GOAL

Quantify the impact (perf costs/overhead) of mis-configuring PCP Power Usage metrics collection (PMDAs).  

We have several flavors of PowerUsage collection mechanisms. Redfish uses Openmetrics based collection, RAPL has a dedicated denki-rapl PMDA and nvidia-gpu is also separate PMDA.  I should probably do some stress testing on a system that would (timeout) on all of these. On perf-arm-11 I could install them all and give it bogus URL for Redfish. The NVIDIA and RAPL would fail on their own. I could run sysbench with PCP (pmlogger.cfg requesting all Power Metrics) & without PCP and see if there's any perf impact.

ACCEPTANCE CRITERIA

Document methodology and results

Scripting and Documentation are available in github repo