Edge-Cloud Collaborative Task Offloading Mechanism Based on DDQN in Vehicular Networks

Abstract

Computation offloading is a promising scheme to alleviate the shortage of vehicle resources facing the explosive growth trend of data computation.Compared with studying cloud computing or edge computing separately, integrating with each other can realize the complementary advantages and improve the overall quality of service.In vehicular networks, a primary challenge is to make offloading decisions, which can adapt to the dynamic environment.During this process, the urgency of tasks cannot be ignored.This paper constructs a collaborative edge-cloud task offloading architecture based on Software Defined Network (SDN), where the metrics of task priority is given.The task offloading problem is then formulated as a Markov Decision Process (MDP), which aims to maximize the utility composed of delay and cost.To solve task offloading decisions, this paper puts forward a task offloading decision algorithm based on Double Deep Q Network(DDQN)and a priority-based resource allocation scheme successively.On this basis, this paper designs a method of computing offloading ratio, which aims to minimize the task processing delay while ensuring that the part of tasks can be uploaded completely within the communication time.Simulation results show that the performance of delay and utility of the proposed algorithm is more than doubled compared to other fixed offloading algorithms such as All Local, All Offloading and Allocating Resources Evenly.Under the condition of moderate numbers of vehicles, the success rate of tasks can be maintained at 100%.