Research On Interval-based QoS Uncertainty-aware Service Selection

Service-Oriented Computing(SOC) is the computing paradigm that utilizes software services as fundamental elements for supporting the development of rapid, multi-platform, interoperable, and massively distributed software application. Service-Oriented Architecture(SOA) is a logical way of designing a software system to provide services either to end-user application or other service distributed in a network. Web service technology is a specific kind of implementation of SOA which contains a bunch of standards and protocols. With the development of application in Internet and Cloud computing, more and more services are deployed into the Internet. As a basic mechanism for implementing of complex SOC applications, Service Composition are widely focused by industry and academia.Since there are more and more web services published online nowadays with identical functionalities but different quality characteristics, QoS-aware service selection is a hot research topic in Service Oriented Computing, which aims at to select atom services from candidate service set to build service composition with optimal performance and satisfies user's end to end QoS constraints. Although the problem of QoS-aware Web Services Selection is well addressed by existing studies, in which the QoSs are provided by service provider or assumed satisfying a probability distribution, an important issue, i.e., the uncertainty of QoS, is still left unconsidered and unsolved. It becomes a crucial problem to lead the failing of service composition.This thesis investigates the uncertainty of QoS, and proposes a QoS uncertainty-aware service selection framework. It focus on how to model the uncertainty of QoS and consider it as a basic characteristics of QoS in the process of service composition, constraint satisfaction and clustering reduction.The main content and contributions of this thesis are summarized as follows:(1) This thesis proposes a QoS interval model and service selection algorithm. According to the QoS history records, QoS interval is defined to measure the uncertainty of QoS. Two dominance relationship, Crisp Dominance and Fuzzy Dominance, are defined for QoS interval comparing. For the fuzzy dominance, a detail computing formula for dominance degree is presented. After, based on PROMETHEE method, a service ranking method for QoS interval is proposed. This novel approach rank the services not only consider the profit of QoS but also the uncertainty of QoS. Finally, an updated Genetic Algorithm is used to optimize the service selection. Empirical studies prove that this approach could find composite service with more QoS stability.(2) This thesis presents a soft constrained two-phase QoS uncertainty-aware service selection. The Constraint Satisfaction Problem and Soft Constraint Satisfaction Problem are introduced first. A Soft Constraint Satisfaction Problem model for service selection is then defined which enable user to setup multiple constraints other than single one in traditional method. Accordingly, a Soft Constraint Service Level Agreement is defined to support soft constraint model. In the two-phase approach, a MIP-based global constraint decomposition method are used to compute the constraint satisfaction probability. After that, services with different constraint satisfaction probabilities are mapped to various satisfaction degrees. Finally, a penalty function-based dynamic fitness function is defined for Genetic Algorithm to pursue near optimal composite service. Empirical studies illustrate that this approach could obtains near optimal composite service in different constraint strength.(3) This thesis proposes a clustering reduction-based QoS uncertainty-aware service selection. Since the performance of service selection is very low when there are numerous candidate services in each service class, a novel clustering reduction method is proposed by using Interval Fuzzy C-means Clustering for QoS uncertainty-aware service selection. The candidate services of each cluster are then replaced by one representative service of the cluster to participate global service selection. Three optimal cluster selection methods are proposed. Finally, in order to choose one service to join the composite service from the optimal cluster, three strategies are suggested for different situations of candidate service class respectively. Synthetic and real world dataset-based studies represent that this method could get better composite service when the number of candidate services in each service class is huge.