Newsgroups: comp.parallel.mpi
From: Michael Pernice <usimap>
Subject: Re: Q:  Performance of collective communications?
Organization: Utah Supercomputing Institute
Date: 25 Sep 1996 17:50:44 GMT
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <52brdk$jmm@news.cc.utah.edu>

Take a look at IEEE Parallel & Distributed Technology, (4)1, Spring 1996.
It contains an article by Xu and Hwang that models communication on a
large IBM SP2.  It is one of the few 'communication performance evaluation'
articles I've seen that evaluates collective as well as point-to-point
communication operations.

BTW that article shows poor latency characteristics for gather/scatter and
total exchange for the MPICH implementation.  Is this being addressed?
-- 
_______________________________________________________________________________

Michael Pernice                          Center for High Performance Computing 
Phone:  (801) 581-7708                   85 Student Services Building  
Fax:    (801) 585-5366                   University of Utah
E-mail: usimap@sneffels.usi.utah.edu     Salt Lake City, Utah 84112
_______________________________________________________________________________


