Neighbourhood Centality Based Algorithms for Switch-to-Controller Allocation in SD-WANs

• Published in: 2023 International Conference on Artificial Intelligence, Big Data, Computing and Data Communication Systems (icABCD) pp. 1 - 6
• Main Authors: Adebayo, Isaiah O., Adigun, Matthew O., Mudali, Pragasen
• Format: Conference Proceeding
• Language: English
• Published: IEEE 03.08.2023
• Subjects: Big Data; capacitated controllers; Control systems; neighborhood centrality; Performance evaluation; propagating capacity; Social networking (online); software defined-wide area networks; switch-controller allocation; Switches; Topology; Wide area networks
• DOI: 10.1109/icABCD59051.2023.10220485

The advent of artificial intelligence and big data makes it nearly impossible for large scale networks to be managed manually. To this end, software-defined networking (SDN) was introduced to provide network operators with the infrastructure for achieving greater flexibility and fine-grained control over networks. However, a critical issue to consider when incorporating SDN technology over large-scale networks like wide area networks (WANs) is the allocation of switches to controllers. In this paper, we address the switch-to-controller allocation problem that considers the heterogeneity of controller capacities. Specifically, we propose two neighbourhood centrality-based algorithms for addressing the problem with the aim of minimizing switch-to-controller latency. We also introduce a weighted centrality function that enables fair distribution of load across capacitated controllers. The proposed algorithms utilize centrality-based measures and heuristics to determine the ideal switch-to-controller allocations that consider the propagating capacity of suitable controller nodes. We evaluate the performance of the proposed algorithms on the internet2 topology. The results show that considering the heterogeneity of controller capacities reduces load imbalance significantly. Moreover, by limiting the exploration of the local centrality for each node to a maximum of two-step neighbours the complexity of the proposed algorithm is reduced. Thus, making it suitable for implementation in real-world SD-WANs.