Indeed. Either approach has it's advantages. The "encode an agent's id into the rule conditions" was exactly how I imagined agents in a shared rete would be differentiated. This has the advantage of scaling well via rete node sharing if the agents are substantially similar. Also communication between agents, in a unicast but especially multicast fashion, is natural in the shared rete approach: just deposit an object into working memory that multiple agent's rules match upon.
The separate rete approach ultimately scales better, though. And if the agents' rules don't all change in a coordinated fashion, it's a must. (I'm pretty sure drools and other engines have to halt processing while rules are added/removed.) And even though drools does have a multi-threaded execution mode, there's no guarantee that multiple agents in the same rete would take advantage of it well, and of course the one agent per thread model would. Also, there's no reason why agents should execute on the same JVM, so mattering on the inter-agent communication requirements it could be embarrassingly parallel. Use some clustering technology like Terracotta and agents wouldn't even know they were executing on separate JVMs. Sounds like a lot of fun. :) --- On Sun, 8/23/09, Mauricio Salatino <[email protected]> wrote: > From: Mauricio Salatino <[email protected]> > Subject: Re: [rules-users] Using Drools for modeling with rule-driven agents > To: [email protected], "Rules Users List" <[email protected]> > Cc: "Greg Barton" <[email protected]>, "HusseinAbbass" > <[email protected]>, "Antony W. Iorio" <[email protected]> > Date: Sunday, August 23, 2009, 12:16 AM > Seems that each agent will have a session > created based on different knowledge that can be dinamically > modified. > > 2009/8/22 Russ Abbott <[email protected]> > > Thanks Greg, That sounds quite > promising. I've never used Drools and can't say > how the Rete network will be organized. In fact, before your > message I wasn't aware that Drools allows multiple Rete > networks. If that's the case it seems to make a lot of > sense for each agent to have its own. Since the agents are > supposed to be individual and disjoint I was imagining that > I'd have to code an agent's id into its rules to > limit those rules from being applied to other agents. But > multiple Rete networks sounds much much better! It would > be even better if each agent could run its rules > asynchronously in its own thread. Is something like that > possible: multiple threads of rule evaluators running on a > single workspace? > > > > > -- Russ > > > > > On Sat, Aug 22, 2009 at 9:26 PM, > Greg Barton <[email protected]> > wrote: > > > > Yes, rules can be manipulated at runtime. The class > org.drools.RuleBase has methods for adding Packages and > removing Packages and individual rules, and can be called at > runtime as demonstrated by Mark in this jira post: > > > > > > > https://jira.jboss.org/jira/browse/JBRULES-1431 > > > > I haven't tried it myself, but now that the subject has > come up I probably will. :) > > > > As for the best approach to do this, I'd suggest that > you have your own rule representation that generates DRL, > then compile that DRL in the standard way using the > PackageBuilder. On first blush it may seem best or more > elegant to try creating Rule objects directly, circumventing > the PackageBuilder, but that's a waste of time in my > opinion. (The main focus for support from the drools team is > the language and other "external" parts of the > system.) > > > > > > > So, how will agents interact in your system? Will each > have it's own rete network? Or will several agents > share a rete? > > > > --- On Sat, 8/22/09, Russ Abbott <[email protected]> > wrote: > > > > > From: Russ Abbott <[email protected]> > > > Subject: Re: [rules-users] Using Drools for > modeling with rule-driven agents > > > To: "Greg Barton" <[email protected]> > > > Cc: [email protected], > "Antony W. Iorio" <[email protected]>, > "HusseinAbbass" <[email protected]> > > > > > > Date: Saturday, August 22, 2009, 10:15 PM > > > That's a possibility. But > the > > > underlying capability comes first. Can we do it in > Drools? > > > Has anyone done it, and if so would you be willing to > share > > > your experience? > > > > > > -- Russ > > > > > > > > > > > > On Sat, Aug 22, 2009 at 6:50 PM, > > > Greg Barton <[email protected]> > > > wrote: > > > > > > > > > This wouldn't happen to involve evolutionary > pressures > > > altering rules, would it? > > > > > > > > > > > > --- On Sat, 8/22/09, Russ Abbott <[email protected]> > > > wrote: > > > > > > > > > > > > > From: Russ Abbott <[email protected]> > > > > > > > Subject: [rules-users] Using Drools for modeling > with > > > rule-driven agents > > > > > > > To: [email protected] > > > > > > > Cc: "Antony W. Iorio" <[email protected]>, > > > "Abbass, Hussein" <[email protected]> > > > > > > > Date: Saturday, August 22, 2009, 5:52 PM > > > > > > > > > > > > > > Hi, > > > > > > > > > > > > > > I'm interested in developing (if it > doesn't > > > already > > > > > > > exist) and then using a framework for > agent-based > > > modeling > > > > > > > in which the agents are rule driven. An > important > > > feature is > > > > > > > that the agents and their rules must be > dyanically > > > > > > > manipulable. That is, both the agents and the > rules > > > must be > > > > > > > first class objects that can be manipulated as > the > > > system is > > > > > > > running. Can Drools support that? > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > As someone not familiar with Drools I did my best > to > > > find > > > > > > > out by looking at the Drools > > > > > > > Expert User's Manual. From what I could tell, > the > > > Template > > > > > > > feature in which one uses Java objects as > Template > > > data > > > > > > > sources comes close. I'd like your advice > > > before > > > > > > > putting too much development effort into > attempting to > > > use > > > > > > > it for that purpose. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > In particular, would it be possible to set up a > > > mechanism > > > > > > > that could: create and delete agents on the fly; > > > associate > > > > > > > rule sets with agents on the fly; and modify the > > > rules > > > > > > > associated with an agent on the fly? After > each > > > such > > > > > > > action the system would change to reflect the > change > > > in > > > > > > > agents and rules. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > Thanks for your help. > > > > > > > > > > > > > > -- Russ Abbott > > > > > > > > > > > > > > P.S. If Drools would not be a good choice, is > there > > > some > > > > > > > other system that you could recommend? > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > -----Inline Attachment Follows----- > > > > > > > > > > > > > > _______________________________________________ > > > > > > > rules-users mailing list > > > > > > > [email protected] > > > > > > > https://lists.jboss.org/mailman/listinfo/rules-users > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > _______________________________________________ > > rules-users mailing list > > [email protected] > > https://lists.jboss.org/mailman/listinfo/rules-users > > > > > > -- > - http://salaboy.wordpress.com > - http://www.jbug.com.ar > - Salatino "Salaboy" Mauricio - > > > _______________________________________________ rules-users mailing list [email protected] https://lists.jboss.org/mailman/listinfo/rules-users
