A Novel Method for Routing Optimization in Software-Defined Networks

Software-defined network (SDN) is a new form of network architecture that has programmability, ease of use, centralized control, and protocol independence. It has received high attention since its birth. With SDN network architecture, network management becomes more efficient, and programmable interfaces make network operations more flexible and can meet the different needs of various users. The mainstream communication protocol of SDN is OpenFlow, which contains a Match Field in the flow table structure of the protocol, which matches the content of the packet header of the data received by the switch, and completes the corresponding actions according to the matching results, getting rid of the dependence on the protocol to avoid designing a new protocol. In order to effectively optimize the routing for SDN, this paper proposes a novel algorithm based on reinforcement learning. The proposed technique can maximize numerous objectives to dynamically update the routing strategy, and it has great generality and is not reliant on any specific network state. The control of routing strategy is more complicated than many Q-learning-based algorithms due to the employment of reinforcement learning. The performance of the method is tested by experiments using the OMNe++ simulator. The experimental results reveal that our PPO-based SDN routing control method has superior performance and stability than existing algorithms.