Machine Learning-Driven Resource Orchestration for Fog and Edge Computing in SDN Environments

Abstract Fog and edge computing, integrated with software-defined networking (SDN), provide a robust framework for supporting latency-sensitive applications in distributed environments. This paper proposes a machine learning-driven resource orchestration framework that optimizes task offloading and resource allocation across fog and edge nodes using a distributed optimization approach. By combining Lyapunov optimization with machine learning techniques, such as reinforcement learning and predictive modeling, and leveraging SDN’s centralized control, our method minimizes latency and energy consumption while ensuring quality of service (QoS). The framework employs a mixed integer non-linear programming (MINLP) model, enhanced with a heuristic-based relaxation for scalability, to manage resources dynamically in fog-enhanced edge networks. Simulations on fog-edge-cloud topologies demonstrate that our approach achieves lower latency and higher resource efficiency compared to centralized methods, validated through extensive performance evaluations, highlighting the role of machine learning in adapting to dynamic workloads and fog computing in extending computational capabilities closer to end-users.