CN115811518A - Vehicle-mounted high-performance computing framework based on SOA communication middleware and IWARP technology and implementation method - Google Patents

Abstract

The invention discloses a vehicle-mounted high-performance computing framework based on SOA communication middleware and IWARP technology and an implementation method thereof, wherein the vehicle-mounted high-performance computing framework comprises a central computing unit, an edge computing unit and a domain controller, which are communicated with each other through the SOA communication middleware; the central operation unit and the edge calculation unit are internally provided with a network controller RNIC (radio network integrated circuit) and are used for sending and receiving iWARP (internet authentication and privacy protocol) protocol messages, and the central operation unit and the edge calculation units form a distributed calculation framework; and the domain controller collects data and sends the data to the central operation unit, the central operation unit divides the data and sends the data to the edge calculation unit for distributed calculation, and then receives and returns the calculation result for data fusion, so that a final result is obtained and sent to the corresponding domain controller. The invention adds the RNIC required by the iWARP technology between the central operation unit and the edge calculation unit, and simultaneously constructs a set of high-availability, high-performance and low-delay distributed calculation framework by using SOA communication middleware, thereby ensuring that the technical grade of the automatic driving application scene is realized more quickly.

Description

Translated fromChinese

基于SOA通信中间件和IWARP技术的车载高性能计算框架及实现方法Vehicle-mounted high-performance computing framework and implementation based on SOA communication middleware and IWARP technologypresent method

技术领域technical field

本发明涉及自动驾驶技术，尤其涉及一种基于SOA通信中间件和IWARP技术的车载高性能计算框架及实现方法。The invention relates to automatic driving technology, in particular to a vehicle-mounted high-performance computing framework and an implementation method based on SOA communication middleware and IWARP technology.

背景技术Background technique

随着车载以太网技术的不断演进以及下一代汽车电子电气架构逐渐集中化，以域架构为基础的硬件架构大行其道。在软件架构层面上，面向服务的软件架构(SOA)是实现集中化汽车E/E架构的必要条件。以DDS、SOME/IP、MQTT组成的车载通信中间件均支持SOA的编程模式。然而，这三类车载通信中间件运行在TCP/IP协议栈上，由于数据在系统内存、处理器缓存和网络控制器缓存之间来回进行赋值移动给处理器和内存造成了严重负担，更加剧了网络延迟效应。而延迟过高对于大部分车载应用场景如自动驾驶而言是难以接受的。With the continuous evolution of automotive Ethernet technology and the gradual centralization of next-generation automotive electrical and electronic architectures, hardware architectures based on domain architectures are gaining popularity. At the software architecture level, service-oriented software architecture (SOA) is a necessary condition for realizing a centralized automotive E/E architecture. The in-vehicle communication middleware composed of DDS, SOME/IP, and MQTT all support SOA programming mode. However, these three types of in-vehicle communication middleware run on the TCP/IP protocol stack, and the assignment and movement of data between the system memory, processor cache and network controller cache has caused a serious burden on the processor and memory, which is even more serious. network delay effect. However, excessive delay is unacceptable for most vehicle application scenarios such as autonomous driving.

同时，自动驾驶分级标准下中高等级标准的实现依赖于MPU芯片的算力发展，这意味着芯片算力一旦遇到瓶颈无法提升则会反过来制约自动驾驶的技术等级。因此，如何设计分布式的车载高性能计算框架则显得尤为重要。基于SOA软件架构的通信中间件和IWARP技术在车载环境下实现分布式高性能计算框架的技术则很好地弥补了上述两方面的空缺。At the same time, the realization of the middle and high-level standards under the automatic driving classification standard depends on the development of the computing power of the MPU chip, which means that once the chip computing power encounters a bottleneck and cannot be improved, it will in turn restrict the technical level of automatic driving. Therefore, how to design a distributed vehicle high-performance computing framework is particularly important. Communication middleware based on SOA software architecture and IWARP technology to realize distributed high-performance computing framework in the vehicle environment well make up for the gaps in the above two aspects.

发明内容Contents of the invention

为解决现有技术中存在的不足，本发明的目的在于，提供一种基于SOA通信中间件和IWARP技术的车载高性能计算框架及实现方法，在MCU/MPU对应的控制器或计算单元内铺设支持IWARP技术的网络控制器RNIC，利用支持SOA软件架构的通信中间件，实现车载分布式高性能计算框架。In order to solve the deficiencies in the prior art, the object of the present invention is to provide a vehicle-mounted high-performance computing framework and implementation method based on SOA communication middleware and IWARP technology, and lay The network controller RNIC that supports IWARP technology uses the communication middleware that supports SOA software architecture to realize the vehicle-mounted distributed high-performance computing framework.

