Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!ames!ncar!gatech!hubcap!stunkel
From: stunkel@uicsgva.csg.uiuc.edu (Craig Stunkel)
Newsgroups: comp.parallel
Subject: need iPSC/2 applications for tracing
Message-ID: <6823@hubcap.clemson.edu>
Date: 19 Oct 89 16:18:53 GMT
Sender: fpst@hubcap.clemson.edu
Lines: 33
Approved: parallel@hubcap.clemson.edu

As part of my thesis work, I am collecting memory addresses traces for
executions of programs on the Intel iPSC/2 hypercube.  I am looking for
more application programs for this purpose.  My main interest is in
numerical programs, but I may be interested in expanding this scope.

For me to be able to collect traces, the node program executions should
conform to the following:

(1) exit normally (not just get killed by the host).
(2) I can figure out how to compile it and run it within my lifetime.
(3) A realistic (not toy) problem must be able to fit in 3 Meg or so even
	when executed on one node, so that I can collect traces across
	all dimensions (we have a 16 node machine) of the iPSC/2.
(4) A realistic (not toy) problem must be able to be solved in about
	3 minutes or less when running on a single node.  I need this because
	my tracing and cache performance analysis can slow the processor
	down by about x 50.

If a program cannot satisfy (3) or (4) for 1 node but would do so if I
	started with 2 or 4 nodes instead of 1 node, I'm also very interested.

Also, if a program "fits" on one node but simply runs too long for me,
that's not too bad of a problem: I can either run less iterations if it's
iterative, or collect only the first 500 Meg memory addresses or so that are
generated.

I will naturally reference any publications concerning your programs
and acknowledge that I got the progs from you !!

Thanks much in advance for any help you can give me !

Craig Stunkel				Coordinated Science Lab
stunkel@bach.csg.uiuc.edu		University of Illinois