Collaborative Mobile Computation Offloading to Vehicle-Based Cloudlets

This paper investigates collaborative computation offloading in a vehicular network. Although there are increasingly more smart vehicles on roads, a significant number of legacy vehicles that are not equipped with powerful computing devices are expected to exist for a long time. When mobile devices located in these legacy vehicles require computation offloading, they can offload the tasks to nearby smart vehicles that are available to serve as cloudlet servers. Due to high mobility of the vehicles, multiple tasks of an application may have to be offloaded to different vehicle-based cloudlets. The offloading problem is formulated as a Markov decision process (MDP) by considering the randomness of the vehicle moving speeds and wireless channel conditions. The objective is to minimize the average completion time of the application. The complexity for solving the problem directly, however, is prohibitively high due to the large size of the state space and state transition probability matrix. The problem is solved by exploring the special structure of the state space, which helps reduce the computational complexity. A heuristic solution, namely, site-by-site and task-by-task (SSTT), is then proposed that makes the offloading decisions for individual tasks with much lower complexity. Simulation results show that the proposed SSTT solution not only achieves much lower average completion time, compared to executing all tasks locally and using distance-based offloading decisions, but also significantly reduces the energy consumption of the mobile device.

Automated summary

This paper investigates collaborative computation offloading in a vehicular network, proposing an optimization method for intelligent task offloading considering vehicle mobility and wireless channel conditions.