为实现本发明的目的，本发明所采用的技术方案是：For realizing the purpose of the present invention, the technical scheme adopted in the present invention is:

一种基于SOA通信中间件和IWARP技术的车载高性能计算框架，包括中央运算单元，边缘计算单元和域控制器；A vehicle-mounted high-performance computing framework based on SOA communication middleware and IWARP technology, including a central computing unit, an edge computing unit and a domain controller;

中央运算单元与边缘计算单元之间通过SOA通信中间件相互通信，中央运算单元与域控制器之间通过SOA通信中间件相互通信；The central computing unit and the edge computing unit communicate with each other through the SOA communication middleware, and the central computing unit and the domain controller communicate with each other through the SOA communication middleware;

中央运算单元内置网络控制器RNIC，边缘计算单元内置网络控制器RNIC，发送并接收iWARP协议报文，中央运算单元与多个边缘计算单元组成分布式计算架构；The central computing unit has a built-in network controller RNIC, and the edge computing unit has a built-in network controller RNIC to send and receive iWARP protocol messages. The central computing unit and multiple edge computing units form a distributed computing architecture;

域控制器采集数据后发送给中央运算单元，中央运算单元将数据分割后发送给边缘计算单元进行分布式计算，然后接收返回的计算结果进行数据融合，得出最终结果并发送给相应的域控制器。The domain controller collects the data and sends it to the central computing unit. The central computing unit divides the data and sends it to the edge computing unit for distributed computing, and then receives the returned computing results for data fusion to obtain the final result and send it to the corresponding domain controller. device.

进一步地，域控制器之间通过SOA通信中间件相互通信。Further, domain controllers communicate with each other through SOA communication middleware.

进一步地，SOA通信中间件为SOME/IP、DDS、MQTT三者中的一种。Further, the SOA communication middleware is one of SOME/IP, DDS, and MQTT.

进一步地，SOA通信中间件适配iWARP技术，中央运算单元与边缘计算单元之间通过支持iWARP技术的SOA通信中间件相互通信。Furthermore, the SOA communication middleware is adapted to iWARP technology, and the central computing unit and the edge computing unit communicate with each other through the SOA communication middleware supporting iWARP technology.

进一步地，中央运算单元与边缘计算单元之间通过Request/Response模型进行通信；具体为：Further, the central computing unit and the edge computing unit communicate through the Request/Response model; specifically:

中央运算单元将分割后的数据作为数据源封装在Request报文内，向边缘计算单元发送Request；边缘计算单元接收到Request报文后，提取数据源调用自身算力和相应的算法进行计算，输出计算结果；The central computing unit encapsulates the divided data as a data source in a Request message, and sends a Request to the edge computing unit; after receiving the Request message, the edge computing unit extracts the data source and invokes its own computing power and corresponding algorithm for calculation, and outputs Calculation results;

边缘计算单元将计算结果作为数据源封装在Response报文内，并向中央运算单元发送Response报文；中央运算单元接收到Response报文后，提取计算结果，并与其他边缘计算单元返回的计算结果一起，通过数据融合得到一个最终结果。The edge computing unit encapsulates the calculation result as a data source in the Response message, and sends the Response message to the central computing unit; after receiving the Response message, the central computing unit extracts the calculation result and compares it with the calculation results returned by other edge computing units. Together, a final result is obtained through data fusion.

进一步地，中央计算单元与域控制器之间通过两类通信模型进行通信，SOA中提供的Notification模型，数据单向流通；或SOA中提供的Request/Response模型，数据双向流通。Furthermore, the central computing unit and the domain controller communicate through two types of communication models, the Notification model provided in the SOA, with one-way data flow; or the Request/Response model provided in the SOA, with two-way data flow.

进一步地，域控制器采集数据，通过SOA通信中间件的Notification模式将数据发送给中央运算单元；中央运算单元通过分布式高性能计算框架计算数据后，再通过SOA通信中间件的Notification模式将计算结果发给域控制器。Further, the domain controller collects data and sends the data to the central computing unit through the Notification mode of the SOA communication middleware; after the central computing unit calculates the data through the distributed high-performance computing framework, it then sends the calculated data through the Notification mode of the SOA communication middleware The result is sent to the domain controller.

进一步地，域控制器采集数据，通过SOA通信中间件的Notification模式将数据发送给中央运算单元；中央运算单元通过分布式高性能计算框架计算数据后，再通过SOA通信中间件的Request模式将计算结果发给域控制器；Further, the domain controller collects data and sends the data to the central computing unit through the Notification mode of the SOA communication middleware; after the central computing unit calculates the data through the distributed high-performance computing framework, it then sends the calculated data through the Request mode of the SOA communication middleware The result is sent to the domain controller;

域控制器收到请求后，将计算结果提取出来并进行二次分析，然后将分析结果通过SOA通信中间件的Response模式发送给中央运算单元。After receiving the request, the domain controller extracts the calculation result and performs secondary analysis, and then sends the analysis result to the central computing unit through the Response mode of the SOA communication middleware.

