Research On Routing Optimization Of SDN Based On Reinforcement Learning Method

With the development of Information Technology,Machine Learning has already become a necessary assistant of modern industry.Reinforcement Learning as an important branch of Machine Learning,it is widely used in intelligent network management.The amount of data forwarding and transmitting on traditional network were increasing tremendously,which made the traditional network to bear massive data stream.In order to solve the traditional network routing problem,with the combination of the Software Defined technology and the Reinforcement Learning,a new SDN(Software Defined Network)framework is created.It has rapidly become one of the main research directions in recent years.SDN as the trend of future network development,it has separated design for control and forwarding modules.Because of SDN advanced features,it has already let many scholars engaging the research of routing optimization in SDN.Reinforcement Learning has excellent ability in choosing actions,it is suitable for the processing network routing optimization.The purpose of the research is to optimize the performance of SDN network without changing the device as much as possible.In the thesis,Reinforcement Learning method which called SARSA Learning was researched in Software Defined Network architecture.SARSA Learning is a typical Reinforcement Learning algorithm.Through the On-Policy exploring and learning of the network environment,SARSA Learning algorithm can derive the optimal routing decision of unknown network environment,in this way,the network data routing and forwarding can be effectively completed.A simulated topology for SDN and the data routing interfaces were designed in this thesis.A SARSA Learning Routing algorithm V-G Routing(variable greedy function within SARSA Learning routing,V-G Routing)was deployed into RYU-SDN Controller.The V-G Routing algorithm preserves the efficiency of the SARSA Learning.The purpose of V-G Routing introduces a variable factor to ?-Greedy function.The Annealing Algorithm was imported to the algorithm to calculate the probability of the agent choosing action.The V-G Routing algorithm can be dynamically calculated to represent the priority of the current state in SDN network to solve the problem of SDN network optimal route selection,which can avoid long package waiting queue and reduce SDN network congestion and improve the link transmission speed.The Variable Greedy function makes the algorithm more suitable to the network environment.The experimental results verify the effectiveness of the algorithm.