SDN Multipath Routing Algorithm Based On Reinforcement Learning

In the face of the emerging demand of Internet applications,the network architecture is extremely fragile and inadequate.The SDN network architecture has emerged in this context,aiming to change the problem that the existing network infrastructure is difficult to adapt to the needs of the new application.The biggest advantage of SDN is the separation of control from the equipment,which is managed by the centralized controller.So,any deployment of a network strategy is not required to modify the configuration of the network device,but rather the way the user programmatically customizes any desired network strategy.SDN network routing is unified by the controller,so provides convenient conditions for the multipath routing design,at the same time,can avoid the traditional algorithm convergence and computational complexity,but so faces some challenges,especially SDN network with distributed controllers.In the distributed controller SDN,the network model of routing problem will be from a single layer of network structure into a dynamic hierarchical network topology,this obviously makes the traditional routing algorithms can not adapt to the network structure change,need under the new network structure design of efficient routing algorithm,to realize the network load balance and performance opitimization.Aiming at these problems,from the perspective of intelligent learning algorithm,a distributed controller SDN network multipath routing algorithm is proposed in this paper,and the simulation test is used to analyze convergence and computational complexity of the algorithm.The basic ideas and steps of the algorithm are as follows:(1)According to the structure of the controller in the distributed controller SDN network,the network structure model of the hierarchical and sub-zone is constructed as the system model of the routing algorithm.(2)Underneath the controller layer,because the multi-zone network topology information can get,shortest path algorithm in graph theory is adopted to calculate the multi-paths meeting the demand of user's QoS demands.(3)On the top floor controller layer,multipaths of each sub-zone network underlying the controller are nodes,the QoS performance of the path is the weight value of node,the link of the path boundary nodes is edge,so the small path network is abstracted.Then the Q-learning algorithm is used to solve the multipath of satisfying the user's need in the path network.(4)In the virtual experiment network environment built by the Mininet,OpenDaylight controller is selected as the realization platform of the algorithm.The convergence and complexity of the algorithm in the different network sizes are analyzed based on the delay of the data packets between source node and target node.Test results show that the convergence time of the algorithm increases with the increase of the number of nodes,but the numbers of sub-zone network has more influence on the convergence than the sizes of sub-zone network.The performance of the algorithm is proportional to the number of nodes and the edge density.