Forward-Backward Inverse Reinforcement Learning for Load Balancing in SDN

Abstract

Software Defined Network (SDN) is a kind of networking technology that dynamically establishes the network for improving network performance. The SDN architecture is composed of three layers which includes application layer, control layer, infrastructure layer. Each layer is subjected to several challenges in terms of improper controller placement, poor scalability, performance degradation, frequent load imbalance among nodes, weak network connectivity, and many more. In this paper a novel Forward Backward Inverse Reinforcement Learning (FBIRL)-based load balancing framework is defined for SDN domain. The load balancing policies are formulated immediately and the possibility of picking the best action at every step is very high. The performance of the proposed FBIRL is tested over Mininet virtual network. The results are found be promising towards the performance metrics like reduced respond time and blocking probability. Simultaneously increase in the success rate achieved in the client tasks completion.