Browse Theses

Web services for Service-Oriented Architecture (SOA) provide a flexible and scalable framework for service composition. Using standard-based protocols (SOAP and WSDL), composite services can be constructed by integrating atomic services developed independently to form complex business processes and workflows. The runtime performance of a composite service is important for most distributed applications with practical significance. The end-to-end performance management is a big challenge for distributed SOA systems due to the flexible and compositional nature of Web services. In this dissertation, we design a broker-based architecture QBroker to provide efficient QoS management for service-oriented distributed system. The main functions of QBroker include service tracking, planning, selection and adaptation. In QBroker, Web service composition with end-to-end QoS assurance includes two steps: service planning and QoS service selection. The efficiency of service selection decides the run-time performance of a QBroker. We design several service selection algorithms for composite services with different composition structures and different number of QoS constraints. The algorithms define the selection problem in two ways: combinatorial model, by modeling the problem as various knapsack problems and graph model, by modeling the problem as multi-constrained optimal path problem. The objective of service selection is to maximize an application specific utility function under the end-to-end QoS constraints. The performances of the algorithms have been studied by extensive simulations. The algorithms are efficient for both offline planning and making online decisions.