An edge intelligent hardware platform architecture method and system suitable for high mobility environments

Abstract

The present invention discloses an edge intelligent hardware platform architecture method and system suitable for a high-mobility environment, the method includes edge intelligent basic hardware environment preparation, edge computing resource virtualization pooling management, edge intelligent computing container full network lightweight deployment, edge computing resource container scheduling allocation; the system includes edge intelligent servers for training, edge intelligent servers for reasoning, edge network communication equipment, edge computing resource virtualization software, edge computing resource management software, edge intelligent computing container deployment service software, edge intelligent computing container deployment agent software and edge computing resource scheduling software deployed in edge service nodes and edge terminal nodes. The present invention not only supports unified description and management of computing resources based on edge intelligent basic hardware environment, but also can realize distributed and lightweight scenario application of edge intelligent computing container deployment in the whole system, and can support the agile and adaptive construction and adjustment of edge intelligent hardware platform in the whole system.

Description

Technical Field

[0001] The present invention relates to the field of edge computing technology, and in particular to an edge intelligent hardware platform architecture method and system suitable for high-mobility environments. Background Art

[0002] With the rapid development of the Internet of Things, Internet of Vehicles, Internet of Ships, Internet of Aircraft and other fields, edge computing technology has made continuous breakthroughs, and various intelligent computing services have been extended to the edge, thereby increasing the scale and complexity of intelligent information systems, bringing challenges to the organization and application of intelligent services in edge environments.

[0003] In today's market, there are a large number of intelligent information devices in the edge environment for building various small and medium-sized intelligent information systems, but the software and hardware resources of these devices are uneven, and the performance varies, and some are even poor, among which terminal devices are particularly prominent. Limited computing, storage, and communication resources will constrain the scale of intelligent services that can be integrated, deployed and utilized in various places; there are problems such as inefficient distributed implementation, deployment and operation and maintenance, and poor utilization of computing and transmission resources due to the complex environment and diverse systems at the edge. In order to provide more quality-assured intelligent service integration and deployment architecture support for various intelligent information systems at all levels at the edge, it is necessary to study a more efficient and available edge intelligent hardware platform architecture method. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide an edge intelligent hardware platform architecture method and system suitable for high-mobility environments.

[0005] Technical solution: The edge intelligent hardware platform architecture method suitable for high mobility environment described in the present invention includes the following steps:

[0006] S1. Preparation of basic hardware environment for edge intelligence: Equip all edge nodes with edge intelligence hardware. Edge nodes include edge service nodes and edge terminal nodes. Edge service nodes are equipped with training edge intelligence hardware and reasoning edge intelligence hardware, and edge terminal nodes are equipped with reasoning edge intelligence hardware.

[0007] S2. Virtual pooling and management of edge computing and transmission resources: Virtualize and uniformly describe the intelligent computing and network communication computing and transmission resources of all edge nodes, and manage them to form an edge computing and transmission virtual resource pool;

[0008] S3. Lightweight deployment of edge intelligent computing containers across the entire network: Through the interactive collaboration between the edge intelligent computing container deployment service software and the edge intelligent computing container deployment agent software, system operation and maintenance personnel can perform lightweight layered decoupling, deployment recommendation, accelerated distribution, and perception aggregation of container images suitable for edge high-mobility environments, and distributed loading and deployment of edge intelligent computing containers to all edge nodes.

[0009] S4. Scheduling and allocation of edge computing resource containers: The edge computing resource scheduling software dynamically allocates and recycles edge intelligent computing and network communication computing resources assigned to edge intelligent computing containers through containerized scheduling of edge computing resources.

[0010] Furthermore, step S1 includes preparation of basic hardware environment for edge intelligence training and basic hardware environment for edge intelligence reasoning.

[0011] Furthermore, the basic hardware environment preparation for edge intelligence training includes system construction personnel installing training-type edge intelligence chips, edge intelligence acceleration cards or edge intelligence computing modules on the training edge intelligence servers in the edge service nodes.

[0012] Furthermore, the preparation of the basic hardware environment for edge intelligent reasoning includes system construction personnel installing inference-type edge intelligent chips, edge intelligent acceleration cards or edge intelligent computing modules on edge intelligent servers for reasoning in edge service nodes and edge terminal nodes.

[0013] Furthermore, the step S2 comprises:

[0014] S21. Computing and transmission resource entity affiliation hierarchical planning: The edge computing and transmission resource virtualization software divides the entity affiliation of the edge node's intelligent computing and network communication computing and transmission resources into two-level mapping descriptions: nodes and servers;

[0015] S22. Virtual pooling of edge intelligent computing resources: The edge computing resource virtualization software virtualizes and uniformly describes the intelligent computing resources of all edge nodes, and the management software manages them to form an edge intelligent computing virtual resource pool. The edge intelligent computing virtual resources in the edge intelligent computing virtual resource pool are represented as follows:

[0016] ,

[0017] in, Indicates the block number in the virtual resource pool; Indicates the edge intelligent computing virtual resource pool Block number edge intelligent computing virtual resources have intelligent computing power; express The entity that provides the block number virtual resource; Indicates the virtual resource pool The entity that provides the block number virtual resource belongs to the node; Indicates the edge intelligent computing virtual resource pool The provider entity of the block number edge intelligent computing virtual resource belongs to the edge intelligent server; Indicates the edge intelligent computing virtual resource pool Block number Description of edge intelligent computing virtual resources;

[0018] S23, virtual pooling of edge network communication resources: virtualize and uniformly describe the network communication resources of all edge nodes, manage them to form an edge network communication virtual resource pool, and the edge network communication virtual resources in the edge network communication virtual resource pool are represented as follows:

[0019] ,

[0020] in, Indicates the block number in the virtual resource pool; Indicates the edge network communication virtual resource pool Block number edge network communication virtual resource virtual network communication attributes; Indicates the edge network communication virtual resource pool Block number: the communication bandwidth owned by the edge network communication virtual resource; express The entity that provides the block number virtual resource; Indicates the virtual resource pool The entity that provides the block number virtual resource belongs to the node; Indicates the edge network communication virtual resource pool Block number Description of edge network communication virtual resource.

