2.

Objectives and Scope

benchmarking for use in WileyPLUS E5. In general the purposes of this

testing are:

z Validate the core Sakai framework and certain tools meet the minimum performance

standards established for this project. The following tools will be measured for

performance:

testing effort outlined in this document will not cover the following:

z Performance testing any new Sakai tools that are developed

existing Wiley performance test process. This test plan will serve as

the basis for Testware to create Silk Performer Test Scripts. These scripts will be run by Leo

Begelman using the Silk Performer software. Unicon, Inc. will watch and measure the CPU

utilization of the web and database servers used during testing. Unicon, Inc. will analyze and

present the performance test results to Wiley at the conclusion of the performance test cycle.

5. Load/Stress Test Types and Schedules

The following tests will be run:

z Capacity Test – Determines the maximum number of concurrent users that the

application server can support under a given configuration while maintaining an

acceptable response time and error rate as defined in section 7.

z Consistent Load Test – Long-running stress test that drives a continuous load on the

application server for an extended period of time (at least 6 hours). The main purpose of

this type of test is to ensure the application can sustain acceptable levels of performance

John Wiley and Sons, Inc.

in Appendices 1 and 2. (Note: there

should be 118 students and 2 instructors).

2. Use the results from the first execution to make a guess as to how many users the

system might support. One possibility might be to run 1000 different users through the

system for one hour, with approximately 240 concurrent users at a time.

3. If the second execution continues to meet the performance goals outlined in section 7,

continue to run new tests with increasing quantities of concurrent users until the
performance goals are no longer met. It is desired that one server will support up to 500

concurrent users.

4. Assuming the maximum capacity is determined, a consistent load test will be run. The

changes can be made to the system under test environment (see section 14) and run the

test again.

6. Test Measurements, Metrics, and Baseline

The following metrics will be collected

Database Server:

• CPU Utilization – Max., Avg., and 95th percentile. This data will be collected using the sar

system utility.

• SQL query execution time: The time required to execute the top ten SQL queries involved

has no cached images, script, or pages on their

machine because the request is a fresh request to the server. Therefore; that

request is expected to be more expensive.

• The cached mode means that images and pages are cached on the client with only

the dynamic information needing to be transmitted for these subsequent actions.

• It is recommended a mix of initial Action and Cached scenarios be included in the

performance test runs.

• This data will be collected using Silk Performer

Client:

John Wiley and Sons, Inc.

WileyPLUS E5 Load/Stress Test Plan

• Time to last byte (TTLB): This is what will currently be measured in the stress tests, as

opposed to user-perceived response time. Time to last byte measures the time between the

request leaving the client machine and the last byte of the response being sent down from

the server. This time does not take in to account the scripting engine that must run in the

browser, the rendering, and other functions that can cause a user to experience poor

performance. If the client-side script is very complex this number and the user perceived

response time can be wildly different. A user will not care how fast the response reaches their

machine (about the user perceived response time) if they cannot interact with the page for an
extended amount of time. This data will be collected using Silk Performer.

Network:

• Network Traffic: Network traffic analysis is one of the most important functions in

performance testing. It can help identify unnecessary transmissions, transmissions which are

larger than expected, and those that can be improved. We need to watch network traffic to

identify the bytes over the wire being transmitted, the response times, and the concurrent

connections that are allowed. This data will be collected using the sar system utility.

7. Performance/Capability Goals (Expected Results) and Pass/Fail Criteria

The following are performance requirements (success criteria) for the performance tests:

1. The average response time (measured by the Time to last byte metric) is less then 2.5

seconds

2. The worst response time (measured by the Time to last byte metric) is less then 30

seconds

3. The average CPU utilization of the database server is less then 75%

4. The average CPU utilization of the application server is less then 75%

5. Each blade server must be capable of handing 500 concurrent users

6. The maximum number of acceptable server errors, non HTTP-200 status codes on client

requests, will be less then 2% of all client requests.

8. Software and Tools Used

Component Software Version

Sakai Sakai 2.5.0 GA

Servlet Container Tomcat 5.5.25

Java Virtual Machine Sun Java Development Kit 1.5.0.14

Database Oracle 10g (version 10.?.?.?)

Load Test Tool Silk Performer

9. Load Descriptions

Each test outlined in section 5 will run with a ratio of 59 students to 1 instructor. There is no

expected difference between users logging in for the first time or subsequent logins given how the

data (outlined in section 10) will be created. The data set these tests will start with, will appear to

be in mid-course for all users.

There will be no ramp up time for any of the Single Function Stress Tests. The ramp up time
for all other tests, should be set to 1 user every 3 seconds. 120 users should therefore be

John Wiley and Sons, Inc.

WileyPLUS E5 Load/Stress Test Plan

running within 6 minutes. The wait time between requests is contained in the test scenarios in

Appendix 1 and Appendix 2.

