Research On The SLA_Oriented QoS Management And Optimization For Service Composition

In an open service-oriented computing environment, effective QoS management is required for service providers, so that dynamically changing QoS demands of service users can be satisfied by flexible configurations on service provisioning. On the other hand, the service level agreement (SLAs) provides contract base for the business relationship between service users and providers, which reflects user’s dynamic QoS demands and provides basis for the QoS management of service providers. Therefore, it is necessary to consider and design the QoS management process and functionality under SLA management framework. In this paper, we focus on the SLA-oriented QoS management and optimization problem for service composition, in which the end user’s QoS demands are defined by fore-end SLA while the back-end SLAs guarantee service qualities of partner services used in the composition. Because of the QoS correlation between fore-end and back-end, the fore-end QoS requirement service composition can satisfy is decided not only by resource allocation, but also by the QoS level contracted in each back-end SLA for exterior partner services. Thus, confronting dynamic QoS requirements of fore-end multiple users, the service composer should perform configurations on computing resources as well as back-end SLA contracting simultaneously, so as to satisfy the QoS requirements while maximum its SLA revenue. And the configurations rely on effective optimization approach.Considering above issues, this paper try to provide systematical solution for QoS management and optimization problem of the service composition on basis of SLA contracting environment. At first, two multi-objective optimization problems are formally defined for two kinds of configurable resources, i.e. the back-end SLAs for constructing service composition instance,and the computing resources for deployment and running. Then, relation between two optimization problems is established, such that the two kinds of resources can be optimized simultaneously when performing configuration. Based on the problem definition, a SLA-oriented QoS management model is presented for service composition, in which the management entities, functions as well as activities are provided for three different system layers respectively, i.e. the SLA contracting layer, the service composition layer and the resource allocation layer. Meanwhile, correlations of management entities and functions between different system layers are built by optimizing the configuration of two kinds of resources, and constructing the entire system model. Finally, two different QoS management strategies, i.e. QoS level based and customized QoS,and their respective configuration process are presented. For the purpose of optimization, a preference-based and a non-preference multi-objective optimization approach and corresponding evolutionary algorithms are presented.In the non-preference optimization approach, concerning about the uncontrollable results scale of pure Pareto ranking, the PROMETHEE model is further introduced to define a combined ranking model for the QoS optimization problem of service composition. Then, an efficient evolutionary algorithm:P_MOEA is presented, which can efficiently find the Top-k optimal solutions for configuration in the QoS management process. In the preference-based approach, focus on following issues, i.e. the difficulties in preference definition and weights assignment, the limitation of linear utility function for identifying preferred skyline solutions, and the requirements of efficiency and scalability of the optimization algorithm. At first the problem is based on the weighted Tchebycheff distance rather than the linear utility function. Then, a fuzzy preference model for preference representation and weights assignment is presented. In which a set of fuzzy linguistic preference terms and their properties are introduced for establishing consistent preference relations upon multiple QoS dimensions, and a weighting procedure is proposed to transform the preference into numeric weights. Finally, two evolutionary algorithms, i.e. single_EA and hybrid_EA are presented to implement different optimization objectives, and have different features that can be used in different QoS management strategies. Sets of experimental studies are conducted for the evaluation of above proposed algorithms.