Methodology

To compare the impact of the various rules for the cpuplugd daemon, a manually optimized setup was prepared.

The objectives for the manual sizing were minimal memory requirements (without swapping) and a CPU utilization of 90% to 95% for all CPUs. The assumption was that the automated sizing using cpuplugd will never create a system with a higher utilization.

Compared to this workload, the additional CPU utilization needed to run cpuplugd, which is activated at intervals potentially as small as 1 second, was expected to be small, and the system will never run faster when the resources are managed by any kind of tool. The question of interest was whether the expected degradation would be large or acceptable.