基于车载以太网的任务卸载方案

Task Offloading Scheme Based on Automotive Ethernet

现代汽车总线的电子电气(E/E)架构开始从基于CAN总线的分布式架构向通信容量更大的基于车载以太网的集中式架构发展,但车内计算资源分布不平均,集中式网络的中央设备计算能力显著高于其他边缘端设备。由于智能汽车边缘端设备的计算任务不断在增长,为应对增长的边缘端计算任务需求,降低计算延时,研究在集中式E/E架构中实现有效的计算任务卸载方案,对车载以太网的通信特点进行分析,以车载中央计算机为核心对车内域控制器的计算任务进行卸载。针对车载以太网AVB通信协议的特点,结合数据分级和整流算法获得计算卸载的延时模型,进而建立边缘计算任务延时最小化的目标优化方程,并采用改进遗传算法进行迭代求解,最终收敛求出任务卸载的最优参数。实现了在满足复杂车载条件约束条件下降低车内计算任务时延的目标,仿真实验结果表明所提方法可以有效降低车内计算任务延时,在集中式E/E架构的车内网络上实现了计算资源的充分利用。

The electrical and electronic(E/E)architecture of modern automotive bus has started to evolve from a distributed architecture based on CAN bus to a centralized architecture based on Automotive Ethernet with larger communication capacity,but the computational resources are not evenly distributed in the vehicle,and the computational capacity of the centralized network is significantly higher than that of other edge devices.Since the computational tasks of edge devices in smart cars are growing,in order to cope with the growing demand of edge computational tasks and reduce the computational latency,an effective computational task offloading scheme is studied in the centralized E/E architecture,and the communication characteristics of Automotive Ethernet are analyzed to offload the computational tasks of in-vehicle domain controllers to the in-vehicle central computer.For the characteristics of AVB communication protocol of Automotive Ethernet,the delay model of computation offloading is obtained by combining data classification and traffic shaping algorithms,and then the objective optimization equation for minimizing the delay of edge computation tasks is established and iteratively solved by using improved genetic algorithm,and finally the optimal parameters of task offloading are converged.The objective of reducing the in-vehicle computation task latency under the constraints of complex automotive conditions is achieved,and the simulation results show that the proposed method can effectively reduce the in-vehicle computation task latency and realize the full utilization of computational resources on the in-vehicle network of centralized E/E architecture.