theoretical company provides web servers that link to the outside world through a firewall and then switch to
the ISP. The ISP then supports global connectivity to the root servers of the Internet, sometimes referred to as
the 13 root servers. This connection enables the global communications that allow data to flow in and out of
the web servers. Consequently, this becomes a critical point for managing traffic access and security with
which the NAS devices become integrated.
In any event, the WAN access as it comes in from the Web is in the form of presentation layer commands, or
HTTP requests. However, the resolution of IP addresses, and then subsequent network and computer
addresses, are provided by the switch connected to the ISP circuits; this is often called an edge switch.
Given the transmission of packets from the edge switch, additional direction is provided, vis--vis our example
in Figure 12-7 through the layer 4 switch where load balancing is facilitated to balance access among the web
servers. Finally, the redirection will take place to initiate the file I/O to the appropriate NAS storage device.
Figure 12-8 illustrates this in more detail. The importance of the transmission path plays an important role in
the placement of the NAS devices within the data center segment. More importantly, placement within the
subsegment provides additional functionality with handling the type of traffic that may be expected with web
access. Our example also points out that there is a scalable path for growth as both web servers and
associated NAS storage devices will undoubtedly grow within the network segment in conjunction with the
network itself (in other words, as web access grows, network bandwidth needs increase along with layer 3 and
layer 4 switching and the need for NAS storage). Given NAS placement in this scenario, the NAS boxes follow
a logical path that is associated with related network infrastructure components.
Figure 12-8: TCP/IP considerations for NAS deployment

NAS Connectivity Architecture: Software
TCP/IP is the third of the three major components supporting communications within a diversity of transport
media; however, the most popular data communications protocol remains Ethernet. Support here is evolving
with increasing bandwidth to support Gigabit Ethernet standards, sophisticated participation within enterprise
networks with quality of service (QoS) and routing participation. However, it should be noted that one of the
limiting factors of file I/O within a TCP/IP network environment is the processing of the TCP layers for
communications between client and server (for example, NAS device) elements. Note that this should reflect
the TCP Off-load Engine (TOE) technology and other initatives such as InfiniBand that are evolving to
optimize this latency.
To understand the relationships of network software and NAS we must describe the basic concepts behind the
transport software that support the transmission of data throughout a network. As we discussed in Part II, the
blueprint for open networks is the OSI layers of network processing. Parts of the OSI layers have been
discussed in the previous sections of this chapter in explaining how the network devices work within LANs. We
left that discussion within the transport layer, or layer 4.
We will begin the software discussion at the network layer and move up through the additional layers of the
OSI model, discussing transport layer, session layer, presentation layer, and application layer. Most of our
time will be spent on network, transport, and session layers due to their contribution to TCP/IP processing and
their effect on I/O processing within the NAS model. However, it is important that we also mention the session
and presentation layers and their effect in transporting the file level protocols, such as HTTP and FTP, and the
subsequent interchange between applications and presentation layers.
The OSI model represents all the required processing for making a connection within a network as well as the
necessary connections to applications that occur within a client/server computing model. Figure 12-9 shows
layer 3, the network layer, and the remaining layers of the model: transport, session, presentation, and
application. Each represents an important aspect of processing applications within the client/server model.
More importantly, they represent a focused set of activities as they relate to NAS devices, along with their
integration and interchange with the NAS micro-kernel.
Figure 12-9: OSI model with layers 4 through 7
An Overview of TCP/IP Processing
We begin the following overview by taking a look at how TCP and IP work to encapsulate and transport both
data and messages throughout a network. TCP/IP supports data transmissions and functions within a
switched packet network. This means that IP provides the packets and associated delivery information to the
network. As discussed in the network device overviews, many of the switch and router devices derive their
function from IP addresses.
Although we have already referenced IP in association with TCP/IP, the function of the Internet Protocol (IP)
component is critically important to the transmission of data. IP functions as the envelope for application
communications within a network. To that end, it provides an address, consisting of both network and
computer addresses, for delivery within the Internet networking topologies. Using the IP address provides a
means of locating any computer within the universe of the Internet and local networks that is designated to
receive the IP packet.
All computers running TCP/IP software have IP addresses. These are assigned when the TCP/IP software is
installed and initialized. The IP assignments become part of the computers network name and specific
address within the network. The addresses, being a collection of numbers, rely on processing by the network
devices to evaluate the IP address and route the IP packet to the appropriate network, and subsequently to
the specific computer.
NAS functions on a TCP/IP communications basis. Therefore, NAS devices run TCP/IP and consequently play
by the same rules as other computers that are addressed through IP address schemes. There are several
considerations to this, howeverthe most important being the placement of NAS within the network segments
due to its IP addressability and transmission of I/O file content within the IP packet.
The considerations are the placement and usage of the layer 3 and layer 4 network devices which can have a
positive or negative effect on an NAS device. Well look at specific performance conditions in Part VI, however
its important to understand these devices as they operate with an NAS device. Figure 12-10 further illustrates
our example network with NAS being fronted with both the layer 4 device, supporting UNIX applications
servers, and the layer 5 device, supporting e-mail and web servers. The layer 4 devices examine the IP packet
as it comes into the network segment, and then route it according to the routing tables it maintains. This allows
some requests to be transmitted directly to the NAS device for I/O file processing without them being
redirected from the server. Such well-performed tasks illustrate the real value of the NAS solution. Some
requests, however, are more generic and are sent to the application server where theyre redirected to the
NAS device. This is a function of the application and its related request.
Figure 12-10: Network operational aspects of NAS deployment
Also a function of the NAS device to support the application server I/O requests, some degree of network
efficiency can be derived by processing I/O requests directly through to the NAS device. However, layer 4
network devices only route the IP address and dont provide any additional intelligence as to its routing
processes.
Layer 5 network devices may be more appropriate when trying to mediate complex data access requirements
regarding application servers supported by NAS devices. One way this is done is through load balancing,
where layer 5 network devices working at the session and presentation level can derive information about its
network traffic and route accordingly. Therefore, in some instances, NAS devices can participate in load
balancing by routing all explicit or some masked inexplicit I/O file requests directly to the NAS server.
A client request for data uses TCP to encapsulate and break the request into packets and then passes them
to the IP functions. IP provides addressing information in the form of headers regarding the destination of the
packets. Another important distinction of the TCP processing is the error checking it provides, especially for
datacentric requests.
TCP uses a common error algorithm called checksum to calculate the data within the packet. When the
packet has reached its destination, the IP functions verify the correct address and pass the packet to TCP.
TCP, on the receiving end, breaks the packet and reassembles the initial client request. On the receiving end,
the TCP checksum error routines verify that the correct amount of data was sent. If not, TCP returns the
packet to IP, which requests a retransmission from the sending client.
The effect of the IP and subsequent TCP processing places an additional workload on both the sending and
receiving computer. In the case of a server, the overhead can be extremely high given the amount of TCP/IP
processing required with multiuser access. Given that the majority of overhead occurs within the TCP and IP
software running on a particular server, this can become critical when considering the case for NAS.
There are important points regarding NAS that need to be thought through when working with TCP/IP
processing. First, is the number of concurrent users possible with the workloads. This is actually the number of
client or server redirect I/O requests planned for the NAS device. Placement within a network segment must
balance the capabilities and resources of the NAS device. An obvious example would be the placement of an
NAS appliance device in a network segment that handles an enterprise workload. In addition, placement within
a network segment is imperative given it could easily become overloaded if the workload expanded beyond
the capabilities of the NAS device. This is characterized by the placement of NAS appliance devices in
network segments served by hub devices which are extended by additional client PCs or servers. The shared
bandwidth of these segments can quickly overwhelm a network segment with NAS workloads.
NAS Connectivity in Operation
Our previous example of a sample network illustrates NAS I/O operations as they relate to connectivity. Figure
12-8 shows how the network operates, using sample workloads that take place in the user departments as
well as in the data center. The Accounting and Sales departments are good examples of NAS appliance
deployment, while the more heavyweight workloads in the data center depict mid-range and enterprise NAS
solutions.
Operational aspects of Figure 12-8 include the following:
NAS AppliancesThese devices are placed within network segments that contain traffic within the local
users. The devices also can be reached through administrators and support personnel for problem and
maintenance activity. They become problematic with data protection (for example, backup and recovery),
and are scalable within their segment.
NAS Mid-RangeData center devices that support the Sales and Marketing departments are greater in
both capacity and data access resources (that is, the number of network interfaces). These devices also
support inquiries through WAN connections and, therefore, must have additional security both prior to
access and as the requests are processed within the NAS device. The scalability is much greater here
given the layer 3 and layer 4 switch and router processing, not to mention the bandwidth of the data
center backbone. Also, the mid-range NAS appliance can be upgraded in place to achieve larger storage
capacities and data access network interfaces.
NAS EnterpriseThe other data center device supports more intense workloads by housing the data
tables for data mart processing of the marketing applications. Although the access is limited to marketing
and executive users, the complexity of the processing and size of I/O access is much greater than what
we see with both NAS Mid-Range and NAS Appliances. Again, the scalability is handled through the
bandwidth of the data center backbone, the limited number of users, and layer 3 routing. However, this
solution becomes problematic with backup and recovery given its relationship to database systems
running on an application server. A related problem is the capability to provide an effective disaster
recovery solution of this magnitude over WAN mirroring functions due to bandwidth and size
considerations.
NAS WebThis solution, although similar to the mid-range NAS devices, is singled out due to its ability to
respond quickly to the Internet type of data access, as well as the volatility of storage requirements. This
solution is well placed to take advantage of both layer 3 and layer 4 network devices by providing levels of
workload balancing and caching throughout the web server configurations.
Network Configurations and Workloads
The previous sample network also provided a basis to illustrate NAS placement as it relates to I/O workloads.
Figure 12-11 shows how the I/O workloads operate within their specific placements. Our sample workloads
form an important aspect to NAS placement decisions as they relate to user departments or placement within
the data center. The accounting and sales departments are good examples of simple file processing (Sfp)
using NAS appliances, while the more heavyweight workloads in the data center provide examples of both
complex file processing (Cfp) and network file processing (Qfp) using a depiction of the mid-range and
enterprise NAS solutions.
Figure 12-11: Network workload aspects of NAS deployment
The workload aspects of Figure 12-11 include the following:
NAS AppliancesWorkload descriptions using Sfp
NAS Mid-RangeWorkload descriptions using Qfp
NAS EnterpriseWorkload descriptions using Cfp
NAS WebWorkload descriptions using Qfp and Cfp

Part IV: Storage Area Networks
Chapter List
Chapter 13: Architecture Overview
Chapter 14: Hardware Devices
Chapter 15: Software Components
Chapter 16: Configuration Options for SANs

Overview
In 1993, the largest data warehouse application supported only 50GB of aggregate data. Although this
appears trivial by todays standards, it drove the client/server and mainframe computing infrastructures to
support larger and larger data capacities. Both mainframe and open server vendors responded by increasing
the number of processors, channels, RAM memory, and bus capabilities. However, these technologies had
reached a level of maturity in their development that limited dramatic improvements in many of these areas.
Coupled with the momentum of online transactional workloads driven by Internet access, web applications,
and more robust Windows-based client/ server applications, the network began to play an increasing role in
supporting the larger data-centric platforms. However, the distribution of work through the network only found
the LAN to be challenged by its own data-centric traffic congestion.
In a different response to this challenge, both new and established vendors moved toward hybrid types of
computing platforms, while methods appeared to handle the growing data-centric problem (see Chapter 8).
These were parallel processing and high-end SMP computing platforms using high-performance relational
database software that supported increased degrees of parallel processing. Vendors of systems and
databases were thus caught in a dilemma, given that applications requiring data storage above 100GB
capacities needed very costly high-end solutions.
Driven by the sheer value of access to data, end-user appetites for data-centric solutions continued to grow
unabated. However, these requirements proved increasingly difficult for IT organizations as they struggled to
supply, manage, and integrate hybrid solutions into their data centers. Obviously, one of the main areas
experiencing dynamic growth within this maelstrom of datacentric activity was storage and storage-related
products. This was an interesting dilemma for the storage vendors, given that the key to any enhancement of
the data-centric application rollout revolved around a high-end I/O system. This drove storage vendors to the
same traditional solutions as their mainframe and open-server brethren: enhance the existing infrastructure of
disk density, bus performance (SCSI, PCI), and array scalability.
As the conservative approach from mainframe and open server vendors drove innovators offering alternative
solutions to start their own companies and initiatives, a similar evolution began within storage companies.
Existing storage vendors stuck to their conservative strategy to enhance existing technologies, spawning new
initiatives in the storage industry.
Several initiatives studied applying a network concept to storage infrastructures that allowed processing nodes
(for example, servers) to access the network for data. This evolved into creating a storage network from
existing architectures and technologies, and interfacing this with existing I/O technologies of both server and
storage products. Using the channel-oriented protocol of Fibre Channel, a network model of packet-based
switching (FC uses frames in place of packets, however), and a specialized operating environment using the
micro-kernel concept, the architecture of Storage Area Networks came into being.
This was such a major shift from the traditional architecture of directly connecting storage devices to a server
or mainframe that an entire I/O architecture was turned upside down, prompting a major paradigm shift. This
shift is not to be overlooked, or taken lightly, because with any major change in how things function, it must
first be understood, analyzed, and evaluated before all its values can be seen. With Storage Area Networks,
we are in that mode. Not that the technology is not useful today. It can be very useful. However, the full value
of Storage Area Networks as the next generation of I/O infrastructures still continues to evolve.
Creating a network for storage affects not only how we view storage systems and related products, but also
how we can effectively use data within data-centric applications still in demand today. Here we are, ten years
from the high-end data warehouse of 50GB, with todays applications supporting 500GB on average. This is
only a ten-fold improvement. What can we achieve if we move our data-centric applications into an I/O
network developed specifically for storage (perhaps a hundred-fold improvement)? Hopefully, ten years from
now, this book will reflect the average database supporting 5 terabytes and being shared among all the
servers in the data center.

Creating a Network for Storage
Network attached storage is very different from a Storage Area Network on many levels. First, SANs denote
an entire infrastructure supporting storage systems. Secondly, (and maybe this should actually be first) SANs
function on their own network developed specifically for shared I/O processing, enhanced storage devices,
and scalability within a data center infrastructure. Thirdly, SANs operate on a protocol different than NAS (NAS
integrates into traditional TCP/IP networks and associated network topologies) by using Fibre Channel. Lastly,
SANs offer complete flexibility within their infrastructure by maintaining the intrinsic I/O communications
protocols inherent with directly attached storage.
Whereas NAS remains a retrofit to existing computer networks, SANs offer an entire I/O infrastructure that
breaks the storage boundaries of traditional storage I/O models and moves the support for data-centric
applications into a new generation of computer processing models.
Several things happen when you create a storage network. The storage becomes accessible through a
network model-for example, nodes logging in to the network can communicate with other nodes within the
network. Nodes operating within the network offer a diverse amount of function depending on their device
types-they can be storage devices, servers, routers, bridges, and even other networks. You can transfer data
faster, with more throughput, and with increased flexibility. Managing the resources in a storage network can
be performed from a centralized perspective across sharable domains.
This results in an inherent value to the Storage Area Network. Servers can now share the resources of
storage systems such as disk arrays and devices. Conversely, storage arrays can be shared among the
servers consolidating the number of devices required. This means increased access to centralized data by
large numbers of applications, with support for larger storage capacities through increased ways of providing
I/O operations.
Storage Area Networks are made up of four major parts. As we discussed with NAS, these parts cover the
major areas of I/O operations, storage systems, and supported workloads. In SAN technology, however, there
are more seemingly disparate parts that must be integrated to form an entire solution. With NAS, most of the
products available are offered as bundled solutions. Not so with SANs. When considering SAN infrastructure,
we must ponder more carefully the separate components that make up the infrastructure, because each
operates independently and interdependently with the Storage Area Network they participate in.
Given that consideration, we can discuss and consider the major components of a Storage Area Network and
resulting architecture. SAN components include the following:
Network PartSANs provide a separated network over and above what the client/server or mainframe
networks utilize in connecting clients, terminals, and other devices, including NAS. This network, as
mentioned, is based on the Fibre Channel protocol and standard.
Hardware PartSANs depend on specific hardware devices just like other networks. Given that SANs
provide a packet-switched network topology, they adhere to a standard layer of processing that the
hardware devices operate within. Again, to avoid confusion between data communications packet
topologies, the FC protocol relies on frames and an architecture more compatible with I/O channel
operations.
Software PartSANs operate within a separate set of software called a fabric that makes up the FC
network. Additional software is required at the server connection where the average Windows or UNIX OS
drivers must communicate within a SAN environment. The same thing is true for storage devices that
must communicate with the SAN network to provide data to the servers.
Connectivity PartMuch of a SAN's value derives from its connectivity, which is made up of several
hardware and software functions. Given the complete shift in I/O operations from bus level
communications to network communications, many components must change or be modified to operate
within this environment. Connectivity options within the SAN determine performance and workload
applicability.
Figure 13-1 offers a glimpse into the SAN infrastructure.
Figure 13-1: SAN components overview
The Network Part
SAN is a network and, as such, its main task is to provide communications among devices attached to it. It
does so through the FC standard protocol, denoted by the FC standard and relative T-11 standards committee
that maintains the protocol. The FC protocol offers a layered approach to communications similar to TCP/IP,
but through a smaller set of processing layers. The FC layers are illustrated in Figure 13-2, showing functions
FC-0 through FC-4. (More detail on FC layers can be found in Chapter 16.) However, what's important is that
each layer is leveraged through different hardware components within the SAN.
Figure 13-2: FC layers and functions
What facilitates the network is the switch component that provides a set of circuits making up the
communications paths for the devices attached. The communications paths are accommodated through the
addressing associated with the devices' communications layers. Therefore, the FC switch device can provide
an any-to-any connectivity matrix that allows communications from one device to another. For example, a
server may require a read from a particular disk. The server's I/O request provides the address for the
requested device, the network fabric develops the connection, and that passes the read I/O to the storage
device. Conversely, the storage responds with the same type of process, whereby the storage device
requesting a connection with the server address returns the block of data from the read operation, the switch
fabric makes the connection, and the I/O operation is completed.
It is important to note that one of the particular values of SANs, and specifically Fibre Channel, is its capability
to provide a way for existing protocols to be encapsulated within the communications. This is especially
valuable, as the SCSI commands for disk operations do not have to change. Consequently, disk and driver
operations can operate within the FC protocol unchanged.
However, it is also important to note that in developing the SAN network, additional components are required
beyond the FC switch for both the server and the storage devices. These are depicted in Figure 13-3, as we
see the effects of our read I/O operation within the switch. These components-HBAs, FC switches, hubs, and
routers-will be discussed in more detail in the section titled 'The Hardware Part' in this chapter, as well as in
Chapter 14.
Figure 13-3: A SAN devices overview and sample read I/O
Inside the FC network, the fabric software operates on a frame-based design. Similar to concepts used in
TCP/IP packets, communications within the network are performed by transmitting FC frames throughout the
fabric. The frames are composed of header, user data, and trailer information. FC layers determine the
operations responsible for the frame development, frame addressing, and transmission (see Figure 13-3).
We would be remiss if we did not mention other ways the FC network can be configured. This is important
because many of these configurations are still operating and some offer specific solutions to particular
situations; they're also cost-effective. However, most can be attributed to implementations that occurred prior
to the development of the FC Switched fabric solutions.
These solutions include the following:
Point-to-PointThis uses the FC to connect one device to another. Employed in initial FC
implementations of FC disk arrays, it leveraged the increased bandwidth, but remained a direct attached
solution. Also, tape devices could be used with an FC point-to-point configuration, increasing the number
of drives supported from a single server.
HubThis uses FC to connect in a loop fashion. Basically, the initial network fabric supported an
arbitrated loop arrangement whereby, like SCSI bus configurations, the devices were configured in loop
topologies which not only shared bandwidth but had to arbitrate for network communications, like a SCSI
bus. It leveraged the speed of FC and the capability to place additional disk arrays within the network
allowing connection of additional servers. FC-AL, however, provided latency issues that never fully
facilitated the bandwidth and performance of FC-switched fabric implementations.
The Software Part
The central control of a SAN is contained within the Fabric Operating System functions (sometimes called the
SAN OS, or just the fabric). Like micro-kernel components, SAN fabric utilizes the micro-kernel implementation
as the basis for its operating system.
The Fabric OS runs within the FC switch. Figure 13-4 shows the typical makeup of the Fabric OS. As we can
see, it truly is an effective implementation of a micro-kernel, supporting only the required functions of the
fabric. This is good news for performance, but bad news whenever future enhancements and management
functions are considered. It has similar limitations to the NAS bundled solution-a micro-kernel conundrum
shared among storage networking solutions.
Figure 13-4: A typical fabric OS micro-kernel
Fabric OS Services
The Fabric OS offers a common set of services provided by any network. However, it also provides services
specific to Fibre Channel and I/O operations. These services are summarized in Figure 13-5 and described
next:
Figure 13-5: Fabric OS micro-kernel components
OS ServicesRequired functions configured with micro-kernel, such as task, memory, and file
management, are included within these services. These form the basic OS functions that provide the
core-level processing specific to supporting a fabric.
FC ServicesThese functions interact closely with OS services to facilitate the FC layer processing of
frame management within the switch. The FC services provide both switched fabric and arbitrated loop
modes of processing, as well as frame processing and protocol routing for the complex interconnections
that occur within the switch.
Simple Name ServicesThe FC network works on the basis of a naming convention that identifies
devices (such as HBA port and storage adapter port) by their names as they are attached to the fabric.
These services provide a rudimentary database (for example, a file that supports the SAN naming
conventions, the status of device connections, and so forth) and functions to identify new devices as they
are attached.
Alias ServicesThe fabric provides functions that allow particular devices to broadcast frames to a set of
aliases as noted within the naming conventions (for instance, a generic set of names that one device can
broadcast to).
Management ServicesFC switches that provide startup and configuration utilities allowing
administrators to set up and configure the switch. These utilities range from direct attached access
through a PC attached to a serial interface of the switch to web browser-based access if the switch is
configured with an Ethernet interface. The management functions over and above configuration and setup
consist of status and activity information stored and accessed through a Simple Network Management
Protocol (SNMP) database. This database referred to as a MIB (Management Information Base) is
specific to SNMP and requires this protocol to extract or store any information. Additional information and
discussion of SAN management can be found in Part VI.
Fabric APIs and Applications
FC-switched fabric systems operate similar to other software operating environments by providing a set of
basic services to the supporting hardware (refer to the previous bullet points within the OS services
discussion). However, like any base operating environment, it must have interfaces to the outside world for
access from administrators and other software. These access points will come from application programming
interfaces, or APIs. This software interface will provide third-party software vendors, or any systems
administrators brave enough to try, the specification for writing a program to work with the fabric OS. We
provide an overview of both API and fabric applications below.
Application Programming InterfacesFC switch vendors all include various levels of Application
Programming Interfaces created specifically to allow other software vendors and complementary hardware
vendors to develop applications that enhance the operation of the switch. These interfaces are generally
so complex that it is beyond most IT administrators to leverage, nor is it recommended by the switch
vendors to facilitate local API utilization by customers. Regardless of complexity, this does provide an
important feature which enables software vendors-mostly management vendors-to supply applications
that interface with the switch and therefore enhance its operation.
Fabric ApplicationsFC fabric supports specific applications that adhere to the vendor's API structure.
Applications generally provided by the switch vendor are characterized by their close operation of the
switch through facilities such as zoning, hardware enclosure services, and advanced configuration
features. Third-party tools, offered as products from software vendors, will interface with the switch to offer
mainly management services such as backup and recovery, device and component monitoring, and
storage services.
Server OS and HBA Drivers
Dealing with I/Os generated from servers requires that the server operating systems be SAN-compliant. This
generally means that the OS is able to recognize the FC Host Bus Adapter driver and related functions
necessary to link to SAN attached disks. Although mostly transparent for disk operations, this aspect of the
software dependencies and functions supporting the SAN configuration is becoming increasingly complex as
file and database management becomes more compliant with SAN network functions.
Most system vendors offering operating systems support SANs. However, it should be noted that further
diligence is necessary to identify specific OS release levels that support a specific SAN implementation. Given
the variety of hardware and network devices supported through operating system compatible drivers, the
support for particular SAN components requires detailed analysis to ensure the attached servers support the
SAN infrastructure.
All HBA vendors provide drivers to support various SAN configurations and topologies. These drivers are
matched to a specific level of OS. Given most SANs are available as a packaged solution, it is critical that the
driver software provides sufficient flexibility to support not only the FC switch configuration, but also the
storage arrays and other devices attached, such as routers/bridges, tapes, and other server HBAs. Refer to
the OS macro guide to cross reference the SAN support for a particular configuration. As with the server OS
consideration, more diligence is necessary to identify specific driver release levels that support a specific SAN
implementation and set of hardware components and fabric.
Storage Intelligence
An additional area of software to note is the functional area inherent with particular storage systems. Storage
arrays are all delivered with particular implementations of FC, SCSI, and added utility applications. These
software implementations must be compatible with both the server OS and fabric OS, as well as the HBA
drivers.
Other Networks: As a final note on software, we must recognize the effects SAN fabrics are having on
TCP/IP and their integration into other enterprise networks. We noted previously that the integration of NAS
devices with FC storage components is now available. This configuration does not require a FC switch to
support Ethernet attachment. However, it does not develop into a storage area network despite the fact that
FC storage can be accessed through TCP/IP stacks and compatible file protocols supported by the NAS
device. The capability for a fully functional SAN to participate within the IP network remains problematic.
This also provides a method for switch-to-switch connectivity, thus allowing remote connectivity from one SAN
configuration to another. This should be given due consideration, given its ability to execute a block I/O
operation through an IP network, something which impacts security, performance, and loading. We examine
this in more detail when discussing SAN Connectivity in Chapter 16.
The Hardware Part
The most striking part of a storage area network is its diversity of components. NAS, as we have discussed in
Part III, is almost polarized in the opposite direction given its bundled and simplistic structure. The SAN, on
the other hand, must have a central network of new devices as well as supporting network appendages
attached to the storage and servers it supports. This is most apparent in the required hardware of switches,
HBAs, and sometimes bridges and routers. In support of our discussion on the SAN, the following overview
will further familarize you with these components.
The Fastest Storage Possible: 100MB/sec
The hardware part of the SAN is made up of components that allow a complete storage network to be
enabled. The minimum devices necessary are an FC switch, an FC-enabled server, and FC-enabled storage
systems. Although the FC switch could be replaced by a FC hub to facilitate the storage network, we will
address it separately (in the section titled 'The Unusual Configuration Club, Optical, NAS, and Mainframe
Channels' later in the chapter), given its legacy position and its enhancement of switches to handle both FC
fabric and arbitrated loop operations.
Figure 13-6 illustrates the minimum components necessary to configure a simple SAN. The FC switch centers
the network as it connects the server and storage array. The FC server is connected through an FC Host Bus
Adapter (HBA). The FC HBA provides the necessary FC protocol processing and interfaces with the server's
operating system. The FC storage array is connected through an integrated FC port attachment that injects
the necessary FC protocol communications into the storage controller's mechanisms.
Figure 13-6: A simple SAN architecture
This forms the simplest SAN configuration. However, it is the one that facilitates the FC SAN architecture and
provides the basis for additional SAN configurations no matter how complex. In addition to the basic devices,
Figure 13-6 also shows the type of connections required for implementation into an existing data center.
The Usual Configuration Gang, RAID, Disk Array, and Tape
Figure 13-7 takes our simple SAN configuration and places into a data center setting. As we can see, the SAN
has been enhanced to include additional servers, two storage arrays both supporting RAID levels, and a tape
device connected through a FC bridge. This characterizes a sample configuration that supports all the
necessary data center operations including database, backup and recovery, and shared access.
Figure 13-7: A simple SAN configuration
The Unusual Configuration Club, Optical,NAS,andMainframe Channels
Figure 13-8 takes our simple SAN configuration and provides some interesting twists. Although these are
unusual implementation scenarios, they are important to point out because they show the increasing levels of
flexibility SANs have. Note that there is FC optical storage involved, as well as connection to file storage
systems through an IP connection, and, finally, the connection to alternative servers such as mainframes
supporting FC connectivity, like IBM's FICON.
Figure 13-8: A specialized/advanced SAN configuration
The Connectivity Part
If we further analyze our simple SAN implementation, we find that switch connections come in two forms: the
physical connection and the logical port configuration. The physical connection supports both optical and
copper connections. Obviously, most switch products operate at optimum using the optical connections;
however, the capability to support several connectivity schemes can be important when integrating into
existing cabling structures.
The logical connection requires understanding the type of port configuration the switch provides. Ports can be
configured as particular types of connections. These are defined according to topology and class of
processing. In some switches, these are performed automatically per port, as different types of processing
configurations are defined for the fabric-for instance, Classes 1, 4, and 6 for connection types of circuits, and
Classes 2 and 3 for connectionless circuits or frame switching (that is, packet-switched networks).
Connecting the Server: The Host Bus Adapters
Connecting the server requires the use of a Host Bus Adapter (HBA). This is a component similar to the
Network Interface Card (NIC) discussed previously in Chapter 8 as network fundamentals in storage
connectivity options, and in Chapter 12. It essentially performs the same function as the NIC in terms of
providing a physical connection to the FC network, using either copper or optical connectors. In addition, it
provides the software functions that translate the FC protocol communications and interfaces these
commands with the server operating system. This is provided (similar to NICs) with a set of drivers and is
configured according to the type of FC topology the SAN is operating with.
Connecting the Storage: The SCSI/RAID Operations
One of the most effective architectural design strategies within the FC protocol and FC fabric elements is its
ability to frame higher-level protocols within communications processing. This means that the SCSI commands
that communicate disk and tape operations can continue to operate. These commands are packaged within
the Fibre Channel frame and transmitted. Upon delivery, the FC frame is translated and the SCSI commands
are reassembled for delivery and execution on the target device.
This is not only effective for server operations, which have historically interfaced with SCSI bus technologies in
direct-attached disks, but also for the disk and related controllers themselves. As the SCSI commands are
developed from I/O operation activities within the server OS, the I/O operations at the disk controller perform
as if they continued to be directly connected. However, this begins to become complex as the reference to
logical units (LUN) within a bus are tracked and configured for an FC network. This then becomes a process
that grows more challenging when operating with LUNs that are dynamically modified or hidden by layer 2 or 3
processing-say, in the case of a bridge or router.
SAN Configurations
SANs evolve within data centers from simple configurations using single switch or hub operations with a
limited number of disk devices. These are often limited traffic projects that provide an entry level of operations
and proof of concept for targeted applications. However, they soon grow to configurations with multiple FC
switches that are interconnected to optimize the reliability or availability of the applications. The following three
levels of configurations best characterize the types of configurations IT will become familiar with.
Entry Level
An entry-level configuration will basically be a proof of concept or beta installation to prove the reliability of the
SAN project. As stated earlier, these are often limited projects handling limited user traffic and designed to
support a targeted single application. The entry-level SAN configuration can also allow IT personnel to
become familiar and comfortable with the new types of hardware, give them some experience with the new
software fabric configurations and operation, and let them experience real traffic within the SAN.
Figure 13-9 shows an example configuration of an entry-level SAN. Note this configuration contains a single
FC switch, two storage arrays, and a tape device. There are three Windows-based servers attached to the
switch which complete the SAN network. In this example, anywhere from three to five relational database
systems provide production test beds for application testing. Each database has its aggregate data stored
across the storage arrays. The storage arrays themselves are configured for RAID level 5 and level 1. This
allows several databases to share storage in a single array; in this case, array A using RAID level 5, while
array B is configured for RAID level 1.
Figure 13-9: An entry-level SAN configuration
This configuration provides an entry-level SAN that provides a test bed for three database applications; some
are for typical OLTP workloads while the third is for a data warehouse application. The systems are set to
allow applications and DBAs to test new query applications against test databases. The SAN configuration
replaces the six servers used to support database testing and the six storage arrays directly attached to their
respective servers.
Consequently, even for an entry configuration, the architecture of the SAN can consolidate a considerable
amount of hardware and software.
Mid-Range: Production Workloads, Multiple Switches
If we move up to supporting production workloads, we must consider all the necessary reliability and
availability configuration requirements. Figure 13-10 shows an example SAN that supports production
applications with ten servers using Windows OSs. The servers are attached to three FC switches configured
in what's called a cascading arrangement. The multiple storage arrays support both RAID 5 and RAID 0 in our
example set of applications. Our production configuration also has the addition of backup with the attachment
of a tape library.
Figure 13-10: A mid-range SAN configuration
Our production workload supports several database applications that utilize five of the ten servers. These
transactional OLTP-type requirements utilize the storage arrays configured for RAID 5. Our other workloads
are a combination of an intranet for internal users of the company's internal web application. This application
utilizes the storage arrays configured for RAID 0, which has the data stripped but has no failover capability.
Enterprise: Director-Level with DuplicatePathsandRedundancy
Moving up to the support of an enterprise level and the support of multiple applications, we have an example
enterprise configuration in Figure 13-11. This much larger configuration shows 20 servers, consisting of both
UNIX- and Windows-based OSs, attached to six FC switches. Storage arrays support 2 terabytes of both file-
and database-oriented data models, given that the workloads range from OLTP database applications to web
servers, and data-centric datamart applications. A datamart, incidentally, is similar to a data warehouse with
limited subject scope; in some industry sectors, however, datamarts have proliferated faster given their
singular and more simplistic database design. This allows for faster design and implementation activities,
which makes them demanding in their storage capacity, access and data sourcing requirements from larger
data warehouses, and operational databases.
Figure 13-11: An enterprise SAN configuration
The storage arrays are configured in RAID levels 5, 1, and 0. As with any production workloads, appropriate
failover has been configured with a more complex cascading switch configuration. There is a new twist,
however-a remote connection within the switch that allows for a disaster recovery operation to be in place. The
usual configuration of a tape library to facilitate the backup strategies of various workloads and applications is
also in place.
As demonstrated here, SAN configurations have an inherent value to storage network architecture. Even
though the architecture reflects a new model of processing, SAN configurations provide the most effective and
efficient method of server consolidation, storage consolidation, improved data access, and scalability of data
storage.

Overview
SAN hardware is divided into three distinct parts: the Fibre Channel switch, the Host Bus Adapter (HBA), and
the storage device itself. In addition, one must consider bridging, or routing, solutions. Chapter 13 is the
continuation of the detailed discussion of SANs started in Chapter 12, which covered their specific
architecture. This chapter will explore the hardware components that make up the SAN, a summary of what
they are and how they operate, and a view of fabric operation from a hardware perspective.
The switch part is a discussion of the Fibre Channel (FC) switch component that essentially becomes the
heart of the SAN. Included will be a discussion of the device attachment options, a brief explanation of
operation, and the role they play in the configuration of the switch, consisting of the foundations for
configuration, port identification, and interswitch links.
The FC Host Bus Adapter (HBA) is a necessary component to attach servers to a SAN. HBAs have ports that
connect to the switch, which play a pivotal role in communicating with the SAN fabric on behalf of the server.
There are also considerable differences in HBA software drivers and OS support, as well as options for HBA
functions and their interaction with FC switch.
Perhaps their most important aspect, given that SANs are 'storage' networks, is the FC-enabled storage itself.
Storage devices for SANs require that they communicate using the Fibre Channel networking standard,
consequently their roles and various operations differ in comparison to direct attached storage devices.
Included with this discussion will be an orientation to tape media into the SAN, where we identify an additional
supporting component, the FC bridge/router, which is necessary in developing a fully functional FC-based
storage area network.
We will conclude the chapter with a brief discussion on several important points in the operation of the SAN
fabric. From a hardware perspective, the modes of processing and the functions of the physical ports become
required knowledge when designing and implementing the SAN. Given that the SAN's heart is the FC switch,
the operational aspects of how the switch provides a circulation of data throughout its system cannot be
overlooked. SAN switches will be used in multiple configurations (covered in detail in Chapter 18). This
requires an initial understanding of the switch-linking functions used for expansion, redundancy, and recovery
configurations. Discussions of such can be found in this chapter, as well as in Chapter 15.

The Fibre Channel Switch
The Fibre Channel switch is the heart of any Storage Area Network configuration. It connects the user to the
data being stored in the SAN. Like a heart, all information traveling through this closed system will sooner or
later pump through the switch. Within the switch are connections, called ports, which function in different ways
depending on the type of device connected. There is an effective Fibre Channel standard, called the T11.3
standard, which is the ANSI standard governed by the X3T11 group and defines the usage of the port via the
Fibre Channel protocol. The T11.3 standard provides a set of guidelines for vendors to follow-a common
language, so to speak. The Fibre Channel standard defines three types of ports:
The F_PortUsed to attach nodes. In storage networking, nodes refer to a network logical term for
attached devices (for example, servers, storage devices, routers, and bridges, which are attached to the
switch).
The E_PortUsed to connect one switch to another switch.
The G_PortA generic port that can be pressed into service as an F_Port or an E_Port depending on the
vendor's implementation.
It should be noted that, in general, Fibre Channel switches are categorized in terms of port capacity and their
capability to sustain port failures. Low-end switches, while cost-effective, support only a base set of
functionality in connecting devices to the network; moreover, the ability to connect to other switches or fabrics
is minimal. Low-end switches offer no real port failure tolerance. The class of switches defined for entry level
and mid-range usage is referred to by the generic term Fibre Channel switch. A glass ceiling of port counts
exists that separates the FC SAN switch from the Director level products. With port counts at or above 64,
including internal path redundancies for fault tolerance, the Director class storage switches are used for large
enterprise class storage configurations. Feature function notwithstanding, there is a clear cost difference
between low-end Fibre Channel switches and Director Class switches. The difference is not only clear, it is
quite large.
Unless a user has a high-end requirement, most users start with a handful of low- end switches, as well as the
myriad problems associated with them, and end up very quickly needing to bump up to Director Class
switches. The point is, you may not need the ports now, but you may very well need the functionality
depending on the service level requirements of the applications being supported.
The F_Port
As stated previously, the F_Port, also referred to as a Fabric Port, connects server and storage devices to the
switch itself. A device plugged into the switch's F_Port is referred to as a node, and, in FC terms, is identified
as an N_Port. If used in an arbitrated loop topology, it becomes an NL_Port. For instance, a server, when
plugged into the switch, creates an N_Port, whereas a storage device using an arbitrated loop is recognized
as an NL_Port. A basic F_Port is shown in Figure 14-1. Operating as FC nodes, they are identified by the
switch as their particular N_Port or NL_Port designation.
Figure 14-1: Port configurations
All devices attached to a Fibre Channel switch must log in. When a device is attached, it accesses a file in the
Name Server database within the switch that contains information explaining to the server just what this device
is. The Name Server informs the switch of the device's name, address, type, and class of service. It is the
formal introduction between switch and device. Anyone putting together a Storage Area Network must make
sure the devices they are planning to attach (or are going to attach in the future) are supported by the switch.
If a device is not supported, it becomes all but useless to the SAN. Keep in mind, too, that arbitrated loop login
will take longer than general fabric login or import login, given the extra loop and device assignments.
The Name Server, shown in Figure 14-2, registers a device's port address, port name, node name, class of
service parameters, and the Fibre Channel layer 4 (FC-4) protocols it supports, as well as the type of port. The
Name Server also provides the information one device, like a server, needs to find another device on the
network, like the storage. Port name and port address are contained within a single table so that a device can
be referred to by its name or by its address, much the way a web site can be called up by its URL or IP
address. This is helpful as Storage Area Networks increasingly have IP ports made available to them. The
Name Server supplies a method of recognition between the two types of ports, regardless of the protocol.
When connecting switches, it is paramount that each switch's Name Server be synchronized and consistent
with one another.
Figure 14-2: The name server contains names, addresses, and information on nodes.
Any changes made to the devices attached to the switch are handled by either voluntary or involuntary
services provided within the switch. As devices or connected switches are upgraded, port usage can become
volatile. This is an important aspect in change management within the port itself. How these changes are
made is dependent upon vendor service and implementation. Certain vendors automatically update the Name
Server using State Change Notification (SCN) or Registered State Change Notification, which is an example of
an involuntary change. A voluntary service consists of, you guessed it, using the two preceding methods to
update the Name Server yourself. At this point, you may be asking yourself-and rightly so-just how is that
voluntary?
The E_Port
E_Ports, also referred to as expansion ports, connect one switch to another. These switch ports have become
critical given the rapid expansion of multiple switch configurations needed to support larger and larger server
and storage capacities. E_Ports are now fundamental to the operation of a Storage Area Network and should
be considered carefully when designing and implementing any SAN configuration.
The E_Port, as shown in Figure 14-3, provides the means by which one switch fabric communicates with
another switch fabric. For this to happen, a compatible connection is required. This allows the E_Port to
successfully coordinate with the Name Servers within each switch, synchronize frame transfers between
switches, and facilitate interswitch access to storage resources. Utilizing arbitrated loop configurations
between switches, while possible, increases the overall risk factor in the efficiency of an E_Port because of
the increased overhead and treatment of the NL port as a separate network, which dramatically increases the
number of addressable units and decreases bandwidth utilization.
Figure 14-3: Extending the switch with an E_Port
The G_Port
The G_Port is a generic port and, depending on the vendor's implementation, can be used as either an F_Port
or an E_Port. In other words, G_Ports are universal ports that can be used for a combination of functions.
Although we haven't covered the necessary overhead in implementing a port configuration for issues such as
recoverability and redundancy, G_Ports are likely candidates for recoverability and redundancy usage in light
of port failure.
G_Port populations will dramatically increase, as ports become multifunctional and autosensing becomes
more prevalent. Switches will come equipped with 64 so-called generic ports, and will distinguish themselves
as either F_Ports or E_Ports when a device or an additional switch is attached. This adds critical switch
flexibility when configuring multiple switch operations-all of which increases the cost per port.

Host Bus Adaptors
Simply put, the Host Bus Adapter (HBA) is the link between the server and the Storage Area Network. Similar
to the Network Interface Card (NIC), HBAs provide the translation between server protocol and switch
protocol. HBAs connect to a server's PCI bus (see Chapter 7) and come with software drivers (discussed in
greater detail in the following chapter) that support fabric topologies and arbitrated loop configurations.
HBAs are available in single port or multiple port configurations. Multiple ports allow additional data paths for
workloads moving between the server and the switch via a single HBA. In addition to containing multiple ports
(a maximum of four, at this point), a single server can hold, at most, four HBAs. Today, any single server can
possess 16 ports (four ports times four HBAs), or 16 separate points of entry into the switch. However, keep in
mind that four discrete ports on one HBA increases the risk for single point of failure along those data paths.
In providing the initial communication with the I/O from the server, HBAs encapsulate SCSI disk commands
into the Fibre Channel layer 2 processing. HBAs communicate within the FC standard through the class of
service defined by the Name Server at the time of login. As such, the HBA plays a key role in providing levels
of efficiency in executing the operating system's I/O operations. Figure 14-4 illustrates an HBA's basic
functions.
Figure 14-4: The basic functions of the FC Host Bus Adapter
One key duty of any HBA worth its bandwidth is discovering and mapping the storage resources available to it
within the switch fabric. This mapping is critical to the devices that are available to any particular server. As
will be discussed in the next chapter, there are various ways to restrict access to storage resources, such as
zoning. It is important to note here, though, that HBAs must deal with this issue in an effort to understand the
devices it must contact on behalf of the server, which it ultimately works for.
As you might expect, all of this becomes even more complex in supporting arbitrated loop devices on the
switch.
Depending on the vendor, additional functionality is bundled with different HBAs. These functions range from
software RAID functionality to advanced management functions (for example, diagnostic functions and the
new enclosure services provided by many vendors). In the future, additional functionality will come to include a
virtual interface that bypasses much of the layered processing currently required. For more information on this,
check out the discussion on InfiniBand in Chapter 20. But don't say I didn't warn you.
The major reliability questions of any HBA are twofold. First, the HBA's compatibility with its server's operating
system is key to the effective operation of the FC network in total because each operating environment has
unique differences in how it handles base I/O, file system, and buffering/caching methods. It is important to
understand at a macro-level the differences between a UNIX integration and implementation of an HBA versus
integration and implementation within a Windows environment. The second factor is the compatibility, at a
software level, of the switch's fabric operating system and the HBA's software drivers. This requires an
understanding of the supported release levels of any given switch vendor against any particular HBA vendor.
Since many SAN components are acquired through OEM relationships, compatibility can become sticky when
intermixing equipment from different vendors. As shown in Figure 14-5, HBAs play a critical role in the
interoperability of a SAN configuration.
Figure 14-5: HBA's critical role in interoperability

Putting the Storage in Storage Area Networks
Given that SANs are Storage Area Networks, perhaps the single most exigent component in the configuration
is the storage itself. SANs require storage that is Fibre Channel- enabled. In other words, FC disks must be
connected to the switch via an FC topology. Fibre Channel-enabled disks come in two configurations: JBOD,
Just a Bunch Of Disks, and RAID, Redundant Array of Independent Disks. Each configuration maintains its
own set of FC features and implementations.
FC storage arrays, which are multiple disks hooked together, are connected in a loop configuration, as
depicted in Figure 14-6. Loop configuration allows each disk to be addressed as its own unique entity inside
the segment. FC storage arrays (depicted in Figure 14-6 as well) also provide additional functions, such as the
stripping of data across the units, which allows a single outsized file to be spread across two, three, or even
four drives in the array. In addition, most provide low-level enclosure management software that monitors the
device's physical attributes (its temperature, voltage, fan operation, and so on).
Figure 14-6: An FC storage array connected in a loop configuration
JBOD
A typical JBOD configuration is connected to the switch through an NL port. Most implementations provide
dual loop capacity whereby redundancy protects against single loop failure. In other words, should one loop go
down, the information on the storage device can be retrieved via the second loop. A dual loop requires four of
the switch's ports. Another, less typical method of attaching a JBOD array to a switch is to split the devices
into separate loops. A JBOD array of eight drives could have one loop serving drives 1-4 and a second loop
for drives 5-8. This method also requires four ports on the switch. The benefits of splitting the devices into
several loops include shorter path lengths and less arbitration overhead within the loop itself.
The disadvantage of any JBOD implementation is its lack of fault resiliency, though there are software RAID
products that allow the stripping of data with recovery mechanisms encompassed in the JBOD enclosure.
Given the number of disks and the transmission of SCSI commands to multiple targets, using RAID software
in conjunction with a JBOD implementation presents its own problems. It is important to keep in mind that
while these are Fibre Channel-enabled disks, the disk drives themselves execute a SCSI command set when
performing read/writes.
RAID
A Fibre Channel-enabled RAID storage array places a controller in front of the disk array that provides and
controls the RAID level for the disk array (RAID 1-5). RAID offers a way of protecting the data by creating an
array of disk drives that are viewed as one logical volume. Inside this logical volume, which may consist of
seven drives, data is partitioned, as is recovery information. In the event of a single drive failure, the array
reassembles the information on the remaining drives and continues to run. The RAID controller uses either an
N_Port or a NL_Port depending on how the vendor put the disk enclosure together.
Because the RAID controller stands in front of the array, the FC-enabled disks, regardless of their
configuration, become transparent. The switch only sees one device, not several linked together. A single
RAID controller can also control several arrays, as indicated in Figure 14-7. For example, four volumes, each
volume containing seven disks, would equal a RAID system of 28 disks. Still, the switch only sees the RAID
controller-just one device. In this scenario, more often than not, the RAID controller will utilize an N_Port, not a
loop.
Figure 14-7: An FC RAID storage configuration
The key advantage of Fibre Channel RAID is its ability to provide levels of fault resistance for hardware
failures, a feature not found in JBOD configurations. For enterprise-level workloads, this feature is all but
mandatory.
Just as the HBA is the critical point between a server's operating system and the switch's operating system,
RAID controllers require a level of specific Fibre Channel software that must be compatible with the switch and
the HBA. As noted previously, it is the HBA's job to inform the server which disks are available. In a JBOD
configuration, this is pretty straightforward. Each disk is an addressable unit. In a RAID configuration, it
becomes the controller's duty to specify which disk is addressable. The RAID controller, via the software
contained within it, has to identify itself to the switch, specifically the Name Server within the switch, as well as
the HBA, in order for the server to know which disks it has access to on the network. This can quickly become
confusing, given that RAID deals in logical units, not independent addressable disks.
One last word about Fibre Channel storage, and this goes for RAID and JBOD configurations: when assigning
multiple servers to the switch (via the HBA), the servers have to be told which storage resources they are
allowed to play with. And this can quickly become tedious. For example, each server has its own file system,
and that file system must reflect the location of the files the server has access to. Problems arise when two or
more servers have access to the same files. What happens when two servers reach out for the same file? You
guessed it trouble, headaches, and a whole lot of shouting. File sharing between servers attached to a
Storage Area Network remains a tedious and problematic issue. Consequently, zoning and masking each
server's authorized resources continues to be a prerequisite for effective operation. Before you start shouting,
and reach for the aspirin, have a look at Chapter 22.
Bridges and Routers
In addition to the challenges posed by file sharing, data sharing, and device sharing, there are standard data
center practices that are required for any type of storage model. Every data center must protect the data
stored within the arrays. Most accomplish this using backup/recovery software and practices, which entails the
use of tape devices as the primary media for archival copy and data copy functions. In a SAN environment,
however, this is easier said than done. Fibre Channel-enabled tape devices have been problematic in their
support of this new storage model.
To overcome this hurdle, a new type of device was required to bridge the FC protocol into a SCSI bus
architecture used in tape media. Because of tape technologies' sequential nature and the resulting complexity
entailed with error recovery, tape media has been difficult to integrate. Solving these difficulties required a
device that not only bridged the Fibre Channel into the tape controller/drive bus system, but also further
required the management of the logical unit numbers (LUNs) that were utilized in the tape's SCSI
configuration. The solution was found in bridges, or routers, for Fibre Channel. This is illustrated in Figure 14-
8.
Figure 14-8: A simple bridge/router configuration
Although they are compatible with other SCSI devices, routers are primarily known for their capability to
facilitate the operation of tape media within a SAN. Routers provide an effective means of establishing a tape
media library for SAN configurations. The alternative would be to copy data from the FC storage arrays onto
the LAN and shoot it off to a backup server with a directly attached SCSI tape drive. Considering the overhead
required, routers provide a much sleeker configuration for data protection. As always though, even the sleek
solutions have their drawbacks. Fibre Channel to SCSI routers bring their own performance issues in matching
the switch's port performance with the SCSI bus attachment at the other end of the bridge. Due to SCSI
transfer rates, speeds across the router will be constantly slower and throughput will be compromised.
When integrating a router solution, it is important to understand what is needed, whether it's a discrete
component or one of the integrated solutions that are increasingly creeping into tape subsystem products. In
looking at either of these solutions, you'll have to distinguish other protocols besides SCSI that may be
necessary to bridge into your SAN configuration. Router technology is beginning to move toward a gateway
type solution where Fibre Channel integrates additional I/O and network protocols.
An additional word of caution: routers add yet another level of microkernel operating environment that must be
compatible with all of the components across the Storage Area Network. Routers must also be compatible with
a significant number of tape systems, which only adds to the complexity of implementation. The surest bet is
to approach the integrated solution supported by the tape manufacturer.

Fabric Operation from a Hardware Perspective
The SANs fabric operates through, and is governed by, the logical components of the Fibre Channel standard
previously mentioned. The standard is broken down into logical constructs of frame, sequence, and
exchanges. FC provides its storage flexibility through the ability to transport other protocols, such as the SCSI
protocol.
FrameThe logical construct of transporting data through the switch.
SequenceA block of numbered or related frames transported in sequence from initiator to target. Error
recovery takes place as the receiving port processes the sequence.
ExchangeA number of nonconcurrent sequences processed bidirectionally (it should be noted, however,
that a port can only process one sequence at a time). Although a single port is limited to single sequence
processing, it can process multiple simultaneous exchanges.
Within the overall FC protocol standard are embedded the primitive sequence protocols used for error
recovery during exchange processing. Fabric login is used to communicate the ports operation
characteristics. Operations are directed by the N_Port login and logout commands, which provide session-
level parameters as ports issue and exchange sequences. The exchanges are made up of frames (as shown
in Figure 14-9) and indicate the large user data area or payload that allow FC protocols to transmit at gigabit
speeds.
Figure 14-9: Anatomy of the FC Frame
Selected by the port at login, the class of service and other parameters the switch will support directs flow
control of the frames. Flow control is determined by connection attributes between communicating devices.
This process is similar to network operations where connections are either mindful of their state and
acknowledge transmission and receipt, or disregard their state, acknowledgment, and effective frame
recovery. Although there are six classes of operation defined by the FC standard, only three are useful to data
storage operations, and therefore worthy of mention.
Class 1A dedicated connection in which two devices establish complete control and dedicated
bandwidth of the link. Until they complete their transactions, the ports remain busy and unavailable to any
other traffic. Although useful for batch type operations and data streaming when the complete bandwidth
of the link can or must be used, Class 1 operation performs best in transactional-intensive workloads like
OLTP, web, or data warehousing.
Class 2A switched bandwidth sharing connection (considered connectionless) providing a form of time-
slicing algorithms that share bandwidth, thus enabling multiple exchange activities to take place. Although
Class 2 service does have acknowledgment of frame delivery, frames do not have ordered, guaranteed
delivery, and thus must wait their turn in the linking schedule to perform the reordering of frames within
the routing mechanism set up by the vendor. Class 2 (illustrated in Figure 14-10) demonstrates its value in
typical application I/O workload situations. It can handle the transaction rates an OLTP or time- dependent
system might throw its way.
Figure 14-10: Typical flow control using Class 2 operation
Class 3A connectionless buffer-to-buffer flow mechanism that offers a quick way to broadcast messages
within a multicast group. However, cutting out the recipient acknowledgment and buffer coherence means
introducing a high risk of frame loss. Class 3 works well for specialized applications that broadcast
requests to devices. Without the retransmission of the request in other class processing, this limits the
useful value to simple data replication and browsing/ searching acknowledgment.

SAN Hardware Considerations
As you might have gathered by this point, SAN hardware is made up of multiple components with many
variables, specifications, and options unique to each. Most hardware decisions are made based on the
features and standards supported by a particular components vendor. It is important when heading into a
purchasing decision that you eliminate as many unknowns as possible.
Of the billions of things to keep in mind, first and foremost will always be ensuring that total I/O workload
matches SAN resources. (Chapter 17 lays the foundation for I/O workload planning, while Chapter 18 sets out
some guidelines in estimating basic SAN resources.) Consider the following.
Match server hardware- and software-supported releases with proposed storage and SAN configuration
components. This will ensure the correct operating system level, HBA, and switch fabric release levels for
compatibility. Also, verify the compatibility and features of the switch against the type of I/O workload you are
supporting. Ensuring you can configure Class 2 support, at a minimum, will be important as you implement the
workload processing into the environment.
It is paramount that a planned SAN configuration be compatible with the operations you are planning. If you
are going to use an external tape library, ensure sufficient port and network resources for transmitting data
across the SAN and LAN network. If instead you are going to employ an integrated tape library, understand
the compatibility issues surrounding the additional equipment. Whether integrated or discrete, routing
hardware must come under the same scrutiny as every other component in the configuration.

Overview
The era of client/server computing ushered in the challenge of dealing with multiple operating environments.
The need for operating systems to communicate has only grown and become more sophisticated. As
discussed earlier regarding NAS hardware and software, its architecture of effective OS communications is at
the heart of what it delivers. Fortunately for IT professionals, the NAS devices were efficient, encapsulated
product bundles and didnt require an in-depth configuration integration effort with other operating
environments. This isnt true of SAN technologies (as Chapter 14 illustrated) with their various hardware
components.
This chapter will cover the multiple software components that make up the SAN, a summary of what and how
they operate, and a view of fabric operation from the software perspective. At the forefront of this discussion
will be the FC switchs operating system, a discussion of the microkernel-oriented set of software services that
provides the base functionality for creating the SAN network and switched fabric of interconnections.
The device driver discussion will cover the necessary software drivers that must be available to allow the
external node devices to connect to the fabric. Their functions, and how they operate, are integral to the
complete configuration of a SAN. Due to their distributed nature, their considerations for design and
implementation are dependent on server, storage, and fabric software components.
The most important is the SAN storage software part, given that the entire value of the SAN components
leverage a stronger storage position. The SAN infrastructure provides a transparent implementation of most
storage media, especially in the area of disk storage arrays and the use of external controller functions
necessary in RAID and tape media functions. These include RAID recovery functions, internal array caching,
and fault-tolerant features that are integrated with the storage component. Their interaction with the switchs
OS, as well as the FC HBA drivers, are critical to the effective configuration and operation of the SAN.
Supporting components part: Additional features inherent with the SAN, but which require additional
components to leverage, are data sharing, device sharing, management, and special routing software
functions. Although supplied by third-party vendors, the milieu of interrelationships with the Fabric OS and
device drivers becomes problematic when implementing these advanced SAN functions.

The Switchs Operating System
If a Storage Area Networks heart is the switch, continuously pumping data between the storage and the
servers, then the networks brain is the switchs operating system. Switch operating systems are similarly
configured to the microkernel RTOS used in NAS solutions. In fact, the operating systems, in many cases, are
a derivative of the same technology used by certain NAS vendors. Switch operating systems, regardless of
implementation, operate with the same core-level operating systems that support other embedded options.
Figure 15-1 illustrates the basic components of a typical switch operating system, the foundation of which is
the core-level, or kernel-level system calls that provide basic system-level functions, such as interrupt
processing, memory management, and task prioritization. The ability to execute Fibre Channel commands
through low-level system calls to the FC chip-set is the specialization encapsulated in the switch, given that
microkernel performance of the switchs operating system is paramount to the successful operation and
effective utilization of switch hardware.
Figure 15-1: The basic components of the switchs operating system
There are two major issues that impact the operation and performance of the switchs OS. The first is the
extensibility of the systems functions to successfully execute existing functionality while being able to
encompass new levels of applications which have become increasingly important to the operation of the SAN
as a whole. Surrounding this level of compatibility and scalability is the microkernels capability to coexist
alongside a commodity-level chip and Fibre Channel microprocessor. Consideration must be given to the
addressability of the operating environment and its ability to multiprocess and multitask the network linking
functions of the switch. The second issue is the operating systems capability to provide reliable yet efficient
Name Service functions for intraswitch operations and interswitch communications. The inherent danger in
mixing and matching different vendors switch components should be duly noted, and strongly heeded.
Figure 15-2 illustrates the importance of add-on applications that must also utilize the operating systems
resourcesfor example, recovery services, management services, external API, and communication
components.
Figure 15-2: SAN applications running within the switchs OS
The following list discusses the additional components that process under the control of the fabric operating
system. The specific processes will be implemented differently depending on the vendor selection of FC
switch products; however, the common functions as explained here should be included to achieve a complete
operational SAN fabric.
RecoveryCore-level services that provide specific class processing within the switch, including link
recovery, sequence retransmission, and frame recovery of, specifically, class 2 processing. This also
includes class 3 and 4, not to mention the added value functions related to Fibre Channel that enable
disk-to-disk communications. These leverage node functions from layer 3, 4, and upwards, along with the
extended copy command that allows devices to communicate without a third-party supervisor. As you
might have guessed, recovery functionality is essential. For additional information, see Chapters 21 and
22.
ManagementNot an easy one. The big difficulty with any embedded system is its inability to be seen
within the environment it operates. SAN configurations are just about the least visible to existing server-
level applications, so it falls to the switchs operating system to provide some level of visibility. This is
accomplished through the use of a Management Information Base (MIB). Given its network-centricity,
MIBs, and its subsequent access method of Simple Network Management Protocol (SNMP), are the
foundation components monitoring the SAN. These unwieldy components are difficult enough within a
network environment, and have proved to be just as obtuse in storage network environments. As hard as
they are, though, MIBs have provided management vendors and those users intrepid enough to brave the
waking nightmare that is SNMP an opportunity to gather information about the operation of the switch. So,
choose the fabric component, but please make sure the operating system is compatible with third-party
management tools. Please
External Application Programming Interface (API)Most switch operating system implementations will
make some accommodation for an API to facilitate third-party programs that enhance the operation of the
SAN. APIs include data sharing software, management software, and communications software.
Communication componentsA way to get from the switch to the outside world. This is yet another
important duty of the switchs operating system, given that port communications are encapsulated within
core-level functions, outside communication to IP networks, serial communication to console functions,
and the leveraging of E_Port switch-to-switch communication. The switchs capability to direct these
functions is dependent on size and scalability. IP overhead communication through a G_Port provides
sufficient latency to affect the performance of the switch. E_Port interswitch communication stacks
additional processing cycles that subtract from the core functionality of the switch itself. As a result,
console configurations through a serial port should have sufficient low-level prioritization to interrupt kernel
functions in order to reconfigure or solve a problem.
Taking all of this into consideration, it becomes plain that the concurrency of port-to-port communication,
outside communications, switch-to-switch communications, and the latency factors for the microkernel can,
and will, quickly become overwhelming.
At the end of the day, the microkernel operating system is not likely to handle a lot of other work on top of what
its core-level job is, which is the linking of ports. Low-end, generic switches of 8 to 32 ports arent likely to
support additional data sharing, management, or switch-to-switch latencies without taking away from core-
level functionality. These switches are designed to provide a switch fabric set of operations for hooking
switches to storage, and thats it. Microkernel operating systems just werent designed to be general operating
systems. So why are we asking them to be general operating systems?

