P_MOEA:一种多目标决策辅助遗传算法用于服务组合QoS优化

Abstract

For the QoS management of service composition, one of the most important problems is how to find the optimal solutions for the service composition under certain SLA constraints. There are two main challenges existing for this optimization problem of SLA constraint service composition. At first, since many QoS dimensions are involved to describe the qualities of service composition, it is a typical multi-criteria decision making problem, where effective ranking model is required to rank candidate solutions and to identify optimal solutions. On the other hand, it is also a combinatorial optimization problem that is NP-hard, where the efficient and scalable algorithm is necessary for finding optimal solutions from large scale service composition candidates. In most traditional approaches, two kinds of techniques have been used to define the optimal solutions for service composition, i.e. the linear utility function and Pareto dominance relation. In the former, precise numeric weights are required for the definition of utility function, which is a hard task for users especially when the number of QoS dimensions increase. In the latter, weights assignments are not necessary and a set of skyline solutions are identified as optimal results. However, the scale of results set is uncontrollable, when the problem scale increases the number of skyline solutions will become too many to provide any insights for users to make the final selection. To cope with these drawbacks of traditional methods, in this paper we firstly introduce the PROMETHEE model into this challenging optimization problem. By combining it with Pareto optimization, we redefine optimization objective as top-k PROMETHEE optimal solutions based on skyline computation. Then, we present an efficient evolutionary algorithm: P_MOEA, which can efficiently find the Top-k optimal solutions for large scale service composition problem. We evaluate the effectiveness, efficiency and scalability of our approach with comprehensive analytical and experimental study.