Hello Leon :) On 05/09/2025 18:18, Leon Woestenberg wrote: > On Fri, 5 Sep 2025 at 01:45, Marc Gonzalez wrote: > >> I am trying to get reproducible / deterministic / low variance results >> using a vendor kernel (6.8.0-79-generic #79-Ubuntu SMP PREEMPT_DYNAMIC). > > Is that your end goal or is your question maybe an XY question? > If so please describe X. I need a rudimentary (but accurate) benchmark framework, to measure the impact of micro-optimizations on codes. These optimizations are expected to make programs 0-3% faster, so the framework should have the lowest variance possible, otherwise I can't discern the result from the noise. > Or just include the measurement inside the program using CLOCK_MONOTONIC. David suggested that as well, but it was not the source of the problem. John pointed out the culprit: /proc/sys/kernel/sched_rt_runtime_us https://www.kernel.org/doc/html/latest/scheduler/sched-rt-group.html Regards
