Figure 9-26. This response-time graph focuses on the IO subsystem, so we chose physical IO as our workload metric. IO subsystems nearly always exhibit some queue time, and this situation is no different. Physical IO requests do include a significant amount of queue time, so we have multiple ways to reduce the physical IO-related response time. However, on this system, physical IOs are satisfied so quickly that the best course of action is to simply eliminate them by increasing the buffer cache.

Figure 9-26 shows the response-time graph based on physical IO processing during our reporting interval. The physical IO workload metric was chosen because it typically has a high correlation to IO requests and directly relates to our application analysis.

As expected, there is significant queue time involved with our IO requests. As presented previously, there are multiple ways to reduce the queue time and also the service time. One of our performance-improving strategies is to virtually eliminate all physical IO requests, essentially changing the arrival rate to zero. While the service time theoretically will not change, because the number of physical IO requests will be drastically reduced, Oracle will not need to spend so much time placing blocks into the buffer cache. This effectively reduces the CPU time spent per logical process, resulting in a reduction in service time. It will be interesting to see what actually occurs!