Performance Modelling and Optimization of Controller Cluster Deployments in Software-Defined WAN

As a novel network paradigm, Software-Defined Networking (SDN) decouples control logic from data forwarding devices, and introduces a separate control plane to manipulate substrate switches via southbound interface typically OpenFlow. This paradigm offers numerous benefits for wide area networks (WAN), like promoting application performance and reducing deployment costs, but gives rise to severe performance bottlenecks of logically centralized control plane. This paper is motivated to propose an accurate performance model of controller clusters in software-defined WAN. In particular, we model the packet-in message processing of a controller cluster as an M/M/n queue based on investigations into its packet-in message arrival process. Subsequently, we build an optimization model of controller cluster deployments, and solve the optimal number of controllers in the cluster based on the proof of the convexity of its objective function. Finally, our proposed queueing model is evaluated with the prevalent controller benchmark OFsuite_Performance by experiments. Experimental results indicate that our proposed model provides a more accurate approximation of controller cluster performance. Moreover, we obtain the optimal number of controllers in the cluster for different varying parameters, which offers guidelines for controller cluster deployments in software-defined WAN.