Routing Optimization using Deep Reinforcement Learning in Wireless Software-Defined Edge Network

Abstract

Technological advancements, increased consumer demands, and the provision of various innovative services have contributed to an increase in network data traffic. On the other hand, the demand for uninterrupted networks and better QoS factors are on the rise. With the increase in these demands, the need for routing optimization has been on the rise, especially in the Wireless Network. Traditional routing protocols are slow to adapt to network changes leading to disruption in the services. The emergence of Software-Defined Networks makes it possible to combine the central management of a controller with artificial intelligence. Besides, the benefits of SDN can be coupled with that of Edge Computing to address scalability issues and reduce the latency of the OpenFlow Communications between the controller and the switches. We propose a routing solution based on Reinforcement Learning to better adapt to the traffic conditions and make dynamic decisions based on various QoS factors such as delay, link utilization, and packet loss in a Wireless Software-Defined Edge Network. The results affirm the superiority of our solution over the conventional routing method reliant on Dijkstra's algorithm (shortest hop count), reducing the delay by a factor of 4. This solution also improves the link utilization of the paths taken.