Research On QOS-based Web Service Selection And Diagnosis

As an ongoing revolution in software engineering, SOA (Service Oriented Architecture) is attracting more and more attention both in industries and academics as a result of its outstanding flexibility, encapsulation and software reusability. In Service Oriented Architecture, software is designed and developed in the form of interoperable and heterogeneous services to meet clients’varying requirements with a low cost. Given the promising prospect, challenges and problems such as how to select and compose services in an efficient manner and how to ensure the software reliability in such a heterogeneous environment when exception occurs still remain in SOA. As the number of service climbs sharply, these problems become the focus topics in corresponding field. Based on previous work, this paper presents sophisticated discusses of issues mentioned above and arrives at corresponding solutions.Two web service selection methods are deviced as a part of our work and entitiled respectively as the simple uncertain QoS based web service selection method (SUQ) and the representative QoS records based web service selection method (RQR). Firstly, these two methods both take the uncertainty of QoS into consideration, thus capable to generate more reliable composition plan. Secondly, these two methods are adapted to different environments. SUQ method needs a high cost to get the optimal plan. Instead, experiment shows that RQR method could get a near-to-optimal plan (more than95%in accuracy) with a much lower cost (less than1%in time consumption) compared to SUQ.A QoS based diagnosis method for SOA is also proposed as another part of our work. This method is capable to locate the reason of failure when runtime exception occurs to support the subsequent exception handling process. Firstly, the proposed method extends an existing architecture in order to get more diagnosis information. Secondly our method operates a probabilistic analysis on different disgnosis explanations to evaluate their importance with posterior probabilities and thus provides more diverse support for exception handling.