A microservice-based cloud-network-edge intelligent collaboration and management orchestration method

Abstract

This invention discloses a microservice-based cloud-network-edge intelligent collaboration and management orchestration method, applied to a cloud-network-edge intelligent collaboration framework system. This system comprises three domains: cloud, network, and edge. Each domain has a four-layer, three-plane architecture. The four layers consist of an infrastructure layer, a virtualization layer, a functional layer, and an application layer, designed from bottom to top. The three planes include a control plane, an intelligence plane, and a MANO plane. The infrastructure, virtualization, functional, and application layers interact with the control plane, intelligence plane, and MANO plane, respectively. The control plane, intelligence plane, and MANO plane also interact with each other. The corresponding methods include: implementing the distributed intelligent algorithm training and inference process based on microservices; and implementing the network function template instantiation process based on Kubernetes. This invention achieves efficient deployment.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, specifically relating to a cloud-network-edge intelligent collaboration and management orchestration method based on microservices. Background Technology

[0002] With the development of cloud computing, 5G networks, and edge computing, artificial intelligence can be moved from the powerful cloud to the edge, closer to the user, reducing latency and providing faster service. However, edge computing is limited by computing and storage resources, supporting only small-scale intelligent algorithm inference and training. Therefore, it is crucial to combine cloud computing, edge computing, and 5G networks for collaborative training and inference of intelligent algorithms.

[0003] However, current research both domestically and internationally focuses on cloud-edge collaborative architectures. Most technologies involve deploying cloud applications to edge nodes and using cloud-edge-device integration to uniformly manage massive edge resources and services on cloud servers. These studies focus on how to deploy intelligent algorithms in cloud computing platforms, but centralized data processing inevitably puts pressure on the network and may leak user privacy. Furthermore, the deployment flexibility and agility are not high. Summary of the Invention

[0004] To address the aforementioned problems in existing technologies, this invention provides a cloud-network-edge intelligent collaboration and management orchestration method based on microservices. The technical problem to be solved by this invention is achieved through the following technical solution:

[0005] This invention provides a microservice-based cloud-network-edge intelligent collaboration and management orchestration method, applied to a cloud-network-edge intelligent collaboration framework system, which includes:

[0006] The system comprises three domains: cloud, network, and edge. Each domain has a four-layer, three-sided architecture. The four-layer architecture includes, from bottom to top, an infrastructure layer, a virtualization layer, a functional layer, and an application layer. The three-sided architecture includes a control plane, an intelligence plane, and a management orchestration plane. The infrastructure layer, virtualization layer, functional layer, and application layer interact with the control plane, intelligence plane, and management orchestration plane, respectively. The control plane, intelligence plane, and management orchestration plane interact with each other, and the corresponding methods include:

[0007] The system implements distributed intelligent algorithm training and inference processes based on microservices;

[0008] The instantiation process of the system's network function template based on Kubernetes;

[0009] In the training and inference process, the microservice concept is used to decouple the training and inference process of the distributed intelligent algorithm into several network functions.

[0010] In one embodiment of the present invention, the functional layer of the intelligent surface utilizes the microservice concept to realize the distributed intelligent algorithm training and inference process of the system.

[0011] In one embodiment of the present invention, the plurality of network functions include five network functions: data collection, data preprocessing, modeling, distribution, and aggregation.

[0012] In one embodiment of the present invention, Network Function Virtualization (NFV) technology is used to virtualize all network functions.

[0013] In one embodiment of the present invention, the plurality of network functions are all mounted on a unified virtual Service-based Interface (SBI) bus, and the Hypertext Transfer Protocol (HTTP) is used to obtain and provide the corresponding services.

[0014] In one embodiment of the present invention, the functional layer of the management orchestration surface utilizes Kubernetes to implement the instantiation process of the system's network function templates.

[0015] In one embodiment of the present invention, client-java and client-go provided by Kubernetes are used for secondary development, and the template instantiation of various network functions and the management and orchestration of resources are realized by interacting with the API Server.

[0016] In one embodiment of the present invention, the instantiation process of the network function template of the Kubernetes-based implementation system includes:

[0017] After receiving the distributed intelligent algorithm service request sent by the application layer, the management orchestration plane sends model request information to the model selector.

