Genetic algorithms with particle swarm optimization based mutation for distributed controller placement in SDNs

Abstract

This paper proposes a distributed controller placement problem that finds out the pareto optimal solutions minimizing the switch-to-controller delay, controller-to-controller delay, and controller load imbalance for wide area software defined networks. We introduce a general model that not only considers the controller placements but also the switch assignments, so that this model can further be used to develop many other multi-objective optimization problems such as energy saving, controller migration, or NFV allocation. To solve this problem with huge search space without losing generality, we introduce a Multi-Objective Genetic Algorithm (MOGA) with a particle swarm optimization based mutation function. It maintains a pre-calculated global best position for each single objective, and choose the global best position of an objective that has the best accordance to a parent to guide the mutation of the parent. Evaluations show that our MOGA can generate a pareto frontier with a larger diversity toward the given global best positions in much shorter convergence time than a general MOGA.