chk@purdue.edu (01/05/89)
[ Sent in answer to a request by me on another board. Thought it
might be interesting.
-- Steve.
]
There is a decent-size collection of benchmark programs at
the National Institute of Standards and Technology. Besides several
benchmark libraries (including pointers to the Argonne stuff) there is
an extensive bibliography on benchmarking and vector/parallel
computation. The following is the documentation I got from them:
WELCOME TO NISTLIB
Formerly NBSLIB
On August 23, 1988, the National Bureau of Standards became the
National Institute of Standards and Technology.
This collection of computer performance benchmark routines is or-
ganized into a number of subdirectories, primarily by contribu-
tor.
Your attention is directed to subdirectory "export" which con-
tains the benchmark routines recommended by NBSIR 88-3795,
"Benchmarks to Supplant Export 'FPDR' Calculations", by Bailey
(NASA-Ames), Brooks (LLNL), Dongarra (ANL), Hayes (LANL), Heath
(ORNL), and Lyon (NIST), June 1988. These are mostly carefully
selected well-known benchmarks which, as a group, are intended to
give a fair picture of the potential performance of a computer in
a number of application fields. This report may be obtained
from:
US Department of Commerce,
Office of Export Administration,
Room 4078, Herbert C. Hoover Building,
14th and Constitution Avenue, NW,
Washington, DC 20230.
Contact (301) 975-5681 if you have difficulty.
Accessing the nbslib at the National Bureau of Standards
(The nbslib is a collection of benchmarks, measurement data, and a
bibliography on benchmarking maintained at NBS as part of a project
on the performance measurement of parallel computers. The collection
exists to provide a pool of measurement data and benchmarks on more
or less 'large' computers. The pool grows by donation from the
interested public. The idea for an automatic, electronic mail-based
software distribution system, as well as the software to make it work,
came from Jack Dongarra at Argonne National Laboratory and Eric Grosse
at AT&T-Bell Labs. The idea for a collection of benchmarks to help
untangle the performance issues in parallel machines came from a group
of scientists who attended a workshop on parallel performance measure-
ment at NBS in June, 1985. Please send comments, criticisms, and
donations to ant@cmr.icst.nbs.gov)
Public domain benchmarks can be obtained free of charge via electronic
mail from
nbslib@cmr.icst.nbs.gov
For an index of the available software libraries send the line
send index
to the above address. Index for a single library is gotten by sending a
request of the form
send index for lanl
and individual programs by
send bmk11 from lanl
The requests wake up a server and the material is sent by return electronic
mail.
The nbslib service is meant for the individual user and does not send
entire libraries.
NOTE: We have changed our official hostname from <icst-cmr.arpa> to
<cmr.icst.nbs.gov>. The old address will continue to work for some
time, but will gradually be phased out.
Maybe you can find some useful material there; I know I did.
Chuck