[0018] The model selector selects a suitable neural network model based on the model request information and requests a predefined neural network model template from the intelligent algorithm model library.

[0019] After receiving the template request information sent by the model selector, the intelligent algorithm model library updates the predefined parameters sent by the model selector to the corresponding neural network model template, and returns the updated neural network model template to the management orchestration surface.

[0020] The management orchestration surface receives the neural network model template sent by the intelligent algorithm model library, sends the neural network model template to the API Server, and requests the creation of the corresponding Pod instance;

[0021] After receiving the Pod creation request information sent by the management orchestration plane, the API Server writes the neural network model template received by the API Server into Etcd;

[0022] After receiving the neural network model template information sent by the API Server, Etcd saves it to the database and returns updated database information to the API Server.

[0023] The API Server sends a Pod creation request to the Scheduler. After receiving the Pod creation request through the Watch mechanism, the Scheduler selects a suitable Node through its internal scheduling mechanism and returns the Node to the API Server.

[0024] After receiving the Node sent by the Scheduler, the API Server writes the information corresponding to the Node into the Etcd and sends Pod binding request information to the Kubelet of the Node.

[0025] After receiving the Pod binding request information sent by the API Server, the Kubelet instantiates the neural network model template into a Docker container using the container runtime based on the neural network model template in the Etcd. It then allocates corresponding resources to the Docker container based on the CPU, Memory, and network resource information in the neural network model template to complete the instantiation of the system's network function template and complete the response to the distributed intelligent algorithm service request sent by the application layer.

[0026] In one embodiment of the present invention, the distributed intelligent algorithm training process based on the microservice implementation system includes:

[0027] Users access the core network through 5G base stations and connect to the cloud-network-edge intelligent collaborative framework system via wireless networks, and send requests to the application layer of the control plane.

[0028] After receiving the request, the application layer of the control plane obtains the service of the functional layer of the intelligent plane through the RESTful interface and sends the service request.

[0029] The functional layer of the intelligent surface receives the corresponding service request, parses the Uniform Resource Locator (URL), and sends a network application request to the management orchestration surface via the HTTP protocol;

[0030] After the management orchestration plane listens to the network application request sent by the functional layer of the smart plane, it obtains the network function of the smart plane that needs to be instantiated by parsing the URL, obtains the template of the network function of the smart plane from the template library, instantiates the template with the help of Kubernetes, deploys the corresponding network function, and sends a request to the distribution network function of the cloud domain.

[0031] After receiving the request, the distribution network function selects suitable edge nodes to participate in this round of distributed intelligent algorithm training through its internal node selection function, obtains the neural network model from the model network function, and distributes the obtained neural network model to the selected edge nodes.

[0032] After receiving the neural network model sent by the distribution network function, the data collection network function of the selected edge node begins to collect the corresponding data and puts it into the database;

[0033] The data preprocessing network function reads the data put into the database by the data collection network function, performs data preprocessing by truncation, zero padding, normalization and labeling, puts the processed data into the database, and sends a training request to the model network function.

[0034] After receiving the training request from the data preprocessing network function, the model network function performs a round of model training using the neural network model sent by the distribution network function and the data processed by the data preprocessing network function, and sends the trained neural network model to the aggregation network function of the cloud domain via HTTP protocol.

[0035] The cloud domain's aggregation network function receives the model parameters from the edge nodes, performs federated aggregation to obtain the aggregated model, and sends the aggregated model to the cloud domain's distribution network function via HTTP protocol. The process of the cloud domain's distribution network function receiving the request is repeated until the neural network model reaches the desired accuracy.

[0036] In one embodiment of the present invention, the distributed intelligent algorithm inference process of the microservice-based implementation system includes:

[0037] Users access the core network through 5G base stations and connect to the cloud-network-edge intelligent collaborative framework system via wireless networks, and send requests to the application layer of the control plane.

[0038] After receiving the request, the application layer of the control plane obtains the service of the functional layer of the intelligent plane through the RESTful interface and sends the service request.

