Node-type edge computing system

Abstract

Provided in the present invention is a node-type edge computing system. The system comprises a micro-service module, a CI / CD module, an application management module, an observation module and a security module. The node-type edge computing module in the present invention has the characteristics of having a low network delay, supporting mass data access, having an elastic infrastructure, etc. Moreover, in addition to reducing the transmission delay, the advantages brought about by shortening a spatial distance further comprise the delay of various PLCs, gateways, routing-transfer, and network device processing in a complex network is also reduced. In addition, since the probability of network link contention is greatly reduced, the overall delay can be significantly reduced. By means of the node-type edge computing module, a distributed capability is added to a traditional cloud center, some service logic is deployed on an edge side, and related data processing is completed, such that the pressure of transmitting data back to a center cloud can be greatly relieved. The node-type edge computing module can also provide an elastic virtualization capability, such as calculation, networking and storage, based on an edge location, and can truly realize "cloud-edge collaboration".

Description

A node-type edge computing system Technical Field

[0001] The present invention relates to the field of cloud computing technology, and in particular to a node-type edge computing system. Background Art

[0002] Cloud computing is currently entering its next technological phase, extending its capabilities to the edge, closer to manufacturing equipment. Through unified management and control across the cloud, edge, and end, cloud computing services are decentralized, providing end-to-end cloud services. This has led to the emergence of the concept of edge cloud computing. This cloud computing platform, built on top of fundamental production equipment, leverages the core of cloud computing technology and the power of edge computing. This platform provides a comprehensive elastic cloud platform with computing, networking, storage, and security capabilities at the edge of production equipment. Together with the central cloud and IoT devices, it forms an end-to-end technical architecture characterized by "cloud-edge-end synergy." Node-based edge computing modules offload network forwarding, storage, computing, and intelligent data analysis to the edge of the device, reducing response latency, alleviating pressure on the cloud, and lowering bandwidth costs. They also provide cloud services such as network-wide scheduling and computing power distribution. The central cloud and node-based edge computing modules work together to achieve center-edge collaboration, network-wide computing power scheduling, and unified management and control, truly realizing a "ubiquitous cloud." Edge cloud computing can also use a distributed architecture to deploy node-based edge computing modules on devices closest to the terminal, providing targeted computing power for terminal-side data sources. This computing power can terminate some data processing on the edge of the device, while other data can be processed and then transmitted back to the central cloud or network-side edge computing modules. At the same time, edge cloud computing can provide a new type of elastic computing power resource for terminal infrastructure through node-based edge computing modules, and then cooperate and coordinate with the central cloud to provide cloud computing services that meet the technical needs of terminal facilities.

[0003] The current concept of cloud computing is based on centralized resource management and control. Even when multiple data centers are interconnected, all software and hardware resources are still managed, scheduled, and sold as a unified resource. Therefore, for production line equipment, cloud computing's real-time operability and the ability to fine-tune production through segmented equipment fall far short of meeting the requirements of manufacturing companies.

[0004] Simply relying on technological advances in the physical and transport layers of wireless and fixed networks will not meet these requirements. Edge computing can help address these issues. Combining edge computing with 5G will effectively reduce costs, and enabling more applications to run at the edge will also reduce workloads on the cloud.

[0005] Node-based edge computing modules are suitable for localized, real-time, and short-cycle data processing and analysis, and feature lightweight, small-scale deep reinforcement learning capabilities. They break the conventional "automation + industrial software" paradigm in large-scale production processes, enabling cross-industry integration between IT and OT. This avoids the waste caused by overlapping functions across various industrial management software and eliminates issues such as poor communication and coordination between software and industrial equipment.

[0006] The lack of privacy that cloud computing brings to the world in industrial production poses high security risks and is prone to technological spillover.

[0007] Currently, industrial automation (OT people) in the manufacturing industry focuses on control, has weak intelligence, has many breakpoints in human-computer interaction, has a heavy workload and lacks control accuracy.

[0008] Summary of the Invention

[0009] The purpose of the present invention is to propose a node-type edge computing system with full coverage, security, stability and high operating accuracy.

[0010] To achieve the above objectives, the present invention proposes a node-type edge computing system, including a microservice module, a CI / CD module, an application management module, an observation module and a security module;

[0011] The microservice modules include service governance, grayscale release, circuit breaking, current limiting, and intelligent routing;

[0012] The CI / CD module includes image repository management, image migration, security scanning and pipeline model;

[0013] The application management module includes application publishing, application deployment, version control, application repository management and application store;

[0014] The observation module includes: cluster and application monitoring, cluster and application log management, alarm notification, audit log, time query and metering and billing;

[0015] [Corrected 20.08.2024 according to Rule 91] The security module includes user management, role management, login authentication, key management and network policy management.

