Feasibility and reliability of peercloud in vehicular networks: A comprehensive study

Abstract

Advanced computing capabilities embedded in modern vehicles enable them to accommodate a variety of intelligent transportation systems and real-world applications that help improve driving safety and compliance with road regulations. However, some of these applications are computationally demanding, and the local processing capabilities of vehicles may not always be enough to support them. To address this issue, existing research has proposed offloading the excessive workload to other computing facilities, such as nearby base stations, roadside units, or remote cloud servers. Still, these facilities have several limitations, including frequent unavailability, congestion, and high fees. In this paper, we explore a more pervasive and cost-effective solution: offloading excessive workloads to nearby peer vehicles via peer-to-peer connections. This approach, referred to as peercloud-vehicle, is an extension of the peercloud approach, which has been proposed for mobile social networks in the literature. The objective of this work is to have a comprehensive study on the feasibility and reliability of vehicle-to-vehicle offloading. First, we analyze two real-world vehicular network datasets to study the robustness of the vehicle contacts and estimate contact durations with deep learning-based regression methods. Second, we design reliable vehicle-to-vehicle offloading approaches based on two optimization objectives: min-delay task offloading to minimize the overall execution delay, and cost-aware task offloading to minimize the cost of task offloading. Experimental results based on real-world datasets demonstrate that peercloud-vehicle significantly outperforms existing approaches.