Research On Multi-control Lers Effective Deployment Strategy In Software-Defined Network

With the development of the Internet,newly emerged network applications bring explosive growth,which makes traditional network infrastructure face great challenges.Thus,Software defined network(SDN)is presented with new idea of separating the data layer and the control layer.However,research at the present stage is still spring up,and so many problems still need to be solved.For the large-scale WAN,researchers focus on the controller numbers and positions,which mainly consider local latency,throughout,controller balancing and so on.Since the controller is selected from switch nodes,and research remains the original topology structure,the controller deployment problem can be transferred into the domains partition of topology where some existing solutions include K-means,spectral clustering and so on.It can be found that some of these solutions are merely based on the graph data clustering scheme for different topology,but the network performance is not fully considered.To address the the problems,this paper proposes an efficient scheme for SDN controller deployment.This scheme consists of two stages:the sub-domain division and the optimal deployment of SDN controller which combines load_balancing,local resiliency and other performance indicators.Actually,this paper proposes a subdomain partition based on the diffusion distance.Different from the traditional metric,diffusion distance-based partition has the great nature which benefits for load balancing and latency:almost equal nodes number in each partition,more links in the same domain,and less in the different ones.Secondly,unlike the past way to choose clustering centroid as controller positon,this paper proposes a subdomain controller deployment algorithm based on the local resiliency which inspired by the idea of protected switches to find the optimal placement of controller and reduce the impact node failure has on the whole network.As for the experiment,our method is evaluated on Internet2 OS3E,Internet Topology Zoo and compared with different method.Experimental results illustrate that our optimal method has a good performance of global latency,load-balancing and local resiliency.