In order to place as much stress on the system as possible with a small number of users, all

users should come from different worksites.

10. Content and User Data Preparation

The Sakai system will need to be preloaded with data before the performance testing will begin.

This data will be created using the Sakai Shell and Sakai Web Services. Once the data is

created it will be extracted from the database with a database dump. The following table

identifies several types of data that will need to be preloaded into the Sakai environment.

Type of Data Amount of Data

Users 93567

Students 92000

Instructors 1557

Administrators 10

Very Large Worksites 2

Large Worksites 5

Medium Worksites 50

Small Worksites 1500

Students to Instructors Ratio 59 to 1

Students per Very Large Worksite 1000

Students per Large Worksite 500

Students per Medium Worksite 250

Students per Small Worksite 50

Announcements per Worksite 13

Forum Topics per Worksite 1/3 of all forum posts in a worksite

Forum Posts per Worksite (spread across topics) ¼ students in worksite *

6.5
Columns (Assignments) in Gradebook 13

Instructor Resource per Worksite 20

11. Load Script Recording

Testware will create new Silk Performer Test Scripts based on the two scenarios outlined in

Appendices 1 and 2. Appendix 1 represents the Student scenario, while Appendix 2 represents

the Instructor scenario. The Test Suite should be set up to accommodate a ratio of 59 students

to 1 instructor. Wait time should be included between each page request, so that the total time

for an activity is equal to the number of page requests * 2.5 seconds + the total wait time.

12. Load Testing Process

This section details the load testing process that will be followed for all performance tests

conducted as described in this test plan.

John Wiley and Sons, Inc.

WileyPLUS E5 Load/Stress Test Plan

1. Start data collection scripts

2. Stop the application

3. Remove any temporary files

4. Reset the database and file system to known starting points

5. Start the application and run a quick sanity test to make sure each application server can

successfully return login screen markup can successfully process a login request.

6. Request the Silk Performer scripts be started

7. Once the Silk Performer scripts have completed, stop the data collection scripts

8. Acquire any database specific data being collected

9. Collate the data for all the metrics specified in section 6, into one report.

10. Make the report available in the Wiley Portal

13. Training Needs

Testware will need to be trained on the use of the Sakai environment. The scenarios outlined in

Appendix 1 and Appendix 2 give some idea about which parts of Sakai will be used, however,

more training will probably be required by the Test Script writers.

14. System-Under-Test (SUT) Environment

Specifying mixes of system hardware, software, memory, network protocol, bandwidth, etc.
• Network access variables: For example, 56K modem, 128K Cable modem, T1, etc.

• ISP infrastructure variables: For example, first tier, second tier, etc.

• Client baseline configurations

• Computer variables

• Browser variables

• Server baseline configurations

• Computer variables

• System architecture variables and diagrams

Other questions to consider asking:

• What is the definition of “system”?

• How many other users are using the same resources on the system under test (SUT)?

• Are you testing the SUT in its complete, real-world environment (with load balances,

replicated database, etc.)?

• Is the SUT inside or outside the firewall?

• Is the load coming from the inside or outside of the firewall?

15. Test Deliverables

The following test deliverables are expected as part of this performance testing effort.

z Test Plan – This Document

z Test Scripts – Silk Performer Test Scripts to implement the scenarios outlined in

Appendix 1 and Appendix 2

z Test Results Data – The data resulting from the performance tests run

z Test Results Final Report – The final report that documents and analyzes the results of

the performance tests that were conducted according to this test plan

John Wiley and Sons, Inc.

WileyPLUS E5 Load/Stress Test Plan

16. Team Members and Responsibilities

The following table defines the different team member responsibilities.

Responsibility Team Member

Test Plan Creation Cris Holdorph (Unicon)

Silk Performer Test Script Creation TestWare

Sakai Test Fixture/Data Generation Cris Holdorph (Unicon)

Silk Performer Test Script Execution Leo Begelman

CPU Monitoring and Test Orchestration Unicon

Database Loading and Analysis Alex Bragg (Unicon)

17. Risk Assessment and Mitigation

Business

Risk: Unsatisfactory Performance of Sakai

Mitigation: Conduct performance testing of core Sakai at the beginning of the project. If Sakai

does not meet the goals established above, project management will have the most time possible

to adjust project plans.

IT

Risk: Limit on the number of virtual users available with Silk Performer

Mitigation: Test only one blade server per 500 virtual users available with Silk Performer

Risk: All Sakai tools needed for testing at this stage may not be available

Mitigation: Tests will be conducted against the core tools that are in the Sakai 2.5.0 release.

Where a tool that is needed is not yet available, and place holder tool has been specified in the

test scenarios in Appendix 1 and 2. (e.g., Calendar will be used in place of Student Gateway for

this testing)