Multi-objective Hybrid Evolution with Information Entropy Awareness for Controller Placement

In software defined wide-area networks, the number and location of controllers in the optimization model have a significant impact on network performance. Compared with the traditional single-objective model or quasi-multi-objective model that can be transferred to the single-objective model, multiobjective models can provide more comprehensive solutions to the problems by concerning controller deployment at one time, which makes network operators use different solutions to accommodate various scenarios better. In this paper, an actual multi-objective model is built to optimize controller deployment by considering deployment cost, load difference, and propagation delay. To solve this model, we propose an algorithm by specially designing the hybrid initialization method to generate an initial population that balances diversity and convergence. After that, we design the mechanisms of encoding conversion, information entropy awareness, hybrid evolution, and perturbation modification. These mechanisms are particularly constructed for the proposed algorithm to solve the problems in the evolution process and to improve the global search ability of the algorithm for obtaining superior Pareto sets. Finally, we validate the effectiveness and generality of the proposed algorithm by comparing its Pareto sets with those of other algorithms in Internet2 OS3E network from various aspects.