Scheduling method and system for distributed heterogeneous computing power in the city brain AI computing center

Through fully symmetric distributed architecture and mirror management technology, a distributed computing power expansion and reduction architecture is built on demand, which solves the problems of low utilization rate of AI computing power and complex scheduling, and realizes flexible expansion and efficient urban brain AI computing center Resource management improves system stability and resource utilization efficiency.

CN118337781BActive Publication Date: 2024-11-05SHANDONG AITE YUNXIANG INFORMATION TECH CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410288050.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-14
Publication Date
2024-11-05
Estimated Expiration
2044-03-14

AI Technical Summary

Technical Problem

The existing AI computing power utilization rate is low, the hardware procurement cost and energy consumption are high, the computing resource scheduling threshold is high, the heterogeneous computing power system has complex interaction and poor adaptability, making it difficult to meet the real-time and flexibility of massive computing power requirements.

Method used

Adopt a fully symmetric distributed architecture to deploy and control the physical nodes and storage networks of the computing cluster, build a distributed computing power architecture that expands and reduces the computing power, quickly creates cloud hosts through mirror management, performs full life cycle management and heterogeneous computing resource scheduling, and predicts resource requirements. Carry out pre-scheduling and allocation to realize the management of massive data computing resources and the handling of urban governance emergency incidents in the urban brain AI computing center.

Benefits of technology

It realizes flexible online expansion of the urban brain AI computing center, avoids performance bottlenecks, ensures balanced distribution of stored data, improves system stability and reliability, improves resource utilization efficiency, ensures business continuity, and supports data center scenarios in multiple architectural forms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118337781B_ABST
    Figure CN118337781B_ABST
Patent Text Reader

Abstract

The present invention provides a method and system for scheduling distributed heterogeneous computing power of a city brain AI computing center, including: deploying and controlling computing cluster physical nodes and storage networks through a fully symmetrical distributed architecture; organizing multiple computing cluster physical nodes into a decentralized cluster that can scale computing power on demand, and building a distributed on-demand computing power architecture; managing cloud host computing resources throughout their life cycle, and visually controlling heterogeneous computing resource management and scheduling through a unified self-resetting scheduling interface control; building a city brain AI computing center based on a distributed on-demand computing power architecture and heterogeneous computing resource management and scheduling, and managing massive urban data computing power resources and handling emergency events for urban governance; directly managing cloud platform physical servers on bare metal; predicting future resource requirements and performing pre-scheduling and allocation in advance; and performing seamless migration management of city brain AI computing resources and elastic scaling of cluster physical node computing resources in distributed heterogeneous computing power scheduling.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of AI computing power scheduling, and more specifically, to a method and system for scheduling distributed heterogeneous computing power in a city brain AI computing center. Background Art

[0002] At this stage, the development of the AI ​​basic layer has provided the core driving force for the development of AI computing power and chips. The market size of AI chips has shown a high growth trend with the demand for cloud training. The scale of the artificial intelligence basic layer technology industry has developed rapidly and has broad space, and it has generally shown a sustained growth trend. Ultra-large model business scenarios such as AIGC, including ChatGPT, StableDiffusion, etc., have driven massive computing power demand. However, the current computing power demand is strong but the utilization rate is low, including: the real-time computing power demand in various industries has increased, but the business and computing power adaptation is difficult; computing power resources are difficult to deploy and schedule; hardware procurement costs and energy consumption costs are high, and the delivery cycle is uncontrollable, and the threshold for computing resource scheduling is high; computing power cluster operations are complex, and various heterogeneous computing power systems have complex interactivity and poor adaptability. These problems have yet to be solved. Therefore, it is necessary to propose a scheduling method and system for distributed heterogeneous computing power in the city brain AI computing center to at least partially solve the problems existing in the existing technology. Summary of the invention

[0003] A series of simplified concepts are introduced in the content of the invention, which will be further described in detail in the specific implementation method section; the content of the invention of the present invention does not mean to attempt to limit the key features and essential technical features of the technical solution claimed for protection, nor does it mean to attempt to determine the scope of protection of the technical solution claimed for protection.

