A survey of deep reinforcement learning based routing optimization in SDN

Software Defined Network (SDN) research has grown due to exponential internet traffic growth. Network flexibility and efficiency are improving as hardware-centric networks become software-centric. SDN is gaining popularity due to cloud computing and 5G technology. Thus, many researchers have worked to strengthen SDN performance. Researchers have also focused on using artificial intelligence in various fields, resulting in much ongoing research. The Deep Reinforcement Learning (DRL) technique has garnered significant attention in multiple industries, particularly the network sector. An understanding of how the traffic load of network switches affects network performance is gained by the Deep Reinforcement Learning (DRL) agent. After that, it determines how to set up the link weights so that the overall network's latency and packet loss are ideal. The routing paths are established by the SDN controller using a predefined collection of link weights. After that, it deploys the flow rules to the SDN-enabled switches. This paper aims to provide a concise introduction to SDN and DRL technology. It will accomplish this by identifying and analyzing recent studies that employ DRL for optimized routing in SDN. Implementing DRL techniques in SDN routing demonstrates enhanced efficiency in enforcing routing within dynamic networks. This approach can achieve optimal resource utilization while delivering exceptional Quality of Service (QoS).