一种基于SOA通信中间件和IWARP技术的车载高性能计算框架的实现方法，包括步骤：A method for realizing a vehicle-mounted high-performance computing framework based on SOA communication middleware and IWARP technology, comprising steps:

1)域控制器进行数据采集后，将采集到的数据发送给中央运算单元；1) After the domain controller collects the data, it sends the collected data to the central computing unit;

2)中央运算单元根据当前边缘计算单元的总数进行数据分割，并依次编排序号，同时启动定时器，以避免边缘计算单元下线对数据计算造成影响；2) The central computing unit performs data segmentation according to the total number of current edge computing units, and serializes the sequence number, and starts the timer at the same time, so as to avoid the impact of edge computing units offline on data calculation;

3)中央运算单元根据编排的序号，封装相应的数据，并通过Request模式发送给边缘计算单元；3) The central computing unit encapsulates the corresponding data according to the programmed serial number, and sends it to the edge computing unit through the Request mode;

4)边缘计算单元获取到数据后，调用自身驻留的算法模块进行数据计算，并输出计算结果；4) After the edge computing unit obtains the data, it calls its resident algorithm module to perform data calculation and output the calculation result;

5)边缘计算单元根据Request报文中提供的序号以及计算结果，封装到Response报文中，并且将报文发送给中央运算单元；5) The edge computing unit encapsulates the serial number and the calculation result provided in the Request message into a Response message, and sends the message to the central computing unit;

6)中央运算单元获取到对应边缘计算单元提供的Response报文，提取出序号和计算结果，并持续等待其他边缘计算单元提交相应计算结果，直到定时器超时；如果定时器超时，则此次计算流程应返回失败；6) The central computing unit obtains the Response message provided by the corresponding edge computing unit, extracts the serial number and the calculation result, and continues to wait for other edge computing units to submit the corresponding calculation results until the timer expires; if the timer expires, the calculation The process should return failure;

7)中央运算单元判断是否收到所有边缘计算单元的Response报文，如果全部收到则进入流程8)，如果尚未全部收到则返回流程6)；7) The central computing unit judges whether the Response messages of all the edge computing units have been received, and if all are received, then enter process 8), and if not all are received, then return to process 6);

8)中央运算单元将收集到的所有计算结果进行融合，得出最终结果，并发送给相应的域控制器。8) The central computing unit fuses all the collected calculation results to obtain the final result and sends it to the corresponding domain controller.

本发明的有益效果在于，与现有技术相比，本发明利用集中化的E/E域架构，在车载中央运算单元和车载边缘计算单元之间增设iWARP技术所需的RNIC，同时利用SOA通信中间件普遍提供的Request/Response通信模式和Notification通信模式，构建了一套高可用、高性能、低延迟的车载分布式计算框架。The beneficial effect of the present invention is that, compared with the prior art, the present invention utilizes the centralized E/E domain architecture, and adds the RNIC required by the iWARP technology between the vehicle-mounted central computing unit and the vehicle-mounted edge computing unit, and utilizes SOA communication The Request/Response communication mode and Notification communication mode commonly provided by middleware build a set of high-availability, high-performance, low-latency vehicle distributed computing framework.

通过这套分布式计算框架，可以有效地通过增加芯片数量而并非提高芯片算力的方式来增强整车的整体算力，从而保证了自动驾驶应用场景技术等级的更快实现。Through this distributed computing framework, the overall computing power of the vehicle can be effectively enhanced by increasing the number of chips instead of increasing the computing power of chips, thus ensuring faster realization of the technical level of autonomous driving application scenarios.

同时，借助互联网广域RDMA协议技术，有效地满足了车载领域下某些应用场景要求对延迟要求苛刻的需求。通过iWARP技术可以快速融入车载以太网，并非替换车载以太网，本质上提供了一个低成本的解决方案。At the same time, with the help of Internet wide-area RDMA protocol technology, it effectively meets the demanding delay requirements of certain application scenarios in the vehicle field. The iWARP technology can be quickly integrated into the vehicle Ethernet, not to replace the vehicle Ethernet, but essentially provides a low-cost solution.

本发明还利用SOA的特性使得车载分布式计算框架可以保障紧急制动、ADAS、智能座舱等应用场景和环境下的高可靠性和低延迟，从而确保了整车运行过程中的安全性和舒适性。The present invention also makes use of the characteristics of SOA to enable the vehicle-mounted distributed computing framework to ensure high reliability and low delay in application scenarios and environments such as emergency braking, ADAS, and smart cockpit, thereby ensuring the safety and comfort of the entire vehicle during operation sex.

