Newsgroups: comp.parallel.pvm
From: cussans@desy.de (David Cussans)
Subject: Re: PVM on multi-processor SGI
Organization: Rutherford Appleton Lab
Date: Thu, 11 Apr 1996 16:48:48 +0200
Message-ID: <cussans-1104961648480001@cussans.desy.de>

In article <4kgvip$d7f@post.gsfc.nasa.gov>, haferman@audry.gsfc.nasa.gov
(Jeff Haferman) wrote:

> I have access to an SGI with 6 processors.  My PVM application 
> actually only needs 4 of the processors.  How do I start up a
> pvmd on the individual processors on a single machine?
> 

   Forgive me if this a (really) stupid question. Why not have just one
pvmd and use it to start up the four tasks?

   I am unsure about how IRIX distributes tasks over processors, but this
method would probably also give you better load-balancing than if you
tried to do it yourself.

   I have run job to simulate high energy electron-proton interactions for
Monte-Carlo studies on a network of ten SGI-Indigo's and a 16 processor
SGI-Challenge machine. The single pvmd on the SGI-Challenge seemed to cope
with the sixteen tasks and the tasks were well distributed over the
processors.

   Admitedly this was an "embarassingly parallel" application with
multiple copies of the same program running.

-- 

   David Cussans
   Experiment H1
   Deutsches Elektronen-Synchrontron
   22603 Hamburg, Germany
   E-mail: cussans@desy.de , d.g.cussans@rl.ac.uk 
   Telephone: {49} (40) 8998 2911

--------------------------------------------------------------------------

