QoS awareness and adaptation in service composition

The dynamic nature of a Web service execution environment generates frequent variations in the Quality of Service offered to the consumers, therefore, obtaining the expected results while running a composite service is not guaranteed. When combining this highly changing environment with the increasing emphasis on Quality of Service, management of composite services turns into a time consuming and complicated task. Different approaches and tools have been proposed to mitigate the impacts of unexpected events during the execution of composite services. Among them, self-adaptive proposals have stood out, since they aim to maintain functional and quality levels, by dynamically adapting composite services to the environment conditions, reducing human intervention. The research presented in this Thesis is centred on self-adaptive properties in service composition, mainly focused on self-optimization. Three models have been proposed to target self-optimization, considering various QoS parameters, the benefit of performing adaptation, and looking at adaptation from two perspectives: reactive and proactive. They target situations where the QoS of the composition is decreasing. Also, they consider situations where a number of the accumulated QoS values, in certain point of the process, are better than expected, providing the possibility of improving other QoS parameters. These approaches have been implemented in service composition frameworks and evaluated through the execution of test cases. Evaluation was performed by comparing the QoS values gathered from multiple executions of composite services, using the proposed optimization models and a non-adaptive approach. The benefit of adaptation was found a useful value during the decision making process, in order to determine if adaptation was needed or not. Results show that using optimization mechanisms when executing composite services provide significant improvements in the global QoS values of the compositions. Nevertheless, in some cases there is a trade-off, where one of the measured parameters shows an increment, in order to improve the others.