Newsgroups: comp.parallel.pvm
From: darren@cs.curtin.edu.au (Darren Brown)
Subject: Load Balancing - PVM ? MPI ? p4 ? LAM ?
Summary: Questions regarding suitability of software systems for algorithm imp.
Keywords: load balancing PVM MPI p4 scheduling
Organization: Curtin University of Technology
Date: 18 Jul 1994 10:53:53 GMT
Message-ID: <30dn01$pe@info.curtin.edu.au>


Hi everyone,

I posted some time ago and made a general enquiry regarding load
balancing and PVM.

I am now in a position to do the work I was considering, which is...

Implementation of a particular dynamic and decentralised load balancing
algorithm suitable for a distributed memory multiprocessor or
multicomputer.

As we do not possess such a machine, but do have an overabundance of
workstations (I guess we are not alone in this), I intend to use some of them
(ethernetted SGI Indys) and a software library (to provide the concurrency
support). Given that I wish to stay as close to the multiprocessor/
multicomputer paradigm as possible, my questions are:

1: Which of the software libraries is most suitable (PVM, p4(5?), MPI,
   LAM, Linda, etc...) and why? PVM appears to be the standard now, but
   what about MPI (is it stable enough?). Also, I have read that p4 is more
   efficient - true?

2: I have heard of Condor, DQS and QM, but has other work been done in this
   area? Is anyone doing anything similar currently?

3: Obviously the network environment necessitates larger task granularity,
   can anyone recommend a suitable application for experimentation? Initially
   I require a simple application whose sub-tasks do not have precedence or
   communication relationships eg raytracing.

Please feel free to any of the above, either directly to me or to the
newsgroup. I will compile replies and post them (if there is enough!)

Yours sincerely,

Darren Brown
-----------------------
darren@cs.curtin.edu.au
Postgraduate research student @
Curtin University
Perth, Western Australia

