Newsgroups: comp.parallel.mpi
From: rdaoud@magnus.acs.ohio-state.edu (Raja B Daoud)
Subject: Re: Performance Issues in MPI
Organization: Ohio Supercomputer Center
Date: 30 May 1995 15:23:15 GMT
Message-ID: <3qfd93$4ff@charm.magnus.acs.ohio-state.edu>

Satya Ranjan Dash <srd320@cs.usask.ca> wrote:
>Clearly Satya Dash goofed.  I was surprised though no one blasted me for
>such ridiculous figures.  MPI-LAM is giving excellent communication
>overheads with 0.1 seconds for 1 Etherenet packet and 2 seconds for 8
>packetes (4096 double precision data).

These numbers, though not ridiculous, are still wrong.  "tping" is not a
timing tool, it's too crude.  Correct timing of an MPI ping pong of 1024
doubles (8K) gives a result in the 15-25 millisecond range on LAM 5.2
(depending on machines and network used).

--Raja

-=-
Raja Daoud				raja@tbag.osc.edu
Ohio Supercomputer Center		http://www.osc.edu/lam.html
-=-
MPI users conference, June 22-23 @ Notre Dame