Device Drivers
Communicating externally with the switchs operating systems requires the use of software drivers that allow
the devices to operate as either N_Ports or NL_Ports. Each type of device requires a specific kind of driver to
enable this functionality. In a Storage Area Network, the two big ones are the HBAs interface with the switch,
and the storages connection.
As discussed in Chapter 14, the HBA provides an FC-interface from a computer device, typically a server, to a
Fibre Channel switch. In addition to the hardware adapter that must be physically connected to the servers
bus interface, the software contained in the HBA must translate the file I/O request and device command set
into Fibre Channel commands. For this to happen, software drivers are installed on the system that execute as
device handlers at the kernel level, requiring a change to the ROM configurations, as well as additions to
recognize the new interface that exist within an HBA. In essence, an application takes SCSI commands found
for particular file destinations and translates these commands into specific requests through the FC protocol
as it logs in to the switch and executes the appropriate services.
HBAs are critical for two reasons. First, for their capability to shield the server from the specifics of the
addressing encapsulated in the Fibre Channel (not to mention in managing the actual physical and logical
units theyre authorized to use within the fabric). The HBA maintains the logical unit numbers (LUNs) for the
devices it can access. It works in tandem with the operating system to determine the bounds of the devices it
has access to, thereby initiating the first level of device virtualization to the server.
The second critical function is providing the server with an effective transmit and receive data path into the
switch, which supplies a way of leveraging the flexibility and bandwidth of the FC network it is connected to.
Figure 15-3 shows both the logical and physical functions that are controlled by an HBAs device drivers.
Figure 15-3: The HBA driver functions
Given an HBAs critical proximity to the I/O path, performance is a direct result of the addressability of the
servers bus interface bandwidth. Providing a high-end HBA interface to a low-level bus wont buy you much
performance. Similarly, hooking a legacy bus into an HBA is only asking for trouble. The additional features
found in todays HBAs parallel much of the added functionality mentioned earlier in our discussion of the
switchs operating system. Some of these features are unique to the device drivers themselves, including
software RAID, support for virtual interface protocol (see Chapter 8), and service features like enclosure
management.
Fibre Channel storage arrays, discussed in the previous chapter, provide yet another embedded system for
consideration when it comes to compatibility, functionality, and servicing. The big hardware vendor feature
offers the capability to differentiate on the basis of software, which means that all storage arrays, whether
JBOD, RAID, tape, or optical, have some level of secret sauce that must be downloaded in order to gain
bonus levels of functionality. In the Fibre Channel arena, it becomes super challenging in just trying to access
the storage array through the FC network on a native basis. This is a required step in configuring either JBOD
or the RAID controller with the proper software (were really talking firmware here). As a result, another
configuration utility must be used against the storage array as it becomes implemented. Having said all this,
never fear, once you work through this a first time, the process can generally be proceduralized into the
functions and servicing of the SAN. Now, heres the fly in the ointment. Each vendors implementation is
different. A heterogeneous storage array installation, as you might have guessed, can become mired in
inconsistencies, incompatibilities, and 25 different ways of doing the same thing.
RAID configurations pose the biggest problem, but also provide the biggest buyback. Given that RAID
software shields, or manages, the local devices it is connected to, it directs through its parameters read/write
operations within the disk subsystems. This is an example of level 2 virtualization. Coupled with level 1, which
has already occurred on the HBA, were then forced to contend with two levels of virtualization. Even after
configurations, care should be taken as to the level of logical unit numbering that occurs when defined within
the RAID configuration chosen.

The Supporting Components
Even with the previous components of switch OS and its multitude of functions, there are yet additional
supporting components that have become increasingly important to the SAN software. The following are
critical to SAN configuration and operation:
Data sharingStill problematic because a global file system operating across a set of servers doesnt yet
exist. Although some existing software products can be employed, they provide an increased level of
latency in switch processing and external server processing. They also require additional hardware
resources which provide a centralized locking mechanism.
Device sharingLess problematic than data sharing, device sharing can be used effectively to some
degree in homogenous environments (for instance, in all Windows servers). By allowing the devices to be
controlled by the operating system, they will not be able to communicate altogether effectively.
ManagementThis continues to evolve as a discipline. Because SANs are still a new technology, no
ones quite sure what to manage, or what to look for, actually. Problem uniqueness, driven by the
overwhelming configuration options, thus makes management a site-specific discipline.
ISL functionsThe Achilles heel of SANs. This type of implemented configuration drives interswitch
communication. Core-edge configurations, meanwhile, demonstrate different switch-to-switch
communications than cascading configurations. As additional hops are needed from switch to switch,
processing latency builds, and performance becomes non-linear with respect to resources applied (see
Chapter 1).
Device Sharing
One of the first dilemmas faced in working with SAN hardware is sharing devices within the fabric. Device
sharing frustration rivals the initial learning curve of setting up network fabric, connecting the devices in the
SAN, or even coming up with a suitable cabling strategy. Device sharing requires the knowledge of switch
software, HBA parameters, and OS storage access.
The problem boils down to sharing devices between operation systems. NAS environments provide this level
of sharing because they separate the native I/O from the application infrastructure (see Chapter 20 for
additional information on the differences between SAN and NAS I/O operations). This separation does not
happen in SAN configurations. The servers communicate through native I/Os (say, block I/O requests) with
the storage devices. This makes the switch, and the switchs operating system, the fundamental global access
supervisor in which the storage administrator is used as the key in establishing who can access what.
Although this could be construed as a hardware function, device access within the switch fabric is handled
through the configuration parameters of the switch OS. This is known as zoning, a control function that
specifies at switch startup time and device initialization time what devices the node login can access. This
happens logically through a set of port addresses associated with a particular server. When implementing a
SAN configuration, careful thought should go into planning the whos who of SAN storage access. Most SAN
switch vendors recommend that the user develop a matrix of server-to-storage access required for SAN
design and implementation.
Figure 15-4 shows the zoning function in partitioning the storage arrays within a simple SAN structure. Sever A
has access to devices on storage array prod and test. Server B, however, can only access QA, with
shared access for both occurring through the test array. Zoning, at what is considered the hardware level,
restricts the access of one device to another, and becomes further complicated when individual devices within
the arrays need to be zoned within the configurations.
Figure 15-4: Storage device segregation through port zoning
The zoning process requires the mapping of LUN addressing at the server (and, by default, the HBA) level, the
array level, and the bridge/router lever, which only makes the whole process that much more complex. Further
LUN zones take place within existing zones, something called LUN masking, which masks off access of
specific LUNs to the nodes. Operationally, this provides a lower level of device sharing and partitioning within
a storage array, or in the context of a tape unit. This process is shown in Figure 15-5, where zoning with LUN
masking accommodates further sharing of the QA array along with the installation of the tape drives, which
can be used by both servers, though only with specific tape drives.
Figure 15-5: SAN configuration with zoning and LUN masking
Data Sharing
Device sharing naturally suggests data sharing. NAS allows a level of sharing because it separates the data
from the actual I/O, allowing ownership and access to the data. SAN, however, as we have discussed,
provides native access to the data from several servers. As such, data sharing is problematic, to say the least.
The major reason is that within a SAN software environment there is no global supervisor to monitor and
control access to shared data. As Figure 15-6 shows, through unrestricted access or liberal zoning, two
servers can access the same data block for write access, which spells significant data integrity trouble.
Figure 15-6: SAN configuration using zoning and masking
Data sharing in a SAN environment is tricky, and, depending on the server operating environment, can be
assembled using multiple components of the server OS, switch-zoning functions, and LUN masking. This level
of complexity within a heterogeneous server OS-SAN environment can be challenging and problematic, given
the differences of file access between UNIX and Windows-based operating systems.
There are a handful of commercial software solutions that provide the psuedo-global file manager to monitor
and control shared data access. Although these products are becoming more robust, mature, and reliable,
they supply an overhead to the processing requests that are subject to data sharing restrictions. Figure 15-7
shows how these products operate in an out-of-band processing mode with the SAN.
Figure 15-7: Out-of-band processing
SAN Management Applications
While device sharing and data sharing could be considered management applications, they are directed from
both the switch OS as well as from external devices, such as attached servers and contollers. These
applications run on external devices and only communicate with the switch to obtain specific data for their
processing requirements.
This also characterizes a SANs management software components, which are categorized from an external
and internal perspective of the switch. Management components are considered out-of-band when processed
external to the switch and in-band when their functions process within the switch. Both sets of software,
whether in- or out-of-band, access the switch OS for their particular function. As you can see, management
functions are quickly becoming imperative to the effective operation of the SAN.
In-Band Management Functions
In-band management functions are distributed within FC switch application functions. However, several
provide base functionality, including activities such as configuration and maintenance. Other functions allow
add-on type applications, like performance, change, and problem management. Several of the functions
generally available through in-band management applications are discussed next. Keep in mind that vendor
selection and third-party availability play a key role in what is actually configured in the switch or SAN product
set.
Console or Workstation Access Console, or workstation, access is the initial entry to the switch OS,
requiring a direct connect console or workstation, and is used in communicating changes through the OS. This
interface provides a base-level entry into the initialization and configuration of a SAN fabric, as well as the set
up of switch parameters and operating specifications.
Configuration Utility Defining fabrics, assigning worldwide names, domain names, port ID addresses, and IP
addresses are just a few of the initial activities performed through the configuration utility console. Others
(although vendor-dependent) include license verification, ISL operation and verification, and FC fabric
validation.
The Console Boot Process The console boot process is very important, unless youre installing or operating
a single-point SAN. The order in which you reboot the switch operating systems is critical. Most vendors
provide some form of boot utility that allows the specification of boot sequencing. Also, given the critical nature
of console boot processing, the configuration must be able to provide alternate paths for redundant
configurations, such as core/edge installations.
Maintenance Loading new firmware releases will exercise boot processes, given that the operating
environment of the switch is no different than any other computer. Most configuration utilities have an
integrated function that loads new releases of the switch OS, service patches, and other new or upgraded
functions. In some cases, this may be a separate function of an application accessed through the console.
However, this brings up an important point of change management, one that is central to OS maintenance in
general, and imperative to SANs. The sophistication of this type of management application is just beginning
to evolve and change. Management is largely a manual tracking process, albeit an important one (see Chapter
22 for a more detailed discussion of management applications that affect SAN availability).
Monitoring Functions Monitoring functions provide a window in which to view the performance of the SAN
components. These same functions can also be accessed through the console and configuration utilities. Like
change management, dont expect much. The ability to view port operations is usually on an on-off status
level, which can also be done by looking at the port status LEDs. Port activity, in terms of I/O workload
performance, is left to third-party tools that process on an out-of-band device. However, ISL operation and
status functions can be found with most switch in-band applications. A separate discussion later in this
chapter, in the section titled ISL Functions, will cover ISL components.
Out-of-Band Management Functions
Managing the SAN externally will require functions and processes that are executed from attached devices.
These, for the most part, will be enabled from attached servers, although some supporting functions will be
necessary for functions that bridge, route, or provide an external process. The following discussion highlights
some of the major functions that should be considered in design and configuration of SAN solutions. These
become critical once the SAN configuration becomes operational.
Configuration Management Configuration management is perhaps the most widely available function, even
though its probably the one that does the least. It scans a name server and configuration files on the switch
and renders a picture of the total SAN environment. Given additional sophistication, the SAN diagram can
become an active diagram in providing a rear view elementary monitor function based on information from
the switchs MIB. Added sophistication or no, it is still a rear-view, so theres not a whole lot you can do
having used it. (Imagine trying to drive using only your rear-view mirror.) The ability to shield the SAN
administrator from the complexities of the line-command configuration utilities or the obtuse nature of the
Windows-based GUI is still evolving. Its great for having a pretty picture of the SAN to show your boss, or
downloading it to his PC so he can show his boss.
Performance Management Like the other five management disciplines, performance monitoring of the SAN is
very elementary. Given the complexities of the switch configuration, fabric software, and diversity of device
attachments, this is a difficult task, one that will require additional development and maturity within SAN
technology and vendor areas before reaching a real-value proposition. However, I remember what the old IT
manager once said about availability of systems information, In the land of the blind, the one-eyed man is
king. Given that analogy, any insight into these complexities will provide a piece of the puzzle, which will help
accurately manage performance. However, be prepared to manually correlate the information from the
application on down to the device.
NoteSystems management has been categorized into five management disciplines that are used to
maintain computer systems regardless of vendor class or technology. These are performance,
change, problem, configuration, and capacity management.
Storage Management Many SAN-specific storage functions that interact with the switch and related software
are becoming increasingly availablefor example, storage applications that provide device and disk
management solutions like volume managers, disk utilities, and failover solutions. This type of software, in the
case of volume managers, executes on the server and interacts through the switch to manage disks into
volumes, providing a level of virtualization for applications. Other software products manage the
backup/recovery processes that handle data protection strategies. They can also execute on an attached
server, and leverage a device-to-device copy operation. Known as server-free or server-less backup, such
operations allow the backup application to turn control of the copying operations over to an FC device such as
the FC switch. Without the extra resource utilization normally imposed by the controlling servers involvement,
the results are a highly optimized data backup operation (see Chapter 22 for more on backup/recovery and the
extended copy operation).
Considerations for Integrated Management Functions
Heres the problem. Each application running as an in-band operation on the switchs operating system takes
away from the processing dedicated to the operation of the switch, as well as the movement of data within the
fabric on behalf of the attached devices. Likewise, any time an out-of-band application accesses the switch, it
too takes away from processing power. As Storage Area Networks are very often built with entry-level
switches, and continue to experience growth through incremental switch upgrades, supporting applications
can easily overrun the switchs operating systems capacity to perform, even at minimum levels. Additionally,
as out-of-band applications become more sophisticated and contact the switch with increasingly complex
requests, major latency issues are going to pop up with greater frequency. Left unattended, this will put the
SAN into a gridlock of overhead versus primary processing responsibilities.
ISL Functions
Interswitch linking (ISL) functions are integral to implementing scalable and highly available SAN
configurations. ISL functions are a combination of hardware and software functions, though software
functionality quickly becomes the predominant component in enabling switch-to-switch connectivity. Be
warned though, ISL functions continue to limit the heterogeneity of configuration flexibility and design due to
the proprietary implementation employed by various vendors. The bottom line is that highly scalable
multiswitch SAN configurations are best built from homogeneous products.
ISL functions provide a communications link between switches and fabrics, which allows switches to be
configured to meet the needs of the I/O workloads they are supporting. Simple cascading connections provide
a way of attaching multiple device nodes within an extended fabric, or among disparate fabrics. Figure 15-8
shows the ISL functions required to support a three-switch configuration. The servers I/O requests for switch
A are processed through the switch it is directly connected to. If the server requests data from storage array
Prod2, though, it has to connect through switch A, then reconnect to switch B through ISL 1 port. The I/O
request is then serviced and the response from the Prod2 storage array is sent back along the same route,
using ISL 1 to return to the server via switch A.
Figure 15-8: ISL functions supporting a simple cascading switch configuration
Taking this a step further, switch C can be accessed via switch A through the ISL 2 port. This configuration
continues to expand using ISL functions to provide additional access to storage device nodes. Additional
operational characteristics to this switch configuration include dynamic switch configuration, cross connections
(also referred to as trunking), fabric recognition, ISL device recognition, and performance.
Many switches support dynamic configurations, where, upon initialization, the switch automatically recognizes
and logs in attached device nodes and E_Port ISL functions, thereby identifying the ISL connection as well as
the devices connected to both.
In many cases, the use of an ISL facilitates the basic connectivity to other node devices, but may not possess
the necessary performance or failover requirements for the workload. Should this arise, multiple ISL functions
are paired to facilitate throughput requirements and act as a backup if one ISL port becomes inoperative. This
is illustrated in Figure 15-9.
Figure 15-9: ISL supporting higher performance and failover
The final factor is the switchs capability to dynamically recognize storage devices. In many cases, this is
performed automatically when the switch configuration is initialized. Sometimes, though, the device nods might
drop off and not be recognized by the primary switch, as in cascading configurations. If this happens, caution
must be taken in analyzing the requirements for the storage device and the switch through which recognition is
passed to the owning server.

Considerations for SAN Software
SAN software components can form a complex set of functions. SAN configurations, given their
multicomponent assembly, require software that provides the distributed communications capabilities of the
fabric, as well as the core-level functionality that drives the internal routing and linking of FC frame processing.
The bottom line to all this is that SAN software supporting a new infrastructure (for example, a Fibre Channel
storage network) is much more complex than NAS environments. Additional discussions regarding these
differences in terms of applicability can be found in Chapters 17, 18, and 19.
Many SAN software functions are accessed outside the switch operating system, which makes the
configuration more complex as processing takes place both externally (out-of-band) and internally (in-band).
The switch, although similar to the internal RTOS in NAS solutions, is not accessible except through the
configuration process. Configuration complexities require that some training be made available prior to the
design and implementation of the SAN software.

Overview
As was shown in Chapters 14 and 15, Storage Area Networks can be assembled and connected in a variety of
ways. Switch options and device configurations are dependent on I/O workloads and recovery requirements.
Once assembled and connected, the network as a whole has to be integrated into the data center
infrastructure. Chapter 16 moves SAN configurations out of the abstract and into the data center.
Regardless of connectivity or configuration, viable SAN solutions are moving from single switch configurations
to multiswitch configurations with numerous device attachments designed to deal with increasing workloads
and even the simplest recovery and redundancy requirements. And though there are many ways to configure
a switch, the basic starting-point configurations remain core/edge, meshed, and cascading configurations. All
three present their own unique challenges when integrated into an existing infrastructure.
This chapter delves further into an external SAN connectivity beyond the node devices, which become the
drivers for application usage. Storage configurations, the workhorses of SANs, are often called upon to
provide multiaccess for the server population within the data center. One of the biggest integration challenges
is in trying to throw the configuration into the support of heterogeneous storage components. Additionally,
external forces driving new and enhanced storage media waiting to be included in the SAN can play a part in
connectivity.

Connecting into the Data Center
Connectivity refers to how the SAN as a whole integrates into the data center. In this discussion of
connectivity, we will draw a theoretical line between SAN components and server processing. As such, the
SAN as a whole includes the components that connect to the FC switch with a dotted line drawn at the HBA.
This allows us to position the FC SAN in a set of server populations and existing configurations. Even if only
used as a logical model, it helps us visualize the inevitable problems. Figure 16-1 shows the logical
demarcation line we will use throughout the chapter.
Figure 16-1: Viewing the SAN as an entire solution
Implementing a SAN configuration into a data center presents three challenges right off the bat. The first
challenge is the need to configure a multiswitch SAN. As I/O workloads, driven by supporting applications, are
applied, the need to provide multiple connections, discrete zones, and paths for performance drive the
configuration into multiple FC switches. Theres just no way around it. Over and above this is the need to
provide some level of recovery and redundancy for data availability and protection scenarios. Existing policies
and practices will drive this challenge; if not, its good to start thinking about recovery, redundancy, and single
point of failureof which, a single switch strategy presents.
The second challenge is in providing support to the server population within the data center, which can be
driven by the justification for the SAN in terms of server consolidations and supported applications. Note,
though, that it also moves into the realm of data ownership, production cycles and processing, and data
sensitivity. More importantly, it pushes the need for a heterogeneous SAN supporting more than one operating
system at the attached server level. In many cases, this becomes the big SAN imbroglio in which
implementation becomes political and ceases being productive, or relatively anxiety-freeor, for that matter,
fun in any way.
The third challenge is supporting external networks, which encompasses all of the issues mentioned
previously. The ability to provide the data stored within the SAN to corporate networks is dependent on the
external network points within the data center, as well as the capability of the attached server to distribute this
data within the infrastructure it is connected to.
Implementing Multiswitch Configurations
The need to configure a multiswitch SAN solution will become evident when designing your first solution. Any
solutions should encompass a rationale for the configuration orientation used (for example, cascading,
core/edge, or mesh). Initial implementations often move from a simple cascading architecture to a multilayered
cascading architecture without the benefit of, or options for, moving into a more complementary configuration.
Keep in mind that SAN configurations are driven by supported I/O workloads. As such, the need to provide
recovery and redundancy should play a significant role in the design. Here are a few ideas to chew on.
OLTP WorkloadsUsing a core/edge configuration enhances performance through storage access at the
edge, while reducing instances of single point of failures for high availability transactions accessing
multiple I/Os through the SAN configuration (shown in Figure 16-2).
Figure 16-2: OLTP workloads supported by a core/edge configuration
Web Applications/Messaging WorkloadsUsing a mesh design provides alternate paths for high traffic,
asynchronous I/O processing, while reducing instances for single point of failure. It relies on effective
switching methods for high traffic management within the SAN configuration (shown in Figure 16-3).
Figure 16-3: Web and messaging applications supported by a mesh configuration
Data Warehouse/Datacentric WorkloadsUsing a cascading design provides the necessary access and
performance of datacentric transactional workloads. Here, availability requirements are reduced, but
access to large data bases remains paramount when it comes to processing time and I/O intensive
transactions (shown in Figure 16-4).
Figure 16-4: A data warehouse application being supported by a cascading configuration
Coming up with the appropriate SAN design for I/O workloads also requires some thoughts on recovery.
Though driven by the applications needs, a working design should encompass current backup/recovery
processes that are in place within the data center. Its good to reconsider recovery issues relative to the
storage capacities of the SAN and the recovery requirements of the data. In a normal backup situation, data is
copied from the storage arrays to an attached server and shuttled onto the network so it can get to the backup
servers attached to tape drives. In all likelihood, these are located close to the tape library. However, if the
SAN starts out with capacities over 500GB of user data, this scenario might just overcome existing backup
practices.
Consideration should be given to backup/recovery within the SAN infrastructure, which means the addition of
external tape mediatape drives and tape libraries providing backup/recovery operations integrated into the
SAN. Designs will have to be modified, regardless of configurations (core/edge, mesh, cascading), in order to
accommodate the backup traffic and backup/recovery softwaremeaning the addition of a bridge/ router
component. This setup adds another benefit (which is largely dependent on recovery requirements and
storage capacities): the evolving capabilities of FC fabric to provide a direct copy operation in support of
backup software will eliminate a significant amount of I/O overhead. Often referred to as server-free backup
or server-less backup, these configurations will be covered again in Chapters 22 and 23.
Supporting the Server Population
The next challenge is determining which servers to support. Not only is this driven by what applications are
moving over to the SAN, but also what strategy justified the SAN in the first place. If consolidation justified the
SAN, then someone will be looking to retire or redeploy a number of servers through the consolidation efforts,
which brings up an interesting set of activities.
First is the modification of existing capacity plans to calculate the requirements of new or redeployed servers,
and to calculate the necessary I/O workloads as they contribute to the design of the SAN (some examples and
guidelines for estimating I/O workloads are upcoming in Chapter 17, with specifics for SANs following in
Chapter 18). This will very likely make you have to think about moving applications that are running on both
UNIX- and Windows-based servers. And though this is possible, and many vendors provide a mountain of
literature on the theory and ability to support this configuration, it significantly raises the cost of the SAN, not
necessarily in terms of added hardware or software, but in complexities encountered implementing and
supporting these configurations. Consider the following, carefully.
UNIX- and Windows-based servers do not access or handle storage in similar fashions. Even though there are
multiple solutions based on POSIX standards, the capability to share devices within the SAN is not yet a
reality, which means strict segregations have to be enforced with system-level zoning and LUN masking
strategies that are both time- consuming and complex.
Applications supported by UNIX- or Windows-based servers have drastically different service levels. Design
and configuration of the SAN will either add resources to the Windows area, which dont need it, or
compromise the UNIX environment by not providing sufficient SAN resources. Plus, recovery requirements are
going to be different, which means they will either share the tape media, likely to result in problems, or they will
have to have their own tape media for backup.
We should stop here and make sure were not overlooking one of the big-time workhorses within the data
center, the IBM and IBM-compatible mainframes. The need to interconnect a mainframe to open computing
has been with us since the first UNIX server deployed its data throughout the enterprise, and the Windows PC
began to develop its own legacy data, while needing to be integrated with the corporate record on the
mainframe. Unfortunately, mainframe solutions continue to elude vendors, IT systems organizations, and
research agendasnot that disparate and site-specific solutions dont exist, they do, if only in order to support
specific business applications. Integration of IBM mainframes into server environments remains elusive, even
counting all the money IBM has thrown at making the old, but reliant, MVS operating system both compliant
with POSIX as well as available to networking infrastructures like Ethernet and TCP/IP.
To its benefit, and its everlasting detriment, IBM mainframe-processing models predate many of the
innovations we are now working with, including storage networking, data sharing, device pooling, and scalable
configurationsboth internally at the high-end SMP level as well as in the multiprocessing, multitasking,
transaction performance that no other processing platform can beat. Sadly, and perhaps tellingly, the IBM
mainframe- processing model was a proprietary system that resisted evolving into the client/server
architecture too long and was thus eclipsed by its own inability to deploy and support new and distributed
applications in a cost-effective manner.
There is, however, hope on the horizon, in the form of a serious revolution in the IBM mainframe world. The
traditional operating environment has given way to a new operating system: the zOS. This next step in the
evolution of MVS provides a multipartitioned operating environment in which you can run an IBM Linux system
on user-defined partitions that leverage the power of the mainframe while ensuring that it remains cost-
effective.
So what does this have to do with storage and, specifically, integrating a Storage Area Network into the data
center? Quite a bit, if you want the truth. Because heres the first question youre going to hear: Great! When
can we connect to the mainframe?
The ability to share storage resources within the complete populations of servers, keeping in mind that zOS
facilitates a logical set of servers, is a big-time perk for data center managers. Now, heres the flipside. IBM
mainframes are moving toward their own orientation to storage area networking and are connected to their
data paths by a fiber optic connection, which the IBM folks refer to as channels. This evolution from legacy
bus, tag cables, and enterprise connectivity (or ESCON cabling: a parallel bus implementation), not to mention
the low-level channel link protocols to Fibre Connectivity (or FICON), supplies the first real direct connect to
the IBM mainframe usable by other open storage networks.
This level of service in SAN requires a FICON connection at the FC switch port level. IBM has been the
pioneer in connecting storage through a switched environment, having integrated director-level switches for
some time. At the end of the day, providing switch-to-switch communication with IBM storage switches may be
an alternative to the evolving mainframe connectivity conundrum. Entry through an IP address will also provide
connectivity, but the bandwidth restrictions make this alternative unlikely.
Figure 16-5 illustrates the many alternatives to supporting a heterogeneous OS data center through a SAN
infrastructure.
Figure 16-5: A fully heterogeneous OS SAN-supported environment
Supporting the External Networks
What if a SAN has to support an OLTP application for a highly available business process in three time
zones? Not a problem. At least its not my problem. Im just a little ol Storage Area Network.
Though maybe not.
The ability to access the SAN remotely, instead of going into the data center every time you need to access
SAN operations and information, can be configured to meet administrators needs through external network
access. The SAN configuration discussed in Chapter 15 required that SAN operations be performed through a
dedicated workstation that is directly attached to one of the SAN switches. Given this is where software
configuration tools can be directly accessed for setup, maintenance, and initialization activities, most switch
products are enabling users with IP access points that allow some of the SAN operational tools to be available
remotely.
There are several innovations waiting in the wings that will drive advanced connectivity options for the SAN.
Key among these is the need to further integrate IP- based storage solutions with FC networks. However this
will require a more sophisticated usage of storage systems driven by user data that is increasingly distributed
throughout a corporation. By analyzing current data usage scenarios, iterations of data must make several
stops within an enterprise infrastructure in order to satisfy distributed application requirements.

The Evolving Network Connections
There are several innovations waiting in the wings that will drive connectivity options for the SAN. Among
these is the need to further integrate IP-based storage solutions into FC networks, which will be largely driven
by the integration of more sophisticated usage of storage as data is distributed throughout the enterprise. If we
look at current data usage scenarios, data still has to make several stops within an enterprise infrastructure.
For example, many organizations still take it upon themselves to distribute data to remote locations, something
largely accomplished by a series of data subset deployments that are both time- and location-dependent.
Using a combination of data replication software services and remote storage devices, these configurations
depend on general- purpose servers to communicate within the enterprise network so they can ship current
data to remote locations. And though some of these can be cleaned up smartly with a NAS configuration,
many remain datacentric and rely on sophisticated RDBMS functions at the corporate data center, as well as
the distributed location. Figure 16-6 is a conceptual picture of how this works and how it lays the groundwork
for future deployments and connectivity strategies for SANs.
Figure 16-6: A data replication strategy and potential SAN solution
There are two basic scenarios that will push these networks into integrated, global SAN infrastructures. First is
the integration through IP storage devices, which offer the capability to store data without the latency and
overhead general-purpose server deployments and direct-attached storage components required. This is
handled in two ways, either through an enhanced NAS device that supports direct communication with the
SAN infrastructure, or with an iSCSI device that supports direct I/O communications with the SAN
infrastructure. Either way, the devices have to be linked through an existing IP connection and be viewed as a
node device within the SAN, which means additional port functionality at the SAN switch to handle the IP file
or block I/O over the IP network.
The second scenario involves the installation of SAN configurations that are both local and remote and can
communicate and be controlled remotely. Figure 16-6 also shows a conceptual view of how this is
accomplished, facilitating timely data replication throughout a typical corporation. Remote configurations need
further extensions of fabric software to encompass remote fabrics as extensions or as recognized external
fabrics.
This SAN-to-SAN scenario assumes that switch port evolution will continue to support the extended functions
of an E_Port communicating with a remote switch. As noted earlier, vendors must further enhance the switch
as users demand full IP connectivity and FC processing between the storage node devices. Users will also
require a management scheme to operate these configurations in a physically distributed fashion.
However, once this is established, the capability to provide a vast infrastructure of storage will be in place,
providing yet another step in the ubiquity of data for applications. This will further enhance effective web
services, thus allowing applications to be synchronized with data.
The feasibility and availability of these solutions is now becoming based on enhancements to existing network
standards, including the addition of Fibre Channel throughout the enterprise, IP standards, and SCSI
integration into IP specifications. All of which have added up to an initial wave of solutions that lead to the
extension of SAN technology. Figure 16-7 shows the potential future IP storage integration and SAN-to- SAN
communications supporting advanced data management strategies.
Figure 16-7: A future SAN-to-SAN configuration

The Evolving Device Connections
Along with network connections, SAN connectivity options continue to evolve, encompassing more storage
devices. In fact, integrating FC communications with the myriad tape solutions available has been underway
for some time. Further enhancements to tape media will likely come in the form of optical and solid state disks,
as well as advanced forms of caching mechanisms.
Integrating Tape Solutions
Tape media is currently integrated using an external bridge/router solution (see Chapter 14). Bridge/router
connections are increasingly integrating solutions whereby FC connectivity is built into the tape controller
devices, allowing for a cleaner installation, not to mention a closer integration into the controller mechanism
that takes advantage of individual tape manufacturer's differentiators and functions. This level of integration
comes closer to allowing tape to function more efficiently by talking directly to the switch port. Developing tape
media is an ongoing effort of the FC standards groups.
Tape media remains a particularly sensitive area within data centers, given the large investments that must be
made in these realms. The capability to change and start additional library areas that support additional media
is no small undertaking. One of the largest bottlenecks in support processing within data centers is backup,
which gets performed on a daily, weekly, and monthly basis. This takes up an inordinate amount of time
compared to the production processing that takes place in direct support of the company's business.
(Additional details on backup/recovery processes are covered in Chapter 22.) Integrating tape library functions
into the SAN is a significant task. Even though it may seem trivial at the beginning, the importance of the
complete analysis of this integration greatly impacts the backup cycle-both positively and negatively.
The positive aspect is enhanced performance when moving backups to FC speeds. Even though you're
throttled by the limitations of the tape device, it remains significantly faster than a slower and copy-redundant
direct-attached model that can quickly congest an IP-based network infrastructure. There is also the potential
value that a 'server-less' backup scenario can be supported within your SAN configurations. Here, the elapsed
time savings in running backup jobs can be as great as 50-75 percent, meaning that a typical backup cycle
which runs seven hours would now take only three and a half. Because of this, significant overhead savings
can be obtained over and above the amount of production processing time placed into the cycle.
The negative aspect is the increased cost to the facilities and recovery operations that result from integrated
SAN tape configurations. The additional costs are needed to allocate tape drives for the FC SAN switch. More
than likely, these will be dedicated given the non-sharing that FC SANs are currently encumbered with. In
addition to this are the extra facilities necessary for extending the tape library functions-even though it may
integrate into the current library system.
Another negative aspect is the consideration of the recovery operation that will ultimately have to take place.
This requires the ability to locate proper backup tapes, log tapes, and control tapes. The tapes have to be
dumped to the appropriate volume within the devices from the attached server, and then applied in the event
of database recovery (which most may be). This isn't necessarily any different than normal recovery, other
than that all of this has to be performed within the SAN environment, which, unless it's taken into consideration
beforehand, can very quickly get out of hand.
Looking for an Optical Connection
Another popular media that will potentially integrate into the SAN is optical, or CD and DVD drives and
libraries. Historically, these devices have seen limited usage as removable media libraries within enterprise
network environments because of their performance as a write-once disposable media and physical read/write
performance at the optical device level. Both of these conditions have improved to the degree that optical
media and drives can be used for special applications of information distribution, reference, and, in some
cases, effective backup media.
Optical devices can be integrated into a SAN using a bridge/router device to perform the same interconnection
as Tape media. The use of SCSI optical drives with a SAN router can integrate the CD and DVD media
sufficiently for production processing. The production of CD or DVD for distribution from digital sources, such
as post processing of entertainment and educational packages, can benefit from the increased speed of SANs
and allow it to process in a connection-oriented environment to facilitate the total bandwidth of the SAN.
Front-end processes of these applications, the preprocess production, where digital entertainment and
educational products are edited and produced, have been an early adopter of SAN technology. This is largely
due to bandwidth and storage capacities, as well as the ability to provide a Class 1 connection, which
facilitates the streaming media production requirements for digital video files. These configurations can form
either the production system for manipulating and creating digital video/audio effects, or the distribution
system to move entertainment or educational files out to a network.
Actually, optical drives and autoloader packages have recently become available in NAS configurations. The
NAS device is bundled with optical media drives and an autoloader system while maintaining the NAS file
access front end and Plug and Play connection to an IP network. Given the future integration of NAS and
SAN, the integration of NAS optical systems into SAN will become another alternative to SAN/ NAS integration
and optical media usage with the SAN environment.

Connecting to the Future
Over and above the connectivity of SAN into existing data center infrastructures, SAN will soon need to
integrate into the continuing evolution of external processing technologies, such as advanced clustering,
InfiniBand, and blade-level computing. Each of these will require the SAN to address the compatibility issues
of how the FC network interfaces with these evolving technologies.
Clustering has been an emerging solution for highly available configurations and is now employed to address
availability levels for real-time transactional workloads. Although clustering solutions utilize the SAN in an
effective manner, the ability to support both sides of a cluster is problematic, causing most configurations to
support two distinct SAN configurations, which provides two distinct system images, with a SAN fabric
supporting each.
The consideration regarding these configurations is the expense in doubling the processing environment to
ensure that upon failover the supporting system can take over the workload and synchronize the data and
continue processing. Secondly, the protocol used to communicate between the two systems, TCP/IP, adds
quite a large overhead for the short communication paths between servers. Finally, the complexities of
synchronizing the databases and data for continued processing upon recovery makes these solutions
expensive and difficult to configure and maintain.
Look for clustering to take advantage of two systems sharing the same fabric, reducing the complexity of
totally discrete systems to failover at the processing level with shared SAN access. This allows the data to be
protected through a RAID-level mirror within the SAN, should it be needed at the storage level. Additional
protection for hardware and data redundancy within the SAN configuration provides for failure recovery within
the storage infrastructure, as shown in Figure 16-8.
Figure 16-8: Advanced clustering with shared storage SAN
If we take our clustering discussion and replace the system-to-system communications (now using TCP/IP)
with a switched fabric network implementation using the InfiniBand protocol, we begin to see the benefits of
InfiniBand. Let's take that one step further and implement a fabric-to-fabric interconnection, tying our SAN
configurations together. Now we're really seeing the benefits of InfiniBand.
An InfiniBand configuration allows system-to-system communications to scale beyond the current 2-, 4-, or 8-
way clusters that are expensive and unwieldy, but nonetheless available, and replace this with low-cost
process node servers communicating through a switch fabric. This allows an infinitely larger number of servers
to effectively communicate. In addition, the ability to tie in storage devices, such as the SAN, through an
InfiniBand fabric to FC fabric interconnection allows some or all the servers to share the storage resources
being managed by the SAN fabric.
InfiniBand technology allows for the creation of both a process infrastructure and a storage infrastructure,
each with the capability to communicate though a common scalable interconnect in order to develop a dense
computing and storage fabric.
If we take our Infiniband and FC SAN fabrics and key in on the process nodes, the application of a switched
fabric internal to the process node provides yet another scalable resource within the process fabric. The use of
internal switched fabrics, such as Rapid I/O and HyperTransport, form a foundation that allows internal
components of the process node to scale new heights, bumping up the number of CPUs and addressable
storage levels.
Given this configuration, we can begin to visualize the requirements for an internal storage function within the
process fabric to support the internal functions of the process node. The communications between storage
and process fabrics are enabled by the InfiniBand protocol.
In conclusion, network technologies will continue to decouple the traditional computing configurations into
process and storage networks. The data center is headed toward the next level of computing, driven by
storage area network technologies that support all aspects of application sets and process environments:
business, academic, scientific, and personal. The future data center is likely to become a logical set of storage
and process fabrics, as illustrated in Figure 16-9, communicating through a public and private network
infrastructure.
Figure 16-9: The future: process and storage fabrics with a common interconnect

SAN Configuration Guidelines
The capability of Storage Area Networks to integrate into the data center can become more complex than
expected. However, given the right orientation, and a realistic view of current solutions, SAN offers a
compatible solution to both UNIX and Windows environments. Not without its challenges, the ability to move
beyond the homogeneous implementation can be successful, though more expensive and encumbering than
the original SAN justification may have entailed. With the current set of products and solutions available, the
data center can expect the following when connecting the SAN into the data center:
Multiswitch configurations will be a way of life. Be prepared to work with multiple switch configurations.
Again, there's just no way around this. Even if the simplicity of a single switch configuration may suit your
requirements, the minimum attention to recovery and availability requirements will force a move to at least
two switches for some level of continuous operations.
Matching I/O workloads to compatible configurations will require additional planning. Part of the
justification for the SAN will likely be server and storage consolidation. As you consider the combined
workloads and their relative I/O, you will begin to see port count requirements, HBA data paths, and
storage sizes themselves moving beyond a simple solution. However, for performance and availability, it
is best to implement a SAN configuration that facilitates your workloads. Given the flexibility and
numerous variables to work with, be prepared to spend considerable time upfront considering these
design activities carefully.
SAN configurations will be subject to rapid change and enhancement. One of the most important practices
to consider is staying flexible, especially considering the evolutionary tracks SANs are on. Major
enhancements will continue to be announced on a constant basis, so a reliable and consistent view and
monitoring of the SAN industry will be an effective tool to understanding the evolution of SAN connectivity.
As Storage Area Network technologies and solutions continue to evolve, connectivity of components to key
elements within the data center infrastructure can be terribly disruptive to existing operations. However, this
same connectivity can also be the most productive in facilitating the scalability of a storage infrastructure as
well as establishing a foundation for the evolving enterprise storage fabric.

Part V: Application-Putting It Together
Chapter List
Chapter 17: Defining the I/O Workload
Chapter 18: Applying the SAN Solution
Chapter 19: Applying the NAS Solution
Chapter 20: Considerations When Integrating SAN and NAS

Overview
Capacity planning has always been something of a black art. Commodity-level distributed computing has
driven IT to conduct a simplistic set of activities meant to plan and implement computing resources. Any
discussion on applying the concepts of storage networking must touch on key workload definition concepts
that are part and parcel to an effective capacity plan and to the possible rejuvenation of a more effective
capacity planning model. Part V orients the reader towards applying storage networking concepts by way of an
initial discussion and overview of workload definition and planning. The following chapters contained herein
focus on identifying existing and anticipated workloads where storage networking solutions can be
implemented for maximum value and productivity to the data center. SAN and NAS configurations will be
addressed in analyzing the particular workloads of each.
We have explored and used the term workload many times already, and in Part V, it remains important to our
discussion. Chapter 17 provides a detailed examination of workload definitions, identifications, and planning.
Workload guidelines will be placed into a real-life context that characterizes both business applications and
support activities, as well as the maintenance of capacity strategies as they relate to storage networking. Many
of the activities that you manage on a daily basis may be enhanced or hurt by the new technologies involved in
SAN and NAS solutions. In an effort to provide as much relative information as possible, generic descriptions
and examples are used in Chapters 18 and 19 to cover several specific workload characterizations.
Part V addresses the usage of SAN and NAS as an integrated solution. Chapter 20, for instance, explores
situations based on current IT experiences and also looks at future uses of integrated SAN and NAS
solutions. This becomes a complex exercise when multiple workload scenarios are cast within the same
environment. The ability of microkernels to communicate effectively and for storage caching mechanisms to
operate in tandem becomes a challenge when considering the future of heterogeneous storage operation.
This is compounded by the ability to maneuver within a distributed file system that incorporates optimum
support for both traditional workloads yet sustains the proprietary nature and operation of database
technologies. The current evolution of common application services through enhancement of web-enabled
solutions will incorporate all the attributes of OLTP, batch, and data-intensive I/O, and will play a key role in
these discussions.
Moving storage networking into the future requires an analysis of integrated storage networking solutions,
using both FC- and IP-based connectivity schemes. Because of this, Chapter 20 addresses key future
technologies, showing how they may affect IT storage networking solutions. Such technologies and standards
include iSCSI, the Internet SCSI standard that allows SCSI commands to be encapsulated and transmitted
through an IP network. In addition, we address the slowly evolving InfiniBand standard and infrastructure
advancements that will result. Finally, well look at some of the internal bus connectivity standards and
strategies that could have a dramatic effect on the data storage hierarchy as we know it today (see Chapter 5)
exploring products such as AMDs Hyper-Transport, the industry standard VI transport mechanisms, and
Intels next bus technology, Rapid-IO.

Storage Planning and Capacity Planning
Planning and installing applications continues to require a big effort within IT, especially when it comes to
applications that are considered enterprise-level (a term which has never seemed appropriate, by the way).
Nevertheless, the term was employed by large system players (read mainframes) to differentiate their
solutions from those of the small system players (read PCs) in the early days of the user revolution. Commonly
known as the PC wars, it was a time in which many business users made attempts to buy their own IT
infrastructures for pennies on the mainframe.
It was about this time that sophisticated capacity planning and workload management started to go out the
doorout the door, out the window, and out of IT standards and practices. All in all though, those tools and
activities provided a certain amount of pride and gratification when the systems (read mainframesagain)
were upgraded and everything not only worked, but worked the same as it had before, and had improved user
response time, balanced utilization, and a batch window that was completed two hours earlier.
Well, we traded those simple days for ubiquitous and prolific computing resources based on everyones
desktop, job, travel agenda, and meeting rooms. We traded the security of centralization that made the
capacity planning a manageable and gratifying entity for a Wild West distributed environment that reflected the
scalability of the box we were operating on, be it UNIX or Windows. The mantra was: If the ones were running
dont work, are too slow, or cant support the network, we can always get new ones. What could be simpler? A
capacity plan based at the box level. If it doesnt work, just get another one. They dont cost much. We dont
need to consider no stinking workloads, we can process anything, and if it grows, well just add more boxes.
The same goes for the network.
A wonderful idea, and to some degree it workeduntil the applications started to take advantage of the
distributed network features of the boxes, as well as the increasing sophistication and power of the boxes.
Yes, the little boxes grew up to be big boxes. Others grew to be giants, the size of mainframes. It should be
pointed out, however, that many boxes had genes that stunted their growth, keeping them from becoming
enormous mainframe-like boxes. Even then the box applications grew in sophistication, resource utilization,
and environmental requirements.
NoteThe gene mentioned here was referred to as the Gatesonian gene, whose scalability would not
grow beyond the confines of a desktop.
Interestingly enough, we seem to have come full circle. Although nothing essentially duplicates itself, our circle
of activities has landed us back near the realm of thegasp! workload. Its not like workloads ever really
went away, we just havent had to recognize them for a while, given we all got caught up in the box level
capacity planning activities, known as the BLCP practice.
BLCP is not unlike the IT practice of going into installation frenzy during certain popular trendsfor example,
ERP, relational databases, e-mail, office automation, web site, intranet, and now the dreaded CRM. Such
things cause us to be driven by current application trends, leaving us to distinguish between CRM
performance, ERP utilization, and e-mail service levels. To a great degree, we handled these installation
frenzies with the BLCP (box-level capacity plan); we can implement anything by using the right box, as well as
the right number of boxes.
Several things within the applications industry have rendered the BLCP practice obsolete. First and foremost
is the sophistication of the application. Today, applications are distributed, datacentric, hetero-data enabled,
and, the up-and-coming process, de-coupled. Second, the infrastructure has become specializedfor
example, there is the network, the server, the desktop (for instance, the client), and the storage. Third, the
infrastructure itself is becoming further de-coupled. More specifically, storage is taking on its own
infrastructure, with storage network freeing its processing bounds and restrictions from the logic of the
application, as well as the network overhead of the client/server model, becoming itself immunized against the
Gatesonian gene.
NoteHetero-data is the characteristic of an application that requires multiple data types to perform its
services.
Finally, the common services that support the application logic infrastructure are quickly becoming both
commodity oriented and public in nature. We are moving into a future where applications are programmed by
end users to employ web services. These services access lower level infrastructures and subsequently
complex and proprietary support products operating within a particular information technology structure. Users
will likely take it for granted that adequate resources exist.
These conditions will further render the storage entity with its own operating infrastructure. However, this will
require that it evolve as a more complex structure, supporting an even more complicated set of common
services that includes application logic, network processing, and common I/O facilities. Each of these entities
will have to understand and configure themselves to effectively operate with existing and future workloads that
will dictate the processing and physical requirements necessary for just minimal performance.
Thats why workloads are important.

The Definition and Characterization ofWorkloads
Workloads are traditionally defined as application processing modules that exhibit similar resource
requirements, processing characteristics, and user expectations. Figure 17-1 further depicts this definition with
its characterization of banking transactions. For example, at one time or other, weve found ourselves in a
bank line with someone whos balancing their bank statement, applying for a car loan, and/or trying to find a
missing check from, say, 1982, all at the same time. Their transaction, obviously, is much more resource-
intensive than the deposit of your Grandmas birthday check.
Figure 17-1: An example banking transaction
So, if there were a categorization for the types of transactions the bank handles, everyone would get serviced
more efficiently within a prescribed set of expectations. However, this would require the tellers to be
categorized by the type of transaction they perform. Tellers would have appropriate resources at their disposal
to deal with the transactions they service. There you have it, a simple workload categorization for the tellers,
who, in this example, can also be characterized as servers (depicted in Figure 17-2).
Figure 17-2: Organizing workloads for maximum performance
The tellers have a set of resources and types of transactions they work on, giving certain kinds first priority.
However, since these are all general-purpose tellers, they can work on any transaction should they run out of
the particular type of work they are optimized for. Keep in mind that they are optimized for a particular type of
workload and prioritize their work on that basis. Because of this, they may not handle a different workload as
effectively as a peer teller that is optimized for a different type of transactional processing.
In this example, a simple transaction like depositing your Grandmas check can be taken care of by a teller,
who functions similar to a computer transaction server that specializes in, and is optimized for, checking
deposit transactions. The teller has the appropriate resources to process your deposit within a service time
you are expecting. Other tellers process different transactions, such as safe deposit transactions (very
physical, employing various resources), bank statement questions (very process-intensive, requiring much
data), and financial portfolio transactions (multiple processes, compute-intensive with high priority).

The Business Application
By using the banking example, we can determine all the necessary considerations for identifying, describing,
and characterizing the workloads. We first define the banking activities as the business application. Within the
application are several subsets of activities, which can be directly related to application programs that support
the banking business application. We will now convert our abstract teller infrastructure, supporting the banking
application to a computer infrastructure with servers and related components that allow users to perform
transactions. This is illustrated in Figure 17-3 and will serve as our basis for workload considerations for this
chapter.
Figure 17-3: Banking configuration example: defining the business application
Using Figure 17-3, we see that identifying parts of the application becomes fairly straightforward, as an
application development group defines the functions performed by the business application. This provides an
inventory of application programs that have been developed and need to be available for meeting the users
needs through their transactions. We evaluate each function, or program, to our computer configuration, as
indicated in Figure 17-3.
We do this by understanding the data organizational structures required for each application program, the data
paths it uses, and the user access requirements for each. For example, our checking deposit program uses a
database and related log files to handle its processes. The user access demands that the deposits be
reflected in the daily database but also in the statement of record for the customer, which is in another
database. Further subsets of this information go toward an activity log used for updating other applications
within the banking business application.
Deposit ApplicationWorkload Attributes
The previous information describing the banking transaction pretty much defines the data paths required for
the deposit program. We then coalesce this information into a set of workload definitions for the deposit
application. Based upon the organizational and user access information, we can define additional attributes of
the data path. This allows us to describe the deposit application with the following attributes and begin
identifying it as a workload.
Data Organizational Method (DOM)Relational database, flat files, and temporary suspense files.
User Access (UA)Online transactions that require immediate update to daily customer tables, with
conditional processing based on secondary transactions to customer system record tables and log
suspense files.
Data Paths (DP)Macro data paths require update access to daily customer tables, read access to daily
suspense tables, and a system of record tables. In addition, there are multiple paths required to facilitate
the reliability, availability, and serviceability of the application.
By further defining the technical attributes of the workload, we begin to make decisions on where this
particular program will be placed within our infrastructure. These are illustrated in Figure 17-4, which are
depicted as updates to our initial Figure 17-3. We can now overlay workloads within this configuration. What
we have in our sample banking application is a classic OLTP workload that supports the deposit subset of
functions. The OLTP workload dictates several configuration characteristics at a minimum, with additional
considerations based upon estimated traffic and growth.
Figure 17-4: Placing workloads within an infrastructure
This defines a simple model for identifying and describing a macro set of attributes for the workload. Having
done this, there are additional details that become necessary as we move to place the application into our
sample configuration. Now it becomes necessary to further understand the resource utilization, traffic arrival
rates, and environmental conditions surrounding the support of this workload. However, before we proceed
with deriving additional details from the deposit application, we must address in summary fashion the analysis
of the complete inventory of the Banking Application. This requires using the same process that created our
initial workload definitions for the deposit application.
Given this iterative process, the importance of analyzing the entire Banking Application in context cannot be
overemphasized. During this process, we find that many of the applications within the Banking Application
have similar workload characteristics, such as same database, similar user access demands, and process
requirements. On the other hand, we also recognize that many applications have completely diverse sets of
attributes. Our example shows two other distinct workloads for placement within our configurationa batch
workload and an archival workloadeach of which has a unique set of user transactional requirements.
NoteIt should be noted that such distinct workloads were chosen to reflect the ubiquitous nature of each
configuration. These can be found in almost all IT data centers and remain applicable across
business and scientific applications.
This requires an accumulation of resource, access, and environmental elements to derive a total picture of the
workload. As such, our original example of the banking deposit becomes only a part of the Banking Online
System, which we have concluded from its accumulated characteristics to be an OLTP workload, with a set of
requirements designed to meet user expectations. Figure 17-5 offers guidelines for identifying workloads and
defining their characteristics.
Figure 17-5: Guidelines for identifying and defining workloads
Another interesting aspect to this exercise is something you may have already concluded as we moved
through our workload identification and definition stage. That is, with few exceptions, the majority of resource
requirements and characteristics center around the I/O of the workload. Certainly I/O plays a pivotal role in
processing transactions, given the necessity of timely execution of data acquisition and transfers in relation to
response time. Within this simple concept lies an interesting conclusion: over 90 percent of the workloads that
IT professionals deal with are I/O-intensive transactions. In other words, the majority of activities encapsulated
within commercial processing consist of I/O-related activities.
The corollary to this observation is that the majority of workload planning and implementation centers on the
I/O system that supports the processing infrastructure. A further realization is the importance storage
architecture plays within the infrastructure and the tremendous impact putting storage on a network will have.
Why? Because it changes the implementation strategy of the workload so dynamically that workload
identification and definition becomes paramount. Otherwise, just throwing disparate workloads within a storage
networking infrastructure can be a hit or miss affairone with a low probability of success. However, making it
too simple and only implementing homogeneous workloads may not leverage the real value of storage network
technology in how it supports disparate workloads.
The conclusion is that workload identification, definition, and planning are critical activities to the effective
application of storage networks.

I/O Content and Workloads
Potential resource utilization and traffic arrival rates play a very important role in integrating the workload into
an existing infrastructure, or in building a new infrastructure, for that matter. As discussed previously, the
categorization and accumulation of workload attributes provided the definition and requirements for the
workload. Continuing with our banking application example, Figure 17-6 illustrates the final inventory,
categorization, and definition of the workloads. This allows us to accumulate a set of estimates of resource
requirements for each workload and provide an estimated sum of the entire set of workloads. In order to
accomplish this, we must look closely at each category of workload.
Figure 17-6: The workload inventory, categorization, and definition of the Banking Application System
The resource requirement details are contained in the major areas of I/O activities, data organization, data
paths, and user access. Putting these areas together forms a picture of the required infrastructure for the
workload and, as we will see, a picture of the total infrastructure.
The Data Organizational Model
The most important aspect of the workload is the data organizational model it uses. In todays inventory of
applications, both internally developed by IT application groups or IT implemented through application
packages, the majority are based upon a relational model. The use of relational database technology
(RDBMSs) defines the major I/O attributes for commercial applications. This is a good thing because it
provides those workloads that use RDBMSs a set of processing metrics for estimating I/O behavior and
utilization. The use of the relational database has become so accepted and widespread its macro behavior is
very predictable. Additional resource utilization is provided by the workload characteristics that define internal
processing requirements such as caching, temporary workspace, and partitioning.
Dont make the mistake of overlaying the storage infrastructure too quicklythe consideration of recovery
scenarios and requirements needs to be considered at the macro level first and then decisions made to
handle specifics of the workload or particular subset of workload (for example, the specific application
program). Therefore, we will have all our workload consideration taken into context before we make any
conclusions about storage system features such as RAID, cache sizes, and recovery strategies.
The data organizational model provides the following sets of information to the workload behavior:
Block SizeThe size of the block of data moving from the application transaction is dictated by the setup
of the database. This can be either file or database attributes.
PartitioningThis attribute defines behavior regarding user access patterns in relation to the data itself. It
influences decisions regarding the type of fault resiliency strategy for the workload (for example, RAID
levels) and software recovery mechanisms.
Physical DesignThe physical design of the database drives how a supporting file system is used. This is
perhaps one of the most important attributes to consider given the type of database and its performance
when using a file system.
NoteRelational databases continue to prefer the use of raw disk partitions. This is important given the
performance penalties one may encounter when performing I/O operations that are duplicated
through the file system and passed off to the RDBMS system for further read/write operations.
Referred to as the double-write penalty, this will be covered in more detail in Part VI.
MaintenanceProbably the most overlooked attribute in todays implementation- crazy environments, this
defines workload behavior itself and includes requirements for backup/recovery, archival, and disaster
recovery.
NoteRegarding the use of RDBMS technology, the backup/recovery group not only includes basic
backup operations but more important recovery operations that have to occur within a
transactional basis. They must therefore include the necessary log files and synchronize the
database to a predefined state.
This articulates a macro view of workload considerations and attributes supported by the analysis of the data
organizational model.
User Access
User traffic defines the data hi-way attributes for the workload. Notice that we address this set of information
prior to our topic on data paths. Obviously, we need to know the estimated traffic to understand the type,
number, and behavior of the data hi-ways prior to associating that with the workload expectations. We
basically need three types of information from the end users or their representatives, the application systems
analysts. This information is shown in Figure 17-7 and consists of a number of transactions, the time-period
transactions needed to execute, and the expected service level.
Figure 17-7: User access information for workload analysis
This activity is the most challenging. Trying to get either end users or their application representatives (for
example, application system analysts) to estimate the number and type of transactions they will execute is
difficult and, although important, can and should be balanced with a set of empirical information. Lets look at
our banking application example and the deposit transactions. If we query the user community, we find that
each teller handles approximately 100 deposit transactions per day. There are ten branch locations and, on
average, five tellers per branch working an eight-hour shift. This results in an estimated 5,000 deposit
transaction per day that need processing between the hours of 9 A.M. to 5 P.M.. Tellers expect their deposit
transactions to be available during the entire eight-hour shift and have transactional response times ranging
from subsecond processes to those lasting no more than five seconds.
However, the deposit transaction comes in three forms: simple, authorized, and complex. Simple requires an
update to a customer table, authorized needs an authorized write to the permanent system of record for the
account, which requires a double-write to access more than one database table. The complex, meanwhile,
requires the authorized transactions but adds an additional calculation on the fly to deposit a portion into an
equity account. Each of these has a different set of data access characteristics yet they belong to the same
OLTP workload.
This is important given the potential for performance degradation if the appropriate data access points are not
taken into consideration. Not all of this information can be expected to come from the end users, although we
can estimate by analyzing the transaction history to determine the mix of subtransactions and thus plan
accordingly. Consequently, the importance of fully understanding the transactional estimates generally goes
beyond the end user or even the application developers estimates, providing critical information regarding
decisions in I/O configuration.
However, its also important because it defines the I/O content of the transactions. I/O content is defined as
the amount of user data transferred during an I/O operation. We have discussed previously that both bus and
network transfer rates differ in the amount of bytes transferred. However, this is dependent on several
variables including operating system, file system, and data partitioning in RDBMSs. Consequently, it is not
always the case that the bus is full when executing an I/O transaction. Therefore, the more intensive the I/O
content, the more throughput occurs, and the less time it takes to complete a transaction based on obtaining
the amount of data needed.
An example is the deposit transaction set where the simple transaction only requires access to a database
record within the database table. Even though this customer record only consists of a small amount of data, it
still requires the server OS to execute an I/O operation. This design hardly makes the trip productive given the
system overhead of the I/O operation using SCSI/PCI configurations or the larger amount of system process
overhead necessary to leverage the tremendous payload of Fibre Channel. However, if the application design
requires that each separate transaction is provided a single transfer packet or frame facilitated by the I/O
operation, then the efficiency of the I/O must be considered to understand the necessary system requirements
to support the I/O workload. Although this may provide an extremely fast I/O operation and subsequent
response time, the amount of system resources dedicated to accomplishing this limits the number of
transactions supported.
Using this example analysis of a single I/O per transaction, the number of transactions processed within the
stated time period becomes very important. In the case of the example deposit application, the simple
transactions make up the bulk of the workload. This leads to the conclusion that the capacity of the system is
simply based upon the number of I/Os required for processing. Nevertheless, we know that the efficiency of
the I/O system is highly inefficient and subject to non-linear response time service should any anomaly in
increased transactions occur. Consequently, basing a capacity estimate simply on the number of I/Os a
system is capable of is not necessarily the only metric required for balancing or building an effective and
scalable I/O infrastructure.
The final piece of critical information regarding user access is the expected service level. This places our
eventual, albeit simple, calculations into a framework that defines the resources needed to sustain the amount
of operations for the OLTP workload. From our initial information, we find that there are two goals for the I/O
system. First, the banking applications data needs to be available from 9 A.M. to 5 P.M. each workday.
Second, the transactions should complete within a time frame ranging from subsecond response times to
those lasting no more than five seconds. For the sake of our example, we will not address the networking
issues until later in Part VI. However, it is important to note that although the I/O can, and will, take up a great
deal of response time factors, network latency issues do need to be considered.
By adding our service-level expectations to the mix, comparing these to user transactional traffic estimates,
and considering the detail of the subset of deposit transactions, we find that the deposit transactions require a
system that enables a transaction rate of 5,000 transactions/8 hours, or approximately 10 transactions per
minute. That adds to our cumulative amount, which provides the capacity for the entire banking application.
The cumulative analysis further concludes that a 500-MBps transfer rate is needed to successfully meet user
expectations.
In addition, the infrastructure must support the 100 percent uptime that users expect in terms of data
availability. This is a foundational requirement for meeting response time (for example, the data must be
available to process the transactions). However, this begins to define the type of storage partitioning and
structure necessary to provide this. Consider also that in meeting these goals the system must continue to
process in the event of erroran area handled by RAID. In particular is the decision to provide level-1 or level-
5 solutions necessary for efficient response time even during recovery processing.
Data Paths
Now, lets look at the data hi-way required for this workload. From an analysis of our first two categories we
create a picture of the logical infrastructure necessary for the workload. By comparing the data organizational
model (for example, the type of database and characteristics) and byte transfer requirements with something
called the concurrent factor, we can begin to formulate the number of data paths needed to meet workload
service levels. The concurrent factor, as mentioned previously during the user access discussion, determines
the minimum, logical set of paths required to sustain our service level given the probability that all tellers at
some point may execute deposit transactions simultaneously.
This calculation provides a more accurate picture of the resources needed to sustain the service level in real
time. In reality, the probability that all the tellers will execute a deposit simultaneously is actually quite high and
is calculated at 90 percent. Therefore, for each time period, 90 percent of the total tellers would be executing a
deposit transaction. From our previous calculation, we estimate a mix of simple, authorized, and complex
deposit transactions, which would be something like 80, 15, and 5 percent, respectively. This calculation
provides for the average number of bytes transferred while taking into account the different I/O content of each
transaction.
Figure 17-8 illustrates the accurate requirement for our workload. With this I/O workload analysis information,
we can evaluate existing configurations to see if any of them will sustain the load, or develop a new model to
configure a system that will sustain the load. Most likely you will want to do both. As in our sample case, we
can see that we need a large amount of sustained workload I/Os for the entire business application. If we
overlay this existing solution of direct-attached SCSI storage systems with capacities of no more than 50
MBps and arbitrarily-based device execution, it is likely this will be completely deficient in meeting workload
service goals.
Figure 17-8: Calculating workload requirements for byte transfer rates
However, if we develop our own model, we find that a simple FC SAN solution with a 100-MBps rate through a
frame-based transport will likely sustain the workload and support the concurrent transactional workload I/Os.
If you add storage controller requirements of RAID, data maintenance, and recovery applications, we can
estimate three data paths totaling 300MB burst rates. An estimated 240MB sustained rate thus will not only
provide sufficient transfer rates but also a safe zone to compensate for peak utilization and growth factors
before additional paths are required. Figure 17-9 illustrates a logical model built from our calculations.
Figure 17-9: Workload analysis logical model
NoteThe 240MB transfer rate referred to previouly is calculated at 80 percent of total capacity (2,000
bytes 80% = 1,600 bytes), which provides for switch and device latency.
From this model, we can begin to assign specific technologies to find the most appropriate fit. Cost will
certainly be a factor in determining the best solution for the workload. However, cost notwithstanding, the
value of workload identification, definition, and characterization starts to become evident when moving the
workload analysis into real implementations.

