Newsgroups: comp.lang.smalltalk
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!udel!gatech!swrinde!cs.utexas.edu!news.uta.edu!utafll.uta.edu!bruce
From: bruce@utafll.uta.edu (Bruce Samuelson)
Subject: "Object Magazine" Comparison--Digitalk outruns PPS et. al.
In-Reply-To: dekorte@symnet.net's message of 30 Oct 1994 00:11:38 GMT
Message-ID: <BRUCE.94Nov2205809@utafll.utafll.uta.edu>
Sender: news@news.uta.edu (USENET News System)
Nntp-Posting-Host: utafll.uta.edu
Organization: UTexas at Arlington, Linguistics
Date: Thu, 3 Nov 1994 02:58:09 GMT
Lines: 40

>> Can anyone give me a hint about the conclusions?
 
> "the ParcPlace tool set is easily the best of it's class"
> Kyle Brown, Sizing Up the SmallTalks, Object Magizine Oct 94

That's not an adequate summary. The reviewers tried to give a balanced
discussion and managed to find some area in which each product was
best. I would have liked to see more negatives discussed. The
positives were fairly evenly distributed.

One point that really caught my attention was that, running my early
1993 Smopstone (Smalltalk Medium OPeration Stone) benchmarks, Digitalk
for Win 32 was the clear winner, surpassing ParcPlace for Win 32 by
factors of 1-2 depending on benchmark. Quasar was just ok, after
taking into account the slower Mac hardware it was run on. IBM for
OS/2 was very slow (because of not yet having a machine code cache?).
Easel Enfin was so slow it was almost off the chart. Those are back
end benchmarks that don't test GUI or database access.

When I originally ran these benchmarks, ParcPlace was the winner, so
Digitalk has made a lot of progress in the last year or so.
Incidentally, on my Windows 3.1 machine, VW 2.0 benchmarks at the same
speed as 1.0. ParcPlace has been adding database features rather than
speed lately.  As I've written elsewhere, my benchmarks have many
flaws, and Smalltalk badly needs something better like C's SpecInt and
SpecFloat, although Spec marks aren't very indicative of real
performance either.


Incidentally, the Intel benchmarks were run on a 486/50, not a 486/33
as reported in the article. Kyle told me that the proofs were
finalized before he could correct the MHz from 33 to 50. That's why
the ParcPlace numbers were 1.4. They should be 1.0 on a 486/33.
--
***************************************************************
* Bruce Samuelson   bruce@utafll.uta.edu                      *
* Wycliffe Bible Translators; Summer Institute of Linguistics *
* Linguistics Program, University of Texas at Arlington       *
***************************************************************

