When 3G Meets VANET: 3G-Assisted Data Delivery in VANETs

In this paper, we consider a sensory data gathering application of a vehicular ad hoc network (VANET) in which vehicles produce sensory data, which should be gathered for data analysis and making decisions. Data delivery is particularly challenging because of the unique characteristics of VANETs, such as fast topology change, frequent disruptions, and rare contact opportunities. Through empirical study based on real vehicular traces, we find an important observation that a noticeable percentage of data packets cannot be delivered within time-to-live. In this paper, we explore the problem of 3G-assisted data delivery in a VANET with a budget constraint of 3G traffic. A packet can either be delivered via multihop transmissions in the VANET or via 3G. The main challenge for solving the problem is twofold. On the one hand, there is an intrinsic tradeoff between delivery ratio and delivery delay when using the 3G. On the other hand, it is difficult to decide which set of packets should be selected for 3G transmissions and when to deliver them via 3G. In this paper, we propose an approach called 3GDD for 3G-assisted data delivery in a VANET. We construct a utility function to explore the tradeoff between delivery ratio and delivery delay, which provides a unified framework to reflect the two factors. We formulate the 3G-assisted data delivery as an optimization problem in which the objective is to maximize the overall utility under the 3G budget constraint. To circumvent the high complexity of this optimization problem, we further transition the original optimization problem as an integer linear programming problem (ILP). Solving this ILP, we derive the 3G allocation over different time stages. Given the 3G budget at each time stage, those packets that are most unlikely delivered via the VANET are selected for 3G transmissions. We comprehensively evaluate our 3GDD using both synthetic vehicular traces and real vehicular 3G traces. Evaluation results show that our approach outperforms other schemes under a wide range of utility function deflations and network configurations.