[0004] In order to at least partially solve the above problems, the present invention provides a method for scheduling distributed heterogeneous computing power of a city brain AI computing center, comprising:

[0005] S100 deploys and controls computing cluster physical nodes and storage networks through a fully symmetric distributed architecture; multiple computing cluster physical nodes are combined into a decentralized cluster that can scale computing power on demand, building a distributed computing power architecture that can scale computing power on demand;

[0006] S200, quickly creates cloud hosts through image management and defines cloud host operating systems; manages cloud host computing resources throughout their life cycle, and performs visual control of heterogeneous computing resource management and scheduling through a unified self-reset scheduling interface control;

[0007] S300, based on the distributed on-demand computing architecture and heterogeneous computing resource management and scheduling, builds the city brain AI computing center to manage massive urban data computing resources and handle urban governance emergencies;

[0008] S400 remotely connects to the cloud host of the City Brain AI computing center to directly manage the bare metal physical servers of the cloud platform; predicts future resource requirements and pre-scheduling and allocation in advance; and manages the seamless migration of City Brain AI computing resources and the elastic scaling of cluster physical node computing resources and distributed heterogeneous computing power scheduling.

[0009] Preferably, S100 includes:

[0010] S101, deploys and controls the computing cluster physical nodes and storage network through a fully symmetric distributed architecture;

[0011] S102, multiple computing cluster physical nodes are combined into a decentralized cluster that can expand and reduce computing power on demand;

[0012] S103, building a distributed on-demand computing power architecture based on the on-demand computing power decentralized cluster;

[0013] Deployment and control of computing cluster physical nodes and storage networks include: computing cluster physical nodes use GPUs and multiple chip models to perform active discovery and scheduling of GPUs / FPGAs / XPUs and computing power job scheduling and deployment, and build an AI elastic computing power resource pool storage network;

[0014] Based on the decentralized cluster of computing power that can be expanded or reduced on demand, a distributed computing power architecture that can be expanded or reduced on demand is built, including: providing underlying computing, storage and network resources through each node to form a software-defined infrastructure resource pool; all controlled system services are deployed to each node in a distributed and microservice architecture, and each node provides cloud service capabilities to the outside world; building a distributed computing power architecture that can be expanded or reduced on demand with redundancy, service self-healing and scalability.

[0015] Preferably, S200 includes:

[0016] S201, based on the distributed on-demand computing architecture, through image management, supports the coexistence of traditional centralized architecture and distributed architecture, quickly creates a cloud host, and obtains the image cloud host;

[0017] S202, defining the cloud host operating system through the ISO image file according to the image cloud host;

[0018] S203, according to the cloud host operating system, the cloud host computing resources are managed throughout their life cycle, and the heterogeneous computing resource management and scheduling are visually controlled through a unified self-resetting scheduling interface control.

[0019] Preferably, S300 includes:

[0020] S301: Build the AI ​​computing center of the city brain based on the distributed on-demand computing architecture and heterogeneous computing resource management and scheduling;

[0021] S302: Manage massive urban data computing resources based on the city brain AI computing center;

[0022] S303, handling urban governance emergencies based on massive urban data computing resource management;

[0023] Building the City Brain AI Computing Center includes: using the City Brain AI algorithm model to perform City Brain intelligent image recognition and build a cloud-edge-end city AI retina platform.

[0024] Preferably, S400 includes:

[0025] S401, remotely connect to the cloud host of the City Brain AI Computing Center and adjust the cloud host configuration by adjusting the CPU, memory, adding virtual disks, adding virtual network cards and the management interface console online;

[0026] S402: Directly manage bare metal physical servers on the cloud platform and perform heterogeneous resource status monitoring and management on physical servers in data centers with various architectures;

[0027] S403, predict future resource demands, perform pre-scheduling and allocation in advance, and monitor the real-time resource usage of the system; perform seamless migration management of the city brain AI computing resources and elastic scaling of cluster physical node computing resources and distributed heterogeneous computing power scheduling.

[0028] The present invention provides a scheduling system for distributed heterogeneous computing power of a city brain AI computing center, comprising:

[0029] The distributed on-demand computing power architecture subsystem deploys and controls the computing cluster physical nodes and storage network through a fully symmetrical distributed architecture; multiple computing cluster physical nodes are combined into an on-demand computing power decentralized cluster to build a distributed on-demand computing power architecture;

[0030] Heterogeneous computing resource management and control subsystem, quickly create cloud hosts through image management, define cloud host operating systems; manage cloud host computing resources throughout their life cycle, and perform visual control of heterogeneous computing resource management and scheduling through a unified self-reset scheduling interface control;

[0031] The computing power resource management and emergency response subsystem builds the city brain AI computing center based on the distributed on-demand computing power architecture and heterogeneous computing resource management and scheduling to manage massive urban data computing power resources and handle urban governance emergency events;

[0032] The cloud platform resource prediction and computing power scheduling subsystem directly manages the bare metal servers of the cloud platform by remotely connecting to the cloud host of the City Brain AI Computing Center; predicts future resource demands and performs pre-scheduling and allocation in advance; and manages the seamless migration of City Brain AI computing resources and the elastic scaling of distributed heterogeneous computing power scheduling of cluster physical node computing resources.

