Par_iter_mut does not seem to be utilizing available (idle) cores? about rayon HOT 2 CLOSED

Try adding a call to weight_max in there? The default -- which I suspect I should change -- is to assume that each task is very cheap. weight_max tells is to treat tasks as very expensive, and hence to avoid lumping them together.

from rayon.

Ahhh, I see. That did indeed fix it, full usage across all cores. Dropped from 40s to 5s :)

I assumed that Rayon assumed the opposite: each task was expensive, and so maybe my tasks weren't expensive enough and were finishing quicker than Rayon could schedule them. I'll read up on the weighting code, perhaps a middle-ground will work better.

Tangentially, would it be possible for each task to set it's own weight? E.g. each Page knows how much work it needs to compute before it starts computing, and is relatively variable. It might be nice if each individual task can set it's weight dynamically. Not sure if that's feasible with Rayon's architecture though.

Thanks for the help (and for Rayon in general)!

from rayon.