Deep Reinforcement Learning-based QoS Aware Routing in Software Defined Networks

Software-Defined Networks (SDN) is the nextgen network technology which helps in efficient management of networks. The important concept underlying SDN is the separation of the data plane and control plane. The data plane devices serve as forwarding devices and decision-making is shifted to a controller which operates on the control plane. Recently, machine intelligence is used by researchers to enhance SDN-based centralised control by routing and flow management which helps in optimal network traffic management. In this work, we propose a machine learning-based system to optimize the routing efficiency of a network. A deep reinforcement learning-based routing algorithm called Advantage Actor-Critic (A2C) is used to compute optimal routes. The results are compared with two algorithms namely Dijkstra’s shortest path and deep neural network algorithm called Actor-Critic (AC) Networks. Simulation results using Mininet emulator reveals that routing with A2C provides higher throughput and a lower average delay in comparison with AC.