Delay-Aware Control Plane Virtual Topology Design of Software Defined-Elastic Optical Network

Abstract

An upper bound on the delay in the control channels is necessary as control state inconsistency significantly degrades the performance of a logically-centralized and physically-distributed control plane. Statistically multiplexing control traffic with the data traffic (in-fiber in-band) or reserving a low-rate control channel does not ensure QoS guarantees in terms of delay specially if the network is congested. Therefore, control traffic must be viewed and treated differently from data traffic. This article presents ac MILP and a heuristic to design distributed SDN-based in-fiber out-of-band control plane for elastic optical network (EON). The objective is to minimize the number of controllers while reserving resources for the control traffic to ensure that delay in the control channel is bounded. The designed inter-controller virtual tree topology avoids control traffic replication. For better resource utilization it requires OEO conversion at every controller node for grooming control traffic and this incurs queuing delay at each controller node. The design takes into account queuing delay along with transmission and propagation delay in control plane links and maintains the total delay within a specified bound. To the best of our knowledge, this is the first work to analyze queuing delay along with transmission and propagation delay in inter-controller links of the control plane. We further present a MILP to make this virtual control plane tree topology survivable against single link failure by utilizing the concept of fundamental cycles. The numerical results show that our proposed control plane design performs better than existing control plane designs where control traffic is statistically multiplexed with data traffic in terms of delay in the control plane and blocking probability in the data plane.