[0021] Furthermore, the step S3 comprises:

[0022] S31. Layered decoupling of edge intelligent computing container images: Decouple edge intelligent computing container images into system layer, dependency layer and business layer according to functional logic: the system layer provides a streamlined operating system base image for the edge intelligent computing operating system; the dependency layer provides several intermediate software base images with the same dependencies for similar applications; the business layer provides business logic application images specific to each application;

[0023] S32, edge intelligent computing container image deployment determination: on the edge service node, based on the edge intelligent computing full network resource utilization details and edge network communication full network resource utilization details, according to the edge intelligent computing service to be deployed information list, autonomously recommend and generate the edge intelligent computing container full network to be deployed recommended list, and manually adjust and determine to form the edge intelligent computing container full network to be deployed list;

[0024] S33, edge intelligent computing container image distribution across the entire network: On the edge service node, according to the edge intelligent computing container to be deployed list across the entire network, select each edge intelligent computing container image to be deployed on all edge nodes from the edge intelligent computing container image central warehouse in turn, and use the edge intelligent computing container image accelerated distribution to quickly and lightly distribute the selected edge intelligent computing container image to the edge intelligent computing container image cache warehouse of the corresponding edge node, and issue the edge intelligent computing container node to-be-deployed list to the corresponding edge node;

[0025] S34. Edge intelligent computing container image deployment and loading: When each edge node receives the list of edge intelligent computing container nodes to be deployed, it pulls and aggregates each edge intelligent computing container image to be deployed in turn through the edge intelligent computing container image perception according to the list of edge intelligent computing container nodes to be deployed, and loads the aggregated edge intelligent computing container image into the corresponding edge intelligent computing container.

[0026] Furthermore, the step S4 comprises:

[0027] S41. When any edge intelligent computing container instance is enabled, according to the assigned intelligent computing power and communication bandwidth of the edge intelligent computing container instance, the edge intelligent server deployed by the edge intelligent computing container instance is scheduled to allocate edge intelligent computing resources equal to the assigned intelligent computing power, and the edge communication device deployed by the edge intelligent computing container instance is scheduled to allocate edge network communication resources equal to the assigned communication bandwidth to the edge intelligent computing container instance;

[0028] S42. When any edge intelligent computing container instance is withdrawn, the edge intelligent server deployed by the edge intelligent computing container instance is scheduled to recover edge intelligent computing resources equal to the assigned intelligent computing power and communication bandwidth of the edge intelligent computing container instance, and the edge communication equipment deployed by the edge intelligent computing container instance is scheduled to recover edge network communication resources equal to the assigned communication bandwidth to the edge intelligent computing container instance.

[0029] The edge intelligent hardware platform architecture system suitable for a high mobility environment described in the present invention includes:

[0030] Deploy edge intelligent servers for training, edge intelligent servers for reasoning, edge network communication equipment, edge computing resource virtualization software, edge computing resource management software, edge intelligent computing container deployment service software, edge intelligent computing container deployment agent software, and edge computing resource scheduling software in edge service nodes;

[0031] Deploy edge intelligent servers for inference, edge network communication equipment, edge computing resource virtualization software, edge intelligent computing container deployment agent software, and edge computing resource scheduling software in edge terminal nodes.

[0032] Furthermore, the edge intelligent server for training is equipped with a number of training-type edge intelligent chips, edge intelligent acceleration cards or edge intelligent computing modules, which provides a basic hardware environment of edge intelligent computing and network communication resources for regular online incremental training and optimization of edge intelligent models, and supports updating the edge intelligent model after iterative optimization to the container image of the edge intelligent service.

[0033] Furthermore, the edge intelligent server for inference is equipped with a number of edge intelligent chips, edge intelligent acceleration cards or edge intelligent computing modules of the inference type, which provides a basic hardware environment of edge intelligent computing and network communication computing resources for the loading and operation of edge intelligent services.

[0034] Beneficial effect: Compared with the prior art, the present invention has the following significant advantages: the present invention not only supports the unified description and management of computing and transmission resources based on the edge intelligent basic hardware environment, but also supports the distributed and lightweight deployment of edge intelligent computing containers in the entire system, and flexibly schedules resources based on the adaptability of application requirements to ensure containerized operation of intelligent services; it is more suitable for business information systems such as motor vehicles, ships, and aircraft, and can support the agile and adaptive construction and adjustment of the edge intelligent hardware platform for the entire system, and support the deployment and operation of intelligent services in the entire system based on the edge intelligent hardware platform, thereby improving the reliability and application efficiency of the implementation of the intelligent information system architecture. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of an application scenario of the present invention;

[0036] Figure 2 Schematic diagram of the process of preparing the basic hardware environment for edge intelligence;

[0037] Figure 3 A schematic diagram of the process of virtualizing and managing edge computing resources.

