Optimal multi‐controller placement strategy in SD‐WAN using modified density peak clustering

Abstract

In software-defined wide area networks (SD-WANs), finding an optimal number of controllers and their placement is termed as the controller placement problem. The existing solutions based on multi-objective optimisation techniques incur high computation time. Heuristic algorithms based methods quickly converge to local optimum solutions and clustering-based approaches demand that the number of controllers required to be predefined. In SD-WANs, computationally simple solutions for controller placement can be developed using non-parametric clustering algorithms. The proposed modified density peak (modified-DP) clustering algorithm decides the multi-controller placement for the selected networks from the Internet Topology Zoo. The computed performance metrics are compared with hierarchical k-means, modified affinity propagation (modified-AP), and basic DP (basic-DP) clustering algorithms. The criteria importance through inter-criteria correlation method calculates weights of each performance metric, and a normalised weighted additive utility function computes the utility value for the controller placement using each algorithm. The modified-DP computes the minimum utility value and optimal placement of SDN controllers compared to hierarchical k-means, modified-AP, and basic-DP. The Cbench tool calculates the maximum throughput for the controller placement in each network topology. The calculated fairness in the utilisation of the controllers justifies the optimal controller placement using modified-DP.