XCraft: A Dynamic Optimizer for the Materialization of Active XML Documents

An active XML (AXML) document contains special tags that represent calls to Web services. Retrieving its contents consists in materializing its data elements by invoking all its embedded service calls in a P2P network. In this process, the results of some service calls are often used as inputs to other calls. Also, usually several peers provide each requested Web service, and peers can collaborate to invoke these services. This implies many equivalent materialization alternatives, with different performance. Optimizing the AXML materialization process is a hard problem, which often involves searching a huge space of solutions. Current techniques for workflow scheduling and distributed query processing are insufficient for this problem, since in AXML materialization: (i) the set of participating peers is not known in advance; (ii) service calls in the result of other calls forbid a simple “optimize-thenexecute” strategy; and (iii) due to the peer volatility in the network, a plan computed by the optimizer may become invalid at the moment of its execution. Moreover, most of the current optimizers are based on centralized coordination. We propose a dynamic, cost-based optimization strategy to efficiently materialize AXML documents considering the volatility of a P2P scenario. We formalize the problem from a performance-oriented perspective, and present an optimization strategy that incrementally generates and executes materialization plans. This enables the optimizer to reduce the size of the search space, get more up-to-date information on the status of the peers, and deliver partial results earlier. Our strategy can handle arbitrarily complex AXML documents, and exploits decentralization in many levels. We also present a service-oriented optimization architecture called XCraft. We evaluated our approach in an XCraft prototype for the ActiveXML system, an open-source P2P platform. Our results show promising performance gains compared to centralized, static materialization strategies.