Controller placement in software-defined WAN using multi objective genetic algorithm

SDN is a promising architecture that can overcome the challenges facing traditional networks. SDN enables administrator/operator to build a simpler, customizable, programmable, and manageable network. In software-defined WAN deployments, multiple controllers are often needed, and the location of these controllers affect various metrics. Since these metrics conflict each other, this problem can be regarded as a multi-objective combinatorial optimization problem (MOCO). A particular efficient method to solve a typical MOCO, which is used in the relevant literature, is to find the actual Pareto frontier first and give it to the decision maker to select the most appropriate solution(s). In small and medium sized combinatorial problems, evaluating the whole search space and find the exact Pareto frontier may be possible in a reasonable time. However, for large scale problems whose search spaces involves thousands of millions of solutions, the exhaustive evaluation needs a considerable amount of computational efforts and memory used. An effective alternative mechanism is to estimate the original Pareto frontier within a relatively small algorithm's runtime and memory consumption. Heuristic methods, which have been studied well in the literature, proved to be very effective methods in this regards. The second version of the Non-dominated Sorting Genetic Algorithm, called NSGA-II has been carried out quite well on different discrete and continuous optimization problems. In this paper, we adapt this efficient mechanism for a new presented multi-objective model of the control placement problem [7]. The results of implementing the adapted algorithm carried out on the Internet2 OS3E network run on MATLAB 2013b confirmed its effectiveness.