Online Optimization And Decentralized Orchestration Research Of Web Service Composition

Web services technology is presented as an effective solutionto inter and intra enterprise applications integration. Servicecomposition is a process that assembling some component servicesaccording to their function, semantic and logic relationshipamong them to a service for providing a new function. Along withthe similar web service increase quickly, the amount of serviceimplementing the same function is large, so selecting thecomponent service considering its QoS is a key technology of webservice composition. Orchestrating fashion of composite serviceis a deploy structure when composite service execute dependenton, and a control and data transfer model and implement machinebetween component services. Current researches focuses on thecentralized orchestrate of composite service, but thisorchestrate fashion leads to some problems such as the runningefficiency, expansibility and applicability of the compositeservice system, and does not adapt to the application contextwhich has the data flow constrains.Currently, most researches only focus on some phase of QoSlifecycle management, this paper propose a frame that supportsthe management of whole QoS lifecycle. Based on this frame, anAHP approach with automatically modified judgment matrix isproposed to implement online strategy of local QoS optimization.Traditionally, consumer can only provider relative importanceof Qos properties, but cannot describe the global Qosrequirement. Aiming at this scenario, this approach provides anonline local optimization solution.Aiming at the application context of data flow constrain, in this paper our tool composes the web services with centralizedmodel, then take the algorithm to partition the centralized codeinto decentralized codes for decentralizing execution. Weapply multilevel algorithms for graph partitioning to partitionthe centralized code in order to satisfy the constrain ofminimizing the data stream between the domains and maximizingthe throughput of the distributed system. Our algorithm canaccomplish the task automatically and quickly, and furthermorebalance the load of the distributed system.