Advancing Fog-Edge Continuum: A Hybrid Approach Using GRL and Stable Matching for Task Offloading

Fog computing has emerged as a solution to overcome the constraints of cloud-centric models by bringing computational resources closer to IoT devices. Fog computing improves the capabilities of IoT devices by offering intelligent services through offloading mechanisms, which help overcome resource limitations. However, the structural complexity and dynamic nature of the Fog Computing environment still pose a challenge in constructing effective task offloading techniques. In order to achieve optimal task offloading within Fog-Edge continuum this study proposed graph reinforcement learning based architecture. This technique leverages an adaptation of the stable matching approach in association with Graph Reinforcement Learning to take cognitive offloading decisions. The GRL framework models the system as an acyclic graph, and effectively captures the dynamic relationships and heterogeneous characteristics of the network layers. The incorporation of stable matching approach ensures an initial allocation of tasks to resources that is both efficient and mutually preferred. Which is then refined using GRL to further optimize the performance. Analytical assessments indicate that proposed methodology SMGRL effectively minimizes latency by 25% in comparison to conventional heuristic and baseline methods. The framework also demonstrates a 44% improvement in energy efficiency and achieves task completion rates 3% to 12% higher than other algorithms, underscoring its effectiveness in fog computing environments.