A promising paradigm called vehicular fog computing (VFC) fully uses the processing capabilities of moving and idle vehicles acting as fog servers to increase computation capacity. Vehicles with limited resources can offload complex and computing-intensive applications to closer fog servers compared with the cloud, reducing the response time and price paid for application execution. Most current mechanisms in the VFC field focus on resources in coverage of a RoadSide Unit (RSU) or local cluster and do not consider idle computing resources in other clusters that lead to an increase in the application failure rate. Vehicular networks also face the challenge of handling applications with varying priorities and processing efficiencies. Therefore, predicting a cost-efficient task offloading solution while considering the priority, deadline, energy consumption, and budget of the application is an issue. We propose a two-tier VFC distributed architecture wherein neighboring clusters can share their idle computational resources for distributed task processing, therefore substantially improving the overall network computing resources. In this hierarchical architecture, we also use local Software-Defined Networking (SDN) and global SDN controllers to optimize intra- and inter-cluster task offloading policies. In addition, the task-offloading system is based on the priority of tasks, and priority queues in fog nodes are used to schedule the tasks of applications. Nevertheless, the global SDN controller uses the Double Deep Q-Network (DDQN)-based approach to reduce processing costs. The results indicate that the proposed strategy improves performance in terms of average processing cost, average response time, failure rate, and average energy consumption compared to other baselines.

Keywords