[0016] Furthermore, the application store supports edge device support by downloading applications and installing components, and installing edge node management components;

[0017] The edge node uses an ARM architecture main chip, and the edge node management component has the following functions:

[0018] Support edge-to-edge cross-network communication;

[0019] Support edge-center cloud cross-network communication;

[0020] Supports stable simultaneous operation of multiple Cloud Core and Edge Core instances;

[0021] The edge supports IOT gateway applications;

[0022] Support cloud-edge collaborative AI framework;

[0023] Support edge-cloud collaborative training and inference;

[0024] Use a standard service mesh for service governance control;

[0025] Supports cloud-edge collaborative monitoring.

[0026] [Corrected 20.08.2024 in accordance with Rule 91] Furthermore, application management components and automatic orchestration algorithms are used to distribute and deploy componentized or microservice-based applications to corresponding node-type edge computing modules through the application publishing and application deployment functions and the underlying edge node management components.

[0027] [Corrected 20.08.2024 according to Rule 91] Further, the principle of the automatic orchestration algorithm is as follows: let W(1, 2, …i, …n) be the initial weight array, abbreviated as W, W(i) be the initial weight of the microservice numbered i, for the selected microservice, its initial weight is 1, the required microservice also has an initial weight of 1, and the remaining microservices have an initial weight of 0; D(1, 2, …i, …n) be the final weight array, abbreviated as D, D(i) is the final weight of the microservice numbered i;

[0028] Let R(i) be the numbered list of microservices associated with the i-th microservice, abbreviated as R; G(1, 2, ...i, ...n) be the processing flag array, abbreviated as G, which is initially all 0;

[0029] The automatic arrangement algorithm is to realize the mapping relationship of function F(W, G, R) = D;

[0030] Loop through the numbers in W with a weight of 1. For each number, recursively loop through the numbers in R. Assuming the current number is j, first check whether G(j) is 1. If it is 1, do nothing and directly process the next number. Otherwise, set D(j) to 1 and also set G(j) corresponding to the processed number from 0 to 1.

[0031] After all the numbers with a weight of 1 in W have been processed, the numbers with a weight of 1 in G are the set of microservice numbers that need to be automatically orchestrated.

[0032] Furthermore, in order to collect, store, and forward field device data and perform data processing or machine learning on the edge, an IoT gateway application is deployed on the node-type edge computing module, running on the ARM architecture-based Linux operating system. Its functions are as follows:

[0033] Data acquisition: supports multiple acquisition protocols, such as MODBUS RTU, MODBUS TCP, OPC UA, BACNet IP, TCP / IP, MQTT, WEBSOCKET and HTTP;

[0034] Data storage: In addition to being stored in memory, the collected data must also be persisted in a database. It must support cache databases, relational databases, and time-series databases, and have data processing and management capabilities.

[0035] Data forwarding: supports multiple forwarding protocols, such as OPC UA, TCP / IP, MQTT, WEBSOCKET and HTTP;

[0036] It has programmable logic control (PLC) function and provides corresponding programming environment, or can communicate with third-party PLC to control related equipment through third-party PLC.

[0037] Furthermore, the edge node management component has a certain computing power of 2.3TOPS and above, supports the cloud-edge collaborative AI framework, receives data collected by the IOT gateway application on the node-type edge computing module, uses edge-cloud collaborative training and inference, and performs distributed heterogeneous parallel training to fully utilize the computing power of the edge module.

[0038] Compared with the prior art, the advantages of the present invention are:

[0039] 1. Full coverage: The distributed architecture of the edge cloud can cover various production scenarios through industrial node-type edge computing modules, thereby achieving one-stop service and agile delivery capabilities.

[0040] 2. Elastic Scaling: Purchase on demand, eliminating the transitional payment for some overlapping functions of various industrial software, realizing the elastic scaling needs of the business, and saving the supply chain management, construction and capital investment costs required for self-built.

[0041] 3. Open and flexible: Provides a "standard open" edge computing platform that can be easily connected to the central cloud system and flexibly deploy various applications according to business needs.

[0042] 4. Security and stability: The secure and stable edge computing ecosystem built using the internal production system as the core technology can be directly disconnected from the external network without affecting the current production rhythm.

[0043] 5. Reduce latency: Whether wired or through industrial 5G, all applications of node-type edge computing modules can provide uninterrupted terminal access latency of less than 5ms.

