Hi Maurice, There is some work done last year by Thomas and me for distributing a PN model on a set of nodes. Although we assume there is one master node for designing, partitioning and distributing a model, after distributed the execution runs as peer-to-peer, i.e. one node can communicate with another node directly. We used Corba for remote communication. In order to resolve the communication dependency loop, we adapted the event channel idea: each node runs an agent to route messages to the corresponding agent on a remote node, and the model on a node uses the agent on the same node (could be some other agent if necessary) to communicate with other models. No centralized event channel is used though, each agent act as a half event channel.
Although we didn't use JXTA or P2PS in this work, I feel we share some of the design goals. It would also be interesting to see how JXTA or P2PS can make the design deferent from above. For example, I think we can get rid of the agent for routing messages easily when we use JXTA or P2PS for the underlining discovery and communication. There is some classes uses JXTA in the Ptolemy tree as Christopher pointed out in his email. These classes are checked in by me long time ago and I am not sure it is still compatible with the new JXTA version. There were two main applications developed. One application supports peer-to-peer actor sharing. Another application is probably more interesting in the sense it uses JXTA and Corba together: it allows one actor to discover other actors uses JXTA, the IOR of the discovered actor is encapsulated in a JXTA message, the discovering actor can get the object reference from the IOR and do remote method call on the discovered actor. Not sure this combination is important for practical use. It may be useful when we prefer dynamic discovery and configuration, but prefer synchronous communication (like remote method call rather than pipe). Anyway, it was for fun... I agree with Matt on the JXTA comments. I haven't tried P2PS. Ian Wang demonstrated it to me in a workshop. It looked cool and much lighter than JXTA. I would be happy to know how the programming with it is like if you experiment with it. Cheers, Yang At 12:52 PM 11/29/2005 -0900, Matt Jones wrote: >Hi Maurice, > >We have been experimenting with JXTA and Jini in Kepler and Ptolemy, but >at this point the work is largely in the experimental phase. Daniel >Cuadrado committed some code that allows for an DistributedSDF director, >which can be used to distribute the computations on a set of Jini nodes. > This isn't fully mature because there is not yet enough control over >what parts of the model are distributed and how the Jini nodes are set >up and configured with Kepler. But it does work. You can compile and >run the jini prototype in Kepler using "ant run-jinidev", although it >would probably take some setup beforehand. > >In addition, I did some work on using JXTA to set up peer-to-peer grids >-- it is in CVS but not tied in to the running application. The JXTA >work is described here: > http://kepler-project.org/Wiki.jsp?page=PeerToPeerKepler > >We haven't been able to work on this stuff lately due to more pressing >issues in Kepler. But I do plan to come back to it as soon as possible. > If you'd like to help out in this regard it would be welcomed. I >think we will probably drop the JXTA framework as it seems very hard to >program to and somewhat fragile. We'll have to evaluate other >frameworks (Jini, P2PS, others) to see if they would be better than JXTA. > >Cheers, >Matt > > >Maurice Yarrow wrote: > > Hello Kepler users > > > > Is there presently any way to interface > > Kepler to either JXTA or P2PS ? > > > > I.e.: I would like to write autonymous > > cli-servs-peers (have properties of both > > client and server) that can communicate > > with the director of a running Kepler instance. > > > > My goal is not well served by using Web > > Services for a variety of reasons, so > > the above model will allow more flexibility > > for the type of real-time conduct I require. > > > > (Triana supports this but Kepler would be a > > preferred platform for a variety of reasons.) > > > > Also apropos of this issue, what is the status > > of the line item in Alpha 8 (Dec 9 release > > indicated): "initial kepler grid support for > > distributed directory/p2p" ? > > > > > > Maurice Yarrow > > > > > > _______________________________________________ > > Kepler-users mailing list > > Kepler-users at ecoinformatics.org > > http://mercury.nceas.ucsb.edu/ecoinformatics/mailman/listinfo/kepler-users > >-- >~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ >Matt Jones Ph: 907-789-0496 >jones at nceas.ucsb.edu SIP #: 1-747-626-7082 >National Center for Ecological Analysis and Synthesis (NCEAS) >UC Santa Barbara http://www.nceas.ucsb.edu/ecoinformatics >~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ >_______________________________________________ >Kepler-users mailing list >Kepler-users at ecoinformatics.org >http://mercury.nceas.ucsb.edu/ecoinformatics/mailman/listinfo/kepler-users
