Newsgroups: comp.parallel.mpi
From: 4er@stretto.iems.nwu.edu (Robert Fourer)
Subject: MPI for SPARC 20/514MP; for ATM
Organization: Northwestern University
Date: 13 Sep 1994 17:22:07 GMT
Message-ID: <354n3v$5u@news.acns.nwu.edu>

We are beginning to set up a "multi-processing computing
facility for large-scale optimization" which will consist
of a network of workstations using, most likely, MPI.  The
idea is to do a lot of learning and prototyping on this
dedicated facility, with finished code eventually being
ported to more powerful MPI-supported configurations.

We would appreciate assistance or pointers that anyone can
provide to enable us to answer the following two questions:

(1) Is there any point to acquiring a multi-processor
workstation such as a Sun SPARC 20/514MP?  Specifically, do
current MPI implementations support message passing between
the processors of such a machine?  And if so, is the
message speed significantly greater than what one can get
using networked single-processor machines?

(2) Is there any known obstacle to using MPI with 
workstations that have been networked using a high-speed 
protocol, specifically ATM?  Would ATM make a significant
difference to the message speed for applications written
using MPI?

Robert Fourer
Industrial Engineering and Management Sciences
Northwestern University
Evanston, IL 60208-3119

(708) 491-3151
(708) 491-8005 (fax)

4er@iems.nwu.edu


