AI-enabled SD-WAN: the case of Reinforcement Learning

Abstract

Traffic Engineering in WAN infrastructures is critical for the efficient management of costly resources and for guaranteeing acceptable QoS levels to applications. SD-WAN has recently emerged as a key solution to manage enterprise WANs, allowing fine-grained, policy-based control over traffic flows. In this paper, we propose a framework based on Reinforcement Learning for the effective use of multiple channels connecting distributed sites of a company. We evaluate it in a realistic, emulated network with a centralized SDN controller. Results show that under heavy load conditions, our approach leads to a 33% reduction in the number of QoS policy violations compared to a benchmark approach. Smaller average latency and connectivity costs are also obtained.