Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!wuarchive!emory!hubcap!anand
From: anand@elsa.top.cis.syr.edu (Rangachari Anand)
Newsgroups: comp.parallel
Subject: simple coarse-grained parallel programs
Message-ID: <12179@hubcap.clemson.edu>
Date: 10 Dec 90 18:21:21 GMT
Sender: fpst@hubcap.clemson.edu
Reply-To: anand@top.cis.syr.edu (Rangachari Anand)
Organization: CIS Dept., Syracuse University
Lines: 34
Approved: parallel@hubcap.clemson.edu

Using a group of workstations connected on a LAN  as a prallel
computer is an idea that is continually being rediscovered. Given the 
high speeds of the available workstations, this is clearly an Idea Whose 
Time Has Come. Indeed, I think that such an arrangement is arguably
at least the equal of dedicated parallel computers in speed for some
problems and definitely very cost effective. I get the feeling however
that there is considerable reluctance to work on coarse-grained
parallel programs as they appear to be too simple. In my view a
coarse-grained algorithm simply means that the number of 
possible methods implementation is simply greater.

Anyway, I wonder if there is a survey about packages developed 
for this task. More to the point, it would be interesting to know the 
speeds obtained for specific problems. Here are a few systems that I am 
aware of:

1. At Caltech they used a group of workstations as a virtual hypercube.

2. One implementation of Linda runs on a group of workstations.

3. There seems to be some work at Brown University in distributing
ray-tracing over workstations.

4. A package called AERO from Berkely. This was recently announced in
comp.os.research. This seems to be a very nicely put togethere system.

5. Over the summer, I wrote a package which combines ISIS and C++ to
simplify the creation of parallel programs. It is actually a compiler
similar to Sun's rpcgen and is available for anonymous ftp from 
mach1.npac.syr.edu

--
R. Anand
anand@top.cis.syr.edu