Considerations for I/O Workloads in Storage Networking
Implementing a solution that meets a set of business applications requires a workload analysis, as we have
demonstrated. It then becomes necessary to analyze an existing configuration or develop a new configuration
that can sustain the workload and meet user service levels. However, in the world of storage networking, IT
faces additional challenges. These issues focus on the increasing separation from the server complex,
something which becomes an infrastructure in and of itself (such as a storage network infrastructure).
The most visible issue will be the discrete activities necessary to analyze the I/O infrastructure. Even though it
must be taken in context with the business application and subsequent workload analysis, I/O becomes its
own entity when moved into a storage networking model. Therefore, it is imperative to plan, model, and
configure storage networking solutions with sufficient detail and I/O specifics that are consistent with the
workloads you are supporting.
In some cases, the I/O workload is only part of a larger capacity planning exercise. Consequently, I/O
workload planning needs to become integrated into existing planning activities and be driven from workload
estimates that have been previously discovered. If that is the case, the minimum elements of I/O workload
planning, as indicated in the earlier section Deposit ApplicationWorkload Attributes, are still required and
should be discovered even though they may come from an existing capacity planning exercise.
The ability to support multiple workload types cannot be overlooked. This will probably be the most challenging
activity required. Certainly SANs and, in some cases, NAS solutions will be required to support workloads that
are disparate in their characteristics and processing service levels. This makes workload analysis imperative
when planning the I/O infrastructure so workloads can be identified and sized before implementation takes
place. This is also necessary in order to defend existing configurations when planning upgrade strategies.
This leads us to the next two chapters in this part of the book, which concern the application of storage
networking solutions and technologies. Given that we have already identified, described, and characterized
our workloads at a macro level, the right solution should become evident.
SAN or NAS Solutions
With current information, you can decide pretty early whether youre looking to leverage an IP-based NAS
solution or a more comprehensive FC SAN configuration. The decisions based upon workload analysis will
provide accurate direction when it comes to choosing a storage networking solution. Keep in mind, it will also
provide accurate justification if staying with current direct-attached storage solutions looks to be the best
alternative. Figure 17-10 offers some guidelines on initial workload analysis when macro decisions need to be
made regarding SAN, NAS, or direct attached.
Figure 17-10: Workload solutions matrix

Overview
The SAN architecture, as previously discussed in Chapter 17, provides a level of storage scalability and
capacity beyond traditional I/O infrastructures. SANs provide a rich set of opportunities to apply to several
workloads that are standard within todays data center. Among these workloads are OLTP, web services, and
data warehousing. OLTP, a foundation of most businesses by supporting their daily operations, is a significant
consumer of storage capacity with strict access requirements. The increasing use of web services to provide
an array of Internet transactional messages between customers, retailers, and suppliers, enables a significant
value benefit to all involved; however, it pushes the performance and scalability envelope of the traditional I/O
infrastructure. Data warehouse applications and their related brethren, the data mart, always move the storage
capacity bar to its limits as the movement of data within these applications accelerates the need for wider
access bandwidth.
This chapter will discuss the guidelines for estimating SAN configurations and explore key concepts
concerning the deployment of SANs into these standard workloads. Key among these will be the identification
of port requirements necessary to sustain a particular workload, as well as particular issues associated with
the often-disparate processing characteristics of OLTP, web services, and data warehousing applications.

SAN Workload Characterization
When characterizing workloads for a SAN, its helpful to consider the inherent value that Storage Area
Networks bring to the data center. This cross-application infrastructure enhancement may be the overriding
factor in the justification of an initial SAN configuration that on the surface could be handled by direct connect
architecture. This further demonstrates the importance of defining I/O workloads, as discussed in Chapter 17.
When describing SAN I/O workloads, it is also important to be aware of the possible integration of system-
level capacity planning as well as network capacity planning for the obvious reason that the SAN architecture
represents a combination of computer system and network characteristics that must work together to sustain
the workloads. This requires that workloads be evaluated with both I/O processing and network configuration
metrics. Even though SANs are young by contrast to other technologies, standard configurations have
emerged that can be applied to most common workloads. The major categories, excluding a single switched
environment, are cascading, meshed, and core/edge configurations.
These categories provide starting points for data centerspecific customization and specialization. SAN
expansion and derivations will likely come from one of these three configurations. A brief overview follows.
Cascading SAN Configuration This configuration provides a switch-to-switch connection that allows the
number of server and server devices to scale quickly. Figure 18-1 shows a simple cascading configuration
with three servers and multiple storage devices using three FC switches.
Figure 18-1: Cascading SAN configuration
Meshed SAN Configuration This configuration provides a performance- oriented system that allows for
the quickest path from server to data. Figure 18-2 illustrates how an I/Os path is reduced as it can reach
the storage array without traversing multiple switches. However, it also provides the multiple connections
to connect to the storage array using alternate paths in the event of heavy traffic or switch disruption.
Figure 18-2: Meshed SAN configuration
Core/Edge SAN Configuration This configuration takes into account I/O optimization, redundancy, and
recovery, as shown in Figure 18-3. By far the most performance oriented, it is also the most complex in
implementation and configuration.
Figure 18-3: A core/edge SAN configuration
These are all important considerations as you begin to evaluate your SAN design and implementation. Some
words of caution, however, before the details of configuration complexities overcome your planning and
design:
Identify and describe the I/O workloads of the applications you expect the SAN to support.
Understand the strengths and weaknesses of each of the major networking configurations.
Reviewing Chapter 17 will help get you started on I/O workload analysis.
Assuming you have completed a general workload analysis and have a total I/O workload transfer rate (see
Chapter 17 for guidelines on estimating I/O workloads), we can estimate the number of ports required to
support the I/O workloads. The sidebar Guidelines for Estimating the Number of Ports Required to Support
I/O Workloads describes a set of activities helpful in estimating the number of ports required.
Guidelines for Estimating the Number of Ports Required to Support I/O Workloads
The assumptions to consider in viewing these guidelines include the following:
The I/O workload transfer rate is available and accurate.
The base configuration consists of FC switch ports using switched fabric.
The base ratio available from the server to a number of data paths (for example, the number of HBAs
for each server connected to the SAN configuration).
Guidelines for estimating SAN port requirements include:
Total I/O Workload Transfer Rate (I/OWTR) / Maximum Port Transfer Capacity (MPTC) = Number of
Port Data Paths (PDPs) required
PDPs Redundancy/Recovery Factor (RRF) = Number of ports for Redundancy and Recovery
(RRPs)
PDPs + RRPs = Total Data Path Ports required
Total PDPs (with RRF) + Server Ports (SPs) {number of servers number of HBAs) = Total Switch
Ports required Where RRF = 3040%; SPs = 14
Once the total estimated numbers of switch ports are calculated, the workload characteristics can be applied
to determine the type of access and performance factors necessary. As shown in our examples, this can be
an OLTP-type application, such as our banking example in Chapter 17, or a typical web transactional workload
that may be used for processing warranty and service information. Finally, the ubiquitous datacentric
application, the data warehouse, demonstrates unique processing characteristics and is enhanced by the
basic architectural value of SAN architectures.
Switch port estimates are applied to each of these workloads, which drives the configuration into connectivity
schemes that best suit the workload. As discussed previously, we then configure and distribute the port count
into a cascading, meshed, or core/edge configuration. Through examples, we apply an I/O workload for each
configuration and discuss the rationalization for our decision, first using OLTP to demonstrate a core/edge
solution. We then move to web transactional applications supported by a meshed configuration, and finally a
data warehouse using a cascading architecture.

Applying the SAN to OLTP Workloads
What are OLTP workloads? Multiple users performing a set of consistent transactions characterize online
transaction processing workloads. The transaction mix will be simple to complex, but consistent in its
processing requirements, usually applying an 80/20 rule (that is, 80% simple, 20% complex). The transaction
type identifies its nature in terms of resource utilization, I/O content, and I/O utilization. If we use our banking
application as an example, the tellers perform most of their transactions using the deposit transaction. This
transaction verifies a customer account and adds the deposit to a suspense balance. The suspense files
would be used as input for later off-shift processing during a batch cycle to update permanent customer
records and accounts.
In this case, the simple transactions I/O content is approximately 80 bytes of customer and account
information. This is accomplished with a minimum of simple math additions and balance calculations within the
application. Although small, these are executed on a transaction-by-transaction basis and require an existing
file or database table be updated. Given that profile, each transaction generates an I/O operation to complete
the transaction process. Meanwhile, every associated I/O completes a read/write operation to a specific disk
within the supporting storage array. In the case of RAID implementation, the physical I/Os will increase based
upon the type of RAID level in operation within the storage array.
Consequently, the I/O workload is characterized by many small transactions where each will require an I/O
operationmultiple I/Os as RAID is introduced. As the transaction rate builds during peak utilization, the
workload is further characterized by additional simple transactions adding multiple I/Os, all requiring
completion within a three- to five-second time period (essentially the service level for the deposit transaction).
The workload and service level demands a configuration that can handle the peak load. This requires a large
number of data paths to handle the I/O transaction rate.
SANs are excellent choices for OLTP workloads because they can provide more data paths than any other
storage model. Given the flexibility of the storage network, the scalability of the I/O workload can be enhanced
without the addition of server resources.
SANs also provide the additional benefit of data partitioning flexibility. Given that most OLTP applications
leverage the services of an RDBMS, the ability to utilize the databases partitioning schemes intra-array and
inter-array provides the additional balance necessary to physically locate data for OLTP access.
Most OLTP applications support an operational aspect of a business. Therefore, data availability is important
given that the day-to-day activities of the business depend on processing transactions accurately. Any
downtime has an immediate effect on the business and therefore requires additional forms of redundancy and
recovery to compensate for device failures and data integrity problems. SANs offer an environment that
enhances a configurations ability to provide these mechanisms in the most efficient manner.
Additional discussion regarding recovery and fault-tolerant planning can be found in Part VI of this book.
However, its important to note that each of the workloads discussed will have various degrees of business
impact and therefore require different levels of recovery management.
The use of relational database technology (RDBMSs) defines the major I/O attributes for commercial
applications. This provides OLTP workloads with a more defined set of processing metrics that enhances your
ability to estimate I/O behavior and utilization. The use of the relational database has become so accepted
and widespread that its macro behavior is very predictable. In recognition of this, additional consideration
should be given to I/O workload characteristics relative to I/O processing requirements such as caching,
temporary workspace, and partitioning.
Dont make the mistake of overlaying the storage infrastructure too quickly. The consideration of recovery
scenarios and availability requirements should be considered at the macro-level first, followed by decisions on
how to handle workload specifics. Particular subsets of a workload, such as a complex set of deposit
transactions, may require different resources or configurations.
OLTP workloads using an RDBMS can be supported through RAID arrays using level 5 configurations. This
provides both redundancy and fault tolerance needs. RAID 5 allows the storage array to continue to function if
a disk is lost, and although running in a degraded mode it permits dynamic repair to the array, or, if needed, an
orderly shutdown of the database itself. Additional storage system features, such as caching strategies and
recovery models, require due consideration given the unique nature of the data contained within the relational
database model.
User Access
User traffic plays a significant role in defining the number of data paths necessary for the OLTP transactions to
be processed within the service level. Obviously, we need to know the estimated traffic to understand the type,
number, and behavior of the data paths prior to assigning the configuration ports. We basically need three
types of information from the end users or their representativesthe application systems analysts. This
information consists of the number of transactions, the time period within which the transactions will be
executed, and the expected service level.
The deposit transaction used in our example banking application comes in three forms: simple, authorized,
and complex. Simple requires an update to a customer table, while authorized requires an authorized write to
the permanent system of record for the account, necessary for a double write to take place when accessing
more than one database table. Complex requires the authorized transactions but adds an additional
calculation on the fly to deposit a portion into an equity account. Each of these have a different set of data
access characteristics even though they belong to the same OLTP workload.
Service levels become critical in configuring the SAN with respect to user access. This places our eventual,
albeit simple, calculations into a framework to define the resources needed to sustain the amount of
operations for the OLTP workload. From our initial information, we find there are two goals for the I/O system.
First, the banking applications data needs to be available from 9 A.M. to 5 P.M. each workday. Second, the
transactions should complete within a time frame of 5 seconds. It is important to note that although the I/O can
and will take up a great deal of the response time factors, network latency issues should not go without
consideration.
In addition, the configuration must support a high availability uptime, usually expressed in terms of the
percentage of downtime for a period. Therefore, 99 percent uptime requires the configuration to only be down
1 percent of the time. For example, if the OLTP time period is 12 hours each workday, the downtime cannot
exceed 7.2 minutes every day. The users expect availability of data, which is a foundation requirement for
meeting the response time (in other words, the data must be available to process the transactions, as well as
be available during the batch cycle to process information for updating other database tables). This also forms
the requirement for supporting the reliability factors and defines the type of storage partitioning and structure
suited to providing this.
Data Paths
Next, we need to understand the data highway required for this workload. This necessitates breaking down our
first two categories into a logical infrastructure for the workload. By comparing the data organizational model
(that is, the database type and its characteristics, as well as the byte transfer requirements) with something
called the concurrent factor, we can begin to formulate the number of data paths necessary to meet workload
service levels. The concurrent factor provides us with a minimal and logical set of paths to sustain our service
level, given the probability of all tellers executing deposit transactions at the same time.
This estimate provides a more accurate picture of the amount of resources needed to sustain the service level
in real time. In reality, the probability of all tellers executing a deposit transaction is actually quite high during
peak hours, and could be calculated at 90 percent. Therefore, for each time period, 90 percent of the tellers
would be executing a deposit transaction. Using our previous calculation, we estimate the mix of simple,
authorized, and complex deposit transactions to be 80, 15, and 5 percent, respectively.
We can develop our own OLTP SAN model by using the preceding considerations as well as the guidelines for
I/O workload port estimates. With our banking example, we estimate that 116 switch ports are required to
support the OLTP application. From this information, we can begin to model the appropriate design topology
for the I/O workloads. Our previous definitions lead us to a core/edge configuration, which supports the
maximum number of data paths into the storage arrays, while minimizing the length of each transfer for each
I/O (for example, the number of interswitch hops is kept to a minimum). This is also the configuration that
provides the best redundancy when alternative paths are needed in the event of a port interruption or switch
failure. Cost not withstanding, we can now evaluate configuration options with this logical model in terms of
switch types, port densities, and recovery options.
The value of workload identification, definition, and characterization becomes evident as we move into the
OLTP SAN implementations.
The Design and Configuration of OLTP-Based Workloads
Figure 18-4 shows an example of our core/edge configurations supporting the OLTP application. This
configuration is comprised of four FC switches, 15 disk arrays, intersystem-link ports, and an integrated FC-
SCSI bridge into a tape library. This supports our workload analysis estimate of ports using three servers with
two HBAs, respectively. It assumes a relational database that is capable of partitioning among the storage
arrays and leveraging a RAID 5level protection scheme within each array.
Figure 18-4: An OLTP workload using a core/edge SAN configuration

Applying the SAN to Web Transactional Workloads
Workloads generated from Internet or intranet sources can be very complex. Internet applications can
generate transactions from simple reads of web pages to complex queries and updates of database
structures. Based upon the processing architectures of web server software, which enable interactive web
services, many of the workloads will be almost entirely transactional in nature. However, within this
transactional model, another type of transaction type will emergethe messaging transaction.
Though messaging transactions are processed in real time, they will have a different set of processing
characteristics than typical OLTP transactions. Messaging provides an asynchronous form of processing
where the client application who has submitted the messaging transaction relinquishes any synchronous
connection with the corresponding application. The most familiar of these is e-mail, although many other
Internet transactions are message-based. This produces a different set of processing characteristics and
certainly a more complex set of I/O operations.
NoteThe emergence of asynchronous processing of applications also produced an additional category of
software termed middleware. The middleware, which has been used in both traditional interactive
application processing architectures as well as Internet/web-based applications, provides the
mechanisms for queuing a transaction for later processing. This can be for effective resource
utilization or handling larger interactive transaction requests where synchronous processing is
impossible. The use of middleware within applications can be a key indicator that storage networking
can also be an effective supporting strategy given its architecture to handle multiple workloads and,
in the case of middleware, handle multiple middleware queues.
Messaging workloads generate asynchronous I/O operations. This is actually a very simple model where an
application transaction is submitted, and a request is queued for later processing. The client, or initiator, of the
request is notified that the transaction has been accepted for later execution. The processing is executed at a
later time or when required resources become available or are provided. All this is accomplished in a manner
that doesnt require the user, or initiator, to wait for the transaction to complete and can continue to execute
other transactions. Although the characterization of the messaging workloads can be estimated similar to
OLTP as we demonstrated previously, care must be taken to account for delayed I/O and resulting byte
transfer requirements that may go unaccounted for when estimating total I/O for web transaction workloads.
NoteThe difference between client and initiator is an important distinction because other software and
internal storage bus processes operating on storage devices are initiators of messaging or
asynchronous operations.
Web Internet transactional workloads, which well refer to as WIT workloads, also contain typical synchronous
transactions, much like those described in our OLTP examples. Many older WIT workloads use file systems
for their data organizational models and specifically their web-based file formats. However, the RDBMS
vendors have made significant advances in providing a relational model database for web servers.
Consequently, many new applications utilize a relational database demonstrating characteristics similar to
databases supporting OLTP processing. The fundamental I/O resource requirement details continue to be
contained in the I/O activities relative to user access, data organization, and data paths. An analysis of each of
these areas begins to form a picture specific to WIT workloads.
In general, WITs require flexibility and multiple paths to sustain their I/O workloads. However, they are not as
sensitive to multiple hop degradation given the amount of messaging contained with the workload. Therefore,
a meshed configuration can provide both transactional synchronous support through single-hop data paths
while messaging data can sustain longer paths due to their asynchronous processing.
WIT workloads simply use the OS file system as the data organizational method. This provides a
straightforward view when determining byte transfer rates for WIT I/Os given the format of web file structures.
However, this must be augmented when considering the amount and type of delayed I/Os generated from
messaging transactions. Similar to the transactional structure of OLTP, the typical WIT I/O may be small but
the random nature of arrival rates and unpredictability regarding the number of users provides a challenging
configuration problem. This defines the need for placing WIT workloads in the most flexible environments
possible.
User Access
Defining user traffic is the most challenging activity in estimating WIT workloads. In web-based processing
environments, the classic time periods do not apply. This is especially true in retail-oriented web sites that rely
on customer interaction for sales and which are open 24/7. The requirement to provide adequate service
levels in an environment where user interaction crosses worldwide time zones means estimating WIT
workloads becomes problematic and challenging at best.
It is not impossible, however. Many software packages and services provide an accounting of user access and
traffic analysis. Though most are oriented toward information access and purchase patterns, there is sufficient
information to provide a macro estimate of I/O workloads. User access in terms of time periods, time zones,
and further seasonal factors demonstrate a SAN configurations ability to provide a flexible storage
environment that supports OLTP and messaging types of I/Os.
Data Paths
The number of data paths, as with any workload, is key to the performance of WIT workloads. The influence of
FC frame capacities may become a factor when analyzing switch port capacities, given that the nature of
OLTP may prove to be overkill in some web transactions. However, in many cases, the FC frame capacity
may become the required transport, given the increasing I/O content of web transactions which continue to
grow with text, image, and other unstructured data needed for transport. Even though the SAN I/O comprises
only a portion of the overall response time factor, given the move to larger Ethernet capacities in order to
support client IP networks, the matching components have begun to come together.
NoteFC frame payloads transport a maximum capacity of user data (approximately 2k).
Data paths for the WITs in a complex web environment, comprised of both OLTP, messaging, and batch
loads, will likely surpass the number for typical OLTP. However, the gating factor to this continues to be the
reliance of the web server file system. Its important to keep in mind that the probability of multiple data paths
may have little or no effect if client requests are queued at the file system. Consequently, there may be a need
to consider the following:
The ratio of channels per server (for example, the number of HBAs per server)
The need for SMP-enabled servers (meaning more concurrent processing)
Ensure workload balancing can be used in some form (for instance, in a workload balance software
package, processing affinity options, or process prioritization)
More detail on these options can be found in Part VI of the book.
The Design and Configuration of Web-Based Workloads
Figure 18-5 shows an example of a mesh configuration supporting the web-based workload. This configuration
is comprised of three FC switches, 18 disk arrays, intersystem-link ports, and an integrated FC-SCSI bridge
into a tape library. This supports our workload estimate using web-based consideration and using seven
servers with two HBAs, respectively. It assumes a combination of web-based e-mail files, a transactional
relational database supporting customer retail transactions, and a complement of web pages. This supports
both transactional activity through the meshed configuration where numerous paths exist to the data while the
messaging workload is supported by sufficient data paths with consideration for delayed I/O.
Figure 18-5: A web Internet transactional workload using a meshed SAN configuration

Applying the SAN to Data Warehouse Workloads
Data warehouses provide strategic and tactical business information used to direct business planning and
forecasting analysis. Data warehouses are built from subsets of operational databases, typically OLTP
workloads that form a historical database of various business activities. Users query these large databases in
the form of What if transactions that range from simple to complex questions. The answer sets of these
queries can be extremely large, complex, and form the basis for additional subsets of databases that warrant
further analysis.
A spin off of data warehouses is the data mart. Theoretically, data warehouses are based upon corporate
information and are used to drive corporate business decisions. Data marts are subsets of data warehouses
that are used to drive lower-level business decisions within the divisional or departmental levels. This is
important because the size and I/O content will be driven by use of the data warehouse at the corporate level
or divisional or departmental level. Given that the true nature of data warehousing is tied to the functions of the
relational database, there are many database rules, guidelines, and practices that determine its design,
operation, and usage. However, most of these are outside the scope of this book and although important, only
those that provide us with clues to their I/O workload analysis will be mentioned.
As you may have experienced or surmised, data warehouses can be very large. Given their sources are the
continuing updates from daily business transactions, their storage utilization has been the most
comprehensive. The capture of operational database recordsknown as the extraction, transformation, and
loading activities are driven by the complexities of the data analysis application. In other words, the more
comprehensive the analysis application, the more data is required. Many data warehouse configurations are
moving into the multiple terabyte range and, consequently, provide the foundation for why data warehouses
and data marts are good candidates for SANs. Of all the workloads examined so far, the data warehouse
provides the most complex I/O activities that we will discuss.
Data warehouses are built around relational databases; therefore, they are under the control of the users
database system of choice. Moreover, this defines the need to provide large storage capacities that scale well
in terms of pure size and I/O content. This is due to the activity of the data warehouse transaction. The DW
transaction in its simplest form accesses information from multiple database tables which span multiple
recordsFor example, in order to analyze the number of customers with account balances over $1,000 for an
entire year with credit card balances requires the database system to access a customer table, deposit
accounts table, credit card accounts, and so on. Without going into the specifics of the database design, we
see that even a simple query generates multiple I/Os, each with a large set of records.
DW transactions are serviced on a synchronous and asynchronous basis. Consequently, its important to
understand the mix of transactions and relative service levels. This brings up an interesting phenomenon as
both online and batch-oriented operations can process simultaneously within these environments. Given that
these are disparate workloads in terms of their resource requirements, it becomes a challenge to configure an
I/O system capable of handling this set of complexities.
This is where SANs come in. SANs provide the most comprehensive solution for data warehousing since the
MPP database machines. SANs, through their increased addressing capacity, provide an extensible and
scalable solution to configuring a very large database. Moving well beyond the addressing limitations of SCSI
bus configurations, the need to supply beyond 16 addressable storage devices has long been a requirement in
DW environments. SANs support a comprehensive workload with considerable I/O content given the
increased FC frame payload capacity for each I/O. Finally, the flexibility to isolate disparate I/Os through SAN
operations and configurations can sustain a disparate and complex workload.
Consequently, taking into account the mix of disparate workload characteristics, there needs to be a balance
between transactional activity, update activity, and data currency. These should be identified in the workload
analysis. The subsequent resource requirement details can be estimated in the major areas of I/O activities
(for example, the data organization method, data paths, and user access).
The use of a relational database system as the data organizational model for the data warehouse is
guaranteed. This not only defines a set of metrics that allows for greater detail in determining the byte transfer
rates, but also the physical placement of data within the storage devices. Working closely with a database
administrator greatly enhances the SAN design and operational factors. As discussed earlier, there are three
key database metrics to consider. These are database size, multiple table accesses, and updates.
Database size determines the number of storage devices necessary for pure capacity. Although important,
physical placement becomes a critical factor. Partitioning within the database, meanwhile, is key to the type
and transaction mix. Consequently, databases that span storage arrays have needed consideration in
determining port capacities and access at the switch level.
Depending on the DW transactional mix, the ability to provide adequate performance for accessing I/O
concurrently across the database is required. Databases partitioned within storage arrays (which is a given),
and those that span storage arrays (a good possibility), need to be analyzed in order to provide sufficient data
paths that meet required service levels. Just as important is the ability to isolate the data paths in terms of
concurrent access to batch operations that may be taking place simultaneously.
The significant challenge is the configuration attributes that handle the updates. DW sources are operational
databases that handle a companys day-to-day operation. Copies of subsets of data from these databases are
used as updates to the DW databases. The operational databases may or may not be part of the SAN
configuration. In most cases, they arent; therefore, care should be taken in how the update process is
executed. Usually, this is a batch process performed during an off shift. If this is the case, an analysis of the
options for getting the data into the DW is key to the success of SAN operations.
User Access
The one break you can expect from DW configurations is that the user community supported is generally
limited. Although somewhat obvious, since there are only a limited amount of company personnel assigned to
strategic planning, the user base is likely to grow significantly as the success of DW applications demonstrate
their value. Although user traffic may be limited, keep in mind that the I/O content of answer sets can be very
large. This is one area where applications like DW push the boundaries of FC transfer rates.
If we consider our previous example of analyzing the number of customers with accounts of $1,000 for the
year, with credit cards, and who have lived in the same place for more than three years, we may find that the
answer set exceeds a gigabyte of data. Dealing with such a large transaction, the number of I/Os generated
requires sufficient resources to meet the expected service level and that may be a real-time requirement.
However, if only two users generate this requirement per day, this will begin to put the requirement into
perspective in terms of analyzing the total byte transfer rates.
Data Paths
Processing transactions that have gigabyte requirements requires multiple data paths. In addition, the
necessity to physically partition the database across storage arrays requires more data paths. If we augment
the requirement further with database updates that may need to be performed in isolation, the number of data
paths becomes significant.
Although the number of data paths needed is dependent on the service level, the ability to access multiple
terabytes of data using complex queries in a reasonable time period will drive the number of data paths
required by a SAN. The alternative is a highly complex MPP environment using sophisticated parallel
processing functionality.
The Design and Configuration of Data Warehouse Workloads
Figure 18-6 shows an example of our cascading configuration supporting the data warehouse application. This
configuration is comprised of three FC switches, 18 disk arrays, interswitch-link ports, and an integrated FC-
SCSI bridge into a tape library. This supports our workload estimate using data warehouse metrics and
considerations, and supports the five servers attached to the end-user LAN. It assumes three relational
database systems supporting two warehouses, the Financial Analysis System and the Financial Planning
System. Note that the Financial Analysis data is large enough that it is partitioned into North American activity
and European activity. The outstanding issue in this configuration is the update process that occurs with
operational databases that are not part of this SAN. Considering our initial example of the banking application
from Figure 18-3, an interesting problem to contemplate is the operational activities necessary in updating the
data warehouse from SAN to SAN. This will be discussed in Chapter 20, which concerns integration issues.
Figure 18-6: A data warehouse workload using a cascading SAN configuration

Conclusions on Applying SAN Solutions
Three major points have emerged from our discussion of applying workloads to SANs. First is the SANs
scope of configurations needed to handle the most common set of workloads within the data center. Second is
the flexibility of the configuration which enables both transactional, messaging, and batch types of workloads.
In some cases, as in data warehouse datacentric loads, two diverse workloads can be processed concurrently
while providing a manageable configuration necessary to meet service levels.
Points to Consider Before Proceeding
It should be noted that SANs consume both manpower and dollar expenses, which in many cases are over
and above your existing direct-attached configurations. Cost notwithstanding, the following additional
considerations should be analyzed when contemplating the use of SANs within a data center.
Enterprise Workloads SANs are justified using enterprise-level workloads that have resource
requirements which exceed the scope and functionality of direct-attached solutions, or which exceed the
capacity of a NAS solution.
Integrating Systems and Networking Skills SANs require existing personnel be educated about
themtheres no getting around it. However, storage expertise, coupled with network expertise, will best
facilitate capacity planning, design, and installation activities.
Plan for Open-Ended Solutions SANs allow the data center to design and plan for the long term, while
making purchases for the short term. This leverages the SANs scalability and long-term viability as a
storage solution.
Preparing for the Solution
All the solutions in these examples follow a macro plan. The following steps are recommended when
implementing a SAN into a production environment. The macro plan is further defined by separate micro plans
that describe actual task-level activities, IT assignments, and target dates. Additional steps may be required
given the level of change control managed within the data center. (Refer to Chapter 23 for more details on
change control processes and practices.)
Design and Configuration
Our examples show SANs supporting three common types of workloads: OLTP, web Internet-based, and data
warehouse. It is beyond the scope of this chapter to illustrate the enterprise application, which will be defined
by your workload planning. Typically, configurations are comprised of combinations of 8-, 16-, and 32-port FC
switches, with disk arrays commensurate with storage capacities; however, its not unlikely to surpass 20
distinct systems. Another important point we have focused on in configuration management is the inclusion of
interswitch-link ports (ISLs), as well as an integrated FC-SCSI bridge into a tape library.
Test Installation
Define a test installation environment. Putting a small configuration in place provides essential first-case
experiences useful in the configuration and operation of a SAN environment. This also provides a test bed for
assessing future software and hardware upgrades while enabling an application testing facility.
Use the test installation to initiate a pseudo-management practice. Management becomes the most
challenging activity when operating the SAN. Its also the fastest moving and most quickly evolving practice
given the rapid change in software tools and accepted practices today. Detailed discussion of SAN
management can be found in Part VI of this book.
Production Installation
Develop a production turnover activity where a formal change window is established. In many cases, this may
need to be integrated into existing change management activity within the data center. Key among these is
tracking the changes made to components of the SAN. It is particularly troublesome if you formalize changes
to the switch configurations and dont upgrade critical components like HBAs, routers, and attached storage
devices.
An important aspect of the production installation is the discipline surrounding establishment of a back-out
practice. Because the SAN is an infrastructure in and of itself, its reliability can be problematic in the
beginning, as with any new technology. However, being able to back out quickly and return the production
environment to a previously existing state will save valuable time as you move into a SAN environment.
Production and Maintenance
If you have established a formal set of production turnover and change window practices, maintaining the SAN
components should become manageable. The key area in providing maintenance to the SAN components is
recognizing their complexities as interoperable components. Upgrading the fabric OS in a switch configuration
may effect interactions with the server HBAs, which in turn may impact storage bridge/ routers and other
attached node devices.
Further establishing a maintenance matrix of SAN components is your best defense against maintenance
ricochet, where upgrading or changing one component affects the operation of others. However, SANs are no
more complex than networks, and as weve discussed several times in this book, they share many of the
processing characteristics of a network. The differences you encounter will be in the devices attached to the
network (for instance, the switches, servers, routers, tapes, and so on). This means that the mean time to
defect recognition is longer than in traditional networks, given the fact that there are no clients directly
attached that will provide instant feedback if the network is down or not operating properly.
Consequently, there is a need to monitor the operation of the SAN in as active a fashion as possible. Although
we will cover this in more detail in the management part of the book (Part VI), it is important that the
information gathered during this activity play an important role in problem identification, tracking, and
resolution.

Overview
NAS is best known as a cost-effective, general-purpose storage solution. Although accurate taken from its
legacy of initial deployment, the definition remains a misnomer as the evolution of this storage device has
outgrown its provincial network orientation. NAS usage has expanded to include solutions ranging from
departmental appliances deployments to strategic data-center solutions. NAS can contribute significant value
within the data center supporting applications with over-1TB (terabyte) capacities and gigabit speeds, and
outside the data center with its bundled nature that survives end-user environments in remote and campus
settings.
NAS devices support an increasing breadth of applications as well as transparent infrastructure storage
solutions. These range from specialized engineering, scientific, or academic applications to multiple device
installations that have become standard in many Internet service provider data centers. Consequently, the
selection of NAS as the best solution for the job requires a bit more effort than throwing NAS against a storage
problem and seeing if it sticks. Applying NAS will require a similar characterization of workloads as we
demonstrated in applying the SAN solution.
This chapter offers both insight into this process and an overview of typical data-center applications where
NAS provides significant value. Well set forth guidelines for estimating NAS resources and configurations,
thereby establishing a foundation for demonstrating key configurations.

NAS Workload Characterization
When characterizing workloads for Network Attached Storage, the first step is to keep in mind the inherent
value that NAS brings to application storage environments. As described in Chapters 3 and 12, these include
the following:
Compatibility with existing network resources
Quick deployment with bundled packaging
Support for multiple file protocols
Cost-effectiveness
These characteristics appear to reinforce NAS as a general-purpose storage solution; however, it also places
some parameters around what NAS can doas well as what it cant do. Some of this was also covered in
Chapters 3 and 12, which touched on processing environments with the following workload characteristics:
An OLTP configuration supporting write-intensive applications
A high-traffic transactional-intensive environment with high I/O content
Messaging environments that use embedded databases
Relational database systems supporting distributed environments
As we review each workload characteristic and environment, both pro and con, well begin to narrow our scope
of supported applications.
What NAS Does Well
NAS provides one of the best solutions for departmental network storage requirements. In many cases, it can
extend beyond these applications and move into an enterprise orientation where user access heads into the
thousands and data size moves into the terabyte range. We examine each of these NAS characteristics next.
Attachment to Existing Network Resources
As discussed previously, this is NASs capability to operate within an existing Ethernet LAN environment. This
allows any server existing on the LAN to connect and utilize the NAS file server functions. It also allows clients
who use network files to connect and utilize the file server functions as a general-purpose file server.
Deduction: A good file server at a greatly reduced cost.
Quick Deployment as a Bundled Solution
Quick deployment as a bundled solution allows NAS to be installed swiftly on a network, where its defined as
an addressable entity and made available for immediate use. In many cases, this can be done within a matter
of minutes given the right network and network functions.
Deduction: A quick fix for storage capacity, once again at a greatly reduced cost.
Support for Multiple File Protocols
NAS supports most common remote file protocols, allowing many applications to exist transparently while
application servers redirect I/O requests to a NAS device. More importantly, this allows both clients and
servers to move storage requirements and I/O requests to an external device.
Deduction: A good device for separating storage requirements from application requirements, and if it supports
HTTP, it can be used as a cost-effective storage device for Web Services.
Cost-Effectiveness
Although I think we have introduced this concept with each of our previous attributes, it can be further
articulated in terms of the NAS bundled solution, which provides both hardware (processors, network
attachments, and storage devices) and software (the OS, file system, and so on) with several alternatives to
RAID protection.
What NAS Cant Do
It is also important to keep in mind what NAS cant do. Trying to get NAS to support things it cant undermines
its value as a cost-effective solution and potentially renders it obsolete. Here are some things to balance
against what NAS does well.
OLTP in a Write-Intensive Application
Remember two things: First, NAS is bundled with a file system; OLTP almost always requires an RDBMS that
restricts NAS from fully participating in this application. The RDBMS engine cannot run on the NAS device
because it has an RTOS and is closed to other applications. Second, the file system will hamper performance
of tables that are stored on the device, and support from database vendors using NAS in this fashion is
problematic.
Deduction: Be cautious when considering NAS storage for high-end OLTP applications supporting multiple
remote end users. These are applications using traditional relational database solutions.
Transactional-Intensive Applications with High I/O Content
NAS devices, although offering multiple network access points and multiple SCSI paths to data, remain limited
in their scalability to meet high-transactional traffic.
Deduction: You probably dont want to use NAS for your corporate data warehouse application, or any
application that requires sophisticated RDBMS partitioning, updates, and complex data access.
Messaging-Intensive Applications Using an Embedded Database
The same can be said for messaging environments like e-mail, where high user traffic coupled with high I/O
operations cannot sustain large user communities with additional deferred I/O operations that may require the
remote execution of transactions.
Deduction: Be cautious when considering NAS for your e-mail operations. Although this is a matter of
scalability, most e-mail systems continue to have a monolithic file architecture that makes remote partitioning
of the data difficult, and scaling problematic.
High-Performance Relational Database Systems
The relational model has yet to fit well with NAS systems given the optimization at the file-system level and the
singular nature of the storage array options. Relational databases at the high end of the product spectrum do
not work well with NAS storage. Given there is some value to using NAS to store database tables in
elementary RDBMS installations, enhanced performance options such as the parallel processing of queries,
sophisticated table partitioning, and adding global table locking mechanisms remain beyond the capabilities of
NAS.
Deduction: You should not consider NAS for applications that have database systems, which may extend their
requirements into advanced RDBMS functionality. Although considerations of scalability is a factor, NAS
storage servers do not have the processing services necessary to handle advanced database functions.
In conclusion, there are a number of applications to which NAS will provide significant value. These
applications are not necessarily the most highly political or visible, but they make up much of ITs
expenditures. Thus, our findings for NAS include the following:
The NAS device can be used as a storage solution for Web Services, because it does speak HTTP, along
with other remote dialects such as NFS, CIF, and future file transfer technologies such as Direct Access
File System.
NAS provides a quick solution for file servers whose storage capacity is constantly being upgraded within
specific network domains.
Given the flexibility in attachment to an existing Ethernet LAN, NAS can be utilized as a remote storage
device while controlling it through the network and vice versa.
NAS provides an excellent cost-effective storage device for archival- and HSM-type activities.
These conclusions establish disembarkation points for departmental solutions, data centerspecific solutions,
and application-specific solutions. Derivations of any one of these solutions will likely come from one of these
three configurations. A brief overview follows:
Appliance/Departmental Configuration This NAS deployment provides a storage-centric solution to
departmental file services. Figure 19-1 shows a simple NAS configuration supporting a network domain of
two servers and multiple client devices that rely on NAS storage for networked file services, shared files
such as application code libraries, and the download of business info (for example, results of large data
warehouse operations).
Figure 19-1: An appliance/departmental NAS configuration
Enterprise/Data Center NAS Configuration This NAS configuration provides a performance-oriented
system that supports multiple web servers for both Internet and intranet storage requirements. Figure 19-2
illustrates multiple NAS storage devices supporting the Web Services servers which connect clients as
they transfer data within their own network segment using high-speed gigabit Ethernet transports.
Figure 19-2: An enterprise NAS configuration
Specialized Application Support NAS Configuration This configuration supports the large storage
requirements for special applications that store unstructured data in existing file formats, archived HSM
data used for near real-time access, or image/video data used in streaming applications. (See Figure 19-
3.)
Figure 19-3: A specialized application support NAS configuration
The following are some important considerations as you begin to evaluate your NAS design and
implementation. Keep in mind that although NAS is easy to implement, and in some cases can be installed as
a quick fix to a critical capacity problem, a level of analysis may be helpful to avert potential problems.
Identify and describe the I/O workloads of the applications you expect NAS to support.
Understand the strengths and weaknesses of each of the major networking configurations.
Assuming you have completed a general workload analysis and now have a total I/O workload transfer rate
(see Chapter 17 for guidelines on estimating I/O workloads), you can estimate the number of ports required to
support the I/O workloads. The sidebar Guidelines for Estimating NAS Requirements describes a set of
activities that can be used to provide NAS implementation estimates given the size and scope of a potential
NAS solution. These should be viewed as general guidelines and used as a worksheet when analyzing your
potential NAS requirements. As always in capacity planning exercises, many data-center specifics should to
be considered in context. Your results may vary according to those specific requirements.
Consider the following when contemplating specific NAS requirements:
The I/O workload transfer rate should be available and accurate.
The base configuration will be using LAN.
Guidelines for Estimating NAS Requirements
(Total I/O Workload Transfer Rate (I/OWTR) / Max Network Performance
[1]
)/2 = Minimum number of
Network Paths (NetPaths) required
Total I/O Workload Transfer Rate (I/OWTR) / Max NAS Channel Performance
[2]
= Minimum number
of data paths (DataPaths) required
Add NetPaths to DataPaths = Total logical paths (Lpaths)
Lpaths Redundancy/Recovery Factor (RRF)
[3]
= Number of paths for Redundancy and Recovery
(RRPs)
Lpaths + RRPs = Minimum logical paths (Lpaths) required
Calculate ratio of Lpaths to DataPaths Lpaths/Dpaths = Size Factor (sf %)
Compare sf % factor to NAS sizing table
NAS Sizing Factor Table
NAS Devaice Type Data Capacity Requirement Sizing Factor
Appliance Aggregate Data is < 500GB 0 30%
Mid-Range NAS Aggregate Data is >500GB <1TB 30% 60%
Enterprise NAS Aggregate Data is > 500GB <5TB < 60 %
Conclusion Use the guidelines to calculate estimates for data paths, (number of SCSI channels), network
paths, (number of IP ports), and the NAS sizing factor. This will build in additional resources for redundancy
and recovery operations/device transfers. Use the NAS sizing factor table to find the NAS device most
appropriate for your workloads by comparing your total data storage requirements to the sizing factor.
For example, if your total data required for NAS workloads is less than 500GB and your calculated sizing
factor is under 30 percent, your best solution will be an appliance-level NAS device. Size the appliance-level
device with estimates that provide the minimal number of data paths and network paths as estimated in your
calculations.
Estimates that do not fit the table or cannot be sized with sufficient network or data paths are probably better
suited to a more scalable solution such as a SAN, a high-end, direct-attached solution using SMP, a clustered
system, or a combination of any of the three.
[1]
The MAX network performance is based upon the type of Ethernet network segment you install on the NAS
device (ranging from a low-speed base 10 to high-speed 10Gb segments). A recalculation of bits to bytes is
required to perform the calculation.
[2]
Max NAS Channel Performance is the bandwidth MB/sec burst rate specification of the SCSI channels
selected. This ranges from 10MB/sec to 5060MB/sec, depending on the SCSI type selected.
[3]
The RRF is made up of an average of 40 percent for transactional workloads, 20 percent for file servers, and
10 percent for archival and special applications.

Applying the NAS to Departmental Workloads
Departmental workloads make up the bulk of NAS implementations. These workloads are defined as users
and applications that are closely integrated in terms of physical office structure, network segments, or
application usage. Most environments are a combination of all three of these attributes. This is characterized
by a physical attribute such as all the engineers on the third and fourth floors, the entire set of architects within
the building, or the customer support representatives in the basement. They share a common attribute in that
they are physically located together and are serviced by servers that provide both network access and a
shared storage space within that server, or a combination of servers supporting that physical area.
However, the physical factor may only exist in terms of network proximity, where all the sales people at the
front of the building share a network segment with, say, the executive secretaries, and half of the customer
service department, which is located in the basement. The common network segment provides an
infrastructure again where these groups likely share file system servers that provide outside network access to
the rest of the company as well as customers and shared files within their own organizational domains.
Certainly, many of the personnel with the same physical environment and network segment are going to share
a common set of applicationsusually the ones directed by the corporation, installed by the IT department,
and used by the sales and support representatives in the front of the office and the customer service
representatives in the basement.
Common activities within the shared environment just discussed are the continual additions, upgrades,
changes, and enhancements to the departmental file servers that service the network segment. This is driven
for the most part by the growing demand for storage on the file servers. Even though the applications are
upgraded and enhanced, physical growth must increase alongside the data accumulated by the sales
department, administration, and, yes, the guys in the basement.
This continual upheaval becomes costly, disruptive, and inefficient, adding, replacing, and enhancing general-
purpose servers to accommodate additional and more sophisticated storage, just to access files.
The solution is a NAS appliance.
The data organizational model has a file system bundled into the solution (yes, thats right, a file system). File
processing characteristics are based upon the requirements that workloads place on the file system and
supporting file transport protocols. Characterizations of file system processing complexities, including simple
file processing (Sfp), quality file processing (Qfp), and complex file processing (Cfp) can be found in Chapter
12.
User Access
User traffic always plays a significant role in defining the number of paths needed on a concurrent basis for a
particular workload. The I/O content of the users transactions compared to the total workload byte transfer
rate will begin to determine the NetPaths required. These estimates are based upon the aggregate network
speeds and the type and characteristics of the Network Interface Cards used in the proposed NAS solution.
Data Paths
Data paths must include the total number of paths that the data travels. Consequently, data paths are
combined with the number of network paths from both servers and users (for instance, the NICs), and the
number of SCSI channels for each storage array that has access to server processing.
Design and Configuration for Departmental-Based Workloads
Figure 19-4 shows an example of NAS appliance configurations supporting the departmental file server
requirements.
Figure 19-4: Departmental workload using a NAS appliance configuration

Applying the NAS to Enterprise Web Workloads
Workloads generated from Internet or intranet sources can be very complex. Internet applications can
generate transactions from simple reads to complex queries and updates of database structures. Built upon
web software and providing interactive web services, many enterprise web-based workloads are almost
entirely transactional in nature. However, within this transactional model much of the traffic is generated from
the read-only access to web pages.
This plays into the strengths of the NAS model, which should provide storage for the rapidly growing number
of web applications accessed only as read I/O and which are loaded on a remote client. Web servers provide
the necessary infrastructure to redirect I/O to storage servers based upon a HyperText Transport Protocol
(HTTP) universal resource locator address. Given that NAS boxes are able to speak HTTP, this client request
along with the subsequent redirection from web indexing software can provide the requested web page, and
thus free the web server for web transactional and indexing tasks.
In some cases, a web services provider may provide a heterogeneous set of servers for web service
deployment. This is characterized by UNIX and Windows servers coexisting within the same data center. The
inherent strength of providing heterogeneous support to both server operating environments through the NAS
multiprotocol services allows us to use the NAS device as a storage solution for Web Services, given it
speaks HTTP and other remote dialects (NFS, CIFs, and so on), as well as future file transfer technologies,
such as the Direct Access File System.
NAS-WIT (Web Internet Transactional) workloads simply use the OS file system as the data organizational
method. This provides a straightforward method of determining byte transfer rates for WIT I/Os given the
format of web file structures. However, this must be augmented when considering the amount and type of
delayed I/Os generated from redirected I/O transactions and asynchronous transactions (for example,
messaging). Similar to the transactional structure of OLTP, the typical WIT I/O may be small but the random
nature of arrival rates and unpredictability regarding the number of users provide a challenging configuration
problem. In terms of NAS devices that are limited in their network and data path capacities per device, we are
driven immediately into a multidevice configuration. Consequently, this requires a logical distribution of the
total workload byte transfer rate among a set of NAS devices.
User Access
Defining user traffic is the most challenging part about estimating WIT workloads. In web-based processing
environments, the classic time periods do not apply. This is especially true in retail-oriented web sites that rely
on customer interaction for sales and are open 24/7. Providing adequate service levels across worldwide time
zones makes attempts at estimating WIT workloads problematic.
Data Paths
The number of data paths, as with any workload, is key to the performance of WIT workloads. The influence of
IP packet traffic becomes a defining factor in analyzing network interface capacities given the OLTP nature of
the WIT workload. Given the rich I/O content of web transactions that continue to grow with text, image, and
other unstructured data, a gigabit transmission may become the required transport, even though its a read-
only environment. NAS I/O comprises only a portion of the overall response time factor; limiting network
latencies with Gbe capacities helps bring the number of network paths and data paths into line.
Design and Configuration for Web-Based Workloads
Figure 19-5 shows an example of a mid-range NAS configuration supporting the web-based workload.
Figure 19-5: A web Internet read-only workload using a mid-range NAS configuration

Applying the NAS to Specialized Workloads
Engineering, scientific, academic, and entertainment segments all demonstrate specialized needs given their
increasing reliance on data that is characterized by its unstructured format. The data organizational model is
an average file system and contains information used in GIS seismic applications, image and video streaming,
and model simulations such as weather patterns and nuclear explosions. The files associated with these
applications typically require large file sizes with multiple read I/Os from many users, both local and remote.
The additional characteristic of this data is the way it gets created. On the high end, the data may come from
seismic studies and surveys where millions of digital measurements are contained within the data to simulate
a geophysical structure. The measurements are contained in files, which are later used by simulation
applications to visualize a projected oil deposit. Consequently, the data is larger and must be accessible to
those engineers and scientists who analyze the visualization. Traditional client/server storage solutions
become very expensive as storage capacities and resources beseech a general-purpose server thats beyond
the necessary computer capacities required.
On the low end, data may be generated from the increasing amount of digital images used in governmental
work. For example, images of land, land surveys, and improvements can provide a significant productivity
improvement in many small governments. The problem is that storage of the images greatly exceeds the
limitations of departmental servers employed at this level. Traditional client/server systems are very
expensive, especially when purchased just to add storage capacity.
File systems continue to be the most common level of data organization for these applications. Given their
straightforward access needs (no relational model is necessary to perform What if analyses on the data
itself), the data is formatted to the mathematical/computational operations of the application. This
characterizes a processing cycle of large data transfers followed by compute-intensive processing.
Consequently, the arrival rate of large data transfers initiated by the user may appear to be random; however,
further investigation of I/O activity may show processing trends defining the level of concurrency required.
User Access
User access is typically much smaller than traditional web, departmental, or general business OLTP
applications, and is characterized by specialists that are processing the data. The major client change in these
areas is the makeup of workstations for specialized end users with increased storage capacity at the client
level. This extends the I/O content of the transactions that are executed as well as the requirements needed
for file transfers between workstation and NAS storage devices.
Data Paths
Processing transactions that have gigabyte requirements necessitates multiple net-data paths. However,
similar to the enormous database structures described in data warehouse applications, the need to physically
partition the file system across storage arrays due to size requires more data paths. If we augment this
requirement further with updates to the files that may need to be performed in isolation, stored or archived, the
number of data paths becomes significant.
Design and Configuration for Specialized Workloads
Figure 19-6 shows an example of specialized NAS configuration supporting the seismic analysis application.
An interesting problem to contemplate is the operational activities involved in updating the seismic files,
archived files, or remotely transmitted files (see Chapter 20 for more on integration issues).
Figure 19-6: A seismic analysis workload using an enterprise-class NAS configuration

Conclusions on Applying NAS Solutions
Several major points have emerged from our discussion of applying workloads to NAS. Consider NAS not only
for cost-effective departmental workloads but also for some enterprise-level workloads that fit the NAS profile.
Integrating network skills and storage skills becomes the best educational platform for effectively supporting
NAS environments. Even though NAS can be justified very simply, and installed quickly with immediate
results, this does not relinquish your responsibility in planning out your NAS solutions.
Consider These Points Before Proceeding
It should be noted that NAS, while being a cost-effective solution, could consume both manpower and
additional dollar expenses as your workload begins to grow. Cost notwithstanding, the following additional
considerations should be analyzed as you contemplate using NAS within departmental, remote, or data-center
environments.
Departmental and Enterprise Workloads NAS configurations are easily justified for departmental
settings where storage resource requirements exceed the growth of general-purpose server processing
elements. In some cases, departmental applications expand beyond the scope and functionality of direct-
attached solutions by their increased I/O requirements. In these cases, consider bringing these inside the
data center with consideration of a mid-range or enterprise NAS configuration. Dont immediately go to
SAN justification unless the total workload I/O transfer rate and user access concurrency factors demand
it.
Integrating Systems and Networking Skills NAS solutions, like SANs, require an educational
requirement for existing personnel. Although vendors would like to downplay this aspect, theres no
getting around it. The problem is that there are few options for this type of storage networking education.
As with SANs, storage expertisecoupled with network expertisebest facilitates capacity planning,
design, and NAS installation activities, and serves as a combined template for storage networking
solutions supporting both SAN and NAS.
Plan for NAS Integration and Migration NAS allows the data center to be implemented quickly and to be
safely purchased for the short term without unduly lengthening the design and planning necessary for
SANs. However, this doesnt mean planning or foresight isnt necessary when it comes to NAS. NAS
should be viewed in the long term more critically as it migrates through various NAS capacities and sizes
with consideration for integration. Integration eventually occurs as NAS workloads work in conjunction with
SAN workloads, or as NAS workloads become integrated into a SAN solution.
Preparing for the Solution
All the solutions within these examples follow a macro plan. The following steps are recommended when
implementing NAS into production environments. As we stated with plans for SAN implementation, a macro
plan is further defined with separate micro plans, tasks, and assignments. Additional steps may be required,
given the level of change control managed within the data center.
Our examples demonstrate an overview of how NAS can support three common types of workloads,
departmental file systems, Web read-only workloads, and specialized engineering seismic analysis
application. Guidelines used in these sections can provide general direction for specific design and
configuration activities.
Test Installation
Although not as critical for NAS, it is helpful and productive to define a test installation environment. Putting a
small configuration in place provides essential first-case experiences in the configuration and operation of a
NAS device. This also creates a test bed for testing future bundled model upgrades while simultaneously
enabling a risk-free application testing facility.
Use the test installation to initiate a pseudo-management practice. Remote management is the most
challenging activity when operating the NAS. It is also the most rapidly evolving practice, with daily changes in
software tools and accepted practices. A detailed discussion of NAS management can be found in Chapter
21.
need to be integrated into existing change management activities within the data center. Key among these is
tracking all the changes within the TCP/IP network infrastructures that support the NAS device. This can save
much time when formalizing changes to the NAS configurations and integrating these into server OS and
application upgrades that modify or change file-level protocols. This includes tracking key components such
as the server NICs, network routers, and attached storage devices.
Production and Maintenance
If you have established a formal set of production turnover and change window practices, maintaining the NAS
components should become very manageable. The key area in providing maintenance to the NAS
components is in recognizing the complexities of serving multiple file-level protocols. Upgrading the server
OS, web software, or new storage management software may affect interactions with NAS NICs, file systems,
and internal storage software.
Further establishing a maintenance matrix of NAS devices is your best defense in quelling maintenance
ricochet, where upgrading or changing one component affects the operation of others. The mean time to
defect recognition can be much shorter in the NAS configurations, given the fact that clients directly interact
with the NAS devices. This provides an immediate feedback loop if the NAS device is down or not operating
properly.
As with any active component within the network, there will be a need to monitor the operation of the NAS
devices. Well cover this in more detail in the management part of the book, Part VI, but its important to note
that the information gathered during this activity, regardless of product or NAS functions, plays an important
role in problem management, capacity planning, and performance and fine-tuning.

Chapter 20: Considerations When Integrating SAN and
NAS
Overview
Chapter 20 addresses the usage of SAN and NAS as an integrated solution. Discussions will cover current
data-center realities and experiences but also look to the future uses of integrated SAN and NAS
technologies. Because many of the challenges to integration are hidden in managing multiple I/O workloads,
just the introduction of this subject can be a complex exercise. This chapter was developed with a broad
scope in order to provide a meaningful orientation for those uninitiated to storage strategies as well as prove
valuable to those experienced with enterprise storage and I/O workloads.
Integrating storage networking models is, in many ways, like developing a client/ server-networking
environment; it maintains some fundamental differences given the extreme architectures of SAN and NAS.
The fundamental differences are block access in SAN operations versus file access in NASblock I/O being
an extension of the server OS, and file I/O as a functional extension to networking technologies. Beyond these
visible differences, the challenges of microkernel compatibility, communications standards, and caching
coherency issues will require server and storage vendors to cooperate with synchronization, technology
integration, and yet-to-be-developed standards. The current evolution of business applications through the
continuing enhancement of web-enabled solutions will incorporate all the attributes of OLTP, batch, and data-
intensive I/O. These application attributes will be key drivers in the SAN/NAS product evolution. Taken in
context with current data-center infrastructures, the evolution of comprehensive SAN/NAS integrated products
will definitely be a future solution.
As combined SAN/NAS solutions support the drive to an application democracy existing in a universal
network, the integration of major components of the I/O infrastructure becomes paramount with the
development of integrated and faster I/O protocols. This chapter discusses some of the key future
technologies that will begin to affect IT storage networking solutions, along the way viewing future I/O
infrastructure technologies from two perspectives. First will be a view external to the server with innovations
such as iSCSI and InfiniBand. This will be followed by a view internal to the processor components with
advancements such as HyperTransport and Rapid I/O initiatives.

