Enhancing QoS for Dynamic Load Balancing in 6G Networks Using SDN and Deep Reinforcement Learning
Mohamed Amine Hechmi, Sonia Ben Rejeb, Sami Tabbane · 2024
The architecture of Software-Defined Networks (SDN) has become crucial for the flexible and dynamic management of modern networks. With the advent of 6G, the complexity and demand for higher performance necessitate sophisticated load balancing mechanisms. SDN controllers are central to resource management, but their efficiency hinges on the optimal distribution of workloads. Effective load balancing enhances resilience, reduces latency, and maximizes resource utilization. In 6G, the diverse applications require adaptive and intelligent methods. Traditional load balancing methods, though robust, face limitations in dynamic environments. The main challenge is to distribute workloads optimally among SDN controllers to avoid bottlenecks, as imbalances can lead to high latency and packet loss, impacting Quality of Service (QoS). Existing methods struggle to adapt quickly to unpredictable traffic variations. This paper proposes a hybrid approach combining Deep Q-Learning (DQN), Long Short-Term Memory (LSTM) networks, and Proximal Policy Optimization (PPO) for load balancing among SDN controllers. DQN learns optimal policies in real-time, LSTM predicts future workloads for proactive distribution, and PPO continuously optimizes strategies based on current and predicted conditions. Simulations and real-world experiments show this hybrid approach significantly improves SDN network performance and QoS, reducing latency, enhancing resource utilization, and increasing resilience to traffic variations.