Analyzing an Overloaded CPU

A CPU is overloaded when a high percentage of active users is found in CPU wait, that is, does not get as much processor time as they want to. This situation is likely to develop on systems which run near to their maximum capacity for much of the time. The things to look for are:

High CPU wait percentages for the users, and a high number of active users in the dispatch list
High CPU load figures

However, the problem can be a bit more complicated:

CPU Wait: The information is obtained by the high-frequency sampling technique of the CP monitor.

CPU Load: The CPU load percentages are based on real elapsed time. For VM systems running in an LPAR or second level under another VM, this may mean that processors may run at maximum capacity even though their utilization is still well below 100%. On the other hand, a high CPU load alone does not necessarily indicate a bottleneck. CP's dispatching algorithm usually takes care of a single CPU-intensive machine, and, as long as the CPU-bound user does not have a high dispatching priority, CP will distribute CPU time so well that only very little degradation is felt even if one machine is in a permanent CPU loop.