Abstract
Software-defined networking (SDN) has become a key architectural enabler for smart city infrastructure, where a logically centralized control plane governs a distributed forwarding fabric. As smart city deployments scale, a single controller can no longer handle the volume of control traffic, and operators must deploy multiple controllers in a coordinated manner. The question of where, and how many, controllers to install was first framed by Heller et al. (2012) for the single-controller case and has since been extended to the multi-controller setting. Most existing formulations assume light-tailed traffic, yet real smart city traffic exhibits heavy-tailed behavior that conventional models underestimate. This paper evaluates multi-controller placement under a hyperexponential traffic model that captures the bursty, heavy-tailed character of urban control traffic. Using a representative metropolitan topology, we compare three placement strategies — greedy latency minimization, k-means clustering, and a load-aware heuristic — across average latency, controller utilization, and packet loss. The hyperexponential model reveals failure modes that light-tailed models miss, particularly under peak-hour load. We show that load-aware placement reduces tail latency by roughly 28% compared to latency-only optimization, while keeping utilization variance below 0.12.
References
This work is licensed under a Creative Commons Attribution 4.0 International License.