Using lexicographic preorder for pareto search in QoS-aware web service composition

In web service composition, there exist various web services that provide the same functionalities but differ in QoS parameters. Hence there exist several execution plans with different QoS attributes which fulfill user constraints. We propose in this paper an algorithm for selecting an execution plan which satisfies as much as possible user end-to-end QoS requirements and preferences. The algorithm we propose is based on pareto search. An execution plan is pareto optimal for a given qos user requirements and preferences if it is not possible to improve a qos attribute without deteriorating at least one qos attribute. Unlike existing propositions which bind to a service qos vector a unique value (utility function), our approach relies on an utility vector to each execution plan where each element of a vector represents the utility of the plan to the corresponding QoS attribute. However, since we can be faced with incomparable utility vectors, we propose to use a lexicographic preorder for pareto search which corresponds to a preference pre-order among quality attributes given in user request. Experiments on large problem instances demonstrate the scalability and the effectiveness of the approach.