The Differences
This book, to a great degree, is about the differences between SAN and NAS architectures, technologies, and
solutions. Looking at SAN and NAS products as an integrated solution within the data center, we must view
both the differences and similarities from a different perspective-that is, how do we put such solutions together
to provide value in the data center. Moreover, what are the applications and, more specifically, I/O workloads
that would benefit from both technologies?
The most visible difference is the method in which SAN and NAS process an application's request for data.
This represents the block I/O versus the file I/O orientation of each solution.
SANs provide a storage interconnection that enables an attached server to process an I/O as if it were
communicating with a direct attached storage device. This allows the attached server's OS to execute a block
I/O even though the operation is transported over the FC protocol within the storage network. Consequently,
SANs enjoy the benefits and efficiencies of block I/O while enabling the flexibility and added bandwidth
performance of being on a network.
From the perspective of the NAS device, the application's request for data is handled through a file request,
which is in turn managed by the operating system, where the application is executed and the NAS operating
system takes the file request. However, as the request is executed from the NAS operating system, it
performs a block I/O to access the data on the NAS device. Block I/O is the native function in how the OS
executes an application I/O, whether the application is local or remote (refer to Part II for more detailed
information).
Figure 20-1 contrasts these differences to show SAN I/O as an intrasystem transaction and the NAS I/O as an
intersystem transaction. The SAN block I/O requires direct communication with the initiating operating system.
The NAS file I/O does not require direct communications with the initiating operating system, but must
communicate with another operating system where it obtains only the file attributes. Having accepted that
information from the initiating OS, the NAS OS executes a block I/O on their behalf. When viewed from an
application standpoint, the application program does not care where the data is, as long as there is some level
of addressability to locate and read or write the data. The internals of that operation, which are a block I/O, are
handled by the NAS OS where the data actually resides.
Figure 20-1: SAN block I/O versus NAS file I/O
The other visible difference is the network where each solution resides. NAS uses existing Ethernet topologies
and the TCP/IP-layered network software model, rendering it an IP-based storage transport. SAN, on the other
hand, uses the Fibre Channel-layered network software model across a variety of media transports.
Nevertheless, it remains a closed system that requires some method of translation to participate in other
networks.
The differences of data encapsulation, speed, and reliability between transport protocols of TCP/IP and FC
illustrate the challenges of integrating IP-based networks with FC-based networks. TCP/IP based upon data
communications architectures is not well suited to the transmission of a native I/O, as indicated in Figure 20-2.
The size of IP packets limits its value in the types of I/O workloads it can handle. Enhanced only through the
increased media speeds, the limitations of TCP/IP for storage equate to many of the limitations of NAS
products. However, the ability to leverage the wide area network capability of TCP/IP networks can provide
access to remote storage locations.
Figure 20-2: NAS I/O local limitations and remote strengths
A SAN's use of FC provides the basis for a channel-oriented architecture that facilitates its use as a storage
transport (shown in 20-3). The FC classes of service provide the levels of reliability to perform in environments
where data integrity is important (for example, most commercial data centers). The frame architecture of
approximately 2KB and its transfer rate of 1 Gbps moving to 2 Gbps and beyond extends FC's ability to handle
a larger set of I/O workloads. The FC protocol functions as a transport for block I/O within the limitations of
SAN networks. The issue with FC SANs is its limitation in providing wide area network usage when accessing
remote storage locations.
Figure 20-3: SAN I/O local strengths and remote limitations

The Similarities
The need to provide access to more storage resources characterizes the similarities of both NAS and SAN
solutions. Both solutions utilize a network to facilitate this requirement. The NAS and SAN products provide
multidevice access that interconnects both servers and storage. With respect to their local operations, they
both perform block I/O, as described previously.
The use of a network ties these storage models together and provides a foundation for integration activities.
Despite the evolution of data communications networks, the integration of disparate network topologies and
protocols remains with us today. Accomplished through the use of bridging mechanisms at the lower levels of
the network transport layers, data networks were able to interoperate, allowing the scalability of multiuser
networks. The evolution of these elementary bridging technologies to router products facilitated the delivery of
client/server data packets as well as sophisticated traffic management and communications reliability.
Although we have yet to see this sophistication within the storage networking world, the use of bridges and
routers to accommodate disparate storage protocols in the SAN has been an accepted and widely used
mechanism (see Chapter 15). Given their nature of communicating at the file level, NAS can leverage existing
bridge, router, and switching technologies as an inherent characteristic necessary to utilize standard Ethernet
TCP/IP networks.
In terms of value, they provide similar justification in what they offerthat is, storage consolidation, server
consolidation, centralized management, larger storage capacities, multiple application access, and storage
scalability. These value propositions center around the effective usage of each storage model, respectively.
This requires the ability to identify and analyze the characteristics of the I/O workload. As discussed in Chapter
17, specific I/O workloads are better suited to either NAS or SAN technologies. Which leads us to the
underlying question: Why does anyone need to combine a NAS and SAN configuration? Part of the answer
lies in the simple rationale as to why disparate data communications networks need to integrate. The reason is
that those users with Macs need to access the same applications and data as PC users. Further, UNIX
workstation users operating within an NFS environment need to access shared information on the PC
networks. Users of mainframe applications require a separation of complexity from legacy applications that are
increasingly integrated with front-end client/server systems. Ultimately, the Internet prompted uniform access
to information both locally and worldwide.

The Need to Integrate
If we accept the premise that storage networking will evolve to some degree like data communications
networks, then we have to accept the state of today's technologies as being at the forefront of this evolution.
Consequently, elementary and sometimes proprietary and localized solutions give way to integration, allowing
applications transparent access to storage resources regardless of the model.
If we view this from a pure user requirements perspective, the evolution of web services and the integration of
local systems inside data centers to accommodate this will drive the need to integrate storage networks to a
uniform access model. Not that any of the complexities will go away; actually, they will be more complex-just
as they were with data communications networks. However, the necessity to consider lower-level access
fundamentals, such as block I/O versus file I/O, will dissipate as increasing layers of storage access fold over
the technical uniqueness of SAN and NAS architectures.
Web services will demand an integration of I/O workload types by their very make-up. Service that combines
transactional synchronous responses will be coupled with secondary and tertiary transactions that respond
with a series of asynchronous message-oriented requirements. Each of these will require a set of I/O
workloads commensurate with its processing characteristics. However, each will be integrated in its
application processing responsibilities. Thus, we may find the future of I/O workloads being driven by web
services. Common personal and business applications will be executed by an integrated I/O system with the
flexibility to route the appropriate I/O requests to the correct set of data using the correct access mechanism-
for example, block, cache, or file.
These are the driving factors to integrate SAN and NAS. The flexibility to handle multiple workload
requirements demands that storage networking have multiple access mechanisms. The industry is starting to
experience this as NAS solutions begin to integrate both file access and block-level access. This is currently
being accomplished through the integration of NAS server configurations with FC disk arrays. This sets the
foundation for bilateral access to the data via file access through existing IP network requests and block I/O
access through the FC disk array. There are many issues surrounding these early solutions, such as the need
to provide extensions to OS and file system functions, application interfaces, and additional levels of
standards, that are necessary to get everyone on the same page. However, it's a start.

Future Storage Connectivity
As the SAN/NAS core technologies evolve, they will be affected by additional external changes happening with
the microprocessor and networking industries. These changes can be viewed from a perspective external to
the server, many of which will be network driven. Other changes will occur from an internal perspective where
technology advancements in chip-to-chip communications takes place within the microprocessor system.
iSCSI, LAN/WAN Storage
One limitation of the SAN I/O architecture revolves around distance. Performing a block I/O remotely requires
the resources of the NAS solution and the inherent overhead of file I/O processing. If there was a method of
extending the server I/O operation remotely, this would allow native I/O operations to take place without the
necessity of an alternative network and required special devices, as in FC SANs, or the special server devices
associated with NAS. This appears to be a very efficient and cost-effective solution for remotely accessing
data in storage arrays.
This also describes an upcoming standard and product solution called iSCSI. The iSCSI solution (iSCSI
denoting Internet SCSI) provides a standard for transmitting SCSI commands through an IP-based network.
This would allow servers to send a block I/O request through an existing TCP/IP network and execute the
SCSI storage read/write operations on a remote storage array.
The iSCSI configurations require special Network Interface Cards (NICs) that provide the iSCSI command set
at the storage array end (as shown in Figure 20-4). This facilitates data and SCSI commands to be
encapsulated into an IP packet and transmitted through an existing IP network, thereby bypassing file-level
processing protocols and additional server-to-server communications inherent to NAS solutions. On the
surface, this solution provides an effective mechanism for disaster recovery operations, data replication, and
distribution.
Figure 20-4: iSCSI configurations showing remote block I/O operation
However, this develops an increasing number of issues in implementing and managing an iSCSI configuration.
First is the security of sending an unsecured block I/O across a public network. An associated aspect is the
reliability of the IP packet transmission process. Overall, the ability to access data within unsecured
environments that can withstand packet transmission errors may be the I/O workload that suits this future
technology.
Extending I/O functionality beyond the server has been a goal of system vendors for many years in their
struggle to replace the aging bus mechanisms that form the foundation of our current computing platforms.
Multiple initiatives suggesting that I/O should be disengaged from the computing elements have coalesced
into the industry initiative of InfiniBand. InfiniBand provides a switched fabric environment very similar to the
FC SAN switched-fabric environment. However, the InfiniBand standard encompasses all I/O from the
processor into the external connectivity of networks and storage devices. InfiniBand on the surface can be
viewed as a replacement for the aging PCI bus technologies (see Chapter 7). However, due to its scalable I/O
infrastructure, it provides a shift in computing fundamentals by providing a switched-fabric architecture that
allows devices to communicate with processor nodes with increased bandwidth, reduced latency, and
throughput.
InfiniBand, the Universal Bus
InfiniBand is an I/O standard developed by seven of the computer industry's key players: IBM, Compaq, HP,
Sun Microsystems, Microsoft, Intel, and Dell. InfiniBand replaces traditional PCI bus technology with a
switched fabric (think network), which allows peripherals (such as storage and client/data networks) to
communicate within a network (think switched fabric) of servers. It also allows InfiniBand-enabled servers to
utilize the same fabrics (network) to communicate amongst themselves.
InfiniBand is poised to cause a fundamental change to the data center, as we know it. As its scalable I/O
infrastructure is integrated, the total cost of ownership (TOC) models and metrics will change, as will how
applications are deployed and managed. IT departments, especially IT management, must understand this
infrastructure if they hope to leverage intelligent adoption strategies. InfiniBand vendors, on the other hand,
must move toward systems solutions in order to overcome the initial wave of anxiety and reduce the
complexities of integration and adoption.
As illustrated in Figure 20-5, the I/O fabric is connected from the server through a Host Channel Adapter
(HCA), while connections to peripherals move through a Target Channel Adapter (TCA). These components
communicate through a switch that routes to all the nodes that make up the fabric. In addition, the I/O fabric
enhances communication within the fabric by using remote direct memory access (RDMA) methods to
facilitate I/O applications operations. InfiniBand links are serial, segmented into 16 logical lanes of traffic.
Given this architecture, the bandwidth for each link eclipses all other data transports now available.
Figure 20-5: InfiniBand's I/O fabric
Why is InfiniBand important? The traditional deployment of servers has come through vertical implementation.
In most cases, expanding an existing application requires the installation of a new server. Each time this
happens, total overhead and the latency needed to support the application increases. Left unchecked, the
operating system services and network overhead required to support the additional servers can consume as
much or more processing power than the application. Scalability of the application ultimately becomes non-
linear as the number of servers increases.
From an I/O standpoint, the vertical deployment of servers poses additional problems. As application data
grows, so must the required I/O bandwidth and paths needed to get to the data. If the server cannot provide
either of these, the application suffers from having to wait on data to be processed. This same phenomenon
happens when users are added to the application, or there is an increase in the users accessing the
application, as often happens with Internet applications. The required bandwidth and paths must be in place,
otherwise the application suffers by having to wait on large queues of user requests.
These conditions are further exacerbated by the inefficient usage of resources-in both hardware and software-
needed to provide an adequate configuration for supporting the workload. All of this places additional burdens
on data-center personnel, who must handle the complexity of installation, maintenance, system upgrades, and
workload management.
What we have achieved is a 'gridlock' effect. Unfortunately, planning and implementing configurations for new
applications is even more daunting.
However, InfiniBand's potential value can be significant. InfiniBand architecture addresses each of the
challenges noted earlier. Scalability enhancements, in terms of link speeds, switched peripheral connectivity,
and server-to-server connections, provide a significant boost to the scalability of an application. If we consider
the prioritization and management within link technologies driven by fabric software, InfiniBand configurations
can process multiple workloads within the same configuration, thereby reducing the vertical implementation
required by traditional server technology.
Once vertical implementation has been addressed, a new model for TCO within the data center emerges. This
is accomplished by utilizing two key characteristics of InfiniBand technology. First, InfiniBand servers only
require a CPU(s), memory, and related interconnects, while using HCAs for outside communication. This
reduces an InfiniBand server to a 'Blade' computer (a computer the size of a Plug and Play board-level
component). Smaller servers will significantly reduce required space, power consumption, and installation
complexities. Secondly, InfiniBand configurations centralize the processing of applications and provide for
more dynamic operations within the fabric. Management of applications and related resources is centralized,
which means software and resources can be shared and used more efficiently as workloads demand.
What will the price be? In realizing an InfiniBand infrastructure within the Data Center, the following
considerations should be noted. InfiniBand will require a learning curve-there's no way around this. InfiniBand
configurations affect all major components within the Data Center. This is especially true in regards to storage
and networking components. Product offerings must support systems adoption, distribution, installation, and
support. Software, as usual, will be the significant detractor. Server OS integration and services and
application optimization and compatibility are a must.
Inside the Box
This section describes the requirements for HyperTransport and Rapid I/O. There are two in-the-box
initiatives, one of which connects processor modules, thus providing the ability to interconnect multiple CPU
configurations. The complementary initiative is to provide a faster, more efficient interconnect between the
CPU components within the constructs of the board.
Rapid I/O
Rapid I/O is an industry initiative that in some ways provides an alternative to the InfiniBand standard, moving
the universal bus concept down to the microprocessor level. It provides a serial connectivity protocol and
physical link standard in connecting processor board-level components. As shown in Figure 20-6, Rapid I/O
provides an alternative to PCI or other bus-connectivity solutions that interconnect components through a
Rapid I/O switching mechanism.
Figure 20-6: Rapid I/O: a future processor module interconnects
Rapid I/O is used to facilitate connectivity of clustered systems, multifunction embedded systems, and
general-purpose systems that must communicate within close proximity of each other (for instance, within the
system chassis). The standard facilitates the connection of an extensive set of internal and external
components. However, Rapid I/O remains one of the few future technologies that takes into consideration
legacy components by providing parallel technologies that offer compatibility for those components which
require multipin physical connections.
HyperTransport
Providing a high-speed interconnect between components within the processor module has long been a goal
of microprocessor designers. The usual bus connections have been eclipsed by new serial connections that
form a switched environment. HyperTransport is a vendor-sponsored initiative that proposes a standard set of
specifications when building an internal switched bus to increase I/O flexibility and performance. Staging and
moving data in and out of CPU processing units is just as important as the performance of external I/O.
The HyperTransport standard defines an I/O protocol that facilitates a scalable interconnect between CPU,
memory, and I/O devices. As a switching protocol, the architecture is defined within a layered network
approach based to some degree on the OSI network model. Consequently, there are five distinct layers of
processing that direct how internal components communicate.
Figure 20-7 depicts an example of the deployment of HyperTransport architecture. In this type of configuration,
an I/O switch handles multiple data streams and interconnections between CPUs and memory. The
configuration is extended to support external bus and network connections such as InfiniBand, PCI, and Gbe
Ethernet. This enables processing and memory components to have scalability and bandwidth flexibility similar
to switched network environments like InfiniBand and FC SANs.
Figure 20-7: HyperTransport: a future high-speed CPU interconnect

Storage Integration
Is it necessary to have a SAN and NAS model? In the near term, both storage networking models will be with
us. As we continue to support the present application mix of OLTP- and batch-type workloads, it's necessary
to provide an optimized access method. This will continue to drive the segregation of data between file and
block level access and require increased duplication and overhead. The results will be a continuation of
storage demand and growth, albeit an inflated use of resources.
The future of SAN/NAS integration will be driven by the increased need to utilize existing data for multiple
purposes or applications. The slowing of duplication requirements is a critical requirement for any integrated
front-end storage access method. As some applications require file access from remote locations, other local
applications will require read/write access on a fundamental block I/O level. The ability to provide these
access levels through a combined storage network utilizing a file or block access will be the next storage
system.
What is missing today are the components to make that happen. Figure 20-8 shows a future configuration
where an integrated file system serves as both file- and block- level access to a common storage network.
Future components that will facilitate this solution will be a shared file system with storage network coherency,
increased state- aware disk devices, and real-time fault resiliency in recovery fabrics. The storage network
components will communicate with storage access appliances that serve applications with either block or file
access. Applications will log in and register with the storage network regarding their preferences for recovery,
access, and security.
Figure 20-8: A storage network of the future
In looking toward the future, use caution when developing storage networking configurations. The useful life of
storage networking technologies has only been a year and a half in the short end and three years in the long
term. The point must be made that we remain at the forefront of storage innovations that will be increasingly
vulnerable to both changes in microprocessor technologies and the evolving universe of web services.

Part VI: ManagementKeeping It Running
Chapter List
Chapter 21: Planning Business Continuity
Chapter 22: Managing Availability
Chapter 23: Maintaining Serviceability
Chapter 24: Capacity Planning
Chapter 25: Security Considerations

Overview
Computer systems depend largely on their keepers, whether they be embedded systems, the latest blade
servers, or the largest super computer. Computer operators and systems programmers have largely given way
to increased areas of specialization for administration, maintenance, and servicing. Although many mundane
tasks have been automated, more complex configurations of distributed systems have increased the need for
system-oriented personnel to monitor and keep the servers running. This is because, no matter how
sophisticated they become, computer systems remain electronically controlled components, susceptible to
both human and mechanical error.
The basis for these activities revolves around the accepted disciplines of managing systems. These are
known as the five systems management disciplines and can be applied to any macro system in operation.
These disciplines include the following:
Performance Management The activities associated with the monitoring, tuning, and analysis of a
computer system's (or subsystem's) performance.
Capacity Management Analysis activities that drive the planning, design, and implementation of
computer resources necessary to maintain established service levels.
Configuration Management Activities that surround the design and implementation of a computer
system; this includes ongoing support of system enhancements and upgrades.
Change Management Driven by performance and configuration management activities, these practices
monitor and supervise changes that occur to computer systems.
Problem Management Activities that track, analyze, identify, and resolve system problems.
Before you object to the absence of applications management, security management, network management,
web management, and others, I should first explain how these topics relate to the subject. The basic
disciplines of systems management, as discussed previously, can be applied to any logical definition of a
system. Consequently, targeting business applications for management can, and should, have the same
disciplines applied to it-in other words, performance, capacity, change, configuration, and problem. By any
other name, they usually do. However, particular topics, such as security, add a different dimension to
management practices given they are so integrated with the supporting infrastructure. Security, which will be
discussed in Chapter 25, should be regarded as a supporting system, like backup and recovery, or database
administration. Even here, foundation disciplines can, and should, be applied.
As discussed in Chapter 1, the difference between enterprise applications and programs and utilities that
support the systems' infrastructures cannot be overlooked. They form synergistic relationships between the
productive and the necessary. Applications drive the business and form the primary components for
workloads. However, the performance of computer systems depends on the effective interaction between
productive work and overhead. Although many systems programs and utilities are referred to as applications
themselves, they often offer productivity enhancements to the systems management activities and therefore to
the effectiveness of the IT staff. However, they do nothing for the bottom line. In analyzing basic systems
management disciplines, its important to remember that they do not exist purely on their own. Without the
bottom line application of computing for some productive processes, the activities that monitor, track, and
analyze computer systems would be non-existent.
Although we often consider the management of business applications to be the pinnacle of systems
management, there are many other scientific, academic, and government endeavors that require activities just
as stringent as general business computing. Consequently, the five systems management disciplines are not
aligned to any vertical market or work orientation.
Computing continues to be a general-purpose discipline where applied technology and operational practices
migrate within the society of computer professionals who perform management operations. Practices that
worked well in scientific areas have migrated into business over time, with both government and military
agencies using unique database management, just to name two.
The point is that computer system management is an iterative process and continues to evolve through cross-
pollination with different application bases, such as storage networking. Given that storage networking is a
different storage model, however, existing management practices, though relevant, need to be molded to
support a new infrastructure, which is what storage networking creates, a new application of storage.
In evaluating management of storage networking, we must apply the five disciplines. Figure 21-1 shows how
these trans-technology practices are applied to the ever-increasing diversity within the data center further
separating the systems into specialized practices. However, as Figure 21-1 also points out, storage
networking draws heavily from data center practices and interfaces with each in order to be an effective
infrastructure.
Figure 21-1: Fundamental systems management applied to storage networking
Planning computer systems to meet business continuity requirements has been one of the most tedious jobs
within Information Technology. In the past 30 odd years of established data centers, business continuity
responsibilities were ill-funded exercises in theoretical computer systems planning. However, as the world's
socio-economic conditions began immersing computer systems within business operations, whether for the
sake of competitive positioning or national defense, the discipline of disaster recovery became a more
imperative science.
Computer storage has always been the linchpin of any recovery scenario. From legacy systems to existing
blade level embedded servers, all are linked to the vital circulation of data. Without it, the computer
infrastructures they create would be worthless. Historically, the storage component was considered part and
parcel to the holistic system that supported the evolving set of automation we have grown to depend on. Only
recently has this link been recast into an infrastructure of its own. Storage networking in its inherent value has
caused many to rethink their data center continuity plans, change their processes, and modify and enhance
their solutions according to this new storage paradigm.

Defining the Environment (Putting Storage in Context)
The objective of business continuity is to facilitate uninterrupted business support despite the occurrence of
problems. This depends on the problem, of course-it could be a small problem only effecting a single user
application through an operative disk drive, or a complete system failure where the outage is caused by the
entire SAN or NAS configuration being down. However, it could also be a site disaster affecting the entire data
center. Certainly, the categorization of the problem is important, but the plans in place should be able to
provide a road map, resources, and ability to hopefully recover from any incident.
Storage has traditionally played a key role in any recovery scenario. Without the data, the applications are
useless. Consequently, most recovery scenarios center on making the data available as quickly as possible so
business applications can continue the company's operations. Without stating the obvious, this requires the
ability to replicate data throughout an infrastructure that enables everything from a micro recovery to a macro
disaster site deployment.
Traditionally, with storage being tied directly to the server, this rendered storage problems and outages as
associated problem instances-meaning they were viewed as a server outage and thus were combined with all
the problems and outages associated with a server that's down or inoperative. As storage networking
configurations create their own infrastructures, supporting more independent external components, they
become active participants in the management disciplines. This redefines the data center environment and
must be reflected in the continuity planning.
Figure 21-2 reflects the change from associated server/storage components to separate and discrete server,
network, and storage components. This describes a data center that has divided its processing environment
into three discrete infrastructures: servers, networks, and storage. Although many data centers have evolved
into viewing and managing their environment this way, the separation of responsibility by operating
environment (for example, UNIX, Windows, MVS, zOS, and Linux) must contend with the integration effects of
heterogeneous storage networks.
Figure 21-2: Integrating storage networking as a major infrastructure
Categorizing Interruptions in Service
The plans that are designed and implemented all depend on the magnitude of the problem incident. Therefore,
it's important to distinguish and categorize the types of storage problems you are likely to encounter as well as
larger problems that affect an entire infrastructure. This analysis provides a realistic view of the amount of
insurance (redundancy, failover, and recovery methods) you need.
The interruptions in service that most commonly reflect outages can easily be lumped into two distinct
categories: hardware failures and software problems. Hardware failures can be categorized through the
inventory of storage networking devices and subcomponents, as well as the subsequent development of a
matrix that reflects the causal relationship to each. Software problems can be organized the same way,
although the causal relationships may assume complexities that are unnecessary to the task at hand.
Because of this, it's best to be reasonable about what the system can and can't do when developing this list.
An enhancement to existing systems information will be to develop a matrix for SAN hardware components.
Among necessary items such as component, name, vendor, serial number, and model, this type of list
provides an excellent location to indicate a priority designation on the component in case of failure. Given our
Plug-and-Play component-driven world, most problem resolutions fall into the category of a field replaceable
unit (FRU), which provides an excellent beginning point to the categorization of the list. The value of having
readily accessible and up-to-date SAN hardware information will enhance all management activities
associated with the SAN hardware.
A corresponding matrix for SAN software configurations can be just as important. The relationship between
switch operating system releases, node firmware, and server OS releases is necessary for problem
identification and resolution. Keep in mind the SAN and NAS will place the overall solution in a multivendor
scenario. The more information you have at hand, the better you can facilitate the problem resolution
activities. Like the hardware matrix, the software list and categorization can articulate a level of priority and
failure relationship to software elements as well as affected hardware components.
Work from the Top Down
Laying the groundwork for business continuity planning in this manner allows you to view the enterprise in a
macro level. This enables the identification of a set of requirements to determine how a storage networking
infrastructure fits into the data-center environment. This guides the level of planning that needs to occur at the
micro level of storage continuity as you deal with storage recovery and, more specifically, the details
surrounding storage networking.
If we refer to Figure 21-1, the ability to analyze the macro level is fairly simple. For example, go through the
failure points from an externalization view first. This can be done, as shown in Figure 21-2, by performing a
'what if' scenario on potential network outages. If the lease lines into the NAS servers go down, what
applications go offline? Asking further questions qualifies the configuration for potential continuity insurance.
For example, is the affected application or application service level compromised? If so, what is the recovery
scenario?
In this manner, by working through the external scenarios, you can begin to develop a picture of potential fault
points, their effect, and the basis for continuity insurance. In our example, with the leased lines being down,
connecting any or all the remote offices would compromise their service level. Therefore, a level of recovery
can be planned as insurance from this type of incident.
Participation
Planning and design of business continuity processes, resources, and environments require the participation
of a diverse set of individuals. Although it's imperative to have the business users present, this may not always
be the case. However, it should be the case that at the very least your colleagues in the data center
participate, support, and even drive the plans and processes you develop. After all, the storage infrastructure
is a key requirement for the continuation of automated business processes.
Without the participation and input from the business users you support, the plans you develop are considered
'out-of-context.' While that should not end your continuity planning, it puts the judgment of business application
value and the assignment of recovery and redundant resources in making the storage infrastructure available
totally within the IT organization. The out-of-context planning usually gets attention when the cost of business
continuity is presented to the company.
The following includes additional detail and best practices when planning within your IT organization or within
the user community:
Planning out of context This provides a level of difficulty in determining the financial and operational
impact placed on application outages, which may be accomplished through effective planning sessions
with application designers and development organizations. It's best, as stated previously, to provide a
robust continuity plan that encompasses all the production processing and related data, and then present
this to company management with the associated cost. This will quickly lead to participation of the end
users who ultimately have to bear the burden of these costs.
Planning within context Planning continuity with the participation of the end users presents a much more
realistic view of recovery and redundant requirements. Aside from a formalized project for business
continuity, this should be integrated into existing capacity planning activities. Aside from formalized
meetings, this offers a less intrusive method of gaining the information. This can be accomplished through
activities such as worksheet-based surveys to end users that provide a macro view of the business impact
and service level requirements.
Planning with our colleagues At a minimum, the participation of your colleagues within the data center is
imperative. As pointed out in dealing with out-of-context planning, the involvement of application design
and development management should be involved to provide additional analysis on the value of the data
and application service levels.

The Role of Storage Networking in Business Continuity
Storage networking, in changing the reality of storage deployment and access, shifts the traditional views of
business continuity planning. Data becomes centralized, consolidated, and maintained within its own
infrastructure (for instance, SAN and NAS configurations), and as a result, it has also become more distributed
within remote areas of the company. Fundamentally, the change in technology requires different tools or at
least a modification of traditional tools such as backup/recovery software, data replication, and archival tools
and products. Regardless of these changes, many of the fundamental storage planning activities should
continue to be applied to storage networking configurations.
The following is an example of the types of information that should be gathered in evaluating the business
continuity planning requirements.
Data ownership Data ownership continues to be more of a political question than a pragmatic one.
However, finding the owner of the application data can provide the keys to the value, integrity, and
necessity of it.
Application usage Finding what programs access the data can provide the means to uncovering data
that may be overlooked in continuity implementations. For example, there is always that one file that only
gets accessed once a month by the inventory control applicationi f its online. Consequently, not
including this file in the regular backups, disaster recovery copies, or archival process results in an
unhappy recovery experience.
Data utilization This exercise provides a similar value to the application analysis. Not only does this
uncover data that has not been accessed in months, quarters, or even years, it also provides a side
benefit of increased storage utilization through the identification of unused data that can be archived to
tape.
End users Relating the data to an end-user department, division, or customer will also put another
important piece of the continuity puzzle into perspective. Although this may seem clerical and
administratively oriented, the proper placement of data within the complex structure of a SAN or the
remote environment of a NAS are important to physical determinations of continuity configurations.
Current infrastructure Understanding the current storage infrastructure, and applying the preceding
bullets, provides a road map into what areas are most at risk. In terms of SAN and NAS, it will provide
valuable insight into the placement of data from a NAS to a SAN and vice versa.

How Much Insurance Do You Need?
At some point, usually after you inventory what you have, estimate your requirements for continuity activities
and resources. This exercise aids in assigning a value to the modifications in the current storage configuration
installed, the additional redundant configurations you need, and other additional resources set for
implementation in the event of trouble.
Level of Redundancy
How much of the production configuration needs to be replicated within what time period will be evident from
your continuity analysis. In estimating the level of redundancy, it is necessary to understand the relationship to
the minimum data capacity required, the number of users, and the level of access. For example, if you plan for
full redundancy, then the data capacity, number of users, and access are simply duplicated in an adjacent
configuration. However, this is rarely afforded given the cost, and therefore requires a level of adjustment that
takes into account the type of outage, and the level of business operation required for a particular time period.
For example, for unscheduled outages, the level of operation must be maintained for deposit transactions,
with the related data available. However, the other transactions may be down for a period of no more than 15
minutes for a single processing cycle. Therefore, configurations have to reflect a level of redundancy to
maintain the deposit suspense transactional data while the other transaction data, customer service, and
portfolio services should have fault tolerant systems with a mean time to repair of 15 minutes per 12 hours.
Level of Requirements-How Much It Will Cost
In calculating these requirements, you must not only account for new configurations (there to meet necessary
service levels), but also those modifications-reflected in existing production configurations-that accommodate
redundant systems. In terms of NAS, this may be the addition of a redundant NAS device that mirrors the
data. Within the more complex SAN configurations, requirements include: upgrades to switches to add
expansion ports, E_Ports, software to enable interswitch linking, trunking, and redundant switch configuration
and related devices.
Level of Associated Requirement- How Much It Will Really Cost
One of the most challenging activities within this exercise is the inclusion of additional hardware and software
components you didn't think about initially in contemplating storage configurations. These expenses include
additional software licenses, upgrades to existing software, and the cost of new devices such as adapters and
redundant servers.

Storage Design and Implementation of the Business Continuity Plan
As you develop a new plan for the business continuity of storage networks, or modify an existing plan, there
are key areas to consider that will enable the network to continue operating in a variety of circumstances.
Among these are the integration of business continuity requirements into capacity planning disciplines, the
effective tracking of storage system and device outages, and a realistic evaluation of sustained outage
assessment.
On the technical side, the ability to effectively monitor performance and capacity of existing configurations,
though challenging, is imperative (see Chapters 22 and 23). This includes an inventory and evaluation of the
existing capabilities of storage functions (for example, RAID level support, environmental limitations, and
remote settings).
SAN
The main design considerations for the SAN are the configuration of core switching and estimates regarding
new redundant switches for recovery processing. Within this category are the number of expansion ports, the
availability of dynamic ports and paths, and port utilization estimates. Secondary to the core switch is the
storage strategy to facilitate replication of recovery data, the time and state requirements of replicated data,
and the synchronous (or asynchronous) nature of this process. Thirdly, is the ability of the reconfigured
infrastructure to handle the adjusted I/O workload.
It should be noted that many additional items are required to develop, design, test, and implement a full
business continuity plan for SAN. These include server sizing, recovery site preparation, application portability,
and required application software and subsystems, such as compatible file systems and database systems.
Although outside the scope of this book, these items should be integrated into a well-designed plan and not
overlooked.
NAS
NAS will be the most deceptive in preparing for business continuity planning. Given its ease of installation and
maintenance, its possible to overlook its potential problems. Design considerations should center on the
requirements for data redundancy, and result in configuration to mirror production data either synchronously or
asynchronously. This requires a NAS-to-NAS device configuration and should be driven by the requirements
for complementary or compatible devices.
In addition, it will be necessary to provide redundant capacities and daily backup protections for respective
devices. Over and above these are fundamental considerations to ensure that capacities are allocated so that
as a failover scenario is implemented, the realities of the continuity plan workload can be handled.
Remote vs. Local
Redundant configurations that are local must adhere to cabling limitations. Remote configurations must
maintain a strict level of monitoring to ensure the data maintains the prescribed level of consistency. Local
redundancy costs can be much less than remotehowever, full failover clustering systems can move costs
well into the range of remote disaster recovery services.
Backup vs. Disaster
It is important to understand whether the configuration will support a local system or unit outage, or have to be
part of an entire disaster recovery scenario. This simple characterization of the business continuity solution will
result in a configuration direction focused on the appropriate support.
Expertise and Configuration Data
Trying to not overlook necessary components in assembling an alternative storage configuration to support a
business continuity solution is a constant worry. However, the successful operation of disaster recovery
storage configuration may never see production processing and hopefully never be used in a disaster
scenario. Even so, these are configurations that certainly should be tested. A successfully tested configuration
should provide the required information of configuration settings, files, and addressing schemes including
storage arrays, switch ports, LUN assignments, and zonings used.
Many components of the backup storage infrastructure may have to be re-configured back into a production
environment or used in an alternative configuration. If this is the case, or might be the case, be sure you
document the configuration information for later access. Its best to rely on procedures and current information
than ad hoc expertise to reinvent the backup configuration during each disaster recovery test.

Overview
Availability is driven by capacity and performance. Without consideration for matching the capacity of the
storage size and configuration to the workload, availability becomes a guessing game. Consequently, there
must be a sound basis for the existing storage capacities and configurations. However, taking reality into
consideration, using the most overlooked work axiom-do it right the first time- the existence of a sound
capacity plan for an initial configuration may not be the case. On the other hand, storage networking, SAN
more than NAS, is a new solution and a learning curve always begins with mistakes. Therefore, this chapter
begins with the assumption that mistakes have already happened, can easily happen, and in the best case,
can be avoided.
The premise of availability for storage networks is the application of the five systems management disciplines
in an iterative process. These will be discussed as they occur in a real-time continuum, but we'll also view
them in a cyclic manner. The following illustration offers a peek into the activities that make up the cycle. As
we can see, a capacity model and implementation may set the basis for the start of the cycle, but may not
have much to do with the reality of daily data-center operations. Performance management is influenced by
both configuration and change activities, and given its investigative characteristics, forms a logical beginning
to availability.
The performance management activities are supported by external events that are tracked through problem
and application management systems. All of these activities develop a synergistic relationship that cycles back
to influence the capacity model. So you see, it is, and continues to be, an iterative process in providing
available systems.
This chapter discusses these interrelationships as they are related to the SAN and NAS environments. By
applying activities and concepts from systems and network management practices, the availability of storage
networks characterizes their fundamental difference compared to existing direct-attached storage models. In
addition, these discussions begin to demonstrate the immaturity and challenges still evident within this area of
storage infrastructures. Focusing on both fundamentals of systems, and storage management, the scope of
these discussions illuminate key issues and offer insight into providing availability in storage networking
environments.

Availability Metrics
Providing a measurement of services within the storage network requires some type of metric. The most
encompassing is the service level. Service levels exist on several levels. There are specific agreements with
end users that define the level of service the IT department provides. There are internal service levels that are
managed to provide a working measurement of systems and networking performance. Indeed, working with
existing service levels for storage, if they exist, may be either too specific or fail to have sufficient depth to
cover performance beyond pure capacity measurements. The key is the establishment of meaningful
measurements for the SAN infrastructure that encompasses both storage systems and network specifications.
Establishing service levels for storage network configurations requires the integration of dependent system
functions, related network activities, and specific storage elements as they come together to form the storage
network infrastructure. First and foremost in the development of these metrics is the association of the
workloads. A discussion of estimating workload I/Os can be found in Chapter 17. This is an important exercise
because it forces discipline regarding establishing the users side of the equation and assists in integrating
storage configuration service levels into supported applications.
As discussed, two forms of service levels should be developed: an external articulation of services to the
supporting business units and end users, and an internal service level that supports both the data center
infrastructure and your application development colleagues (DBAs, other systems administrators, and
business continuity teams). The external service to the business units is explained in the following section.
External Service Level for Storage Networks
The following can provide a data availability framework for working with external service levels.
Data Availability
This component defines which data is referred to with the service and when it will be available. In addition will
be a understanding as to when the data will be unavailable. All data becomes unavailable at some point
because of the need to back up files and volumes periodically, and because of the occasional system upgrade
that makes all data unavailable while hardware, operating system, or application software is upgraded. These
are typical of general service levels and should be included within those commitments. For storage service
levels, however, the data center should note that particular availability requirements will be subject to
additional or similar instances where the hardware, switch OS, or a management application will be upgraded.
These may require the user data to be unavailable during those times.
Data availability is extremely important to the data center. The ability of meeting external service levels will
center upon IT activities to define the service levels along with supporting the service levels.
Uptime This metric is usually expressed as a percentage of the data availability period that the data
center has committed to. This is driven by the need to keep data available for particular applications, user
transactions, or general-purpose access (as with network file systems), and reflects the sensitivity of the
business application to the companys operations. Examples are expressed in terms of 99.999, the five
nines to distinguish that the data is committed to be available for that percentage of the overall time period
it has been committed to. In other words, if an OLTP application requires a 99.999 uptime during the
transaction period of 6 A.M. to 6 P.M. PST to support the operations of both coasts, then the percentage
is applied to that 12-hour period. Therefore, in this example, the OLTP application requires the data be
available 719.9928 minutes of that 12-hour (720 minutes) time periodeffectively rendering this a 100-
percent uptime for the data.
Scheduled Downtime This metric defines the known time required to perform periodic maintenance to
the data. Knowing this up front provides additional credibility to the storage infrastructure and reflects the
time the data is unavailable to end users and applications. The important distinction is that the same data
used in the transaction system very likely will be the same data that is accessed in batch mode later that
night. Although transparent to the end users, this can be critically important to internal service levels.
NoteIts important to note the distinction between data availability uptime and response time. The
main purpose of the storage infrastructure is to keep the data available for the required
applications. Given that actual application response time is made up of a series of components
that include the operating system, database, and network, the ability to provide metrics beyond
the availability of users is beyond the storage infrastructures scope. However, the commitment
to I/O performance must be reflected in the internal service levels leveraged as a component of
the overall response time.
Data Services
This component of the service level defines the services that the storage administration offers.
Although the services listed next are traditionally accepted services of the data center, the storage
infrastructure will accept increasing responsibility in meeting service levels surrounding these items.
Backup Service Operations to copy specific sets or groups of data for recovery purposes due to data
corruption or service interruptions.
Recovery Service The corresponding recovery operation that restores data compromised due to data
corruption or service interruption.
Replication Services Services provided by the storage infrastructure that send copies of user data to
other processing infrastructures throughout a company. Generally used to copy data to remote offices for
local processing requirements.
Archival Services Services providing periodic copies of data for necessary legal, local policy, or
governmental archival purposes. Used in conjunction with the backup service, they provide a duality of
service.
Disaster Recovery Commitments
As part of the business continuity plan, these services and service levels offer data availability during the
execution of the disaster plan.
Consider the key storage items listed next when developing service levels for business continuity and disaster
recovery plans.
Data Recoverability Matrix Its unlikely that all data will be covered under a disaster site recovery plan.
Therefore, a matrix, list, or report should be available regarding what data is available at any point in time
prior to executing the disaster site recovery plan. The service levels and services here are to some
degree a subset of services available during normal operations.
Data Uptime The same metric as the normal operating data uptime percentage and predefined
processing time period.
Data Scheduled DownTime The same metric as the normal operating percentage, but using a
predefined processing time period.
Restoration of Services A new level of service driven by the schedule for total restoration of services for
a disaster site recovery plan.
User Non-Aggression Pact
It is always helpful to reach an agreement with a user community that accounts for the unexpected. This
allows both parties to agree that unexpected circumstances do happen and should they occur, they should be
resolved with mutual understanding and changes in the existing plan. This can be critical given the volatility of
SAN installations and NAS data-volume requirements.
The following are a few of the unexpected circumstances that are worth discussing with end-user clients.
These can form the basis for a non-aggression pact that will benefit both parties in managing to the agreed
upon service levels.
Unexpected Requirements The most common set of unexpected circumstances is the unforeseen
application that requires support but which hasnt been planned for. In terms of affecting the SAN
installation, this can be costly as well as disruptive, given the scope of the new application storage
requirements. Within the NAS environment, this is one of the strengths NAS brings to the data center. If
the requirements can be handled through file access and NAS performance and capacities, then the NAS
solution can be used effectively during these circumstances.
Unforeseen Technology Enhancements This is the most common circumstance when the initial SAN
design proves insufficient to handle the workload. The requirement to retrofit the SAN configuration with
enhanced components means additional cost and disruption. This can at least be addressed with the
understanding that new technology installations are available.
Mid-term Corrections It is likely that any enterprise storage installation will experience either one or both
of the preceding conditions. Consequently, it is extremely important to build into the user agreements an
ability to provide mid-term corrections that are an evaluation of the current services and corrections.
Internal Service Levels for Storage Networks
Within the storage infrastructures, and especially the storage networking areas, the support is both to end
users as well as other individuals in the data center. Therefore, there will be an internal set of service levels
that support the other infrastructures within the data center. These at a macro level are the systems
organizations, of which storage may be a component, the support of systems administrators, and from which
web masters, systems programmers, and system-level database administrators (DBAs) will be critical. On the
applications side, storage remains an integral part of applications programmers, analysts, and maintenance
programmers requirements. Certainly, we cant forget the network administrators, help desk, and network
maintenance people.
The following is a more detailed discussion on best practices for negotiating and determining service levels for
internal IT co-workers and management.
Storage Capacity The most common requirement for the data center is the raw storage capacity
necessary for applications, support processing, and database support, just to mention a few. A service
level here will prove very productive in staying ahead of user requirements and system upgrades that
require additional raw storage capacity. This is critical with SAN and NAS configurations, given that each
has attributes that make upgrades more support-intensive. With the manipulation of switch configurations,
zoning, and LUN management upgrades, storage capacity in the SAN environment is not a snap on
activity. Although NAS provides a much easier way of increasing storage, the bundling of the storage
device may provide more storage than required and external network performance effects may be more
intense and challenging than they first appear.
Data Availability This service level is similar to the end-user service level, and in many ways forms the
foundation for the application service level with the same end user. Over and above the commitments for
production data, which should be addressed with the supporting applications personnel, is the need for
data to be available for testing, quality assurance functions, and code development.
Data Services Again, similar to the end-user services specified in the external functions, are the services
provided to assure backup/recovery and data archiving. These services include operating system copies,
code archives, and backups for network configuration data to name a few. Relating these to the SAN
infrastructure initiates an increase in the complexities of recovery operations. Given these installations are
new and require additional procedural recovery operations to be developed, a learning curve is expected.
Storage Reporting Key to internal constituencies, this provides the basis for monitoring data usage and
activity within the storage network configurations. It sets the stage for effective housekeeping activities by
establishing either new or preexisting quotas, archival policies, and ownership tracking. In addition to
these administrative tasks, it establishes a tracking mechanism to help fine-tune the storage configuration.
This is key to the SAN environment where volume allocation, physical data placement, and
systems/application access are controlled through the storage support personnel and your configuration.

Implementing the Plan
Now comes the fun part: the implementation of the storage network configuration. Its time to develop or
redevelop the SAN or NAS configurations to suit your particular service levels. The configuration options,
driven by service levels and both internal and external requirements, demand a particular level of availability.
These are generally categorized as on-demand availability, highly available systems, or fault-tolerant systems.
In the end, sensitivity to the end user is an external call and end users obviously have final say in determining
their value. These categories are discussed next.
On-Demand Availability These systems support 80 to 95 percent availability for defined processing
periods. They generally support transactional systems that do not significantly impact the daily operations
of the company if they are down for a reasonable time. Most of these systems are data warehouse
systems that are prone to outages given the level of data and their reliance on background processing.
Others are those that provide purely informational services to internal users of a company, such as
intranet systems, inventory, and educational systems. Figure 22-1 shows a typical cascading configuration
supporting a data warehouse.
Figure 22-1: A cascading SAN configuration supporting on-demand availability
Highly Available These systems require 99.999 percent uptime, and given the right resources they can
achieve this. They are characterized by applications that have a significant impact on the companys
business if they are down. This includes systems such as OLTP financial and banking applications (for
obvious reasons), retail and point-of-sale transactional systems, and customer relationship systems that
allow customers to place orders, as illustrated in Figure 22-2.
Figure 22-2: A core/edge configuration supporting a highly available CRM application
Fault Tolerant These systems cannot afford any downtime and must be available 24/7. This includes
transportation systems such as those supporting air, emergency, mass transit, medical, and emergency
911 systems. For obvious reasons, these need to be supported by fully redundant systems that can
provide full failover capability so operations remain uninterrupted. Figure 22-3 depicts a redundancy mesh
SAN configuration supporting an emergency medical response application.
Figure 22-3: A redundant mesh SAN configuration supporting a 24/7 application
If we contrast these availability categories to SAN and NAS configurations, we find some simple guidelines
can be applied to orient them toward appropriate configurations.
On-Demand AvailabilityNAS and SAN
NAS configurations for on-demand availability can be contained with entry- and workgroup-sized NAS devices.
Availability levels in the 80 to 90 percent uptime range are generally acceptable in supporting shared files and
network file services for PCs. Given the reliability of the NAS devices performing this type of work, the storage
availability will likely be close to the five 9s; however, the variable in keeping these service levels is the
external effects of application servers and the local network. Given that scenario, if the external effects are
managed within the internal service levels through agreements with your colleagues in the systems and
network areas, and external agreements with end users (see Chapter 24) make these effects known, the NAS
entry and workgroup devices will perform well.
SAN configurations supporting these areas can be simple or extended cascading configurations where the
level of path redundancy is not required, and the performance factors are more oriented toward availability of
the data. This will largely be a matter of the reliability of the FC storage devices rather than the switch
configuration; however, with todays RAID array functions and their support of larger data segments, the
reliability of these devices will be quite high. Because of this, data uptime can exceed 90 percent.
Highly Available SystemsSAN maybe NAS
For systems that require the five 9s, the more robust the underlying devices and the more flexible the
configuration, the better the solution. As these systems will likely support the use of RDBMSs for data
organizational models, the use of NAS becomes somewhat problematic. In terms of SANs, this means an
entry point at which the robust nature of its storage devices and the flexibility of its switched fabric begins to
show real value. However, the trade-off is increased complexity, as support for these availability models
requires a core/edge configuration to ensure performance that includes consideration for interswitch duplicate
paths that are necessary for redundancy. In some cases, the entire SAN configuration may need to be
duplicated within a cluster operation, whereby a standby SAN configuration holds the failover system. In the
end, this is very expensive and not recommended unless absolutely necessary.
Fault-Tolerant SystemsSAN only
Systems requiring full redundancy may either be highly proprietary systems that provide multiple levels of
failover (such as 4- to 8-way clustering systems with OS extensions to synchronize the failover functions), or
systems with a single failover used solely as a standby. Other systems may be fully fault tolerant in a tightly
coupled system of MPP or high-end SMP architectures (see Chapters 7 and 8 for additional information on
MPP and SMP configurations). Regardless of their configuration, these systems require storage. Many of
these systems, being UNIX-based, can participate in SAN configurations leveraging the ability to access either
shared storage at the hardware level, more sophisticated (and proprietary) shared data systems at the
software level, or a combination of both. This brings the value of SAN configurations further into focus.
A word about NAS when used in fault-tolerant environments. Its not true that NAS solutions cannot be used in
these environments; there are several caveats that must be considered. Among these are the uses of an
RDBMS within the application, and subsequently in housing the database system and user data. If thats the
case, NAS is more likely to be problematic given it has limited abilities with the relational data organizational
model (see Chapters 9 and 11 on NAS architectures and NAS Software, respectively). In addition is the
interconnection using a TCP/IP-based network with a speed fast enough to support failover operations
providing a mirrored data environment. If this requires the installation of a short hop special 10Gbe switch and
cabling, the cost may or may not be worth the effort. Lastly will be the flexibility affording the switchs network
configurations. Each switch should have accommodations for redundant paths to handle additional as well as
alternative paths for data communications.
Storage AvailabilityConsidering RAID
RAID storage configurations are available for both SAN and NAS configurations. Certainly from an availability
perspective, RAID has become a requirement for most data centers. Given its implementation, reliability, and
cost, it has become a defacto standard for applications. Thus, the question becomes: what level of RAID to
use for particular applicationsor in terms of this book, what level to use for the I/O workload.
Here, RAID is used in basically two configurations: level 5 and level 1. Other RAID levels are used for specific
application support like RAID level 10, also referred to as 0 +1, where files are duplicated with higher
performance. However, this comes at the sacrifice of non-stop reliability given there is there is no parity
information calculated or stored in this configuration. Other types, such as RAID level 4, are bundled into
some NAS vendor solutions.
Considerations for RAID and its associated applications and availability include the following:
RAID Level 5 This level of RAID provides data striping on each device within the storage array. The parity
informationthe information the RAID software or firmware uses to reconstruct the data if a disk is
lostis also striped across the array. This configuration is generally used for high-transactional OLTP
applications that require a high-availability service level. The caveat is the write-intensive nature of the
application that will increase I/O latency given the increased write operations, as well as the storage array
for the multiple writes necessary with the data and the parity information.
RAID Level 1 This level of RAID provides data mirroring with parity so the application can continue to run
if one of the disk pairs is inoperative. This requires more storage capacity given its a true duplication of
the data, thus allowing a 2 to 1 ratio for data capacity (in other words, one byte of user data requires a
duplicate or mirror). Consequently, a 100GB database table that uses RAID level 1 would require 200GB
of storage on different disks. Additional data requirements for parity information are not significant and
should not be taken into consideration. RAID level 1 is used for applications that require failover
availability, or data-sensitive applications that require data to be available in case of corruption. This
comes in handy with applications such as data warehousing, and some OLTP using failover
configurations.

Finding the Holes
The most challenging aspect to managing storage networking availability is collecting, reporting, and analyzing
the information, while the most important aspect of business application performance is accessing data. The
data may be stored on disk, tape, optical, temporary cache, or other device. The location of the data becomes
irrelevant; however, if the application cannot get to it, or the paths to the data are congested or critically
impacted, the performance of the application will degrade, or worse, become unavailable.
The key to successfully monitoring and managing the performance of business applications is the consistent
and proactive management of the critical paths to application data. Although software tools are becoming
available, they provide a disparate, incompatible, and inconsistent view of storage information, not to mention
storage networking configurations. No single tool provides consistent, proactive management functions that
associate business applications with application data. IT management must choose from an assortment of
tools that provide only discrete levels of empirical information, ranging from operating system metrics and
database metrics, to I/O and disk metrics. IT users bear the burden of correlating these seemingly unrelated
sets of information in an attempt to understand the effects of workloads on storage networking resources.
Where to Look
The deficiencies within storage software management tools are compounded by the requirements, costs, and
expertise needed to support an increasing set of server platforms, operating systems, and major application
subsystems such as relational database management, messaging, and transactional systems. The following
points illustrate some of the challenges in managing storage network configurations, as well as the
inefficiencies that contribute to business application availability.
Correlation Functions among Distributed Components Storage networks are distributedmeaning
events happening on one node device or fabric switch can seriously degrade performance throughout the
entire system. The ability to correlate important aspects of performance information as it effects the
business application currently must be performed as a set of manual tasks.
Proactive Trending Todays IT managers are expected to effectively drive the bus while monitoring
performance through the rearview mirror. Literally all reporting and trending is historical. The information
which ultimately reaches the IT user is past tense and provides little value in determining real-time
solutions to poorly performing business applications. Consequently, a response to an availability problem
with a production configuration may require significantly more time to address.
Identification of the Root Cause of a Problem The effects of the conditions stated previously means
it is unlikely that the information discovered and reported to the IT user will provide any sort of root cause
analysis. This makes the problem management aspect of availability problematic. The information
provided to identify and correct the problem would likely address only the symptoms, leading to
reoccurring results.
Finding, collecting, and reporting performance information as indicated in the preceding section will be difficult.
As such, this should enter into the availability formulae when considering the appropriate configuration and
resources. There are multiple sources that IT storage administrators can access to find this information. Given
that it remains a manual effort to coalesce the data into something of value, the sources can provide a key to
building a historical database of performance and availability information. Particular to the storage networks,
these sources are the management information base or MIB (provided within the switch fabric operating
system or the NAS RTOS), the hardware activity logs within the specific devices (such as the SCSI enclosures
services (SES)), and the OS-dependent activity logs and files that are part of the operating systems attached
to the SAN or NAS configurations.
This appears as a loose confederation of information, and it is. However, if one sets up a logical organization
of sources, the ability to find, select, and utilize existing tools, and develop internal expertise in utilizing these
resources will help a great deal in monitoring availability. The following are some guidelines for both NAS and
SAN:
NAS RTOS Essentially everything that is active within the NAS device is available through the NAS
operating system. Although vendors have enhanced their offerings for manageability services for
monitoring and reporting, they remain proprietary to the vendor and closed to user customization.
NAS Configuration Looking at NAS devices as a total storage configuration requires that external
sources of information be identified and accessed. These are multiple and can be defined through
network and remote server logs. Again, these require the necessary manual activity to coalesce the data
into meaningful information for availability purposes.
Fabric Operating System Information Available within the switch MIBs. (Note that with multiple switches
there will be multiple MIBs.) In addition, many switch vendors are enhancing their in-band management
utilities and will provide more sophisticated services to access FC MIBs. The same can be said for
storage management software vendors that are increasing their efforts to provide in-band software
management services that communicate with external products running on the attached servers.
HBA Information Available within the attached node server. However, its important to note that HBA
information coming from system log files must be integrated somehow into a single view. Although this
may require additional IT activity, it can and should be accomplished with third-party storage management
tools and system collection repositories such as the Common Information Model (CIM).
Storage Information Available within the attached node device through either MIBs, SES utilities, or
activity log files generally operating within the RAID control units.
OS Information Available within the attached node server and dependent on the OS (for example, UNIX
or Windows). There are multiple sources within UNIX environments to support, find, and select storage
performance information. This also includes existing storage management products that support out-of-
band management processing. Within Windows environments is a relatively new source: the Common
Information Management database. This is an object database that provides activity information for all
processes within the server and has become compatible with a large number of storage networking and
system vendors.
Application Information Available within the attached servers are multiple sources of information
depending on the type of application and its dependent software elements (for example, databases, log,
and configuration files). Some of this information is being identified within the CIM database for Windows
environments to augment third-party management software suites. For UNIX environments, there are
third-party systems management software suites with CIM implementation just beginning. The inherent
challenge of these approaches is the correlation of the application information to storage management
activity and information.
NoteThe problem with any of these strategies, although well intentioned, is the increasing overhead
these applications have on switch performance, cost, and flexibility. As pointed out in Chapters
11 and 15, the operating capacities within the NAS RTOS and SAN FC fabric operating systems
are quite limited.
Data Recovery
An advantage of SAN architecture is the leverage of node communications within the fabric to increase
availability within the data maintenance processes. Within the data center are multiple maintenance and
support applications necessary to maintain platform environments. Of these, none are more basic than
backing up files and data for later recovery. The historical problem with these activities is the time lost to copy
the data from online media to offline media, using tape media in most cases.
This integral data center practice can be broken into two major activities, each with their own problematic
characteristics. First is the process of copying data from disk volumes and writing out the data to a tape
volume. Given the disparity of the devices (see Chapter 6 for more on disk and tape devices), a performance
problem is inevitable. However, it goes beyond device disparity and is exacerbated by the software
architecture of the copy process that has been integrated into most backup and recovery software products.
The problem is simple. The traditional, though arcane operation, requires data to be copied from the disk and
buffered in memory within the initiating server. The server then issues a write operation for the data in the
buffer and the subsequent I/O operation copies the data to the tape media mounted on the tape drive. This
double-write and staging process places a tremendous I/O load on the server executing the copy operation
while reserving both the disk device and tape drive during the operation. Figure 22-4 shows how this impacts
operations during a typical backup portion of the backup/recovery operation.
Figure 22-4: Traditional backup processes and their overhead
The second, and most important, part of this process is the recovery operation. As illustrated in Figure 22-4,
the backup is the insurance premium to cover any disruptions and corruptions to the current online data, while
the recovery operation is the claim payoff so to speak, when a problem has occurred and data needs to be
recovered to both an uncorrupt condition and any previous state.
The recovery operation is different from the copy, even though it appears to be the reverse of the operation; it
is far more selective regarding the data that needs to be written back to disks. This requires additional and
specific parameters to recover data, such as the specific data to be recovered, from a specific time period,
which should be restored to a specific state. The most complex of these operations begins when RDBMSs are
involved. This is due to the state condition that needs to be restored during the recovery operation in order to
bring the database table to a specific state through the processing of transactional log files.
Enter the Storage Area Network. Figure 22-5 demonstrates the capability of devices within the SAN
configuration to communicate with each other, thereby allowing many of the server-based, data-centric
maintenance/support applications to be optimized. The tremendous I/O load from typical backup operations
can now be offloaded from the initiating server. This requires the data copy functions to be performed from
storage network node device to storage network node device (in other words, disk-to-disk, disk-to-optical,
disk-to-tape, and so on). Given that the bulk of elapsed time during the traditional backup operation is the
double writing of data to the server and then to the backup device, such as tape, this time is optimized through
a direct copy operation under the control of the FC fabric working in conjunction with the backup/recovery
software that still controls the overall process.
Figure 22-5: Storage network backup and recovery
For example, if a typical copy operation used 100 I/O operations from the disk drive/ controller to the server
path and a subsequent 200 I/O operations to the tape unit, that requires a net 300 I/O operations that the
server must perform, not to mention the elapsed time in reserving the disk and tape units. By employing the
SAN operation of the direct node communications using the extended copy operation of FC fabric operations,
the server I/O can be reduced to a minimum of two to initiate the operation of the fabric in order to directly
copy data from the disk to the tape units. The copy operation can thus be performed one time at the speed of
FC hardware: 100MB/s with latency for tape bandwidth operations and buffers. Performance depends largely
on the specifics of the data, as well as the switch and tape configurations, but suffice to say, the savings will
be significant.
Keep in mind that the preceding example was only for the backup portion of the operations. We must also
factor in the recovery part. Nevertheless, the savings will be similar to the copy operationsthat is, copying
the information from the tape to the disk. The operation executes the same way: by the backup/recovery
software communicating with the SAN fabric, and the execution of the extended copy command through the
fabric. The extenuating circumstances will be the destination of the recovery, the extent of the post processing
of transaction logs, and the activity and location of the tape drives that must mount and process the log files.
The significant savings of copying data within a SAN should be taken into context regarding both backup and
recovery, because the value in the backup and recovery operation to external service levels is the R part, or
the recovery. In the end, the time it takes to restore data and application services is key.
Most business applications suffer during this maintenance process because data is unavailable during the
time it is being copied. Although necessary for maintaining storage backup procedures and policies, this type
of operation (that is, copying data directly from device-to-device) can greatly improve the availability of
business applications by reducing the time in which data is unavailable.
Unfortunately, NAS differs in its server-based architecture (albeit a thin server) and its attachment to an
Ethernet network. Backup and recovery operations are generally handled by a dedicated server within the
subnetwork. NAS vendors have extended their solutions to include SCSI tape attachment. This allows for a
self-contained NAS solution which includes its own backup/recovery system. Although its important for NAS
devices in remote locations to have an automated backup process, the capability of NAS to participate within
the data center tape library solutions is likely to be a preferred solution for storage administrators.
NAS integration into enterprise-level applications is driving vendors to include data mirroring, snapshot
capability, and self-contained backup/recovery operations. These functions have been extended into the NAS
hardware and software solutions as well as their integration with FC storage. As this evolves, the ability to
participate in device-to-device communication through an extended SAN fabric will make the extended copy
operation possible.