[0044] 6. Business localization: With the adoption of the cloud-edge-end three-body collaborative architecture, a large amount of processing and response occurs locally, and the frequency of terminal-to-cloud access will be reduced by more than 80%, thereby reducing costs and increasing efficiency while ensuring privacy.

[0045] 7. Reduce costs: After introducing node-type edge computing modules, computing, storage, network and other costs can be saved by more than 30%.

[0046] 4. Agile delivery: By adopting node-type edge computing module services, you can achieve "agile delivery in seconds".

[0047] 8. High security: It has high security capabilities integrated with traditional cloud services, including DDoS cleaning and black hole protection capabilities, multi-tenant isolation, automatic detection and cleaning of abnormal traffic, and center-edge security management and control channels.

[0048] 9. Open and easy to use: including an open operating environment, flexible deployment of various cloud services and applications, online remote management, and visual monitoring of operating indicators.

[0049] 10. For OEM customers, single-machine equipment operation does not require external communication through any network, but directly connects internally and intelligently determines the production mechanism.

[0050] 11. Because the node-type edge computing module has the ability of deep reinforcement learning, it not only reduces the pressure on the central cloud or industrial main server group, but also improves the real-time operation and product quality accuracy.

[0051] 12. The PLC and intelligent gateway functions built into the core of the node-type edge computing module make communication methods safer, faster, and more reliable, while also making control more convenient. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] FIG1 is a structural diagram of a node-type edge computing system according to an embodiment of the present invention;

[0053] FIG2 is a schematic diagram of a production process management (PPM) subsystem in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be further described below.

[0055] The present invention proposes a node-type edge computing system, as shown in Figure 1, which includes a microservice module, a CI / CD module, an application management module, an observation module, and a security module;

[0056] The microservice module includes service governance, grayscale release, circuit breaking, current limiting, and intelligent routing;

[0057] The CI / CD module includes image repository management, image migration, security scanning, and pipeline model;

[0058] The application management module includes application publishing, application deployment, version control, application repository management and application store;

[0059] The observation module includes: cluster and application monitoring, cluster and application log management, alarm notification, audit log, time query and metering and billing;

[0060] [Corrected 20.08.2024 according to Rule 91] The security module includes user management, role management, login authentication, key management and network policy management.

[0061] It is a distributed operating platform for cloud-native applications that supports multi-cloud and multi-cluster management, provides full-stack IT automation and operation capabilities, and simplifies DevOps workflows. It can easily integrate third-party applications with the platform through plug-and-play. As a full-stack container platform, it provides an operation and maintenance-friendly operation interface to help users quickly build a powerful and feature-rich container cloud, providing users with the many functions they need: such as multi-cloud and multi-cluster management, Kubernetes resource management, DevOps, application lifecycle management, microservices governance, log query and collection, services and networks, multi-tenant management, monitoring and alarm, event and audit query, storage management, access control, edge node management, network policies, image repository management, and security management.

[0062] The app store supports a variety of applications and components. Install the edge node management component to support edge devices. Edge nodes usually use ARM architecture chips. The edge node management component should have the following basic functions:

[0063] ●Support edge-to-edge cross-network communication

[0064] ●Support edge-center cloud cross-network communication

[0065] ●Supports multiple Cloud Core and Edge Core instances to run stably at the same time

[0066] ●Edge supports IOT gateway applications

[0067] ●Support cloud-edge collaborative AI frameworks, such as KubeFlow / Pytorch / Mindspore, etc.

[0068] ●Support edge-cloud collaborative training and inference

[0069] ●Use standard service mesh for service governance control

[0070] ●Support cloud-edge collaborative monitoring

[0071] [Corrected 20.08.2024 in accordance with Rule 91] Using application management components and automatic orchestration algorithms, componentized or microservice-based applications are distributed and deployed to corresponding node-based edge computing modules through application publishing and deployment functions and underlying edge node management components. For example, the production process management (PPM) subsystem in the manufacturing execution system (MES), as shown in Figure 2, consists of multiple modules:

[0072] After being componentized or microserviced, the PPM subsystem can be deployed and managed at a finer granularity. For example, the process data acquisition module can be deployed to the corresponding node-type edge computing module, making it easier to control the beat and adjust the processing parameters.

[0073] The basic principles of the automatic arrangement algorithm are as follows:

[0074] Let W(1, 2, …i, …n) be the initial weight array, abbreviated as W, W(i) be the initial weight of microservice numbered i, for the selected microservice, its initial weight is 1, the required microservice also has an initial weight of 1, and the remaining microservices have an initial weight of 0; D(1, 2, …i, …n) be the final weight array, abbreviated as D, D(i) is the final weight of microservice numbered i.

