In modern networking, software-defined networks (SDNs) have emerged as a powerful paradigm that separates the control plane from the data plane, enabling centralized and distributed network management. SDNs provide flexibility and efficiency in handling large-scale networks, aiming to optimize resource utilization, reduce energy consumption, and enhance quality of service (QoS). Given the rapid growth in data traffic and the increasing need to minimize response time and energy consumption, developing efficient load balancing strategies has become a critical challenge to ensure network performance and stability. Load balancing plays a vital role in optimizing data traffic distribution across servers and network nodes, preventing congestion, and improving system efficiency. This is especially crucial in large and complex environments such as cloud data centers and distributed networks, where handling high request volumes efficiently is essential. To address these challenges, this paper introduces SDN-PG, a novel dynamic load balancing strategy for SDNs that integrates policy gradient (PG), a reinforcement learning-based optimization method, with dynamic voltage and frequency scaling to enhance energy efficiency and network performance. SDN-PG dynamically optimizes traffic distribution by continuously adapting network policies to real-time fluctuations, significantly improving QoS while minimizing energy consumption. The proposed approach consists of three primary components. The first component is a distribution policy learned via Policy Gradient, enabling adaptive load balancing decisions. The second component involves real-time network monitoring, allowing the system to track and respond to dynamic traffic changes. The third component is an efficient decision-making mechanism, which leverages PG-based policies to reduce computational overhead and optimize response time. To validate its effectiveness, SDN-PG is compared against state-of-the-art methods, including CCA-PSO and DRL-SMS, through simulation experiments. The results demonstrate significant improvements in key performance metrics. SDN-PG achieves a 45.47% and 46.22% reduction in response time, a 14.09% and 11.98% decrease in computational overhead, a 19.47% and 16.38% improvement in energy efficiency, and a 7.6% and 4.24% enhancement in load balancing effectiveness. These findings highlight the practical applicability of SDN-PG in large-scale SDN environments, demonstrating its ability to efficiently balance energy savings and QoS while maintaining optimal load distribution and network stability.

Keywords