Newsgroups: comp.parallel
From: Christopher Marshall <chm@ahpca.hpc.org>
Subject: [Q] scaLAPACK for meta-computing
Organization: Maden Tech Consulting
Date: 29 May 1995 15:32:59 GMT
Message-ID: <3qcpfb$23l@usenet.srv.cis.pitt.edu>

Hi-

I would like to use multiple HPC platforms (vector supercomputers,
MPPs...) to compute the LU factorization and backsolve of a large
dense linear system.  I would like to hear about other experiences
with this type of meta-computing strategy.

An implementation being considered would be to build upon the
scaLAPACK routines with modifications to the underlying BLACS layers.
One aspect of concern is to reduce the communications requirements
between the HPC platforms since the performance of those links can be
orders of magnitude less than the internal memory and communications
bandwidths. 

Pointers to other research approaches in this area or alternative
strategies would be excellent.  If interest is shown (via e-mail), I
will post a summary of the replies.

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

