From: "Alain Coetmeur" <alain.coetmeur@icdc.caissedesdepots.fr>
Newsgroups: comp.parallel.pvm,comp.parallel.mpi
Subject: Re: MPI vs. PVM: What to choose?
Date: Wed, 17 Mar 1999 16:49:01 +0100
Organization: Informatique-CDC
Message-Id: <7coite$6e71@puligny.idt.cdc.fr>
References: <36EFAD85.EB4ABB4@postech.ac.kr>
Mime-Version: 1.0
Content-Type: text/plain; charset="iso-8859-1"
Content-Transfer-Encoding: 8bit
Xref: ukc comp.parallel.pvm:8137 comp.parallel.mpi:4771


Rob Lahaye a écrit dans le message <36EFAD85.EB4ABB4@postech.ac.kr>...
>Hi all,
>According to their analysis, PVM seems to perform better
>in case of different platforms, while MPI is better at handling
>many CPUs in one machine.
>I should choose for PVM in that case.
>
>I wonder whether that report is still up-to-date and people
>generally still agree with the conclusions.


I've read the manual of WMPI (on NT)
that is based on MPI-CH over p4
and support interoperability with
Unix version of MPI-CH.
so interoperability is
possible if the implementer takes care.
MPI-CH seems to support interoperability
if well configured.

but MPI1.1 does not offer dynamic
process creation, and this is
a problem on a network of workstation
where machine are not dedicated.

MPI2 address this problem, but nobody
implement MPI2 fully and I'm not sure
that MPI Process creation is available.

All I've read say PVM is slower that
tweaked MPI implementation on dedicated
platforms (Beowulf, SMP, or supercomputers).