[0038] Figure 4 This is a schematic diagram of the lightweight deployment process of edge computing containers across the entire network;

[0039] Figure 5 A diagram of the recommended process for deploying edge computing container images;

[0040] Figure 6 Schematic diagram of the process of accelerating the distribution of edge intelligent computing container images;

[0041] Figure 7 This is a schematic diagram of the process of edge intelligent computing container image perception aggregation;

[0042] Figure 8 Schematic diagram of the process of scheduling and allocating edge computing resource containers. DETAILED DESCRIPTION

[0043] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0044] like Figure 1 As shown, in this embodiment, an edge intelligent hardware platform architecture method suitable for a high mobility environment includes the following steps:

[0045] S1. Preparation of edge intelligence basic hardware environment. System builders install edge intelligence hardware on all edge nodes to provide edge intelligence basic hardware environment for edge intelligence services. Edge service nodes are equipped with training and reasoning types, and edge terminal nodes are equipped with reasoning types.

[0046] The edge nodes are divided into edge service nodes and edge terminal nodes, wherein the edge service nodes are service-type edge nodes that provide services to the outside, and the edge terminal nodes are terminal-type edge nodes that enjoy services to the outside. The edge service nodes and edge terminal nodes are mainly connected by wireless communication;

[0047] The edge intelligence service refers to software that provides services with the edge intelligence model as the core, and supports packaging and deployment in the form of container images.

[0048] like Figure 2 As shown in the figure, the detailed steps for preparing the basic hardware environment for edge intelligence are as follows:

[0049] S11. Prepare the basic hardware environment for edge intelligence training. System builders install several edge intelligence chips, edge intelligence acceleration cards or edge intelligence computing modules on the edge intelligence servers for training in edge service nodes to provide the basic hardware environment for edge intelligence computing, network communication and other computing resources for regular online incremental training optimization of edge intelligence models, and support the update of edge intelligence models to the container image of edge intelligence services after iterative optimization;

[0050] S12. Prepare the basic hardware environment for edge intelligent reasoning. System builders install several edge intelligent chips, edge intelligent acceleration cards or edge intelligent computing modules for reasoning on the edge intelligent servers for reasoning in edge service nodes and edge terminal nodes, and provide the basic hardware environment for computing and transmission resources such as edge intelligent computing and network communication for the loading and operation of edge intelligent services.

[0051] S2. Virtual pooling and management of edge computing and transmission resources. The edge computing and transmission resource virtualization software virtualizes and uniformly describes the computing and transmission resources such as intelligent computing and network communication of all edge nodes, and the edge computing and transmission resource management software manages them to form an edge computing and transmission virtual resource pool.

[0052] The edge computing and transmission resource virtualization software virtualizes and uniformly describes computing and transmission resources such as intelligent computing and network communication of edge nodes;

[0053] The edge computing and transmission resource management software pools and manages computing and transmission resources such as intelligent computing and network communication of all edge nodes;

[0054] The edge computing virtual resource pool includes an edge intelligent computing virtual resource pool and an edge network communication virtual resource pool.

[0055] like Figure 3 As shown in the figure, the detailed steps for managing edge computing resources in a virtual pool are as follows:

[0056] S21. Computing and transmission resource entity affiliation hierarchical planning. The edge computing and transmission resource virtualization software divides the entity affiliation of computing and transmission resources such as intelligent computing and network communication of edge nodes into two-level mapping descriptions of nodes and servers;

[0057] S22. Virtual pooling of edge intelligent computing resources. The edge computing resource virtualization software virtualizes and uniformly describes the intelligent computing resources of all edge nodes. The edge computing resource management software manages the edge intelligent computing virtual resource pool. The unified description of the edge intelligent computing virtual resources in the edge intelligent computing virtual resource pool is as follows:

[0058] ,

[0059] in, Indicates the block number in the virtual resource pool; Indicates the edge intelligent computing virtual resource pool Block number edge intelligent computing virtual resources have intelligent computing power; express The entity that provides the block number virtual resource; Indicates the virtual resource pool The entity that provides the block number virtual resource belongs to the node; Indicates the edge intelligent computing virtual resource pool The provider entity of the block number edge intelligent computing virtual resource belongs to the edge intelligent server; Indicates the edge intelligent computing virtual resource pool Description of the block number edge intelligent computing virtual resource; including the intelligent computing power that owns the edge intelligent computing virtual resource and its provider entity;

[0060] S23, virtual pooling of edge network communication resources. Virtualize and uniformly describe the network communication resources of all edge nodes, manage them to form an edge network communication virtual resource pool, and the unified description of the edge network communication virtual resources in the edge network communication virtual resource pool is as follows:

[0061] ,

[0062] in, Indicates the block number in the virtual resource pool; Indicates the edge network communication virtual resource pool Block number edge network communication virtual resource virtual network communication attributes; Indicates the edge network communication virtual resource pool Block number: the communication bandwidth owned by the edge network communication virtual resource; express The entity that provides the block number virtual resource; Indicates the virtual resource pool The entity that provides the block number virtual resource belongs to the node; Indicates the edge network communication virtual resource pool The description of the block number edge network communication virtual resource includes the virtual network communication attributes of the edge network communication virtual resource, the communication bandwidth it owns and its provider entity.

[0063] S3, lightweight deployment of edge intelligent computing containers across the entire network. Through the interactive collaboration of edge intelligent computing container deployment service software and edge intelligent computing container deployment agent software, system operation and maintenance personnel can perform lightweight layered decoupling, deployment recommendation, accelerated distribution, and perception aggregation of container images suitable for edge high-mobility environments, and distributed loading and deployment of edge intelligent computing containers to all edge nodes.

