Reducing Redundant Control Messaging for Fast Response in SDN

摘要

The control mechanism of SDN tends to be distributed when the network scale expands. As the number of controllers grows, how to efficiently respond to network events in a timely fashion becomes challenging. Specifically, distributed controllers should reach a consensus on network states before making a network update decision in legacy consensus algorithm, which causes unexpected latency. While eventual correctness algorithms relies on Gossip schemes for fast recovery from network failures, leading to highly redundant controller responses when handling one single network event. In this paper, we aim at developing efficient control algorithm that eliminates unnecessary controller processing, facilitating rapid response to network events. To this end, we design a joint controller selection and message forwarding scheme, which identifies a subset of controllers for control message dissemination. An optimization problem is formulated with the objective of minimizing the total control overhead, which is found to be reducible to the K-center problem. Simulation results show that our method reduces up to 61% of the control overhead compared with legacy benchmarks.