Enterprise Content Management workload

The workload for this case study was based on the IBM® Rational® Performance Tester (RPT) application. RPT is a product for automating the load and scalability testing of server-based software applications.

RPT can be used to prove that an application or an entire system under test (SUT) will scale and perform to meet defined load levels. The machine (x86_64) on which RPT ran, acted as an external workload generator for the SUT implemented in the z Systems™ server.

The users emulated by RPT are virtual users. Because of the short think times in the RPT scripts that were used, each virtual user generated the load-equivalent of many human end-users who would use ICN to accomplish real work tasks. Therefore, 1000 virtual ICN users probably represented a work mix equivalent to many thousands of real end users.

Using RPT, it was possible to record user actions on graphical user interfaces (GUIs) and replay them at a later date. The workload consisted of user actions that followed a defined activity flow.

During the rampup phase, virtual users logged in to ICN one-by-one, with a small delay. Instantly after login, the virtual users started to execute a few one-time transactions.

Once the one-time transactions were completed, the users entered a loop and randomly executed transactions from a pool of transactions with a well-defined distribution of transaction types.

The stable workload phase was reached when all virtual users were logged in and were executing their transaction loop. Between the transaction execution was a thinktime, where the virtual user was idling for a few seconds before starting the next transaction. The metrics transaction throughput and average response time were taken from this stable workload phase.

After a certain time period, the transaction loop ended and the virtual users started to log out from ICN. This was the rampdown phase of the workload, lasting until all virtual users were logged out.

The more virtual users were logged in parallel during the stable workload phase. The smaller the thinktime, the higher was the workload level for the SUT.

The z/VM® hypervisor had 30 IFL processors in total and for the case study four load levels were defined. Starting from low to medium, medium-to-high and high (shown in Figure 1). These load levels were used throughout the scale-out study.