Research On Probabilistic Analysis Of Web Service Composition Based On Random QoS

Web service as a new kind of distributed computing model and collaborative relationship has created an unprecedented opportunity for the dynamic and cross-organizational resource sharing. It has got the attention of the whole industry and academia. With the gradually mature of Web service standards and technology, the development and deployment of Web service has become much easier. When a single Web service can not meet the demand of users, the organic combination of a number of function limited Web services is necessary to improve the reusability of Web services and form a large particle size and the composite service which has the internal process logic. The execution of the composite service can complete the business objectives and create the value-added services.With the spread of the Web Service in multiple areas, the services meet the same function requirements are becoming more and more, we should not only consider the functional requirements but also the demand of non-functional attributes QoS (Quality of Service), such as response time, cost, reputation, throughput, etc. Open and dynamic network environments lead to the stochastic nature of Web service QoS. Therefore, the Web service compositions which composed of these services will be inevitably random of the QoS. And the QoS requirements of Web service composition may not be certainly satisfied. Therefore, the research to the source, expression, quantification, identification and control of the risk of composite service QoS will be very important to improve the ability that the composite service can satisfy the user’s QoS requirements and reduce the using risk of composite service.This paper mainly research the representation and acquisition of the composite service QoS risk information. It is necessary to understand the random characteristics of composite service to complete the management of service risk. And the accurate probability distribution is an effective means to describe QoS stochastic properties. Therefore, we analyze the probabilistic distribution of the stochastic QoS of Web service composition based on a simulation approach named Importance Sampling in stochastic PERT network. And we calculate the unbiased estimation of the target distribution function through a weighted to a simple distribution function. Experiment has been done to show that the Importance Sampling approach is much more rational and efficient than Monte Carlo simulation based on the actual QoS data of Web services. Finally, we design a a prototype system to obtain the probability distribution of composite service QoS, and the prototype system is tested.