Closing the Loop
As available service levels are negotiated, identifying the appropriate and specific information to ensure their
availability commitments will become the main responsibility of performance management. Through effective
monitoring and reporting of performance factors of the storage network, effective storage network
configurations can build from the foundation of information and tools used in this process. With performance
management in place, the next phase can be begin to evolve: the effective management of performance
anomalies, or problems.
As problems start to be discovered, it will become necessary to ensure that these performance anomalies are
accurately identified, tracked, and resolved. Thus begins the problem management phase that illuminates and
justifies configuration changes, increases and decreases storage resources, and further modifies and
enhances those resources necessary to ensure availability.
Chapter 23 will build on the activities, practices, and challenges illustrated and discussed in performance
management and availability.

Overview
The failure of a switch port, the inoperative disk drive, and the new fabric software all require service to the
SAN. Upgrading the NAS storage array, the need to add Ethernet ports, and the installation of the new NAS
device all require a change to the storage network configuration. Each of us is familiar with the call at 2:00
A.M. when the production application is down and problems point to the SAN or a suspected NAS device.
These are definite problems and require an effective problem identification, tracking, and resolution system.
Servicing is defined by the activities necessary to maintain the hardware and software configurations of
computer systems. Storage networks like other infrastructures require servicing. SANs, still being a relatively
new technology, are more intensive than most for many reasons, not to mention the diversity of components
that make up the configuration. NAS configurations, on the other hand, being somewhat more mature, require
very little servicing, although this depends on their application and usage volatility. Regardless of either
application of storage networking, each storage infrastructure benefits from a servicing system that provides
significant configuration reliability and, subsequently, availability.
Servicing is often driven by problems that occur within the storage network configuration as well as by external
problems that affect the performance of the storage network. Secondly, servicing is required for any changes
to the configuration meant to fix problems and to enhance or upgrade configuration components. Considering
another oft-used work axiomif it aint broke dont fix itthe necessity of using a variety of sometimes-
compatible hardware and software components brings another dimension to this colloquialism.
From a systems management discipline, the next phase of activities are change management, problem
management, and, as its cornerstone as indicated in the following illustration, configuration management.
Each plays a pivotal role in the effective servicing of the storage network. Traditional storage management
and elements of network management make up many of the elements that are necessary to service the
storage network. This requires an integration of traditional configuration management and change
management activities. Not to mention the effective development and integration of problem management into
the data-center operations and help desk activities.
This chapter discusses the application of these management disciplines to the storage network environment.
This includes the complexities of hardware/software compatibility issues that affect SAN configurations and
the added value of keeping NAS release levels in sync with each other and the network software. Within the
data center, one of the many technical issues to deal with is the integration into other system management
disciplines that are already in place. Of these, problem management is one of the most challenging.
The activities to track changes within the storage network are beneficial in order to keep the devices
compatible, correlate problem identification, and schedule maintenance activities. However, the benefits of
effective change management within the storage network also provide value to external elements within the
data-center environment. Integrating yet another level of change into the data center is a challenge. This
chapter provides some insight into these potentially volatile situations and offers some guidelines to get things
started.

Tracking the Configurations
Computer systems require the installation of multiple components that, when working together, form some
type of system for producing effective results. Generally, those results are productive for the IT organization
responsible for the system. The ability to manage the diversity of hardware and software components is
referred to as configuration management. Configuration management establishes the inventory of a system,
naming the components, their release and maintenance levels, and connectivity. Although configuration
management is suited for both hardware and software, it is usually used in a segregated manner to track the
hardware configuration and the software release levels, respectively.
Although the premise is that by keeping track of the components within the hardware or software configuration
one can understand the economics in terms of tracking capital equipment, the alternative but oft-unused value
of this discipline is the tracking of the inventory to understand what changes can be made, under what
circumstances, in order to either correct a system deficiency or enhance the performance of the system to
meet established service levels. Given that both hardware and software are likely to change over the effective
life of a system, configuration management serves as both a driver to changes and a tool to identify problems.
Storage networking, being that its a system, and a relatively new system at that, can benefit greatly from
configuration management. Establishing an effective servicing practice for storage networking systems, and
keeping track of the diversity of components that make up SAN or NAS infrastructures, can be extremely
valuable.
Whats different about SANs in configuration management is the number of separate components that make
up the configuration. In addition to its diversity, the shared characteristics of both networking and storage
technologies gives SANs a hybrid view to many people uninitiated in storage networking. At the same time, the
SAN infrastructure is driven by the external effects of other compatible and collaborative systems, such as
servers, networks, and software.
NAS poses a more chameleon-like problem. NAS configurations can be single or multiple in number, and
while looking like storage, they are really a specialized server. This provides a configuration management
discipline with some interesting problems in determining where to categorize this component and what
elements make up the components. If its a storage device, then it has no OS. However, NAS most definitely
has an OS that needs upgrades, changes, and has the potential for problems. However, tracking as a storage
device overlooks these critical elements. In addition, NAS configurations are driven externally by the network
and remote servers or clients that contact it through remote communications.
The point is that storage networks, although computer systems in and of themselves, are different animals
when it comes to categorizing them through traditional configuration management practices. Some thought
must be given to these configurations as they are integrated into either existing configuration management
tools or are used in establishing new practices that integrate both an inventory tracking for capital expense
purposes and a foundation for tracking problems and change.
Physical Configuration Management for SANs
Configurations are likely viewed from a hardware and software perspective. Physical views of the SAN should
provide the inventory of all equipment installed within the configuration. This requires all the typical information
necessary for asset and capital inventory tracking, such as serial numbers, location, and ownership. In
addition, there should be the same current information on firmware or micro-code releases and install dates.
Many components in the SAN area come with some type of firmware, micro-code, and micro-kernel software
that is release- and hardware-specific. Its important that this information is collected and stored, at a
minimum, on a file, although a database is better for later access.
Thus, a decision will have to be made here in terms of categorization of software. Given that firmware and
micro-code are hardware-specific, the micro-kernel that functions as the fabric OS must have a place here or
in the software configuration tracking system. Consequently, the choice depends on the size of the
configuration and what type of distinction there is between hardware and software support for servicing. In a
large installation, this could be critical when it comes to avoiding confusion in tracking microkernel release
levels in two places, or dodging confusion with other firmware and driver software that is specific to HBA and
storage array controllers.
Figure 23-1 indicates the types of categories that may be considered in establishing the configuration
management repository for SANs. Of course, this information can and should drive the digital rendering of the
configuration for general reference and access.
Figure 23-1: SAN configuration management categories
Logical/Software Configuration Management for SANs
The logical configuration refers to the SAN software, but also should include the logical relationships that exist
within the SAN configuration. These are the fabric zones, the LUN assignments at the server/HBA node, and
the storage controller array nodes. In addition, there are additional elements within the fabric operating system
that relate to the port IDs, fabric logins, worldwide naming services, and the management information base
dictionary and layouts.
Configuration management for the logical and software functions is critical. Due to the diversity and complexity
of the SAN software components, much time can be saved and downtime avoided through the effective
documentation and monitoring of the fabric micro-kernel release and maintenance level, the HBA driver code
release and maintenance level, and the storage array RAID driver and firmware release level. There should be
an established configuration management for the attached servers. If not, the server OS release and
maintenance levels will be mandatory.
Essentially, the software components of the SAN form a matrix of interdependent functions. A change in any
has the potential to cascade problems throughout the SAN configuration. You can rely on interoperability labs
of vendors, industry groups, and testing labs, to gather macro level compatibility information. However, the
number of permutations far exceeds what these well-intentioned groups can provide. If that means providing a
test bed for your applications within the configurations specific to your site, then its well worth considering. In
fact, its no different than other test beds that you may require for application and system testing and should
be integrated into testing facilities.
Figure 23-2 illustrates the configuration management categories and example specifications a data center may
consider for SAN physical and logical components. The interrelationships and dependencies between
components can quickly become overwhelming, which provides additional justification for building a repository
of physical and logical components. However, one of the most difficult events to track are the dependencies
from hardware to software components. Consequently, the ability to begin an integrated repository may save
critical time as configuration changes can be tracked against the related hardware and software device
dependencies.
Figure 23-2: Integrating SAN physical and logical categories into a single repository
Physical Configuration Management for NAS
Managing the physical configurations for NAS might seem like a simpler process than SAN, but its not. The
physical connections within the NAS configurations may be simple if they are installed in close proximity and
operate within the confines of the data center. Even then, it can become confusing and a challenge because
of the diversity of networks and network components. The network becomes the critical factor in establishing
the configuration, and consequently, the NAS configurations pose a relatively difficult problem in regards to
configuration managementthat is, how to articulate and track individual components of the device and the
external characteristics of the network hardware within the context of a system (in other words, the
infrastructure).
Figure 23-3 illustrates the complexity of the external relationships between the network, the NAS device, and
the attached servers. In a sense, it can be viewed as a loosely connected storage network, with dependencies
associated with network topologies and protocols, server nodes, and file communications protocols. Unless
the hardware and firmware release levels are inventoried and tracked in conjunction with the network, the NAS
systems become unassociated storage servers unbound to the confines of the networks in which they operate.
Bottom line, if there is a problem within the NAS device thats being perpetuated through the operations of a
network device, the problem will be hard to diagnose and will likely repeat regardless of any changes within
the offending NAS device.
Figure 23-3: The complexity of NAS external relationships
Consequently, its paramount that some type of relationship be established as the configuration management
is instituted for the NAS configurations. This may require the NAS device being subsumed into the network
configuration management disciplines, if they are established; however, this may not take into account the
other relationship of the server component that is actually generating the I/O request. Given this third variable,
the NAS configuration management must rely on two external factors for effective configuration management:
the network and the server. Whether these are accounted for in the network disciplines or the system
disciplines or a new storage network discipline, the benefits to problem investigation and upgrade analysis, as
well as overall availability, are large.
Figure 23-4 indicates the types of categories that may be considered in establishing the physical configuration
management matrix for NAS. Note that due to the complexities of the external relationships of the network
devices, links into a network configuration repository enhance the value of the NAS physical configuration
information. Of course, this information can, and should, drive the digital rendering of the configuration for
general reference and access.
Figure 23-4: NAS configuration management categories
Logical/Software Configuration Management for NAS
Software for NAS becomes both simple and problematic. Simple in terms of the standardization of the NAS
micro-kernel OS used, given that the configuration is a homogeneous vendor solution. Problematic in terms of
how the NAS OS reacts to a network environment of heterogeneous components (for example, hubs, routers,
and switches). Coupled with this is the reaction to a heterogeneous environment of server OSs that may cover
both UNIX and Windows variants.
Given that the NAS micro-kernel forms a barrier to direct communications from remote clients to the storage
arrays through file access, the storage arrays associated with the NAS solution are bundled and protected
from extraneous and external access. The NAS storage upgrade path may be non-existent on entry-level
devices and limited to specific vendor upgrades with larger enterprise-level devices. The RAID controller
firmware will thus be important to problem investigation associated with vendor servicing.
Configuration management within the NAS infrastructure must encompass the NAS micro-kernel release and
maintenance level, the storage RAID controller firmware release and maintenance level, and the network
driver release and maintenance levels. These fundamental parts can be associated with network software,
network components, and server OS release and maintenance levels to form an effective NAS configuration
relationship matrix.
The logical constructs of NAS configuration can be valuable in micro views of the data storage strategies used
in the NAS devices. In other words, these are the logical views of RAID deployment within the arrays with
configurations like local mirroring and remote mirroring illustrating the synchronization methods used.
Integrated with the logical storage information, the logical views of file system extensions supported by the
NAS device should be documented.
Figure 23-5 indicates the types of categories that may be considered in establishing the software and logical
configuration management matrix for NAS. The complexities of integrating both hardware and software should
be taken in context with other configuration management files, repositories, and tracking mechanisms. Figure
23-5 suggests the linking of not only the network inventory but the systems inventory of application and
database servers and related information. Of course, this information can, and should, drive the digital
rendering of the configuration for general reference and access.
Figure 23-5: NAS software and logical views with example categories

Investigating the Changes
What is change management and integrating SAN/NAS changes into an existing data center CM system?
When was the last time someone made a change to the configuration you were supporting? Dont know? Well,
probably about five minutes ago, if youre operating in an unregulated environment. The chief creator problem
is change. Something changed, and finding out what changed constitutes the bulk of problem analysis.
Finding who changed what becomes more of a people management issue than technical. However, the
discipline of change management within computer systems came about for just these reasons. Who changed
what, when, and where is the mantra of change management.
If tracking and prioritizing changes is already an established discipline within the configurations you support,
then youll understand the value of integrating an unwieldy infrastructure like storage networking into the
change management activities. The central value of managing change is that in the case of young technology
solutions like SANs, almost 90 percent of the problems can be eliminated just by understanding the
configurations and implementing reasonable changes.
The chameleon-like nature of NAS and the ease and attraction of Plug-and-Play provide a tremendous force in
circumventing the change discipline. Well, you say, its just some network storage and I can put it on the
network in less than five minutes. Yes, you can, and you can also bring the network to its knees, lose all the
data on the NAS, and corrupt the data on the redirected server. Wow, thats a lot of potential problems for
Plug-and-Play. Regardless of whether were accustomed to this type of service, it doesnt have to happen.
Forget the formalization of change committees and change windows for the SOHO environments. All you need
is some common sense to discipline your changes. Developing a change management discipline can work just
as well for small environments as for large environments. No one appreciates the use of the defragmentation
process that runs during the peak of the servers day, especially when it crashes and all the sales transactions
have to be reentered because the mirror on the NAS device didnt work. Consequently, even if youre not
familiar with change management, starting with storage networks can be extremely beneficial, all in the name
of availability.
Driving Change Through Configuration Management
An ongoing program of regular monitoring and analysis of the storage network configuration ultimately reveals
changes that must be made to enhance performance or provide sufficient reliability to meet service levels.
Whether this comes through changes in software or hardware activities, the ability to schedule, track, and
implement these changes effectively, and in a non-disruptive manner, requires the disciplines of change
management.
In many cases, however, there will be changes to the configuration that require a timely response.
Consequently, there must be a prioritization within the change management system to handle these issues. In
terms of SAN serviceability, the inoperative switch that impacts the service levels of a production application
system must be analyzed given its relationships to the entire configuration and its effects on other components
and systems being supported through the SAN configuration. This may require a quick response, but without
sufficient and accurate knowledge of the existing configuration, the ability to identify and resolve the problem
becomes difficult. Either the change is made without knowledge of its impact, or a lengthy analysis to
document the configuration may elongate the outage or degrade the service.
Consequently, configuration management drives changes by documenting storage network hot-spots and
quickly referencing the relationships within the configuration. In addition, an effective change management
program can leverage the information within configuration management system diagrams to facilitate the
problem resolution process in conjunction with change prioritization. Helping to facilitate the establishment of
change prioritization is an added benefit of configuration management that drives change management in a
positive direction.
How Problem Management/Resolution Becomes the Ying and Yang of
Storage Networks
You cant fix a problem unless you change the configuration. You dont have problems unless there is a
change to the system. This is the yin and yang of change management. Although we have yet to discuss
problem management, we have described the effect of changes within the configuration being driven by
problems that occur. Because storage networks are implemented using diverse components, they are
exposed to a disparate set of external stimulus. They become the critical elements to production systems,
however, and as such, they are the prime focus when diagnosing severity level one problems. In other words,
storage is a key element to evaluate during system outages.
Consequently, managing change can drive the balance of change within the overall data center activities. After
all, its easy to blame everything on the new kid (say, storage network configuration), if they cant defend
themselves. The need to enroll storage network configurations into change management activities allows
many things, including the visibility of changes within the configuration, visibility to external changes that affect
the configuration, and the ability to maintain a credible defense in problem identification activities.
The Effects of Capacity Upgrades on Change Management
One of the major factors that disrupt storage configurations are upgrades. Whether therere extending
capacities or enhancing performance, these changes should affect existing service levels in a positive way;
however, in many cases, capacity upgrades can provide unforeseen results and performance, generally in the
wrong direction. These situations evolve from unscheduled or unmanaged change plans that do not account
for adequate time to complete or consider the effects to related systems.
This is evident in the installation of additional storage capacity using NAS devices. Without sufficient
consideration for the related network capacity and existing traffic, NAS devices can solve the storage capacity
problem but create a network problem as a result. The other issue with NAS upgrades is the related file
protocol services that need to be connected or reconnected after an installation. Installing or upgrading NAS
device capacity may be quick in terms of configuring and identifying the network; reestablishing the drive
mappings and device mounts for the new NAS devices are sometimes an overlooked detail.
SAN configurations provide a more complex activity for upgrades and capacity upgrades. Due largely to the
diversity of components and their interrelationships, SAN upgrades and installations will require a significantly
longer planning and analysis cycle than other storage configurations. For example, upgrading storage capacity
has a ricochet effect within the configuration. Adding a significant amount of capacity is driven by increased
access. The increased access may require additional node servers to be attached, and ultimately will place
more traffic within the switch configurations.
If the extenuating situations as mentioned in the preceding example are not addressed, the SAN configuration
can become compromised in the name of increasing service. Addressing these situations requires a discipline
that is supported through a comprehensive change management system. The articulation of changes within
the storage infrastructure drives the discipline to analyze the following: whats being changed, why, the desired
result, and the plan for reestablishing service. Managing changes highlights the division of labor between
software and hardware changes. It also highlights the division of responsibility in large configurations where
the network part and the storage part are separated and supported by different groups, each having their own
hardware and software expertise. Integrating change management into the storage configuration management
process provides a forum to highlight all the required activities for servicing the SAN or NAS configurations.
Tracking Changes to Physical and Logical Configurations
Tracking of changes within a storage network is similar to the tracking of changes within the data center: a
separation of hardware and software changes. Another view is to divide them into physical and logical
categories. This mediates the traditional separation of hardware versus software, thereby integrating the
component responsibility into combined or shared responsibilities. For example, the typical enhancement of
storage capacity within a storage network is the installation of a new and larger storage array. This could be
performed in an off-shift time period, or on a weekend, and no one would be the wiser.
However, it could also be scheduled within a change management system where the information regarding the
upgrade, proposed activity, and scheduling are communicated to other areas that may be effected. This allows
others to either consider the effects of the change and plan accordingly, postpone the change until issues are
addressed, or, at a minimum, be aware that the possibility of problems could exist after the change date.
Because both NAS and SAN configurations are closely associated with hardware and software functions (see
Parts III and IV for more details on the major components of these storage systems), the tracking of physical
changes provides a method that encompasses both the physical aspect of the component as well as the
bundled software configuration parameters that need to be addressed. For example, the recabling of a SAN
switch to enable the installation of a new switch requires both physical changes and subsequent fabric
changes. Viewing and administrating changes in this way forces the change process to encompass both
hardware and software skills while ensuring the changes are analyzed in an integrated fashion.
The tracking of logical changes performs the same function within the change process, except that logical
configurations are generally driven by changes to software components within the storage network. These
may be enhancements to various features and functions, or maintenance of existing software elements such
as the fabric OS, NAS micro-kernel, and storage and adapter drivers and firmware.
An example is the upgrade to the NAS micro-kernel software. This change, viewed from a logical perspective,
should require an analysis of configuration management information regarding which relevant NAS devices
are affected by an upgrade. The NAS target configurations would then be evaluated for subsequent firmware
changes within the storage array controllers that may be required to support the new NAS kernel software. In
addition, the micro-kernel change may have different levels of IP network support and will require the
evaluation of the potential network components. Consequently, the change to the logical configuration brings
in all of the NAS components, as well as the external elements that may be affected. This also forces the
necessary storage and networking hardware skills to participate in the scheduling and implementation of the
change.
Guidelines for Establishing Change Management in Storage
Networking
The following guidelines can be used to provide the beginning of a change management program for storage
networking. Even though a formal change management function may exist in your data center, the
establishment and integration of storage networking as a valid change management entity is key to
participation at a data-center level.
Establish a link between storage configuration and change management. This forces the evaluation of
each change to be analyzed for the entire configuration, as well as other systems in the data center.
Develop a relationship matrix between the key components among the storage network configurations.
For example, the relationship between the FC switch and the storage nodes is a fundamental and
interdependent one. If the switch becomes inoperative, the storage is inaccessible and potentially
magnified by several storage arrays supported by the switch. Using this example, it also provides a
roadmap of Inter Switch Links (ISL) dependencies and other fabric nodes that are participating in the
fabric.
Establish an internal (within the storage network) change prioritization scheme and an external (outside
the storage network) change effects scheme based on a storage network availability matrix. The storage
change prioritization scheme will define the levels of change based upon their potential effects on the
storage configuration. The external change effects scheme will define the levels of how the storage
configuration is affected by outside configurations of servers, networks, and other storage configurations.
Establish a link between storage change management and problem management. Additional discussion
on problem management follows; however, its important to mention the critical link between problems and
change given their synergistic relationship. For example, if a storage configuration denotes a FC switch
that is attached to five storage arrays and another two switches where information drives the relationship
matrix that depicts its interdependencies. These component interdependencies drive the change
prioritization scheme to reflect this component as a critical availability element within the configuration.
The external changes affect change prioritization schemes that, in turn, reflect the risk to external systems
supported by these components. Consequently, this will drive a level of change prioritization if and when a
problem arises within this component.
Whose Problem Is It Anyway?
Storage problems within the data center can ricochet through production applications faster than any other
component. Obviously, this is a reflection of their critical position within the diversity of components that
support production processing. When problems occur, there generally is some type of trouble ticket system
that logs problems for assignment and potential resolution. These systems provide an invaluable service when
it comes to documenting and tracking problems, but does little to facilitate the entire aspect of problem
management. Consequently, applying a trouble ticket system or help desk software to storage problems only
addresses a small portion of problem management.
Problems, especially within the storage infrastructure, are resolved in a number of ways, depending on the
data-center facility, administrator preferences, and management, or even the establishment of service levels.
Regardless of data centerspecific attributes and management styles, there are fundamental activities that
reflect problem management disciplines as well as critical activities that must be performed to respond to
storage problems effectively. The problem management disciplines are: problem prioritization, problem
identification, and problem resolution. Overlaid with problem documentation and tracking, they form the
fundamental parts of problem management. How these are implemented and integrated with storage networks
is critical to the success and reliability of the storage infrastructure.
The Problem Prioritization Debate
There is no debate more dynamic within the data center than that surrounding problem prioritization. Someone
must decide what constitutes a severity level one problem, as opposed to severity level two, three, and so on.
This discussion becomes even more vigorous when dealing with storage, since storage contains the data, and
ownership is an attribute of that data. If the data becomes unavailable, then the application that uses the data
is unavailable and the level of impact is driven by the effect this situation has on the activities of the company.
Consequently, end users drive the establishment of severity definitions with the data availability, vis--vis,
storage availability. However, most are ill-equipped to handle or understand the level of complexity within the
storage infrastructure to effectively provide this.
These discussions will become either a very complex issue or a very simple equation given the level of data
ownership taken by the end user. If end users take data ownership seriously, then discussions regarding the
definitions of data, usage, and management will become involved. However, if end users simply see the data
center as the service supplying the data, the discussion can be reduced to a simple level of on or off. In
other words, either the data is available, or it isnt.
Regardless of these two perspectives, and their highly charged political engagement with end users, the data
center must view problem prioritization in a more utilitarian fashion. Simply put, the complexities of providing
data delivery and availability infrastructures must be managed within a matrix of interconnecting systems that
are beyond the level of knowledge or interest of end users. Moving end users toward a utilitarian approach
allows the data center to deal effectively with storage ownership and management of the infrastructure.
Defining problems from an end-user perspective within this increasingly complex set of infrastructures allows
data centers and underlying areas to define the prioritization required for the data utility. Key to establishing
prioritization mechanics is storage networking.
The relationship between data availability and storage networking problem prioritization increases the
complexity of this activity. Given the characteristics of data delivery and the evolution of decoupled storage
infrastructures such as storage networks, the chances that problems will arise, given the level of
interrelationships within the storage area, increases dramatically. This results in environments with
increasingly complex storage problems while trying to maintain a simple utility data delivery system.
Problem Identification
Investigating and identifying problems within storage networking configurations is an ongoing activityfor
three reasons. One, these technologies are still new and continue to be temperamental. Two, storage
networks consist of multiple components that must interact both internally within themselves and externally
with other components and systems within the data center. And three, the industry has not yet produced the
level of tools necessary for effective operational problem evaluation and identification.
However, there are some key items to consider when looking at problem identification in these areas. Some
activities dont require tools so much as basic problem-solving practices. Although this is beyond the scope of
this book, the necessary activities should be mentioned in context with storage networks.
Accurate and Current Information Configuration management for installed storage networks should be
accurate and readily available.
Diagnostic Services A minimum level of standard operational procedures should document the entry-
level activities for problem investigation and resolution. This should include the dispatching of vendor
service personnel in order to facilitate problem resolution.
Service Information Accurate and current information should be available to contact vendors and other
external servicing personnel.
Tool Access Access to consoles and other logical tools should be made available to support and
operations personnel. This includes current physical location, effective tool usage, diagnostic processes,
and authorized access through hardware consoles and network-accessible tools.
In SAN environments, an effective guideline is to investigate from the outside in. Essentially, this requires the
evaluation of problems that affect all devices versus problems that affect only specific devices within the
storage network.
For problems that affect all devices, its necessary to view components that are central to the network. Specific
to SANs is access to the availability of the fabric operating services. The directly connected console is the
most effective tool if network access is not available, while fabric configuration routines provide status
information regarding port operations and assignments. There may be additional in-band management
routines as well, which can be used for extra information.
For problems that affect only some of the devices, it becomes necessary to first view components that are
common to the failing devices. If the total SAN configuration is not completely down, then out-of-band
management routines can be accessed to provide a port analysis, port connections, and performance checks.
In addition, the ability to telnet or peek into the attached node devices and node servers verifies the basic
operations of the switch fabric.
Specific to the node devices are the mostly out-of-band tools which evaluate and verify the operation of
storage devices, server HBA ports, and server operating systems. However, access through the fabric can
also verify connectivity, logins, and assignments. Key to this area are the continued network drops from the
fabric by node devices, a practice which can be traced more effectively through fabric access using
configuration and diagnostic tools.
In NAS environments, the same approach can be used to identify failing components; however, NAS
configurations are more accessible to typical performance tools. Therefore, the most effective approach is to
view problems from the outside in by focusing on access points common to all NAS devicesfor example, the
network. This is followed by the investigation of individual NAS devices and related components that are
failing.
The centralized approach can be used with existing network tools to monitor and respond to NAS network
drops where access within a specific area may be the key to the problem (for instance, a network problem
versus a NAS problem). This demonstrates how NAS devices require less overhead to identify problems
because of their bundled self-contained status and their integration into existing networks as IP-addressable
devices.
If problem identification focuses on an individual device, then the failing components may be simple to find
given the bundled nature of the NAS solution and the available tools within the NAS microkernel operating
system. All NAS vendors provide a layer of diagnostic tools that allow investigation of internal operations of the
device. This becomes a matter of a failing server or storage components given the simple nature of the NAS
device.
NAS devices are less expensive and easier to integrate into existing problem identification activities. Given
their simple configuration and mode of operation as an IP addressable network component, they can be
identified quickly through network tools. In addition, internal diagnostic vendor tools provide performance
statistics and notification of failures.
Problem Tracking
Once a problem has been reported, the ability to document its particulars is critical to the identification
process. Most data centers will have an existing problem documentation system associated with their help
desks. Integration of these systems is important for setting the foundation for problem tracking within the
storage network. This doesnt automatically mean that all problems will be documented through the help desk,
however. Many storage network problems will come from internal sources where theyre tracked and reported
by data-center administrators, data base administrators, and network administrators, to name a few.
Even though these positions may report the bulk of problems, it does not guarantee those problems will be
reported through the same trouble ticket system used by help desk and operations personnel. Thus, the
foundation is laid for problem duplication and multiple reporting. This demonstrates just one of the challenges
regarding implementation of the storage network into the problem tracking system. The same can be said for
other key areas of the problem reporting process when it comes to critical and accurate reporting of problems.
Storage network problem documentation is made more effective by following a few general guidelines.
Consistent Reporting Process Ensure that problem reporting uses the same system for all external and
internal users. This requires a certain level of internal discipline to get all administratorsboth in systems
and applications support to utilize the same system. This is important because a problem reported
externally may be linked to a problem reported internally, which will have more technical detail. An
example is a network-shared drive that continues to be inaccessible, as reported by external users. An
internal problem report indicates that the network HUB associated with that device has experienced
intermittent failures and is schedule to be replaced. These two sets of information provide a quick level of
diagnostic information that should lead the storage administrator to verify the NAS device internally and
take steps toward a resolution that may require a different path to the storage.
Historical Reporting and Trending One of the most valuable assets problem tracking systems provide is
a database to view the history of a component or device, its past problems, and resolution. This can be
extremely important as both SAN and NAS components are tracked and analyzed for consistent failures
or weaknesses. For example, historical trends regarding complex switch configurations can be very
helpful in identifying weaker components or trends toward failure as port counts become higher and higher
and interswitch linking becomes more sophisticated. This proves a valuable tool in anticipating potential
problems and thereby scheduling changes prior to a failure. This approach offers a proactive answer to
problem management and subsequent downtime statistics.
Experience and Expertise Tackling problems is one of the quickest ways to learn a system or
technology. Using problem tracking information and configuration management information provides
effective training for IT personnel new to storage networks.

Closing the Loop on Serviceability
Effective serviceability requires the establishment of both configuration management and problem
management. Starting with a set of user requirements that result in the capacity plan, followed by the
installation, the critical information contained with the physical and logical configuration documentation drives
the effective response to problems.
Problem management establishes a method to effectively deal with problem servicing and the proactive
servicing of the configuration. Key among problem management activities is the definition of prioritization.
Although this can be a volatile activity, its best kept as an internal data center activity where end-user data
utility is a key goal. Problem identification is central to establishing credibility within the data center. Given
SAN and NAS configurations, both can be analyzed from an outside-in problem identification perspective.
Although SANs demonstrate a much higher level of complexity, they nevertheless lack the necessary tools to
enhance the problem identification process. On the other hand, NAS configurations, given their operation as
an existing network component, can take advantage of current network tools. NAS also has a level of maturity
that offers internal diagnostic information, thus providing efficiencies in problem identification and resolution.
Without an effective practice for servicing the storage network, the complexities and potential confusion within
configuration analysis and problem diagnostics will exacerbate system outages. Both configuration
management and problem management are critical elements to the cycle of storage network management.

Overview
Planning for storage capacity requires the analysis of future end-user demands and the effective management
of existing storage resources, both of which support the ongoing processing demands of the business.
Provisioning storage can be as simple as adding new storage devices to existing application servers, or as
complex as the calculation of new storage demands for supporting a global Internet services application.
Regardless of complexity or application, storage demands are end-user driven and reflective of the dynamics
within the economics of application usage. For example, an upward trend in sales orders may generate an
increase of 100GB of storage to the online application to handle the new business. New customers may
almost always spark an increase in storage capacities as the systems expand to handle the customer and
related files.
This expansion, and sometimes contraction, of user demands creates the necessity for planning the
processing configurations necessary to meet these demands. This chapter brings our cycle of systems
management disciplines to the starting point (see the following illustration): the beginnings, development,
and/or evaluation of a capacity plan. Capacity planning for storage networks is a subset of overall storage
planning, which supports a macro capacity plan for the enterprise, or at least the major components of the
data center computer configurations. Consequently, our discussions within this chapter will reflect the support
of the entire data-center infrastructure, while articulating the value of developing a capacity plan for storage
network infrastructures.
Over time, data centers develop activities and practices necessary to keep up with computing capacities. This
is generally referred to as the capacity plan. These practices and activities have matured and congealed into
areas of specialization driven by the diversity that exists even in the most conservative IT organizations.
These activities are generally divided along physical areas of responsibilities, as well as technological spheres
of influence. Figure 24-1 indicates both the physical division of responsibilities within most data centers, as
well as the expected technology segregation. This shows applications, systems, and networks as the
dominant areas of responsibility. However, supporting the end-user community is generally an operational
area that supports end-user computing and help desk functions, and which acts as a conduit of information
driven by day-to-day operations of the major areas of specialization (for example, applications, networks, and
systems).
Figure 24-1: The traditional division of responsibilities within the data center
Hidden within these organizational divisions are the sometimes invisible spheres of technology influence and
expertise. These are characterized by the segregation of the application elite and the back room systems
mechanics. This forms the major factions that deal with capacity planning activities. Given that capacity is
driven by end-user demands, the application elite is called upon to use their systems analyst skills to
communicate end-user future plans in terms that can be translated by the systems mechanics. Within this
milieu of communications, the operational front liners form the ad hoc reality-based view of capacity
requirements.
Consequently, one of the major segments of this chapter is the discussion of how user requirements can be
interpreted properly and used as valid input into the storage planning process. As with any human
communications endeavor, this requires a bit of diplomacy and tact to accomplish. Given that storage, and in
particular storage networking, is often viewed as a third-level component within the enterprise capacity plan,
care and understanding of the communications process provides a certain amount of value in accomplishing a
successful storage networking capacity plan.
The importance of integrating storage into a larger plan cannot be overlooked. The need to articulate and
justify complete resources (this includes hardware/software, facilities, and the human element) can make the
initial enterprise capacity plan (that includes storage) a success. The reason is that storage has often been
viewed as a commodity and is purchased in a manner similar to all processing commodities like memory, end-
user software, and printers, by cost. However, the storage network changes that significantly, by offering a
complete infrastructure that includes the support of multiple processing server configurations. Storage
networks can hardly be seen as a commodity, regardless of what vendors want you to think. On the other
hand, NAS, although easy to slip through the official capacity planning exercises, often has extenuating effects
on other supporting infrastructures within the data center, such as the network.
Finally, the justification of a SAN or NAS configuration based solely on consolidation of storage, servers, or
recovery purposes does not make a capacity plan. These can be characterized as internal user requirements
that may or may not justify the use of either storage network solution. This is not to say that these are
superfluous or self-indulgent requirements; however, they must relate to the business supported by the data
center. Justification on a cost basis, which is usually the case for consolidation and recovery strategies, needs
to be balanced by a long-term plan to allow the storage infrastructure to grow along with the storage demands
of the business.
Also, the establishment of a plan does not dictate its success. The data center must manage the plan. This is
especially important with technologies that are volatile, support dynamic growth, and which are critical to an
applications success. This is what storage networks are, and thats why they need a storage networking
capacity plan of their own. Once integrated into the enterprise plan, and having the confidence of data center
management, the storage plan can play a vital and productive role in the long term.

Developing a Plan for Storage
Storage provisioning begins with a sound capacity plan to establish a reasonable configuration that supports
the business applications and constituency of usersboth end users and data-center users. Thus begins a
chicken-and-egg scenario. It also provides the justification for things like beta installations prior to committing
new technology platforms like SAN and NAS configurations to production environments.
To calculate and commit to availability metrics, the following guidelines should be considered in establishing a
reasonable plan to support SAN or NAS configurations.
Analyzing End-User, Application, and Internal Systems Requirements
Develop a method of communications with end users, applications designers and developers, operations,
and systems infrastructure colleagues.
Define a capacity storage metric and level of precision. This could be capacities in number of bytes with a
precision to the nearest megabyte, or it could be a more esoteric work unit such as an application work
unit calculated as 1 gigabyte per unit, or a relational database unit such as 1 megabyte per unit. Whatever
you use, keep consistent with the metric and level of precision. Keep in mind that you should allow for
incongruities in media, which may be reflected in disk and tape densities, sector sizes, and blocking
architectures.
When collecting storage capacity requirements, remember to calculate both end-user aggregate data and
the necessary overhead to support the data. For example, if end-user data for a database is 100GB, then
the total data requirement will be 100GB plus the necessary overhead of database system, index, and
temporary tables. Depending on the type of application, this could go as high as 50 percent of the
aggregate user data storage requirement, rendering our example to 150GB required for current database
physical storage. Other application systems such as e-mail, ERP, and CRM all have additional storage
overhead requirements that are important to consider along with the aggregate user data.
Planning and Establishing Adequate Capacity Based Upon I/O
Workload Estimates
An important consideration to storage capacity is the processing requirements that together with capacity
requirements drive the configuration elements necessary to meet end-user demand. In order to accurately
estimate these requirements, its necessary to translate user requirements into I/O workload estimates
(see Chapter 17).
Using I/O workload identification recommendations (again, see Chapter 17) the application workload types
begin to distinguish themselves. Consider size versus access as the applications demonstrate their
workload characteristics such as OLTP, batch, messaging, and combinations of these. The size factor, as
depicted in our discussion about capacity, must be considered along with the calculation of the I/O
workload factors. This allows for the initial elements of the configuration to start to take shape.
Recovery factors also influence the configuration elements. My I/O workload guidelines provide a recovery
factor for calculating the configuration elements (see Chapter 17). However, this calculation should reflect
the necessary insurance required for particular data-center business continuity requirements (see Chapter
21).
Although a subset of recovery, the redundancy factor must not only be taken into account for recovery
factors, but also for security and individual failover. Largely driven by system failover requirements, the
redundancy in both NAS and SAN can be implemented within themselves, and thus can be a substantial
factor in configuring redundant systems. Therefore, they should be considered as separate items even
though theyre related to recovery and business continuity at a macro level.
Establishing an External Capacity Driver Matrix
Application requirements drive most of the capacity requirements; however, they provide little if any insight
into the storage configuration to support the business application. By developing an external capacity
driver, you have identified the major influences that drive both capacity and performance.
System requirements drive the rest of the capacity requirements with overhead to OS, subsystems, and
network requirements. However, these spheres of technology also exert influence on the configuration
they feel are most appropriate to the processing of the business applications. By developing the external
capacity driver, you have also identified the influences on the storage infrastructure.
Establishing a Reporting Mechanism
Once the end user, application, and systems requirements are translated and calculated into
requirements for a storage configuration, develop a reporting mechanism on the activity of their storage
usage. The system must be able to be as proactive as possible in order to show trends that are outside
the requirements, and to compare them against the storage capacity plan.
Managing to Storage Capacity Plans and Established Service Levels
The establishment of the reporting mechanism begins the management of the plan. Given the translation
of requirements to real storage resources, the users must be held accountable for their estimates.
However, given the volatility of the technologies and the changes in business climates, things will change.
Therefore, its important to also establish an effective communication system track requirements to actual
usage without retribution (see establishing a non-aggression pact).
The User Non-Aggression Pact: It is always helpful to reach an agreement with a user community that
accounts for the unexpected. This allows both parties to agree that the unexpected does happen, and
should it occur, it should be resolved with both mutual understanding and changes to the existing plan.
This can be critical given the volatility of SAN installations and NAS data volume requirements.
Unexpected Requirements The most common set of unexpected circumstances is the unforeseen
application that requires support but which has not been planned for. In terms of affecting the SAN
installation, this can be costly as well as disruptive, given the scope of the new application storage
requirements. Within the NAS environment, this is one of the strengths NAS brings to the data
center. If the requirements can be handled through file access, NAS performance, and capacities,
then the NAS solution can be used effectively during these circumstances.
Unforeseen Technology Enhancements This is the most common circumstance when initial SAN
design proves insufficient to handle the workload. The outcome of retrofitting the SAN configuration
with enhanced components is additional cost and disruption.
Mid-term Corrections Its likely that any enterprise storage installation will experience either one or
both of the preceding conditions. Consequently, its extremely important to build into the user
agreements the ability to provide mid-term corrections that are an evaluation of the current services
and corrections to requirements in order to continue to meet the committed services.

The Storage Analysis
Analysis of storage is divided into two distinct parts: new storage demands and existing storage allocations.
Although these two different activities culminate in the same place, which are reasonable configurations that
support end-user demands, the reason they are distinct is the following. The demands for new storage provide
an opportunity to consider alternative methods to meeting the demands, rather than extending the current
configuration. This is most appropriate in storage networking given the sometimes overlapping solutions that
exist between NAS and SAN solutions. It is also appropriate when considering moving from direct-attached
configurations where the storage demands are not closely tied to existing configurations.
Leveraging a Storage Upgrade
When considering a storage upgrade for capacity and performance, the analysis of a storage network should
be part of the storage capacity planning activities. However, this does provide a challenge to the existing
server infrastructure and installed storage components. Moving away from any of these requires additional
hardware, software, and training. It also requires concurrence from the external drivers, such as applications
and systems. Likely, each of these may be resistant to change and the ability to rely on information from the
external driver matrix will assist in the justification of the whats in it for me scenario.
Analysis of existing storage can provide greater justification into storage networking solutions given the
scalability limitations within the client/server direct-attached model (see Chapters 1 and 2). More often than
not, the move to storage networking provides a longer term solution in supporting increased user demands.
Using the previously described user requirements translations and I/O workload calculations, the justification
can prepare the way for the necessary increase in expenditures as well as showing the short-term fix that
adding storage to existing server configurations will have.
Establishing a New Storage Network
Analyzing new storage demands provides an opportunity to leverage a storage network driven by user
requirements and I/O workload analysis, the justification can be compelling. New storage capacity can be
depicted in terms of scalability of capacity and performance, but also in its ability to consolidate some of the
legacy storage into the storage network. This provides the first articulation and integration of internal
consolidation factors that are so popular in justifying SANs. However, the same can be said for NAS devices if
the storage and application characteristics are justified in this solution.
Working in conjunction with your systems colleagues, there can be real synergy in establishing a storage
network strategy. First, is the consolidation of servers. That, in itself, is a large cost factor reduction in overall
systems responsibility and administration. This will be augmented by added savings of OS and application
license fees associated with multiple servers with direct-attached strategies. Finally, there is the added benefit
of managing fewer server entities and the processing consolidation that occurs with the collapse of application
processes into a single, albeit larger, server. These cost savings start to mediate the increased costs
associated with a storage network solution.
A hidden benefit to systems administrators is the performance and problem management factors that come
with storage networking. The consolidation of servers and the collapsing of the storage arrays translates to
less network connectivity, minimum servers to manage, and consequently fewer things to go wrong.
Establishing an agreed upon and reasonable metric for this allows a quantifiable benefit to be monitored when
adopted. In other words, if putting in four NAS devices can collapse file servers on a 20:1 basis, then the
quantifiable benefit will be losing 20 general-purpose servers for every single NAS device. If that were the
case, then once the plan is implemented, the redeployment or retirement of the 80 servers, to use our
example, would create excellent credibility for the plan and storage network.
Tools for Storage Analysis
Storage networking can be analyzed in two ways: the physical capacity and the performance of the
configuration. Just as direct-attached storage configurations are monitored for storage allocation, usage, and
access, storage networks need to provide the same information. The problem comes with both the lack of
tools that account for multiple usages of networked storage arrays and the immaturity of system tracking
databases that provide historical data. Performance monitoring in storage networks provides the same
challenge regarding tool deficiencies and lack of historical data collectors.
There are several storage software choices when considering capacity and access monitoring tools. It is
beyond the scope of this book to analyze or recommend any of these tools and should be a data-center
specific choice dependent on the specific needs of the entire storage infrastructure. These tools fall into two
categories related to a wide variety of tools known as storage resource management tools. Subcategories
include the quota management and volume management tools.
Storage Resource Management
Quota management tools provide a mechanism to assign storage capacity quotas to end users or specific
applications, or a combination of both. These provide a safety net in terms of storage usage and the potential
of errant users or applications to utilize the majority of storage capacity in an unregulated environment. The
quotas are generally set by administrators, either from a storage or systems perspective, and managed from a
central location.
The majority of tools in these categories provide some level of reporting on storage utilization. The more
sophisticated tools provide views from several perspectives: by end user, by application, by logical volume, or
by logical device. These tools work well with file-oriented storage implementations; however, they become
problematic when attempting to understand the storage usage of relational databases or applications
employing an embedded database model.
Its important to note that in these cases, and given the installation of databases within the data centers (which
is likely to be large), the monitoring of storage utilization and access needs to rely on the database monitoring
and management tools. This provides another important element within the external capacity driver matrix
(see establishing an external capacity matrix), which is the influence of database administrators, designers,
and programmers. Given the transparent nature of the relational and embedded databases to the physical
storage, the usage of these application subsystems needs to be managed in conjunction with the database
expertise. However, its also important to become familiar with these tools to understand the underlying
physical activity within the storage infrastructure.
Volume Managers
Another category of tools that can provide storage analysis information are volume managers. These tools
provide the ability to further manage storage by allocating the physical storage in virtual pools of capacity. This
allows file systems and applications to access particular volumes that are predefined with specific storage
capacity. This can be extremely valuable when allocating storage for particular applications that have specific
requirements for storage needs and may become volatile if storage becomes constrained.
Like quota management tools, most volume managers have reporting mechanisms to track both usage and
access. They also provide an important function and level of detail that enhances both performance analysis
and problem determination. Volume managers work in conjunction with storage controller and adapter
hardware to separate the logical unit (LUN) numbering schemes used by the hardware functions. As such,
they provide a physical-to-logical translation that becomes critical in understanding the actual operation of a
storage configuration. Given the complexities of SANs, these functions can be extremely important in problem
determination and monitoring performance.
We have used volume manager examples in many of the figures within this book. As Figure 24-2 shows, they
add relevance to naming storage devices within a configuration. As seen in this figure, a typical storage area
network, prod01, prod02, and prod03 are disk volumes that contain the production databases for the
configuration. The storage administrator through the services of the volume manager software assigns the
specific volume a name. The volume manager, meanwhile, manages the storage pools, prod01, prod02, and
prod03, transparently. Working in conjunction with storage controllers and adapters, the volume manager
works to translate the LUN assignments within each of their storage arrays. In viewing the actual physical
operation of the storage configurations, one must understand and look at the LUN assignments and activity
within each storage configuration.
Figure 24-2: Volume management working with storage arrays
The quota and volume management tools provide a snapshot of storage usage and access. However, looking
for a longer term historical usage becomes problematic. One of the continuing difficulties within the open
systems operating environments is the lack of historical collection capabilities, although this situation
continues for a number of reasons, such as disparate processing architectures, box mentality, and distributed
computing challenges. Each of these provides a roadblock when systems personnel attempt to gather
historical processing information.
Collecting Storage Information
Disparate operating systems provide the first problem, as the major contributor to the differences between
vendors, but especially between UNIX and Windows operating environments. Historically, however, the
centralized mainframe computing systems of IBM and others provided a unique architecture to gather
processing data. Most examples in this area point to the IBM System Management Facility (SMF) and
Resource Management Facility (RMF) functions as models for what should be expected within the open
systems area. Overall, this is not the case, and taken out of its centralized and proprietary context, the SMF
and RMF models for historical information gathering do not fit well into the area of client/server and distributed
computing.
The first area not covered by the IBM model is the collection and support of distributed computing
configurations that have become commonplace within client/ server installations. This is exacerbated by the
open architectures of UNIX systems and lack of any standards in processing nomenclature. Consequently,
there are multiple ways of viewing a process within the UNIX variants of the UNIX open systems model. This
permeates into the storage space, as the nomenclature and operation of UNIX storage models are, by design,
very flexible and open to vendor interpretation. So gathering data about a SUN configuration is a different
exercise than gathering processing information about an AIX environment. If we throw in the increased open
area of storage, vendor support for each of these environments makes the common historical database such
as IBMs SMF and RMF impossible.
The second area that makes the centralized proprietary environment different is the distributed nature of
processing. As servers became specialized, the ability to provide a historical record in context with the
application became problematic at best. Consider a database transaction that executes partly on application
servers, where application logic creates and submits a database query, which is executed on another server.
The historical log of the application processing becomes difficult to track as it migrates from one server to
another. With the distributed processing activity, the combined activity information transverses from one
operating system to another with the related resource utilization. This example points out the problematic
recording of database transactions within the context of the application.
We will integrate the complexities of a SAN into the collection discussion and provide yet another source of
processing and resource information into the capacity planning equation. In terms of a SAN, this becomes the
fabric and related micro-kernel operations in moving the data within the storage network. If we integrate the
NAS environment, we integrate yet two more sources of processing: the NAS micro-kernel processing and the
operating systems that run the network fabric. The collection of data for both SAN and NAS adds additional
data collection points that are again distributed in nature.
The point is that the collection of data within the distributed environment continues to be problematic. Although
there are some methods that attempt to deal with these issues, the need to develop an alternative to the
historical collection of data becomes evident. Currently, only a limited number of software tools are effectively
addressing this area using emerging standards that not all vendors have accepted. This exacerbates the
situation for storage analysis within storage networks, given the difficulty in collecting valid storage activity
information.
However, there are two basic initiatives to consider when collecting historical storage activity information. The
first is the industry initiative of the common information model (CIM) that is an object-oriented database for
collecting system processing details and device status. CIM is a Microsoft-initiated standard that has achieved
some level of acceptance although its complexity continues to make it impractical for storage network usage.
The other item is the usage of network-oriented tools which many storage- networking vendors include with
their products: the management information base (MIB). These provide the quickest way of collecting
information in-band for the SAN and within the NAS micro-kernel (although this isnt used as often in the NAS
micro- kernel levels).
The Common Information Model and Storage Networks
The common information model (CIM) is an object specification that provides a uniform way of describing a
computer and its components. The specification is used as a standard to write applications that access the
objects as described within the model. The CIM initiative first started as a Microsoft object-oriented structure
used mainly by the operating system to build a repository of general and specific information about the
computer it was running on.
The objective of this functionality was to give hardware companies the ability to access their components
within a network. Used for support and problem management, vendors quickly contributed to the Microsoft-led
initiative with specifications for general computer components, concentrating chiefly on internal components
for processors, adapter cards, and memory structures. However, it was also Microsofts objective to evaluate a
configuration in order to determine the requirements needed for particular operating system functions,
applications packages, and licensing verification. Oddly enough, the initial releases of the CIM specification
did not reflect any of the major peripheral and processing software components necessary to complete a
computer configuration.
The Microsoft CIM initiative was soon passed to the Distributed Desktop Management Task Force, DMTF, a
standards-based body that manages similar initiatives for distributed PCs. DMTF had already recognized the
CIM initiative and included it as part of the initial releases of DMTF standards. The move to integrate CIM as a
DMTF standard came about the time another consortium standard initiative was passed into the DMTF this
was WBEM, or web-based enterprise management.
The WBEM initiative began the development of particular standards for enterprise management of computers
over the Web. As Microsoft was one of the original members of the WBEM consortium, one of the standards
initiatives it began with was the CIM standard. So any vendor wishing to develop products that could be
managed uniformly, regardless of vendor association, and managed over the Web, must meet the CIM and
WBEM standards.
Consequently, specifications for general storage devices (say, IDE and SCSI drives) and offline media, such
as tape and optical peripherals, were not added until later. The component descriptions for Storage Area
Networks are very new and have only been added as of the writing of this book. However, given vendor
acceptance and cooperation, CIM and WBEM do provide a uniform description for storage devices and
storage network components. As these specifications are incorporated within vendor products, the data center
begins to see a consistent view of the storage infrastructure.
The important points here are the eventual scope of the CIM specifications and the use of CIM as a standard.
As a standard, that means sufficient vendors must integrate their product offerings into the CIM standard
before the concept becomes relevant to the data center. As a meaningful solution, both vendors that develop
management products as well as vendors that develop storage networking products must adhere to the CIM
standard for this to be productive. In other words, it would be difficult if you have vendor As disk products that
conformed to the CIM standard, and vendor Bs disk products that did not but conform to their own
specifications of storage devices and activities. The same can be said for vendors who provide products for
performance management and capacity planningthe standard must support all instances of CIM
implementation at the storage hardware and software level. In addition, the CIM specification itself must have
sufficient detail to provide value.
SNMP and MIBs
MIBs, on the other hand, have come through the standards process and been accepted as a valid solution for
collecting and describing information aboutget ready for this networks. Yes, the MIB specification came
from the network environment and continues to be used today as a fundamental element of network
management activities.
Management information bases (MIBs) are very complex file-oriented databases that describe a network and
its components and act as a repository for activity information that occurs within the network. The MIBs were
the database for a distributed protocol that is used to access remote networks for management purposes. This
protocol is known as the Simple Network Management Protocol, or SNMP. Anyone with networking experience
should be quite familiar with this concept and protocol.
SNMP and their related MIBs create a way of collecting information for inclusion in performance, problem, and
capacity management. However, SNMP and their related MIBs are complex systems that require specific
programming to derive value. Many network management products base their functions on SNMP and MIB
standards. This proximity of network technologies within the SAN environments prompted the inclusion of
MIBs within the SAN switch software.
Leveraged by SAN vendors as an accepted network management tool, the SNMP and MIB combination for
SAN management laid the groundwork for todays management repository for SANs. Although not as likely to
be accessed or included within the micro-kernel applications, the NAS configurations can also be included in
SNMP and MIB solutions. The caveat is the complexities of the solution and the long-term viability of the MIB.
The data center must rely on third-party software products that integrate SNMP functions while allocating their
own MIB files, or face writing their own SNMP scripts and defining the MIB files themselves. The latter is a
complex, time-consuming task within the data center.

Modeling Performance and Capacity Requirements
One objective of the capacity planning activities is to estimate as accurately as possible the future needs of
users. This generally can be accomplished from modeling the potential workloads and configurations for
adequate fit. In general-purpose capacity planning terms, there are levels of modeling that can fit the needs of
most data-center requirements. These solutions range from expensive and specialized software tools that
model workloads and potential performance based upon hardware configurations, to vendor benchmarking
that generally uses subsets of the actual data-center workloads, and the official performance councils that
provide third-party benchmarks of vendor solutions.
The first set is the capacity planning modeling tools. Although these provide the greatest detail, their accuracy
depends on the data applied to the model, and the expertise in analyzing the output of the benchmark results.
Unfortunately, these tools have yet to be developed for storage networks, but through creative efforts could
probably be modeled given sufficient time and money. Even then, the model would be an integrated simulation
of workloads that share common storage arrays, or in the case of NAS workloads, that provide additional
latencies as remote network drives.
The second set is the vendor benchmarks, although these by their very nature will be suspect given their
inability to replicate the specifics of an individual data center. These simulations dont always have the
disparate facilities that make up production data centers, and as a result, the benchmark may be skewed
toward the vendors solution. Wouldnt that be a surprise? However, vendors benchmarks provide valuable
insight into understanding the potential capacity and performance of an expensive storage infrastructure
installation. The additional aspect is that many first-tier vendors have user benchmark centers where they test
potential customer solutions as well as conduct their own interoperability testing.
The third set is the third-party benchmarks by non-profit corporations that sponsor testing and performance
benchmarks of real-life configurations. These companies are likened to the insurance safety councils that
perform crash tests. The performance councils take off-the-shelf equipment from vendors and build a real-life
configuration in order to run a simulated workload based upon end-user applications, such as OLTP and data
warehouse transactions. In other words, they test out the configuration in real-life scenarios so as to validate
all the factors a data center would consider when purchasing the configuration. Two are relevant to the
storage industry: the Transaction Processing Performance Council (TPC) and the Storage Performance
Council (SPC).
The TPC provides benchmark testing of computer configurations using standard transactional sets. These
benchmarks execute transactions that characterize database queries which simulate everything from simple
queries to complex data warehouse queries that access multiple databases. The tests are run on vendor-
supplied hardware and software configurations that range from homogenous hardware systems to
heterogeneous software operating environments and database systems. The test results are generally
published and available for purchase through the council. This allows data centers to monitor different levels
of potential configurations at arms length while obtaining information about potential cost-to-operating
environment requirements. This provides an evaluation of storage from an integrated view, as storage
configurations become part of the systems overall configuration.
The SPC is specific to storage and is the new kid on the block when it comes to evaluating vendor storage
configurations. This is the most specific and productive modeling available to date for storage networking and
capacity modeling. Their job is to be the insurance safety council for the storage industry and protect the data
center from products that continue to be problematic in real-life implementations, while providing an effective
feedback mechanism for vendors who strive for better goods.
The SPC-specific objectives are meant to provide both the data center and systems integrators with an
accurate database of performance and price/performance results spanning manufacturers, configurations, and
products. They also use these experiences to build tools that help data centers analyze and effectively
configure storage networks.
They do this through a series of configuration requirements, performance metrics, and tests. The services can
analyze small subsets of storage, from JBOD and RAID storage arrays to large-scale SAN configurations.
However, all configurations must meet the following criteria prior to testing:
Data Persistence Storage used in an SPC test must demonstrate the ability to preserve data without
corruption or loss. Equipment sponsors are required to complete audited tests that verify this capability.
Sustainability A benchmark configuration must easily demonstrate that results can be consistently
maintained over long periods of time as would be expected in system environments with demanding long-
term I/O request throughput requirements.
Equal Access to Host Systems All host systems used to impose benchmark- related I/O load on the
tested storage configuration must have equal access to all storage resources.
Support for General Purpose Applications SPC benchmarks provide objective and verifiable
performance data. Specifically prohibited are benchmark systems whose primary purpose is the
performance optimization of the SPC benchmark results without corresponding applicability to real-world
applications and environments.
Vendors who submit their products to these benchmarks must have their systems available to ship to
customers within 60 days of reporting the SPC benchmark tests.
Probably the most valuable aspect of the SPC benchmarks is the actual test. The SPC has developed two
environments that depict many of the workload demands we have previously discussed (see Chapter 17). The
following describes two test scenarios that are run.
SPC1 IOPS (I/Os per Second) Metric An environment composed of application systems that have many
users and simultaneous application transactions which can saturate the total I/O operations capacity of a
storage subsystem. An OLTP application model makes up the benchmark where the success of the
system rests on the ability of the storage system to process large numbers of I/O requests while
maintaining acceptable response times to the end users.
SPC1-LRT (Least Response Time) Metric This environment depicts a batch type of operations where
applications are dependent on elapsed time requirements to complete. These applications provide
multiple I/O requests, which are often serial in naturein other words, they must complete in a predefined
order. The success of the storage system in these processing environments is dependent on its ability to
minimize the response time for each I/O request and thereby limit the elapsed time necessary.
The SPC carefully audits and validates the results of benchmarks. Configurations and testing criteria is
audited and validated either onsite or remotely through an audit protocol. This serves to provide the vendor
with audit certification that the tests and configurations meet the SPC standards and testing criteria. A peer
review is conducted upon the completion of benchmark results. Results are considered validated and become
official upon the completion of the 60-day peer review process if no compliance challenges have been brought
forward. Official results are available to SPC members on their web site and open to certain publication rights.

