Abstract

In the areas of cloud computing, big data and internet of things, composite services are designed to effectively address complex levels of user requirements. A major challenge for composite services management is the dynamic and continuously changing run-time environments that could raise several exceptional situations such as service execution time that may have greatly increased or a service that may become unavailable. Composite services in this environmental context have difficulty securing an acceptable quality of service (QoS). The need for dynamic adaptations to be triggered becomes then urgent for service-based systems. These systems also require trust management to ensure service level agreement (SLA) compliance. To face this dynamism and volatility, context-aware composite services (i.e., run-time self-adaptable services) are designed to continue offering their functionalities without compromising their operational efficiency to boost the added value of the composition.
The literature on adaptation management for context-aware composite services mainly focuses on the closed world assumption that the boundary between the service and its run-time environment is known, which is impractical for dynamic services in the open world where environmental contexts are unexpected. Besides, the literature relies on centralized architectures that suffer from management overhead or distributed architectures that suffer from communication overhead to manage service adaptation. Moreover, the problem of encountering malicious constituent services at run-time still needs further investigation toward a more efficient solution. Such services take advantage of the environmental contexts for their benefit by providing unsatisfying QoS values or maliciously collaborate with other services. Furthermore, the literature overlooks the fact that composite services data is relational and relies on propositional data (i.e., flattened data containing the information without the structure). This contradicts with the fact that services are statistically dependent since QoS values of service are correlated with those of other services.
This thesis aims to address these gaps by capitalizing on different methods from software engineering, computational intelligence and machine learning. To support context-aware composite services in the open world, dynamic adaptation mechanisms are carried out at design-time to guide the running services. To this end, this thesis proposes an adaptation solution based on a feature model that captures the variability of the composite service and deliberates the inter-dependency relations among QoS constraints. We apply the master-slaves adaptation pattern to enable coordination of the self-adaptation process based on the MAPE loop (Monitor-Analysis-Plan-Execute) at run time. We model the adaptation process as a multi-objective optimization problem and solve it using a meta-heuristic search technique constrained by SLA and feature model constraints. This enables the master to resolve conflicting QoS goals of the service adaptation. In the slave side, we propose an adaptation solution that immediately substitutes failed constituent services with no need for complex and costly global adaptation. To support the decision making at different levels of adaptation, we first propose an online SLA violation prediction model that requires small amounts of end-to-end QoS data. We then extend the model to comprehensively consider service dependency that exists in the real business world at run time by leveraging the relational dependency network, thus enhancing the prediction accuracy. In addition, we propose a trust management model for services based on the dependency network. Particularly, we predict the probability of delivering a satisfactory QoS under changing environmental contexts by leveraging the cyclic dependency relations among QoS metrics and environmental context variables. Moreover, we develop a service reputation evaluation technique based on the power of mass collaboration where we explicitly detect collusion attacks. As another contribution of this thesis, we introduce for the newcomer services a trust bootstrapping mechanism resilient to the white-washing attack using the concept of social adoption. The thesis reports simulation results using real datasets showing the efficiency of the proposed solutions.