In this paper, we propose a direction-based vehicular network model, named DVNM, to provide efficient and reliable service to vehicles in Vehicular Fog Computing (VFC). In DVNM, the vehicles in vehicular fog cell are divided into three vehicular subnetworks according to their turning directions in next crossroad. We also propose a task-offloading strategy to improve the task-offloading performance for vehicular network models. In simulations, we analyze the task-offloading performance of vehicular network models. Numerical results show that, for different cases, the average response time in DVNM is saved over 80% and 30% respectively compared with two existing models.