[Wrf-users] wrfv3 & mpi

[Jan Ploski]

Gerardo Cisneros <gerardo at sgi.com> schrieb am 08/27/2008 10:37:47 PM:

> On Wed, 27 Aug 2008, Jan Ploski wrote:
> > [...]
> > Here's one scalability comparison we did - 03 (OpenMP) vs. 05 (MPI 
over 
> > GigE) vs. 19 (MPI over InfiniBand):
> > 
> > https://srvgrid01.offis.uni-oldenburg.de/wrf-benchmark/case.pl?
> case=conus12km_2001
> > 
> > Up to 8 processes you should get a good speedup. [...]
> 
> I can scale 12km CONUS (computational costs only) using
> WRFV3 out to 1700 cores, using a system with two quad-core 
> processors per node, IB interconnect, and running on all
> cores of each node used.  I can't point you to a plot
> because the benchmark page isn't ready for release.
> 
> Higher resolution models have been known to scale to
> over 10,000 cores.

Good to know! I think that more people should release this sort of data - 
how else is a new user supposed to know whether the performance he sees is 
good, bad, or average?

My above remark about 8 processors referred to Gigabit Ethernet only. Can 
you confirm that GigE doesn't scale (because of latency, I suppose)?

Regards,
Jan Ploski