[comp.parallel] survey of production parallel programs

mizell%venera.isi.edu@usc.edu (David Mizell) (08/19/89)

Dear Colleagues:

I am conducting a survey aimed at collecting information about
parallel application programs that are receiving routine, production
use on parallel computer systems.  I call the collection the "Blue
Collar List," to signify that these are the parallel programs that
earn their keep (or perhaps that of the multiprocessors on which they
are executed) through "daily labor:" regularly performing computations
that are in some way integral to the basic purpose of the using
organization.  This is the first iteration of what I intend to be an
annual survey, continued until the number of such applications becomes
unmanageably large.  Indications are that there are not very many such
programs now, but it should be interesting to track the growth of the
list for the next few years and watch for significant trends with
respect to application domain, architecture, programming language,
etc.

I am writing to ask that you let me know about any such programs of
which you are aware.  If you can put me in contact with a person
familiar with the development aspects of a production parallel
program, I will send him or her (by either e-mail or regular mail) a
short questionnaire to fill out and return, that asks for a brief
description of the program, the application domain, what
multiprocessor it runs on, the implementation language, etc.  Some of
you may have already seen the questionnaire; I sent it to the
marketing directors of all of the U.S. parallel computer manufacturers
of which I am aware, asking them to pass copies on to their software
developers or their customers for whom it would be appropriate.

To be as specific as possible, I'm looking for application programs
that are run on a commercially available, programmable parallel
computer system for the sake of increasing execution speed via
parallelism (as opposed to sequential programs that may just happen to
execute on a multiprocessor), and that have passed the
experimental/developmental/testing stages and are receiving regular
production use.  Either explicitly parallel programs or those for
which the parallelism was automatically discovered by the compiler may
qualify.  Strong evidence that a program is receiving production use
would be that the program is being used regularly by a group other
than the one that developed it, especially if the users don't
particularly care that it's a parallel program, but only that it meets
their performance (or cost-performance) requirements.

If you know of a candidate for the "Blue Collar List," please put me
in touch with a knowledgeable person by the end of August, so that
I'll have time to get them a questionnaire and possibly ask some
follow-up questions.  I hope to publish the results of this survey at
the "Computers in Aerospace" conference in Monterey in early October,
as my contribution to a scheduled panel discussion on the topic of
whether parallel computers are "worth the hassle."

Thanks in advance for your help.

David Mizell

USC Information Sciences Institute
4676 Admiralty Way
Marina del Rey, CA 90292-6695

mizell@isi.edu