Dr. Forgy has just about finished up his new algorithm - there's always some last minute thing that needs to be done. While testing his new TECH algorithm on my variousl platforms I have found it to be about an order of magnitude faster then Rete 2 or Rete III on both my MacIntosh, Core2Duo Unix environment and the Windows 64-bit Dell i7. On the i7 (that is effectively 8 CPUs) it scales nicely and provides even faster performance. I can't wait to get my 8-core Mac Pro (the Nehalem 2x4 CPU) and see how it runs there. Dr. Forgy has agreed to discuss the new algorithm after hours during Pub Nights in the Walt Garrison restaurant and bar that is located in The Adolphus Hotel.
OH! You wanted numbers? OK, the only comparison that I have right now is OPSJ versus TECH but here is what I have so far - lots more to come later and I'll publish it all at ORF. We had to use a new benchmark because TECH runs so blooming fast that it's hard to compare the difference in 1.2 seconds and 0.57 seconds as anything meaningful when running WaltzDB-16. OPSJ runs that one 2.7 seconds on the Core2Duo Mac and in 0.433 seconds on the Dell i7. TECH is not really measurable.
WaltzDB-200 (New Benchmark for 2009) - in seconds
Core2Duo Mac, Leopard, 3GB RAM:
Dell i7 64-bit Vista, 6GB RAM
So, that's pretty much a 10:1 (or thereabouts) speed improvement firing 246,233 rules. Remember, OPSJ is the underlying engine in Java that incorporates Rete 2 / Rete III, the fastest Java rulebase on the planet Earth. Until now, that is. Also, remember that I have NOT finished the full comparisons checkout and double-checked everything to be sure that they are the same and firing the same rules and doing everything. But, so far, it looks pretty fantastic. As soon as I get a chance I'll do the other BRMS and rulebased systems including Drools, Blaze Advisor, Jess, CLIPS and (possibly) JRules.
Depending on when his patents are approved and finalized, Dr. Forgy will be happy to talk with you at ORF about this ground-breaking technology.