yade-users team mailing list archive

[Bruno Chareyre <bruno.chareyre@xxxxxxxxxxx>]

You are wiki member Eugen,

> Here is a first screenshot:
> http://s7.directupload.net/images/130315/ns9gpq39.png
>
> On the x-axis of the stacked bar charts you'll find "count of Collider vs. Iterations done".
> For example: 
>  ->one core and 1000 particles: collider runs 17 times out of 1100

Very interesting!
The last column (512k particles) is irrelevant I'm afraid.
The collider running at each step indicates that something wrong is
happening, probably a bad timestep leading to scheme instability and
particles moving at high speed, or something like that. I suggest to not
publish the results with 500k, or fix the script and run again.
This aside, the results are logical: collider % increases with the
number of threads and number of particles, this is the identified
bottleneck.
Still, it may be possible to modify collider settings to reduce its
frequency at the price of more virtual interactions. For, let's say,
500k particles and 8 threads, there is an optimum to find in terms of
verletDist.

> I hope this will help improving Yade and exposing why yade wont benefit
> from a lot of cores.

Yes, this is very clear.

B