Implementing the Plan
Taking into account a comprehensive analysis of the existing or proposed storage infrastructure,
implementation becomes the critical factor. A big factor in implementation is the acquisition of upgrade
hardware and software or a completely new solution. Coupled with acquisition of the storage networking
components is the scheduled installation, testing, and transfer of production storage data and workload I/O
processing. This is followed by the initialization of the systems management cycle of activities.
Acquisition of storage capacity is an exercise in commodity purchases driven in most data centers by price.
The best price per MB usually gets the business and drives storage vendors to differentiate their offerings in
the firmware and software solutions that become part of the system. However, the storage networking
business has turned this upside-down, with the necessity of new networking devices and the enhanced
storage arrays and tape systems that must be FC-enabled. This has placed price in a more balanced
perspective with other major purchase factors, such as reliability, service, and quality.
Consequently, the following guidelines can be used to assist in data-center acquisition strategies when
purchasing storage networking solutions.
Competitive Bids Its important to gather at least three competitive bids for the configuration you are
seeking. Ensure that the requirements for storage networking, be it SAN or multiple NAS solutions, are
available to the vendors you have decided to work with. Be advised that total solution offerings that are
provided through storage vendors and systems integrators allow them an additional revenue source.
OEM Knowledge Most storage networking vendors, especially larger system vendors, provide their
solutions as a composite of external components supplied by the third-party companies. These
companies provide their component, be it hardware or software, as an Original Equipment Manufacturer
(OEM) supplier, and place their name on the device. This is commonplace in storage networking with
most SAN solutions being made up of OEM suppliers. This is not a bad thing, but its important to
compare apples to apples when considering competitive bids. In other words, dont pay extra for the same
FC equipment, such as switches, HBAs, or routers that is available through a competitive solution.
Storage vs. the Network Try to separate the storage part from the network part. This is especially helpful
in SAN acquisition, given that the network portion is likely to expand at a different rate than the actual
storage arrays. The storage capacity plan should provide for incremental upgrades through the planning
cycle, generally a year in length. During this time, storage array acquisition can be part of the competitive
bidding process, but should be negotiated with as little bid lock-in as possible. In other words, have the
vendors consider bidding on the entire plan (for example, a complete year of requirements) to obtain a
better price, service, and vendor commitment.
Many of the examples shown in the book depict storage networks supporting three common types of
workloads: OLTP, Web Internet Based, and Data Warehouse. More details on these I/O workloads can be
found throughout the book with additional guidelines on identification and estimating in Chapter 17. Typically,
SAN configurations are comprised of combinations of 8-, 16-, and 32-port FC switches, with disk arrays
commensurate with storage capacities that have been estimated with workloads. Within NAS installations,
typical configurations can encompass both departmental solutions and support within the data center for more
sophisticated applications. Another important point to consider is the external factors that influence the
installation. These are Ethernet network modifications within NAS installations and should be considered prior
to installation. SAN configurations are subject to the inclusion of intersystem-link ports (ISLs) and an
integrated FC-SCSI bridge into a tape library.
Test Installation
It is important to define, configure, and install a permanent test installation environment. Putting a small
configuration in place provides essential first-case experiences in the configuration and operation of both SAN
and NAS installations. This also provides a test bed for testing future software and hardware upgrades while
enabling an application testing facility.
Use the test installation to initiate a pseudo-management practice. Management becomes a very challenging
activity, especially when operating the new devices and complexities in the SAN. It also is the most rapidly
evolving practice, with constant change occurring in software tools and accepted practices. Additional
discussion of storage network management topics can be found in Part VI.
need to be integrated into existing change management activity within the data center. Key among these is
tracking the changes made to all components of the storage network (see Chapter 23). It becomes particularly
troublesome if you formalize changes to the switch configurations and not to upgrades of critical components
such as HBAs, routers, and attached storage devices.
An important aspect of product installation is establishing a backout practice. Because the storage network is
an infrastructure in and of itself, the reliability can be problematic in the beginning, as with any new
technology. However, being able to back out quickly and return the production environment to an existing state
saves valuable time as you move into a storage networking environment.

Closing the Loop
Here, we come full circle. Back to the place we started in this part of the bookthat is, a sound and well-
thought-out capacity plan for storage networking. The benefits are evident in the increased availability,
reliability, and success of the storage networking infrastructure. In addition to setting the basis for a
manageable set of service levels, the capacity plan supports the external systems and applications
departments in the data center. More important is the credibility it establishes for that support.
The rationale that storage networking is a new technology and does not lend itself to traditional or summary
planning for capacity is incorrect. As we have demonstrated, the process of estimating resources and
configurations can be done. However, it is an iterative process and one that becomes more accurate as
mistakes and related information arise. As new information and tools become available, the accuracy and time
required to plan for effective storage networking configurations will increase. Key among these will be the
increased information collection methods and the ability to provide effective statistical predictions without the
collection of months and months of detailed activity information.
This must be balanced against the fast-moving storage networking industry where new technologies become
available almost every calendar quarter. The difficulties behind planning without the knowledge of technology
advancements will move both the vendor community and the data center to reach some level of product
planning communications that is beneficial to both parties. As discussed in Chapter 20, new innovations
continue to affect the storage networking solution. Among these is the coalescing of SAN and NAS
architectures and the advancement of traditional computer bus technologies into switched fabrics.
Storage networking has become the cornerstone of new applications within the data center and continues to
deliver the scalable performance necessary for the future. However, these new storage functions must be
applied in an organized and planned fashion. Capacity planning for storage networking infrastructures may
form the foundation for the next level of storage management, culminating in a true paradigm shift within the
data center.

Overview
As storage networks continue to grow and become more complicated, management processes and security
procedures will play an integral role. While management is a key requirement in any type of storage
infrastructure, security is the one component that could undermine the availability of both management and
storage. This chapter provides an introduction to many of the existing and future security problems in storage
networks and the correlating best practices that can be implemented today. Although this chapter will not
cover every single storage security problem or provide in-depth coverage on each security exposure identified,
it will discuss the architectural and tactical problems that most often plague storage networks and how to
approach a solution for each of them.
While various trends have occurred in the digital marketplace over the past five decades, both storage and
security have received more attention in the last decade (early 1990s to present) than the first four combined
(excluding major governmental agencies focus on security). Furthermore, with the growth of the Internet,
security has been the focus of many conversations concerning network architecture, application design, and
now storage technology.
The terms storage technology and security architecture would not have been mentioned in the same sentence
a decade ago, or even five years ago. Today, however, with significant changes in the storage industry (the
expansion of the storage network into the WAN and beyond the LAN) and due to education gained from the
security mistakes in Internet Protocol (the numerous security problems with IP version 4), the relationship
between storage technology and security architecture has become an important one. Todays storage industry
has an opportunity to pursue and support a healthy security posture before storage security problems are
widely identified and, more importantly, widely exploited.
Its important that you understand why storage and security must coexist so that you can anticipate an attack
and understand an attackers (hackers) mindset. For example, consider a home with a front door, back door,
and garage door as three legitimate entryways. The front door has three security devices: a door-handle lock,
a dead-bolt lock, and a chain lock to prevent an intruder from entering. In addition, the homeowner has
secured the back door with two security devices: a Master lock combination on the fence leading to the back
door and a dead-bolt on the back door. The garage door is opened and closed only with a garage door
opener, which is in the vehicle at all times. Additionally, the homeowner has purchased a security system to
alarm the authorities in the event of a front or back door break-in (similar to an intrusion detection system, or
IDS, in a network). This adds an extra obstacle for access to the front and back doors.
NoteAn IDS is a device that passively monitors networks to determine and report any type of malicious
activity that is being conducted.
Now consider the security of the house (or the computer network): The intruder must consider which is the
best route, in terms of time and success, to enter the home (or network) and gather priceless goods. The front
door contains three security devices. The door lock and dead-bolt can be picked, but this takes a significant
amount of time and skill. Furthermore, even if the two locks were picked successfully, the chain lock would
have to be cut with an industrial sized chain-cutter, requiring more skill and time. And then theres the alarm to
consider. As a result, the front door sounds like a tough option.
For the back door, the fence Master lock combination will also take a significant amount of time because an
infinite amount of combinations would need to be attempted. Even after the combination is finally brute-
forced, the back door dead-bolt would have to be picked, also taking time and skill. Even if this door were
opened, the alarm would sound. This option also seems pretty unpromising.
The garage door is secured with the garage door opener. However, most garage door openers do not use an
overly complicated infrared architecture to open the garage door. In fact, a local hardware store usually carries
a device that spans several channels that might connect on the correct channel and open any garage door. In
addition, most hand-held devices, such as Palm and PocketPC, can have applications that capture the
infrared signals of a garage door opener to open it exclusively with the hand-held device. The purchase of this
device, and the fact that the homeowner will probably be opening/closing her garage at the same time every
morning, can result in an intruder using a hand-held device to capture the infrared signal and eventually open
the garage. Once inside the garage, the only thing stopping the intruder is the house garage door but many
homeowners do not lock that door because of the assumption that an intruder would not be able to enter the
locked garage.
This home example relates to the digital storage network and the mindset of a digital attacker. Although the
storage network is not the traditional access point for most attacks, a savvy hacker can avoid the hassle of
subverting multiple firewalls, switches, router Access Control Lists (ACLs), Virtual Private Network (VPN)
devices, and IDS sensors to gain access to data via the less-protected storage network that has direct access
to all of the important data.
NoteRouter ACLs are used to allow or deny access to networks based on IP addresses. VPN devices
allow for remote networks or individual users to connect to internal networks in a safe and secure
manner. A VPN allows multiple networks in different geographic locations to exist as a large virtual
network.
Attackers are not interested in gaining administrator rights or even root access to a given host; rather, they are
interested in access to data. The fact that the storage network contains sensitive data and is not adequately
protected (similar to the house garage doors) leaves the perfect opportunity for an attacker. Furthermore,
many of the storage protocols in place today, such as Fibre Channel and iSCSI, are bandwidth and
throughput-focused protocols with security usually absent (similar to the poor infrared channels on many
garage door openers). This scenario leaves the door wide open for attackers to subvert storage protocols
and get direct access to data, without compromising one firewall or encryption device.
Even though the storage network is often not protected thoroughly, it has access to the companys critical data
and intellectual property. The myth that the storage network cannot be reached by attackers is easily
subverted with a variety of techniques that are usually easier than going through traditional networks with
multiple firewalls, switches, router ACLs, VPN devices, and IDS sensors. This fact makes the decision easy
for the attackers on which route they should take to access data quickly and easily.
This chapter introduces the concept of security in storage networks and discusses basic principles and best
practices to consider. The following topics are discussed:
Overview of computer security
Security methods
Storage security technology
Storage security challenges
Fibre Channel SAN security
NAS security
Best practices

Overview of Information Security
While security threats used to mean hackers defacing web sites and short-term embarrassment for an
organization, new security threats are more serious matters corrupting or deleting intellectual property,
which results in the loss of revenue. The old model of viewing information security as an end result, such as a
product, was not successful because it did not support the core principles of a strong security posture. In
todays model, security is viewed as a process of good policies and procedures and is more successful and
realistic. Security is not a product for an auditor to check off as he or she reviews a list of items. A strong
security process in an organization provides a strong foundation to build upon and supports a stable security
platform.
To support a good security process, basic security elements should be addressed. These elements can be
applied to different aspects of storage networks, such as devices, applications, protocols, appliances, and so
on. The following are typical security elements that must be addressed by any secure solution:
Authentication
Authorization
Auditing
Integrity
Encryption
Availability
Authentication
Authentication is the process by which an entity is verified. The entity can be a packet, a frame, a login
request, or another entity. In all cases, the entity is identified and then authorized or unauthorized.
Authentication and authorization are heavily dependent on each other, because one can be subverted easily
in the absence of the other.
Authorization
Authorization is the process of determining which privileges are granted to which authenticated entity. Note
that authorization is not the same as authentication. Authorization simply allows or denies actions based on a
set of assumed authenticated credentials. Whether the authenticated credentials are valid is not possible to
verify with authorization. Authorization views a set of authenticated entities and allocates rights to those
entities.
Auditing
Auditing is the ability to capture and retain events that occur within the network or specific devices or
applications. While auditing is considered a passive security element, it can make a network aware of a
security incidence, which is often half the battle. In the absence of a pure security technology in the storage
network, such as a firewall, it is imperative that auditing on storage devices and applications be increased and
enabled to the fullest extent. An unsuccessful attack left unnoticed can leave an organization crippled in
security.
Integrity
Integrity is the assurance that unauthorized parties have not modified an entity. Furthermore, integrity confirms
that the data has not been altered in transit from the source to the destination. It allows a network to depend
on other security elements, such as authentication and authorization.
Encryption
Encryption is the process of protecting information from unauthorized access or modification by converting it
into cipher-text that can be accessed only through appropriate credentials or keys. Encryption also allows an
untrusted entity, such as a network, to be used without additional security elements for support. For example,
using encryption with a VPN device allows remote users to use the untrusted Internet as a medium for
business operations. Similarly, the use of encrypted protocols, such as Secure Shell (SSH), allows users to
use in-band data networks for management functions.
Availability
Availability means ensuring that resources are on hand for legitimate users, applications, or network devices
when requested. Security enables availability by ensuring that unauthorized user access or denial-of-service
attacks will be unsuccessful in a given storage object. If an attackers goal is simply to affect a loss of revenue
for a given organization, stealing the organizations data is not the only method that can be used to
accomplish this; simply making part of the storage network unavailable would result in loss of business
operations, which equates to the loss of revenue.
Overall, an organizations storage environment must address these security elements in some form to support
a strong storage security posture. The solution must also enable a process to grow and maintain the storage
security posture over an undefined period of time. A strong presence in security at one point in time does not
necessarily equate to a strong security presence in the future. A successful security plan will foster growth and
provide stability to the storage network over a period of time.

Security Methods
Critical for security is a layered security model, also known as defense in depth. In the defense-in-depth
model, layers of security are built in conjunction with one another in a complementary fashion. Many networks
are built with the M&M model,that is, hard on the outside and soft on the inside. This model crumbles after
a single penetration of the outer perimeter. A defense-in-depth model would not crumble if any devices were
subverted, such as the outer perimeter, because it would contain security layers behind each device.
An example of the defense-in-depth model in the storage network is an ACL on a storage node, such as an
EMC or Network Appliance NAS head, that restricts or permits access according to IP address or subnet.
These ACLs complement the required authentication and authorization procedures by the storage appliances
and/or operating systems. With this model, if one or both of the security elements were subverted, the
attackers would still be denied access if they were not sending the request from the correct IP address. Figure
25-1 illustrates the defense-in-depth model.
Figure 25-1: The defense-in-depth model
The defense-in-depth model allows organizations to protect their critical and sensitive storage data by
eliminating any single point of failure. As shown in Figure 25-1, the model can be as simple as enabling
security features on operating systems, storage switches, NAS heads, and even back-end disk arrays.
Additionally, security controls can be placed on host-bus adapters (HBAs), network interface cards (NICs),
client workstations, storage nodes, and storage applications. Because the storage industry has not yet come
up with any pure security device, such as a firewall, security features need to be enabled and explored at
other devices to support a defense-in-depth architecture to the fullest extent.
A single layer of security does not adequately protect an organizations storage network, proprietary data, or
intellectual property. In addition, identified security weaknesses in one area, such as a storage application,
can actually nullify strong security measures in other areas, such as a storage switch. A layered security
model, in which security is emphasized at key segments throughout the storage network rather than one or
two segments, supports a strong storage security posture.

Storage Security Technology
The storage security technology industry is currently in the development phase, and in this process more
questions than answers have been uncovered. However, emerging storage security organizations are
designing solutions to help secure storage networks from both unauthorized users and accidental
configuration mistakes. Existing storage vendors are also extending their product features to include security
solutions. The following sections discuss the following storage security technologies that are under
development as well as those that have already been deployed:
Inline encryption
Encryption at rest
Key-based authentication
Two-factor application authentication
Inline Encryption
Currently, the T11 committee is working on an initiative to incorporate IP Security (IPSec) technology
established in the IP industry into the storage industry with iSCSI and Fibre Channel Security (FCSec). T11
has established a framework that will use ESP (Encapsulating Security Payload) in Fibre Channel layer 2 to
protect frame data. ESP within the Fibre Channel frame will be able to provide confidentiality between two end
nodes, such as a switch and an HBA.
Additionally, ESP will be able to enhance frame security by providing authentication, integrity, and anti-replay
protection for each frame. This will prevent common attacks at the frame/packet layer, such as session
hijacking, spoofing, and Man-In-The-Middle (MITM) attacks (described later in the chapter). The adoption of
FCSec technology into the storage community will allow for greater flexibility for a storage network.
Specifically, the key goals for the FCSec architecture are as follows:
Node-to-node authentication
Node-to-switch authentication
Switch-to-switch authentication
Frame-level encryption
Node-to-node authentication will allow each entity to authenticate frames transmitted between two nodes. This
allows for the integrity of each frame between the two nodes and ensures against spoofing or replay attacks in
the fabric.
Node-to-switch authentication will be similar to node-to-node authentication in that it ensures integrity in each
frame. Additionally, node-to-switch authentication will allow switch management tools to contain added
strength in their allocation mechanisms (for example, zone allocation).
Switch-to-switch authentication will allow each switch to ensure the integrity of data and management, such as
Simple Name Server (SNS) information, to be protected against transport layer attacks. In addition, switch-to-
switch authentication will reduce the possibility of a rogue switch gaining instant control of the entire fabric.
In addition to the authentication capabilities of FCSec at the frame level, confidentiality can be provided by
using encryption of each frame as it exchanges data between two entities, including both node and switch
entities.
Encryption at Rest
The encryption of data at rest uses a different approach than inline encryption. Standards committees are
developing technologies such as FCSec to protect data and authentication for inline encryption. Encryption of
data at rest is being addressed by new storage vendors and new storage products.
Data at rest is one of the most commonly overlooked risk areas in storage security. While most security efforts
focus on protocols and architecture, the security of data at rest tends to be incorrectly assumed as secure.
Data on disk is often considered secure because it is kept deep inside the storage network, with multiple
storage devices restricting access. However, these types of security assumptions expose the data both to
accidental actions and unauthorized activity.
Several new organizations are designing products to encrypt the data before it reaches the disk. At that level
of encryption, if the storage node is compromised, the attacker will not have readable access to any of the
data residing on the disk. In the event of a configuration error, the data on the disk may not be fully exposed
due to the encrypted format of the data.
Various storage devices protect the data at rest in different ways, but all of them try to accomplish the same
goal of protection. The encryption solutions that encrypt data on disk protect against possible configuration
errors or mistakes commonly found in storage networks. Configuration errors are the most common problems
identified in storage networks today. If controls are not in place to mitigate possible configuration issues, the
ability to compromise data in the storage network might be larger than expected. Additionally, since storage
networks are complicated by nature, the likelihood of a configuration error incorrectly exposing the SAN
(Storage Area Network) is higher than for many other types of networks.
Figures 25-2 and 25-3 demonstrate two scenarios of configuration errors that might lead to the compromise of
data. Figure 25-2 shows how configuration errors can occur at the storage switch. Figure 25-3 shows how an
encryption device can protect against storage configuration errors.
Figure 25-2: Configuration errors at the storage switch
Figure 25-3: Encryption protects against storage configuration errors.
Key-Based Authentication
Fibre Channel-Generic Services-3 (FC-GS-3) is another technology that T11 standards organizations have
developed. Key server technology is discussed in FC-GS-3 to enable key-based authentication, which
supports the best practice of two-factor authentication in storage devices, such as switches, appliances, and
clients. Unlike the FCSec project, the FC-GS-3 project is not solely a project about security. FC-GS-3 includes
specifications dealing with management, directory service, time service, alias service, and other services.
However, we will be referring to FC-GS-3 only in terms of the key server technology that it addresses.
The Key Distribution Service discussed in FC-GS-3 will provide a method for secure distribution of public and
private keys in a storage network used for authentication, similar to a KDC (Kerberos Distribution Center) in IP
networks. The security key server in the storage network would use a level of encryption to protect the keys
and underlying protocols for verification purposes. By using a key server, data would be signed, providing a
secure means of authenticating the original node as well as verifying the data was not modified during transit.
This method of using a key server is often referred to as CT (Common Transport) authentication.
CT authentication is the process of authenticating requests between a client and a server. The CT
authentication mechanism creates a hash based on an algorithm and secret key to represent the message
signature associated with the sender. The hash is transferred as an extended preamble to the receiver. Upon
receiving the hash, the receiver computes the hash before authenticating it, using the same algorithm and
secret key, to make sure the hashes match.
The key server in the storage network would contain a copy of each nodes unique distribution key. The key
server would then authorize the security connections between two nodes and would generate a key to use for
authentication. This authentication key is then used by one node to authenticate to the other node. The single
point of failure in the specification is the key server. If the key server does not contain the proper amount of
security, the authentication model that the key server supports cannot be considered secure. Figure 25-4
shows an example of an authentication procedure using FC-GS-3.
Figure 25-4: Key distribution service
A practical example of FC-GS-3 would be the authentication of a World Wide Name (WWN) of an HBA in a
storage network. In the Fibre Channel SAN Security section of this chapter, the security problems of WWN
are discussed. It would not be best practice to use a WWN as the sole entity for authentication. The process
of using FS-GS-3 to identify HBAs uniquely would allow the use of WWN without concerns of spoofing or
replay attacks being performed successfully. Since WWNs are not reliable as a secure means of host
identification, using a key server will minimize the risk of spoofing or session hijacking and allow the use of
WWN with an acceptable level of integrity in the authentication process.
Currently, many switch vendors, such as Brocade and McData, are using a hybrid of the FC-GS-3
specification by using switch-to-switch key authentication. In this approach, each switch contains a unique key
to authenticate to another switch. Instead of using a third key server device, the two nodes are directly
authenticated to each other using their own keys. While this approach may not be as secure as having a key
server, it does prevent a variety of attacks, primarily the ability for a rogue switch to take control of a given
fabric, its simple name server, and its zoning architecture.
Two-Factor Application Authentication
Two-factor authentication for key applications, such as storage management applications, is becoming a
significant requirement for protecting storage networks. The fact that storage management applications can
subvert the majority of other security controls within the storage environment makes the management
applications a significant target for most attackers.
Currently, many management applications rely on a username/password combination for authentication. As
best practice, a single username and password should not be the only security entity protecting a storage
network. For example, an unauthorized user should not be able to plug into a network with his favorite backup
program and successfully manage storage agents because he was able to guess the username/password
combination or spoof an IP address.
Requiring more than a username and password or an IP address for authentication is possible. To address
this issue, certain vendors have included a second level of authentication within the management application
itself. These management applications use public and private keys for authentication with authenticated
Secure Sockets Layer (SSL) or Kerberos technology to mitigate some of the exposures described so far.
Therefore, if an unauthorized user has plugged into the network and successfully guessed a password or
spoofed a management IP address, the attackers rogue backup software would not be able to gain access to
the storage agents since a second level of authentication is required. The second level of authentication
required from a client application to the server application and/or agents could use tokens, digital certificates,
Kerberos tickets, public keys, or other forms of security. This type of application design may utilize
usernames/passwords and IP addresses, but it also utilizes credentials implicit in the client application to
authenticate to the server and/or storage agents.
NoteTokens, digital certificates, Kerberos tickets, and public keys can be used for authentication. All
these security entities can be delivered digitally to authenticate a sending node to a receiving node,
such as storage management software, in place of or in addition to username/ password
authentication.

Storage Security Challenges
As the storage industry grows, with both SAN and NAS architectures, the requirements for storage security will
also grow. Storage security will need to accomplish a number of milestones and challenges to become
mainstream in storage networks.
Unfortunately, a typical milestone that uncovers the need for security in a particular industry is often some
type of successful and well-publicized attack. For example, wireless security technology, specifically 802.11b,
would not have been viewed as such a severe security issue if hackers were not easily able to break Wireless
Equivalency Protocol (WEP) encryption. Despite the weakness of 802.11b, many organizations did not invest
in wireless security until well after the fact of the discovery of the security problem.
While a break-in might have encouraged organizations to invest in wireless security, to approach security after
a breach costs an organization significantly more time, money, and resources. Unlike the wireless industry, the
storage industry has more to protect than a possible credit card number going over the wireless network in
semi-clear-text. The storage system has an entire organizations intellectual property, sensitive data, and
possibly customer information to protect and maintain. Therefore, a successful storage security attack would
be more severe than a typical network attack.
Another major challenge in storage security is industry acceptance. Many storage administrators and experts
believe security issues in Fibre Channel, storage devices, and storage architectures are not important issues,
for a variety of reasons. However, because storage security has significantly more to protect than other types
of networks, the storage industry should be more aggressive in fixing and solving security issues.
For example, consider the security problems encountered in Cisco routers and switches in the early 1990s.
These routing and switching devices were developed based on availability and functionality. However, due to
the security weakness of IPv4 combined with IP device weaknesses, the devices were easy to subvert and/or
compromise. The same argument can be made for storage products that focus on bandwidth or speed and not
necessarily security. The weakness of IPv4 and Fibre Channel, combined with potential storage device
weaknesses, may well lead to the same results that the IP industry faced.
Industry must face the following key storage security challenges:
Products and solutions
Encryption
Security standards and specifications
Bandwidth/functionality tradeoffs
Products and Solutions
The first and major storage security challenge is the lack of security product solutions and storage security
features in current products. As described earlier, no pure security products are currently available for storage
networks to permit or deny access to certain storage nodes. Furthermore, a firewall or similar device would
have to support 2 Gbps throughput to hold any type of acceptance and success in the storage industry.
Currently, only certain firewalls can support near gigabit throughput, and these are available only for IP
networks. There is no such thing as a Fibre Channel firewall.
An additional storage security challenge is the lack of security features in storage products today and the
quality of security features that are included in existing products. A good example to demonstrate this problem
is the zoning and Logical Unit Number (LUN) masking capabilities in Fibre Channel networks. The zoning
capabilities in Fibre Channel switches and LUN masking capabilities in storage nodes were originally
developed for segmentation, not security. While the segmentation tools are good for secondary security, they
are not adequate to use as primary security tools. In addition, using zoning and LUN masking as the only
security tools does not support a best practice storage architecture. In fact, using zoning and LUN masking as
security tools is similar to using virtual LANs (VLANs) as the only security tools in an IP network which would
exclude the use of better security devices, such as firewalls, router ACLs, and VPN devices to protect the
network. Cisco has stated many times that VLANs should not be viewed as security tools, but rather as
segmentation tools. VLANs hopping ability (the ability to jump across one VLAN to another) has been
demonstrated by security professionals in a variety of tests.
In addition, most storage products provide only Telnet and web (HTTP) capabilities for a management, both of
which are clear-text protocols that can easily be sniffed. The use of encrypted management protocols, such as
Secure Shell (SSH) or encrypted web (HTTPS), is still in the process of being adopted.
Encryption
Encryption technology is another major challenge in the storage industry. Inline encryption in both IP and Fibre
Channel mediums is difficult to implement without significant bandwidth penalties. IP networks that have
bandwidth capabilities of Gbps are reduced to Mbps transfers once encryption technology is in place.
Similarly, Fibre Channel networks with 1.0 to 2.0 Gbps capacities would also be reduced to probably less than
half that amount. Considering the fact that Fibre Channel is often deployed specifically because of bandwidth
capacities, the fact that encryption would directly negate those capabilities is a significant security problem for
storage security engineers.
As mentioned, encryption of data at rest is another security challenge. The major challenge of data at rest is
interoperability with different types of storage devices. In addition, interoperability of all types of storage
appliances should be a requirement supported by standards bodies; unfortunately, competing vendors may
not share this perspective.
Security Standards and Specifications
Lack of security standards and specifications is also a significant problem in the storage security industry. A
good example is zoning definitions in Fibre Channel switches. The term hard zoning is defined one way for
certain Fibre Channel switch organizations and another way for other organizations. Some switch vendors
refer to hard zoning as the act of locking physical port numbers to a particular zone on a Fibre Channel switch.
Other switch vendors refer to hard zoning as a routing tool in which routes will not be broadcasted to non-
participants of the zone. This disparity between definitions is a standards problem. The lack of storage security
standards, from encryption specifications to zone definitions, leads to technologies that do not coexist easily
across various storage vendors. In addition, the lack of security standards and specifications will lead to
competing vendors developing technology that will most likely not interoperate with one anothers products.
This will make an end users storage security architecture a frustrating process to design and implement.
Standards in storage security need to define terms and protocol specifications before storage vendors grow
impatient. For storage security features, tools, and products to be successfully adopted by end users, a clear
classification of any type of architecture needs to be in place. Standards need to be developed in the following
areas:
Encryption standards
In-line and at-rest standards
Standardization of storage security terminology and definitions
Authentication standards
Bandwidth/Functionality Tradeoffs
Bandwidth and functionality tradeoffs are key challenges in creating storage security. Any type of functionality
or bandwidth loss incurred to gain added security will not likely be successful in the storage industry,
especially since many storage networks are deployed for the mere purpose of bandwidth and functionality
offerings. Any type of solution must have a minimal affect on storage functionally while demonstrating clear
interoperability with existing storage features.

Fibre Channel SAN Security
Fibre Channel SAN security deals with a variety of issues ranging from Fibre Channel frame authentication to
Fibre Channel switches. Fibre Channel SAN is an emerging technology that still has many items under
development, such as management, virtualization and interoperability. Furthermore, although some progress
has been made regarding security in Fibre Channel networks, several opportunities exist for growth.
In this section, we discuss some of the security problems in Fibre Channel SANs, at both the frame and
device levels. We continue to address some solutions to these security problems and best practices. The
following items will be discussed:
Zoning
WWN spoofing
Fibre Channel (FC) frame weaknesses
Sequence ID and control number
Disruption of flow control
MITM attacks (SNS pollution)
E_port replication
LUN masking
Zoning
Zones are to Fibre Channel switches as VLANs are to IP switches. Zones are used to segment the fabric into
groups, in which certain nodes are a part of a zone and have access to other nodes in the same zone.
Storage nodes could be a part of multiple zones or part of a single zone.
Two types of zoning are used, soft zoning and hard zoning. In soft zoning, the SNS in a switch uses an HBA's
WWN for zoning. An SNS is a table located in each switch and shared among switches in the same fabric,
which separates WWN into the correct zones. Figure 25-5 shows an example of soft zoning. In the figure,
WWN 9382108xxxx, 3859658xxxx, and 3582968xxxx have full access to each other. However, if any of these
WWNs try to access zone B members, such as 0038283xxxx, access would be denied.
Figure 25-5: Soft zoning in SANs
Hard zoning is not only different from soft zoning but is another term that has multiple definitions. One
definition for hard zoning is 'the act of locking physical ports to certain zones.' For example, physical ports 1
through 4 would be zone A, physical ports 5 through 8 would be zone B, and so on. If a WWN was connected
to port 2 and tried to access port 6 in zone B, it would be denied access. Figure 25-6 shows an example of this
definition of hard zoning.
Figure 25-6: Hard zoning in SANs
Another definition of hard zoning is a routing procedure. If route-based zoning is implemented, certain routes
will be publicized for certain storage nodes. For example, if node 1 were allowed to access node A, node 1
would be notified about the route to node A. If node 1 were not allowed to access node A, node 1 would not be
notified about node A. However, if node 1 knew the existence and route to node A, hard zoning, according to
this definition, would not prevent node 1 from this access. Hard zoning, in this reference, is not a restrictive
tool; rather, it's an information tool.
World Wide Name Spoofing
WWN is used to identify an HBA in a storage area network. WWN is used for authorization of data from one
particular zone to another. Additionally, WWNs are used in soft zoning procedures on storage switches to
separate certain servers and data. Soft zoning separates WWNs into different zones. Because a switch
contains an SNS table that matches up each WWN to a particular zone, a particular WWN would be granted
or denied access to another storage node based on the results of the SNS zone table.
WWN spoofing is a high-risk problem in storage area networks because WWNs are often used as the only
tool for granting authorization. WWN spoofing is as simple as loading up the device drivers included with the
HBA and changing the WWN. An attacker can then spoof a WWN and gain access to data. The switches that
contain the zone information would grant access to this spoofed WWN because it authorizes the WWN and
does not authenticate it. Figure 25-7 illustrates this type of attack.
Figure 25-7: SAN World Wide Name table spoofing
As best practice, hard-zoning procedures can be used to deter spoofing problems. The physical definition of
hard zoning would not use WWN for authorization, but rather the physical port numbers. If an attacker spoofs
another WWN, the hard-zoning switch would still not grant it access since the physical port on the switch is
trying to gain access to another port to which it does not have authorization, regardless of what WWN it claims
to be. However, using the route-based definition of hard zoning, WWN spoofing would not be a good
alternative. Because the route-based definition is not a restricted tool, but rather a tool used to deliver routing
methods, a hard-zoning switch could still be subverted if the attacker knew the address and the route to its
target. Therefore, if a WWN were spoofed with route-based hard zoning, access would be permitted.
Frame Weaknesses
Fibre Channel architecture contains five different layers, numbered 0 to 4. Despite the differences between IP
and Fibre Channel, Fibre Channel frames contain weaknesses that are similar to current weaknesses in IPv4
packets. These IPv4 weaknesses have been turned into vulnerabilities and exploited at a variety of levels. The
weaknesses in Fibre Channel frames are specifically in Fibre Channel layer 2, known as the framing
protocol/flow control layer. Fibre Channel layer 2 is where most security issues have been identified. Fibre
Channel layer 2 contains the frame header, which contains the 24-bit source address of the frame, the 24-bit
destination address of the frame, the sequence control number (Seq_Cnt), sequence ID (Seq_ID), and the
exchange information (Exchange_ID).
As shown as Figure 25-8, layer 2 contains the sequence control number and the sequence ID. A sequence is
a series of one or more related frames transmitted unidirectionally from one port to another. The sequence
series is responsible for keeping the connection alive between two nodes. A given frame must be a part of
some sequence between two end nodes. Frames within the same sequence have the same sequence ID
(Seq_ID) in the frame header, which is kept to identify frames that belong in the same transmission process
between two nodes.
Figure 25-8: Fibre Channel layer 2 and sequence control
In addition, as each frame is transmitted in a sequence, the sequence count number (SEQ_CNT) is
incremented by one. Therefore, that transmission can remain initiated and frames can remain ordered by the
incrementing process. However, the fact that the sequence ID is a constant value, thus predictable, and the
sequence control number is incremented by a predictable number, which is one, makes the sequence series a
predictable value. Because the sequence series is responsible for maintaining the session between two
nodes, an attacker would be able to launch a session-hijacking attack and take control of the session.
This attack was made popular with IPv4 packets with similar problems of predictability in Initial Sequences
Numbers (ISNs). In ISNs, the hijacking packet would simply need to guess the predictable sequence, similar
to the Fibre Channel frame, and take control of a management or data session. Figure 25-9 illustrates this
attack at a high level.
Figure 25-9: Fibre Channel frame header attack
The use of predictable session information leaves Fibre Channel frames vulnerable. Additionally, in-band
management protocols, such as SCSI Enclosure Services (SES), can allow management sessions to be
hijacked and exposed to unauthorized users. As in-band management methods and applications grow, the
weaknesses of the session information in Fibre Channel frames could amplify and result in loss of data.
The solutions for predictable sequence numbers in Fibre Channel frames needs to be similar to the solutions
for IPv4 vulnerability. Operating system settings and encryption technology enabled IPv4 weaknesses to
reduce the likelihood of successful hijacking attacks. Similarly, Fibre Channel frames need to make session
information unpredictable. This change can come from the HBA drivers that are installed on the operating
systems or within the fabric itself. Either solution should be able to address the static nature of sequence IDs
and the predictably of sequence control numbers.
While encryption may be a solution for an IPv4 network, using encryption as a solution for Fibre Channel
frames is not as easy, since encryption involves extra overhead and bandwidth costs that may not be
acceptable in SANs. As a mitigating solution to encryption, separate out-of-band management networks
should be deployed to eliminate the possibility of an attacker hijacking management frames to access storage
devices and their data.
Disruption of Flow Control
Flow control is responsible for message flow between two Fibre Channel nodes. In Fibre Channel networks,
devices transmit frames only when each device is ready and able to accept them. Before these devices can
send and receive frames, they must be logged in to each other and must have established a credit level to
send and receive the correct amount of frames. The establishment of credit from one storage node to the
other is conducted at the Exchange_ID level of the frame header. This credit refers to the number of frames a
device can receive at a time. This value is exchanged with another device during login, so each node knows
how many frames the other node may receive.
The problem with flow control is that Fibre Channel frames are unauthenticated. An attacker can therefore
send out incorrect Exchange_ID information to a variety of addresses, thus creating a temporary denial-of-
service between two nodes. For example, because frames are unauthenticated, an attacker could replace
valid frames with incorrect frames that contain a higher Exchange_ID number between two nodes. Because
one node will now be sending more frames than can be received, the receiving node will be unable to accept
the frames properly and will not be able to respond appropriately. Similarly, instead of creating a higher
Exchange_ID number, an attacker could lower the number for the Exchange_ID information, thus leaving one
node waiting for more frames to process. This would result in the frames being passed slower from one node
to the other, delaying the process of data communication.
The solution for the disruption of flow control is authenticated frames. Because the process of exchanging ID
information requires logging in to the other node, authentication of the node should also be addressed at this
point. However, while this attack is successful for the disruption of flow control, the disruption is actually quite
minimal and the fabric has the ability to correct itself quickly.
Man-in-the-Middle Attacks
MITM attacks have plagued IPv4 networks for some time. The problem in IPv4 is the use of Address
Resolution Protocol (ARP) packets. ARP is a protocol that matches a machine's NIC address-known as the
Media Access Control (MAC) address of the machine-to an IP address. However, both the IP address on a
network and the MAC address can be easily spoofed. Because no authentication of ARP packets is required
in IPv4, a spoofed IP address to another MAC address can redirect information to unauthorized users. The
attacker could send out ARP packets that match their MAC addresses to a target's IP address. Therefore,
when data is sent to the target, the data will be sent to the attacker since the target's IP address matches the
attacker's MAC address.
This attack can be partially replicated in Fibre Channel networks. Fibre Channel switches include a table,
referred to as the Simple Name Server. The SNS matches up WWN (similar to a MAC address) to the 24-bit
address of the storage node. Since Fibre Channel frames are also unauthenticated, an attacker can use any
traffic analyzer program to generate frames in the fabric with a target's 24-bit address and the attacker's
WWN, thus changing the SNS information. The SNS would have an updated entry for the 24-bit address with
a new WWN. When traffic is being sent to the target's 24-bit address, it would go to the attacker's machine
since it matches up to the attacker's WWN.
As shown in Figure 25-10, an attacker sends out a modified frame to xFFFFFE to log in to the fabric (FLOGI).
The modified frame has a source address of another trusted entity on the fabric, such as another trusted
switch, and the WWN of the attacker. The fabric assumes that the attacker is now the legitimate host since
the switch's 24-bit address is matched to the attacker's WWN. All frames destined for the real node are
passed to the attacker and then to the legitimate node.
Figure 25-10: Man-in-the-Middle attack on SANs
An MITM attack is simply polluting the SNS table with switches. Polluting the SNS could also be conducted
when joining the fabric. An N_Port would send a FLOGI (fabric login) to the well-known address of xFFFFFE
(similar to a broadcast in the IPv4). The switch receives the frame at xFFFFFE and returns an accept frame
(ACC). The service information would then be exchanged. Because no validation is required to receive an
accept frame, an attacker could send a modified 24-bit address to xFFFFFE to attempt to corrupt the SNS
table in the switch. As soon as the attacker receives the accept frame, the attacker knows that the SNS table
has been modified.
LUN Masking
A Logical Unit Number (LUN) is a unique identifier that is used to differentiate between devices, including
physical or virtual disks volumes. LUN masking allows nodes to be aware only of other files or data blocks that
they are authorized to access. This is implemented by masking off LUNs that may be available. Similar to
zoning, LUN masking is a segmentation tool that has been used for security. LUN masking can be
implemented at three levels: at the HBA, at the storage controller itself, and using a third-party device in the
storage network.
HBA-based LUN masking uses driver utilities that map WWN to LUNs. The HBA drivers contain masking
features that allow the WWN to view only a certain set of authorized hosts. Masking at the HBA level offers
easy implementation since a third-party device would not be required in the SAN; however, it does require
additional system administration resources since each HBA in a storage network would have to be individually
configured. This solution may work well for small storage environments, but it is virtually unmanageable in
large SANs. In addition, since LUN masking occurs at the HBA level, if the operating system were
compromised, an attacker could change the masking information without any other security controls
preventing this. This would allow an attacker to load up device drivers and change the allocation to include as
many LUNs as desired, thereby negating any security gained by the masking rules.
Conducting LUN masking at the storage controller is another way of implementing LUN security. The storage
controller would map access privileges from each HBA's WWN to the authorized LUN. LUN masking at the
storage controller level is probably the best alternative for most SANs, since it has the ability to scale in large
storage networks and does not require an extra storage device to add undue overhead. In this scenario,
however, LUN masking is dispersed across many storage devices instead of being centrally managed at one
device.
The third method of implementing LUN security is using a third-party storage device for allocation. The third-
party device would be placed between the client nodes and the storage nodes. The device would handle
requests for LUNs before they proceeded to the storage controller. Once the requests have been made, the
device would grant or deny access. If the request is granted, it is forwarded to the storage devices. This
solution scales appropriately to enterprise-sized storage networks and provides a central resource for LUN
management. However, it places yet another storage device in the storage network. Furthermore, it adds
another layer of overhead from the client node to the storage controller.
Table 25-1 compares and contrasts each type of LUN masking technique and the pros and cons of each.
Table 25-1: LUN Masking Techniques Compared
LUN Masking Type Pros Cons
Storage controller Scaleable to large SANs.No
undue overhead with another
storage device.Security
cannot be subverted by an
operating system
compromise.
Management of LUN allocation is
dispersed throughout the SAN instead of
being centrally managed.
Third-party device Scaleable to large
SANs.Centralized
management of LUN
allocation.Security cannot be
subverted by an operating
system compromise.
Added overhead by placing another
storage device in the SAN.
HBA level No undue overhead with
another storage device.
Only possible in small SANs (not
scaleable).Management of LUN
allocation is dispersed throughout the
SAN instead of being centrally
managed.Security can be subverted by
LUN Masking Type Pros Cons
an operating system compromise.
E_Port Replication
E_port (FC extenstion ports) replication is a feature in many Fibre Channel switches. Each port on a switch
has a particular function (refer to Chapters 14 and 15 for additional information on FC ports). Each type of port
holds a different function, whether it connects two client nodes together or connects two switches together. An
E_port, for example, connects two switches together, enabling the connected switches to exchange routing
and zoning information.
Because no authentication is required, E_port replication security is not guaranteed. If a storage server
provider (SSP) were extending its fabric to a client using unauthenticated E_port replication, the simple act of
plugging in the storage provider's E_port to the client E_port would exchange routing and zoning information
for the entire fabric, not just for a specific zone. This would allow an unauthorized switch to access
information.
To prevent accidental E_port replication, port-type locking and key-based authentication between switches
can be used. Products from major switch vendors, including Brocade, McData, and Qlogic, can lock port types
rather easily. For example, if physical ports 1-11 connect only client nodes to storage nodes, the port type can
be locked to n-type. Similarly, if only port 12 is used as an E_port, it can be locked using port-type locking.
This would prevent an attacker from replacing a client node with a switch and have an E_port exchange
information from the two switches automatically.
Key-based authentication between the switches is also a best practice to prevent unauthorized E_port
replication. Key-based authentication would require each switch to authenticate before being added to the
fabric. This would disable any type of communication to and from the switch until validation keys or certificates
have been exchanged from the joining switch to the existing switches.

an operating system compromise.
E_Port Replication
E_port (FC extenstion ports) replication is a feature in many Fibre Channel switches. Each port on a switch
has a particular function (refer to Chapters 14 and 15 for additional information on FC ports). Each type of port
holds a different function, whether it connects two client nodes together or connects two switches together. An
E_port, for example, connects two switches together, enabling the connected switches to exchange routing
and zoning information.
Because no authentication is required, E_port replication security is not guaranteed. If a storage server
provider (SSP) were extending its fabric to a client using unauthenticated E_port replication, the simple act of
plugging in the storage provider's E_port to the client E_port would exchange routing and zoning information
for the entire fabric, not just for a specific zone. This would allow an unauthorized switch to access
information.
To prevent accidental E_port replication, port-type locking and key-based authentication between switches
can be used. Products from major switch vendors, including Brocade, McData, and Qlogic, can lock port types
rather easily. For example, if physical ports 1-11 connect only client nodes to storage nodes, the port type can
be locked to n-type. Similarly, if only port 12 is used as an E_port, it can be locked using port-type locking.
This would prevent an attacker from replacing a client node with a switch and have an E_port exchange
information from the two switches automatically.
Key-based authentication between the switches is also a best practice to prevent unauthorized E_port
replication. Key-based authentication would require each switch to authenticate before being added to the
fabric. This would disable any type of communication to and from the switch until validation keys or certificates
have been exchanged from the joining switch to the existing switches.

NAS Security
Network attached storage (NAS) security addresses security at a device level. Traditionally, NAS devices,
such as filters, are marketed as pure storage devices. Even though many are also deployed as storage
devices, most, if not all, have built-in operating systems, usually based on some sort of UNIX flavor, that
include services and protocols that need to be secured in a way similar to any other type of device on a
network. Additionally, NAS devices are usually based on IP technology, using Ethernet or gigabit Ethernet,
which is threatened by the IPv4 exposures. The two most common implementations of the NAS are Common
Internet File System (CIFS) and Network File System (NFS). The following section focuses on security
concerns of CIFS and NFS NAS devices.
Common Internet File System
CIFS is a standard protocol that is used for data sharing on remote storage devices on IP networks. CIFS
security focuses on two security elements: authentication and authorization. Two types of access rights can be
used for CIFS, each with its strengths and weaknesses: these are share-level authentication and user-level
authentication.
Share-level authentication is based on share points and passwords. Because share points can be made
available to multiple users with required authentication of each user, no accountability or authentication is
required on a user-by-user basis.
Following are its specific vulnerabilities:
User accountability or authentication is not required by individual users (no record of the access).
A single password, which is shared by design, is responsible for the entire share point.
Passwords must be changed every time an employee is terminated or resigns.
Passwords are transmitted in clear-text (however, all Windows 2000/XP machines and NT 4.0 service
pack 3 machines support non-plaintext password authentication).
Per-user password authentication for a NAS appliance is best practice in terms of security. Some key options
are required to attain a desired level of security. Both LAN Manager and NT LAN Manager (NTLM) have
associated severe security weakness, making them capable of being reversed engineered and reduced to the
equivalent of a clear-text protocol. A good solution is to use Windows support of NTLM v2.
In addition to NTLM v2 for username/password protection, Server Message Block (SMB) signing should be
used for CIFS SMB communication. SMB signing places a digital security signature into each SMB packet,
which is then verified by both the client and the server. Without SMB signing, the server authenticates the
client, but the client never truly authenticates the server, so mutual authentication is not completed. As best
practice, mutual authentication should be enabled, especially in an IP network, to eliminate the possibility of
IPv4 attacks.
Network File System
NFS is also a standard protocol used for data sharing on remote storage devices on an IP network. NFS is
basically the equivalent of CIFS, but in the UNIX environment. NFS is used to allow a computer to mount
(share) files over a network. NFS began as a User Datagram Protocol (UDP), but many Transmission Control
Protocol (TCP) implementations exist today.
Like CIFS, most NFS implementations have security weaknesses:
NFS communication is not encrypted (all traffic is in clear-text).
NFS clients (hosts) have a limited level of authentication.
NFS users are not easily authenticated.
As best practice, NFS storage devices should not solely grant or deny access to NFS mounts based on the
host name of the client. IPv4 attacks allow host name and IP spoofing to be accomplished quite easily,
possibly exposing the entire storage appliance to unauthorized users. Furthermore, this type of NFS mounting
does not require passwords for authentication, exposing the data significantly.
Furthermore, per-user password authentication for an NFS appliance is best practice in terms of security.
However, some key options need to be enabled to secure this method properly. Obviously, clear-text
communication should not be used from an NFS client to NFS appliance. Certain NFS appliance vendors do
support RSA/DES encryption with NFS communication. This will eliminate the transfer of sensitive information,
such as usernames, passwords, NFS mounts, file handles, and contents of data.
NoteRSA (Rivest, Shamir, Adleman) and DES (Data Encryption Standard) are algorithms used for
encryption and authentication systems.
In addition, Kerberos (v5) is supported by many appliance vendors, which significantly reduces the risk of
username and password credentials being compromised or replayed. Here are some specific solutions:
Encrypt NFS communication with either RSA/DES algorithm (supported by most vendors).
Do not authenticate based solely on host name.
Enforce Kerberos (v5) for username/password authentication.

Best Practices
As mentioned, a defense-in-depth architecture protects a storage network's critical data. However, a layered
security model can be subverted if simple best practices are not followed when designing storage security. For
example, consider the home security example. The homeowner had spent several thousand dollars buying
locks and security systems for her home. However, a simple best practice, such as not locking the house
garage door, could negate all the other strong security measures that she had placed in the home.
Simple best practices in storage security can help prevent many of the common security attacks that
unauthorized users will attempt. Adhering to the majority of these best practices can leave an attacker
frustrated or bored and motivate him to move to another target or stop the attack altogether.
Following are some basic best practices to follow:
Strong passwords
Configuration
Strong management
Clear-text management
Remote connectivity
Operating system security
Storage appliance services
Passwords
Using strong passwords and changing default passwords on storage devices is good security practice.
Whether it is an EMC Cellera, a Network Appliance filter, or a Brocade switch, basic and simple passwords
can be easily obtained and guessed. In addition, because most storage appliances have some sort of web
management capability that is enabled by default, accessing the storage appliance and guessing passwords is
not difficult for an accomplished attacker. Several tools exist to brute-force web-based authentication forms
that can attempt more than 100 passwords in only a few minutes.
Weak passwords lead to a bigger problem, subscribing to the myth that storage networks are isolated and
cannot be accessed from regular corporate networks. Many storage administrators believe that the storage
network is not easily accessible. However, one of the primary purposes of a storage network is to connect
devices on the network to back up data. Therefore, since these connections are permitted, storage networks
are accessible, either indirectly or directly.
Using weak passwords can also significantly cripple storage security. While many organizations change weak
passwords from prom or password to admin or manage, both admin and manage are common passwords that
are contained in most tools that attackers use to crack accounts.
Following is a list of some common passwords. If these passwords are used in your storage network, consider
changing your password policy for storage devices:
password monitor <<switch vendor>> Config
admin temp <<company name>> Test
manage root Letmein secret
prom backup Secureme keepout
filer KuSuM abcd1234 Test123
netcache momanddad Money green
Configuration
Good configuration practices can build a solid security foundation in a storage environment. Furthermore, poor
configuration practices, an abundance of configuration errors, or the lack of configuration options can
significantly decrease the security posture in a storage network.
Most often, an attacker will complete a successful attack not because of some elite security problem that she
has discovered on the spot, but rather because of poor configuration decisions and/or configuration errors that
lead to wide open and unprotected devices. For example, a minor configuration error, such as a storage
administrator allocating the incorrect LUN to a given WWN, can expose data that is not authorized. A major
configuration error, such as a storage appliance backing up an entire file system with world-readable access
permissions, can also expose all data to unauthorized users. Both have a significant impact on the security
posture of the storage network, despite one being major and one being minor.
The lack of understanding of security options and/or features in existing storage products may also lead to
poor configuration decisions. For example, many end users are not aware of several security options that are
readily available on their storage appliances and/or storage switches today. Not only are these security
features not well advertised by vendors, but end users do understand how to use them. This combination often
leads to end users making configuration decisions that don't work for security and not considering
configuration decisions that are secure.
Management
Management is a critical component to storage security architecture. Since functional storage networks rely
heavily on storage management practices, it is imperative that a strong management environment exist.
Because storage network management contains a significant amount of control, a compromise of any
management entity would give an attacker a considerable amount of privileges.
Protecting management interfaces for a storage network is a significant best practice. It is possible to attack
many storage solutions through the Web, Telnet, or SNMP management interfaces. In many cases, gaining
access to management interfaces is as simple a wire sniffing, session hijacking, and replay attacks. In other
cases, it is as simple as loading an appropriate management program and logging in to the management
application.
Many storage devices and applications rely on unsafe and clear-text protocols (such as Telnet, SNMP, FTP,
CIFS, or NFS) to communicate both data and management commands to and from storage devices. Support
for encrypted data channels, such as SSH, have not been adopted universally. Issues also exist with the
networking devices that are used to support the storage environment. These devices are subject to attack and
may be managed in unsafe ways.
In addition to using unsecure protocols, many organizations make a common mistake by plugging their
management interfaces on storage devices, such as switches and storage appliances, into the internal
corporate network. Connecting the management interface of a storage appliance into the internal LAN
potentially gives any internal employee, external VPN user, third-party business partner, or external onsite
contractor/ consultant the ability to connect to the device and attempt to log in. In addition, if the management
methods use clear-text technology, such as Telnet or web browsers (HTTP), the exposure is amplified. Table
25-2 lists a set of best practices for storage security.
Table 25-2: Best practices for storage security
Risk Solution
Insecure channels for
management
Use encrypted management protocols such as SSH and SSL. SSH
with port forwarding can be used with many storage applications.
HTTPS (SSL) is available on some storage devices that offer web
management. Also, HTTPS provides the ability to wrap clear-text
web management, such as HTTP, around an SSL tunnel.
Hard coded (a user
name/password that does not
change) or weak username
and passwords
Enforce two-factor authentication for all management to reduce the
likelihood of compromise due to a username and password being
lost.
Shared channels for
management
Do not plug management connections to normal, internal networks.
Segment the management network by isolating it from any other
network in the organization, especially the internal LAN.
Shared accounts When possible, limit authenticated users to perform functions within
the job responsibility (e.g., backup administrators versus storage
administrators). Avoid complete authorization of all management
functions to every authenticated user.
Share applications When possible, use filtering to restrict management of storage
devices to a limited number of management clients. Filtering can
occur at the operating system or application level, limiting the
accessibility of any users loading a management application and
managing the storage network.
Remote Connectivity
Remote connectivity, such as dial-in lines, modems, and call-home capabilities of storage devices, are often
overlooked when considering security issues in storage networks. Many storage network engineers have not
explored the remote side of dial-in lines, modems, or call-home devices to ensure that the security of the
remote network is not undermining the security of the storage network. Remote connectivity can leave a well-
secured storage network vulnerable to an attack.
It is important to know and understand which storage devices can perform what network actions without any
user interaction. The following questions should be asked when using any kind of device that uses remote
connectivity:
Are the storage devices using secure protocols?
Do the remote connections require two-factor authentication?
What kinds of controls are placed on the remote connections (other than username and password)?
Are any IP address limitations or requirements necessary to attempt a valid connection?
Operating System Security
Operating system security is a best practice that is often ignored in storage networks. An operating system
can act as a gateway into the storage network since it is connected to the internal LAN and to the back-end
SAN. However, little concentration of security is placed on these systems-it's similar to having a firewall that
connects the outside, untrusted Internet with the internal corporate network with a rule that allows any IP
address to access any IP address.
In a storage network, operating systems are often the only 'firewalls' that protect access to data. A
compromised operating system can enable an unauthorized user to attack storage devices, such as switches
and storage appliances, directly, and to attempt sniffing techniques. This makes the operating system an easy
attack target. For example, consider a Fibre Channel SAN that has an HBA in an operating system for backup
purposes. In addition to the HBA, the operating system also has a NIC that is plugged into the internal LAN. If
proper operating system security has not been placed on the server that has both a connection to the storage
network and a connection to any other network, the operating system may be responsible for partial or even
complete unauthorized access to the SAN.
Many storage networks consist of unsecured and default installations of several types of operating systems,
from all flavors of UNIX to all versions of Windows. Many environments do not contain a host hardening or a
secure build process for operating systems that exist in storage networks. This gives an attacker an
opportunity to compromise a system in such a way that he can be placed directly inside the storage network,
making further attacks easier. Figure 25-11 is a graphical representation of the importance of operating
system security.
Figure 25-11: Operating system security in storage networks
Storage Appliance Services
Storage devices such as Network Appliance (NetApp) and EMC devices contain a significant amount of
system services that are not all storage related. For example, most storage vendors support storage protocols
such as CIFS and NFS. However, in addition to running CIFS or NFS services, it would not be unusual to see
FTP (File Transfer Protocol), Telnet, Simple Network Management Protocol (SNMP), mount, portmapper,
Domain Name System (DNS), HTTP, NetBIOS, RSH, syslog, and others running on these storage devices.
Similar to an operating system with default installations, storage devices can be unsecure in the default state.
As a best practice, storage devices, such as filers, data movers, and NAS heads, should be deployed with a
minimal amount of services and with unnecessary default modules disabled. This process not only secures the
storage device, but it also allows the core service of the system, such as the storage protocol, to run as the
sole module. A minimalist approach to storage devices leaves less room for error and less possibility of
security exposure. An accidentally enabled SNMP daemon could leave the storage appliance vulnerable to
information disclosure. Even a general SNMP vulnerability, which would affect all SNMP devices, would leave
the storage node vulnerable to direct compromise from this single service enabled. In general, it is best
practice to disable all storage device services except the required storage protocols and management
protocols.

Part VII: Appendixes
Appendix List
Appendix A: NAS Case Study: The International Image Processing Company
Appendix B: SAN Case Study: The Import Auto Industry
Appendix C: SAN/NAS Management Case Study: The Southwestern CD Company
Appendix D: Glossary

Appendix A: NAS Case Study: The International Image
Processing Company
Overview
The International Image Processing Company, or IIP, provides a proprietary digital imaging service to
academic, educational, and government institutions throughout the United States. Its computer imaging
services provide clients with the ability to transfer physical documents, such as historical books, documents,
and newspapers, as well as photographic images, such as historical photos, video images, and medical
images, into digital media. The companys clients range from major universities to some of the largest archival
government institutions.
Despite the scope of its clients, IIP is a small business with less than $5 million in revenues, with plans to grow
threefold in five years if it can increase its capacities to bring on more clients. IIP has a small but distributed
business organization with headquarters in Texas and major client work sites in New York City, Washington,
D.C., and San Francisco. Due to the nature of IIPs work, its IT staff is integrated into the specialized work
force that the company employsgiven the digital scanning software and processes that are integral to their
work. The company also employs two dedicated IT techniciansone for hardware and another responsible for
infrastructure software. Additional IT-related activities are handled through the imaging software support and
configuration personnel.
The company was challenged by its dependencies on storage, created through its software scanning
processes that were increasingly exacerbated by new clients. Daily operations could quickly utilize a terabyte
of data storage. Because the imaging scanning product was archived to clients using both optical and tape
storage devices, if operations ran short on available storage, the imaging processes would slow and
eventually stop until sufficient space was freed up for operations. This slowed billing and eventually cash flow.
The storage situation reached critical levels that started to impact the companys ability to handle new clients
and subsequently impacted its planned business growth.
IIP was in the classic predicament of needing a high-performance solution, while being restricted by a
minimum budget and a limited IT staff. The evaluation of potential storage solutions ranged from storage area
networks (SANs) to IDE RAID solutions. While a network attached storage (NAS) solution would satisfy the
size requirement, NAS remained a problematic solution because it also had to be managed remotely at the
New York, Washington, D.C., and San Francisco sites.
IIP chose a general-purpose solution, even though its processes were largely proprietary, which was a matter
of thoughtful strategy on IIPs part to utilize standard hardware and OS platforms. The NAS solution was
chosen for its ability to integrate easily into a small IP environment, while being able to deploy remotely with
some level of remote management.
This case study will discuss the IIP storage and related processing problems and challenges in detail. These
drove the activities involved in identifying the companys workload and analyzing potential storage solutions.
Finally, we discuss how IIP concluded with the decision to implement NAS and the subsequent transition to
the new storage model.

