Dynamic switch migration algorithm with Q-learning towards scalable SDN control plane

Abstract

Multiple controllers are connected to each other to construct a scalable SDN (Software Defined Network) control plane, and every switch is connected to a dedicated controller in data plane. The load on controllers changes dynamically in accordance to the services varying in switches; and switches need to be migrated from their master controllers to slave controllers to make the load balanced for SDN controllers. A dynamic switch migration algorithm with Q-learning towards scalable SDN control plane is proposed in this paper, which models switch migration problem, redefines parameters of Q-learning based on the switch migration model, and gets decisions of switch migration dynamically with the changing load on controllers using the proposed algorithm in SDN.