Research On Key Technologies Of QoS-based Web Service Selection

The emergence of Web services represents the remarkable progress in Internet technology since Web service greatly extends the function of applications by defining how applications interact. Now, Web services are widely distributed on the Internet. Being the de facto standard, they tend to become the basic modules of the next generation’s application development based on SOA (Service Oriented Architecture).Existing Web service system provides support for Web service registration, discovery and invoke. However, the deployment of Web service based SOA applications on the Internet has led to a massive increase in the number of services with identical or similar functions but different QoS (Quality of Service) values. In these circumstances, there comes a serious challenge on how to select and composite proper services precisely and efficiently to meet users’ complex business requirements. In recent years, Web services with QoS assurance have become one of the hot issues among scholars and research institutes all over the world. Although research in this area has been developed progressively, Web service selection is still a challenging issue due to the uncertainty of the QoS value, the characteristics of the composite service QoS, the performance requirements of services selection algorithms and the benefit conflicts between the service requester and the service provider.This dissertation intends to solve the aforementioned problems. The main contributions include:First of all, aiming at the problem of strong uncertainty the Web Services’ QoS have in the real execution environment, a QoS model with its random QoS parameters described in interval-probability is introduced. Compared with the traditional deterministic QoS models, the proposed model has powerful expression ability to reflect the true performance of Web services and more closely match practice needs of the Internet. Moreover, it helps to make flexible QoS commitments to users.Secondly, based on the analysis of the composite service characteristics, a composite service QoS model supporting services relations is proposed. A composite service is organically made up of various service components which are required to coordinate and adapt each other to ensure the correct execution of the composite services. Therefore, besides the general QoS parameters of the basic services, a composite service should have additional QoS parameters reflecting its composition characteristics and these parameters need to be considered in services selection. In consideration of the relations between service components, both the interface composability and the condition composability are introduced into the QoS model of the composite services. The extended composite services QoS model can better reflect the overall quality of services, thereby more effective support for Web services selection can be provided.Thirdly, to solve the problem of neglecting the service providers’benefits in service selection, a win-win strategy based on game theory is given.Most existing service selection methods serve the users with a "best effort" strategy which maximizes users’benefits by providing them with the "best" service while neglecting the service providers’benefits. Base on game theory, an evaluation method on the utility under different users’strategy and providers’strategy is proposed. On this basis, a mathematical model on Web services selection aiming at utility equilibrium is put forward. The proposed win-win strategy helps to improve the synthetic utility of users and service providers.Fourthly, aiming at the problem of premature convergence of PSO (Particle Swarm Optimization) in searching the optimal composite service, a discrete PSO algorithm based on dynamic boundary is proposed. The selection of optimal composite service with overall QoS assurance is a NP-Hard problem, and PSO is an effective method to solve such problems. The PSO works well at fast convergence, but easily falls into local optimum. According to the movement characteristics of particles in PSO evolution process, a strategy fighting premature convergence is proposed in this dissertation. By tracking the distribution of the particles’locations, the proposed algorithm adjusts the boundary of search space dynamically, which could guide the particles to more promising region. Thereby premature convergence is reduced and the search ability is improved.Finally, a prototype system for Web service composition and selection is designed and implemented. The feasibility and effectiveness of the proposed method are proved by the experiments carried out on the system.