[0039] The functional layer of the intelligent surface receives the corresponding service request, parses the URL, and sends the network application request to the management orchestration surface via the HTTP protocol;

[0040] After the management orchestration plane listens to the network application request sent by the functional layer of the smart plane, it obtains the network function of the smart plane to be instantiated by parsing the URL, obtains the template of the network function of the smart plane from the template library, instantiates the template with the help of Kubernetes, deploys the corresponding network function, and sends a request to the data collection network function of the edge domain.

[0041] After receiving the request sent by the management orchestration plane, the data collection network function begins to collect the corresponding data and puts it into the database;

[0042] The data preprocessing network function reads the data put into the database by the data collection network function, performs data preprocessing by truncation, zero padding, and normalization, and requests the calculation results from the model network function via the HTTP protocol.

[0043] After receiving the request from the data preprocessing network function, the model network function parses the URL, selects the neural network model chosen in the request, and inputs the data processed by the data preprocessing network function into the neural network model to obtain the final result, which is then returned to the user.

[0044] The beneficial effects of this invention are:

[0045] This invention proposes a microservice-based cloud-network-edge intelligent collaboration and management orchestration method. This method abstracts the wireless network into a "three-domain, four-layer, three-plane" cloud-network-edge intelligent collaborative architecture system, comprising cloud domain, network domain, and edge domain; application layer, functional layer, virtualization layer, and infrastructure layer; and control plane, intelligence plane, and MAMO plane. Based on the microservice concept, the system's distributed intelligent algorithm training and inference process is decoupled into several network functions. These network functions can be reconstructed into different intelligent services, and Kubernetes is used to instantiate the system's network function templates, thereby achieving efficient deployment based on microservices and real-time monitoring of the entire lifecycle of network functions. Therefore, this invention achieves efficient deployment by combining cloud computing, edge computing, and 5G networks and introducing microservices into the cloud-network-edge intelligent collaborative architecture.

[0046] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the cloud-network-edge intelligent collaborative framework system provided in an embodiment of the present invention;

[0048] Figure 2 This is a flowchart illustrating a cloud-network-edge intelligent collaboration and management orchestration method based on microservices provided in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of the four-layer, three-sided structure of the edge domain provided in an embodiment of the present invention;

[0050] Figure 4 This is a schematic diagram of the functional layer of the intelligent plane based on a microservice architecture provided in an embodiment of the present invention;

[0051] Figure 5 This is a schematic diagram illustrating the instantiation process of the network function template for the Kubernetes-based system provided in this embodiment of the invention.

[0052] Figure 6 This is a schematic diagram of the distributed intelligent algorithm training process based on a microservice-based system provided in an embodiment of the present invention;

[0053] Figure 7 This is a schematic diagram of the distributed intelligent algorithm inference process of a system based on microservices, provided in an embodiment of the present invention.

[0054] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0055] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0056] To achieve distributed data processing and highly flexible, agile deployment, please refer to [link to relevant documentation]. Figure 1 This invention proposes a novel cloud-network-edge intelligent collaborative framework system. This system comprises three domains: cloud, network, and edge. Each domain has a four-layer, three-sided structure. The four-layer structure includes, from bottom to top, an infrastructure layer, a virtualization layer, a functional layer, and an application layer. The three-sided structure includes a control plane, an intelligence plane, and a management orchestration plane. The infrastructure layer, virtualization layer, functional layer, and application layer interact with the control plane, intelligence plane, and management orchestration plane, respectively. The control plane, intelligence plane, and management orchestration plane also interact with each other. Based on... Figure 1 The cloud-network-edge intelligent collaborative framework system shown includes the following methods:

[0057] The system implements distributed intelligent algorithm training and inference processes based on microservices;

[0058] The instantiation process of the system's network function template based on Kubernetes;

[0059] In the training and inference process, the microservice concept is used to decouple the training and inference process of the distributed intelligent algorithm into several network functions.

[0060] Here, the management and orchestration aspect is referred to as MANO.

[0061] In this embodiment of the invention, the network domain mainly consists of network devices and nodes, such as 5G base stations, home routers, and switches. They provide network resources for always-available services, security, and real-time privacy. As the next hop for the Internet of Things (IoT), these devices act as proxies connecting the IoT to the network and the cloud. In the architecture proposed in this embodiment, the network domain acts as a medium connecting the edge domain and the cloud domain.

