Research On SDN-Based Multi-controller High Reliability Deployment Algorithm

Software Defined Network(SDN)separates the control power from the network equipments and sends them to the centralized controller for unified management.As the core of SDN,the controller not only responds to requests of upper-layer applications,but also controls the underlying data plane.However,with the increasing of network scale,single-controller architectures face enormous challenges in terms of reliability and scalability.Therefore,multi-controller architectures have emerged.In the multicontroller architecture,the number,deployment locations and the mapping relationship with the switches of controllers directly affect the performance of SDN.Therefore,this paper focuses on the multi-controller deployment problem.The main contents of this paper is as follows:(1)This paper proposes a Reliable Deployment Algorithm based on Optimized Kcenter(RD-KC).In view of the lack of the current static deployment algorithm to optimize the deployment of the controller from two aspects of delay and reliability,this paper first analyzes the impact of controller deployment on the system reliability and design a path reliability optimization strategy to find reliable low-latency transmission paths between nodes.Subsequently,the K-center algorithm is optimized to improve the solution accuracy and reduce the running time.And in the iterative process of the optimized K-center algorithm,the node mapping adjustment step is added to improve the system reliability.The simulation results show that the RD-KC algorithm has better comprehensive performance in terms of transmission delay and system reliability,and the running time is relatively small.(2)We propose a Highly Available Dynamic Deployment Algorithm based on Leader-Follower Mode(HADD-LFM).In order to meet the controller limitation and reduce the transmission delay after failure recovery,this paper adopts the Leader/Follower mode to solve the problem of controller failure.Firstly,in order to maintain the consistency of the migration state among controllers,this paper determines a master-slave controller node through a neighborhood master controller election algorithm based on Paxos protocol;Secondly,a Time Optimal Allocation Algorithm based on Load Constraint is designed to complete the allocation of switch sets under the fault controller.Finally,the follower performs the switch migration operations based on the allocation to complete network failure recovery.In order to reduce the average response delay of the controller,this paper also proposes a network mapping restoration machanism for restoring the mapping relationship between the switch and the controller when the failure controller recovers.For the remapping process between the switch and the controller,an active four-phase smoothing protocol is proposed to reduce the impact of remapping on the network operation.The simulation results show that the HADD-LFM algorithm can ensure that the controller is not overloaded during the fault recovery due to the migration of the switches,and can obtain a lower controller average response delay than other similar algorithms.