[0033] Preferably, the distributed on-demand computing power expansion and reduction architecture subsystem includes:

[0034] The computing cluster physical node deployment subsystem deploys and controls the computing cluster physical nodes and storage network through a fully symmetric distributed architecture;

[0035] The on-demand computing power cluster subsystem combines multiple computing cluster physical nodes into a decentralized cluster that can scale computing power on demand.

[0036] The distributed on-demand computing power architecture subsystem builds a distributed on-demand computing power architecture based on the decentralized cluster of on-demand computing power expansion and reduction.

[0037] Deployment and control of computing cluster physical nodes and storage networks include: computing cluster physical nodes use GPUs and multiple chip models to perform active discovery and scheduling of GPUs / FPGAs / XPUs and computing power job scheduling and deployment, and build an AI elastic computing power resource pool storage network;

[0038] Based on the decentralized cluster of computing power that can be expanded or reduced on demand, a distributed computing power architecture that can be expanded or reduced on demand is built, including: providing underlying computing, storage and network resources through each node to form a software-defined infrastructure resource pool; all controlled system services are deployed to each node in a distributed and microservice architecture, and each node provides cloud service capabilities to the outside world; building a distributed computing power architecture that can be expanded or reduced on demand with redundancy, service self-healing and scalability.

[0039] Preferably, the heterogeneous computing resource management and control subsystem includes:

[0040] The image management cloud host subsystem, based on the distributed on-demand computing architecture, supports the coexistence of traditional centralized architecture and distributed architecture through image management, and quickly creates cloud hosts and obtains image cloud hosts;

[0041] The cloud host operating subsystem defines the cloud host operating system through the ISO image file according to the image cloud host;

[0042] The lifecycle management control subsystem manages the computing resources of the cloud host throughout their lifecycle according to the cloud host operating system, and performs visual control of heterogeneous computing resource management and scheduling through a unified self-resetting scheduling interface.

[0043] Preferably, the computing power resource management emergency response subsystem includes:

[0044] The City Brain AI Computing Center construction subsystem is based on a distributed on-demand computing architecture and heterogeneous computing resource management and scheduling to build the City Brain AI Computing Center;

[0045] The city data computing power resource management subsystem manages massive city data computing power resources based on the city brain AI computing center;

[0046] The urban governance emergency response subsystem handles urban governance emergency events based on massive urban data computing resource management;

[0047] Building the City Brain AI Computing Center includes: using the City Brain AI algorithm model to perform City Brain intelligent image recognition and build a cloud-edge-end city AI retina platform.

[0048] Preferably, the cloud platform resource prediction computing power scheduling subsystem includes:

[0049] The cloud host configuration subsystem adjusts the CPU, memory, virtual disk, virtual network card and management interface console online, and remotely connects to the cloud host of the City Brain AI Computing Center to adjust the cloud host configuration;

[0050] The server direct management heterogeneous resource monitoring subsystem directly manages the bare metal physical servers of the cloud platform and monitors and manages the heterogeneous resource status of physical servers in data centers with various architectures;

[0051] The resource prediction and computing power pre-scheduling subsystem predicts future resource demands, performs pre-scheduling allocations in advance, and monitors the real-time resource usage of the system. It also manages the seamless migration of the city brain AI computing resources and the elastic scaling of cluster physical node computing resources and distributed heterogeneous computing power scheduling.

