Research On Controller Deployment And Routing Optimization In SDN

Software Defined Network is a new Network architecture proposed under the challenging environment of rapid expansion of Network scale and explosive growth of traffic.In this network environment,how to guarantee Quality of Service effectively becomes a key problem.Thanks to its separation of control and forwarding functions,centralized control in logic and support of network programmability,SDN has a broad application prospect in the field of the next generation Internet.The deployment position of controllers is closely related to the overall network performance.In addition,unintelligent routing rules of traditional routing protocols may cause network congestion.Therefore,how to deploy the controller position reasonably and avoid network congestion has become a key issue in the process of improving SDN network service quality.In order to solve the above problems,the distributed multi-controller architecture in the process of controller placement is studied deeply in this thesis.On the condition of satisfying the dual optimization objectives of delay and load balancing,it improves the traditional clustering algorithm of Affinity Propagation and invented a new multi-controller placement method.The advantage of this algorithm is that the number of controllers and deployment locations can be determined automatically.Firstly,the algorithm calculates the shortest distance by implementing Dijkstra algorithm to measure the similarity relation between controller and switch,and then applies this distance to construct the similarity matrix of clustering algorithm.Secondly,the similarity matrix is used for iterative calculation,and the deployment number and location of the controller are preliminarily obtained.In addition,heuristic algorithm is used to find the optimal controller placement scheme which minimizes the load difference between controllers,So that the whole network to achieve load balance.Finally,in the simulation experiment,different controller deployment algorithms are compared according to the two performance indexes of delay and load,which shows that the algorithm proposed in this thesis can achieve better performance.In addition,when SDN officially enters the stage of use,this thesis proposes a routing optimization framework based on Deep Reinforcement Learning for the optimization of SDN routing,aiming at optimizing the overall network performance,and designs a routing optimization algorithm based on DRL according to the Deep Deterministic Policy Gradient algorithm.Different from other traditional routing algorithms,firstly,this algorithm can detect the network status in normal environment and congestion environment in real time.Second,by rapidly interacting with the environment to try to maximize the cumulative reward,so as to intelligently optimize routing and ensure QoS in different network environments.Finally,it can be verified in the simulation experiment that the algorithm proposed in this thesis can achieve better QoS performance optimization in the normal network environment and in the environment of heavy traffic load.