If you were to draw the response-time curve, it would look something like Figure 1-4. (Keep in mind this is an abstraction, and I took certain liberties, which will be fully explained in Chapters 5 and 9.) Looking at the response-time curve in Figure 1-4, you can see that, at an arrival rate (workload) of 1,510 uc/sec, there is a clear problem, because the system is operating in the elbow of the curve. There is a possibility that service levels are still being met, so this situation does not represent a problem. However, just a slight increase in the arrival rate (workload) will very likely cause a service level breach.

Figure 1-4. This response time curve is based on 12 CPU cores with a service time of 0.0074 sec/uc. In this example, the arrival rate is 1,510 uc/sec, which mathematically equates to an average queue time of 0.0059 sec/uc, an average CPU utilization of 93%, and a response time of 0.0133 sec/uc.

As noted in the preceding list, during the one-hour period, total queue time (wait time) was 32,022 seconds. Since over the one-hour period, 5,400,000 user calls occurred, the average queue time per user call is 0.0059 sec/uc, or 32,022/5,400,000. For service time, Oracle consumed 39,960 seconds, and therefore the average service time for each user call is 0.0074 sec/uc, or 39,960/5,400,000.16