Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!think.com!spool.mu.edu!agate!ucbvax!ucdavis!caldwr!rfinch
From: rfinch@caldwr.water.ca.gov (Ralph Finch)
Newsgroups: comp.sys.isis
Subject: Re: Performance benchmarks for isis
Message-ID: <293@venice.water.ca.gov>
Date: 20 May 91 21:25:48 GMT
Article-I.D.: venice.293
References: <1991May15.230622.1378@digi.lonestar.org>
Organization: Calif. Dept. of Water Resources, Sac.
Lines: 27

In article <1991May15.230622.1378@digi.lonestar.org> bwoster@digi.lonestar.org (Bill Woster) writes:
>
>  We are evaluating isis for message speed and efficiency
>on a SUN sparc 2 network.

We are interested in the same sort of thing.  Here's what we want to
do; if anybody knows a group doing something like this we would really
like to hear from them.

We will be coming to a major decision within the next 6-9 months about
how to parallelize our mathematical models.  The choices we see right
now are vectorizing them, parallelizing them with some sort of
multiple cpu/shared memory machine with parallelizing compiler, or
running them on existing hardware (Sun network) with software such as
ISIS.

We would prefer the latter, because we could use existing hardware,
and as new machines are added to the net the model run just becomes
that much faster.  However, we are not sure if the communication
overhead would be too great relative to the computational overhead.
We would really like to talk to someone who has done distributed
scientific programming/numerical analysis/linear equation solving
using ISIS or something similar.
-- 
Ralph Finch			916-445-0088
rfinch@water.ca.gov		...ucbvax!ucdavis!caldwr!rfinch
Any opinions expressed are my own; they do not represent the DWR