Date: Tue, 10 Nov 87 16:26:45 cst
From: Jack Dongarra <dongarra@anl-mcs.ARPA>
To: na.dis@score.stanford.edu
Subject: Gordon Bell Award

The Gordon Bell Awards Guidelines for Judges

Beginning in 1988, two $1000 awards will be given each year for 10
years to the person or team that demonstrates the greatest speedup on a
multiple-instruction, multiple-data parallel processor.

One award is for the most speedup on a general-purpose
(multiple-application) MIMD, the other for most speedup on a
special-purpose (single-application) MIMD.  Speedup can be accomplished
by hardware or software improvements, or by a combination of the two.

The awards are intended to recognize the best, operational scientific
or engineering program with the most speedup, not including
vectorization on a vector processor.

Speedup is measured against a similar program run sequentially on one
processor of the same system.  To be considered for the award,
submitted programs must have a factor of two more speedup than a
previous winning program.

The program should run at near the peak speed of any computer available
(including various supercomputers) and be a genuine, cost-effective 
solution; "toy" programs will not be considered.

Hardware simulations are not permitted.  The speedup must be
demonstrated on a running piece of hardware.


Definitions

Speedup:

The time taken to run an application on one processor using
the best sequential algorithm divided by the time to run the same
application on n processors using the best parallel algorithm.

General-purpose:

The parallel system should be able to run a wide variety of
applications.  For the purposes of this test, the machine must run
three different programs that demonstrate its ability to solve
different problems.  Suggestions are a large, dynamic structures
calculation; a transsonic fluid flow past a complex barrier; and a
large problem in econometric modeling.

Operational:

A program used to produce a useful scientific or engineering result.

Application:

The problems run for this test must be complete applications; no
computational kernels are allowed.  They must contain all input, data
transfers from host to parallel processors, and all output.  The
problems chosen should be the kind of job that a working scientist or
engineer would submit as a batch job to a large supercomputer.

Judging and Deadlines

Contestants must submit documentation of their base system and their
speedup by Dec. 1, 1987.  Documentation should be in the form of
verified timings, description of the application, and any other support
data.  Whether or not the application qualifies and verification of the
results will be determined by the panel of judges.

Judges will be selected by IEEE Software Editor-in-Chief Ted Lewis.
Winners will be announced in the March issue of IEEE Software.  Winners
will be notified in advance and may be asked to submit a short summary
of their program for publication.  

Send contest submissions to:  Ted Lewis, Editor-in-Chief, IEEE
Software, c/o Computer Science Dept., Oregon State University,
Corvallis, OR  97331.

The judges for the 1988 awards are Alan Karp (chairman) of IBM 
Research, Jack Dongarra of Argonne National Laboratories, and Ken
Kennedy of Rice University.