Thursday, March 12, 2009

RETE Topology Cost Function and dynamic typing

I thought the general public might find this paper interesting. A few weeks back I posted a paper on duck and dynamic typing RETE, which tries to explore some of the challenges of supporting duck/dynamic typing.

The first half of the paper covers the typing issue. The second half explains a topology cost function. I've been using this approach for several years, but didn't bother to formalize it until now. A little bit of history on the topology cost function. Back in 2001-2003 Said Tabet and I were asked by numerous people to quantify and qualify RETE performance. One of the things we did was to compile some rules in JESS and show which nodes would be visited if a set of facts are asserted. We did this manually at the time, since JESS doesn't provide topology cost function out of the box.

I was inspired to write this stuff down when Johan Lindberg asked for more clarification on duck type RETE performance. Thanks to Johan and Joe for their invaluable assistance revising the paper. The paper is available here.

http://jamocha.svn.sourceforge.net/viewvc/jamocha/morendo/doc/dynamic_typing.pdf?revision=2486