[0062] In this embodiment of the invention, the edge domain mainly consists of edge computing located close to the user, collaborating with the network domain and cloud domain to provide users with computing, storage, and network services. Common services can be deployed in the edge domain to provide agile and reliable services to end users, improving service quality. Deploying artificial intelligence technology in edge computing to achieve edge intelligence ensures user data privacy and security while alleviating the pressure on 5G network data transmission.

[0063] In this embodiment of the invention, the cloud domain is mainly composed of cloud computing, and the 5G core network is deployed in a cloud-native manner within the cloud domain, providing powerful computing, storage, and network resources. It also provides global long-term decision-making capabilities. The cloud domain primarily handles massive amounts of data and non-real-time services, and performs task scheduling and management globally. The cloud domain also needs to ensure the robust operation of the 5G core network, provide resources for 5G network functions, and manage and orchestrate them throughout their entire lifecycle.

[0064] Figure 3 The overall architecture of the edge domain is a "four-layer, three-face" structure. The aforementioned "three domains," and the overall architecture of each domain's "four-layer, three-face" structure, are related to... Figure 3 The architectures shown are similar, but the specific network functions of each domain differ. All three domains follow a "four-layer, three-plane" architecture design. The "four layers" refer to the infrastructure layer, virtualization layer, functional layer, and application layer; the "three planes" refer to the control plane, intelligence plane, and MAMO plane. Inter-domain interaction is achieved through a unified virtual SBI bus, using the HTTP protocol as the logical bus implementation.

[0065] Next, we will introduce the “four-layer, three-face” structure of the cloud-network-edge intelligent collaborative framework system in detail.

[0066] In this embodiment of the invention, the infrastructure layer is located at the bottom layer of the entire architecture, providing the necessary computing, caching, and network resources for edge computing and other applications.

[0067] In this embodiment of the invention, the virtualization layer encompasses the control plane, the intelligence plane, and the MAMO plane, primarily utilizing NFV technology to abstract and virtualize the underlying physical hardware resources. Specifically, the control plane virtualization layer interacts with the infrastructure layer via a remote southbound interface using Docker Engine and Hypervisor, abstracting and virtualizing the physical hardware resources provided by the infrastructure layer. The intelligence plane virtualization layer provides a machine learning algorithm library and interacts with the control plane and MAMO plane virtualization layers respectively through RESTful APIs. The MAMO plane virtualization layer is mainly managed by the Docker Virtualized Infrastructure Manager (VIM) and the Hypervisor VIM through RESTful APIs, managing the resources of the MAMO plane and control plane virtualization layers respectively to achieve better resource utilization and load balancing.

[0068] In this embodiment of the invention, the application layer mainly comprises the various services provided by the control plane, the intelligence plane, and the MAMO plane. Specifically, the application layer of the control plane primarily provides edge computing, edge caching, resource allocation, and intelligent services; the application layer of the intelligence plane mainly provides various intelligent algorithm services; and the application layer of the MAMO plane interacts with the application layers of the intelligence plane and the control plane via a RESTful API, centrally managing and orchestrating all application layers according to the business needs of edge computing, thereby providing better services to users.

[0069] In this embodiment of the invention, the functional layer encompasses the control plane, the intelligence plane, and the MAMO plane. This functional layer is designed using a microservices approach. First, it utilizes NFV technology to virtualize all network functions. Second, it mounts all functional networks onto the same virtual SBI bus, using the HTTP protocol for service acquisition and provision. Finally, the functional network functions of the control plane, intelligence plane, and MAMO plane can flow freely along the SBI bus, without being limited to any specific plane.

[0070] In this embodiment of the invention, the functional layer of the control plane mainly includes application selection function, network function storage function, unified data management function, service registration function, communication protocol conversion function, and user plane function (UPF). These network functions mainly provide edge computing with services such as service selection, service registration, and service discovery.

