This talk reviews the current situation in service oriented computing and presents a stratified service architecture (SSA) that combines the elementary service operations describe, publish, discover and bind with complex operations associated with functional and non-functional service composition and service management. It also links the traditional roles of service provider and requester associated with elementary service operations to more complex roles such as those of a service aggregator and market maker.
Subsequently, the talk concentates on the middle layer of the SSA, namely service composition, and describes the requirements and necessary infrastructure support for service composition. It then relates the concept of AI planning to service composition and delineates requirements for planning for services on demand. These include non-deterministic, interleaved and reactive planning.
Finally, the talk proposes the use of service request language (XSRL) developed on the basis of XML and AI planning techniques. XSRL allows the planner to select suitable actions for service compositions created in response to a user request and plan its actions so that it meets the original constraints of the user. XSRL can express information about a request, the succession of activities needed to satisfy a request and about the parameters of the planned actions. In addition, the plans generated by the language are capable of dealing with non-determinism and interleaving and planning of actions under uncertainty on the basis of refinement and revision as new service-related information is accumulated (via interaction with the user or UDDI) and as execution circumstances necessitate change.