Newsgroups: comp.parallel
From: Christopher Marshall <chm@ahpca.hpc.org>
Subject: scaLAPACK for multiplatform LU solve
Organization: Maden Tech Consulting, Inc.
Date: 29 May 1995 15:25:16 GMT
Message-ID: <3qcp0s$22d@usenet.srv.cis.pitt.edu>

     We would like to solve a large, dense linear system by spreading
the work across multiple HPC platforms.  A framework being considered
is to use scaLAPACK with an implementation of the BLACS across the
multiple platform.  Given the relatively slow communications between
the separate HPC platforms (relative to the internal bandwidths) this
would only make sense if there were a way to partition the problem to
reduce the communications required between the HPC platforms even at
the expense of additional computations and communication within each
platform.  Pointers to related work, information on BLACS usage in
scaLAPACK, or other suggestions are welcome.  If there is interest,
I would be glad to post a summary of the responses received.

Sincerely yours,
Christopher Marshall
chm@ahpca.hpc.org

