Testing Siebel Business Applications > Execute Performance Tests >

Measuring System Metrics

Results collection should occur during a measurement period while the system is at a steady state, simulating ongoing operation in the course of a business day. Steady state is achieved once all users are logged in and caches (including simulated client-side caches) are primed. The measurement interval should commence after the last user has logged in and completed the first iteration of the business scenario.

The measurement interval should last at least one hour, during which system statistics should be collected across all server tiers. We recommend that you measure the following statistics:

• CPU
• Memory
• System calls
• Context switches
• Paging rates
• I/O waits (on the database server)
• Transaction response times as reported by the load testing tool

  NOTE:  Response times will be shorter than true end-user response times due to additional processing on the client, which is not included in the measured time.

The analysis of the statistics starts by identifying transactions with unacceptable response times, and then correlating them to observed numbers for CPU, memory, I/O, and so on. This analysis provides insight into the performance bottleneck.