Scalable Load Balancing Scheme for Distributed Controllers in Software Defined Data Centers

We propose in this paper a hysteresis multiple-threshold-based load balancing (LB) system intended to distributed Software-Defined Networking (SDN) control plane network architecture. The considered LB system is based on Markov chain model and is governed by a control policy using a set of scaling out/in thresholds to evenly distribute traffic between an overloaded SDN controller (client controller) and a lightly loaded neighboring controller (server controller). This is achieved to circumvent large discrepancy in resource utilization through dynamically adapting the global available capacity. The proposed LB scheme aims to achieve multi-objectives tradeoff relevant to scalability, high availability, agility, flexibility, resource utilization, blocking probability and power saving without incurring significant overhead. We highlight through numerical investigations the effectiveness of our proposed model. This is achieved by means of transient and steady state analysis, based on appropriate performance metrics such as average aggregated capacity, transition rate (between client and server) and blocking probability. We show also how the proposed LB scheme performs the right scaling and resource provisioning decisions with respect to specific requirements.