On Deep Reinforcement Learning for Traffic Engineering in SD-WAN

The demand for reliable and efficient Wide Area Networks (WANs) from business customers is continuously increasing. Companies enterprises use WANs to exchange critical data between headquarters, far-off branches cloud centers. Many solutions have been proposed over the years, such as: leased lines, Frame Relay, Multi-Protocol Label Switching (MPLS), Virtual Private (VPN). Each solution positions differently in trade-off reliability, Quality of Service (QoS) cost. Today, emerging technology WAN Software-Defined Networking (SD-WAN) that introduces (SDN) paradigm into enterprise-network market. SD-WAN can support differentiated services public by dynamically reconfiguring real-time network devices at edge according measurements service requirements. On one hand, reduces high costs guaranteed QoS (as MPLS), without giving away reliability practical scenarios. other, it brings numerous technical challenges, as implementation Traffic Engineering (TE) methods. TE critically important not only efficiently orchestrate traffic among devices, but also keep their always available. In this work, we develop different kind algorithms with aim improving performance an based terms availability. We first evaluate baseline algorithms. Then, implement deep Reinforcement Learning (deep-RL) overcome limitations approaches. Specifically, three kinds deep-RL algorithms, which are: policy gradient, TD- λ Q-learning. Results show a algorithm well-designed reward function capable increasing overall availability guaranteeing protection restoration SD-WAN.