GDN-Based Dependency Task Offloading for Latency-Sensitive Vehicular MEC Networks

This paper investigates the critical challenge of dependent task offloading in vehicular edge computing, a scenario where computational tasks with complex interdependencies must be efficiently distributed across vehicular networks while simultaneously meeting stringent low-latency requirements and adapting to dynamic network conditions. Conventional task offloading approaches often fail to adequately balance the intricate topological relationships among interdependent tasks with the real-time state of heterogeneous edge resources, leading to suboptimal scheduling performance, increased latency, and inefficient resource utilization. These limitations are particularly pronounced in highly mobile environments characterized by rapidly changing network topologies and unpredictable vehicle movements. To address these challenges, we propose a novel Graph Neural Network-based Dependency-Task Offloading (GDTO) algorithm, which introduces a two-stage optimization framework. First, it intelligently prioritizes subtasks by comprehensively analyzing both task-specific attributes and system information. Second, it leverages a hybrid Graph Convolutional Network-Deep Q-Network (GCN-DQN) deep reinforcement learning framework to dynamically model task dependencies as graph structures while enabling adaptive decision-making in real time. This innovative integration allows GDTO to efficiently coordinate offloading decisions across multiple vehicles and roadside units (RSUs), accounting for spatial-temporal constraints and resource heterogeneity. Finally, comparative simulation experiments with four comparative algorithms demonstrate that the GDTO algorithm can effectively reduce the total task offloading and scheduling latency across various parameter environments.