Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!zaphod.mps.ohio-state.edu!rpi!dali.cs.montana.edu!ogicse!pturner
From: pturner@ese.ese.ogi.edu (Paul Turner)
Newsgroups: comp.unix.aix
Subject: distributed computing on a network of RS6000s?
Keywords: Distributed computing, Parallel processing
Message-ID: <13997@ogicse.ogi.edu>
Date: 14 Nov 90 21:33:55 GMT
Sender: news@ogicse.ogi.edu
Organization: Oregon Graduate Institute - Department of Env. Science and Eng.
Lines: 29


It doesn't take a rocket scientist to figure out that when you 
have 10-20 RS6000s you are well on your way to having a
supercomputer. Given that a bare 320 sells for app. $8k means
a 10 unit setup has a 400 MFLOP theoretical peak, 80 Mb of memory,
1.2 Gb disk, etc. All for $80k. Compared with a similarly
configured parallel machine of your choice, I would imagine it is
cheaper to buy RS6000s - not to mention that the RS6000s
would have other uses. The recent article by Dr. Ron A. Bell,
"IBM RISC System/6000 Performance Tuning for Numerically
Intensive FORTRAN and C Programs," indicates that a substantial 
portion of the theoretical peak can be realized in specific applications.

The obvious problems are a programming interface, load balancing,
and communication bandwidth.

My questions, then, are along the lines of software to make
use of this horsepower, Linda or others, and something other
than ethernet to improve communication. Any ideas welcome.

BTW, does anyone have performance figures for the 550?

I'll summarize any email I receive to the net.

Thank you,

--Paul
Paul J. Turner
pturner@ese.ogi.edu