Newsgroups: comp.parallel
From: News Owner <news@waage.rz.uni-ulm.de>
Subject: Re: Parallel platform
Organization: University of Ulm, SAI, Germany
Date: Fri, 4 Aug 1995 08:42:54 GMT
Message-ID: <3vsmie$io8@waage.rz.uni-ulm.de>

Kevin Cameron (kevin@chronologic.com) wrote:
: In article <3tte8h$qda@convex.convex.com>, Jens Brauer <brauer@convex.convex.com> writes:
: |> Our group plans to procure a distributed-memory parallel
: |> computer platform to perform large-scale MD simulations.

: Personal opinion: just buy Sun multiprocessor sparc stations and try to get
: FDDI net between them, and write your code in C/g++ so it's easily ported to 
: other hardware later. I tried doing a distributed simulator on transputers 
: once - the tools where awful, it's much easier doing it on general purpose 
: hardware.

I tend to agree. Furthermore, use MPI for the communications. This results
in a program that will run on very different architectures, including
heterogeneous WS cluster, and dedicated MPP computers (Parsytec GC/PP, 
nCUBE, Paragon, ...). The latency is fairly low on WS, even if you don't have
a FDDI net, and on the nCUBE e.g. the routines map almost 1:1 to the native
communication calls.

I am currently writing a parallel molecular dynamics simulation on
a DEC Alpha WS-Cluster, which is to run on the GC/PP and the nCUBE3 also.
If you have further questions, please send an email.

Bye for now ...
                   Andy

-- 
Andreas Leibl / grad. student / Dept. of Physics, University of Ulm, Germany
Theoretical Physics I / Room O25 - 406 / Phone +49 731-502-2918
leibl@physik.uni-ulm.de             http://www.mathematik.uni-ulm.de/~atl/ 
atl@mathematik.uni-ulm.de           irc: croc (see you later, alligator :-))
-----------------------------------------------------------------------------
Wenn einer behauptet, er verstehe die Welt nicht mehr, so ist das reine
Koketterie - kein Mensch hat die Welt je verstanden.


