An Effective Clustering Algorithm for Controller Deployment in SDN

Abstract

Software-defined networks(SDNs) can effectively solve the problem of scalability of current network system. The deployment of multi-controllers in the control plane is a very important task that can directly affect the network performance. In this paper, we research the controller deployment problem based on density clustering algorithm. The clustering by fast search and find of density peaks (CFSFDP) is a clustering algorithm by finding tall density peaks. In this paper, we put forward a controller deployment algorithm based on the CFSFDP algorithm in SDN to decrease propagation delay and communication consumption between switches and between switch and its controller. The delay density is calculated by introducing the k-nearest neighbors and the nearest neighbor with lower density. And the number of controllers is obtained automatically through the number of side lobes of $\gamma$ density curve by the kernel density estimation(KDE). Extensive simulations are conducted on Internet2 OS3E network topology. Compared to other methods, the results certify that our proposed algorithm has better performance considering communication consumption and propagation delay.