[0075] Let R(i) be the numbered list of the associated microservices of the i-th microservice, abbreviated as R. G(1, 2, ...i, ...n) is the processing flag array, abbreviated as G, which is initially all 0.

[0076] The automatic arrangement algorithm is to realize the mapping relationship of the function F(W, G, R)=D.

[0077] Loop through the numbers with a weight of 1 in W. For each number, recursively loop through the numbers in R. Assuming the current number is j, first check whether G(j) is 1. If it is 1, do nothing and directly process the next number; otherwise, set D(j) to 1, and also set G(j) corresponding to the processed number from 0 to 1.

[0078] After all the numbers with a weight of 1 in W have been processed, the numbers with a weight of 1 in G are the set of microservice numbers that need to be automatically orchestrated.

[0079] To collect, store, and forward data from field devices and perform data processing or machine learning on the edge, IoT gateway applications can be deployed on node-type edge computing modules. These applications typically run on an ARM-based Linux operating system. Their main functions are:

[0080] ●Data acquisition: Supports multiple acquisition protocols, such as MODBUS RTU, MODBUS TCP, OPC UA, BACNet IP, TCP / IP, MQTT, WEBSOCKET, HTTP, etc.

[0081] Data storage: Collected data generally needs to be stored not only in memory but also in a persistent database. This database must support cache databases, relational databases, and time-series databases, and must have data processing and management capabilities.

[0082] ●Data forwarding: supports multiple forwarding protocols, such as OPC UA, TCP / IP, MQTT, WEBSOCKET, HTTP, etc.

[0083] ●It has programmable logic control (PLC) function and provides the corresponding programming environment, or can communicate with third-party PLC to control related equipment through third-party PLC.

[0084] The edge node management component has a certain computing power, usually 2.3TOPS or above, and supports the cloud-edge collaborative AI framework. It can receive data collected by IoT gateway applications on node-type edge computing modules, use edge-cloud collaborative training and inference, and perform distributed heterogeneous parallel training to fully utilize the computing power of edge modules:

[0085] ●Based on different parallel principles and modes, the mainstream parallel types in the industry are as follows:

[0086] ■Data Parallel: A parallel mode that splits data, generally by batch dimension, and distributes data to various computing units (workers) for model calculations.

[0087] ■ Model Parallel: A parallel mode that partitions a model. Model parallelism can be categorized into operator-level model parallelism, pipeline model parallelism, and optimizer model parallelism.

[0088] ■Hybrid Parallel: refers to a parallel mode that covers data parallelism and model parallelism.

[0089] ● Considering this edge application scenario, a data parallel model is suitable. Based on this model, we adopt the parameter server architecture, which is a widely used architecture in distributed training and has better flexibility, scalability and node disaster recovery capabilities. Specifically, the parameter server supports both synchronous SGD (Stochastic Gradient Descent) and asynchronous SGD training algorithms; in terms of scalability, the model calculation and model update are deployed in two types of processes, Worker (edge ​​computing module) and Server, respectively, so that the resources of Worker and Server can be independently scaled horizontally (adding or deleting Worker and Server resources); in addition, in the environment of large-scale data centers, various failures often occur in computing equipment, networks, and storage, causing some nodes to be abnormal. Under the architecture of the parameter server, such failures can be handled more easily without affecting the tasks in training.

[0090] The parameter server architecture consists of three independent components: Server, Worker, and Scheduler. Their functions are:

[0091] ■Server: Saves the model's weights and reverse-calculated gradient values, and updates the model using the gradient values ​​uploaded by the optimizer through the Worker.

[0092] ■Worker: Executes forward and reverse calculations of the network. The gradient value of the reverse calculation is uploaded to the server through the Push interface. The updated model of the server is downloaded to the local Worker through the Pull interface.

[0093] ■Scheduler: Used to establish communication between Server and Worker.

[0094] ●Basic steps:

[0095] ■First, enable parameter server training mode.

[0096] ■Then initialize distributed training, including networking of the three nodes: Server, Worker, and Scheduler, and initialize collective communication.

[0097] ■Control whether training parameters are updated through the parameter server by calling the interface, and control the parameter initialization location.

[0098] ■Based on the original training script, set all weights of the model to be trained by the parameter server.

[0099] ■For large data, you can configure the cache size of the operator and enable the cache function in the parameter server training mode. This function uses data tables of the same cache size for training on the device, and stores the full data on the server. The data tables used for training are swapped into the cache in advance, and expired data tables are returned to the server to achieve the purpose of improving training performance.