[0064] The edge intelligent computing container deployment service software provides an edge intelligent computing container image central warehouse for unified management of all edge intelligent computing container images, and an edge intelligent computing container deployment tool for unified loading and deployment of edge intelligent computing containers to all edge nodes;

[0065] The edge intelligent computing container deployment agent software provides an edge intelligent computing container image cache warehouse for managing edge intelligent computing container images for local edge nodes, and an edge intelligent computing container deployment agent for loading and deploying edge intelligent computing containers to local edge nodes;

[0066] The edge intelligent computing container refers to a containerized entity that loads the edge intelligent computing service and the operating system and middleware it depends on, and its application software is the edge intelligent computing service.

[0067] like Figure 4 As shown in the figure, the detailed steps for lightweight deployment of edge computing containers across the entire network are as follows:

[0068] S31. Layered decoupling of edge intelligent computing container images. The edge intelligent computing container deployment service software decouples the edge intelligent computing container images into the system layer, dependency layer, and business layer according to functional logic: the system layer provides a streamlined operating system base image for the edge intelligent computing operating system; the dependency layer provides several intermediate software base images that are applicable to the same type of applications with the same dependencies; the business layer provides business logic application images that are specific to each application, where the business logic application image is configured with dependencies on the operating system base image and the intermediate software base image, and the name of the business logic application image is standardized to be the same as that of the corresponding edge intelligent computing service;

[0069] S32, edge intelligent computing container image deployment is determined. On the edge service node, the edge intelligent computing container deployment service software autonomously recommends and generates a recommended list of edge intelligent computing containers to be deployed in the entire network based on the edge intelligent computing full network resource utilization details and edge network communication full network resource utilization details, according to the edge intelligent computing service to be deployed information list, and manually adjusts and determines to form the edge intelligent computing container full network to be deployed list;

[0070] The edge intelligent computing network-wide resource utilization details include the total server intelligent computing power, server intelligent computing power usage, server intelligent computing power surplus, and server intelligent computing power pre-usage information of each edge intelligent server in each edge node of the entire network, where the server intelligent computing power pre-usage is enabled and initialized to zero when the edge intelligent computing container is deployed;

[0071] The edge network communication network resource utilization details include the total node communication bandwidth, node communication bandwidth usage, node communication bandwidth margin, and node communication bandwidth pre-usage information of each edge node in the entire network, where the node communication bandwidth pre-usage is enabled and initialized to zero when the edge intelligent computing container is deployed;

[0072] The list of edge intelligent computing services to be deployed includes the name of each edge intelligent computing service to be deployed, the number of required instances, the best deployment node, the required intelligent computing power, and the required communication bandwidth information;

[0073] The recommended list of edge intelligent computing containers to be deployed across the entire network includes the name of each edge intelligent computing service to be deployed and recommended deployment location information;

[0074] like Figure 5 As shown, the edge intelligent computing container image deployment recommendation: The edge intelligent computing container deployment service software is based on the edge intelligent computing full network resource utilization details and edge network communication full network resource utilization details, and according to the edge intelligent computing service to be deployed information list, it autonomously recommends and generates the edge intelligent computing container full network to be deployed recommended list. The detailed steps are as follows:

[0075] S321. The best node for direct deployment of edge intelligent computing services. The edge intelligent computing container deployment service software is based on the edge intelligent computing full network resource utilization details and the edge network communication full network resource utilization details. It determines in turn whether the best deployment node of each edge intelligent computing service in the edge intelligent computing service to be deployed information list has a node communication bandwidth surplus that can still meet the edge intelligent computing service's required communication bandwidth after deducting the node communication bandwidth pre-usage, and whether the server intelligent computing power surplus of several edge intelligent servers can still meet the edge intelligent computing service's required number of instances after deducting the server intelligent computing power pre-usage. Intelligent computing power: If yes, the deployment location of the corresponding edge intelligent computing service is recommended to be its best deployment node. The edge intelligent server records the name and recommended deployment location information of the corresponding edge intelligent computing service into the recommended list of edge intelligent computing containers to be deployed in the entire network, and accordingly increases the node communication bandwidth pre-use of the best deployment node by the required communication bandwidth of the corresponding edge intelligent computing service, and increases the server intelligent computing power pre-use of its several edge intelligent servers by the required intelligent computing power of the corresponding number of required instances of the edge intelligent computing service; otherwise, the name information of the corresponding edge intelligent computing service is recorded into the recommended list of edge intelligent computing service locations to be deployed, and the recommended list of edge intelligent computing service locations to be deployed includes the name information of each edge intelligent computing service recommended for the location to be deployed;

[0076] S322. The edge intelligent computing service can be deployed near the best node. The edge intelligent computing container deployment service software is based on the edge intelligent computing full network resource utilization details and the edge network communication full network resource utilization details. It sequentially determines whether there are edge nodes of the highest priority same-level and second priority direct superiors near the best deployment node of each edge intelligent computing service in the recommended list of edge intelligent computing service to be deployed. Is there a node communication bandwidth surplus that can still meet the required communication bandwidth of the edge intelligent computing service after deducting the node communication bandwidth pre-usage, and whether there are several edge intelligent servers whose server intelligent computing power surplus can still meet the required number of instances of the edge intelligent computing service after deducting the server intelligent computing power pre-usage: If so, it is recommended that the deployment location of the corresponding edge intelligent computing service is its best deployment node. Several edge intelligent servers adjacent to the edge node will be named, The recommended deployment location information is recorded in the edge intelligent computing container's recommended list of all network to be deployed, and the node communication bandwidth pre-usage of the optimal deployment node and the adjacent edge node is increased by the required communication bandwidth of the corresponding edge intelligent computing service, and the server intelligent computing power pre-usage of its several edge intelligent servers is increased by the required intelligent computing power of the corresponding number of required instances of the edge intelligent computing service; otherwise, the name, name, number of required instances, optimal deployment node, required intelligent computing power, and required communication bandwidth information of the corresponding edge intelligent computing service are recorded in the edge intelligent computing service's list of resources to be deployed and utilized for expansion, which includes the name, name, number of required instances, optimal deployment node, required intelligent computing power, and required communication bandwidth information of each edge intelligent computing service to be deployed and utilized for expansion;

