Summary for the WebSphere Application Server cluster with caching tests

Edit online

This is a summary of results and recommendations as a result of running the Trade workload on the four node WebSphere® Application Server cluster with caching enabled.

These test results and recommendations are Trade specific. Parameters useful in this environment might be useful in other environments, but because caching benefits and costs are highly dependent on application usage and system configuration, you will need to determine what works best for your environment. For detailed test results information, refer to Test case scenarios and results.

The following are summary results related to hardware and hardware configuration:
  • The WebSphere proxy server obtained a better overall level of throughput than the WebSphere Edge Service Caching Proxy Server.
  • The IBM® System z10® system obtained a significant throughput advantage over the IBM System z9® system.
  • The z/VM® VSWITCH LAN configuration resulted in higher throughput than the Guest LAN feature.
The following are summary results related to software and caching:
  • Any form for caching: Command caching, Command and servlet caching, or Distributed Map caching, resulted in a significant throughput improvement over the no caching case.
  • Distributed Map caching resulted in the highest throughput improvement.
  • A Dynamic Cache size of 10000 or greater produced the best throughput.
  • The smaller Dynamic Cache sizes benefit from enabling DynaCache disk offloading. This is especially true for a transaction mix that is mostly read only.
  • Configuring disk offload is not related to overhead in the tests that were run. This feature could be used to run with a significantly smaller cache size. For example, 2000 statements would save memory without causing a performance degradation.
  • For this environment, the default JVM garbage collection policy resulted in the highest throughput.
  • In this environment with the Trade application, the non-shared cache replication policy resulted in the highest throughput.
  • The shared-pull and shared-push-pull replication policies resulted in significant throughput degradation.
Based on the results of the test scenarios, the following is recommended:
  • Use some form of caching whenever possible. The additional development effort to add Distributed Map caching in the application can result in a measurable throughput improvement.
  • Enable disk offloading when using small Dynamic Cache Sizes.
  • Use z/VM VSWITCH instead of a guest LAN.