附图说明Description of drawings

图1是本发明所述的基于SOA通信中间件和IWARP技术的车载高性能计算框架示意图；Fig. 1 is the vehicle-mounted high-performance computing frame schematic diagram based on SOA communication middleware and IWARP technology of the present invention;

图2是中央运算单元与边缘计算单元之间的SOA通信交互示意图；Fig. 2 is a schematic diagram of SOA communication interaction between the central computing unit and the edge computing unit;

图3是中央运算单元与域控制器之间的SOA通信交互示意图；Fig. 3 is a schematic diagram of SOA communication interaction between a central computing unit and a domain controller;

图4是分布式计算框架的具体执行流程。Figure 4 is the specific execution flow of the distributed computing framework.

具体实施方式Detailed ways

下面结合附图和实施例对本发明的技术方案作进一步的说明。以下实施例仅用于更加清楚地说明本发明的技术方案，而不能以此来限制本申请的保护范围。The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments. The following examples are only used to illustrate the technical solutions of the present invention more clearly, but not to limit the protection scope of the present application.

RDMA是一种远程直接内存访问技术，可以使得计算机之间可以直接读取对端的内存，而不需要经过处理器处理，大大降低了处理器拷贝数据的开销。拷贝行为的存在是增大延迟的重要因素。RDMA is a remote direct memory access technology that enables computers to directly read the memory of the opposite end without processing by the processor, which greatly reduces the overhead of the processor copying data. The presence of copying behavior is an important factor that increases latency.

互联网广域RDMA协议(iWARP)是RDMA over TCP，允许通过TCP/IP协议栈执行RDMA的网络协议，与其他RDMA技术不同的是，iWARP技术只要求网卡是支持iWARP的RNIC，而允许在标准以太网基础架构(交换机)上使用RDMA。通过iWARP技术可以快速融入车载以太网，并非替换车载以太网，本质上提供了一个低成本的解决方案。Internet Wide Area RDMA Protocol (iWARP) is RDMA over TCP, which allows RDMA network protocols to be executed through the TCP/IP protocol stack. Unlike other RDMA technologies, iWARP technology only requires the network card to be an RNIC that supports iWARP, while allowing RDMA is used on the network infrastructure (switches). The iWARP technology can be quickly integrated into the vehicle Ethernet, not to replace the vehicle Ethernet, but essentially provides a low-cost solution.

如图1所示，为一个典型的汽车E/E架构的硬件结构。其中，域控制器A与域控制器B、域控制器C之间通过SOA通信中间件相互通信。SOA通信中间件指的是SOME/IP、DDS、MQTT这三者中的一种。SOA通信中间件基于IP协议，无论其上的传输协议是TCP还是UDP。As shown in Figure 1, it is the hardware structure of a typical automotive E/E architecture. Wherein, domain controller A communicates with domain controller B and domain controller C through SOA communication middleware. SOA communication middleware refers to one of SOME/IP, DDS, and MQTT. SOA communication middleware is based on IP protocol, regardless of whether the transport protocol on it is TCP or UDP.

中央运算单元内置网络控制器RNIC，用于发送并接收iWARP协议的报文，中央运算单元与域控制器A、域控制器B、域控制器C之间通过SOA通信中间件相互通信。中央运算单元与4个边缘计算单元共同组成了分布式计算架构。每个边缘计算单元均内置RNIC用于发送并接收iWARP协议的报文。The central computing unit has a built-in network controller RNIC, which is used to send and receive iWARP protocol messages. The central computing unit communicates with domain controller A, domain controller B, and domain controller C through SOA communication middleware. The central computing unit and four edge computing units together form a distributed computing architecture. Each edge computing unit has a built-in RNIC for sending and receiving iWARP protocol messages.

SOA通信中间件需适配iWARP技术，在此基础之上，中央运算单元与边缘计算单元之间通过支持iWARP技术的SOA通信中间件相互通信。The SOA communication middleware needs to be adapted to iWARP technology. On this basis, the central computing unit and the edge computing unit communicate with each other through the SOA communication middleware that supports iWARP technology.

