Re: [not about] gc-by-area and host uptimes

[Keith Price <PRICE%DWORKIN.usc.edu@OBERON.USC.EDU>]

Lisp benchmarks are a funny thing.  I have run Lisp on teh Symbolics,
Explorer (and II), Sun 3/2xx, 4/2xx (Lucid and KCL).  I have numbers
that indicate that the Sun 4 is anywhere from 1/10 the speed of the symbolics
to twice the speed.  (Also numbers that give the Sun4 as equivalent to the
Sun 3 or better.)  These range from tests run to illustrate points in
the class I teach to large collections of programs.  My general conclusion
is that if your Lisp is classical numerical processing then the Sun wins
(even a Sun 3), if you hit the GC limits then the Sun starts to lose.
If you exercise many of the features of Lisp that are not common to all
languages then the Suns really lose.  In all of the tests the Symbolics was
at a disadvantage since it was being charged for other system overhead
(network processing, etc.) and it has much less memory than the Suns
(which never paged on these jobs).   (Essentially the Symbolics I run on
costs less than some of the Suns, the 3/60 with KCL was so much slower
that it can't count, and generally performs better.  It certainly degrades
better when running several (i.e. greater than 1) real processes, but
that is a different story.)  The Sun is a fast box, but it is easy to
push over its limit and then it is a disaster.
	Keith Price.
-------