[0071] In this embodiment of the invention, the functional layer of the intelligent surface, leveraging the microservice concept, decouples the training and inference processes of distributed intelligent algorithms, such as federated learning, into several functional networks, i.e., constructs corresponding network functions for them. In this embodiment, the several functional networks include five network functions: data collection, data preprocessing, modeling, distribution, and aggregation, to achieve intelligent services, such as... Figure 4 This invention presents the design of the functional layer of the intelligent surface based on a microservice architecture. In the diagram, "Collector" represents the data collection network function, "Pre-Producer" represents the data preprocessing network function, "Model" represents the model network function, "Distributor" represents the distribution network function, and "Aggregation" represents the aggregation network function. These network functions are all mounted on a unified virtual SBI bus and utilize the HTTP protocol for service acquisition and provision, thereby enabling collaborative training and inference of distributed machine learning models based on a microservice-based cloud-network-edge architecture, and allowing for more flexible and agile deployment of distributed intelligent algorithms.

[0072] In this embodiment of the invention, the MANO surface functional layer interacts with the intelligent surface functional layer and the control surface functional layer via a RESTful API. It is primarily responsible for resource allocation, health monitoring, full lifecycle management, and possesses a certain degree of autonomy for the network functions of the intelligent and control surfaces. The MANO surface functional layer is implemented using Kubernetes. Furthermore, to better manage and orchestrate the network functions of the control and intelligent surfaces, such as edge computing, this embodiment proposes using Kubernetes' client-java and client-go tools for secondary development. This involves interacting with the API Server to implement edge computing network functions, as well as template instantiation and resource management and orchestration for other network functions.

[0073] based on Figure 1 and Figure 3 For the structure shown, please refer to [link / reference]. Figure 5 This invention proposes an optional solution for implementing the network function template instantiation process of the system based on Kubernetes, including:

[0074] The MANO face, as a daemon process, always listens to the port number for communication with the application layer. After receiving the distributed intelligent algorithm service request sent by the application layer, the MANO face sends the model request information to the model selector.

[0075] The model selector selects a suitable neural network model based on the model request information and requests a predefined neural network model template from the intelligent algorithm model library;

[0076] After receiving the template request information sent by the model selector, the intelligent algorithm model library updates the predefined parameters sent by the model selector to the corresponding neural network model template, and returns the updated neural network model template to the MANO face.

[0077] MANO receives the neural network model template sent by the intelligent algorithm model library, sends the neural network model template to the API Server, and requests the creation of the corresponding Pod instance.

[0078] After receiving the Pod creation request information sent by the MANO face, the API Server writes the neural network model template received by the API Server into Etcd;

[0079] After receiving the neural network model template information sent by the API Server, Etcd saves it to the database and returns the updated database information to the API Server.

[0080] The API Server sends a Pod creation request to the Scheduler. After receiving the Pod creation request through the Watch mechanism, the Scheduler selects a suitable Node through its internal scheduling mechanism and returns the Node to the API Server.

[0081] After receiving the Node sent by the Scheduler, the API Server writes the information corresponding to the Node to Etcd and sends the Pod binding request information to the Kubelet of the Node.

[0082] After receiving the Pod binding request information sent by the API Server, Kubelet instantiates the neural network model template into a Docker container using the container runtime based on the neural network model template in Etcd. It then allocates corresponding resources to the Docker container according to the CPU, Memory, and network resource information in the neural network model template to complete the instantiation of the system's network function template and complete the response to the distributed intelligent algorithm service request sent by the application layer.

[0083] It should be noted that network functions are obtained by decoupling the neural network. Different neural networks will result in different network functions.

[0084] Furthermore, based on Figure 1 , Figure 3 and Figure 4 For the structure shown, please refer to [link / reference]. Figure 6 This invention proposes an optional solution for implementing a distributed intelligent algorithm training process based on microservices, including:

[0085] Users access the core network through 5G base stations and connect to the cloud-network-edge intelligent collaborative framework system through wireless networks, and send requests to the application layer of the control plane.

[0086] After receiving the request, the application layer of the control plane obtains the service from the functional layer of the intelligent plane through the RESTful interface and sends the service request.

