Enhanced Placement-Aware Dynamic Load Balancing for Scalable and Resilient SDN Controller Placement Optimization
TL;DR
A detailed comparative evaluation of six controller placement strategies like random placement method, degree centrality, latency-aware, hierarchical, k-median, and genetic algorithm using two widely adopted SDN controllers shows that load balancing improves latency in most configurations and consistently enhances throughput.
Abstract
Software defined networking (SDN) introduces a flexible network control paradigm by separating the control plane from the data plane. Due to its improved programmability and scalability, it raises two important challenges: determining optimal controller placement and ensuring efficient load distribution among multiple controllers. These factors directly influence key performance metrics such as latency, throughput, and recovery time during failures. This work presents a detailed comparative evaluation of six controller placement strategies like random placement method, degree centrality, latency-aware, hierarchical, k-median, and genetic algorithm using two widely adopted SDN controllers. They are Open Network Operating System(ONOS) and OpenDaylight (ODL). Experiments are conducted on a 500-node fat-tree topology under two operating modes: without load balancing (No-LB) and with dynamic control-plane load balancing (With-LB). In addition, a failure-aware extension of the genetic algorithm is incorporated to evaluate time-to-recovery (TTR) under simulated controller failures. The results show that load balancing improves latency in most configurations and consistently enhances throughput. The best performance is achieved by ONOS combined with genetic algorithm or latency-aware placement under load balancing, reaching a minimum latency of 5.16 ms and an average RTT of 2.11 ms. OpenDaylight demonstrates stable and predictable behaviour across all placement strategies, although with slightly higher latency compared to ONOS. A notable observation is the degradation in performance under hierarchical placement when load balancing is enabled, where latency increases by approximately 11% for ONOS and 13% for OpenDaylight. Under failure conditions, the recovery time ranges between 242 ms and 277 ms depending on the controller and configuration. These findings provide practical insights into selecting suitable controller-placement and load-balancing combinations for both latency-sensitive and reliability-focused SDN deployments.