Based on v$osstat data shown in Figure 9-6 and the reporting interval shown in Figure 9-2, we calculated in the subsequent sections the server is running at 100% CPU utilization. While the wait event situation is not shown, the Statspack report shows the top wait event is clearly CBC latch contention. Based on the instance CPU consumption data shown in Figure 9-3, the reporting interval shown in Figure 9-2, and the CPU core number shown in Figure 9-6, we calculated an Oracle CPU utilization of 94%. Clearly, there is a CPU bottleneck.

When you choose an appropriate unit of work, the response-time graph will be a good representation of the real system. This will make presenting the graph very natural and understandable, and will naturally lead into your performance solutions discussion.

Following our example of a raging CPU bottleneck, we will use logical IOs as our unit of work. Logical IOs consist of all buffer touches. Oracle distinguishes current mode buffer touches by the db block gets instance activity statistic, and the consistent mode buffer touches are signified by the consistent gets statistic. These two statistics will be combined to produce a single logical IO statistic. Based on the instance statistics shown in Figure 9-2, 1,307,632,010 logical IOs occurred during the Statspack reporting interval.