[0087] The functional layer of the intelligent face receives the corresponding service request, parses the URL, and sends the network application request to the MANO face via the HTTP protocol;

[0088] After the MANO facet detects network application requests sent by the functional layer of the intelligent facet, it parses the URL to obtain the network functions of the intelligent facet that need to be instantiated, retrieves the templates for the network functions of the intelligent facet from the template library, instantiates the templates using Kubernetes, deploys the corresponding network functions, and sends requests to the distribution network functions of the cloud domain. Here, the deployment of the corresponding network functions utilizes the intelligent facet's distribution network functions, model network functions, and aggregation network functions in the cloud domain, as well as the edge domain's data collection network functions, data preprocessing network functions, and model network functions. For details on instantiating templates using Kubernetes, please refer to [link to documentation / reference]. Figure 5 The instantiation process is shown below;

[0089] After receiving a request, the distribution network function selects suitable edge nodes to participate in this round of distributed intelligent algorithm training through internal node selection functions, such as the NodeSelector function, and distributes the obtained neural network model to the selected edge nodes. Here, if it is the first training, that is, when the training is initialized, it is necessary to select a suitable neural network model from the model network function. It is not necessary to obtain a neural network model from the model network function in the later stages.

[0090] After receiving the neural network model sent by the distribution network function, the data collection network function of the selected edge node begins to collect the corresponding data and puts it into the database;

[0091] The data preprocessing network function reads the data put into the database by the data collection network function, performs data preprocessing by truncation, zero padding, normalization and labeling, puts the processed data into the database, and sends a training request to the model network function.

[0092] After receiving the training request from the data preprocessing network function, the model network function uses the neural network model sent by the distribution network function and the data processed by the data preprocessing network function to perform a round of model training, and then sends the trained neural network model to the cloud domain aggregation network function via HTTP protocol.

[0093] The cloud domain's aggregation network function receives the model parameters from the edge nodes, performs federated aggregation to obtain the aggregated model, and sends the aggregated model to the cloud domain's distribution network function via HTTP protocol. The process of receiving the request by the cloud domain's distribution network function is repeated until the neural network model reaches the desired accuracy.

[0094] Furthermore, based on Figure 1 , Figure 3 and Figure 4 For the structure shown, please refer to [link / reference]. Figure 7 This invention proposes an optional solution for implementing the distributed intelligent algorithm inference process of the system based on microservices, including:

[0095] Users access the core network through 5G base stations and connect to the cloud-network-edge intelligent collaborative framework system through wireless networks, and send requests to the application layer of the control plane.

[0096] After receiving the request, the application layer of the control plane obtains the service from the functional layer of the intelligent plane through the RESTful interface and sends the service request.

[0097] The functional layer of the intelligent face receives the corresponding service request, parses the URL, and sends the network application request to the MANO face via the HTTP protocol;

[0098] After the MANO face detects the network application request sent by the functional layer of the intelligent face, it parses the URL to obtain the network function of the intelligent face to be instantiated, retrieves the template of the network function of the intelligent face from the template library, instantiates the template with the help of Kubernetes, deploys the corresponding network function, and sends a request to the edge domain data collection network function. Here, the deployment of the corresponding network function uses the edge domain data collection network function, data preprocessing network function, and model network function. For template instantiation with Kubernetes, please refer to [link to documentation]. Figure 5 The instantiation process is shown below;

[0099] After receiving a request from the MANO surface, the data collection network function begins to collect the corresponding data and puts it into the database;

[0100] The data preprocessing network function reads the data put into the database by the data collection network function, performs data preprocessing by truncation, zero padding, and normalization, and requests the calculation results from the model network function via the HTTP protocol.

[0101] After receiving the request from the data preprocessing network function, the model network function parses the URL, selects the neural network model chosen in the request, and inputs the data processed by the data preprocessing network function into the neural network model to obtain the final result, which is then returned to the user.

[0102] It should be noted that the embodiments of the present invention achieve efficient deployment under the "three domains, four layers, and three faces" framework. During the deployment process, the basic knowledge of Kubernetes and microservices will not be elaborated in detail here. For example, secondary development is carried out using client-java and client-go provided by Kubernetes, and the network function template instantiation function required by the embodiments of the present invention is implemented using existing client-java and client-go functions.

[0103] In summary, the cloud-network-edge intelligent collaboration and management orchestration method based on microservices proposed in this invention abstracts the wireless network into a "three-domain, four-layer, three-plane" cloud-network-edge intelligent collaborative architecture system, comprising cloud domain, network domain, and edge domain; application layer, functional layer, virtualization layer, and infrastructure layer; and control plane, intelligent plane, and MAMO plane. Based on the microservices concept, the distributed intelligent algorithm training and inference process of the system is decoupled into several network functions. These network functions can be reconstructed into different intelligent services, and Kubernetes is used to instantiate the system's network function templates, thereby achieving efficient deployment based on microservices and real-time monitoring of the entire lifecycle of network functions. Therefore, this invention achieves efficient deployment by combining cloud computing, edge computing, and 5G networks and introducing microservices into the cloud-network-edge intelligent collaborative architecture.

