Newsgroups: comp.parallel.mpi
From: smith@epcc.ed.ac.uk (A Gordon Smith)
Subject: Re: MPI on T3D
Keywords: t3d,mpi,psc
Organization: Edinburgh Parallel Computing Centre
Date: Sun, 9 Oct 1994 13:16:14 GMT
Message-ID: <CxEq73.Et9@dcs.ed.ac.uk>

In article <3746m4$in9@vixen.cso.uiuc.edu>, gjertsen@uxa.cso.uiuc.edu (Rob Gjertsen) writes:

> I was wondering if any T3D users at PSC have already installed either
> the OSU or ANL version of MPI that runs on top of the Cray version of
> PVM. I plan on doing it myself real soon, but if someone else has done it,
> then it seems silly to do it again.
> 

I'm not aware of any attempt to install the ANL/MSU MPI implementation
(MPICH) on the Cray T3D installed here at Edinburgh Parallel Computing
Centre, University of Edinburgh. My understanding of Ohio State's LAM
is that MPI and PVM interfaces lie at the same level, both implemented
on Trollius' network layer; I don't think that LAM can be installed
trivially on a Cray T3D, but you should confirm this with the LAM team.


> I'm not sure when Cray plans on coming out with a vendor version of MPI,
> but I doubt it will be in the near future.

We at Edinburgh Parallel Computing Centre are currently working on a
native implementation of the full MPI interface for the Cray T3D in 
collaboration with Cray Research U.K. and with support from Cray
Research in the U.S. In particular, we have been in close contact with
Peter Rigsbee of Cray Research in Eagan, who developed the existing PVM
implementation on the Cray T3D. We currently expect to complete
development and testing of native T3D MPI during December of this year.

For more information, contact us by e-mail:

   <t3dmpi-requests@epcc.ed.ac.uk>
   James G. Mills, T3D MPI Project Manager


> As a matter of fact, are there any vendors so far that have actually come
> out with an MPI implementation?
> The vendors so far seem to be catering to PVM and ignoring MPI,
> e.g., Power Challenge, Convex Exemplar, Cray T3D.
> 

Most supercomputer and workstation vendors are aware of MPI and seem
to take it seriously, judging by the attendance of an MPI Implementor's
workshop at Argonne National Lab. last month. I believe that Rusty Lusk of
ANL will report on the workshop at SPLC '94 this week. The report paper is
available through Argonne's WWW pages -- http://www.mcs.anl.gov/mpi


> Thanks,
> 
> Rob Gjertsen
> 
> -- 
> ----------------------------------------------------------------------
> Rob Gjertsen
> gjertsen@ncsa.uiuc.edu
> ----------------------------------------------------------------------


hope this helps,
-- 
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
  =-=- A. Gordon Smith -=-=- smith@epcc.ed.ac.uk -=-=- E.P.C.C.-=-=
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

