Agent-based QoS management for web service compositions

Service oriented computing (SOC) is fast becoming the core platform for software applications and integrations. In SOC, applications are modeled as collections of loosely-coupled, interacting services that communicate using standardised interfaces and protocols. These services can be combined together to form new value-added services, or compositions which potentially span across many administrative boundaries. Managing the quality of Web service compositions is important. However, it is difficult in practice. The difficulty lies in the complexity of composition dependencies such as sharing computing resources or component services. The distributed nature ofWeb services presents another challenge where a service composition may involve service providers from different organizations and administrative domains. A viable collaboration between these service providers is necessary for any successful management. This thesis studies the Quality of Service (QoS) management for multiple service compositions. Despite many approaches having been proposed for the management recently, those approaches are limited to an individual composition and management functionalities executed by a single organization. This thesis aims at providing solutions for those limitations. It has the following contributions: - A new framework supporting a distributed approach to solve the QoS management problem for multiple inter-related compositions is proposed. This problem has not been well investigated in the current Web service research. In the thesis, fresh ideas from Distributed Constraint Satisfaction (DisCSP) research are employed and applied practically to the Web service environments. - Extensions and improvements to existing DisCSP algorithms are suggested. This enables existing DisCSP algorithms to be used practically in the framework. In the adaptations and improvements, existing DisCSP algorithms are reviewed. Their possible limitations when being used in the Web service environments are analysed and addressed adequately. - New DisCSP algorithms to solve the QoS management problem are proposed. These newly proposed algorithms can model different preference levels of service providers and work in dynamic environments where compositions and constraints on QoS variables can be added or removed. Optimization of preferences of service providers can also be achieved by using some of the algorithms. - A novel verification mechanism, for checking the conformance of service providers to a DisCSP algorithm, is developed, and thus allowing the algorithm to be used practically in an untrusted environment such as that of the Web services.