[0104] Please see Figure 8 This invention provides an electronic device, including a processor 801, a communication interface 802, a memory 803, and a communication bus 804, wherein the processor 801, the communication interface 802, and the memory 803 communicate with each other through the communication bus 804.

[0105] Memory 803 is used to store computer programs;

[0106] When the processor 801 executes the program stored in the memory 803, it implements the steps of the above-mentioned cloud-network-edge intelligent collaboration and management orchestration method based on microservices.

[0107] This invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of the above-described microservice-based cloud-network-edge intelligent collaboration and management orchestration method.

[0108] For the electronic device / storage medium embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be referred to in the description of the method embodiment.

[0109] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0110] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the specification and accompanying drawings, will understand and implement other variations of the disclosed embodiments in carrying out the claimed invention. In the specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. While certain measures are described in different embodiments, this does not mean that these measures cannot be combined to produce good results.

[0111] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A cloud-network-edge intelligent collaboration and management orchestration method based on microservices, characterized in that, This system is applied to a cloud-network-edge intelligent collaborative framework system, which includes: The system comprises three domains: cloud, network, and edge. Each domain has a four-layer, three-sided architecture. The four-layer architecture includes, from bottom to top, an infrastructure layer, a virtualization layer, a functional layer, and an application layer. The three-sided architecture includes a control plane, an intelligence plane, and a management orchestration plane. The infrastructure layer, virtualization layer, functional layer, and application layer interact with the control plane, intelligence plane, and management orchestration plane, respectively. The control plane, intelligence plane, and management orchestration plane interact with each other, and the corresponding methods include: The system implements the distributed intelligent algorithm training and inference process based on microservices; in the training and inference process, the training and inference process of the distributed intelligent algorithm is decoupled into several network functions by leveraging the microservice concept. The instantiation process of network function templates for the Kubernetes-based system includes: After receiving a distributed intelligent algorithm service request from the application layer, the management orchestration plane sends a model request to the model selector; the model selector selects a suitable neural network model based on the model request and requests a predefined neural network model template from the intelligent algorithm model library; after receiving the template request information from the model selector, the intelligent algorithm model library updates the predefined parameters sent by the model selector to the corresponding neural network model template and returns the updated neural network model template to the management orchestration plane; the management orchestration plane receives the neural network model template from the intelligent algorithm model library, sends the neural network model template to the API Server, and requests the creation of a corresponding Pod instance; after receiving the Pod creation request information from the management orchestration plane, the API Server writes the neural network model template received by the API Server into Etcd; after receiving the neural network model template information sent by the API Server, Etcd saves it to the database and returns updated database information to the API Server; the API... The server sends a Pod creation request to the scheduler. The scheduler receives this request via a watch mechanism, selects a suitable Node using its internal scheduling mechanism, and returns the Node to the API server. Upon receiving the Node from the scheduler, the API server writes the corresponding information to the Etcd and sends a Pod binding request to the Node's Kubelet. Upon receiving the Pod binding request from the API server, the Kubelet instantiates the neural network model template in the Etcd into a Docker container using container runtime, and allocates corresponding resources to the Docker container based on the CPU, Memory, and network resource information in the neural network model template. This completes the instantiation of the system's network function template and responds to the distributed intelligent algorithm service request sent by the application layer.

2. The cloud-network-edge intelligent collaboration and management orchestration method based on microservices according to claim 1, characterized in that, The functional layer of the intelligent surface utilizes the microservice concept to realize the distributed intelligent algorithm training and inference process of the system.

3. The cloud-network-edge intelligent collaboration and management orchestration method based on microservices according to claim 1, characterized in that, The network functions include five functions: data collection, data preprocessing, modeling, distribution, and aggregation.

