Since we will be also using the sched_switch methodology to get context switches we should also compare our overhead to tools like kepler.
Even in newer iterations they seem to have quite some issues with strong overhead. Since people do drill-downs we can copy the methodology:
sustainable-computing-io/kepler#1670