Failover mechanisms for distributed SDN controllers

Distributed SDN controllers have been proposed to address performance and resilience issues. While approaches for datacenters are built on strongly-consistent state sharing among controllers, others for WAN and constrained networks rely on a loosely-consistent distributed state. In this paper, we address the problem of failover for distributed SDN controllers by proposing two strategies for neighbor active controllers to take over the control of orphan OpenFlow switches: (1) a greedy incorporation and (2) a pre-partitioning among controllers. We built a prototype with distributed Floodlight controllers to evaluate these strategies. The results show that the failover duration with the greedy approach is proportional to the quantity of orphan switches while the pre-partitioning approach, introducing a very small additional control traffic, enables to react quicker in less than 200ms.