Varying virtual CPUs on the WebSphere Application Server

Response time

Observations

Figure 1 shows the CPU utilization of the LPAR for the DistributedMap cache mode when increasing the number of virtual CPUs on the WebSphere Application Server from two to four. The throughput increased by 40%, while the response time decreased by 30% to below 20 ms, which is a very good value.

CPU Utilization of the LPAR

Observations

Figure 2 shows the CPU utilization of the LPAR (eight physical CPUs) from the Performance Toolkit for z/VM® from a phase during the measurement. The utilization of the eight physical CPUs from the LPAR increased from 55% to 79% by adding two more virtual CPUs to the WebSphere Application Server.

Observations

Figure 3 shows the physical CPU consumption per throughput of 1,000 pages per second. The higher throughput with four virtual CPUs on the WebSphere Application Server adds very small additional virtual CPU overhead to the cost of 4.5%.

Conclusions

It is important to assign enough CPUs to the WebSphere Application Server, which could easily be done under z/VM, because these are only virtual CPUs. In our case, the additional CPUs for the WebSphere Application Server resulted in significantly higher throughput and better response times. The physical CPU utilization scales very close with the throughput with only a small overhead of 4.5%. The LPAR load of 79% indicates that it should be possible to increase the throughput further by using a cluster.

Be aware that this scenario (WebSphere Application Server runs CPU constrained) might not be obvious because of the methodology of the monitoring software. The measured values are averaged over a particular time space, so the fluctuations of the load are mostly flattened out. The reported values might not indicate that there is a bottleneck so it is very important to monitor the CPU utilization on a very granular level, if this needs to be identified.