汽车工程 ›› 2024, Vol. 46 ›› Issue (3): 431-437.doi: 10.19562/j.chinasae.qcgc.2024.03.006

• • 上一篇    

域控制器分布式并行计算模型研究

张本西(),史家涛,刘晓波,葛云东,赵星阳,宫珊珊   

  1. 潍柴动力股份有限公司,潍坊 261000
  • 收稿日期:2023-06-29 修回日期:2023-10-11 出版日期:2024-03-25 发布日期:2024-03-18
  • 通讯作者: 张本西 E-mail:zhangbenxi@weichai.com

Research on Distributed Parallel Computing Model of Domain Controller

Benxi Zhang(),Jiatao Shi,Xiaobo Liu,Yundong Ge,Xingyang Zhao,Shanshan Gong   

  1. Weichai Power Co. ,Ltd. ,Weifang 261000
  • Received:2023-06-29 Revised:2023-10-11 Online:2024-03-25 Published:2024-03-18
  • Contact: Benxi Zhang E-mail:zhangbenxi@weichai.com

摘要:

随着汽车域集中式架构诞生,搭载智能芯片的汽车域控制器成为新一代电子电气架构的主控单元。域集中式架构依靠车载以太网相连接,计算节点广泛分布在多个域控制器的智能芯片内,需要一套面向服务架构的分布式并行计算模型将计算节点性能充分释放出来。本文开发了一套面向服务架构的轻量级通信中间件,并基于该服务通信中间件实现了分布式并行计算模型,得出了分布式计算分割系数计算表达式,研究了计算节点负载阶跃波动时对分布式并行计算模型的影响。结果表明:域控制器面向服务架构的分布式计算模型具备负载自平衡能力,能够对智能芯片运行时阶跃负载实现快速任务重分配,保证循环任务的计算开销耗时最小,计算周期最短。研究结果为汽车域控架构下分布式计算框架设计与开发提供了重要参考依据。

关键词: 域控制器, 服务架构, 分布式计算

Abstract:

With the emergence of automotive domain centralized architecture, the automotive domain controller equipped with AI chips has become the main control unit of the new generation of electrical and electronic architecture. The domain-centralized architecture relies automotive Ethernet connection, and computing nodes are widely distributed within AI chips of multiple domain controllers, which requires a distributed parallel computing model with service-oriented architecture for full release of the performance of computing nodes. In this paper, a lightweight communication middleware based on service-oriented architecture is developed, and a distributed parallel computing model based on this middleware is realized. The expression for calculating the partition coefficient of distributed computing is derived, and the effect of computing node step load fluctuations on the distributed computing model is studied. The result shows that the distributed computing model of domain controller service-oriented architecture has the ability of load balancing, which can achieve fast task reassignment for step load during the runtime of AI chips, ensuring the minimum computational cost and cycle time of cyclic tasks. The results provide important reference for design and development of distributed computing frameworks on automotive domain control architecture.

Key words: domain controller, service-oriented-architecture, distributed computation