[0052] Compared with the prior art, the present invention has at least the following beneficial effects:

[0053] The present invention provides a method and system for scheduling distributed heterogeneous computing power in the city brain AI computing center. Through a fully symmetrical distributed architecture, physical nodes and storage networks of the control computing cluster are deployed; multiple physical nodes of the computing cluster are organized into a decentralized cluster of computing power that can be expanded or reduced on demand, and a distributed computing power architecture that can be expanded or reduced on demand is constructed; cloud hosts are quickly created through image management, and cloud host operating systems are defined; full life cycle management of cloud host computing resources is performed, and heterogeneous computing resource management and scheduling are visually controlled through a unified self-resetting scheduling interface control; based on the distributed computing power architecture that can be expanded or reduced on demand and heterogeneous computing resource management and scheduling, a city brain AI computing center is constructed to manage massive urban data computing power resources and handle emergency events in urban governance; by remotely connecting to the city brain The AI ​​computing center cloud host directly manages the physical servers of the cloud platform; predicts future resource requirements and pre-schedules and allocates them in advance; performs seamless migration management of the city brain AI computing resources and elastically scalable distributed heterogeneous computing power scheduling of cluster physical node computing resources; the cluster node scale of the solution of the present invention can be quickly and flexibly expanded online as needed with business development to avoid performance bottlenecks; the storage capacity and performance of the distributed cluster can be linearly expanded with the increase in the number of nodes, and can ensure that the stored data is evenly distributed to all nodes; excellent stability and reliability, avoiding single point failures, improving system robustness, and ensuring business continuity; high operability, and reasonable organization and arrangement of functional modules; the platform adopts a fully symmetrical distributed architecture to deploy control computing Cluster physical nodes and storage network cloud hosts, no single point of failure risk, adding or reducing nodes is more flexible; with high availability of cluster physical nodes, cloud hosts, systems and data, any node has a problem, the cloud host on the corresponding node can be directly migrated to other normal nodes, the migration period does not affect the upper business system data reading and writing, parallel platform management interface use and computing resource scheduling management, providing reliable security for business systems; supports the creation, modification, start, pause, restore, sleep, restart, shutdown, clone, migration and snapshot of cloud hosts, supports the creation of cloud hosts by image and iso customized cloud host operating system; supports the adjustment of cloud host configuration, including: online adjustment of CPU, memory, adding virtual Virtual disks and adding virtual network cards; support remote connection to cloud hosts through the management interface console; the cloud platform can not only manage other virtualization platforms in a unified manner, but also support direct management of physical servers, and can realize resource status and scheduling management functions of physical servers, which is suitable for data center scenarios with various architecture forms; when the cloud hosts corresponding to the resource pool are unevenly distributed on the physical host, it may cause the resources of the physical nodes of the cloud host where the main application is located to become exhausted, thereby affecting the performance of the entire business system; the resource balancing algorithm used by the dynamic resource scheduling function can realize automatic cloud host resource allocation and load balancing functions based on the resource usage of each physical node, improve resource utilization efficiency, and ensure the normal operation of key businesses;According to the user's business needs, by setting scaling rules to automatically increase / reduce business system computing resources, seamlessly increase the elasticity of the business system, and significantly improve the efficiency of computing resource utilization. ;

[0054] The scheduling method and system for distributed heterogeneous computing power of the urban brain AI computing center described in the present invention, and other advantages, objectives and features of the present invention will be reflected in part through the following description, and in part will also be understood by technical personnel in the field through research and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0056] Figure 1 This is a diagram of an embodiment of the scheduling system for distributed heterogeneous computing power of the city brain AI computing center described in the present invention.

[0057] Figure 2 This is another embodiment diagram of the scheduling system for distributed heterogeneous computing power of the city brain AI computing center described in the present invention.

[0058] Figure 3 This is a diagram of an implementation example of resource balancing for the scheduling system and method for distributed heterogeneous computing power of the city brain AI computing center described in the present invention. DETAILED DESCRIPTION

[0059] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments so that those skilled in the art can implement the invention according to the instructions. Figure 1-Figure 3 As shown, the present invention provides a method for scheduling distributed heterogeneous computing power of a city brain AI computing center, including:

[0060] S100 deploys and controls computing cluster physical nodes and storage networks through a fully symmetric distributed architecture; multiple computing cluster physical nodes are combined into a decentralized cluster that can scale computing power on demand, building a distributed computing power architecture that can scale computing power on demand;

[0061] S200, quickly creates cloud hosts through image management and defines cloud host operating systems; manages cloud host computing resources throughout their life cycle, and performs visual control of heterogeneous computing resource management and scheduling through a unified self-reset scheduling interface control;

[0062] S300, based on the distributed on-demand computing architecture and heterogeneous computing resource management and scheduling, builds the city brain AI computing center to manage massive urban data computing resources and handle urban governance emergencies;

[0063] S400 remotely connects to the cloud host of the City Brain AI computing center to directly manage the bare metal physical servers of the cloud platform; predicts future resource requirements and pre-scheduling and allocation in advance; and manages the seamless migration of City Brain AI computing resources and the elastic scaling of cluster physical node computing resources and distributed heterogeneous computing power scheduling.

