Path: utzoo!mnetor!tmsoft!torsqnt!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!wuarchive!uunet!mcsun!unido!uklirb!shell
From: acha@CS.CMU.EDU (Anurag Acharya)
Newsgroups: comp.ai.shells
Subject: Re: shells & Windows
Message-ID: <7533@uklirb.informatik.uni-kl.de>
Date: 8 Feb 91 15:09:20 GMT
References: <7505@uklirb.informatik.uni-kl.de>
Sender: shell@uklirb.informatik.uni-kl.de
Organization: Carnegie Mellon University
Lines: 21
Approved: shell@dfki.uni-kl.de
Posted-Date: Thu Feb 14 08:56:00 GMT 1991
In-reply-to: carasso@aludra.usc.edu's message of 6 Feb 91 16:10:45 GMT

In article <7510@uklirb.informatik.uni-kl.de> carasso@aludra.usc.edu (Roger RDC Carasso) writes:
   ART-IM is probably the fastest expert
   system for real systems.  Most of the others are toys in that they
   don't scale up well in performance.  

Are there any benchmarks available that provide evidence along those lines ?
Ditto for the scalability. If you are looking for raw speed, you might wish
to take a look at OPS83. Lanny Forgy (clf@cs.cmu.edu) will be willing to 
provide you with all the information  you want. 

   ART-IM's speed is independent to
   the number of rules (dependent on the number of unique patterns),.....

True for *all* systems that use the Rete algorithm to perform the match.
Rete shares the common tests in an directed acyclic graph framework.
In practice, however, only a test is shared by only a small number of 
productions. therefore, the match time ends up being proportional to the 
number of the rules albiet with a smaller coefficient. I would be greatly
(and happily) surprised if it was otherwise with ART-IM.

anurag