to web services composition is presented in Ref. [56]. In this paper, we present a service plan generation and execution architecture in the context of web services deployment that addresses problems associated with the open world assumption and dynamic service capabilities. The ISP&E architecture is based on a domain-independent planner called Simple Hierarchical Ordered Planning (SHOP) [25
newport cigarettes coupons,43,44] for dynamic service planning. SHOP is based on the Hierarchical Task Network planning technique. Our choice of this methodology for planning-based service composition was guided by several considerations which we discuss below. Firstly
newports cigarettes wholesale, we provide a brief background on the stateof-art in planning methodology before outlining our rationale. Two generic classes of planning algorithms have been developed in the literature, namely, partial order planners and total order planners [54]. Partial order planners develop project networks, consisting of directed networks of actions, whereas total order planners develop a plan with sequential actions. A given partial order plan may be linearized into multiple
sequential plan alternatives. Partial order planners search in plan space
cheap newports online, whereas total order planners search in state space. A variety of algorithms have been developed for both types of planners, based on techniques such as constraint satisfaction, logic resolution and theorem proving, satisfiability and model checking [54]. Partial order planners embed a leastcommitment strategy and the plans they generate are inherently flexible because of their networked nature. However, developing such partial order plans incurs higher plan search costs unlike state-space planners which have exhibited excellent performance on large problems. Recent research in AI planning has been focused on developing plans involving concurrent and durative actions for both partial and total order planners. However, developing such plans with parallel actions requires explicit temporal models for actions such as their durations and metric models of resources in the plan representation. Consideration of such numeric quantities considerably increases the size of the search space [6
marlboro gold pack,54]. Developing efficient and scalable algorithms for temporal and metric reasoning in planning domains is an active research area whose results may be incorporated in our framework in future. An alternative approach used in our framework is first the generation of totally ordered sequential plans followed by the relaxation of precedence constraints to generate partially ordered plans [7,66]. Details are further discussed in Section 4. In the web services domain, our choice of a planning approach was guided by the need for (a) performance, (b) the ability to embed domain knowledge to guide and control the search for a viable plan, and (c) an explicit state representation to enable interleaving of planning and execution. These criteria guided our decision to develop ISP&E based on a state-space, linear-order planning algorithm. The benefits of exploiting concurrency (in paral