Using SDN and reinforcement learning for traffic engineering in UbuntuNet Alliance

Software Defined Networking (SDN) provides opportunities for dynamic and flexible traffic engineering. This paper discusses how UbuntuNet Alliance National Research and Education Networks (NRENs) can improve bandwidth utilization and reduce inter-NREN latencies through implementation of SDN-based traffic engineering and applying network metrics in selection of inter-NREN paths. Additionally, the paper looks at the utility of applying Reinforcement Learning to path selection, using network data obtained through an SDN controller. Results from simulations using the UbuntuNet topology show an increase in total throughputs when multipath is employed. Furthermore, simulation results show that where latency is the key metric for computing rewards, lower latencies are achieved.