Reinforcement learning-based DoS mitigation in software defined networks

Abstract

A software defined network (SDN) is an OpenFlow-based network that initiates innovative traffic engineering and also simplifies network maintenance. Network security is still as stringent as that of traditional networks. A denial of service (DoS) attack is a major security issue that makes an entire network’s resources unavailable to its intended users. Blocking the flows based on the number of flows per port threshold was the most common method employed in the past. At some occasions legitimate traffic also takes the huge flow will punish by default rules. In order to address this issue, I proposed a reinforcement learning-based DoS detection model that detects and mitigates huge flows without a decline in normal traffic. An agent periodically monitors and measures network performance. It also rewrites the flow rules dynamically in the case of rule violation.