DRLNPS: A deep reinforcement learning network path switching solution

This paper proposes a solution to the problem of switching between different network paths. We choose to switch between multiprotocol label switching (MPLS) and software-defined wide area networking (SD-WAN) connections specifically as they are the mainstream currently. The solution should maintain a service license agreement (SLA) while choosing SD-WAN as long as possible to save cost. Therefore, a deep reinforcement learning solution is proposed that predicts when to switch based on bandwidth availability and quality of service (QoS) parameters like jitter and delay. Results show that double deep Q learning in combination with these parameters are suitable to make a sophisticated decision on link switching between MPLS and SD-WAN.