如图2所示，中央运算单元与边缘计算单元之间通过Request/Response模型进行通信。具体而言，1)中央运算单元将分割后的数据作为数据源封装在Request报文内，向边缘计算单元发送Request；2)边缘计算单元接收到Request报文后，提取出数据源，调用自身MPU的算力和相应的算法进行计算，然后输出计算结果；3)边缘计算单元将计算结果作为数据源封装在Response报文内，并向中央运算单元发送Response报文；4)中央运算单元接收到Response报文后，提取出计算的结果，并与其他边缘计算单元返回的计算结果一起，通过数据融合得到一个最终的结果，然后将该结果作为数据源通过Notification模式或者Request/Response模式与域控制器之间通信。As shown in Figure 2, the central computing unit and the edge computing unit communicate through the Request/Response model. Specifically, 1) The central computing unit encapsulates the divided data as a data source in a Request message, and sends a Request to the edge computing unit; 2) After receiving the Request message, the edge computing unit extracts the data source and calls itself The computing power of the MPU and the corresponding algorithm are calculated, and then the calculation result is output; 3) The edge computing unit encapsulates the calculation result as a data source in the Response message, and sends the Response message to the central computing unit; 4) The central computing unit receives After receiving the Response message, the calculation result is extracted, and together with the calculation results returned by other edge computing units, a final result is obtained through data fusion, and then the result is used as a data source through Notification mode or Request/Response mode and field Communication between controllers.

因此，边缘计算单元需驻留算法，而中央运算单元提供数据。Therefore, the edge computing unit needs to reside in the algorithm, while the central computing unit provides the data.

如图3所示，中央计算单元与域控制器之间通过两类通信模型进行通信，第一类是SOA中提供的Notification模型，数据单向流通。第二类是SOA中提供的Request/Response模型，数据双向流通。图3提供了一个典型案例，具体而言有三种情况：As shown in Figure 3, the central computing unit and the domain controller communicate through two types of communication models. The first type is the Notification model provided in SOA, and the data flows in one direction. The second type is the Request/Response model provided in SOA, where data flows in two directions. Figure 3 provides a typical case, specifically there are three situations:

情况一，域控制器C进行数据采集，如激光雷达、摄像头视频等数据，通过SOA通信中间件的Notification模式将数据发送给中央运算单元。中央运算单元通过分布式高性能计算框架计算数据后，再通过SOA通信中间件的Notification模式将计算结果发给域控制器C。这种情况一般在ADAS应用场景中会出现。Case 1, domain controller C collects data, such as lidar, camera video and other data, and sends the data to the central computing unit through the Notification mode of the SOA communication middleware. After the central computing unit calculates the data through the distributed high-performance computing framework, it sends the calculation result to the domain controller C through the Notification mode of the SOA communication middleware. This situation generally occurs in ADAS application scenarios.

情况二，域控制器C进行数据采集，如激光雷达、摄像头视频等数据，通过SOA通信中间件的Notification模式将数据发送给中央运算单元。中央运算单元通过分布式高性能计算框架计算数据后，再通过SOA通信中间件的Notification模式将计算结果发给域控制器A。这种情况域控制器A一般作为车身域或者动力域需要进行紧急制动时会发生。In the second case, the domain controller C collects data, such as laser radar, camera video and other data, and sends the data to the central computing unit through the Notification mode of the SOA communication middleware. After the central computing unit calculates the data through the distributed high-performance computing framework, it sends the calculation result to domain controller A through the Notification mode of the SOA communication middleware. This situation occurs when the domain controller A generally serves as the body domain or the power domain and needs to perform emergency braking.

情况三，域控制器C进行数据采集，如激光雷达、摄像头视频等数据，通过SOA通信中间件的Notification模式将数据发送给中央运算单元。中央运算单元通过分布式高性能计算框架计算数据后，再通过SOA通信中间件的Request模式将计算结果发给域控制器B。域控制器B收到请求后，将计算结果提取出来并进行二次分析，然后将分析结果通过SOA通信中间件的Response模式发送给中央运算单元。这种情况，域控制器B一般作为座舱域，需要反馈相应的座舱和仪表信息时会发生。In the third case, the domain controller C collects data, such as laser radar, camera video and other data, and sends the data to the central computing unit through the Notification mode of the SOA communication middleware. After the central computing unit calculates the data through the distributed high-performance computing framework, it sends the calculation result to domain controller B through the Request mode of the SOA communication middleware. After domain controller B receives the request, it extracts the calculation result and performs secondary analysis, and then sends the analysis result to the central computing unit through the Response mode of the SOA communication middleware. In this case, domain controller B generally acts as the cockpit domain and will occur when corresponding cockpit and instrument information needs to be fed back.

下面简述分布式计算框架的具体执行流程。如图4所示。The specific execution flow of the distributed computing framework is briefly described below. As shown in Figure 4.

