A Policy Gradient Based Offloading Scheme with Dependency Guarantees for Vehicular Networks

With the rapid development of Internet of Things (IoT) and mobile devices, the IoT applications are getting much more computation-intensive and latency-sensitive, which brings severe challenges to the resource limited devices. In order to enhance the computing power of the network, Mobile Edge Computing (MEC) has served as the key promising method, which can enable resource constrained devices to offload tasks to the edge server. Especially with the advancement of 5G technology and the Internet of Vehicles (IoV), more and more applications need to be offloaded for the ultra-low latency. However, one of the main challenges has been ignored is the dependency between different subtasks, which exerts a great impact on the decision of offloading. In this paper, we focus on the topological structure of subtasks with dependency for IoV applications. First, we employ directed acyclic graphs (DAGs) to explore the dependency of subtasks and introduce the priority of task scheduling. Further, an offloading scheme utilizing policy-based deep reinforcement learning (DRL) is put forward for minimizing the total task latency with dependency guarantees for all IoV applications in multi-vehicles scenario. The experimental results demonstrate that, compared with the state of the art offloading schemes, the proposed offloading scheme can reduce the delay of the whole system efficiently.