Deep Q-Network and Traffic Prediction based Routing Optimization in Software Defined Networks

Software Defined Networking (SDN) is gaining momentum not only in research but also in IT industry representing the drivers of 5G networks, due to its capabilities of increasing the flexibility of a network and address a variety of network challenges, by logically centralizing the intelligence in software-based controllers. However, QoS aware routing is an open issue of SDN-based networks, due to the very large traffic spikes caused by a multitude of short and bursty sessions and large file transfers. Thanks to Machine Learning (ML) techniques, the network performances and utilization can be optimized and enhanced. Neural Networks (NN) and Reinforcement Learning (RL), in particular, have demonstrated great success to cope with complex problems arising in network operation and management. To this end, we exploit in this paper, an SDN-based rules placement approach that aims to dynamically predict the traffic congestion by using mainly NN and learn optimal paths and reroute traffic to improve network utilization by deploying a Deep Q-Network (DQN) agent. In this way, we mathematically formulate the QoS-aware routing. The corresponding optimization problem is to minimize the total delay and link utilization. To solve this optimization problem, a simple yet efficient heuristic algorithm is proposed. The numerical results through emulation demonstrate that the proposed approach can significantly improves the performances in terms of decreasing the link utilization, the packet loss and the delay.