1)域控制器或者其他车载单元进行数据采集后，将采集到的数据发送给中央运算单元；1) After the domain controller or other vehicle-mounted units collect data, they send the collected data to the central computing unit;

2)中央运算单元根据当前边缘计算单元的总数进行数据分割，并依次编排序号，同时启动定时器，以避免边缘计算单元下线对数据计算造成影响；2) The central computing unit performs data segmentation according to the total number of current edge computing units, and serializes the sequence number, and starts the timer at the same time, so as to avoid the impact of edge computing units offline on data calculation;

3)中央运算单元根据编排的序号，封装相应的数据，并通过Request模式发送给边缘计算单元；3) The central computing unit encapsulates the corresponding data according to the programmed serial number, and sends it to the edge computing unit through the Request mode;

4)边缘计算单元获取到数据后，调用自身驻留的算法模块进行数据计算，并输出计算结果；4) After the edge computing unit obtains the data, it calls its resident algorithm module to perform data calculation and output the calculation result;

5)边缘计算单元根据Request报文中提供的序号以及计算结果，封装到Response报文中，并且将报文发送给中央运算单元；5) The edge computing unit encapsulates the serial number and the calculation result provided in the Request message into a Response message, and sends the message to the central computing unit;

6)中央运算单元获取到对应边缘计算单元提供的Response报文，提取出序号和计算结果，并持续等待其他边缘计算单元提交相应计算结果，直到定时器超时。如果定时器超时，则此次计算流程应返回失败。6) The central computing unit obtains the Response message provided by the corresponding edge computing unit, extracts the serial number and calculation result, and continues to wait for other edge computing units to submit corresponding calculation results until the timer expires. If the timer expires, the calculation process should return failure.

7)中央运算单元判断是否收到所有边缘计算单元的Response报文，如果全部收到则进入流程8)，如果尚未全部收到则返回流程6)。7) The central computing unit judges whether all the Response messages of the edge computing units have been received, and if all the Response messages are received, then enter the process 8), and if not all of them are received, then return to the process 6).

8)中央运算单元将收集到的所有计算结果进行融合，得出最终结果，并发送给相应的域控制器。8) The central computing unit fuses all the collected calculation results to obtain the final result and sends it to the corresponding domain controller.

综上所述，通过图1、图2、图3描述了基于SOA通信中间件和IWARP技术的车载高性能计算框架。通过图4描述了基于SOA通信中间件和IWARP技术的车载高性能计算框架的实现方法。In summary, Figure 1, Figure 2, and Figure 3 describe the vehicle-mounted high-performance computing framework based on SOA communication middleware and IWARP technology. Figure 4 describes the implementation method of the vehicle-mounted high-performance computing framework based on SOA communication middleware and IWARP technology.

本发明的有益效果在于，与现有技术相比，本发明利用集中化的E/E域架构，在车载中央运算单元和车载边缘计算单元之间增设iWARP技术所需的RNIC，同时利用SOA通信中间件普遍提供的Request/Response通信模式和Notification通信模式，构建了一套高可用、高性能、低延迟的车载分布式计算框架。The beneficial effect of the present invention is that, compared with the prior art, the present invention utilizes the centralized E/E domain architecture, and adds the RNIC required by the iWARP technology between the vehicle-mounted central computing unit and the vehicle-mounted edge computing unit, and utilizes SOA communication The Request/Response communication mode and Notification communication mode commonly provided by middleware build a set of high-availability, high-performance, low-latency vehicle distributed computing framework.

通过这套分布式计算框架，可以有效地通过增加芯片数量而并非提高芯片算力的方式来增强整车的整体算力，从而保证了自动驾驶应用场景技术等级的更快实现。Through this distributed computing framework, the overall computing power of the vehicle can be effectively enhanced by increasing the number of chips instead of increasing the computing power of chips, thus ensuring faster realization of the technical level of autonomous driving application scenarios.

同时，借助互联网广域RDMA协议技术，有效地满足了车载领域下某些应用场景要求对延迟要求苛刻的需求。通过iWARP技术可以快速融入车载以太网，并非替换车载以太网，本质上提供了一个低成本的解决方案。At the same time, with the help of Internet wide-area RDMA protocol technology, it effectively meets the demanding delay requirements of certain application scenarios in the vehicle field. The iWARP technology can be quickly integrated into the vehicle Ethernet, not to replace the vehicle Ethernet, but essentially provides a low-cost solution.

本发明还利用SOA的特性使得车载分布式计算框架可以保障紧急制动、ADAS、智能座舱等应用场景和环境下的高可靠性和低延迟，从而确保了整车运行过程中的安全性和舒适性。The present invention also makes use of the characteristics of SOA to enable the vehicle-mounted distributed computing framework to ensure high reliability and low delay in application scenarios and environments such as emergency braking, ADAS, and smart cockpit, thereby ensuring the safety and comfort of the entire vehicle during operation sex.