[0064] The principle and effect of the above technical solution are as follows: The present invention provides a method for scheduling distributed heterogeneous computing power in an urban brain AI computing center, including: deploying and controlling computing cluster physical nodes and storage networks through a fully symmetrical distributed architecture; organizing multiple computing cluster physical nodes into a decentralized cluster that can scale computing power on demand, and building a distributed on-demand computing power architecture; quickly creating a cloud host through image management and defining the cloud host operating system; managing the computing resources of the cloud host throughout its life cycle, and visually controlling the management and scheduling of heterogeneous computing resources through a unified self-resetting scheduling interface; building an urban brain AI computing center based on a distributed on-demand computing power architecture and heterogeneous computing resource management and scheduling to conduct massive Urban data computing resource management and urban governance emergency event handling; directly manage the bare metal of the cloud platform physical server by remotely connecting to the cloud host of the city brain AI computing center; predict future resource needs and pre-schedule and allocate in advance; perform seamless migration management of the city brain AI computing resources and elastic scaling of cluster physical node computing resources and distributed heterogeneous computing power scheduling; the cluster node scale of the solution of the present invention can be quickly and flexibly expanded online as needed with business development to avoid performance bottlenecks; the storage capacity and performance of the distributed cluster can be linearly expanded with the increase in the number of nodes, and can ensure that the stored data is evenly distributed to all nodes; excellent stability and reliability, avoid single point failures, and improve system robustness, Ensure business continuity; high operability, and reasonable organization and arrangement of functional modules; the platform adopts a fully symmetrical distributed architecture, deploys and controls the physical nodes of the computing cluster and the storage network cloud host, without the risk of single point failure, and is more flexible in adding and reducing nodes; it has the characteristics of high availability of cluster physical nodes, cloud hosts, systems and data. If any node has a problem, the cloud host on the corresponding node can be directly migrated to other normal nodes. During the migration, the data reading and writing of the upper-level business system will not be affected. The parallel platform management interface and computing resource scheduling management provide reliable security for the business system; it supports the creation, modification, start, pause, recovery, sleep, restart, shutdown, and cloning of cloud hosts. , migration and snapshot, support for creating cloud hosts by image and customizing cloud host operating systems by iso; support for adjusting cloud host configuration, including: online adjustment of CPU, memory, adding virtual disks and adding virtual network cards; support for remote connection to cloud hosts through the management interface console; the cloud platform can not only manage other virtualization platforms in a unified manner, but also support direct management of physical servers, and can realize resource status and scheduling management of physical servers, which is suitable for data center scenarios with various architecture forms; when the cloud hosts corresponding to the resource pool are unevenly distributed on the physical hosts, it may cause the resources of the physical nodes of the cloud hosts where the main applications are located to become exhausted, thereby affecting the performance of the entire business system;The resource balancing algorithm used by the dynamic resource scheduling function can realize automatic cloud host resource allocation and load balancing functions based on the resource usage of each physical node, improve resource utilization efficiency, and ensure the normal operation of key businesses; according to the user's business needs, the business system computing resources can be automatically increased / reduced by setting scaling rules, seamlessly increasing the elasticity of the business system, and significantly improving the efficiency of computing resource utilization. ;

[0065] In one embodiment, S100 includes:

[0066] S101, deploys and controls the computing cluster physical nodes and storage network through a fully symmetric distributed architecture;

[0067] S102, multiple computing cluster physical nodes are combined into a decentralized cluster that can expand and reduce computing power on demand;

[0068] S103, building a distributed on-demand computing power architecture based on the on-demand computing power decentralized cluster;

[0069] Deployment and control of computing cluster physical nodes and storage networks include: computing cluster physical nodes use GPUs and multiple chip models to perform active discovery and scheduling of GPUs / FPGAs / XPUs and computing power job scheduling and deployment, and build an AI elastic computing power resource pool storage network;

[0070] Based on the decentralized cluster of computing power that can be expanded or reduced on demand, a distributed computing power architecture that can be expanded or reduced on demand is built, including: providing underlying computing, storage and network resources through each node to form a software-defined infrastructure resource pool; all controlled system services are deployed to each node in a distributed and microservice architecture, and each node provides cloud service capabilities to the outside world; building a distributed computing power architecture that can be expanded or reduced on demand with redundancy, service self-healing and scalability.