[0077] S323, feedback on the resource requirements for edge intelligent computing services deployed without nodes. The edge intelligent computing container deployment service software determines whether the list of resources to be deployed and utilized for the edge intelligent computing service is empty: if so, it ends directly; otherwise, based on the list of resources to be deployed and utilized for the edge intelligent computing service and the list of information to be deployed for the edge intelligent computing service, the user is prompted with the name, number of required instances, optimal deployment node, required intelligent computing power, and required communication bandwidth information of each edge intelligent computing service in the list of resources to be deployed and utilized for the edge intelligent computing service;

[0078] S324, node-free deployment of edge intelligent computing service demand resource implementation. The user follows the prompts and manually adds edge intelligent servers and edge network communication equipment to meet the required intelligent computing power and required communication bandwidth of the corresponding edge intelligent computing service in the best deployment node of each edge intelligent computing service in the resource expansion list of the edge intelligent computing service to be deployed, or (when the best deployment node does not meet the implementation conditions) the edge nodes of the highest priority level and the second priority level that are directly above the best deployment node;

[0079] S325. Update of edge intelligent computing and communication network resource utilization details. The edge intelligent computing container deployment service software updates the edge intelligent computing network resource utilization details and edge network communication network resource utilization details. The edge intelligent computing network resource utilization details add the total server intelligent computing power of each edge intelligent server in the edge node of the edge intelligent computing service deployment and resource expansion, and the edge network communication network resource utilization details add the total node communication bandwidth of the edge node of the edge intelligent computing service deployment and resource expansion;

[0080] S326. The edge intelligent computing service update can be directly deployed on the best node. The edge intelligent computing container deployment service software again determines, based on the edge intelligent computing full network resource utilization details and the edge network communication full network resource utilization details, whether there is a node communication bandwidth surplus that can still meet the edge intelligent computing service's required communication bandwidth after deducting the node communication bandwidth pre-usage in the best deployment node of each edge intelligent computing service in the edge intelligent computing service deployment location recommendation list, and whether there are several edge intelligent servers whose server intelligent computing power surplus can still meet the edge intelligent computing service's required number of instances after deducting the server intelligent computing power pre-usage. If so, the deployment location of the corresponding edge intelligent computing service is recommended as its best deployment node. The edge intelligent servers of the edge intelligent servers shall record the name and recommended deployment location information of the corresponding edge intelligent computing service into the recommended list of edge intelligent computing containers to be deployed in the entire network, and accordingly increase the node communication bandwidth pre-use of the best deployment node by the required communication bandwidth of the corresponding edge intelligent computing service, and increase the server intelligent computing power pre-use of its edge intelligent servers by the required intelligent computing power of the corresponding number of required instances of the edge intelligent computing service; otherwise, record the name information of the corresponding edge intelligent computing service into the recommended list of edge intelligent computing service locations to be deployed, and the recommended list of edge intelligent computing service locations to be deployed includes the name information of each edge intelligent computing service recommended for the location to be deployed;

[0081] S327, feedback on the demand for nodeless deployment of edge intelligent computing services. The edge intelligent computing container deployment service software determines again whether there is still no node to deploy the edge intelligent computing service: if not, it ends directly; if yes, according to the resource expansion list of edge intelligent computing services to be deployed and the information list of edge intelligent computing services to be deployed, it feeds back to the user the name, number of required instances, optimal deployment node, required intelligent computing power, and required communication bandwidth information of each edge intelligent computing service in the resource expansion list of nodeless deployment edge intelligent computing services, and then goes to step S324.

[0082] S33, edge intelligent computing container image distribution across the entire network. On the edge service node, the edge intelligent computing container deployment service software selects each edge intelligent computing container image to be deployed on all edge nodes from the edge intelligent computing container image central warehouse in turn according to the edge intelligent computing container to be deployed list across the entire network, and uses the edge intelligent computing container image accelerated distribution to quickly and lightly distribute the selected edge intelligent computing container image to the edge intelligent computing container image cache warehouse of the edge intelligent computing container deployment agent software on the corresponding edge node, and then sends the edge intelligent computing container node to-be-deployed list to the edge intelligent computing container deployment agent software on the corresponding edge node. The edge intelligent computing container node to-be-deployed list includes the required number of instances of all edge intelligent computing containers that the edge node needs to deploy and the required loading business logic application image name;

[0083] like Figure 6 As shown, the edge intelligent computing container image accelerated distribution: through the edge intelligent computing container image accelerated distribution, the selected edge intelligent computing container image is lightly and quickly sent to the edge intelligent computing container image cache warehouse of the edge intelligent computing container deployment agent software on the corresponding edge node. The detailed steps are as follows:

[0084] S331, whether the edge node to be deployed with the edge intelligent computing container image is local. The edge intelligent computing container deployment service software determines whether the edge node to be deployed with the edge intelligent computing container image is local: if yes, there is no need to redundantly send the edge intelligent computing container image to the edge node to be deployed, and the process ends directly; otherwise, go to step S332;

[0085] S332, determine whether the edge node to be deployed of the edge intelligent computing container image is offline. The edge intelligent computing container deployment service software determines whether the edge node to be deployed of the edge intelligent computing container image is offline: if yes, wait until the edge node to be deployed is online, receive and update the version information of the edge intelligent computing container image in the edge container image cache warehouse of the edge node to be deployed, and then go to step S333; otherwise, go to step S333;

