An efficient DHT-based elastic SDN controller

Abstract

Elasticity in distributed SDN Controllers [1], [2], [3] improves fault tolerance, power efficiency, cost efficiency and scalability. We find that the Elasticon [1] algorithm has high time complexity since it balances the load equally on all the controllers for every event. In this paper, we propose an efficient algorithm for elastic controllers using Chord [4], a Distributed Hash Table (DHT). In our scheme, we do not balance the load on the controllers until a controller is overloaded. The switch ID space forms the ring in the DHT. The controller is assigned an ID based on the range of switch IDs it is responsible for which changes as the switches migrate from an overloaded controller to its adjacent controllers in the ring. If the average load on the controllers falls below a given threshold, we consolidate the switch allocation to controllers so that some of the controllers can be powered off. We compared the performance of Elasticon and the proposed algorithm in terms of the number of migrations needed, the average number of controllers that are powered on, the time taken to run the algorithm and the standard deviation of the load on the controllers. We find that the number of controllers is similar for our algorithm and Elasticon, the migrations are fewer in our algorithm while the time taken by our algorithm is orders of magnitude less than Elasticon.