Synchronization analysis for decentralizing composite Web Services

Web Services are emerging as the standard mechanism for making information and software available programmatically via the Internet, and as building blocks for applications. A composite web service may be built using multiple component web services. Once its specification has been developed, the composite service may be orchestrated either using a centralized engine or in a decentralized fashion. Decentralized orchestration brings performance benefits, and improves scalability and concurrency. Dynamic binding coupled with decentralized orchestration adds high availability and fault tolerance to the system. However in such systems, the coordination between components needs to be carefully designed to ensure correct execution of the composite and to limit the synchronization overheads. In this paper, we categorize different forms of concurrency and provide an algorithm to identify these forms in a composite service specification. We explore different mechanisms for transferring data between the components in the presence of different forms of concurrency. Then we experimentally evaluate the efficiency and scalability of each mechanism. We also analyze the coordination requirements of a decentralized orchestration in the presence of dynamic binding and fault propagation.