- From: <Joachim.Peer@unisg.ch>
- Date: Thu, 26 Feb 2004 10:52:59 +0100
- To: "Charlie Abela" <abcharl@keyworld.net>
- Cc: public-sws-ig@w3.org, public-sws-ig-request@w3.org
- Message-ID: <OF3422DEAC.FB53C817-ONC1256E46.003470FF-C1256E46.00364A4A@unisg.ch>
hi Charlie, i think the constraints that you are describing are basically capturing the kind of information that is typically capturered by a "plan". lets take for instance plans produced by partial order planners: in POP planning, a plan is represented by a set of action instances A representing the steps to be taken by the agent; a set of causal Links L indicating dependencies among those actions; a set of ordering constraints O defining a partial order of the actions; and a set of binding constraints B on the action parameters. Your constraints <output=input> could be represented by a *binding* constraint. It should be noted, however, that most planners do not support those constraints in their final plans, because they assume that all parameters are ground (= fully instantiated). This is what makes them, in my view, rather unusable for Web service compositions, where certain pieces of information CAN not be known at planning time. On the other hand, planners that do support binding constraints *can* be used for WSC quite effectively, at least as far as my preliminary experiments (toy examples) have shown. In the next time, i am planning to investigate the boundaries of this approach, and if you are interested we can work together and share example scenarios etc. I am using VHPOP currently (which supports partially instantiated actions / binding constraints) in plans (http://www-2.cs.cmu.edu/~lorens/vhpop.html) cheers Joe "Charlie Abela" <abcharl@keyworld An: <public-sws-ig@w3.org> .net> Kopie: Gesendet von: Thema: RE: Composition as planning public-sws-ig-req uest@w3.org 26.02.2004 10:06 Hi, I've been thinking about this for some time and would like to get some feedback. I have considered, to some extent, work on planning and how it can be applied to Web services composition. From earlier discussions over this list it is clear that traditional classic planning techniques have to be in some way adapted to handle the diversity issues over the Web. I want to put forward the following thoughts: Various planning techniques where devised around specific problems and domains (correct me if I'm wrong): for example STRIPS planning. When considering the planning required for Web services, there are issues which seem to crop up in every composition hence there is place for capturing these commonalities into some form of operator-like descriptions. For instance consider a simple sequential service composition: Two services can be composed sequentially if the outputs/effects of one service match with the inputs/preconditions of the other. In the real world, apart from checking this, some composer should also make sure that once the message with the outputs/effects from the first service has been received then another message that includes this information is sent to the second service. Hence though an initial plan that matches IOPE can be performed in some preliminary phase of composition, then monitoring of message passing is required to make sure that the initial plan can actually be executed in the real world. Returning to my initial point, one can say that these actions are common to most compositions that can occur. So it would be possible to capture these into a set of operators that represent the actions that are available for service composition. For example: if the output/effect from service A have been received then a message to service B can be sent Operator: sendInvocationMsg Preconditions: (and (output=input)(effect=precondition)) (some constraint here over the issue of services having both output/effects and requiring input/precond) Effect: (service A executed) Other operators can be defined in a similar way to capture other common behaviours related to service composition. Now this is just an idea and might have been explored by others, but IMHO it gives some standard way of composing services and gives a planner the possibility to monitor real world compositions. Would really like to get feedback on this and possibly other ideas that confirm/or not the above. Regards, Charlie
Attachments
- image/gif attachment: graycol.gif
- image/gif attachment: ecblank.gif
- image/gif attachment: pic09894.gif
Received on Thursday, 26 February 2004 04:53:06 UTC