[0071] The principle and effect of the above technical solution are as follows: through a fully symmetric distributed architecture, the physical nodes and storage network of the control computing cluster are deployed; multiple physical nodes of the computing cluster are formed into a decentralized cluster for on-demand expansion and reduction of computing power; based on the decentralized cluster for on-demand expansion and reduction of computing power, a distributed on-demand expansion and reduction of computing power architecture is constructed; the physical nodes and storage network of the control computing cluster are deployed, including: the physical nodes of the computing cluster use GPUs and multiple models of chips to perform GPU / FPGA / XPU active discovery scheduling and computing power job scheduling and deployment, and build an AI elastic computing power resource pool storage network; based on the decentralized cluster for on-demand expansion and reduction of computing power, a distributed on-demand expansion and reduction of computing power architecture is constructed, including: each node provides underlying computing, storage and network resources to form a software-defined infrastructure resource pool; all system services controlled are deployed to each node in a distributed and microservice architecture, and each node provides cloud service capabilities to the outside world; a distributed on-demand expansion and reduction of computing power architecture with redundancy, service self-healing and scalability is constructed; redundancy, service self-healing and scalability are significantly improved, and the computing power architecture can be expanded and reduced on demand, greatly improving flexibility.

[0072] In one embodiment, S200 includes:

[0073] S201, based on the distributed on-demand computing architecture, through image management, supports the coexistence of traditional centralized architecture and distributed architecture, quickly creates a cloud host, and obtains the image cloud host;

[0074] S202, defining the cloud host operating system through the ISO image file according to the image cloud host;

[0075] S203, according to the cloud host operating system, the cloud host computing resources are managed throughout their life cycle, and the heterogeneous computing resource management and scheduling are visually controlled through a unified self-resetting scheduling interface control.

[0076] The principles and effects of the above technical solution are as follows: based on the distributed on-demand computing architecture, through image management, it supports the coexistence of traditional centralized architecture and distributed architecture, and quickly creates cloud hosts to obtain mirrored cloud hosts; based on the mirrored cloud hosts, the cloud host operating system is defined through ISO image files; based on the cloud host operating system, the cloud host computing resources are managed throughout their life cycle, and heterogeneous computing resource management and scheduling are visually controlled through a unified self-resetting scheduling interface control; the level of intelligent control is significantly improved.

[0077] In one embodiment, S300 includes:

[0078] S301: Build the AI ​​computing center of the city brain based on the distributed on-demand computing architecture and heterogeneous computing resource management and scheduling;

[0079] S302: Manage massive urban data computing resources based on the city brain AI computing center;

[0080] S303, handling urban governance emergencies based on massive urban data computing resource management;

[0081] Building the City Brain AI Computing Center includes: using the City Brain AI algorithm model to perform City Brain intelligent image recognition and build a cloud-edge-end city AI retina platform;

[0082] The intelligent image recognition of the city brain includes: face recognition, license plate recognition, cross-camera tracking of people, abnormal behavior detection, specific scene detection, and intelligent analysis of fireworks and drowning alarms; building a cloud-edge-end city AI retina platform includes: through a full-area intelligent camera monitoring network, deploying edge data centers, and using the computing power sharing platform as the AI ​​algorithm warehouse and computing power support base of the city brain's urban visual cloud platform to build a cloud-edge-end city AI retina platform; using the cloud-edge-end city AI retina platform to perform high-concurrency streaming media services and encoding and decoding, support massive video storage, national standard protocols and mainstream video protocols, and ten-thousand-channel video aggregation management.

