RTCEE Performance Testing: EE Workload Scenario

This scenario tests the performance of the enterprise Dependency Based Build under high volume workload, simulating concurrent build operations that developers would execute in a real case scenario.
The tests itself are based on observations of the usage performed by our customers and our recommended practices to them.

Scenario Detailed Description

RTC Parametrization

Test assets configuration:

• Personal builds requested will have the Minimum Load option set
• Several streams are used in these tests. These streams have not been generated using the “Duplicate” stream feature. Duplicating a stream causes the Source Code Data to be reused, while for the tests information gathering it is preferred that every stream has different set of SCD.
• System Definitions: 4.0.3 Link-edit feature will not be used.
• Build Engines configuration: Build engines should be configured in such a manner that it is assured that a team build request doesn't impact personal builds concurrency. To accomplish that, the following property is added to the build engines that will be serving personal builds: “requestFilter” with value ”personalBuild=true”:

RTC Server configuration options:

• Set advance property: 'Consider only outgoing change sets in personal builds = true'
• Customization of the Asynchronous Task Pool Size parameter for the amount of concurrent builds that will be executed.
• Source Code Data Collection: customize the Max Number of Files in Source Code Data Resource parameter to a maximum value of 100:

Virtual users characterization

To emulate the load and the main tasks expected during real solution usage, the following roles and operations have been identified.
These are used in the testing framework to emulate real users operations at the load and rates finer specified later in this topic:

• Developer role A:
  • Change a COBOL program
  • Check-in the changes
  • Request Personal Build to verify changes
  • Accept incoming changes
  • Deliver his change sets

• Developer role B:
  • Change a copybook source
  • Check-in the changes
  • Request Personal Build to verify changes
  • Accept incoming changes
  • Deliver his change sets

• Contributor role:
  • Login to repository
  • Create and save a work item
  • Execute a query (25%) -- just execute query every 4 iterations
  • Logout

Scenario execution preparation

Following are the main steps for preparing the environment for the test scenario run. The steps are beyond creating and configuring the assets already described:

1. Assign/verify Build System definitions with sample resources
2. Perform a full build of the applications
3. Load the virtual developers workspaces in preparation for the test run.

Scenario Variances

Based on the previous detailed configuration, two variances of the scenario are tested:

Test Scenario1

The following SCM structure is used:

• 50 components distributed among 25 streams:
• Each stream with 2 of the components with the test data sample distribution: one component will hold the “common” project of called “MortgageApplication-Common”, where the other component contains the rest of projects

The following screenshot shows a sample configuration of one of the streams:

• 30 build engines targeting the same build agent. This will allow the concurrency described later and the team build.
  • 25 build engines will be configured to just accept personal builds
• One build definition for each stream representing the build of the test data application
  • Each stream will have a team build with associated:
    • Build workspace
    • Defined HLQ and load directory for every team / build definition
• Each virtual user will need to have defined:
  • A personal repository workspace flowing to the assigned stream
  • A personal HLQ and load directory on the build machine
  • Personal workspace, HLQ and load directory will be used in the automated build requests processing

The scenario is meant to be performed during a long run of 8 hours, trying to emulate a system load during a working day, with the following details:

• The emulation will try to reproduce a scenario during a work day and variations of work load to different applications and streams. Similar to what would be normal lifecycle of development enterprise work, where some applications are more active than others.
  • From the 25 defined streams, 10 would be identified as more active ones, with users and load described in next subpoints.
  • Simulation of 28 concurrent developers.
  • Following distribution of emulated developers and streams:

2 streams will have 7 concurrent developers with the following roles and files to be modified in each iteration:

3 streams will have 3 concurrent developers with the following roles and files to be modified in each iteration:

5 streams will have 1 developer assigned: 1:1 relationship between developers and streams:

Additionally, one team build will be requested at every hour for each of the streams: not scheduled to start at same hour every time but with a difference of 10 minutes., so not all the team builds are executed at same hour time.

#Test Configuration

• Mortgage Application *500 which is 500 duplicates of the Mortgage sample application

We use this configuration for RTC EE scenario 1 test:

#Test Report of Scenario1

• For 8 hours long run of scenario1, 46 times(iterations) concurrent builds have been launched, total 1288 personal builds and 44 team builds have been processed successfully.

