A Two-Stage Controller Placement Approach for Software-Defined Networking: Toward Metaheuristic-Driven Multi-Objective Optimization

Software-defined networking (SDN) decouples the control plane from the data plane, enabling centralized network management and programmability. However, as the network scale continues to expand, a single controller becomes a performance bottleneck and a single point of failure. To address this, multi-controller architectures have been widely adopted, where the number and placement of controllers critically determine overall network performance. Existing controller placement approaches face three key limitations, which are optimizing a single objective, requiring manual specification of controller counts, and relying on random initialization. The approaches may then lead to suboptimal latency, poor load balancing, and an excessively large maximum control domain size. To overcome these challenges, this paper proposes a two-stage metaheuristic optimization approach for controller placement in SDN. The proposed approach integrates Modified Density Peak Clustering (MDPC) with an improved Non-Dominated Sorting Genetic Algorithm (NSGA-II) to overcome the limitations of conventional approaches in terms of latency, load balancing, and maximum control domain size. In the first stage, the MDPC algorithm adaptively partitions the network into subdomains and identifies candidate controller locations, which autonomously determines the initial number and placement of controllers while establishing the initial switch-to-controller mapping, by comprehensively considering node density and distance factors. In the second stage, the improved NSGA-II algorithm co-optimizes controller locations and switch assignments with respect to multiple objectives, including average latency, worst-case latency, load imbalance, and maximum control domain size. Extensive experiments conducted on real-world network topologies demonstrate that the proposed approach achieves competitive performance compared with traditional Density Peak Clustering, K-means, standard NSGA-II, and random placement algorithms in the evaluated metrics. The results confirm the effectiveness of the proposed optimization approach for controller placement in SDN, offering improvements in network performance and domain balance.