Adaptive Cross-Cloud Service Composition with the Bidirectional QoS Updates

With the increasing number of cloud services from different providers, service composition across clouds in dynamic environments presents significant challenges. The fact that each of the cloud services has different values of QoS attributes further increases the complexity. Particularly, current approaches cannot efficiently deal with situations when failure between a pair of composed services occurs or the QoS values of the cloud services change in distributed service executions. In this paper, an adaptive cloud service composition in distributed service executions is proposed to improve the traditional Q-learning model and algorithm. It uses a method called the bidirectional QoS updates on composite services which simultaneously considers the QoS update probability and the time cost of selecting the optimal cloud service pairs. The experimental results show that this approach has higher average cumulative rewards and lower average service composition time than the existing approaches.