INTELLIGENT TRAFFIC ENGINEERING IN SOFTWARE-DEFINED NETWORKS USING FEDERATED DEEP REINFORCEMENT LEARNING

Abstract

In this context, Software-Defined Networking (SDN) has been recognized as a key enabling platform for future programmable networks. Nevertheless, traditional traffic engineering techniques often struggle to meet the requirements of large-scale, multi-domain environments while ensuring the privacy of network traffic. In this paper, we propose a novel framework for intelligent traffic engineering using Federated Deep Reinforcement Learning (FDRL). In this context, the key novelty of this paper is that the proposed technique allows distributed SDN controllers to collaborate and learn the optimal routing strategies without the need to exchange raw network traffic information. This is achieved by integrating federated learning and deep reinforcement learning techniques. Simulation results demonstrate that the proposed FDRL framework significantly improves network performance compared with traditional shortest-path routing and centralized DRL methods, achieving higher throughput, lower latency, and reduced packet loss. These results indicate that federated deep reinforcement learning provides an effective solution for intelligent traffic engineering in next-generation SDN environments.