Path: utzoo!mnetor!uunet!lll-winken!lll-tis!ames!ncar!noao!stsci!sims
From: sims@stsci.EDU (Jim Sims)
Newsgroups: comp.lang.lisp
Subject: Re: Gabriel benchmark results
Message-ID: <283@mithras>
Date: 14 Apr 88 19:51:34 GMT
References: <1327@PT.CS.CMU.EDU> <7799@alice.UUCP>
Organization: Space Telescope Science Institute, Baltimore, MD 21218
Lines: 42
Keywords: LISP, Gabriel
Summary: benchmarks?????
	 sounds more like bench racing to me

I've seen the Gabriel benchmarks and I've seen the results on several machines
I've also seen tests of applications run on multiple machines (see below).
The ONLY conclusion you can draw from ANY - repeat ANY benchmark of this sort
is a ballpark figure. Gabriel would tell you this, I believe. The problem
I see with the Gabriel LISP benchmarks is that they don't Garbage Collect or
Page significantly (does your code??).


Application - Scheduling Hubble Space Telescope

machine        notes                          relative rating
-------------------------------------------------------------
TI Exploder     plain jane			1.0
TI Exploder II  new box				3.5
Slimebolics	36xx?				3.0
MAC II		Allegro LISP			.75
MAC II + LISP CHIP   hah! fooled yah - it crashed and burned when we tried to
			do a directory on the LISP side of the fence


(following are unsubstantiated "rumors")

Application - Expert Systems Tools

machine		notes				relative rating
---------------------------------------------------------------
vaxstation xxxx					3
Slimebolics 36xx				3
TI Exploder II					3.5
SUNN 3/260					3

Basically all this stuff says is that the APPLICATION makes or breaks the
system, and if you want to do benchmarks, run the code you plan to use -
real applications - not what someone else thinks is a godd "general purpose"
mixture of example code.


						happy LISPing
-- 
    Jim Sims     Space Telescope Science Institute  Baltimore, MD 21218
    UUCP:  {arizona,decvax,hao,ihnp4}!noao!stsci!sims
    SPAN:  {SCIVAX,KEPLER}::SIMS
    ARPA:  sims@stsci.edu