Research On SDN Multi-controller Deployment Problem

Different from the traditional network,Software Defined Network separates control and forwarding functions.The controller is the core device of the control plane and is mainly responsible for routing decision.The switch is an important part of the data plane and is mainly responsible for data forwarding.At present,the architecture of control plane is divided into single controller architecture and distributed multi-controller architecture.Compared with the single controller architecture,the distributed multi-controller architecture effectively improves the service quality of the network,but it also brings some problems.For example,in distributed multi-controller architecture,the number and placement position of controllers will affect the performance of the whole network.Therefore,how to properly place the controllers has become one of the key research issues.For the flat control mode in distributed multi-controller architecture,by improving the density peak clustering algorithm,this thesis proposes a flat multi-controller placement algorithm MCPDP(Multi-controller Placement Algorithm based on Density Peak).Firstly,the proposed MCPDP algorithm uses the idea of information entropy to determine the dc value,so as to find the clustering center better.Secondly,the number and position of controllers are determined automatically by finding the jump points in the decision graph.Finally,the reliability of the control path from all switches to each controller is calculated,and each switch is assigned to the management domain of the corresponding controller,so as to improve the reliability of the entire network control path.Simulation results show that MCPDP algorithm can effectively reduce network delay and improve network reliability.For the hierarchical control mode in distributed multi-controller architecture,by improving the Affinity Propagation Clustering Algorithm,this thesis proposes a hierarchical multi-controller placement algorithm MCPAP(Multi-controller Placement Algorithm based on Affinity Propagation).Firstly,the proposed MCPAP algorithm uses the shortest path distance between nodes to construct the similarity matrix,and preliminarily determines the deployment number and location of local controllers through iterative calculation.Secondly,the reliability of the control path from all switches to each local controller is calculated,and each switch is assigned to the management domain of the corresponding local controller,so as to improve the reliability of the entire network control path.Then,the heuristic algorithm is used to change the preference p to find the local controller placement scheme which minimizes the load difference between controllers,and this scheme is taken as the final deployment scheme of the local controller to ensure the load balance of the whole network.Simulation results show that MCPAP algorithm can reduce delay,improve reliability,and ensure network load balance.