An Efficient Reinforcement Learning Based Approach For SDN Controller Placement Optimization

Abstract

Software-defined networks (SDN) are a new networking paradigm that decouples the control plane from the data plane. This boosts the flexibility of management but also introduces new challenges, such as the placement of the controllers. The controller placement problem (CPP) is concerned with determining the optimal location of controllers in a network. Reinforcement learning models can offer an effective learner for the optimal controller placement in a dynamic environment. Controller placement optimization typically makes use of Deep Q Network (DQN), one of the reinforcement learning algorithms. Double DQN solves DQN’s inherent overestimation. It enhances the learned Q-values’ quality, resulting in more accurate policy decisions. The prioritized Double Deep Q Network (PDDQN) is presented in this paper to enable the model to learn from the most informative experiences. The results of the experiments demonstrate that the proposed model performs 22% better in load balancing and 52% better in latency than the controller placement model based on DQN.