A semi-Markov decision process-based computation offloading strategy in vehicular networks

Mobile computation offloading (MCO) is an emerging technology to offload the resource-intensive computations from smart mobile devices (SMDs) to nearby resource-rich devices (i.e., cloudlets) via wireless access. However, the link duration between a SMD and a single cloudlet can be very limited in a vehicular network. As a result, offloading actions taken by a SMD may fail due to link breakage caused by mobility. Meanwhile, some vehicles, such as buses, always follow relatively fixed routes, and their locations can be predicted much easier than other vehicles. By taking advantage of this fact, we propose a semi-Markov decision process (SMDP)-based cloudlet cooperation strategy, where the bus-based cloudlets act as computation service providers for the SMDs in vehicles, and an application generated by a SMD includes a series of tasks that have dependency among each other. In this paper, we adopt a semi-Markov decision process (SMDP) framework to formulate the bus-based cooperation computing problem as a delay-constrained shortest path problem on a state transition graph. The value iteration algorithm (VIA) is used to find the efficient solution to the bus-based cooperation computing problem. Experimental results show that the proposed SMDP-based cloudlet cooperation strategy can improve the performance of computation on the SMD in the cost (i.e., energy consumption and application completion time) and the offloading rate.