Research On QoS-based Web Service Composition And Fault Tolerance Approaches

Service computing pattern supports composing various Web services to acquire a new composite Web service, which achieves Web service reuse. However, as the number of Web services dynamically increases, it results in different performances on quality of service (QoS). While executing the application of a composite Web service, some cases may lead to failure such as various malicious Web services (Web services with viruses), QoS violation that means service providers renege on the QoS pledge to service users, and node failure in a cluster by environmental changes. Hence, there exist many challenges in building the QoS-based service composition and achieving fault tolerance among the large number of Web services. This dissertation focuses on the key problems and the contributions are shown as follows.(1) For the problem of Web service composition with trustworthy QoS, a trustworthy Web service identification approach is proposed, which makes use of the two-phase neural network model for guaranteeing the trustworthy QoS, then a QoS ranking-based composition model and the corresponding solving algorithm are designed, while meeting the global constraints and guaranteeing the Web services with trust-worthy QoS in the composite service. The simulation-based experimental results show that the identification ratio of untrustworthy Web services reaches 90%, high-er than the typical approach such as CorrelationLens. In addition, the proposed approach achieves almost optimal result and offers up to a 3-fold increase in speed for obtaining solution over the previous integer programming approach, while guar-anteeing the optimization and trustworthiness of the composite service.(2) For the problem of Web service composition in a large number of service can-didates, a fast Web service composition algorithm based on QoS-aware sampling (FAQS) is proposed. Firstly, the QoS database and utilities of Web services are in-vestigated by using statistical approach, then the equal number of Web services are sampled from each interval of utility values. Secondly, the model of Web service composition is built by considering the utilities of Web services and their frequen-cies of utilizations. Finally, according to the result of Web service composition among the samples, the further optimal result is achieved by mapping the initial results to all service candidates, while meeting the global constraints and guaran- teeing the utility and frequency to maximum. The simulation-based experimental results show that FAQS can accurately and efficiently achieve the almost optimal result in a large number of service candidates.(3) For the problem of the failure of one or some of web services in a composition plan, a framework (FTDes) is proposed, which includes decision making and opti-mization selection of fault tolerance strategies to enhance the reliability of service composition plan. The process of decision making is to select the decision pa-rameters and to build the decision matrix, then to design the decision model for finding a set of the substitutes of failure services. In the process of optimization, the problem of selecting optimal fault tolerance strategy is defined as 0-1 integer programming, then the convex hull approach combined with a standard integer programming solver (CPLEX) is proposed to efficiently solve the problem. The simulation-based experimental results show that the FTDes can achieve more ac-curate and less response time than the traditional approaches including Analytical Hierarchy Process approach (APH), Evidential Reasoning approach (ER), Tech-nique for Order of preference by similarity to ideal solution approach (TOPSIS), and Linear Assignment Method (LAM). Meanwhile, the impaction analysis is per-formed, such as the impact of process of optimization, TopK algorithm, and deci-sion parameters. The proposed framework (FTDes) maintains and optimizes the overall QoS of the composite service and enhances the reliability of Web service composition plans.(4) For the problem of the node failure in a cluster for Web composition applications, a ring-based heartbeat detection mechanism and a light-weight recovery mecha-nism are proposed to enhance the reliability of Web composition applications. In the ring-based heartbeat detection mechanism, to reduce the burden on central n-ode, a distributed detection algorithm is designed based on the ring topology, and the algorithm is deployed to each node in the ring. To reduce overhead, the light-weight recovery mechanism by considering load balance assignment is proposed. In the light-weight recovery mechanism, a load reassignment algorithm is designed, which can add a new node in the ring topology to balance the load when the cluster load is beyond the threshold. In this way, the failure of the node caused by the high cluster load may be avoided. The experimental results show that the proposed mechanisms can achieve lower overhead and less latency compared with log anal-ysis approaches and system-level recovery mechanisms. Meanwhile, the proposed mechanisms avoid the devastating impact that caused by the failure of Web servers and improve the reliability of Web service composition applications.Above all, this dissertation researches the composition approaches and fault toler-ance approaches in the field of QoS-based Web service composition. To enhance the efficiency, accuracy, and reliability of Web service composition, the ranking-based ap-proach, the fast sampling approach, the ring-based heartbeat mechanism, and the light weight recovery mechanism are proposed. The research has important values in theory and wide prospects in application.