• Personal build time/aversage build time of concurrent builds in 8 hours test;

• No build(both personal and team build) fails in EE workload scenario 1 test;
• CPU usage of CLM server is < 60% ;
  

• CPU usage of BST & DB2z is < 60% ;
  

• CPU usage of DB2z is smoothly under 6% ;
  

• 140.4 Contributors are simulated in 2nd run - Overall Count Attempted Tasks For Run = 23990
• Size of repository volume don't increases in 8 hours exeuction(compare end with start condition).
• Indice incremental rate:

  Indice Directory Name	Before execution	After execution	Incresed Rate
  _Indices	2115620	2242828	6.01%
  ./jfs-rdfhistory	308	424	37.66%
  ./jfs-rdfindex	1986436	2112940	6.37%
  ./jfs-texthistory	52	252	0.00%
  ./jfs-textindex	128816	129404	0.46%

  Test Scenario2

  The following SCM structure is used:

  • 150 components distributed among 25 streams:
  • Each stream with 6 components with the test data distribution, using a different set of qualified file names

  The following screenshot shows a sample configuration of one of the streams:

  

  • 60 build engines targeting the same build agent. This will allow the concurrency described later and the team build.
    • 50 build engines will be configured to just accept personal builds
  • Three build definitions for each stream for building each set of the testing data
    • Each stream will have a team build with associated:
      • Build workspace
      • Defined HLQ and load directory for every team / build definition
  • Each virtual user will need to have defined:
    • A personal repository workspace flowing to the assigned stream
    • A personal HLQ and load directory on the build machine
    • Personal workspace, HLQ and load directory will be used in the automated build requests processing

  The emulation will try to reproduce a scenario during a work day and variations of work load to different applications and streams. Similar to what would be normal lifecycle of development enterprise work, where some applications are more active than others: as Scenario 1 but emulating a system with heavier load:

  • • From the 25 defined streams, 11 would be identified as more active ones, with users and load described in next subpoints.
    • Simulation of 75 concurrent developers.
    • Following distribution of emulated developers and streams:

  3 streams will have 7 concurrent developers with the following roles and files to be modified in each iteration:

  Role	#users	Files to modify (1 line per developer)
  Developer role A	9 (per stream)	AAACMORT.cbl / AABCMORT.cbl / AACCMORT.cbl AADCMORT.cbl / AAECMORT.cbl / AAFCMORT.cbl AAGCMORT.cbl / AAHCMORT.cbl / AAICMORT.cbl AAJCMORT.cbl / AAKCMORT.cbl / AALCMORT.cbl AAMCMORT.cbl / AANCMORT.cbl / AAOCMORT.cbl AAPCMORT.cbl / AAQCMORT.cbl / AARCMORT.cbl AAQCMORT.cbl / AARCMORT.cbl / AASCMORT.cbl AATCMORT.cbl / AAUCMORT.cbl / AAWCMORT.cbl AAXCMORT.cbl / AAYCMORT.cbl / AAZCMORT.cbl
  Developer role B	3 (per stream)	AAAMLIS.bms / AAAMTCOM.cpy / AABMLIS.bms / AABMTCOM.cpy AACMLIS.bms / AACMTCOM.cpy / AADMLIS.bms / AADMTCOM.cpy AAEMLIS.bms / AAEMTCOM.cpy / AAFMLIS.bms / AAFMTCOM.cpy

  5 streams will have 6 concurrent developers with the following roles and files to be modified in each iteration:

  Role	#users	Files to modify (1 line per developer)
  Developer role A	5 (per stream)	AAACMORT.cbl / AABCMORT.cbl / AACCMORT.cbl AADCMORT.cbl / AAECMORT.cbl / AAFCMORT.cbl AAGCMORT.cbl / AAHCMORT.cbl / AAICMORT.cbl AAJCMORT.cbl / AAKCMORT.cbl / AALCMORT.cbl AAMCMORT.cbl / AANCMORT.cbl / AAOCMORT.cbl
  Developer role B	1 (per stream)	AAAMLIS.bms / AAAMTCOM.cpy / AABMLIS.bms / AABMTCOM.cpy AACMLIS.bms / AACMTCOM.cpy / AADMLIS.bms / AADMTCOM.cpy

  3 streams will have 3 developers assigned: 1:1 relationship between developers and streams:

  Role	#users	Files to modify (1 line per developer)
  Developer role A	3 (per stream)	AAACMORT.cbl / AABCMORT.cbl / AACCMORT.cbl

  
  

  Repeating the role steps at cycles of 30 minutes. Additionally, one team build will be requested at every hour for each of the streams: not scheduled to start at same hour every time but with a difference of 10 minutes., so not all the team builds are executed at same hour time.

  Test Topologies

  The following topologies are being used for the execution of this scenario in order to test the performance and gather relevant data.

  Dual Tier Topology: z/OS

  

  Test Data

  From the existing scalability test data based on Mortage Application sample, the following will be used:

  500 times replication of Mortage Application sample.

  Assets	Overall dev assets
  3000 COBOL programs 2000 Copybooks 1000 BMS 3 others	6003

  Test Reports

  Fill with a brief introduction and links to relevant data

  Related topics:

  External links:

  • IBM

  Additional contributors: