Scaling the size of the DynaCache

Edit online

This is a series of test cases that compare the various dynamic cache sizes.

These tests were run with the Trade caching option set to Distributed Map. The dynamic cache size was varied from 2000 to 20000 statements. The values used were 2000, 5000, 10000, 15000, and 20000. These runs were done using the WebSphere® Proxy Server and bypassing the Apache Web server.

These tests were run with workload clients starting at 25 and increasing by 5 until a total of 60 clients.

All runs were done with the cache configuration described in Table 1.
Table 1. Cache definitions - Scaling the size of the DynaCache
DynaCache size Caching mode Servlet caching Replication domain Replication type
2000, 5000, 10000, 20000 Distributed map Disabled Entire domain Not shared
Figure 1 illustrates the throughput when using various dynamic cache sizes.
Figure 1. Throughput - Dynamic Cache Size Scaling, WebSphere Proxy Server
This figure shows the throughput for Dynamic Cache Size Scaling and WebSphere Proxy Server.

Observations

A dynamic cache size of 2000 (the default) had the lowest throughput. Dynamic cache sizes, 5000, 10000, 20000 are very close, where the throughput with a cache size of 5000 statements is mostly the lowest from these.

Conclusions

The default cache size is not the best for this environment. Any cache size above 10000 seems to be suitable for this environment.

Figure 2 illustrates the CPU utilization when using various dynamic cache sizes.
Figure 2. CPU utilization - DynaCache sizes z/VM®
This figure shows the CPU utilization for DynaCache sizes z/VM

Observations

The total CPU utilization results for the various DynaCache sizes are all within 2 % of each other. The DynaCache size of 2000 has a higher WebSphere CPU utilization and corresponding lower throughput. This would seem logical because there is a smaller amount of memory available to hold cached data and would require more WebSphere processing to satisfy the cache misses.

Conclusion

Reducing DynaCache sizes from 20,000 statements to smaller values for this environment did not produce any significant change in either CPU utilization or throughput.