Newsgroups: comp.parallel.mpi
From: salo@mrjones.engr.sgi.com (Eric Salo)
Subject: Re: MPI on SGI PowerChallenge
Organization: Silicon Graphics, Inc.  Mountain View, CA
Date: 1 Jun 1995 17:30:17 GMT
Message-ID: <3qktf9$5ji@fido.asd.sgi.com>

> I would like to get some input on the use of MPI on Silicon Graphics Power 
> Challenge. In particular, I am interested in hearing about experiences with 
> respect to software stability and efficiency. Being a shared memory 
> environment, I guess that a message-passing protocol is not the optimal tool.
> However, we would like to benefit from the portability offered by a common
> standard. Any opinions on how bad (or good) the MPI protocol will be for
> this particular computer?

Ah, my favorite subject...

We are developing an implementation of MPI that is optimized specifically for
the Power Challenge. It is available today but requires IRIX 6.1 or later. It
is based on MPICH, and we hope to include it as part of the next MPICH release.

We have found the the Power Challenge makes a very good engine for MPI. The
latency on a 90 MHz machine is currently 10-12 usecs for short messages, and
peak bandwidth is about 64 MB/sec.

As far as we can tell, the software for running MPI within a single box is
stable. The multi-host code is still being developed. The efficiency is also
very high, since we are able to deliver messages with a single bcopy().

Please let me know if I can be of further help.

Eric Salo         Silicon Graphics Inc.             "Do you know what the
(415)390-2998     2011 N. Shoreline Blvd, 7L-802     last Xon said, just
salo@sgi.com      Mountain View, CA   94043-1389     before he died?"

