Newsgroups: comp.parallel.mpi
From: gdburns@osc.edu (Greg Burns)
Subject: Re: mpi availability & performance
Organization: Ohio Supercomputer Center
Date: 22 May 1995 13:25:26 -0400
Message-ID: <3pqhe6$8t1@tbag.osc.edu>

In article <3piq5e$4aa@NNTP.MsState.Edu> tony@aurora.cs.msstate.edu (Tony Skjellum) writes:

>CampbellM (campbellm@aol.com) wrote:
>
>: 1. What platforms is mpi *currently* available on?
>
>We know that there are many MPI's around.  The key ones are as follows:
>	LAM - for clusters, not emphasizing performance but rather environment
>	for development

To elaborate, LAM does not de-emphasize performance.  From the beginning
we have paid a great deal of attention to performance.  As a daemon based
implementation like PVM, we have a built-in overhead versus a
library based implementation.  The standard nature of MPI allows different
implementors to concentrate on different segments of the market.  LAM is
focussed on cluster computing, where its performance is quite comparable
with MPICH.

Choosing "an MPI" is not like choosing between P4, PVM and PARMACS as in
days gone by.  It is more like choosing an ANSI C compiler.  You might
use GNU to develop a piece of code on your workstation and then you
might switch to a proprietary compiler on the target parallel computer.
The idea is that the code is 100% portable.  It is correct that we are
trying to make LAM the best MPI development environment.  However, we
expect that users will run portable debugged codes on whatever implementation
provides acceptable performance for that application.

-=-
Greg Burns				gdburns@tbag.osc.edu
Ohio Supercomputer Center		http://www.osc.edu/lam.html

