Reserve cpus for benchmarking

[blackgeorge-boom]

Hello, thank you for the great tool!

I was wondering whether it's better to combine this tool with something like cpuset, to reserve the cpus the benchmark and hyperfine run on.

Would this make a difference? I understand that hyperfine will catch outliers stemming from, e.g., competing programs, but would this be avoided in the first place by reserving the cpus?

Thank you for your time.

[sharkdp]

This is a great question. I don't have the answer. I have experimented with hyperfine + cpuset in the past and I believe that it may have helped reducing outliers, but I can't provide definitive proof.

Note that context switches are not the only source of noise. There's also I/O, which often plays a huge role when benchmarking CLI apps, but that obviously spends a lot on your use case. cpuset will not help with this.

[blackgeorge-boom]

Ok, so my understanding is that if someone sees notable outliers/variability in their results, they should consider running hyperfine with cpuset as an option, but remember that this would help only if the variability is due to context switches.