Hello! Sergio Pastor Pérez <[email protected]> skribis:
> Oh, I did not think that `-c` is a configuration option that applies per > derivation. It would be nice to have an option to assign a pool of CPUs > that the daemon can assign to any of the, in this case, 10 concurrent > derivations, that would alleviate the CPU core context switching, right? > > I'm guessing that doing `-c $(nproc)` alone would not always be faster, > since there are operations done while building a derivation that are not > particularly heavy on CPU, such as fetching sources or writing to disk. > > Do you think we could implement this kind of feature? Or perhaps is > there a better way to balance the load of derivation building? The > `-c $(nproc) -M 10` has quite a major speed up in my experience. It would be nice to have that, yes. I would personally like it if we could stop adding stuff to the C++ code base, but I realize there’s still a gap to overcome before we can fully do that. Thanks, Ludo’.
