Software-Defined Networking Application with Deep Deterministic Policy Gradient

Software-Defined Networking (SDN) simplifies hardware-centric network architecture by employing forwarding devices (switches), SDN controller, and SDN applications. Depending on the SDN application that manages the controller, controller can turn a switch to act as a switch, router, firewall, etc. SDN also enables switch to access more network information through controller. Our proposed SDN application consists of traffic monitoring module and routing module to optimize the network. Traffic monitoring module will monitor switches' port utilization and routing module uses deep reinforcement learning agent trained with deep deterministic policy gradient to manage switch forwarding. Relevant flow entries are issued to OpenFlow switches to manage switch forwarding. Agent changes the routing according to the switches' port utilization and learn the best routing to minimize packet loss. The proposed method will be implemented and evaluated in the future work using Mininet to emulate data layer (hosts, switches) and open source SDN controller Ryu.