Newsgroups: comp.parallel.mpi
From: eddemain@neumann.uwaterloo.ca (Erik Demaine)
Subject: Re: performance modeling for MPI
Organization: University of Waterloo
Date: Thu, 30 Nov 1995 22:08:30 GMT
Message-ID: <DIvMu6.Iz@undergrad.math.uwaterloo.ca>

Ying Chen (ying@cs.uiuc.edu) wrote:
: I'm wondering if there is any work that has been done 
: on performance modeling of MPI message passing library. 
: Any pointers, references, or comments would be great!

I'm doing some work in this area, in particular, when running an MPI program
on a uniprocessor workstation.  From this it is possible to extract parallel
performance of a program (even though it's not getting any parallel benifits)
given some information like the speed and topology of the network.  Initial
results for full matrix Cholesky factorization and merge-sort algorithms are
very promising, even for fairly fine-grain problems.

Feel free to ask questions.  This work is currently in progress.
An initial release may be possible soon.

Erik
--
Erik Demaine                 \/  e-mail: eddemain@neumann.uwaterloo.ca
Dept. of Computer Science    /\  URL   : http://barrow.uwaterloo.ca/~eddemain/
University of Waterloo       \/  Most jugglers do poorly while drunk, especially
Waterloo, ON Canada N2L 3G1  /\  in a large wavelength gravity field.