[0086] S333, determine whether the edge node to be deployed with the edge intelligent computing container image has an operating system image. The edge intelligent computing container deployment service software determines whether the edge node to be deployed with the edge intelligent computing container image has an operating system basic image (file) to be deployed: if not, based on the UDP connectionless transmission protocol applicable to the edge high-mobility environment, the packet sending response timeout retransmission mechanism above it, and the file block breakpoint continuation mechanism above it, the operating system basic image to be deployed is lightweight, fast and reliable to be sent to the edge node to be deployed; if yes, go to step S334;

[0087] S334, determine whether there are intermediate software images on the edge nodes to be deployed for the edge intelligent computing container image. The edge intelligent computing container deployment service software determines whether there are intermediate software basic images (files) to be deployed on the edge nodes to be deployed for the edge intelligent computing container image: if not, based on the UDP connectionless transmission protocol applicable to the edge high-mobility environment, the packet sending response timeout retransmission mechanism above it, and the file block breakpoint continuation mechanism above it, several intermediate software basic images to be deployed are lightly, quickly and reliably sent to the edge nodes to be deployed in sequence; if yes, go to step S335;

[0088] S335. Determine whether the edge node to be deployed with the edge intelligent computing container image has a business application image. The edge intelligent computing container deployment service software determines whether the edge node to be deployed with the edge intelligent computing container image has a business logic application image (file) to be deployed: if not, based on the UDP connectionless transmission protocol applicable to the edge high-mobility environment, the packet sending response timeout retransmission mechanism above it, and the file block breakpoint continuation mechanism above it, several business logic application images to be deployed are lightly, quickly and reliably sent to the edge node to be deployed in sequence; if yes, it ends directly.

[0089] S34. Edge intelligent computing container image deployment and loading: When the edge intelligent computing container deployment agent software on each edge node receives the list of edge intelligent computing container nodes to be deployed, it pulls and aggregates the edge intelligent computing container images to be deployed for each edge intelligent computing container to be deployed in turn through the edge intelligent computing container image perception according to the list of edge intelligent computing container nodes to be deployed, and loads the aggregated edge intelligent computing container image into the corresponding edge intelligent computing container.

[0090] like Figure 7 As shown, the edge intelligent computing container image perception aggregation: pull and aggregate the edge intelligent computing container image to be deployed for each edge intelligent computing container to be deployed through edge container image perception and load the aggregated edge intelligent computing container image into the corresponding edge intelligent computing container. The detailed steps are as follows:

[0091] S341, edge intelligent computing container operating system loading. The edge intelligent computing container deployment agent software compares the number of edge intelligent computing container instances required for deployment and the number of deployed instances to learn the number of edge intelligent computing container instances to be deployed, and creates and enables edge intelligent computing container instances in sequence according to the number of edge intelligent computing container instances to be deployed, including generating edge intelligent computing container instances and allocating intelligent computing power and communication bandwidth according to their intelligent computing power and communication bandwidth requirements, pulling the operating system base image from the edge container image cache warehouse and loading it into the edge intelligent computing container instance, deploying the edge intelligent computing container instance to the edge intelligent server at the recommended deployment location, starting the streamlined operating system in the edge intelligent computing container instance, and configuring the virtual network communication properties of the edge intelligent computing container instance;

[0092] S342, edge intelligent computing container intermediate software loading. The edge intelligent computing container deployment agent software compares the name and version information of the intermediate software basic image to be deployed (the number of intermediate software basic images to be deployed business logic application images to be deployed) of each edge intelligent computing container instance to be deployed, and the name and version information of the deployed intermediate software basic image, according to the number of edge intelligent computing container instances to be deployed. The list of intermediate software basic images to be deployed includes the name, version, and deployment execution operation information of each intermediate software basic image to be deployed, wherein the deployment execution operation of the intermediate software basic image to be deployed includes uninstallation, update, and installation. According to the list of intermediate software basic images to be deployed, each intermediate software basic image to be deployed is uninstalled, updated, or installed in the edge intelligent computing container instance to be deployed in turn;

[0093] S343, edge intelligent computing container business application loading. The edge intelligent computing container deployment agent software compares the name and version information of the business logic application image to be deployed of each edge intelligent computing container instance to be deployed, and the name and version information of the deployed business logic application image, according to the number of edge intelligent computing container instances to be deployed, and obtains the name, version, and deployment execution operation information of the business logic application image to be deployed corresponding to the edge intelligent computing container instance to be deployed, wherein the deployment execution operation of the business logic application image to be deployed includes update and installation, and accordingly updates or installs the business logic application image to be deployed to the edge intelligent computing container instance to be deployed;

[0094] S344, the edge intelligent computing container instance is restarted to take effect. The edge intelligent computing container deployment agent software restarts the newly created and updated edge intelligent computing container instance to take effect.

[0095] S4. Scheduling and allocation of edge computing resource containers. The edge computing resource scheduling software dynamically allocates and recycles edge intelligent computing, network communication and other computing resources assigned to edge intelligent computing containers through edge computing resource container scheduling.

[0096] The edge computing resource scheduling software can dynamically allocate and recycle computing resources such as edge intelligent computing and network communication assigned to edge intelligent computing containers.