本发明申请人结合说明书附图对本发明的实施示例做了详细的说明与描述，但是本领域技术人员应该理解，以上实施示例仅为本发明的优选实施方案，详尽的说明只是为了帮助读者更好地理解本发明精神，而并非对本发明保护范围的限制，相反，任何基于本发明的发明精神所作的任何改进或修饰都应当落在本发明的保护范围之内。The applicant of the present invention has made a detailed description and description of the implementation examples of the present invention in conjunction with the accompanying drawings, but those skilled in the art should understand that the above implementation examples are only preferred implementations of the present invention, and the detailed description is only to help readers better To understand the spirit of the present invention rather than limit the protection scope of the present invention, on the contrary, any improvement or modification made based on the spirit of the present invention shall fall within the protection scope of the present invention.

Claims (9)

Translated fromChinese

1.一种基于SOA通信中间件和IWARP技术的车载高性能计算框架，其特征在于，包括中央运算单元，边缘计算单元和域控制器；1. A vehicle-mounted high-performance computing framework based on SOA communication middleware and IWARP technology, is characterized in that, comprises central computing unit, edge computing unit and domain controller;中央运算单元与边缘计算单元之间通过SOA通信中间件相互通信，中央运算单元与域控制器之间通过SOA通信中间件相互通信；The central computing unit and the edge computing unit communicate with each other through the SOA communication middleware, and the central computing unit and the domain controller communicate with each other through the SOA communication middleware;中央运算单元内置网络控制器RNIC，边缘计算单元内置网络控制器RNIC，发送并接收iWARP协议报文，中央运算单元与多个边缘计算单元组成分布式计算架构；The central computing unit has a built-in network controller RNIC, and the edge computing unit has a built-in network controller RNIC to send and receive iWARP protocol messages. The central computing unit and multiple edge computing units form a distributed computing architecture;域控制器采集数据后发送给中央运算单元，中央运算单元将数据分割后发送给边缘计算单元进行分布式计算，然后接收返回的计算结果进行数据融合，得出最终结果并发送给相应的域控制器。The domain controller collects the data and sends it to the central computing unit. The central computing unit divides the data and sends it to the edge computing unit for distributed computing, and then receives the returned computing results for data fusion to obtain the final result and send it to the corresponding domain controller. device.2.根据权利要求1所述的基于SOA通信中间件和IWARP技术的车载高性能计算框架，其特征在于，域控制器之间通过SOA通信中间件相互通信。2. The vehicle-mounted high-performance computing framework based on SOA communication middleware and IWARP technology according to claim 1, wherein the domain controllers communicate with each other through the SOA communication middleware.3.根据权利要求1所述的基于SOA通信中间件和IWARP技术的车载高性能计算框架，其特征在于，SOA通信中间件为SOME/IP、DDS、MQTT三者中的一种。3. The vehicle-mounted high-performance computing framework based on SOA communication middleware and IWARP technology according to claim 1, wherein the SOA communication middleware is one of SOME/IP, DDS, and MQTT.4.根据权利要求1所述的基于SOA通信中间件和IWARP技术的车载高性能计算框架，其特征在于，SOA通信中间件适配iWARP技术，中央运算单元与边缘计算单元之间通过支持iWARP技术的SOA通信中间件相互通信。4. The vehicle-mounted high-performance computing framework based on SOA communication middleware and IWARP technology according to claim 1, wherein the SOA communication middleware is adapted to iWARP technology, and the central computing unit and the edge computing unit support iWARP technology The SOA communication middleware communicates with each other.5.根据权利要求1所述的基于SOA通信中间件和IWARP技术的车载高性能计算框架，其特征在于，中央运算单元与边缘计算单元之间通过Request/Response模型进行通信；具体为：5. the vehicle-mounted high-performance computing framework based on SOA communication middleware and IWARP technology according to claim 1, is characterized in that, communicates by Request/Response model between central computing unit and edge computing unit; Be specifically:中央运算单元将分割后的数据作为数据源封装在Request报文内，向边缘计算单元发送Request；边缘计算单元接收到Request报文后，提取数据源调用自身算力和相应的算法进行计算，输出计算结果；The central computing unit encapsulates the divided data as a data source in a Request message, and sends a Request to the edge computing unit; after receiving the Request message, the edge computing unit extracts the data source and invokes its own computing power and corresponding algorithm for calculation, and outputs Calculation results;边缘计算单元将计算结果作为数据源封装在Response报文内，并向中央运算单元发送Response报文；中央运算单元接收到Response报文后，提取计算结果，并与其他边缘计算单元返回的计算结果一起，通过数据融合得到一个最终结果。The edge computing unit encapsulates the calculation result as a data source in the Response message, and sends the Response message to the central computing unit; after receiving the Response message, the central computing unit extracts the calculation result and compares it with the calculation results returned by other edge computing units. Together, a final result is obtained through data fusion.6.