Tuesday, November 14, 2006

Parallel Rulebase Systems

Whether you're doing a small rulebase (say, 200 rules or less), a medium (1K rules) or an enterprise system (20K rules) you should, as a Knowledge Engineer, be aware of all of those giants who have gone before; meaning that you should do some research into the mists of time. At a recent Business Rules Forum I heard several vendors say that they were investigating the problems associated with using parallel rulebased systems.

Well, a team of guys at CMU (Gupta, Forgy, Newell and Wedig) started solving this problem in 1984. Anoop Gupta (along with Charles Forgy and Allen Newell) finished that investigation in 1989 after Anoop got his Ph.D. and moved to Stanford. I guess that if you read the paper and put it in terms of today's computers, you "could" say the problem was solved. Gupta et al said that they could only get 10:1 improvement using parallel systems, but that seems good enough to at least start down that path, don't you?

SDG
Yaakov
Copyright (c) KBSC

1 comment:

Bubba De Katt said...

I have an agent system using multiple threads so that I have one rules engine that communicates with 21 other agents and each has its own rules engine. Each of these rules engines can run a multitude of systems, e.g. Fuzzy Logic, Neural networks, etc. in an implicitly parallel fashion.

All of this run in the same execution space as the original rules engine.

All of that being said, it might be interesting to someday see each of the 21 (rules) agents running Miss Manners in Dr. Forgy's Parallel OPSJ environment.