[0097] like Figure 8 As shown in the figure, the detailed steps of edge computing resource container scheduling and allocation are as follows:

[0098] S41, dynamic allocation of edge computing resource container scheduling. When any edge intelligent computing container instance is enabled, the edge computing resource scheduling software schedules the edge intelligent server deployed by the edge intelligent computing container instance to allocate edge intelligent computing resources equal to the assigned intelligent computing power, and the edge communication device deployed by the edge intelligent computing container instance to allocate edge network communication resources equal to the assigned communication bandwidth to the edge intelligent computing container instance according to the assigned intelligent computing power and communication bandwidth of the edge intelligent computing container instance;

[0099] S42, dynamic recycling of edge computing resource container scheduling. When any edge intelligent computing container instance is withdrawn, the edge computing resource scheduling software schedules the edge intelligent server deployed by the edge intelligent computing container instance to recycle edge intelligent computing resources equal to the assigned intelligent computing power, and the edge communication device deployed by the edge intelligent computing container instance to recycle edge network communication resources equal to the assigned communication bandwidth to the edge intelligent computing container instance according to the assigned intelligent computing power and communication bandwidth of the edge intelligent computing container instance.

[0100] (II) The present invention also provides an edge intelligent hardware platform architecture system suitable for a high mobility environment, comprising:

[0101] In the edge service nodes, deploy edge intelligent servers for training, edge intelligent servers for reasoning, edge network communication equipment, edge computing resource virtualization software, edge computing resource management software, edge intelligent computing container deployment service software, edge intelligent computing container deployment agent software, and edge computing resource scheduling software;

[0102] In edge terminal nodes, deploy edge intelligent servers for inference, edge network communication equipment, edge computing resource virtualization software, edge intelligent computing container deployment agent software, and edge computing resource scheduling software;

[0103] The training edge intelligent server is equipped with several training edge intelligent chips, edge intelligent acceleration cards or edge intelligent computing modules, which provide the basic hardware environment of edge intelligent computing, network communication and other computing resources for regular online incremental training optimization of edge intelligent models, and support the update of edge intelligent models after iterative optimization to the container image of edge intelligent services;

[0104] The edge intelligent server for inference is equipped with several edge intelligent chips, edge intelligent acceleration cards or edge intelligent computing modules of the inference type, and provides a basic hardware environment of computing and transmission resources such as edge intelligent computing and network communication for the loading and operation of edge intelligent services;

[0105] The edge network communication device provides external network communication bandwidth for edge nodes;

[0106] The edge computing and transmission resource virtualization software virtualizes and uniformly describes computing and transmission resources such as intelligent computing and network communication of edge nodes;

[0107] The edge computing and transmission resource management software pools and manages computing and transmission resources such as intelligent computing and network communication of all edge nodes;

[0108] The edge intelligent computing container deployment service software provides an edge intelligent computing container image central warehouse for unified management of all edge intelligent computing container images, and an edge intelligent computing container deployment tool for unified loading and deployment of edge intelligent computing containers to all edge nodes;

[0109] The edge intelligent computing container deployment agent software provides an edge intelligent computing container image cache warehouse for managing edge intelligent computing container images for local edge nodes, and an edge intelligent computing container deployment agent for loading and deploying edge intelligent computing containers to local edge nodes;

[0110] The edge computing resource scheduling software can dynamically allocate and recycle computing resources such as edge intelligent computing and network communication assigned to edge intelligent computing containers.

Claims

1. An edge intelligent hardware platform architecture method suitable for a high mobility environment, characterized in that: The method comprises the following steps: S1. Preparation of basic hardware environment for edge intelligence: Equip all edge nodes with edge intelligence hardware. Edge nodes include edge service nodes and edge terminal nodes. Edge service nodes are equipped with training edge intelligence hardware and reasoning edge intelligence hardware, and edge terminal nodes are equipped with reasoning edge intelligence hardware. S2. Virtual pooling and management of edge computing and transmission resources: Virtualize and uniformly describe the intelligent computing and network communication computing and transmission resources of all edge nodes, and manage them to form an edge computing and transmission virtual resource pool; S3. Lightweight deployment of edge intelligent computing containers across the entire network: Through the interactive collaboration between the edge intelligent computing container deployment service software and the edge intelligent computing container deployment agent software, system operation and maintenance personnel can perform lightweight layered decoupling, deployment recommendation, accelerated distribution, and perception aggregation of container images suitable for edge high-mobility environments, and distributed loading and deployment of edge intelligent computing containers to all edge nodes. S4. Scheduling and allocation of edge computing resource containers: The edge computing resource scheduling software dynamically allocates and recycles edge intelligent computing and network communication computing resources assigned to edge intelligent computing containers through edge computing resource container scheduling. The step S2 comprises: S21. Computing and transmission resource entity affiliation hierarchical planning: The edge computing and transmission resource virtualization software divides the entity affiliation of the edge node's intelligent computing and network communication computing and transmission resources into two-level mapping descriptions: nodes and servers; S22. Virtual pooling of edge intelligent computing resources: The edge computing resource virtualization software virtualizes and uniformly describes the intelligent computing resources of all edge nodes, and the management software manages them to form an edge intelligent computing virtual resource pool. The edge intelligent computing virtual resources in the edge intelligent computing virtual resource pool are represented as follows: ， in, Indicates the block number in the virtual resource pool; Indicates the edge intelligent computing virtual resource pool Block number edge intelligent computing virtual resources have intelligent computing power; express The entity that provides the block number virtual resource; Indicates the virtual resource pool The entity that provides the block number virtual resource belongs to the node; Indicates the edge intelligent computing virtual resource pool The provider entity of the block number edge intelligent computing virtual resource belongs to the edge intelligent server; Indicates the edge intelligent computing virtual resource pool Block number Description of edge intelligent computing virtual resources; S23, virtual pooling of edge network communication resources: virtualize and uniformly describe the network communication resources of all edge nodes, manage them to form an edge network communication virtual resource pool, and the edge network communication virtual resources in the edge network communication virtual resource pool are represented as follows: ， in, Indicates the block number in the virtual resource pool; Indicates the edge network communication virtual resource pool Block number edge network communication virtual resource virtual network communication attributes; Indicates the edge network communication virtual resource pool Block number: the communication bandwidth owned by the edge network communication virtual resource; express The entity that provides the block number virtual resource; Indicates the virtual resource pool The entity that provides the block number virtual resource belongs to the node; Indicates the edge network communication virtual resource pool Block number Description of edge network communication virtual resource.

