Use numNodes instead of numLocales in benchmarks

[e-kayrakli]

Based on #5021, and should be merged after that.

This PR changes benchmarks to use numNodes config instead of numLocales for properly sizing input data.

[jabraham17]

consider printing both numNodes and numLocales for debugging purposes.

If you do this, should should do it in all benchmarks

[e-kayrakli]

For nightly runs: we have a separate config for co-locale and no-co-locale runs. So, it'd be very easy to know what a particular result is from. Running manually, you already have more control over number of (co-)locales, so it feels less important there.

I do agree that more information printed out from tests is better in general, though. Probably a more principled way is to have a helper printConfigInfo of sorts that are printed out by tests/benchmarks, or by pytest at the beginning of each benchmark.

That's my reasoning to not take this suggestion. Do you want to push more for it?

[jabraham17]

I think its a quick change that provides value, when reading the logs without knowing how the server was run you can easily tell what is being measured.

I agree a more principled way to print this information is in order, but that shouldn't stop us from making incremental improvements.

That's my reasoning to not take this suggestion. Do you want to push more for it?

I fear if the improvement is not made now, it will never be made :). I won't push super hard for it, but I do think you should make that change

[e-kayrakli]

Fair enough. I will do that.

[e-kayrakli]

df7a33a makes the change. I did it with sed and scrolled around a bit as a spot-check.