Ernest:
Would you be so kind as to publish the stats on the tests that you ran?
Machine, speed, RAM, RAM available to JVM, number of rules, time, etc., for
both CLIPS and Jess? I, for one, would find this most interesting.
Ernest Friedman-Hill wrote:
> Modules are something that has been discussed in the past. I'm
> certainly interested in supporting them, but it hasn't been a priority
> so far (I can only do so much :) )
>
> There are two levels of module support one could envision. One would
> be, as one correspondent wondered, just syntactic. Nothing really
> significant would change in the rule engine, except that only rules
> from the current module would be fired. Changing focus would just mean
> telling the engine what name prefix a rule would have to have to fire.
> This conveys one benefit of modules (the software engineering benefit
> of a modularized system), but not the other one, efficiency.
>
> In full module support, the engine would do only computations that
> affected the current module; this could be a significant efficiency
> concern for a large system. Conflict resolution would be made more
> efficient because each module would have its own separate agenda, and
> LHS execution would be more efficient since nodes from non-focused
> modules would not need to be kept up to date (you could update them
> during the 'focus' process.)
>
> In any case, the related concept of multiple engines in one JVM using
> explicit message passing is viable; a number of users have reported
> doing this with success.
>
> I would caution anyone who might be feeling ambitious that this is NOT
> a good time to sit down and pound out these mods yourself, in the
> hopes of contributing them to the greater good. The internal changes
> in the engine between 5.0a4 and 5.0a5 are DRASTIC. The 5.0a3 to 5.0a4
> changes inspired me to do similar housecleaning all throughout the
> code, and the results have been impressive. Jess can now run the
> University of Texas benchmarks (using a stock Microsoft JVM)
> noticeably faster than CLIPS 6.04 (compiled with -O2) on the same
> machine. The speedup scales nonlinearly with problem size, and is most
> apparent for very large systems. For the "Manners" benchmark, Jess now
> beats CLIPS by 25% at 64 persons, and by a factor of two at 80
> persons. I'm using a new smart indexing system in the Rete memories to
> get this speed.
>
> In any case, as usual, the goal is to make Jess leaner and meaner. The
> simpler I can make the innards, the easier itwill be to add features
> like modules.
>
> ---------------------------------------------------------
> Ernest Friedman-Hill
> Distributed Systems Research Phone: (925) 294-2154
> Sandia National Labs FAX: (925) 294-2234
> Org. 8920, MS 9214 [EMAIL PROTECTED]
> PO Box 969 http://herzberg.ca.sandia.gov
> Livermore, CA 94550
>
> ---------------------------------------------------------------------
> To unsubscribe, send the words 'unsubscribe jess-users [EMAIL PROTECTED]'
> in the BODY of a message to [EMAIL PROTECTED], NOT to the
> list. List problems? Notify [EMAIL PROTECTED]
> ---------------------------------------------------------------------
--
ttfn
IHN
Jim
---------------------------------
James C. Owen
Knowledge-Based Systems Corporation
4817 Buckskin Drive
Fort Worth, Texas 76137
817.314.0584 Office
817.314.0585 FAX
817.247.8976 Cellular
mailto:[EMAIL PROTECTED]
http://www.kbsc.com
"Keep Using My Name In Vain And I'll Make Rush Hour Longer. -God"
(Seen on a billboard in DFW MetroPlex.)
---------------------------------------------------------------------
To unsubscribe, send the words 'unsubscribe jess-users [EMAIL PROTECTED]'
in the BODY of a message to [EMAIL PROTECTED], NOT to the
list. List problems? Notify [EMAIL PROTECTED]
---------------------------------------------------------------------
