[comp.ai] estimating performance of rule based / knowledge based system

jkimball@SRC.Honeywell.COM (John Kimball) (02/28/90)

We're evaluating some ideas about how to get a software development
environment to behave more "intelligently," through enhancing its data
model and adding some inference/rule-processing capability to individual
tools and to the user shell.  One criterion in evaluating these ideas is
performance.

I need some info on how to estimate the performance of a hypothetical
rule-based / knowledge-based system.  I welcome pointers, references to
papers/books, or off-the-cuff remarks.

o Are there any books or papers which discuss how to design
  rule-based/knowledge-based systems such that the performance
  characteristics are good?

o What factors influence the performance of such a system?  (EG language
  choice, size of rule base, etc, etc?)

o What things are Really Bad Ideas -- a novice might think of trying it,
  but it would become really expensive?

o Where could I find a discussion of the algorithmic complexity of the
  various features provided by systems for rule-based programming?

Please reply by mail and I'll summarize if indicated.

Thanks!
						John Kimball

domain: jkimball@src.honeywell.com        Honeywell Systems and Research Center
        postmaster@src.honeywell.com      Computer Sciences/Software Technology
uucp: <any-smart-host>!srcsip!jkimball    3660 Technology Drive, MN65-2100
voice: 612/782-7343  fax: 612/782-7438    Minneapolis, MN  55418-1006
disclaimer: I said *what*???