Towards Cost-based Optimization for Data-intensive Web Service Computations

The recent popularity of XML and Web services has lead to a surge in models and platforms for distributed XML data management applications. This work investigates performance issues involved in the deployment of the ActiveXML (AXML) platform for such applications. AXML documents are XML documents, part of which is extensional (present in the document), while part is intensional (specified as calls to Web services). Materializing an AXML document thus requires activating all service calls, and gathering the call results in the document. In this work, we demonstrate that many distributed materialization strategies exist for a given AXML document; basically, these strategies may differ in the choice of the peer that executes each service call, or of the peer that makes the call. The AXML system has to choose among the strategies in order to generate efficient materialization plans. We formally characterize the optimization search space, and provide some heuristics to improve plan generation. We describe our optimization model and how it fits in the actual AXML P2P architecture. We describe the possible decision models; the parameters of this model, and the required infrastructure to provide them. Finally, we present empirical results that validate the proposed methodology.