根据权利要求1所述的基于SOA通信中间件和IWARP技术的车载高性能计算框架，其特征在于，中央计算单元与域控制器之间通过两类通信模型进行通信，SOA中提供的Notification模型，数据单向流通；或SOA中提供的Request/Response模型，数据双向流通。6. the vehicle-mounted high-performance computing framework based on SOA communication middleware and IWARP technology according to claim 1, is characterized in that, communicates by two types of communication models between the central computing unit and the domain controller, and the Notification provided in the SOA model, one-way flow of data; or the Request/Response model provided in SOA, two-way flow of data.7.根据权利要求6所述的基于SOA通信中间件和IWARP技术的车载高性能计算框架，其特征在于，域控制器采集数据，通过SOA通信中间件的Notification模式将数据发送给中央运算单元；中央运算单元通过分布式高性能计算框架计算数据后，再通过SOA通信中间件的Notification模式将计算结果发给域控制器。7. the vehicle-mounted high-performance computing framework based on SOA communication middleware and IWARP technology according to claim 6, is characterized in that, domain controller collects data, sends data to central computing unit by the Notification pattern of SOA communication middleware; After the central computing unit calculates the data through the distributed high-performance computing framework, it sends the calculation result to the domain controller through the Notification mode of the SOA communication middleware.8.根据权利要求6所述的基于SOA通信中间件和IWARP技术的车载高性能计算框架，其特征在于，域控制器采集数据，通过SOA通信中间件的Notification模式将数据发送给中央运算单元；中央运算单元通过分布式高性能计算框架计算数据后，再通过SOA通信中间件的Request模式将计算结果发给域控制器；8. the vehicle-mounted high-performance computing framework based on SOA communication middleware and IWARP technology according to claim 6, is characterized in that, domain controller collects data, sends data to central computing unit by the Notification pattern of SOA communication middleware; After the central computing unit calculates the data through the distributed high-performance computing framework, it sends the calculation result to the domain controller through the Request mode of the SOA communication middleware;域控制器收到请求后，将计算结果提取出来并进行二次分析，然后将分析结果通过SOA通信中间件的Response模式发送给中央运算单元。After receiving the request, the domain controller extracts the calculation result and performs secondary analysis, and then sends the analysis result to the central computing unit through the Response mode of the SOA communication middleware.9.一种基于SOA通信中间件和IWARP技术的车载高性能计算框架的实现方法，基于权利要求1-8任一所述的基于SOA通信中间件和IWARP技术的车载高性能计算框架，其特征在于，包括步骤：9. An implementation method based on the vehicle-mounted high-performance computing framework of SOA communication middleware and IWARP technology, based on the arbitrary described vehicle-mounted high-performance computing framework of claim 1-8 based on SOA communication middleware and IWARP technology, its feature is, include steps:1)域控制器进行数据采集后，将采集到的数据发送给中央运算单元；1) After the domain controller collects the data, it sends the collected data to the central computing unit;2)中央运算单元根据当前边缘计算单元的总数进行数据分割，并依次编排序号，同时启动定时器，以避免边缘计算单元下线对数据计算造成影响；2) The central computing unit performs data segmentation according to the total number of current edge computing units, and serializes the sequence number, and starts the timer at the same time, so as to avoid the impact of edge computing units offline on data calculation;3)中央运算单元根据编排的序号，封装相应的数据，并通过Request模式发送给边缘计算单元；3) The central computing unit encapsulates the corresponding data according to the programmed serial number, and sends it to the edge computing unit through the Request mode;4)边缘计算单元获取到数据后，调用自身驻留的算法模块进行数据计算，并输出计算结果；4) After the edge computing unit obtains the data, it calls its resident algorithm module to perform data calculation and output the calculation result;5)边缘计算单元根据Request报文中提供的序号以及计算结果，封装到Response报文中，并且将报文发送给中央运算单元；5) The edge computing unit encapsulates the serial number and the calculation result provided in the Request message into a Response message, and sends the message to the central computing unit;6)中央运算单元获取到对应边缘计算单元提供的Response报文，提取出序号和计算结果，并持续等待其他边缘计算单元提交相应计算结果，直到定时器超时；如果定时器超时，则此次计算流程应返回失败；6) The central computing unit obtains the Response message provided by the corresponding edge computing unit, extracts the serial number and the calculation result, and continues to wait for other edge computing units to submit the corresponding calculation results until the timer expires; if the timer expires, the calculation The process should return failure;7)中央运算单元判断是否收到所有边缘计算单元的Response报文，如果全部收到则进入流程8)，如果尚未全部收到则返回流程6)；7) The central computing unit judges whether the Response messages of all the edge computing units have been received, and if all are received, then enter process 8), and if not all are received, then return to process 6);8)中央运算单元将收集到的所有计算结果进行融合，得出最终结果，并发送给相应的域控制器。8) The central computing unit fuses all the collected calculation results to obtain the final result and sends it to the corresponding domain controller.

Images