Software-defined networking (SDN) has become a key architecture for next-generation networks, offering centralized control and flexible management. However, existing static switch-to-controller assignment policies often lead to load imbalance, causing long-tail latency and degrading the performance of SDN systems.
To address these issues, a research team led by Deke Guo recently published their new research in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.
The team proposed TACC (Time-Adaptive Changing Controller), a novel load balancing mechanism that empowers switches to dynamically probe and select controllers based on real-time load conditions. By adaptively adjusting the probe window, TACC maintains load balance among controllers under dynamic traffic patterns.
In this research, the authors observed that static assignment between switches and controllers fails to accommodate variable traffic loads and heterogeneous controller processing capacities, leading to significant response time variance. The prior works always use the fixed period to change the mapping relationship to balance the load among controllers. TACC allows each switch to proactively probe two randomly selected controllers and choose the less-loaded one for the next period. The probe interval itself is dynamically adjusted based on feedback delays, helping reduce unnecessary probing overhead while promptly reacting to overloads.
Experiments were conducted using Python-based simulations with a network of 300 switches and 40 controllers under varying traffic conditions. The results demonstrated that TACC achieves superior load balancing and ensures that over 99% of requests meet the target delay requirement of 200 ms, outperforming several representative baseline methods.
Future work could include extending TACC for deployment in real SDN platforms and evaluating its effectiveness in large-scale production environments. The team also plans to investigate how TACC could be enhanced to incorporate additional metrics, such as energy efficiency and fault tolerance, into the load balancing decision process.
Frontiers of Computer Science
Experimental study
Not applicable
Adaptive load balance scheme for the distributed control plane in SDN
15-Sep-2026