Bayesian Network based Optimal Load Balancing in Software Defined Networks

Abstract

Due to the exponential increase in data volume, network complexity necessitated the requirement of software defined networks (SDN). But with SDN, network and service expansion significantly affect resource management. An efficient resource allocation method is mandatory to account for the random nature of network traffic and the load management across several controllers. However, due to the imprecise and dynamic relationships between resources, reinforcement learning has not been well-served in the context of real-time load in SDN. This paper presents deep reinforcement learning (DRL) technique to a Bayesian network to provide a smart optimization framework for SDN resource management. The Bayesian network use a reinforcement learning method, self-adjusting parameter weight, and automatic parameter weight adjustment to regulate the controller load congestion and anticipate the level of load congestion necessary to achieve load balance. By utilizing the prediction results from reinforcement learning, this algorithm selects the best possible next step. Theoretical analysis of the proper load balancing strategy for SDN is supported by a concurrent examination of existing datasets.