[0083] The principles and effects of the above technical solutions are as follows: Based on the distributed on-demand computing architecture and heterogeneous computing resource management and scheduling, an urban brain AI computing center is built; based on the urban brain AI computing center, massive urban data computing resources are managed; based on the massive urban data computing resources management, urban governance emergency events are handled; building an urban brain AI computing center includes: using the urban brain AI algorithm model to perform urban brain intelligent image recognition and build a cloud-edge-end urban AI retina platform; urban brain intelligent image recognition includes: face recognition, license plate recognition, cross-camera tracking of personnel, abnormal behavior For detection, specific scene detection, fireworks alarm and drowning alarm intelligent analysis; build a cloud-edge-end city AI retina platform including: through the full-area intelligent camera monitoring network, deploy edge data centers, and use the computing power sharing platform as the AI ​​algorithm warehouse and computing power support base of the urban visual cloud platform of the city brain to build a cloud-edge-end city AI retina platform; use the cloud-edge-end city AI retina platform to perform high-concurrency streaming media services and encoding and decoding, support massive video storage, national standard protocols and mainstream video protocols, and ten thousand-channel video aggregation management; provide reliable security for business systems.

[0084] In one embodiment, S400 includes:

[0085] S401, remotely connect to the cloud host of the City Brain AI Computing Center and adjust the cloud host configuration by adjusting the CPU, memory, adding virtual disks, adding virtual network cards and the management interface console online;

[0086] S402: Directly manage bare metal physical servers on the cloud platform and perform heterogeneous resource status monitoring and management on physical servers in data centers with various architectures;

[0087] S403, predict future resource demand, pre-schedule and allocate in advance, and monitor the real-time resource usage of the system; perform seamless migration management of AI computing resources in the city brain and elastic scaling of distributed heterogeneous computing power scheduling of cluster physical node computing resources;

[0088] Carry out seamless migration management of AI computing resources in the city brain and elastically scale distributed heterogeneous computing power scheduling of computing resources in cluster physical nodes, including: remotely connecting to the cloud host of the AI ​​computing center of the city brain through online adjustment of CPU, memory, adding virtual disks, adding virtual network cards and management interface console, and adjusting the configuration of the cloud host; scheduling and managing the resource status of physical servers in data centers with various architectures, and directly managing the bare metal of physical servers on the cloud platform; processing AI computing task requests of the AI ​​retina platform of the city brain, matching them with AI algorithms in the preset AI algorithm warehouse, and scheduling them to available computing resources according to the specific needs of the task;

[0089] By comparing historical data with current conditions, we can predict future resource demands, pre-schedule and allocate resources in advance, prevent resource bottlenecks, and monitor the real-time resource usage of the system; use machine learning and big data analysis to detect database changes of physical servers in real time, and create an appropriate amount of edge data computing instances according to preset rules; collect and analyze historical data, count the operation of various AI algorithms on physical nodes in different clusters, and perform intelligent scheduling of computing power; use a zero-trust model to standardize system resource access rights, and perform system resource scheduling verification and authorization; use a microservice architecture to split each physical server into independent microservice clusters; detect and discover new edge data computing instances in real time; when the cloud hosts of the corresponding resource pool are unevenly distributed on the physical host, and the physical node resources of the cloud host where the main application is located are exhausted, use dynamic resource scheduling, through the resource balancing algorithm, according to the resource usage of each physical node, automatic cloud host resource allocation and load balancing resource dynamic scheduling;

[0090] By setting elastic scaling rules for computing resources, the computing resources of the business system can be automatically increased or automatically reduced. When business demand increases, the elastic business system computing resources can be seamlessly increased; when business demand decreases, the elastic computing resources can be automatically reduced. When any node fails, the cloud host on the corresponding node is directly migrated to other normal nodes, and the upper-level business system data is read and written in parallel during the migration. The city brain AI computing resources are seamlessly managed and the distributed heterogeneous computing power scheduling of cluster physical node computing resources is elastically scalable.

[0091] The principles and effects of the above technical solutions are as follows: by adjusting the CPU, memory, adding virtual disks, adding virtual network cards and management interface consoles online, remotely connecting to the cloud host of the City Brain AI computing center, and adjusting the cloud host configuration; directly managing the bare metal physical servers of the cloud platform, and monitoring and managing the heterogeneous resource status of physical servers in data centers with various architectures; predicting future resource requirements, pre-scheduling and allocating in advance, and monitoring the real-time resource usage of the system; and performing seamless migration management of the City Brain AI computing resources and elastic scaling of distributed heterogeneous computing power scheduling of cluster physical node computing resources;

