Newsgroups: comp.parallel.mpi
From: lederman@b125.super.org (Steve Huss-Lederman)
Subject: Re: request for performace results
Organization: Supercomputing Research Center, Bowie MD, USA
Date: Tue, 27 Sep 1994 17:51:36 GMT
Message-ID: <LEDERMAN.94Sep27135137@b125.super.org>

Hello MPIers!

I have received several private requests about MPI performance results
and now Michele Freed has made a public request.  Therefore, I will
now generally post this information.

We have written a paper for a minisymposium to be given at the SIAM
Parallel Processing conference in February.  It is available via
anonymous ftp from ftp.super.org in pub/prism.  It is wn22.ps.  This
article is only 6 pages long since that was the conference limit.  It
contains results analyzing the speed of broadcast on the Delta,
Paragon, SP1 and CM5 using the MPI from Argonne/Mississippi State.
The MPI_Bcast is compared to our own broadcast using MPI send/recv as
well as an optimized NX version on the Intel machines.

If people look at this paper I ask them to keep in mind what is
intended.  It is a snapshot of the status as of the date the results
were obtained.  They also only represent a modest subset of all the
timing results we had obtained up to that time.  Since then we have
gotten results on more systems and versions of MPI.  We may update and
expand this paper in the future.  The minisymposium talk will have the
latest results.

Hope this is of help to some.

Steve


