We can mathematically determine batch process segment elapsed time by simply dividing the required resource time by the available parallelism. For example, suppose a CPU-intensive batch process segment consumes 120 seconds of CPU. When run serially, this process takes 120 seconds. After some analysis, it was determined the process could be split into three parallel streams without any Oracle concurrency issues. The anticipated elapsed time becomes 40 seconds. The formula is as follows:

For this example, 120 seconds of CPU is required and three parallel streams are available, so the anticipated elapsed time is 40 seconds. If we looked at the average CPU utilization, it would now be around 75% busy, because only three of the four available CPU cores are being used.

The used and available parallelism are obviously very important. Just because four CPU cores or 100 IO devices exist does not mean they can be used. In our example, there are four CPU cores available, but the application developers were able to create only three parallelism streams. And, of course, if the application creates ten parallel streams, yet only four cores are available, the CPU subsystem will become a bottleneck, operating deep in the elbow of the response-time curve.