A parallel refined probabilistic approach for QoS-aware service composition

Abstract Service composition integrates existing online services to provide a value-added service. With the rapid growth of web services with similar functionalities, Quality of Service (QoS) has emerged as an important quantitative criterion on non-functional aspects. The optimization of QoS-aware service composition, depending on different aggregated QoS attributes has attracted significant attention. The dynamic nature of QoS-aware service composition adds further challenges to the optimization problem. Most existing approaches ignore the diversity of solutions, which have the potential to provide alternative compositions when changes occur. A few works only partially explore the search space and do not consider the optimality of solutions and the computational cost concurrently. To address these issues, we propose a novel reactive approach, called MrEDA, which integrates the estimation of distribution algorithm (EDA), restricted boltzmann machine (RBM), and multi-agent technology. It constructs a refined probabilistic model to diversify alternative solutions and guide the search by adaptively capturing the promising information of a service composition. Meanwhile, multiple agents make use of a flexible parallelism with distinct explorations and adaptive sampling to improve the global optimization and speed up the optimization. The effectiveness and efficiency of our approach for adaptive service composition is validated through an extensive experimental evaluation.