yade-users team mailing list archive

[Václav S(milauer <eudoxos@xxxxxxxx>]

-
> Yes, clumps have been implemented in Yade to simulate agregates of
> spheres.
With a few caveats, though: they are nut supported by NewtonsDampedLaw
and they are not wrapped from python. I can do the latter if it is
needed, it is quite simple. The former would however require a consent
on the order of clump/clumped bodies in BodyContainer
>
>> - and my last question: is it  available to conduct a parallel
>> processing on a network to decrease the computation time?  
> We are working on it, but I don't think it is fully functional yet.
Hmm, by accident I was just composing a call for developers on that few
days ago. Let me post it here, then:

Yade is not able to run parallel computation, even though all papers
mention that DEM, as an explicit method, parallelizes very well. While
processors have more cores, but the speed doesn't increase much,
supporting computation on multiple cores is becoming critical. We have
quad-cores now, there will be 6/8 cores in a year or so, Intel is
working on 96-core Larrabee platform, there is new standard (finally)
OpenCL for running computations on graphics cards that have easily over
500 stream processors, there are Cell processors with 6 (32 in the
future) vector computation units. Yade is going to lose the game if this
trend is not caught up with.

A decent implementation of parallelism in Yade is an effort adequate for
a PhD. Up to now, there has only been isolated efforts (lock-free
containers, ParallelEngine), but only as as part-time jobs to scratch an
itch; someone devoted to this for a few years is needed. I think it is
maybe 30% physics, the rest being quite low-level (and interesting)
software engineering effort.

If you (anyone) are in position to take PhD students, please consider
this as something of big importance for Yade future. I believe I am
speaking for other developers here.

Vaclav