[0100] Use the shell scripts corresponding to the three roles of Worker, Server, and Scheduler to start training.

[0101] The above description is merely a preferred embodiment of the present invention and does not limit the present invention in any way. Any person skilled in the art who, without departing from the scope of the present invention, makes any equivalent substitution, modification, or other changes to the technical solution and technical content disclosed in the present invention shall be deemed to be within the scope of the present invention and still fall within the scope of protection of the present invention.

Claims

1. [Corrected 20.08.2024 in accordance with Rule 91] A node-based edge computing system, characterized in that: Includes microservice module, CI / CD module, application management module, observation module and security module; The microservice modules include service governance, grayscale release, circuit breaker, current limiting and intelligent routing; The CI / CD module includes image repository management, image migration, security scanning and pipeline model; The application management module includes application publishing, application deployment, version control, application repository management and application store; The observation module includes: cluster and application monitoring, cluster and application log management, alarm notification, audit log, event query and metering and billing; The security module includes user management, role management, login authentication, key management and network policy management.

2. The node-based edge computing system according to claim 1, characterized in that: The application store supports edge device support by downloading applications and installing components, installing edge node management components; The edge node uses the ARM architecture main chip, and the edge node management component has the following functions: Support edge-to-edge cross-network communication; Support edge-center cloud cross-network communication; Supports stable operation of multiple Cloud Core and Edge Core instances at the same time; The edge supports IOT gateway applications; Support cloud-edge collaborative AI framework; Support edge-cloud collaborative training and reasoning; Use a standard service mesh for service governance control; Supports cloud-edge collaborative monitoring.

3. [Corrected 20.08.2024 in accordance with Rule 91] The node-based edge computing system according to claim 2, characterized in that: Using application management components and automatic orchestration algorithms, componentized or microservice applications are distributed and deployed to corresponding node-type edge computing modules through the application publishing and application deployment functions and underlying edge node management components.

4. [Corrected 20.08.2024 in accordance with Rule 91] The node-based edge computing system according to claim 3, characterized in that: The principle of the automatic arrangement algorithm is as follows: let W(1, 2, ...i, ...n) be the initial weight array, abbreviated as W, W(i) is the initial weight of the microservice numbered i, for the selected microservice, its initial weight is 1, the required microservice has an initial weight of 1, and the rest of the microservices have an initial weight of 0; D(1, 2, ...i, ...n) is the final weight array, abbreviated as D, and D(i) is the final weight of the microservice numbered i; Let R(i) be the number list of the associated microservices of the i-th microservice, abbreviated as R; G(1, 2, ...i, ...n) be the processing flag array, abbreviated as G, which is initially all 0; The automatic arrangement algorithm is to realize the mapping relationship of function F(W, G, R) = D; Loop through the numbers with a weight of 1 in W. For each number, recursively loop through the numbers in R. Assuming the current number is j, first check whether G(j) is 1. If it is 1, do nothing and directly process the next number. Otherwise, set D(j) to 1, and also set G(j) corresponding to the processed number from 0 to 1. When all the numbers with weight 1 in W have been processed, the numbers with weight 1 in G are the set of microservice numbers that need to be automatically orchestrated.

5. The node-based edge computing system according to claim 4, characterized in that: In order to collect, store and forward data from field devices and perform data processing or machine learning on the edge side, an IoT gateway application is deployed on the node-type edge computing module, running on the Linux operating system based on the ARM architecture. Its functions are as follows: Data acquisition: supports multiple acquisition protocols, such as MODBUS RTU, MODBUS TCP, OPC UA, BACNet IP, TCP / IP, MQTT, WEBSOCKET and HTTP; Data storage: In addition to being stored in memory, the collected data also needs to be persisted in a database; it needs to support cache databases, relational databases, and time series databases, and have data processing and management capabilities; Data forwarding: supports multiple forwarding protocols, such as OPC UA, TCP / IP, MQTT, WEBSOCKET and HTTP; It has programmable logic control (PLC) function and provides corresponding programming environment, or can communicate with third-party PLC to control related equipment through third-party PLC.

6. The node-based edge computing system according to claim 2, characterized in that: The edge node management component has a certain computing power of 2.3TOPS and above, supports the cloud-edge collaborative AI framework, receives data collected by the IOT gateway application on the node-type edge computing module, uses edge-cloud collaborative training and inference, and performs distributed heterogeneous parallel training to make full use of the computing power of the edge module.

Images