The Situation Analysis
IIP developed its storage solution from years of experience in scanning and duplicating photographic images,
ranging from military applications to historical documents. The challenges had always been the time required
to scan an image, versus the quality required for large clients. This had given way to the amount of space
required for each image, given that the IIP client base dealt with millions of images.
The type of work IIP performed was not directed toward the market of flatbed scanners nor even larger
scanners in terms of quality and production process requirements. Instead, IIP developed a process and
proprietary software that used specialized digital cameras to scan images of various dimensions, physical
states, and types-various papers, photographic techniques and types, and so on. The process, coupled with
the scanning stations, provided a production-oriented environment where imaging could take place 24/7, if
required. The process and software included a fully complete life cycle of imaging capture, correction, and
quality assurance before it was placed on a CD or tape for shipment to the client.
Clients of IIP had requirements to digitize documents and photographs to provide a wider distribution and
availability of these items through the Internet. Consequently, these clients had become part of the growing
movement within both the academic community and public sector to save historical documents. For the most
part, these markets are just emerging, given the tremendous amount of material that remains to be scanned
and digitized.
IIP's client requirements could run into the 500,000-plus number of images for a single project. That type of
work drove the production-oriented environment introduced by IIP in the late 1990s.
IIP Systems Infrastructure
IIP maintained a distributed infrastructure with its headquarters in Texas and field locations in New York,
Washington, D.C., and San Francisco. Figure A-1 illustrates the configurations at the headquarters' locations.
Here you see the integration of capture stations, process servers, correction/quality assurance workstations,
database servers, and archive servers. Each field location is set up in an identical fashion. Each is linked to
the headquarters' network and web server through a virtual private network (VPN). E-mail and FTP services
are handled in this manner. Architecturally, this setup was designed for future development of remote scan
processing and diagnostic imaging services.
Figure A-1: IIP systems configuration
The imaging process will describe the storage utilization scenarios and why the process is so data-centric.
Figure A-2 depicts the process from start to finish.
Figure A-2: IIP imaging application process
1.
Physical images are entered into a database. The database drives and tracks the processing of images
from capture to output on media. In IIP's case, the database is a relational database that tracks the
location and status of the scanned image so that at any one time the image can be located and
accessed for specific purposes. The initial step in the process is to have the image scanned into the
system. An important note regarding the database: Because of certain restriction and challenges with
unstructured data and relation technology, the scanned images are actually stored as files within the
system. Therefore, the database is actually quite small, as its main job is tracking the locations and
characteristics of the scanned image.
1.
At the capture station, the images are scanned into the system. Capture stations are high-end
workstations that communicate with the specialized digital cameras. A camera scans the image into the
capture station; then the capture software, running on Windows 2000 Professional, writes the scanned
image to a network drive on an available process server over the network and moves to the next
image. Average images range from 300 to 500MB per raw image.
2.
The process server automatically corrects the image for perspective and positioning and updates the
database. The process creates another set of files for the corrected image. This set of files is smaller
because it contains only the changes made to the image. The image is then ready to be evaluated and
checked out.
3.
The image evaluators view each image for correct quality of preset parameters, as discussed with
clients. If the images meet the quality criteria, they are sent to checkout for delivery to media servers.
However, they may need manual correction, which is performed at the evaluation workstations or sent
back for rescanning, as required. This is all tracked by the database.
4.
The delivery point is the media servers where the completed images are batched together and written
to optical media, CDs, or tape, per the requirements of the client. At this point, a final verification is
performed, and if passed, the storage becomes available for more scanned images on the process
servers.
5.
Problem Detail
What IIP had not foreseen were the systems infrastructure requirements for this type of production work. This
required the calculation of both processing cycles and, most of all, the amount of storage space that would be
needed on a daily operational basis. Because IIP is a small business, it had resisted a formal capacity plan
and had relied on its ability to respond quickly when additional capacities were needed. That meant the
additional servers were purchased on an 'as-needed' basis, with most of the hardware being 'do-it-yourself'
built chassis and motherboard configurations.
With this orientation to developing and maintaining the hardware portion of the systems infrastructure, the
storage challenges were met with larger and higher speed internal IDE disks. This gave rise to additional
server installations that were needed to handle the post scan image processing (see Figure A-2, Steps 2
through 4). This then prompted the acquisition of dedicated media servers to write out the client images using
CD or tape media. This is the archival system on the backside of the process (see Figure A-2, Step 5-the
delivery process). This meant that a faster network was necessary to speed the transmission of scanned raw
files to the process servers, and ultimately it placed the problem back at the storage infrastructure once again
as the image scans overtook the capacities on the servers.
A stopgap effort was a move to IDE RAID to provide adequate storage for the process servers. This was
largely driven by the 'do-it-yourself' mode of the hardware and severe limitations of budgets constraints.
Although the IDE RAID facilitated a quick fix, IIP's flexibility in providing reliability and backup protection was
problematic. In many cases, the volatility of the data movement over the period of one week could easily
surpass more than five terabytes running through a single process server. As the tremendous write activities
continued, the IDE drives generally failed twice a month, with minimal success running data protection with
RAID level 1. However, the space it provided offered a brief interlude to the space problems that shut down
the process entirely.
Given that additional business was coming in with more restrictive time constraints for completion, IIP
concluded that a longer term solution had to be found.
The Search and Evaluation
The IIP hardware and software IT personnel researched the solution along with assistance and input from the
imaging software specialist. They found that a SAN was a valid consideration, since it appeared to be the
choice of others working with unstructured data such as video and audio projects. However, they found that
the imaging system, although proprietary by its methodology, used open and commodity levels of hardware
and operating environments and was further open to additional solutions that integrated well into the small
business environment. Another alternative was to move to larger process servers with external SCSI drive
arrays, to scale up in both process and storage power. Yet another alternative was the consideration of a NAS
solution, which would integrate easily with the existing network, would use file systems, and would have the
capacity they needed.
Estimating the Workload
The IT personnel, working with an outside consultant, used the guidelines mentioned in Chapter 17 in
identifying the company workloads. They further moved to understand the types of configuration needed for all
three possible alternatives. First looking at the SAN configuration, followed by the larger server with external
RAID, and finally the potential NAS configuration. The results are summarized in the following sections.
Workload Identification Looking at a year's history of scanning images, the IIP team concluded that the
workload was complex and data-centric, and it fit somewhere between online transaction processing (OLTP)
and data warehousing. The workload encompassed OLTP characteristics when scanning the image and then
transmitting the write transaction to the process server. Although developed as a synchronous process, it was
recently changed to an asynchronous process to facilitate greater throughput at the capture station. However,
this still required a sequential write process at the process server as each image was scanned.
On average, the image scans were 300MB in size. The calculation of 300 images per shift three capture
stations working two shifts provided the necessary throughput. It was determined that at least 540GB of free
space was needed to accommodate the daily scanning process. This required that the overall storage
infrastructure be able to accommodate a 566MB per second throughput rate.
Workload Estimates for SAN Using the guidelines described in Chapter 19, we can quickly calculate that the
required components for a SAN could be handled by one, 16-port switch, given that a single point of failure is
acceptable for the installation; or it could be handled by two, 8-port switches for some level of redundancy.
Three HBA adapters with 2 ports each for redundancy and performance would be required for the process
servers. However, not to be overlooked, this configuration will require additional Fibre Channel storage arrays
to accommodate and be compatible with the new Fibre Channel storage network. Given that the total capacity
of 540GB needs to be available every 24 hours, we can estimate that two storage arrays of 500GB each would
provide the necessary capacity with sufficient free space to handle peak utilization as images are processed
through the system.
Workload Estimates for Direct Attached Aligning the requirements to new servers, we find that all the
process servers would have to be upgraded. This would also require that the storage capacities be carefully
aligned with each process server. Even with this alignment, specific workload affinity would have to be
observed to utilize the storage effectively. On the other hand, the process server could more easily share
storage across the network but would have to reflect some level of duplication for storage requirements to
accommodate the total capacity, essentially doubling the entire storage requirement.
This would require, in addition to new servers installed, OS software upgrades, with appropriate maintenance
and all the necessary activities of a major system installation. It would result in a normal disruption of service
and reliability characterized by new system installations. However, the new servers would have to be
configured to handle the I/O throughput of an aggregate of 566MB per second. This would require each server
to handle 188MB per second if the workload is evenly distributed, which in most cases will not be the case;
however, we will use this for estimating purposes. That relates to a minimum of six Ultra-wide SCSI-3 adapters
necessary to handle the sustained rate of 188MB per second. This requires the total storage to be divided
among the servers, and subsequently the adapters, and places a limitation of approximately 120GB per LUN.
Thus, a more complex management problem in terms of flexibility of reconfiguration based on required storage
would be necessary, given that one capture station could generate 180GB of images every 24 hours.
Workload Estimates for NAS Using the worksheet and guidelines in Chapter 19, we can calculate that our
workload requirements are definitely within the mid-range NAS device configuration and probably just under
the enterprise NAS solutions. Our calculations indicate that the workload requires the following minimum
requirements:
Two network paths
Eleven data paths
An additional path for redundancy (calculated using 10 percent special applications category)
13.2 total logical paths
Comparing total logical paths to data paths = 83 percent
Using the quick estimate NAS Sizing Factor Table in Chapter 19, we select mid-range even though our sizing
factor is within the enterprise range. This is based on the special application circumstances and because the
aggregate data is below a terabyte and would be physically segmented within the aggregate data capacity
estimate. In addition, we considered the workload being further characterized by limited users working with an
almost dedicated Gbe network.
The NAS solutions also offer the flexibility of storage incremental selection-for example, installing two large
NAS servers and one small server, or one large and two medium-sized servers. These solutions also provide
the flexibility of RAID processing, network compatibility, and non-disruption to the existing server
configurations. In addition, these solutions can be easily configurable to support the scanning projects and
mapped as network drives with the same flexibility. They will also provide a closed, yet remotely accessible,
solution for the remote network configurations.
One last word on our estimating process: We recognize the characteristics of the small integrated IT staff and
the company's lack of any formal capacity planning activities. The process of workload identification and
estimates provides this company a level of direction and planning. The result of this exercise has identified
that the mid-range NAS devices can meet the company's workload now and within a limited planning period.
However, it also provides an insight into future challenges IIP will encounter, as its staff has become aware
that it borders on moving into enterprise solutions of either the NAS type or probably a SAN if the budget for
infrastructure can support either.

IIP Storage Solutions
IIP chose to acquire a NAS solution for headquarters and each of its remote work sites. The evaluation proved
that the requirements for capacity, access, and budget were well suited to the NAS solution.
Given the growth and support required for IIP for the next 12 months, the ability for the organization to absorb
a new technology such as Fibre Channel would have been beyond the current budget limitations. In addition,
the learning curve mistakes with the SAN could not be afforded with a small company such as IIP, where there
is no time, space, or budget for test or beta machines. The IIP IT evaluators determined that SANs would
probably be in the companys future as bandwidth catches up to the current network capacity and the need to
provide capture stations with at least 100MB-per-second speeds just to keep up with the increased scanning
capture operations.
IIP Storage Evaluation Results
The requirements and evaluation are further depicted in Table A-1. The evaluation is based on a score of 1
through 10, with 10 being the highest, or best, score. Vendor selection is specific to the customer and the
competitive nature of an acquisition, although it is recommended that a minimum of three bids be requested
once workload estimates have been concluded.
Table A-1: Storage Solution Evaluations for IIP
Criteria SAN Direct
Attached
NAS
Meets I/O workload performance and capacity
specs
9 5 8
Ease of installation and maintenance 3 4 7
Scalability of storage capacity 7 2 7
Compatibility with existing storage
infrastructure
5 6 8
Composite score 24 17 30
System price $150,000
($175,000
w/training)
$235,000 $75,000
NoteWe have estimated pricing only as a general reference point. Included in these pricing estimates are
total components for system operation, additional software required, and storage array costs.
Training for the SAN solution is an add-on, however its a highly recommended item.
IIP Storage Acquisition and Implementation
Along with the storage requirements, IIP provided a list of service requirements it had developed in
conjunction with an outside consultant. These requirements, along with the capacity, access, and
management parameters, drove the decision to go with NAS. Although cost was a factor, the costs of the
proposals were all within 10 percent of each other.
Implementation of the NAS devices provided additional activities. Although minimal, these still provided
additional time before the devices were used in production. The items are summarized in the following list.
Note that any of these are reflective of the operating environment and maintenance status of both operating
systems and the network.
Network Directories Not Recognized on Network Required additional upgrade to network for
compatibility with NAS server network software.
Network Directories Being Dropped Required additional upgrade to Windows 2000 Server software to
reflect NAS devices.
NAS RAID Not Working RAID configured improperly, and reconfigured to reflect recovery parameters.
NAS Integrated Backup Used as archival to tape media, however, requires update to database to reflect
new delivery device.
NAS Slower Performance from Capture Stations NAS set up to use NFS file protocol, overhead to
write much higher than before. Investigating solutions or optimization.
The IIP NAS Solution
The final configuration provides increased storage capacity and enhanced performance. The IIP scanning
process is now being upscaled with new customers and imaging projects. In addition, the space is being
utilized on an automated basis from new project initiation to deletion after delivery. Figure A-3 illustrates the
NAS configurations that support the revised IIP systems infrastructure.
Figure A-3: IIP NAS storage infrastructure
Figure A-3s NAS configuration illustrates the increased capacity for storage of images, but it also provides the
foundation for image access through the Internet. This was designed into the system to allow the remote work
sites that have identical installations to upgrade to NAS devices. Over and above the increased capacity at all
sites will be the potential ability to access images across the IIP storage network infrastructure. In other words,
images scanned in New York could be evaluated in Texas, Washington, D.C., San Francisco, or other remote
offices as they begin operation. This is an added business efficiency that takes advantage of the imaging
expertise across the company without regard to location. It also allows clients eventually to be provided test
and quality images through the Internet to further facilitate the delivery and client acceptance process.
In summary, the NAS solution turned out to be a good fit for IIP. Even though its application appeared to have
many proprietary processes and software processes, NAS operated within the bounds of leveraging
commodity infrastructures for networks, systems, and storage. With the exception of the database server, all
the servers within the IIP infrastructure now share the available storage in the NAS devices. The
acknowledgment of this value further expands the use of the NAS solution as the company began to configure
its web and imaging software code development on the NAS devices.
The NAS solution for IIP provided a cost-effective solution for a small company, but it also provided the
company with the necessary storage resources for expanding its business, which had grown dependent on a
storage-centric product. The companys storage infrastructure is now poised to grow into the future with either
more sophisticated products such as a SAN, or to continue to scale economically with further uses of NAS
devices. Either way, IIP now has the business flexibility to meet the growing dynamics of the imaging
business.

Appendix B: SAN Case Study: The Import Auto
Industry
Overview
The import auto industry provides distribution services for automobiles manufactured in both Europe and Asia.
Most successful international automakers have expanded to provide an extensive presence in North America,
as most have populated manufacturing plants in the United States, Canada, and Mexico.
The activities that surround the distribution of products within the U.S. have provided a rich datacentric set of
activities that has prompted leading-edge data centers in the U.S. Because the automakers' IT organizations'
data centers have had to address data-processing activities on an international level, they have consistently
been ahead of many other U.S.-based companies in their abilities to handle data on an international basis as
well as becoming involved with leading computer technologies. In addition, government regulations
exacerbated the amount of the data these companies must collect, analyze, and address.
Although the international aspect of these companies requires that specific additional data center issues be
addressed, they must also deal with the basic systems of business in terms of reporting, tracking, and
analyzing data. What makes the auto industry unique in many respects is the various sources of revenue that
it derives from its products, much of this transparent to the consumer. Key among these activities is the sale of
parts products, service contracts, and the increasing amount of technical support provided to dealerships.
Many of these are reflective of the increased technological advances within the automobiles themselves. In
addition, warranty products and services must be tracked and accounted for. For example, each time a
warranty claim is presented and processed within a dealership, the manufacturer pays for the service, parts,
and support. Although these activities are not restricted to just the auto industry, it does set the foundation for
a set of required information that means that an extremely large amount of receivables be managed by the
U.S. distribution companies. This necessitates a synchronized transaction between dealer, distributor, and
manufacturer to resolve.
The results of all these operations and interactions make this industry an extremely datacentric and data-
sensitive enterprise. The composite collection of IT organizations within the U.S. that represents these
companies has become the key user and consumer of high-end computer systems and technologies.
However, in many cases these organizations must work in tandem with headquarters data centers in both
Europe and Asia. This requires the U.S. data centers to implement or become compatible with a strategy that
has been developed outside the United States. Working with the levels of autonomy, each of the U.S. data
centers has laid the foundation for the tactical types of technology trends they can justify and implement. This
issue has both advanced and stifled some of the most productive trends within these data centers.
This case study discusses the activities surrounding the analysis, justification, and implementation of a
storage area network (SAN) to support a major restructuring of key application systems within a major
international auto distributor in the United States. The company was advanced in its installation of state-of-the-
art mainframes and distributed systems. These systems supported the U.S. headquarters, parts distribution
centers, and its extensive network of dealerships. However, the major applications supporting these areas
lagged the rest of IT, with largely centralized systems without the necessary distribution of data from the
mainframes to enhance or synchronize the data that existed on the distributed systems. Consequently, a
consistent and timely analysis of the key areas of the enterprise remained problematic.
This evolved as the company executives mandated more data to analyze portions of the business-key among
these are parts, service/warranty, and sales lines of business. They placed hard requirements to the IT
executives to facilitate the development of and restructuring of current data warehouse applications into a
more responsive set of data marts as well as a corporate data warehouse that provides relationships to the
key revenue indicators of the company.
The industry's IT area had been restricted by a static budget for the last two years as the auto business has
tried to stabilize its operations and financial positions. IT employs 350 personnel, with an additional 50 to 60
contractors employed at any one time. Within this organization, the technical support area has 35 personnel,
with enterprise storage administration being supported by one full-time and one part-time systems
programmer. This small contingent works to administrate and support storage operations in conjunction with
the assigned mainframe systems programmers and UNIX and Windows systems administrators.

The Situation Analysis
The current application model consists of all critical application systems being run on the mainframes. These
are generally OLTP-type applications that provide support to all business functions within the headquarters
and distribution centers. These online applications are supported by a heavy batch cycle of processing within
a very restricted online and batch-processing model. The dealership network is a separate entity with
restricted access to limited sets of inventory and sales data. The applications targeted for redesign and
redeployment are some of the few that are operated from the UNIX open systems area. These are the current
data warehouses for sales, warranty, service, and parts. They are currently loaded from the operational
databases on the mainframe during a nightly batch cycle.
The new application requirements are to restructure each data warehouse and distribute its data into a smaller
data mart supporting each of the individual lines of business (LOB). The demands require that the warranty
and service data warehouses be consolidated into a larger integrated data warehouse, along with a smaller
but integrated data mart for each LOB. In addition, the larger, but longer term requirements are to build a
corporate data warehouse to provide key executives with real-time analysis on sales and issues affecting
sales. Although this requirement is beyond the scope of this study, the impact will be affected by the storage
infrastructure that is ultimately selected by the storage administration team.
Figure B-1 depicts the current systems configuration overview and data capacities as provided by the storage
administrator.
Figure B-1: An overview of the current systems configuration
The challenge is to develop an infrastructure that will house the information and provide online access to the
predefined analysis that will be run against the data. A common set of service levels has been agreed to in the
requirements meetings with end users. However, these were approved by the application and database
designers without the concurrence or involvement of the storage admin team, much less the technical support
group. Consequently, the storage teams job is to develop the infrastructure that will support the predefined
capacities, processing requirements, and online service levels. These are summarized in Figure B-2 and Table
B-1.
Figure B-2: Data warehouse/data mart requirements
Table B-1: Processing and Service-Level Requirements for Data Warehouse and Data Marts
Processing Req. Sales DW/DM War/Service
DW/DM
Parts DW/DM
Age of Data << 24 hours
Age of Data >> 24 hours

Batch Loads MonFri: nightly MonFri: nightly MonSat: nightly

Response Time << 3 seconds << 10 seconds << 10 seconds
Availability MonFri: 6 A.M.7
P.M.SatSun: 8
A.M.5 P.M.
MonSat: 24x7Sun:
as available
MonSat: 67
P.M.Sun: as available
Armed with the following information, the storage team had the task of supporting the IT applications area in
developing these business systems. Working with the application designers, the timeline was approximately
six months, in which the last three were given over to development and beta testing of the new data
warehouse configurations.
The Search and Evaluation
The storage team concluded that any solution had to scale to the multi-terabyte level. This was deduced by a
quick but careful analysis of current utilization, historical trending of growth, and current estimates. Current
utilization of the data warehouse applications was around 750 to 800GB of aggregate data, combining both
user and database overhead. This was accommodated through the current installation of 1 terabyte of external
SCSI-based storage capacity installed over three UNIX servers (summarized in Figure B-1).
The other quick conclusion was scalability of access given the large amounts of storage; end user access and
capability to load and update the databases would be key to the feasibility of any solution that would meet the
processing requirements and service levels. The new requirements, as provided by the application design
team, called for an additional capacity of 1.125 terabytes, moving the total requirement for the data warehouse
and data mart applications to well over 2 terabytes. The team had calculated that the data warehouse storage
utilization was growing at 40 percent per year, with no indication that it would decrease.
The storage team had to validate key pieces of information before embarking on its search for solutions. First,
it was asked, was the validity of current estimates it received from the application designers for aggregate user
data, or did it include overhead estimates for database overhead from the DBAs? Second, were these
estimates for day-one installation or have some growth factors been built in? The answers turned out to be
good news and bad news, and some new news, which called for even new challenges.
The good news was that the estimates included the overhead as estimated by the DBAs. Consequently, the
aggregate data sizing could be used as is. The bad news was that these were day-one estimates. In other
words, this was the capacity needed when production started and reflected the actual storage required for the
sourcing of the data from the mainframe databases. The new news was that the consolidation of warranty and
service data warehouses would continue to use two different database products; therefore, the consolidation
would not occur immediately and the two data warehouses would run in tandem until they would be
consolidated later in the year. Consequently, the storage team needed to add an additional factor for two
databases; doubling the current capacity estimate accomplished this.

Available Alternatives
Considering all the available estimates, and validating the capacities, processing requirements, and services
levels, the new data warehouse infrastructure would need to support a 3-terabyte capacity. Further
investigation indicated that the user community-although comparatively small, with less than 250 end users-
would require an I/O workload of multiple gigabytes per transaction. This also carried over into I/O workload
for loading and updating the database, creating an additional multiple-gigabyte workload per transaction. In
summary, calculations indicated a total I/O workload of 29,300GB for a 24-hour period. This was reduced to
1200GB per hour and subsequently an aggregate of 20GB per second.
Using an external direct attached storage strategy that would simply enhance the current configurations would
require additional servers and additional SCSI storage arrays. A quick estimate revealed a requirement of
approximately 50 new servers with the maximum of SCSI adapters for each server to meet the aggregate I/O
workload of 20GB per second. An alternative to this scenario would be to replace existing servers with a much
larger server machine, such as a UNIX mainframe server, where the machine could be partitioned to run the
individual data warehouses with dedicated resources. However, this solution also would require the new
storage arrays, although the existing arrays could possibly be used in conjunction with these.
By utilizing the guidelines for estimating SAN capacity (see Chapter 18), it turns out that a SAN configuration
will require a total of 240 ports, including calculations for recovery and redundancy factors. Given the type of
workload-datacentric (high I/O content), transactional with batch updates, and less than 250 end users-the
configuration could be handled by 15, 16-port switches, or 4, 64-port director class switches. However, this
solution also requires that new storage arrays be purchased for compatibility and performance within the
storage network. This strategy does provide the possibility of utilizing the existing servers.
The storage team, in conjunction with technical support management, chose to evaluate the UNIX mainframe
alternative compared with the SAN configuration. This evaluation took place as a mini-capacity planning
exercise, since this requirement was addressed from an existing capacity plan that was in place.
Consequently, the IT executives, to provide the flexibility to evaluate solutions outside the context of existing
budget plans, invoked the end-user non-aggression pact. (Refer to Chapter 23 regarding storage network
capacity planning and Chapter 22 for end-user non-aggression pacts.) This exercise revealed an interesting
comparison.
Storage Evaluation Results
The UNIX mainframe was a high-end SMP machine that was capable of being partitioned into 2 to 16 logical
processing partitions. The SAN configuration was analyzed as 4, 64-port director class switches, with 12 HBAs
installed across the existing data warehouse UNIX servers-for example, 4 HBAs each. The storage arrays
were configured with 3 terabytes of capacity with specific configurations for each solution-for example, one set
for the SCSI-based UNIX mainframe and a different set of arrays for the Fibre Channel-based SAN.
The requirements and evaluation is further depicted in Table B-2. Our evaluation is based on a score of 1
through 10, with 10 being the highest, or best, score. Vendor selection is specific to the customer and the
competitive nature of an acquisition. It is recommended that a minimum of three bids be requested once
workload estimates have been concluded.
Table B-2: Storage Solution Evaluation Table for Data Warehouse/Data Mart Applications
Requirement SAN Direct Attached
Requirement SAN Direct Attached
Meets I/O workload performance and capacity specs 9 8
Ease of installation and maintenance 3 4
Scalability of storage capacity 7 4
Compatibility with existing storage infrastructure 5 8
Composite score 24 24
System price $700,000 $1,850,000

Meets I/O workload performance and capacity specs 9 8
Ease of installation and maintenance 3 4
Scalability of storage capacity 7 4
Compatibility with existing storage infrastructure 5 8
Composite score 24 24
System price $700,000 $1,850,000

Solution Acquisition and Implementation
The storage admin team recommended the SAN solution with subsequent approval from IT executives and
reluctant concurrence from the application design team. Because this would be the first installation of the
storage networking technology for the data center and because of the visibility of the applications, the design
teams reluctance was understandable. However, an integrated plan was developed to provide beta support
for the application testing of new data warehouse and data mart prototypes.
The decision was based upon both price and performance. Given the increased throughput with the Fibre
Channelbased storage arrays, the SAN solution appeared to be one of the most adaptable solutions for this
type of high-end throughput application that is, an aggregate 20GB per second of data. Specifically, the
ability to source the data warehouses from the mainframes into the UNIX servers for subsequent
preprocessing, loading, and updating was appealing. Moreover, the shadow of the impending corporate data
warehouse project provided the pivotal requirement that pushed the decision toward the SAN. The decision
makers realized that if the UNIX mainframe strategy was chosen, another would be required to handle the
additional load of the corporate data warehouse. Finally, the ability of scaling the SAN or adding another SAN
would be more cost-effective, but it would also be more responsive to the application needs in the short and
long term.
The installation of the SAN provided additional internal requirements and surprises, but none that were
unmanageable or extremely cost intensive. These centered on the new operational characteristics of the SAN,
the additional software tools required, and the responsibility for maintaining the SAN software and hardware
(from an IT organizational view). Each of these issues was handled within the beta period of installation and
subsequent prototype testing of the applications. These issues are summarized in the following section.
SAN Operational Issues
The new SAN required a learning curve and subsequent integration into the existing data center hardware and
software processes. The storage admin team and select members of the systems administration team took
vendor classes, which provided an overview specific to the vendor product selection and additional guidelines
on installation and preinstallation planning.
The installation was somewhat more problematic because of the problems any new infrastructure has in
moving into the data center. The issues were centered on three areas: facilities and location; fundamental
management processes, such as meeting the existing data center operations rules; and integration into
existing server, network, and storage wiring mechanisms.
Facilities and Location This was handled by accommodating more space within the existing server and
storage areas. However, given the less restrictive lengths, the storage was initially planned for installation
in the mainframe storage area on a different floor. This was scratched because local proximity (for
example, server to switch to storage) was better during the initial production cut-over period. However, the
flexibility of subsequently moving the storage to another area, given the increased length capability of
Fibre Channels, turned out to be an additional plus for the data center facilities planners.
Management Processes Perhaps the most troublesome issue was the operational integration into the
data center. Because of the fundamental processes of each SAN director, this required that additional
data center operational policies and processes be developed, followed by training and orientation of key
operational staff members. The ability to provide configuration information and real-time performance
information continues to hamper this aspect of the operation; however, this is becoming more adaptable
as the operations of SAN become familiar to both operations and systems/storage admin staff.
Integration into Data Center One of the most difficult activities surrounding the installation and
subsequent expansion of the SAN is the integration of wiring complexities and integration into the existing
complexities of server and network wiring and switching structures. The ability to implement, manage, and
track this process remains quite difficult as more than 200 ports are routed to appropriate points of
contact. Although not unlike their Ethernet counterparts, and comparatively small next to the LAN/WAN
wiring closets, the critical sensitivity to a lost, unconnected, or inoperative port can affect the highly visible
application such as the companys data warehouses. Existing plans call for the SAN infrastructure to be
separated from the existing LAN/ WAN configurations.
New SAN Software Tools
Once prototype testing was in place, it became apparent that new tools were necessary for managing the
storage across the SAN. The requirements specifically came from the need to manage the storage centrally
across the supported servers. This required some level of both centralization and drill-down capability for
specific array and individual devices access. This was accomplished through the acquisition of new volume
management tools and vendor-specific management tools for the storage arrays, both having specific
functionality for Fibre Channelbased SANs.
The next level of tool that proved to be more problematic was the backup/recovery tool. Due to the nature of
the application, data warehouses generally dont require the stringent restore functions that an OLTP
application would need. Therefore, the ability to perform standard backups would impact the nightly update
function or prove to have little value to an analysis that uses a 24-month rolling summary of data.
Consequently, rebuilding or reloading the data warehouse can be done on a basis that is less time
sensitivefor example, performing a rolling backup once a week and probably on the day when the least
processing occurs. Given the large amount of data, the specific challenge to relational databases, a snapshot
function is being pursued for those database tables that are the most volatile and time consuming to rebuild
and reload. In the case of a complete volume or array outage, the snapshot would allow the DBAs to go back
to a specific time period and reload and update the database within a minimum amount of time.
IT Organizational Integration
The responsibility to evaluate and justify the SAN initially was given to the storage administration team. The
installation and support of the SAN beta testing was accomplished through a team approach, using the
expertise of systems admins, application designers, and DBAs, with leadership from the storage admin team.
As the SAN became a production entity, the team dissolved into its daily responsibilities and the nebulous
nature of some of the SAN components began to affect the total mean time to recovery (MTTR) for any
problem scenario.
Consequently, IT management is faced with an interesting dilemma of providing shared responsibility for the
SAN or developing a new infrastructure group focused on storage. There are pros and cons for either
direction. This study indicates the somewhat conservative nature of the industry to continue to share
responsibility across the SAN components. The network, the systems, and the storage staff are all responsible
for the SAN or various components of the SAN. This unfortunately results in the increase of the MTTR, since
this type of arrangement creates the its not my problem scenario. (Refer to Appendix C for details on how
this manifests itself in a storage networking management business case.)
The SAN Solution
The final configuration provides the increased storage capacity and enhanced performance that was expected
of the solution. The data warehouse/data mart project was accomplished on time and has proved to be
responsive to the I/O workload. Because the estimates were accurate and allowances were built into the
configuration for expansion, the storage capacity and access performance is ahead of the I/O workload
estimates and business requirements, making both of these execute within service levels.
The current configurations are moving into the planning for the next highly visibly project, the corporate data
warehouse. However, this time it has complete concurrence from the application design teams, the DBAs, and
system administrators. Further planning is being considered to consolidate additional storage and servers
within the data centers. The two outstanding caveats remain the organizational fluidity of responsibility and the
continued organizational challenges to effective wiring management. The first continues to elongate any
problem associated with the SAN configuration, and the second continues to facilitate the singular failure of
port mismanagement, exacerbating the first issue.
The configuration shown in Figure B-3 illustrates a summarization of the data warehouse/data mart application
systems. The ability to portray a 200-plus port configuration is beyond the scope of this study. However, its
important to point out the results that the SAN solution provided this company with the ability to manage its
business more effectively through the use of multiple technologies. The functionality of the relational database
to facilitate the analysis and relationships among business functions has provided a 40 percent increase in the
amount of parts sold and a 25 percent cost containment in the area of technical support. More importantly, the
applications have been used to reduce the warranty receivables significantly, and that results in an important
improvement to the company bottom line.
Figure B-3: SAN storage infrastructure
The SAN solution provided the infrastructure for these applications. The ability to move and access data at the
increased levels of Fibre Channel SAN configurations has provided this company with a significant competitive
edge. This case study illustrates the success that storage networking can have with an adequate and
reasonable application and I/O workload planning set of activities.

Appendix C: SAN/NAS Management Case Study: The
Southwestern CD Company
Overview
The most important aspect of business application performance is the ability to access data. Data stored on
disk, tape, optical media, or temporary cache, using the traditional direct attach storage models will be less
susceptible to performance and operational anomalies than data stored through the network model. Although
the location of the data should be irrelevant, if the application cannot get to it or the paths to the data are
congested or critically impacted, the performance of the application will degradeor worse, become
unavailable. The ability of an IT staff to locate, identify the cause, and correct these anomalies becomes more
problematic as storage configurations move into SAN and NAS configurations.
The key to successfully monitoring and managing the performance of any application starts with the consistent
and proactive management of the critical paths to application data. In most cases, this is done with an
assortment of vendor- and user-developed tools that support management of storage configurations. In Part
VI of this book, we discussed many of the accepted management disciplines that become the foundation for a
well-run IT organization, none of which will be more important in this case study than performance
management. Having said that, performance management is tied to the necessary yet mundane activities of
configuration management and the need for an orderly problem management system of processes and
procedures.
In this study you will discover that in the storage networking area, the availability of tools and coordinated
systems will be lacking greatly. As a general industry trend, management tools lack the implementation of
successful technologies by three to five years. This situation is becoming even more significant and
problematic as the integration of infrastructures needs to be taken into account in both small and large
configurations. As the chapters on connectivity options for both SAN and NAS point out, the effects and
counter-effects of storage networks are experienced on an enterprise-wide basis.
Unfortunately, as this case study will demonstrate, todays storage management industry remains in an often
misguided and confused startup mode with point products and integrated storage management architectures
and standards useful only to those companies selling industry controversies. However, storage management
has become an effective catch-up mechanism for large vendors, especially those in the software management
sector, who have more integration challenges than the small startups.
Nevertheless, regardless of vendor size, the current assortments of products are characterized by the
following deficiencies, to name only the major ones:
No correlation of functions among distributed components
No collection of statistically useful data
No proactive performance, configuration, or capacity trending
No identification of root cause to problems
All this makes for a challenge when you are faced with installing, managing, and maintaining a storage area
network and network attached storage solutions, in addition to the challenge the SAN and NAS vendors have
in compensating for these inadequacies. This study will illustrate many of the problems and potential solutions
when the Southwestern CD Company deals with its initial SAN experiences.

The Deferred Responsibility Scenario or Whos Doing What to
Which Data?
The Southwestern CD Company distributes all types of entertainment CDs to both wholesalers and retail
outlets large and small across North America. From its headquarters in Scottsdale, Arizona, the company
ships on average 250,000 CDs a week. By purchasing directly from several manufacturers at optimal prices,
the company has achieved revenues of approximately $51 million per annum. Its profit margin is based upon
moving product effectively to its retailers to leverage the current entertainment fashion for both music and
movies. Because it competes with other, larger distributors, Southwestern must offer better services and a
more competitive price structure.
The CD distribution industry is based upon time to market and can rely heavily on technology to facilitate and
provide a competitive advantage to the ordering process. CD products are purchased from manufacturers with
a return policy that allows the retailers and, subsequently, the distributor to return unsold products. These
market dynamics are reflected in the price, generally at the retail level, although the competitive positioning
based on availability and amounts can be experienced at the distributor level.
An IT department of 15 people supports the company. Ten permanent personnel and five contract
programmers staff the department. They do not have specific storage expertise, but they rely on their systems
administrators to handle the operating system infrastructures that include the storage configurations. Their
configuration is Microsoft-based with the exception of the database servers, which are UNIX machines running
a popular relational database system. They have recently implemented a SAN to consolidate their databases
and provide more responsive order-entry and sales-entry applications.
In conjunction with order-entry and sales-entry applications, the company runs an accounting application, a
sales analysis application, and corporate e-mail, all from within its Metro Data Area (see Chapter 2). The
recent implementation of the SAN has consolidated data from all of these applications within the available
storage resources. All applications are now centrally accessible through the SAN.
The companys configuration, business applications, and client network is depicted in Figure C-1.
Figure C-1: Southwestern CDs Metro Data Area

The Problem
The companys new sales-entry application has experienced periodic slowdowns in its performance. In certain
instances, the database has had to be restarted and reorganized to resume operation. Although the rationale
behind this has yet to be analyzed or understood, restarting appears to restore performance for a timeall of
which takes time away from the sales department, which cannot enter sales data while the system is down.
Todays scenario, however, is further complicated. Its the end of the month and the day begins with a handful
of frustrated phone calls from sales personnel, who either cant log in to the application or are suffering
because the application is running extremely slowly.
Business Situation
At the end of the month, the sales department really needs to hit its numbers so that the orders can be booked
for this quarter. Order quotas come from the Southwestern executives, are part of the quarterly goals, and are
the basis for the financial incentives for the sales management staff. They rely heavily on the system to be up
and performing optimally to enter as many orders as possible. Over and above is the new pricing structure on
the hot new CD products they can sell to their largest retailers, to which they are still trying to sell.
Technical Scenario
The IT department is short staffed, because the database programmer recently quit to take a job at a bank.
The application database administrator (DBA) now has new and additional responsibility for the systems
aspects of the production databases. The periodic slowdowns happened before the SAN installation; however,
they were never completely debugged and root causes were never identified. The problems appear to have
gotten worse as the SAN went into production. All of the user table data has been transferred to the SAN disk
arrays, which was the last activity the recently departed database programmer accomplished.
Problem Dialogue
The following is a summarization of the IT departments events and activities in determining the cause and
solution to the problem.
9:45 a.m.After numerous calls to the Help Desk over the past hour, the IT operations manager fills out a
trouble ticket and assigns it to the DBA.
10:45 a.m.Busy implementing the next release of the database while parsing additional database design
changes, the DBA, having worked through most of the night, arrives to work late. The DBA initiates the
database management program, which provides performance metrics on the database, leading to the
following conclusions: the sales-entry database is up and running, and a number of deferred writes are filling
up the temporary storage. The DBA decides the problem is an operating system anomaly and reassigns the
trouble ticket to the Windows administrator.
11:00 a.m.The Windows administrator, who is also the LAN administrator, is busy implementing a new LAN
segment to facilitate the expansion of the accounting department. Consequently, he must be paged when
there is no response to the trouble ticket within the allotted time.
11:15 a.m.The Windows/LAN administrator logs in to the sales servers and initiates the OS management
software, which provides performance metrics on each of these systems. Two anomalies become apparent:
client transactions are being queued at an increasing rate, and this is putting stress on the system as memory
and temporary space within the database becomes saturated. System-level paging has increased to an
alarming rate. (See Figure C-2.)
Figure C-2: Southwestern CDs Systems Configuration
11:30 a.m.After a brief discussion with the DBA, the Windows/LAN administrator and DBA concur that a
reboot of the sales system and reinitialization of the databases will clear up the problem. By this time, the
entire sales-entry system is down. A message is broadcast via e-mail and voice-mail stating that the sales-
entry system will be rebooted during lunch. The VP of sales estimates $500,000 in sales has not been
entered. He is not pleased.
12:30 p.m.The DBA and the Windows/LAN administrator discover that although the servers are rebooting,
the database is extremely slow coming up. The Windows Performance Monitor tool now indicates that few
I/Os are being serviced from the sales servers. After further discussion, the two hit upon the possibility that
some external application is impacting the reboot of the databases.
12:45 p.m.Inquiries are made and the two discover a large archive job has been running since 8:00 A.M.
Initiated by the Sales Analysis department, the job was to dump two years of sales history from the SQL
Server database running on the ADMIN server. Not surprisingly, the sales history database is stored within the
same storage subsystem as the sales-entry database. The DBA and the Windows/LAN Administrator (whove
ordered in lunch by this point), attempt to contact Sales Analysis management to get permission to stop the
archive job and restart it later that night or over the weekend. The entire Sales Analysis management team is
out to lunch, of course.
2:00 p.m.The DBA finally contacts a Sales Analysis manager and negotiates a stop to the archive job, which
will restart later that nightbut with the mandate that he (the DBA) have the data archived and reorganized for
his department by 9 A.M. the next morning. The DBA is not pleased.
3:00 p.m.The Windows/LAN administrator and DBA terminate the archive job and restart the sales servers
and sales-entry databases, all of which come up running at optimum levels. The second message is broadcast
via e-mail and voice mail: Systems up. Alls well that ends well.
3:30 p.m.Sales personnel are finally back online and able to enter sales data into the system.
Epilogue
The analyses surrounding activities of problem identification, determination, and correction are indicative of
the state of management within storage networking environments. The following epilogue, or post analysis,
can provide some insight into the hidden problems that can develop as storage configurations move into a
networked setting.
Root Cause
In this case, unexpected I/O activity from a single server impacted the capacity of the I/O subsystem and
storage network to the degree that it severely limited the operation of the sales-entry application and access to
their data sharing the same network.
Solutions
Unscheduled archive jobs were terminated and rescheduled, sales servers were rebooted, sales-entry
databases were reinitialized, and recovery programs were run.
Downtime
In effect, the sales-entry application was down from 8:30 A.M. to 3:30 P.M., a total of 7 hours. Due to the
sales-history database not being archived and reorganized, it was unavailable for a longer term, some 12
hours.
Effect on Business
As a result, sales was unable to log all of the days orders, subsequently losing an estimated $1 million in
backlog business. Downtime of the sales-entry database also adversely affected other departments, including
purchasing, whose buyers were unable to meet order deadlines to record manufacturers, which will increase
costs by approximately $2 million for the next month.
Effect on IT
IT management reviews the steps taken to resolve this problem and concludes the following:
More correlation of information across the systems is needed.
Greater attention should be given to the root cause of a problem.
No integrated tools are available to coordinate the actions of the Windows/LAN administrator and the
Oracle DBA.
The Windows administrator and the DBA must work toward proactive trending in estimating storage
resource requirements.
Additional storage restrictions must be put in place to prohibit processing intrusion into the Sales
production environment.
Evaluation
This scenario can be evaluated in three ways:
Business rules and policies
The processing configuration limits and choke points
What available tools and practices are currently available that are relevant and appropriate to this
scenario
Business Rules and Policies
The Southwestern CD Company, like many companies, lacked a complete evaluation and definition of the
business applications it was supporting. Prioritizing sales entry, sales analysis, and other workloads would
have provided metrics for dealing with many of the timing issues. Defining these rules and policies is the
responsibility of the business user and the IT department and should be performed up front, if only on a
macro-level (that is, sales-entry must have highest availability of all applications).
Knowledge of Processing Configurations
Not knowing the processing configuration, especially the storage network configuration, was especially harmful
in this case. More importantly, knowing and understanding the limitations of the storage network in context to
the supported servers masked the root cause of the problem. Finally, and most importantly, was the lack of
association between the storage network and related devices to the business application. Had this association
been made, the problem would have been identified within 15 minutes and resolved within 30.
Available Tools and Practices
Unfortunately, in this case, any available tools are probably of little use in identifying root causes. The
following points provide a quick summary of the value of the major tools currently available:
Storage resource management (SRM) Defensive, reactionary information received after the fact. SRM
tools are resource-focused, operate without business policy, and provide no correlation of information
across distributed servers; in addition, they have no relationship to the business application.
Backup/recovery Adding new and/or bolstering existing backup/recovery software to the storage
network. This is complementary, at best, to resolving the problem. The benefit of faster recovery time is
not part of the root cause of the outage. Although necessary, backup/recovery bears no relation to the
business application.
Although more software tools are becoming available, they continue to provide disparate, incompatible, and
inconsistent levels of information on the storage infrastructure. No single tool provides consistent, proactive
management functions that associate business applications with application data. IT management must
choose from an assortment of tools that provide only discrete levels of empirical information, ranging from
operating system metrics, to database metrics, to I/O and disk metrics. IT users bear the burden of correlating
these seemingly unrelated sets of information in an attempt to understand the effects of resources on
business applications.
The deficiencies within these tools are compounded by the requirements, costs, and expertise needed to
support an increasing set of server platforms, operating systems, and major application subsystems such as
relational database management, messaging, and transactional systems.
The following points illustrate only a few of the major deficiencies challenging todays management tools, as
well as the inefficiencies surrounding business application availability:
No correlation functions among distributed components. Todays business applications are distributed,
which means that events happening on one server can seriously degrade performance throughout the
entire system. The ability to correlate important aspects of performance information as it effects the
business application is currently unavailable.
No proactive trending. Todays IT managers are expected to drive the bus effectively while monitoring
performance through the rearview mirror. Literally all reporting and trending is historical. The information
that ultimately reaches the IT user is about incidents that have already occurred and provides little value
in determining real-time solutions to business applications that perform poorly.
No identification of root cause to problems. The effects of the conditions stated previously means it is
unlikely that the information supplied to the IT user will provide any sort of root cause as to why a
business application is down, not to mention any information intended to identify and correct the problem.
Most activities address only the symptoms, leading to reoccurring results.
As far as practices go, this case study illustrates a competent and responsive IT department with processes in
place for problem notification and assignments. However, the specific expertise and focus of the individuals
contributed to the time it took to pinpoint the problemthat is, the database was up but was not processing
any transactions.
The following points present alternatives to consider when providing a long-term solution:
Reconfiguration/isolation Adding hardware (servers, switches, and so on)also known as throwing
money at it. Although a possible long-term solution, this can essentially convolute the problem and put
storage isolation as part of the IT solution matrix, thus limiting the inherent benefits of storage networking.
Also, reconfiguration/isolation provides no direct association or long-term benefit to the applications.
Increased manpower Hiring more people to monitor and manage the storage infrastructure by business
applicationalso known as throwing money and people at it. This could be a solution, but a highly
improbable one, given the monitoring requirements, expertise required, and associated costs. This
solution does in fact relate to the business application, but only at a prohibitive cost.
Final Analysis
More powerful servers and storage subsystems are quickly evolving into Metro Data Areas characterized by
high data traffic, inevitable congestion, and complex problems relating to data availability. The Internet is also
driving the Metro Data Area theory by increasing exponentially the number of data highways leading in and out
of these already congested areas.
Business applications are distinguished from other applications through their direct association to the
business, their size, and their mission-critical requirements. They are supported within storage systems by
practices and products that manage various aspects of maintaining the storage devices and data. Business
applications can be enhanced with storage networking technology, although a host of problems remains that
will cause difficulties relating to data availability. Currently, no tools directly relate the business application to
the storage network.
Through a composite IT scenario, the realities behind operating with a storage networked environment were
illustrated in this case, as well as the challenges in problem identification, isolation, and solution within these
configurations. Although a SAN configuration was used, a NAS configuration could just as easily be
substituted for the same effect. Evaluation of the scenario indicated that best practices and current tools would
not minimize or stop these types of availability problems.
Steps to Better Storage Networking Management
As demonstrated in this case study, the association of business applications to computing resources remains
for the most part at macro-levels, at best, and not at all within new technologies, such as storage networks.
Storage networks, given their disparate components and critical interrelationships, have evolved into an
infrastructure category within the data center. As such, they have become a tremendous challenge to manage
as a complete entity, much less integrated into the logical set of applications and supported systems. The
following provides additional information on whats required for new tools. This is further augmented by best
practices for managing storage networks.
The Need for Next-Generation Performance Management Tools
The most critical element of performance for a business application is its availability to its own data.
Consequently, a level of performance tools is needed that closely associates the business application with
storage networking resources. These tools should provide a level of functionality to perform the following:
Correlate business application performance information across servers and storage networks
Provide metrics and monitoring for proactive trending of performance and take appropriate action based
upon previously established business rules
Identify root cause areas, if not components, to IT support personnel and make appropriate
recommendations based upon previously established business rules
Such a set of tools would enhance IT managements ability to manage business applications within
increasingly complex application and storage networked configurations. These tools would move IT
professionals closer to the critical elements of business application performance.
Storage networking best practices to observe include the following.
Eliminate unnecessary downtime. Decrease the negative business impact of storage networking and
enhance the inherent abilities of storage networking technologies.
Automate performance management tasks. Decrease the cost to IT management and staff, and maintain
specialization in storage networking as implementation and production use increases.
Learn/associate business to infrastructure. Root cause analysis, proactive processing, and management
can provide solutions before outages occur.
The following guidelines can provide some visibility into dealing with the Southwestern CD company scenario:
Focus maintenance efforts on critical resourcesbefore they go inoperable. Identify the critical resources
through effective configuration management activities and processes. These can help in focusing efforts
on current configuration and processing scenarios.
Proactively manage growth and change within the storage networked configuration. As the storage
network moves into production, be proactive through an effective problem management and ongoing
capacity plan. This will allow the identification of problems and changes to capacity. Further
accommodations can be helpful through an effective change management program that will identify and
focus on processing anomalies and activities.
Gather value oriented performance data. This will probably be the most intrusive and time-consuming
task, given the lack of availability and the programming necessary to gather information from disparate
sources such as MIBS, Common Information Model, and vendor statistical collections. This can also
provide valuable input to vendors and industry during problem-resolution scenarios and activities.

Appendix D: Glossary
A-B
ALU
A component of the computers central processing unit where mathematical calculations are
assembled and executed. CPU-intensive devices and applications rely on the ALU heavily, which
in turn utilizes temporary storage in locations of CPU registers and system-level cache.
application development life cycle
Activities that surround the design, development, and implementation of computer applications.
These are generally characterized by logical phases of work that progress in a contiguous
fashion. Business justification, external and internal design, coding, testing, and deployment
characterize these activities.
archival
The activities of moving data from an online storage device to an offline or removable storage
device for long-term storage. Archiving data generally results from data that has not been
accessed in a specific time period, consequently taking up valuable online disk space. It is
distinguished from backup and recovery by being available for reloading into an online system for
later access.
backup/recovery
Copying data from an online storage device to an offline device for protection against possible
data corruption, storage outage, or site disaster. These functions differ from archival activities and
functions by making a copy of the data and their time requirements in order to provide a specific
data state and other information for recovery purposes.
bandwidth
A set of frequencies that designate the band of electronic wavelengths used to transmit signals
within a computer system and reflect the transmission capacity of the link. This is measured by
the address of the data that can be transmitted and the speed of the signals through various cycle
times measured in MHz (megahertz).
blade computing
A method of loosely connecting discrete computer systems through a common bus. Blade
computing is made up of small, powerful, fully functional, general-purpose computers that are
optimized with only CPU, memory, and I/O bus components. Server blades, as they are known,
are used in conjunction with an I/O backplane that supports connectivity to a common set of I/O
devices, including various storage devices, network devices, and control functions.
block
A unit of data used with storage devices and systems that requires a contiguous block of data
accessed with each I/O instruction of the system. Block sizes are dependent on both operating
system and storage device types, but have direct relationships to the efficiency of the systems
I/O operations.
bus
A system that provides the interconnection between various devices within computer systems.
Generally characterized by a singular and priority-driven operation between two devices within the
busone transmitting instructions and data, the other receiving them. Various bus mechanisms
are used within a computer system such as the PCI bus to interconnect CPU, memory, and I/O
adapters, with storage buses such as SCSI providing interconnections to multiple storage
devices.
business application
A set of computer programs that automates the functions or operations of a business. This is
different from utility or management applications that are computer programs which automate or
enhance the management of the computer systems. Examples include systems such as
backup/recovery and disk defragmentation systems.

C
capacity plan
Activities that identify business elements and translate these into workload requirements. These
include the projection and configuration of systems that effectively process the workload
requirements. Historically, capacity planning has included processor, memory, and storage
capacity as a holistic system. However, storage networking separates the I/O workload from the
computer systems and demands a new set of activities that estimate I/O workload but which are
integrated back into the supported processor complex.
CD-ROM
Storage type that utilizes a polymer material for encoding information which is then subsequently
read using a laser light source. CD-ROM has yet to be fully integrated into FC SAN; however, it
can be used in some configurations with a SCSI bridge/router. CD-ROM drives are available as
NAS devices and can be configured as network storage devices, with certain limitations.
cluster
In terms of computing, it's a collection of interconnected CPU resources that provide shared
efficiencies, such as workload balancing, reliability, availability, and serviceability to the
supporting infrastructure. In terms of storage, it's the smallest addressable unit on a disk drive as
used by Windows operating systems. Cluster sizes are dependent on the total storage capacity of
the disk-therefore, the larger the disk, the larger the cluster. This has a direct relationship to the
efficiency of the system's I/O. Writing a single file to a large cluster uses the entire cluster.
Consequently, writing a 1k file to a 32k cluster utilizes the entire space.
Common Internet File System (CIFS)
A distributed file system used with Windows operating environments that allows files to participate
within networked file systems and network attached storage. CIFS has become an ad hoc
standard in enabling Windows clients to participate and share files within networked
environments.
controller
A device which controls and manages multiple devices that communicate with a computer
system. Storage systems utilize controllers extensively to control and manage multiple storage
devices that communicate within a bus system.
CPU
The central processing unit (CPU) is the mechanism that processes instructions which invoke all
other actions within the computer and ultimately produce the results of the end-user application.
The CPU is dependent on its basic architecture in how it operates and processes instructions.
Two popular architectures are the Reduced Instruction Set (RISC) and Complex Instruction Set
(CISC). Each processes instructions differently; however, both utilize temporary locations, such
as registers and system-level cache, and initiate I/O operations.
cylinder
A measurement of disk geometry, where the association of tracks is measured vertically within the
multiple platters of a disk drive. Cylinder allocations can have significant performance benefits for
applications that access large amounts of information on a read-only basis.

D
data mart
Related to the data warehouse type of application, the data mart is generally designed around a
single subject. Consequently, the scope of the information is reduced, along with its relationships,
within a greater set of information. For example, a data warehouse application that supports
customer financial accounts will have all customers, their accounts, and their related attributes.
However, a data mart providing secondary support may only have information that relates to a
subset of this information such as customers in a particular region, or a set of financial products
and their status. These subset applications are datacentric, and although they utilize a smaller set
of data, they remain challenging given their distributed requirements and use of relational
databases.
data warehouse
A type of application that allows end users to analyze historical information on a set of related
subjects. For example, a data warehouse for a financial application may have information for the
past three years on all customers, their accounts, and any related attributes. Therefore, using
relation data base technology, an end user can query all customers in a particular set of states
that have moved within the last year and have brokerage accounts. These applications are
datacentric and can be the largest users of both storage and processing capacity within a data
center.
Digital Linear Tape (DLT)
A format for magnetic tape that has become an ad hoc standard for open systems. The
architecture is based on the DLZ1, a special compression algorithm where data is written on the
tape in dozens of straight-line (linear) tracks, usually 128 or 208. Capacities are vendor-
dependent but range to >>50GB of data when compression is used. A variant of DLT technology,
Super DLT, provides capacities to move beyond the 100GB on a single cartridge.
Direct Access File System (DAFs)
A file system used specifically with NAS devices, where file access bypasses the normal file
location services through the OS and storage controllers and utilizes a new service called virtual
interface (VI). This allows files to be transferred through direct-memory transfers, which
circumvent the overhead of TCP processing for file access. This, however, requires that a number
of new components be implemented, including a new type of network card which enables the VI
protocols and NAS devices that process direct memory addressing.
director
A class of storage switching devices that are generally defined as having the attributes of >>64
ports, additional hardware functions that enable redundancy within the switch, and related
software that configures recovery and redundancy features.
disk compression
Operations that allow data on a disk to be compressed for efficiency of space utilization.
Compression algorithms usually find all spaces and duplicate characters and remove these from
the storage data source. These are then replaced when the data is accessed.
disk defragmentation
Disks read and write information in a random manner, which is part of their value in accessing
data. However, as write operations occur over time, the disk becomes fragmented as the drive
writes related data blocks out to available blocks, clusters, or sectors, depending on the type of
disk device and operating environment. This fragmentation of the disk, with related data spaced
all over the geometry of the drive, begins to cause performance problems since it takes the disk
head longer to read the segments of the file, or find available space within a fragmented disk.
Defragmentation, sometimes referred to as defragging, moves all the scattered blocks, clusters,
or sectors into a contiguous space on the drive, essentially reorganizing the drive so it can
operate more efficiently.

E-F
embedded database
Many computer applications come with a database integrated with each function. Other
applications are unattended and utilize databases to collect or store data for its operations, such
as an automobile, HVA systems, and aeronautical systems. Each database included in these
applications integrates an embedded database. The embedded database is tightly integrated with
the application programs and cannot be used for other general- purpose applications. These
applications, and moreover their embedded database elements, can be challenging to storage
given they have limited performance and allocation parameters. Therefore, their inherent
limitations have to be compensated for by the processing infrastructure.
F_Port
A FC switch port designated to connect participating nodes, known as N_Ports, to the FC
network. These nodes can be devices such as storage arrays and servers.
fabric
A Fibre Channel switch, or multiple Fibre Channel switches, that are interconnected in a
networked topology to facilitate the physical transmission of FC frames between any two N_Ports.
This includes the software operating within the control of a microkernel that facilitates the
operation of a FC switch. The software functioning within the control of a microkernel operating
system provides the base level services that form the switch fabric network. The fabrics are
vendor-dependent and although compliant with FC standards, they have levels of differences in
how they implement many of the FC operations.
Fibre Channel (FC)
Fibre Channel is a network connectivity standard for a serial I/O bus, with the capability of
transmitting data between two N_Ports at 100 MBps. Principally, the X3T11 committee of ANSI
governs this standard.
file allocation table
Disks have tables of contents so they can access the data stored in a responsive manner. Most
operating environments support some type of file allocation table that resides on the disk and
stores information regarding the files and physical locations of the data on that disk.
file system
File systems are the operating system functions that manage the stored data within the computer.
Most file systems allow users to organize their data into a defined naming and hierarchical
structure. File systems function in networked environments by permitting other computers to
access their data by providing access to segments of the file system. File systems maintain
various levels of data integrity by monitoring and managing the processes that allocate and write
data to the files. This becomes problematic with storage network environments as multiple
systems with their own file systems attempt to manage the files they own-a particularly difficult
problem in a SAN where multiple servers may access the same storage array, device, and
subsequent file. NAS devices have less of a problem with this since they provide remote file
services and offer centralized ownership of files that reside on NAS storage devices.
frame
A component of the Fibre Channel standard that describes the method and specifications of the
basic transmission packet that encapsulates the user data.

G-J
G_Port
A port on a FC switch designed to function as an F_Port or an E_Port that connects both nodes.
This can be external to FC switches through interswitch links (ISLs), and SCSI-to-FC
bridges/routers. This type of universal port is becoming increasingly popular in developing simpler
port utilization structures. These ports are vendor-dependent on implementation.
Head and Disk Assembly (HDA)
The mechanical assembly that houses the read and write heads for a magnetic disk. The
assembly moves mechanically over the spinning platters of the media as it performs its
operations. Additional functionality has been moved into the HAD to facilitate greater read/write
performance.
InfiniBand
An I/O standard that provides a switched fabric network replacing the traditional bus networks of
computers. The InfiniBand standard requires that all I/O be connected through an InfiniBand
switch with the added benefit that connecting computers together makes clustering a more
scalable and efficient process.
iSCSI
A new I/O standard that provides for the execution of a SCSI- based I/O operation over a TCP/IP
network. This allows a server to execute a block I/O operation to a remote storage device. The
new standard provides additional specifications for encapsulating SCSI-based commands into
TCP communications and transferring them through an IP packet to corresponding SCSI-based
storage devices that are IP addressable and have compatible TCP processing capabilities. Initial
implementations have required new NIC cards to support these functions at both the server and
storage device level.
JBOD
Defined as 'just a bunch of disks,' this term defines a storage array in which all devices within the
array are addressable independent units. These provide additional capacity but do not offer any
fault resiliency in the event of an inoperable drive. Partitioning data throughout the disks and
providing a layer of virtualization services is generally part of the I/O management of the operating
system. These functions are also available through third-party software applications.

L-N
Linear Tape Open (LTO)
A format for magnetic tape that competes with digital linear tape. LTO uses a linear multichannel
bidirectional format but adds an enhanced timing-based servo (a device that automates a process
of error correction for a mechanism), hardware data compression, and efficient error correction.
Initially developed by a consortium of IBM, HP, and Seagate, it now has two different formats
one for fast data access and another for greater storage capacity. The Accelis format uses 8mm
tape on a two-reel cartridge that loads at the mid-point to provide fast data access, specifically for
read-intensive applications. The other is the Ultrium format that uses a single reel of half-inch
tape to maximize storage capacity, specifically for write- intensive applications, such as archival
and backup functions. Ultrium-based products offer a >>100GB capacity, while some vendors are
approaching the 1TB capacities.
LUN
A term used with addressing SCSI devices within a SCSI bus system. A logical unit number (LUN)
is defined at the controller level to access, control, and manage the devices within the connected
bus. LUNs provide a level of virtualization when communicating with the operating system, and
subsequently the application, by shielding the application from the complexities of knowing
physical addressing and access information for disk or tape I/O operations.
LUN management
The activities, functions, and capabilities that make up the manipulation, control, and assignments
of LUNs within SCSI bus systems. These generally take the form of configuration parameters for
controller initialization through microcode, but can assume higher levels of functionality, especially
when managing multiple SCSI bus systems.
LUN masking
Part of LUN management where certain servers access specific SCSI devices within the shared
SCSI bus environment, while other servers are blocked from accessing specific LUNs.
micro-kernel
A special type of operating system that can be optimized for a particular application. Both SAN
and NAS devices use micro-kernels to provide base level operating system services to the SAN
switch and NAS server, respectively.
MPP
A term used to describe a massive parallel-processing configuration. These configurations are
made up of multiple processors (often numbering in the hundreds), sharing no system resources
except a high-speed interconnect. They are currently used to address complex scientific,
academic, or business analytical processing such as measuring earthquakes, nuclear explosions,
and very high-end data warehouse applications, by breaking the application tasks into activities
that can be processed in parallel. The processing nodes within MPP systems often have their own
storage arrays; however, in some cases, some nodes have a common I/O configuration.
N_Port
The FC port that is designated for participating nodes, such as servers and storage arrays. These
ports are configured through a Host Bus Adapter (HBA) and used to connect to the FC switch
F_Ports.
NVRAM
A type of shared memory that is used by multiple system processors. Non-volatile random access
memory is a portion of memory that ensures data integrity with the data that is written to the disk
as multiple processor I/O operations take place across the same data.