4. The cloud-network-edge intelligent collaboration and management orchestration method based on microservices according to claim 3, characterized in that, By using NFV technology, all network functions can be virtualized.

5. The cloud-network-edge intelligent collaboration and management orchestration method based on microservices according to claim 4, characterized in that, All of the aforementioned network functions are mounted on a unified virtual SBI bus and utilize the HTTP protocol to obtain and provide the corresponding services.

6. The cloud-network-edge intelligent collaboration and management orchestration method based on microservices according to claim 5, characterized in that, The functional layer of the management orchestration plane uses Kubernetes to implement the instantiation process of the system's network function templates.

7. The cloud-network-edge intelligent collaboration and management orchestration method based on microservices according to claim 6, characterized in that, We can use the client-java and client-go tools provided by Kubernetes for secondary development, and interact with the API Server to realize the template instantiation of various network functions and the management and orchestration of resources.

8. The cloud-network-edge intelligent collaboration and management orchestration method based on microservices according to claim 7, characterized in that, The distributed intelligent algorithm training process based on the microservices implementation system includes: Users access the core network through 5G base stations and connect to the cloud-network-edge intelligent collaborative framework system via wireless networks, and send requests to the application layer of the control plane. After receiving the request, the application layer of the control plane obtains the service of the functional layer of the intelligent plane through the RESTful interface and sends the service request. The functional layer of the intelligent surface receives the corresponding service request, parses the URL, and sends the network application request to the management orchestration surface via the HTTP protocol; After the management orchestration plane listens to the network application request sent by the functional layer of the smart plane, it obtains the network function of the smart plane that needs to be instantiated by parsing the URL, obtains the template of the network function of the smart plane from the template library, instantiates the template with the help of Kubernetes, deploys the corresponding network function, and sends a request to the distribution network function of the cloud domain. After receiving the request, the distribution network function selects suitable edge nodes to participate in this round of distributed intelligent algorithm training through its internal node selection function, obtains the neural network model from the model network function, and distributes the obtained neural network model to the selected edge nodes. After receiving the neural network model sent by the distribution network function, the data collection network function of the selected edge node begins to collect the corresponding data and puts it into the database; The data preprocessing network function reads the data put into the database by the data collection network function, performs data preprocessing by truncation, zero padding, normalization and labeling, puts the processed data into the database, and sends a training request to the model network function. After receiving the training request from the data preprocessing network function, the model network function performs a round of model training using the neural network model sent by the distribution network function and the data processed by the data preprocessing network function, and sends the trained neural network model to the aggregation network function of the cloud domain via HTTP protocol. The cloud domain's aggregation network function receives the model parameters from the edge nodes, performs federated aggregation to obtain the aggregated model, and sends the aggregated model to the cloud domain's distribution network function via HTTP protocol. The process of the cloud domain's distribution network function receiving the request is repeated until the neural network model reaches the desired accuracy.

9. The cloud-network-edge intelligent collaboration and management orchestration method based on microservices according to claim 7, characterized in that, The distributed intelligent algorithm reasoning process of the microservice-based system includes: Users access the core network through 5G base stations and connect to the cloud-network-edge intelligent collaborative framework system via wireless networks, and send requests to the application layer of the control plane. After receiving the request, the application layer of the control plane obtains the service of the functional layer of the intelligent plane through the RESTful interface and sends the service request. The functional layer of the intelligent surface receives the corresponding service request, parses the URL, and sends the network application request to the management orchestration surface via the HTTP protocol; After the management orchestration plane listens to the network application request sent by the functional layer of the smart plane, it obtains the network function of the smart plane to be instantiated by parsing the URL, obtains the template of the network function of the smart plane from the template library, instantiates the template with the help of Kubernetes, deploys the corresponding network function, and sends a request to the data collection network function of the edge domain. After receiving the request sent by the management orchestration plane, the data collection network function begins to collect the corresponding data and puts it into the database; The data preprocessing network function reads the data put into the database by the data collection network function, performs data preprocessing by truncation, zero padding, and normalization, and requests the calculation results from the model network function via the HTTP protocol. After receiving the request from the data preprocessing network function, the model network function parses the URL, selects the neural network model chosen in the request, and inputs the data processed by the data preprocessing network function into the neural network model to obtain the final result, which is then returned to the user.

Images