[Wien] Wien2k with load-distributing systems

Peter Blaha pblaha at zeus.theochem.tuwien.ac.at
Wed Sep 17 11:26:20 CEST 2003

> I'll try to recompile WIEN using the shared memory architecture. As I don't
> really get that section of the userguide: how do I tell WIEN to use the
> finegrained parallel instead of k-point parallelization?

I would NOT do that on a regular basis.

Finegrain mpi parallelization works ONLY reasonable when your matrixsize is
above 5000 (better 10000 or more!!!), and you need very fast
communication!! (shared memory), but even then you will never get 100% speedup.
If you have bad luck (slow communication, syncronization problems), you
might run on 4 processors NOT faster than on ONE !!

In essence:

k-point parallelism can run with 100% speedup.

Fine grain only for cases NMAT>10000 and shared memory (IBM SP4,....)

Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/

More information about the Wien mailing list