P-Z
PCI Bus
A bus system to connect peripheral devices, which has become a standard within the industry.
PCI buses provide the interconnections to the internal and external storage devices through
adapter connectivity points, but also provide the interconnections between the CPUs, RAM, and
caching systems.
Relational Database Management System (RDBMS)
A database system that allows for the organization and access of data using relational operations.
RDBMS architectures are developed from tables where entities are defined with each table and
attributes are defined as columns or fields within the table. These databases have become the
preferred method of storing information for both transactional and analytical systems. RDBMS
architectures provide their own file systems and special I/O operations given they are table-driven
and have different data access and storage characteristics than file systems.
SCSI
A standard term for a type of bus storage architecture used with block-level devices such as disks
and tape. The Small Computer Systems Interface (SCSI) defined the standards used with
connecting storage through parallel bus architectures. Forming the largest install base of storage
devices today, the SCSI command structure has superseded its bus application and is used with
both FC protocols and the newly defined iSCSI standard.
service levels
Service levels exist on several levels. There are specific agreements with end users that define
the level of service the IT department provides. Internal service levels are managed to provide a
working measurement of systems and networking performance. Service levels for storage, if they
exist, are often integrated into the internal system service levels and are comprised only of
capacity measurements for end users. The establishment of meaningful measurements for the
storage- networking infrastructure that encompasses storage, systems, and network
specifications have yet to be developed.
SMP
Symmetrical multiprocessing defines a configuration that provides multiple computing nodes
supported by both shared memory and shared I/O configurations. These are generally
characterized by multiple CPUs within the same cabinet. The scalability restrictions have long
centered on the limitations of the shared I/O structure and related performance problems with
storage scalability. SMP configurations are greatly enhanced by the inclusion of storage networks,
especially SANs that provide additional bandwidth and speed to the I/O storage subsystem.
snapshot
A term used for copying online data at a synchronized time and providing a copy of a data
volume, file system, or database for later access. Used in response to user data requiring 24/7
access, snapshot configurations generally require a duplicate volume to be operating so that the
actual copy is made from the duplicate volume. This allows uninterrupted backup operations, and
provides a consistent snapshot of the data at the time it was executed. Once the snapshot
(backup) is taken, the duplicate volume then resynchronizes itself with the primary online volume
and continues processing.
T11.3 standard
The ANSI Fibre Channel standard specifications. All vendors whose products support Fibre
Channel work through this set of specifications. The ANSI Fibre Channel X3T11 Committee
governs these.
tape compression
A setup of functions and operations that allows data on a tape to be compressed for space
efficiency purposes. Compression algorithms usually find all spaces and duplicate characters, and
remove these from the storage data source. These are then replaced when the data is accessed.
temporary storage
These consist of various locations within a computer system where data is stored temporarily until
it can be used by the CPU or written on a storage device. These locations are considered internal
when described as system cache, CPU registers, and memory buffers in RAM. They are external
when considered as part of a peripheral such as a block device.
track
A fundamental measurement of a disk's geometry that spans the circle made by the disk head as
it makes a single rotation of the media platter. Track sizes, therefore, are relative to the size and
density of the media platter. Tracks are formatted for use with an operating environment by way of
the operating system dividing the tracks into clusters or sectors. Tracks aligned vertically with
multiple platters make up another measurement called a cylinder.
zones
Functions within a FC switch that allow the ports to be segmented into individual areas for node
access. Authorized access lists for each port provides login services to the switched fabric
defined zones. Zoning can further be determined regarding the attributes of the ports, node login
addresses, and login attributes. This allows the SAN to define access to authorized devices such
as servers to attached storage devices such as disk and tape arrays.

Index
16-bit addresses, 98
32-bit addresses, 98
A
Access Control Lists (ACLs), 423, 426
adapters
controllers for, 8384
host, 70, 8182, 99
Host Bus Adapters. See HBAs
Host Channel Adapters, 344
NIC, 135
SCSI, 103105
Target Channel Adapters, 344
Address Resolution Protocol (ARP), 440
address space, 98
addresses
16-bit, 98
32-bit, 98
I/O devices, 145
IP, 189191
LUN, 104105
MAC, 178179, 440
APIs (Application Programming Interfaces)
external, 239
fabric, 207208
SANs, 239
Apple servers, 122123
appliance-level NAS networks, 181184, 319320, 326
Application Programming Interfaces. See APIs
applications
access to. See data access
business. See business applications
capacity planning, 277278
data mart, 471
data warehouse. See data warehouse applications
data-centric, 12, 5051
deposit application, 281284
design of, 416, 2526
development cycle, 2528, 490
enterprise-level, 2325, 277
fabric, 207208
HSM-supported, 76
information about, 379
maintenance, 2425, 3132
management, 246249
middleware, 304
performance, 716, 71
personal, 23
RAM vs. online storage, 78
response time, 368
SAN management, 246249
storage planning, 277278
support, 2425, 3132
types of, 2425
usage of, 361
workload attributes, 281284
arbitrated loop topology, 106108
archival storage, 70, 7273, 369, 490
archive tools, 24
ARP (Address Resolution Protocol), 440
ARPANET (Advanced Research Projects Agency Network), 36
arrays
device drivers and, 241
disk, 8586, 210, 241
fault tolerance and, 228
JBOD, 8687, 227229, 496
RAID, 8790, 228229
storage, 209, 241
auditing, 425
authentication
Common Transport, 430
described, 424
key-based, 430431, 443
node-to-node, 428
passwords, 432, 444445
share level, 444
two-factor, 431432
user level, 432, 444
usernames, 432
authorization, 425
auto-loader libraries, 93
availability, 365382
business application data, 3132
capacity planning and, 404407
change management and, 396
data centers and, 368
data recoverability matrix, 369
described, 425
disaster recovery and, 369
fault tolerant systems, 372, 374376
highly available systems, 371373, 375
metrics for, 367371
monitoring, 378379
on-demand, 371375
overview, 366
RAID storage configurations, 376377
response time and, 368
restoration of services, 369
scheduled downtime, 368369
security and, 425
service levels, 367371
uptime, 368369

Index
B
backplane, 79, 99
backup operations
availability and, 380382
described, 490
NAS devices, 151, 174, 372
SAN devices, 256, 266
server-free, 248, 256
software for, 24, 485
tape media and, 266
tools for, 24, 485
backup service, 368
bandwidth
buses, 97, 101
data transfers, 22
described, 97, 490
I/O, 344345
PCI buses, 103
security and, 434435
serial connections, 111
storage networking and, 22
basic input/output (BIOS), 81
batch processing storage systems, 7173
benchmarks, 414417
best practices
security, 445450
storage networking, 488
BIOS (basic input/output), 81
BIOS microcode, 81
black box architecture, 133, 156
blade computing, 346, 490
BLCP (box level capacity planning), 277278
block devices, 146
block I/O operations, 337338, 342343
block size, 286
blocks, 92, 286, 491
boot files, 121
boot processes, 247
box level capacity planning (BLCP), 277278
bridges, 178179, 229231
buffers
hard disk controller, 70
host adapter, 70
I/O buffer pool, 146
RAM, 69, 91, 148
See also cache
bus mastering, 99
buses
address space, 98
arbitration scheme, 100
bandwidth, 97, 101
clock speed, 97
controller cache, 99100
cross-talk interference, 101
data width, 97
described, 491
differential, 101
electromagnetic interference, 9899
electro-mechanical, 9899
evolution of, 99
expansion, 9698
Firewire standard, 110111
I-bus, 97
internal, 97
interrupt processing, 98
I/O, 9798, 109113
operations, 99101
overview, 9699
parallel connections, 100101
PCI, 99, 103, 498
SCSI, 103105
shielding, 101
single-ended, 101
universal, 9697, 343346
USB standard, 110111
vs. networks, 101105
wide, 101
business application data
access to, 31
availability of, 3132
consolidation of, 31
NAS configurations, 31
scalability of access, 30
speed of access, 31
storage strategies, 416
business applications
benefits of storage networks on, 2832
building, 2528
data. See business application data
described, 491
design of, 416, 2526
enterprise applications, 2325
post-implementation analysis, 28
production mode, 26187
requirements document, 25
system implementation, 2528
testing, 2526
workloads and, 281284
business continuity, 353364

Index
C
cache
buses, 99100
controller, 99100
disk, 78
hard disk controller, 70
host adapter, 70
L1/L2, 69
NAS devices, 148
system, 78, 68
See also buffers; memory
capacity management, 354
benchmarks, 414417
box level, 277278
capacity requirements, 404407, 414417
collecting storage information, 410414
described, 276, 491
established service levels, 406407
external capacity driver, 406
implementation of plan, 417419
I/O workload estimates, 405406
modeling tools, 414415
new storage networks, 408409
overview, 277278, 402404
production installation, 418419
quota management, 409
reporting mechanisms, 406
storage analysis, 407414
storage provisioning, 404407
test installations, 418
volume managers, 410411
workloads and, 277278
cascading configurations, 250, 259, 294295, 310311
case studies
import auto industry, 465476
International Image Processing Company, 453464
Southwestern CD Company, 477488
CD media, 267
CD-ROMs, 94, 491
central processing unit (CPU)
CISC vs. RISC, 146148
data storage and, 6669
described, 492
Cfp (complex file processing) mode, 153, 155, 173
change management, 354, 392400
channels, 260
CIFS (Common Internet File System)
data organization and, 122124
described, 443444, 492
evolution of, 42
CIM (common information model), 412413
CISC (Complex Instruction Set Computing), 146148
client computers, 152
client/server environment, 1819, 40
client/server storage
database servers and, 1112
limitations, 1516
overview, 1015
vs. storage networking, 1820
clock speed, 97
clock, system, 145
clustering, 267269
clusters, 120, 491492
common information model (CIM), 412413
Common Internet File System. See CIFS
Common Transport (CT) authentication, 430
compilers, 126
complex file processing (Cfp) mode, 153, 155, 173
Complex Instruction Set Computing (CISC), 146148
compression, 493, 499
computer systems
performance of, 355
processing, 5153
See also specific systems
concentrators, 178
concurrent factor, 302
configurations
cascading, 250, 259, 294295, 310311
core/edge, 256257, 297298, 303
data warehouse workloads, 310311
errors, 429430, 446447
HBA, 225226
iSCSI, 342344
ISP, 1415
JBOD, 227229
management, 248, 354
mesh, 256, 258, 296, 306307
NAS. See NAS configurations
near-line storage, 7476
OLTP, 76, 302303
RAID, 228229, 241, 376377
SAN. See SAN configurations
web-based workloads, 306307
connectionless flow mechanism, 232233
connectivity
buses. See buses
devices, 7980
direct vs. network server connections, 16
Ethernet, 3132, 102, 135
Fibre Channel, 105109
frame, 232233
fundamentals of, 95113
high-speed Internet topologies, 111113
motherboard, 99
NAS. See NAS connectivity
NAS devices, 176188
networks, NAS, 138139, 180184, 186187
networks, SAN, 262264
parallel, 100101
peripheral-to-server, 111113
remote, 448449
SAN. See SAN connectivity
SAN devices, 264267
serial, 100101, 111
console access, 247
console boot process, 247
controller cache, 70, 99100
controllers
adapters, 8384
described, 492
device, 8384
functions, 78, 83
overview, 8384
PCI, 103
SCSI, 104105
tape, 9091
core/edge configurations, 256257, 297298, 303
CPU (central processing unit)
CISC vs. RISC, 146148
described, 492
CPU registers, 68
cross-talk interference, 101
CT (Common Transport) authentication, 430
cylinders, 492

Index
D
DAFS (Direct Access File System), 124, 493
data
access to. See data access
availability of. See availability
business application. See business application data
encryption, 425, 428430
hetero-data, 278
mirroring, 30, 174, 376
non-volatile, 66
organization model. See data organizational model
organizing, 115128, 285286
ownership of, 360
protection of, 7374, 174
recovery of. See recovery
sharing, 241, 244246
size of, 22, 4546, 6061
storage of. See data storage
striping, 376
temporary storage, 499
utilization of, 361
volatile, 6667
data access
challenges, 316, 2223
database systems, 125127
NAS, 4445
SANs, 60
servers, 16
storage networks and, 1819, 2223
See also user access
data blocks, 146
data centers
acquisition strategies, 417418
backup services, 368369, 380382
data availability, 368
data recovery, 369, 379382
division of responsibilities within, 402403
help desks, 399
problem management, 396400
SAN connectivity, 254262
services, 368369, 371
storage capacity, 370
storage problems within, 396400
data language compiler functions, 126
data mart applications, 471
data marts
vs. data warehouses, 308
data organizational method (DOM), 282
data organizational model
departmental workloads, 325
I/O workloads, 285290
OLTP workloads, 300301
specialized workloads, 329330
WIT workloads, 327
data paths
described, 282
OLTP workloads, 302, 310
WIT workloads, 328329
data protection workloads, 74
data replication, 262263, 369
data services, 368369, 371
data storage
application design, 416
architectures, 6576
challenges, 316
client/server. See client/server storage
elements of, 6670
hierarchy, 6870
Metro Data Areas, 21
need for, 4
Network Attached Storage. See NAS
online. See web-based storage
problems, 16
Storage Area Networks. See SANs
web-based. See web-based storage
See also storage systems
data warehouse applications, 1213, 471
data warehouse transactions, 308310
data warehouse workloads
applying SAN to, 308311
configurations, 310311
data organizational model, 309
data paths, 310
SANs, 256
data warehouses
data size and, 6061, 308
described, 308, 493
RDBMS technology and, 130132
SANs, 6061
vs. data marts, 308
data width, 97
database management systems (DBMS), 116
database manager functions, 126
database schema, 127
database servers, 1112
database systems
abstraction, 125126
access, 125127
components of, 126127
considerations, 127128
overview, 125128
performance, 117
databases
business application design, 56
challenges, 127128
concurrent factor, 302
data warehouse, 130
described, 116
embedded. See embedded databases
hierarchical data model, 127
network data model, 127
OLTP, 60
physical design of, 286
relational. See RDBMS
size of, 22
Static Database Model, 173
Very Large Databases, 51
data-centric applications, 12, 5051
data-centric workloads, 256
DBMS (database management systems), 116
departmental architecture, 139140, 319, 324326
deposit application, 281284
DES algorithm, 445
device drivers. See drivers
devices, 7794
block, 146
controllers, 8384
disk systems. See disk systems
drivers. See drivers
Firewire connections, 110111
hardware failures, 357359
host adapters, 8182
magnetic disk storage, 8485
NAS. See NAS devices
peripheral connectivity, 7881
SAN. See SAN devices
SCSI, 103105
sharing, 242244
USB connections, 110111
differential bus, 101
Digital Linear Tape (DLT) format, 92, 493
Direct Access File System (DAFS), 124, 493
Direct Memory Access (DMA), 69, 124
director devices, 493
directories, 165167
director-level enterprise, 216
disaster recovery, 363364, 369
See also recovery
disk arrays, 8586, 210
disk cache, 78
disk drives, 15, 8485
disk systems, 8590
disks
capacity, 85
compression, 493
described, 70
fragmentation, 494
magnetic disk storage, 8485
partitions, 286
physical vs. logical locations, 8283
technical specifications, 85
virtual, 85
Distributed Desktop Management Task Force (DMTF) standard, 413
DLT (Digital Linear Tape) format, 92, 493
DMA (Direct Memory Access), 69, 124
DMTF (Distributed Desktop Management Task Force)standard, 413
DOM (data organizational method), 282
DP (data paths), 282
drivers
device I/O, 146
HBA, 208209, 240241, 441
host adapters, 81
RAID and, 241
SANs, 239241
software, 239241
switch OS, 239241
drives
disk, 15, 8485
tape, 9092
DVD media, 94, 267
DW (data warehouse) transactions, 308310

Index
E
EISA (Extended Industry Standard Architecture), 99
electromagnetic interference, 9899
electro-mechanical buses, 9899
embedded databases
described, 494
NAS solutions and, 318
Encapsulating Security Payload (ESP), 428
encryption, 425, 428430, 434
end users, 361
enterprise business applications, 2325
enterprise connectivity (ESCON), 261
enterprise NAS architecture, 141142
enterprise-class file systems, 117118, 122
enterprise-level applications, 277
enterprise-level NAS networks, 183184, 319321, 327329
enterprise-level workloads
NAS, 327329
SAN, 312
E_Port (Expansion Port), 221, 223224, 443
ERC (error recovery code), 91, 101
errors, configuration, 429430, 446447
ESCON (enterprise connectivity), 261
ESP (Encapsulating Security Payload), 428
Ethernet connections, 3132, 102, 135
events, external, 98
exchanges, 231
expansion buses, 9698
Expansion Port (E_Port), 221, 223224, 443
Extended Industry Standard Architecture (EISA), 99

Index
F
fabric APIs, 207208
fabric applications, 207208
fabric architecture
described, 494
Fibre Channel standard, 108109
I/O operations, 344345
SANs, 59, 231233
Fabric operating system, 378
Fabric OS micro-kernel, 205207
Fabric Port (F_Port), 108, 221223, 494
fabric switch, 108
fault tolerance
availability and, 372, 374376
enterprise-class file systems, 118
NAS and, 375376
Quality File Processing, 152153
storage arrays, 228
fault-tolerant systems, 372, 374376
FC hub, 204205
FC layers, 202204, 437438
FC networks, 204, 337, 439441
FC (Fibre Channel) Protocol, 220224
arbitrated loop topology, 106108
architecture, 106
classes, 232233
connections, 105109
fabric operations, 381382
fabric topology, 108109
fabric transmission, 55
frames, 107, 204, 231233, 437441
latency issues, 105
operations, 107110
overview, 5456, 105107
point-to-point topology, 106107
SAN security, 428, 435443
FC switch device, 202203
FC switches, 207210, 220224, 434436
FC-GS-3 (Fibre Channel-Generic Services-3, 430432
FCSec (Fibre Channel Security), 428
fiber optic connections, 260261
Fibre Channel. See FC
Fibre Connectivity (FICON), 261
FICON (Fibre Connectivity), 261
field replaceable unit (FRU), 357358
FIFO files, 165
file allocation table, 495
file manager functions, 126
file servers
Microsoft, 40, 42
NAS and, 139140
UNIX, 38
file systems, 119124
allocation, 119120
CIFS. See CIFS
components of, 119121
DAFS, 124, 493
described, 38, 116, 119, 495
enterprise-class, 117118, 122
file access, 120
file allocation, 120
illustrated, 39
management, 120121
mounting, 165167
NAS. See NAS file systems
NFS. See NFS
NTFS, 121
overview, 119121
performance, 117
POSIX-compliant, 121122
remote, 158
shared, 39
types of, 121124
UNIX, 3839, 121124, 165167
volume definition, 120
Windows, 121124, 159
files
boot, 121
FIFO, 165
I/O operations, 337338, 342
log, 121
virtual, 39
Firewire standard, 110111
firmware, 79, 247
formats
DLT, 92, 493
LTO, 92, 496
tape drives, 92
formatting process, 85, 120
F_Port (Fabric Port), 108, 221223, 494
fragmentation, 494
frames
described, 55, 231, 495
FC, 107, 204, 231233, 437441
flow control of, 232233
sequences, 438
weaknesses, 437441
FRU (field replaceable unit), 357358

Index
G
generic port (G_Port), 221, 223224, 495
Gigabit Ethernet (Gbe), 102
glossary, 489500
G_Port (generic port), 221, 223224, 495

Index
H
hard disk controller cache/buffer, 70
hard disks. See disks
hard zoning, 434, 436437
hardware devices. See devices
hardware failures, 357359
hardware interrupt handler, 145
HBA drivers, 208209, 240241, 441
HBA information, 378
HBA-based LUN masking, 441442
HBAs (Host Bus Adapters)
basic functions, 225
overview, 224226
reliability of, 226
SAN/server connections, 212
HCA (Host Channel Adapter), 344
HDA (Head and Disk Assembly), 85, 495
help desks, 399
hetero-data, 278
Hierarchical Storage Management. See HSM
historical reporting, 400
historical storage activity information, 410414
host adapters, 70, 8182, 99
Host Channel Adapter (HCA), 344
HSM (Hierarchical Storage Management) storage systems, 7476
HSM workloads, 7576
HSM-supported applications, 76
HTTP (Hyper Text Transport Protocol), 122124, 327
HUB architectures, 5758
hubs, 5758, 178
Hyper Text Transport Protocol (HTTP), 122124, 327
HyperTransport standard, 347349

Index
I
I-bus (internal bus), 97
IDS (intrusion detection system), 422
IEEE1394 standard, 110111
image processing case study, 453464
import auto industry case study, 465476
in-band management functions, 246247
Industry Standard Architecture (ISA), 99
InfiniBand standard, 268, 343346, 496
infrastructure, 361
Initial Sequences Numbers (ISNs), 438439
inodes, 120, 166
instances, 127
integrity, 425
internal bus (I-bus), 97
International Image Processing Company case study, 453464
Internet
Common Internet File System. See CIFS
data stored on. See web-based storage
servers on, 14, 21
WIT workloads, 304307, 327328
Internet NAS architecture, 139141
Internet Protocol. See IP
Internet SCSI. See iSCSI
Internet service providers (ISPs),1415, 185
interrupt processing, 98
interswitch linking (ISL) functions, 249251
intrusion detection system (IDS), 422
I/O bandwidth, 344345
I/O buffer pool, 146
I/O bus standard, 18
I/O buses, 9798, 109113
I/O channels, 9698
I/O content, 156, 285290
I/O device addressing, 145
I/O fabric, 344345
I/O manager
described, 144145
features, 145146
NAS hardware as, 144152
NAS software as, 171174
I/O operations
block, 337338, 342343
distributing, 5356
NAS, 134137, 144152, 156, 339
Rapid I/O, 346347
SANs, 5356, 337338, 340
shared I/O, 112113
WIT I/O, 327
I/O operations performed per second (IOPS), 156
I/O priority interface, 146
I/O subsystem, 145146
I/O systems, 78
attributes, 281284
business applications, 281284
characterization of, 278280
data paths, 289290
defined, 278
integration with web services, 341
NAS, 173174
storage networking and, 290292
user access, 287289
workload solutions matrix, 292
See also workloads
IOPS (I/O operations performed per second), 156
IP (Internet Protocol), 80
IP access points, 262
IP addresses, 189191
IP communications, 102
IP packets, 102
IP ports, 222
ISA (Industry Standard Architecture), 99
iSCSI configurations, 342344
iSCSI devices, 264
iSCSI (Internet SCSI) standard, 276, 496
ISL functions, 242, 249252
ISL (interswitch linking) functions, 242, 249251
ISNs (Initial Sequences Numbers), 438439
ISPs (Internet service providers), 1415, 185
IT organizations
effects of storage networks on, 3334
storage networking needs, 2123
IT utility programs, 56

Index
J
JBOD arrays, 8687, 227229, 496
JBOD configurations, 227229
JES (Job Entry Subsystems), 52
Just a Bunch of Disks. See JBOD

Index
K
Kerberos algorithm, 445
kernels
Fabric OS micro-kernel, 205-207
micro-kernel definition, 497
NAS. See NAS micro-kernel
SANs, 205-207
UNIX, 161-162
key server technology, 430-432
key-based authentication, 443

Index
L
LAN storage, 342344
LAN topologies, 177188
latency issues, 54, 204
Linear Tape Open (LTO) format, 92, 496
local area network. See LAN
log files, 121
logical unit number. See LUN
LTO (Linear Tape Open) format, 92, 496
LUN (logical unit number), 104105, 441442, 497
LUN addresses, 104105
LUN management, 497
LUN masking
described, 497
SANs, 242245
security and, 433, 441442

Index
M
MAC addresses, 178-179, 440
Mac clients, 152
Mac OS, 122-123
magnetic disk storage, 84-85
magnetic disks, 68, 78, 94
magnetic tape, 78, 94
magnetic tape drives, 90-92
magnetic tape storage, 90-94
formats, 92
optical storage, 93-94
read/write mechanisms, 92
tape libraries, 93
tape systems, 92-93
mainframes, 210-211, 277
maintenance
applications, 24-25, 31-32
NAS workloads, 333
SAN software, 247
SAN workloads, 313-314
workloads, 286
management information bases (MIBs), 238, 412, 414
Man-in-the-Middle (MITM) attacks, 440-441
Massive Parallel Processing Systems (MPP)
described, 497
I/O processing and, 53-54
MIMD and, 106
overview, 52-53
shared nothing systems, 55-56, 111-112
master file tables (MFT), 120
MCA (Micro Channel Architecture), 99
memory
I/O bottlenecks and, 148
NAS and, 137, 148
NAS devices, 148
NV-RAM, 164, 497
shared, 112-113
See also cache; RAM
mesh configurations, 256, 258, 296, 306-307
messaging workloads, 256, 258
metadata, 120-121, 126
Metro Data Areas, 21
MFT (master file tables), 120
MIBs (management information bases), 238, 412, 414
Micro Channel Architecture (MCA), 99
microcode, 79, 81, 83
micro-kernels
defined, 497
Fabric OS micro-kernel, 205-207
NAS. See NAS micro-kernel OS
SANs, 205-207
Microsoft file servers, 40, 42
middleware, 304
MIMD (multi-instruction multi-data), 54, 105-106
mirroring, 30, 174, 376
MITM (Man-in-the-Middle) attacks, 440-441
monitoring functions, SAN, 247
motherboard connections, 99
MPP (Massive Parallel Processing Systems)
described, 497
I/O processing and, 53-54
MIMD and, 106
overview, 52-53
shared nothing systems, 55-56, 111-112
multi-instruction multi-data (MIMD), 54, 105-106
Multiple Virtual Systems. See MVS
multiswitch SANs, 256-258, 270
MVS (Multiple Virtual Systems), 52
MVS operating system, 260
MVS-JES2/MVS_JES3 systems, 52

Index
N
name server, 221223
NAS (Network Attached Storage), 129195
advantages, 1920, 317318
architecture, 131142
caveats, 47
complex file processing (Cfp), 153, 155
concepts, 3547
configurations. See NAS configurations
connectivity. See NAS connectivity
cost-effectiveness, 317318
data access, 4445
data protection and, 174
data sharing, 244
departmental architecture, 139140, 319, 324326
described, 132
design considerations, 320322, 363
devices. See NAS devices
enterprise architecture, 141142
evolution of, 3843
fault tolerant environments and, 375376
file I/O, 337338
file systems. See NAS file systems
hardware. See NAS devices
illustrated, 20, 30, 44, 46
integrating with SANs, 335350
Interent architecture, 139141
introduction to, 3638
I/O operations, 134137, 144152, 156, 339
memory, 137, 148
networks. See NAS networks
operating system. See NAS micro-kernel OS
operational overview, 4446
Plug and Play technology, 132
problems. See problem management
processing models, 152155, 172173
quality file processing (Qfp), 152154
RAID and, 168171
recovery operations, 372
security. See security
server component, 133134
similarities with SANs, 338341
simple file processing (Sfp), 152153
software. See NAS software
solutions. See NAS solutions
storage component, 134135
UNIX and, 159161
vs. SANs, 50, 201202, 336338
web-based storage, 139141
workloads. See NAS workloads
NAS configurations
department, 139140, 319, 324326
enterprise-level, 141142, 149151, 319321
entry-level, 149
Internet, 139141, 319, 327329
management, 386, 390391, 394395
mid-range, 149151
networks, 194195
specialized, 320, 322, 329330
types of, 139142
WAN topologies, 184188
workloads, 316330
NAS connectivity, 175195
hardware, 176188
NAS appliances, 192193, 195
NAS Enterprise, 192195
NAS Mid-Range, 192193, 195
NAS Web, 192, 194195
networks, 138139, 180184, 186187
operational aspects, 193194
overview, 176
software, 189195
TCP/IP processing, 138139, 189193
NAS devices, 143156
advantages, 176
application support and, 316
architecture, 133135, 151
backups, 151, 174, 372
configurations. See NAS configurations
connectivity. See NAS connectivity
data size, 4546
described, 144
early versions of, 39, 41
enterprise-level, 149151
entry-level, 149
I/O manager, 144152
I/O operations, 134137
I/O processing, 144152
major functions of, 172
marketing data, 184
memory, 148
mid-range, 149151
NAS sizing factor table, 136138
NAS workloads and, 154156
processing models, 152155, 172173
processor architecture, 146148
processor component, 146148
sales data, 184
serviceability, 398399
storage system element, 149152
UNIX environments, 43
upgrades, 394
Windows environments, 43
workloads for, 173174
NAS file systems, 158171
mounting file directory, 165167
operations, 165168
overview, 158161
types of, 137138
NAS micro-kernel OS, 161165
components, 161165
example of, 163164
overview, 136137, 161164
UNIX and, 161164
upgrade to, 395
Windows and, 164165
NAS networks
appliance level, 181184, 319320, 326
architecture, 135
connectivity, 138139, 180184, 186187
deployment of, 180184, 186187
enterprise level, 183184, 319321, 327329
TCP/IP processing, 138139, 189193
workloads, 194195
NAS ports, 136137
NAS RTOS, 378
NAS servers, 44, 133134
NAS software, 157174
components, 157174
connectivity, 189195
described, 158
as I/O manager, 171174
NAS solutions, 315333
advantages, 317318
conclusions, 330333
data protection, 174
described, 316
disadvantages, 318319
embedded databases and, 318
enterprise web workloads, 327329
estimating NAS requirements, 136138
marketing data, 184
OLTP and, 318
preparing for, 332333
sales data, 184
specialized workloads, 320, 322, 329330
workload characterization, 316324
NAS workloads, 154156
characterizing, 316324
departmental, 324326
described, 154
enterprise web, 327329
maintenance, 333
NAS devices, 173174
NAS hardware and, 154156
NAS-WIT workloads, 327328
port requirements, 136137
production installations, 332333
specialized, 320, 322, 329330
vs. SAN workloads, 292
See also workloads
NAS/SAN solutions, 335350
NAS-WIT workloads, 327328
near-line storage configurations, 7476
network file system. See NFS
Network Interface Card. See NIC networks
connectionless services, 108
connectivity, 7980
described, 101
FC, 204, 337, 439441
LANs, 177188, 342344
NAS. See NAS networks
SAN. See SAN networks
switched fabric, 55
TCP/IP, 102
vs. buses, 101105
WANs, 184188, 342344
See also storage networks
NFS (network file system)
emergence of, 39, 160
illustrated, 40
internal operations, 167168
NAS and, 160161
overview, 165168
UNIX and, 165167
NIC (Network Interface Card), 7880, 139, 176177, 342
NIC adapters, 135
NIC connections, 135
NL_Port, 221, 227
nodes, 108, 221
non-aggression pact, 369370
non-volatile random access memory (NV-RAM), 164, 497
Novell Netware, 159
N_Port, 221, 497
NTFS (NT File System), 121
NV-RAM (non-volatile random access memory), 164, 497

Index
O
OEMs (Original Equipment Manufacturers), 417418
OLTP (Online Transaction Processing)
configurations, 76, 302303
data-centric nature of, 5051
NAS and, 318
OLTP databases, 60
OLTP storage systems, 71
OLTP workloads
attributes, 282283
data paths, 302, 310
data warehouses, 308311
described, 299
HSM applications and, 76
SANs, 256257
user access, 301
See also workloads
online storage systems, 7172, 78
See also web-based storage
Online Transaction Processing. See OLTP
Open Systems Interconnection. See OSI
operating systems (OS)
Mac OS, 122123
micro-kernel, 136137, 161165
MVS, 260
NAS. See NAS micro-kernel OS
Real-Time Operating System, 43, 162
SAN, 236251
security, 449450
server OS, 208209
storage performance and, 379
switch, 236251
UNIX. See UNIX environment
Windows. See Windows environment
optical libraries, 94
optical storage systems
overview, 9394
SANs and, 210211, 267
OS. See operating systems
OSI (Open Systems Interconnection), 42
OSI model, 177180, 189190
outages, 357359, 361, 363
out-of-band management functions, 247248
out-of-band processing, 245246

Index
P
packets, FC, 107
packets, IP, 102
parallel connections, 100-101
partitioning, 119, 286
passwords
authentication and, 432, 444-445
guidelines, 446
per-user, 444-445
security and, 446
paths, data
described, 282
I/O workloads, 289-290
OLTP workloads, 302, 310
WIT workloads, 328-329
PCI buses, 99, 103, 498
PCI controllers, 103
PCI (Peripheral Component Internet) standard, 99, 103, 186
performance
applications, 7-16, 71
capacity planning and, 414-417
computer systems, 355
data storage systems, 117
database systems, 117
file systems, 117
HBAs and, 240-241
management, 248, 354, 366
non-linear, 7-16
operating systems, 379
tools for, 487-488
Peripheral Component Internet. See PCI
peripheral devices, 78-81, 111-113
See also devices
physical addressing, 103
Plug and Play technology, 132, 156
point-to-point topology, 106-107
port zoning, 242-243
ports
Expansion Port, 221, 223-224, 443
Fabric Port, 108, 221-223, 494
generic, 221, 223-224, 495
G_Port, 221, 223-224, 495
IP, 222
NAS, 136-137
NL_Port, 221, 227
N_Port, 221, 497
replication of, 443
SAN workloads, 297, 299
switch, 297, 299
POSIX standard, 122, 161
preprocessor compiler functions, 126
problem management
configuration errors, 429-430, 446-447
described, 354, 400
diagnostic services, 398
disciplines, 396
documentation, 399-400
interruptions in service, 357-359
NAS environments, 398-399
problem identification, 393, 397-400
problem prioritization, 397, 400
problem resolution, 396, 400
reporting problems, 399-400
root cause analysis, 378
SAN environments, 398
service information, 398
serviceability and, 393
storage change management and, 396-400
storage problems, 396-400
tool access, 398
tracking problems, 399-400
processors
CISC vs. RISC, 146-148
data storage and, 66-69
described, 492
production installations
NAS workloads, 332-333
SAN workloads, 313
production workloads, 215
programs. See applications

Index
Q
Qfp (quality file processing) mode, 152154, 172
query processor functions, 126

Index
R
RAID (Redundant Arrays of Independent Disks), 18, 8587, 134, 210
RAID arrays, 8790, 228229
RAID configurations, 228229, 241, 376377
RAID controller, 228
RAID functionality, 30
RAID levels, 8790, 168171, 376377
RAID storage systems, 168171
RAID/SCSI operations, 213
RAM, 91, 148
See also memory
RAM buffers, 69, 91, 148
RAM storage, 78
Rapid I/O, 346347
raw I/O, 146
raw partitions, 119
RDBMS (Relational Database Management System)
challenges, 127128
data warehouses and, 130132
I/O workloads and, 285286
NAS and, 318319
OLTP workloads and, 300301
RDBMS model, 11
SANs and, 262263
read/write routines, 146
Real-Time Operating System (RTOS), 43, 162
recovery
activities for, 490
availability and, 369
data, 379382
data centers and, 369, 379382
disaster, 363364, 369
NAS, 372
SAN, 256, 266
scenarios for, 357, 359
software/tools for, 485
switch OS, 238
recovery service, 369
Reduced Instruction Set Computing (RISC), 146148
redundancy, 361362, 406
Redundant Arrays of Independent Disks. See RAID
Relational Database Management System. See RDBMS
Remote Procedure Calls (RPCs), 40, 159
repeaters, 178
replication services, 262263, 369
requirements document, 25
resource management, 409
Resource Management Facility (RMF), 411412
resource usage, 24
RISC (Reduced Instruction Set Computing), 146148
RMF (Resource Management Facility), 411412
router ACLs, 423
routers
NAS, 179180
SAN, 229231
SCSI, 230231
RPCs (Remote Procedure Calls), 40, 159
RSA algorithm, 445
RTOS (Real-Time Operating System), 43, 162

Index
S
Samba program, 137
SAN configurations
cascading, 250, 259, 294295, 310311
core/edge, 256257, 297298, 303
director-level, 216217
enterprise, 216217
entry-level, 213215
guidelines, 270271
JBOD, 227229
levels of, 213217
management, 385388, 394395
meshed, 256, 258, 296, 306307
mid-range, 215
multiswitch, 256261, 270
RAID, 228229
SAN-to-SAN communications, 264265
serviceability and, 386387, 394395
utility for, 247
workloads, 312313
SAN connectivity, 253271
challenges, 254255
data center connections, 254262
future of, 267270
guidelines, 270271
multiswitch configurations, 256261, 270
networks, 262264
optical devices, 267
overview, 254
SCSI/RAID operations, 213
server connections, 212
server support, 258261
tape media, 266
SAN devices, 219234
backups, 256, 266
bridges, 229231
communications between, 3132
connections, 264267
considerations, 234
described, 220
fabric operation, 231233
Fibre Channel switch, 220224
overview, 204205
recovery, 256, 266
routers, 229231
sharing, 241244
storage component, 227231
upgrades, 394
SAN networks
connectivity, 262264
external, 262
TCP/IP and, 209
SAN ports, 297, 299
SAN software, 235252
components, 235252
configuration management, 248
configuration utility, 247
considerations, 252
console access, 247
console boot process, 247
data sharing, 244246
device drivers, 239241
device sharing, 242244
in-band management functions, 246247
maintenance, 247
management applications, 246249
monitoring functions, 247
out-of-band management functions, 247
performance management, 248
storage management, 248
tools, 473474
workstation access, 247
SAN solutions
data replication, 262263
enterprise-level workloads, 312
optical media solutions, 267
tape media solutions, 266
SAN switches, 58
SAN workloads, 293314
enterprise-level, 312
maintenance, 313314
planning for, 312313
port requirements, 297
production installations, 313
production practices, 313314
vs. NAS workloads, 292
WIT workloads, 304307
See also workloads
SAN/NAS solutions, 335350
SANs (Storage Area Networks), 197271
advantages, 1920
architecture, 199217, 294
caveats, 6162
complexity of, 6162
components, 201213
concepts, 4962
configurations. See SAN configurations
connectivity. See SAN connectivity
cost issues, 61
data access, 60
data sharing, 244246
data size, 6061
data warehouse environment, 6061
data warehouse workloads and, 308311
data-centric applications, 5051
design considerations, 362363
device drivers, 239241
device sharing, 242244
devices. See SAN devices
distributing computer processing, 5153
distributing I/O processing, 5356
evolution of, 5758, 200201
Fabric OS, 206207
Fibre Channel Protocol, 5356
hardware. See SAN devices
illustrated, 20, 30, 57
integrating with NAS, 335350
interswitch linking (ISL) functions, 249251
introduction, 5657
I/O operations, 5356, 337338, 340
I/O workloads. See I/O workloads
management, 62, 242, 246249
Massive Parallel Processing Systems (MPP), 5253
networks. See SAN networks
OLTP workloads and, 299303
operational overview, 5861
production workloads, 215
SAN management applications, 246249
security. See security
shared nothing architecture, 5556
similarities with NAS, 338341
SMP and, 5152
software. See SAN software
supporting components, 241251
switch operating system, 236251
switched fabric networks, 55
switches, 211
TCP/IP and, 209
vs. NAS, 50, 201202, 336338
web-based workloads and, 304308
workloads. See SAN workloads
scalability, 30
SCSI adapters, 103105
SCSI buses, 103105
SCSI commands, 213, 239240
SCSI controller, 104105
SCSI devices, 103105
SCSI IDs, 103104
SCSI routers, 230231
SCSI (Small Computer System Interface) standard, 18, 59, 103, 498
SCSI tape drives, 90
SCSI/RAID operations, 213
sectors, 85, 90
security, 421450
authentication. See authentication
best practices, 445450
challenges, 432435
CIFS, 443444
configuration practices and, 446447
default modules and, 450
defense-in-depth model, 426427
encryption, 425, 428430, 434
E_port replication, 443
Fibre Channel, 435443
firewalls, 433
frame weaknesses, 437441
layered security model, 426
LUN masking, 433, 441442
management practices and, 447448
Man-in-the-Middle attacks, 440441
methods for, 426427
NAS, 443445
NFS, 444445
operating system, 449450
overview, 424426
passwords, 446
products, 433434
remote connectivity and, 448449
SANs, 435443
solutions, 433434
standards/specifications, 434
storage and, 25, 422424
storage application services and, 449450
storage security technology, 427432
tools for, 25
WWN spoofing, 431, 436437
zoning, 433437
sequences, 231, 438
serial connections, 100101, 111
server-free backups, 248, 256
servers
Apple, 122123
connecting peripherals to, 111113
connecting to SANs, 212
data access, 16
database, 1112
database size and, 22
direct vs. network connections, 16
file. See file servers
Internet and, 14, 21
name, 221223
NAS, 44, 133134
SAN, 256261
server OS, 208209
specialization of, 11
thin, 31
UNIX-based, 38, 260
web, 1415
Windows-based, 40, 42, 260
service levels, 367371
described, 71, 498
Sfp (simple file processing) mode, 152153, 172
shared I/O systems, 112113
shared memory, 112113
shared nothing architecture, 52, 5556, 111112
sharing items
data, SAN, 241, 244246
devices, SAN, 241244
silos, 93
simple file processing (Sfp) mode, 152153, 172
Simple Name Server (SNS), 440441
Simple Network Management Protocol (SNMP), 238, 414
Small Computer System Interface. See SCSI
SMB signing, 444
SMF (System Management Facility), 411412
SMP (symmetrical multiprocessing), 5154, 498499
snapshots, 93, 174, 499
SNMP (Simple Network Management Protocol), 238, 414
SNS (Simple Name Server), 440441
soft zoning, 435436
software
backup, 485
drivers, 239241
maintenance, 247
NAS. See NAS software
problems with, 357359
recovery, 485
SAN. See SAN software
solid-state disks (SSD), 68
Southwestern CD Company case study, 477488
SPC (Storage Performance Council), 415417
SRM (storage resource management), 24, 485
SSD (solid-state disks), 68
Static Database Model, 173
storage application services, 449450
storage arrays. See arrays
storage capacity. See capacity planning
storage data management tools, 25
storage information, 379
storage infrastructure, 361
storage integration, 348350
storage intelligence, 209
storage locations, 6667
storage management, SAN, 248
storage networking model, 1820
storage networks
advantages, 1819
best practices, 488
business application. See business applications
business continuity and, 360361
change management, 392400
component relationship matrix, 388389, 395
configuration management, 385393
data access, 2223
data size, 22
effects on IT organizations, 3334
external service levels, 367370
implementing plan for, 371377
internal service levels for, 367, 370371
interruptions in service, 357359
I/O workloads in, 290292
IT organization needs and, 2123
management challenges, 377382
reasons for, 2123
remote vs. local, 363
upgrades. See upgrades
See also networks
Storage Performance Council (SPC), 415417
storage reporting, 371
storage resource management (SRM), 24, 485
storage security tools, 25
storage systems
archival, 70, 7273
arrays. See arrays
batch processing, 7172
business applications for, 2325
data organization, 115128
data protection, 7374, 174
Hierarchical Storage Management (HTM), 7476
historical information, 410414
magnetic disk, 8485
magnetic tape, 9094
managing, SAN, 248
near-line configurations, 7476
OLTP workload, 71
overview, 7076
performance, 117
planning for. See capacity planning
recovery scenarios, 357, 359
resource management, 409
typical configurations, 7176
See also data storage
switch operating system, 236251
switch ports, 297, 299
switched fabric networks, 55
switched networks, 58
switches
10Gbe, 376
Director Class, 221
fabric, 108
FC, 207210, 220224, 434436
Fibre Channel, 220224, 434436
low-end, 221
multiple, 215, 256261, 270
NAS, 178180
SAN, 58, 211, 236
symmetrical multiprocessing (SMP), 5154, 498499
system cache, 78, 68
system clock, 145
System Management Facility (SMF), 411412
systems, computer
performance of, 355
processing, 5153
See also specific systems
systems management, 354356

Index
T
T11.3 standard, 220221, 499
tables, 128
tape compression, 499
tape controller, 9091
tape drives, 9092
tape libraries, 9293
tape media, 9192, 210, 266
tape systems, 9293
Target Channel Adapter (TCA), 344
TCA (Target Channel Adapter), 344
TCP (Transmission Control Protocol), 80, 191192
TCP layers, 138139
TCP off-load engine (TOE), 139, 189
TCP stacks, 172
TCP wrappers, 102
TCP/IP networks
described, 102
evolution of, 4243
NAS and, 138139, 189193
SANs and, 209
vs. FC networks, 337
TCP/IP stacks, 80, 189
temporary storage, 499
test installations
NAS workloads, 332
SAN workloads, 313
thin servers, 31
throughput, 85
TOC (total cost of ownership), 344, 346
TOE (TCP off-load engine), 139, 189
TPC (Transaction Processing Performance Council), 415
tracks, 92, 499
transactions
application, 78
banking, 278282
data warehouse, 308310
deposit, 287291
user, 2223
web-based storage, 78
transfer rate, 85
Transmission Control Protocol. See TCP
trending, 377, 400
troubleshooting
configuration errors, 429430, 446447
described, 354, 400
diagnostic services, 398
disciplines, 396
documentation, 399400
interruptions in service, 357359
NAS environments, 398399
problem identification, 393, 397400
problem prioritization, 397, 400
problem resolution, 396, 400
reporting problems, 399400
root cause analysis, 378
SAN environments, 398
service information, 398
serviceability and, 393
storage change management and, 396400
storage problems, 396400
tool access, 398
tracking problems, 399400

Index
U
UA. See user access
universal bus, 9697, 343346
universal resource locators (URLs), 327
Universal Serial Bus (USB), 110111
UNIX clients, 152
UNIX environment
application information, 379
directories, 165167
HBAs and, 226
NAS and, 159161
networking storage issues, 3638
OS information, 379
UNIX file servers, 38
UNIX file system, 3839, 121124, 165167
UNIX kernel, 161162
UNIX-based servers, 260
upgrades
change management and, 393395
storage configurations and, 393395
URLs (universal resource locators), 327
USB (Universal Serial Bus), 110111
user access (UA)
challenges, 89, 1415, 2223
described, 282
number of users, 89, 22
OLTP workloads, 301
storage networks and, 1819
web servers, 1415
WIT workloads, 327
User Non-Aggression Pact, 407
usernames, 432
users
end users, 361
non-aggression pact, 369370
transactions, 2223

Index
V
Very Large Databases (VLDB), 51
virtual disks, 85
virtual files, 39
Virtual Private Networks (VPNs), 423
virtualization, 117
VLDB (Very Large Databases), 51
volume managers, 410411
VPN devices, 423
VPNs (Virtual Private Networks), 423

Index
W
WAN storage, 342344
WBEM (web-based enterprise management) standard, 413
web application workloads, 256, 258
Web Internet transactional (WIT) workloads
NAS, 327328
SAN, 304307
web servers, 1415
web services, 341
web-based enterprise management (WBEM) standard, 413
web-based storage
application transactions, 78
capacity of, 7
Metro Data Area evolution, 21
NAS and, 139141
vs. RAM storage, 78
web-based workloads
data paths, 305
user access, 305
wide area network. See WAN
Windows environment
application information, 379
file systems, 121124, 159
HBAs and, 226
NAS micro-kernel and, 164165
OS information, 379
Simple File Processing, 152
Windows-based servers, 260
WIT I/O, 327
WIT (Web Internet transactional) workloads
NAS, 327328
SAN, 304307
workloads
analysis, 287, 290291
attributes, 286
business applications, 281284
data organizational model for, 285286
data protection, 74
data warehouse. See data warehouse workloads
defined, 71, 278, 283284
estimating, 459461
HSM, 7576
identification of, 283284, 459461
importance of, 278
I/O content and, 285290
messaging, 256, 258
multiple types of, 292
NAS. See NAS workloads
for NAS devices, 173174
OLTP. See OLTP workloads
production, 215
requirements, 289290
SAN. See SAN workloads
SAN vs. NAS, 292
solutions matrix, 292
user access. See user access
web-based. See web-based workloads
workload solutions matrix, 292
See also I/O workloads
workstation access, SAN, 247
World Wide Names (WWNs), 431, 436437, 440441
WWN spoofing, 436437
WWNs (World Wide Names), 431, 436437, 440441

Index
Z
zones, 500
zoning
SANs, 242245

List of Figures
Figure 1-1: Database choices for applications
Figure 1-2: Components that effect application performance
Figure 1-3: Online storage for applications
Figure 1-4: The optimum configuration of user transactions versus storage resources
Figure 1-5: The increasing diversity of application data
Figure 1-6: The client/server storage model
Figure 1-7: An overview of database server configuration
Figure 1-8: Data moving through a data warehouse application
Figure 1-9: An overview of web server storage
Figure 2-1: Directly connected storage in traditional client/server environments
Figure 2-2: NAS example showing enhancements to data access
Figure 2-3: SAN example showing enhancement to data size
Figure 2-4: The application development lifecycle
Figure 2-5: Application implementation challenges
Figure 2-6: The realities of application implementation
Figure 2-7: Moving to networked data access with a SAN
Figure 2-8: Device-to-device communications within a SAN
Chapter 3: Decoupling the Storage Component: Putting Storage on
the Network
Figure 3-1: Early UNIX networking storage issues
Figure 3-2: Early UNIX file servers
Figure 3-3: The file system
Figure 3-4: The network file system
Figure 3-5a: Early NAS scientific applications
Figure 3-5b: Early NAS code development applications
Figure 3-6: Early Microsoft file servers
Figure 3-7: Schematic of the NAS configuration
Figure 3-8: A NAS for data access
Figure 3-9: A NAS for data size
Chapter 4: Decoupling the Storage Component: Creating a Network
for Storage
Figure 4-1: SMP processing
Figure 4-2: MPP processing
Figure 4-3: Fibre Channel Protocol
Figure 4-4: Switched Fabric Network
Figure 4-5: Shared Nothing Architecture
Figure 4-6: A storage area network
Figure 4-7: Early SAN configurations using HUB architectures
Figure 4-8: Current SAN configurations using fabric architectures
Figure 4-9: SANs for an OLTP database environment
Figure 4-10: SANs for a data warehouse environment
Figure 5-1: The diversity of storage locations
Figure 5-2: The data storage hierarchy
Figure 5-3: The definition of workloads
Figure 5-4: An online storage system
Figure 5-5: A batch storage system
Figure 5-6: The data protection workload
Figure 5-7: A typical configuration for HSM workloads
Figure 6-1: Peripheral connections to the main computer elements
Figure 6-2: Connecting to the network
Figure 6-3: Overview of connectivity inside the network
Figure 6-4: Host adapter functions
Figure 6-5: Path limitations through single adapter strategies
Figure 6-6: Controller functions
Figure 6-7: The controllers view of reality
Figure 6-8: The disk drive anatomy
Figure 6-9: Disk arrays
Figure 6-10: The basic RAID architecture for a storage array
Figure 6-14: An overview of the SCSI tape drive
Figure 7-1: A universal computer bus
Figure 7-2: The internal functions of a typical I/O bus
Figure 7-3: Bus operations: arbitration for control
Figure 7-4: The standard TCP/IP layers and functions
Figure 7-5: SCSI configuration single host
Figure 7-6: SCSI configuration multiple hosts
Figure 7-7: Fibre Channel layered architecture
Figure 7-8: The FC point-to-point topology
Figure 7-9: The FC arbitrated loop topology
Figure 7-10: The FC fabric topology
Figure 7-11: FC operations
Figure 7-12: A shared nothing high-speed interconnect
Figure 7-13: A shared I/O high-speed interconnect
Figure 7-14: A shared memory high-speed interconnect
Figure 8-1: The components of a file system
Figure 8-2: An enterprise-class file system
Figure 8-3: Hypertext Transport Protocol (HTTP)
Figure 8-4: The Common Internet File System (CIFS)
Figure 8-5: Direct Access File System (DAFS)
Chapter 9: Putting Storage on the Network: A Detailed Discussion
Figure 9-1: The NAS hardware components
Figure 9-2: The NAS software components
Figure 9-3: NAS file system configurations
Figure 9-4: The departmental NAS configuration
Figure 9-5: The Internet NAS architecture
Figure 9-6: Enterprise NAS configurations
Figure 10-1: The I/O manager components
Figure 10-2: RISC versus CISC
Figure 10-3: NAS appliance-level storage configurations
Figure 10-4: NAS mid-range and enterprise storage configurations
Figure 10-5: The NAS Sfp model
Figure 10-6: The NAS Qfp model
Figure 10-7: The NAS Cfp model
Figure 11-1: Overview of the NAS components
Figure 11-2: What is an RPC?
Figure 11-3: An initial NFS configuration
Figure 11-4: An early NFS implementation
Figure 11-5: Overview of the NAS micro-kernel components
Figure 11-6: A typical micro-kernel
Figure 11-7: An example of an NAS micro-kernel
Figure 11-8: The UNIX hierarchical directory system
Figure 11-9: Mounting a file system
Figure 11-10: Mounting a file directory
Figure 12-1: OSI Layersfrom the physical to the transport
Figure 12-2: Network devices and their corresponding OSI layer
Figure 12-3: A typical network infrastructure
Figure 12-4: The network segment supporting Accounting
Figure 12-5: Network segments within the data center
Figure 12-6: WAN connectivity options
Figure 12-7: NAS deployment using WAN connections
Figure 12-8: TCP/IP considerations for NAS deployment
Figure 12-9: OSI model with layers 4 through 7
Figure 12-10: Network operational aspects of NAS deployment
Figure 12-11: Network workload aspects of NAS deployment
Figure 13-1: SAN components overview
Figure 13-2: FC layers and functions
Figure 13-3: A SAN devices overview and sample read I/O
Figure 13-4: A typical fabric OS micro-kernel
Figure 13-5: Fabric OS micro-kernel components
Figure 13-6: A simple SAN architecture
Figure 13-7: A simple SAN configuration
Figure 13-8: A specialized/advanced SAN configuration
Figure 13-9: An entry-level SAN configuration
Figure 13-10: A mid-range SAN configuration
Figure 13-11: An enterprise SAN configuration
Figure 14-1: Port configurations
Figure 14-2: The name server contains names, addresses, and information on nodes.
Figure 14-3: Extending the switch with an E_Port
Figure 14-4: The basic functions of the FC Host Bus Adapter
Figure 14-5: HBAs critical role in interoperability
Figure 14-6: An FC storage array connected in a loop configuration
Figure 14-7: An FC RAID storage configuration
Figure 14-8: A simple bridge/router configuration
Figure 14-9: Anatomy of the FC Frame
Figure 14-10: Typical flow control using Class 2 operation
Figure 15-1: The basic components of the switchs operating system
Figure 15-2: SAN applications running within the switchs OS
Figure 15-3: The HBA driver functions
Figure 15-4: Storage device segregation through port zoning
Figure 15-5: SAN configuration with zoning and LUN masking
Figure 15-6: SAN configuration using zoning and masking
Figure 15-7: Out-of-band processing
Figure 15-8: ISL functions supporting a simple cascading switch configuration
Figure 15-9: ISL supporting higher performance and failover
Figure 16-1: Viewing the SAN as an entire solution
Figure 16-2: OLTP workloads supported by a core/edge configuration
Figure 16-3: Web and messaging applications supported by a mesh configuration
Figure 16-4: A data warehouse application being supported by a cascading configuration
Figure 16-5: A fully heterogeneous OS SAN-supported environment
Figure 16-6: A data replication strategy and potential SAN solution
Figure 16-7: A future SAN-to-SAN configuration
Figure 16-8: Advanced clustering with shared storage SAN
Figure 16-9: The future: process and storage fabrics with a common interconnect
Figure 17-1: An example banking transaction
Figure 17-2: Organizing workloads for maximum performance
Figure 17-3: Banking configuration example: defining the business application
Figure 17-4: Placing workloads within an infrastructure
Figure 17-5: Guidelines for identifying and defining workloads
Figure 17-6: The workload inventory, categorization, and definition of the Banking Application System
Figure 17-7: User access information for workload analysis
Figure 17-8: Calculating workload requirements for byte transfer rates
Figure 17-9: Workload analysis logical model
Figure 17-10: Workload solutions matrix
Figure 18-1: Cascading SAN configuration
Figure 18-2: Meshed SAN configuration
Figure 18-3: A core/edge SAN configuration
Figure 18-4: An OLTP workload using a core/edge SAN configuration
Figure 18-5: A web Internet transactional workload using a meshed SAN configuration
Figure 18-6: A data warehouse workload using a cascading SAN configuration
Figure 19-1: An appliance/departmental NAS configuration
Figure 19-2: An enterprise NAS configuration
Figure 19-3: A specialized application support NAS configuration
Figure 19-4: Departmental workload using a NAS appliance configuration
Figure 19-5: A web Internet read-only workload using a mid-range NAS configuration
Figure 19-6: A seismic analysis workload using an enterprise-class NAS configuration
Chapter 20: Considerations When Integrating SAN and NAS
Figure 20-1: SAN block I/O versus NAS file I/O
Figure 20-2: NAS I/O local limitations and remote strengths
Figure 20-3: SAN I/O local strengths and remote limitations
Figure 20-4: iSCSI configurations showing remote block I/O operation
Figure 20-5: InfiniBands I/O fabric
Figure 20-6: Rapid I/O: a future processor module interconnects
Figure 20-7: HyperTransport: a future high-speed CPU interconnect
Figure 20-8: A storage network of the future
Figure 21-1: Fundamental systems management applied to storage networking
Figure 21-2: Integrating storage networking as a major infrastructure
Figure 22-1: A cascading SAN configuration supporting on-demand availability
Figure 22-2: A core/edge configuration supporting a highly available CRM application
Figure 22-3: A redundant mesh SAN configuration supporting a 24/7 application
Figure 22-4: Traditional backup processes and their overhead
Figure 22-5: Storage network backup and recovery
Figure 23-1: SAN configuration management categories
Figure 23-2: Integrating SAN physical and logical categories into a single repository
Figure 23-3: The complexity of NAS external relationships
Figure 23-4: NAS configuration management categories
Figure 23-5: NAS software and logical views with example categories
Figure 24-1: The traditional division of responsibilities within the data center
Figure 24-2: Volume management working with storage arrays
Figure 25-1: The defense-in-depth model
Figure 25-2: Configuration errors at the storage switch
Figure 25-3: Encryption protects against storage configuration errors.
Figure 25-4: Key distribution service
Figure 25-5: Soft zoning in SANs
Figure 25-6: Hard zoning in SANs
Figure 25-7: SAN World Wide Name table spoofing
Figure 25-8: Fibre Channel layer 2 and sequence control
Figure 25-9: Fibre Channel frame header attack
Figure 25-10: Man-in-the-Middle attack on SANs
Figure 25-11: Operating system security in storage networks
Appendix A: NAS Case Study: The International Image Processing
Company
Figure A-1: IIP systems configuration
Figure A-2: IIP imaging application process
Figure A-3: IIP NAS storage infrastructure
Figure B-1: An overview of the current systems configuration
Figure B-2: Data warehouse/data mart requirements
Figure B-3: SAN storage infrastructure
Appendix C: SAN/NAS Management Case Study: The Southwestern
CD Company
Figure C-1: Southwestern CDs Metro Data Area
Figure C-2: Southwestern CDs Systems Configuration

List of Tables
Table 25-1: LUN Masking Techniques Compared
Table 25-2: Best practices for storage security
Appendix A: NAS Case Study: The International Image Processing
Company
Table A-1: Storage Solution Evaluations for IIP
Table B-1: Processing and Service-Level Requirements for Data Warehouse and Data Marts
Table B-2: Storage Solution Evaluation Table for Data Warehouse/Data Mart Applications

List of Sidebars
Metro Data Areas-A Look to the Future of Storage Infrastructures
Guidelines for Estimating the Number of Ports Required to Support I/O Workloads
Guidelines for Estimating NAS Requirements

Complete Reference Books - Storage Networks

Uploaded by

Copyright:

Available Formats

Complete Reference Books - Storage Networks

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Complete Reference Books - Storage Networks

Uploaded by

Copyright:

Available Formats

What topics does the book cover?

What are some of the considerations when integrating SAN and NAS?

Storage Networks: The Complete Reference

byRobert Spalding ISBN:0072224762

Age of Data >> 24 hours

Batch Loads MonFri: nightly MonFri: nightly MonSat: nightly

You might also like