Hitachi

uCosminexus Application Server System Design Guide


8.1.2 Tuning procedure

Performance tuning is a task that involves detecting the best settings for system performance. In an environment that is already built, revising parameters, identifying and removing bottlenecks during actual processing and mock loading can continuously enhance the performance.

The procedure of tuning a number of concurrent executions is shown below as an example of performance tuning procedure:

Figure 8‒1: Procedure of performance tuning (tuning the number of concurrent executions)

[Figure]

For tuning, first of all, you need to decide the target value. In the above example, it could be any value such as the CPU usage.

Next, measure the throughput where the default value is set in each parameter, then while applying the simulated load, revise each parameter and search for the optimum value that is closest to the target value. Using a special tool for this purpose can simulate load-increase.

You can use tools such as the monitoring tool of the OS for estimating the CPU usage for the purpose of tuning. You can use tools such as the load generation tool for estimating the throughput. Moreover, you can use the statistics collection functionality to collect statistic information of Application Server such as the number of statistic threads. For details about confirmation methods, see 3. Monitoring the Statistics (Statistics Collection Functionality) in the uCosminexus Application Server Operation, Monitoring, and Linkage Guide.

The performance tuning is complete when throughput reaches a value specified as the target value. If the CPU usage is slightly less than 100%, bottlenecks are likely in the I/O and exclusion processing of the system. Identify and remove the bottlenecks and then retry performance tuning. You can use performance analysis trace to identify the bottlenecks in the Application Server system. For details about the functionality of performance analysis trace, and how to use the trace file acquired using the performance analysis trace, see 7. Performance Analysis by Using Trace Based Performance Analysis in the uCosminexus Application Server Maintenance and Migration Guide.