Closed ayllon closed 2 years ago
I have re-run develop over the sample from the challenge data, and as I expected, it is not slower.
But this branch isn't substantially slower either.
Still, to be left on hold. The preliminary from the thread pool with priorities and with a limited queue look very promising.
Obsoleted
Helps #361.
They seem to reduce the memory footprint quite considerably for some use cases, without impacting the performance. This is for problem #380.
For other cases, it does reduce a little, and (surprisingly) it is faster. Likely I just got lucky with the node lesta assigned me (though they are all supposed to be the same and I do not think shared, since I use every physical cores: 16). This is the benchmark I use for multi-threading (from the challenge data):
At the very least, it doesn't seem to slow down things.
Anyhow, maybe wait for merging. I will try with now limiting the queue size and bumping the priority of measurements. It may help more, it may even be better if combined with this (?)