| Service Oriented Architechure (SOA) may will be the answer for the forever-change requirements and difficult integration problem of software development. But, to full-fill what industry expecting from it, SOA need a mechanism to automaticly integrate services. The efforts of combining Semantic Web technology with Web Service created Semantic Web Service technology, which can provide a machine understandable languge to decribe knowledge of services. This make automaticly composing services to new funtional granularity according to what they will do possible.For the last 2 years, lots of works on this topic came out, but in fact most of them, from our view, are nonedeterminated model with dynamic binding composition —to which you may call them "automatic", but have no chance to make them the one SOA required. We think only the truely automatic technology like automated planning can achieve this task. This work can be taken as a practical shot onto the automated semantic webservice composition using automated planning technology.After carefully study on numbers of usecases of Web Service composition, we think traditional planning can not match the requirements: most of todays semantic webservices defined and described using the viewpoint of KR and Ontology, not FOL; requirements of most usecases have too much non-classic planning characters; the problem scale is really large for none-trivial usecases; in some of them, the enviroment is almost the most dynamic you can see; there is requirement for data flow composition for the RPC type Web Services.So, we propose a modification basing on Ontology viewpoint on thetraditional planning description: we do not use propersitions to describe world state but "individuals"; We separate the congnized world state Sa from real world state S; we take the service execution as producing, eliminateing or sensing process. And then introduce a concept VCB( variable constrain block) to character a individual. Use VCB's binding relationship to express i-o binding, result-precond cause link. Then convert the problem of finding a solution plan to finding a set of VCBs with bindings satisfied a certain constrain.Base on these work we try to extract a simple subset of Web Service composition usecases, and try to find acceptable solution to them. We start working on 2 planner: liquidus.ffplanner, almost finished FF planner, using enforced hill-climbing algorithm to find solution as fast as it can and a under developing POP planner liquidus.popplanner, using plan space heuristic search to find solution when ffplanner can't.One important result we have got when we study these usecases is :Web Service Composition trends to use schedule technology more than general planning. So we expand the service description of OWL-s to support unique resource occupy declaration. We also set a mechnism to get feedback of service execution time from execution engine. And we use these two thing to make our partial-ordered solution plan can be concurrently executed for usecases not having restricted temporal requirments.We think the combination of Static Modeling and Dynamic Binding is the most practical solution to most of automated composition process, but the Binding must be through a composition. So our work is used in compositing binding of executing an OWL-s process model by our Execution Engine.To resolve the problem of large scale, we write a filter program to filter related operations by: operations' taxonomy if the composition provided one; extracting backward from goal concept using a limited greed search through operations if not. And the relaxed plangraph heuristic mechanism also help us to determinate current helpful operations.
|
PDF Full Text Request