Gabriel Benchmarks

[boyer@froggy.ics.utexas.edu (Bob Boyer)]

It should be clear from the remarks of Jonl and Schelter that although
the ``Gabriel'' times I sent out may provide some interesting evidence
comparing the speed of three Lisps (identical tests were run on three
different Lisps on a lightly loaded machine about the same time of
day), nevertheless the times presented should not be taken absolutely,
only comparatively.  That is, those times do not reflect the best
times that any of those Lisps may have on the official Gabriel
benchmarks.  For one thing, some of the official tests were run
several times or grouped together; this was not made evident in my
note; mea culpa.  For another thing, turning off other users,
networks, screens, interrupts, and so forth always seems to improve
best times.  (Often unrealistically in my experience; but some
manufacturers do these things to get their best times and then others
are more or less forced to do similar things.)  Finally, different
Lisps do better with different Common Lisp declarations.

I continue to believe that running those tests (available by anonymous
ftp as the file rascal.ics.utexas.edu:/pub/gabriel.tar.Z) to time
various Lisps on a single, given machine will provide interesting
comparative performance statistics for anyone wondering whether to use
KCL.  I think that the Gabriel benchmarks still provide an excellent
picture of how good a Lisp's performance is, even after years of
manufacturers knowing that they need to do their best on those
benchmarks.  If I receive tests deemed more fair to alternatives to
KCL, I certainly will try to incorporate them.

Bob