yade-dev team mailing list archive

[Václav Šmilauer <eudoxos@xxxxxxxx>]

> I did some performance tests of Yade and they are very good. But I am
> confused about OpenMP because the results with the 1 and 2 threads are
> almost identical. Maybe I have something wrong doing? I attach the used
> scripts and results.

Hi, I put together a simpler benchmark based on examples/cyl.py. I put
it to bzr branch at https://code.launchpad.net/~yade-dev/yade/cyl-bench.

For the one that you sent:

1. if you don't use InteractionDispatchers, therefore the parallel
performance will be MUCH worse (3 loops instead of 1)
2. SQLiteRecorder is just leftover, I assume.
3. First step should not be measured, since the collider is being
initialized (not proportional to N)
4. Use InsertionSortCOllider::velocityBins, it eliminates lots of time
spend otherwise in the collider.

I will send the same for a 6-way Opteron machine. I think it is really
RAM speed (and architecture) that makes the most difference.

Cheers, Vaclav