Developing Optimal SDN Controller Placement in WAN Networks Based on Clustering Approach

Abstract

This research looks into controller placement as a key factor in software-defined wide area network (WAN) deployment. We are trying to figure out the best number of controllers and where to place them in order to meet certain performance objectives, such as delay and cost of deployment. We are using ML algorithms and mathematical approaches to find the optimal placement of controllers to get the most out of the network. We set up two metrics to measure the network: average propagation latency between controllers and switches, and cost of deployment for each controller. Simulations were run on a real-world network called SANReN to test the minimum number of controllers and appropriate placement in order to improve the network's performance without needing expensive and time-consuming physical testing. We applied k means clustering, and compared our approach to previously used methods on the same dataset. Our results from SANReN showed that deploying the k means clustering method drastically decreases computational time used for controller deployment and reduces the chances of failure of controller based on delay of communication as it is one of controller failure reason. These findings will be helpful for network operators to deploy the right number of controllers and their locations for optimal performance.