Based on Figure 9-3, the total service time is 16,881.6 seconds, or 16,881,600 milliseconds. Determine the service time by dividing the total service time by the unit of work value. Here is the service time math:

Determine the queue time by dividing the total queue time by the unit of work value. For Oracle systems, the total queue is all the non-idle wait time that occurred during the sample interval. Most Statspack and AWR reports have a Top 5 Timed Events section near the top of their reports. This is simply the top four most time-consuming wait events and also the CPU time. Usually the top four wait events account for 90% or more of all the non-idle wait time. For our required level of precision, we can simply sum the wait time for the top four wait events. While the details are not shown, their combined wait time is 45,672 seconds during the sample interval. Here is the queue time math:

For CPU subsystems, you can compare the actual CPU utilization (perhaps gathered from v$osstat or vmstat) with the classic utilization formula. If you picked a good unit of work, the difference should be within 15%. If the bottleneck is the IO subsystem, because of caching and batching, utilization comparison may be interesting, but it is unlikely to closely match or provide much value.