2. The edge intelligent hardware platform architecture method suitable for high mobility environment according to claim 1 is characterized in that: The step S1 includes the preparation of the basic hardware environment for edge intelligence training and the preparation of the basic hardware environment for edge intelligence reasoning.

3. The edge intelligent hardware platform architecture method suitable for high mobility environment according to claim 2 is characterized in that: The basic hardware environment preparation for edge intelligence training includes system construction personnel installing training-type edge intelligence chips, edge intelligence acceleration cards or edge intelligence computing modules on the edge intelligence server for training in the edge service node.

4. The edge intelligent hardware platform architecture method suitable for high mobility environment according to claim 2 is characterized in that: The preparation of the basic hardware environment for edge intelligent reasoning includes system construction personnel installing inference-type edge intelligent chips, edge intelligent acceleration cards or edge intelligent computing modules on edge intelligent servers for reasoning in edge service nodes and edge terminal nodes.

5. The edge intelligent hardware platform architecture method suitable for high mobility environment according to claim 1, characterized in that: The step S3 comprises: S31. Layered decoupling of edge intelligent computing container images: Decouple edge intelligent computing container images into system layer, dependency layer and business layer according to functional logic: the system layer provides a streamlined operating system base image for the edge intelligent computing operating system; the dependency layer provides several intermediate software base images with the same dependencies for similar applications; the business layer provides business logic application images specific to each application; S32, edge intelligent computing container image deployment determination: on the edge service node, based on the edge intelligent computing full network resource utilization details and edge network communication full network resource utilization details, according to the edge intelligent computing service to be deployed information list, autonomously recommend and generate the edge intelligent computing container full network to be deployed recommended list, and manually adjust and determine to form the edge intelligent computing container full network to be deployed list; S33, edge intelligent computing container image distribution across the entire network: On the edge service node, according to the edge intelligent computing container to be deployed list across the entire network, select each edge intelligent computing container image to be deployed on all edge nodes from the edge intelligent computing container image central warehouse in turn, and use the edge intelligent computing container image accelerated distribution to quickly and lightly distribute the selected edge intelligent computing container image to the edge intelligent computing container image cache warehouse of the corresponding edge node, and issue the edge intelligent computing container node to-be-deployed list to the corresponding edge node; S34. Edge intelligent computing container image deployment and loading: When each edge node receives the list of edge intelligent computing container nodes to be deployed, it pulls and aggregates each edge intelligent computing container image to be deployed in turn through the edge intelligent computing container image perception according to the list of edge intelligent computing container nodes to be deployed, and loads the aggregated edge intelligent computing container image into the corresponding edge intelligent computing container.

6. The edge intelligent hardware platform architecture method suitable for high mobility environment according to claim 1, characterized in that: The step S4 comprises: S41. When any edge intelligent computing container instance is enabled, according to the assigned intelligent computing power and communication bandwidth of the edge intelligent computing container instance, the edge intelligent server deployed by the edge intelligent computing container instance is scheduled to allocate edge intelligent computing resources equal to the assigned intelligent computing power, and the edge communication device deployed by the edge intelligent computing container instance is scheduled to allocate edge network communication resources equal to the assigned communication bandwidth to the edge intelligent computing container instance; S42. When any edge intelligent computing container instance is withdrawn, the edge intelligent server deployed by the edge intelligent computing container instance is scheduled to recover edge intelligent computing resources equal to the assigned intelligent computing power and communication bandwidth of the edge intelligent computing container instance, and the edge communication equipment deployed by the edge intelligent computing container instance is scheduled to recover edge network communication resources equal to the assigned communication bandwidth to the edge intelligent computing container instance.

7. An edge intelligent hardware platform architecture system suitable for a high mobility environment, implemented by the edge intelligent hardware platform architecture method suitable for a high mobility environment as described in any one of claims 1 to 6, characterized in that: include: Deploy edge intelligent servers for training, edge intelligent servers for reasoning, edge network communication equipment, edge computing resource virtualization software, edge computing resource management software, edge intelligent computing container deployment service software, edge intelligent computing container deployment agent software, and edge computing resource scheduling software in edge service nodes; Deploy edge intelligent servers for inference, edge network communication equipment, edge computing resource virtualization software, edge intelligent computing container deployment agent software, and edge computing resource scheduling software in edge terminal nodes.

8. The edge intelligent hardware platform architecture system suitable for high mobility environment according to claim 7, characterized in that: The training edge intelligent server is equipped with a number of training-type edge intelligent chips, edge intelligent acceleration cards or edge intelligent computing modules, which provides a basic hardware environment of edge intelligent computing and network communication resources for regular online incremental training optimization of edge intelligent models, and supports updating the edge intelligent model after iterative optimization to the container image of the edge intelligent service.

9. The edge intelligent hardware platform architecture system suitable for high mobility environment according to claim 7, characterized in that: The edge intelligent server for inference is equipped with a number of edge intelligent chips, edge intelligent acceleration cards or edge intelligent computing modules for inference, providing a basic hardware environment of edge intelligent computing and network communication computing resources for the loading and operation of edge intelligent services.

Images