[0092] The seamless migration management of the AI ​​computing resources of the city brain and the elastic scaling of the computing resources of the cluster physical nodes include: adjusting the cloud host configuration by remotely connecting to the cloud host of the city brain AI computing center through online adjustment of CPU, memory, adding virtual disks, adding virtual network cards and management interface console; scheduling and managing the resource status of physical servers in data centers with multiple architectures, and directly managing the bare metal physical servers of the cloud platform; processing the AI ​​computing task requests of the AI ​​retina platform of the city brain, matching them with the AI ​​algorithms in the preset AI algorithm warehouse, and scheduling them to the available computing resources according to the specific requirements of the task; predicting future resource requirements by comparing historical data with the current situation, pre-scheduling and allocating in advance, preventing resource bottlenecks, and monitoring the real-time resource usage of the system; using machine learning and big data analysis to detect database changes of physical servers in real time, and create an appropriate amount of edge data computing instances according to preset rules; collecting and analyzing historical data, counting the operation of various AI algorithms on different cluster physical nodes, and performing intelligent scheduling of computing power; using Based on the zero-trust model, the access rights to system resources are standardized, and system resource scheduling verification and authorization are performed; the microservice architecture is used to split each physical server into an independent microservice cluster; new edge data computing instances are detected and discovered in real time; when the cloud hosts of the corresponding resource pool are unevenly distributed on the physical host and the physical node resources of the cloud host where the main application is located are exhausted, dynamic resource scheduling is used to automatically allocate cloud host resources and dynamically schedule load balancing resources based on the resource usage of each physical node through the resource balancing algorithm; by setting elastic scaling rules for computing resources, the computing resources of the business system are automatically increased or automatically reduced, and when business demand increases, the elastic business system computing resources are seamlessly increased; when business demand decreases, the elastic computing resources are automatically reduced; when any node fails, the cloud host on the corresponding node is directly migrated to other normal nodes, and the upper-level business system data is read and written in parallel during the migration; seamless migration management of urban brain AI computing resources and elastic scaling of cluster physical node computing resources and distributed heterogeneous computing power scheduling are performed; it is suitable for data center scenarios with various architectural forms.

[0093] The present invention provides a scheduling system for distributed heterogeneous computing power of a city brain AI computing center, comprising:

[0094] The distributed on-demand computing power architecture subsystem deploys and controls the computing cluster physical nodes and storage network through a fully symmetrical distributed architecture; multiple computing cluster physical nodes are combined into an on-demand computing power decentralized cluster to build a distributed on-demand computing power architecture;

[0095] Heterogeneous computing resource management and control subsystem, quickly create cloud hosts through image management, define cloud host operating systems; manage cloud host computing resources throughout their life cycle, and perform visual control of heterogeneous computing resource management and scheduling through a unified self-reset scheduling interface control;

[0096] The computing power resource management and emergency response subsystem builds the city brain AI computing center based on the distributed on-demand computing power architecture and heterogeneous computing resource management and scheduling to manage massive urban data computing power resources and handle urban governance emergency events;

[0097] The cloud platform resource prediction and computing power scheduling subsystem directly manages the bare metal servers of the cloud platform by remotely connecting to the cloud host of the City Brain AI Computing Center; predicts future resource demands and performs pre-scheduling and allocation in advance; and manages the seamless migration of City Brain AI computing resources and the elastic scaling of distributed heterogeneous computing power scheduling of cluster physical node computing resources.

[0098] The principles and effects of the above technical solutions are as follows: The present invention provides a distributed heterogeneous computing power scheduling system for the AI ​​computing center of the city brain, including: a distributed on-demand computing power architecture subsystem, which deploys and controls the physical nodes and storage network of the computing cluster through a fully symmetrical distributed architecture; multiple computing cluster physical nodes are combined into a decentralized cluster of computing power that can be scaled up and down on demand to build a distributed on-demand computing power architecture; a heterogeneous computing resource management and control subsystem, which quickly creates a cloud host through image management and defines the cloud host operating system; manages the computing resources of the cloud host throughout its life cycle, and performs visual control of heterogeneous computing resource management and scheduling through a unified self-reset scheduling interface control; a computing power resource management emergency disposal subsystem, based on the distributed on-demand computing power architecture and Heterogeneous computing resource management and scheduling, building a city brain AI computing center, managing massive urban data computing resources and handling urban governance emergencies; cloud platform resource prediction computing scheduling subsystem, directly managing the bare metal of the cloud platform physical server by remotely connecting to the cloud host of the city brain AI computing center; predicting future resource needs and pre-scheduling and allocating resources in advance; performing seamless migration management of city brain AI computing resources and elastic scaling of cluster physical node computing resources distributed heterogeneous computing scheduling; the cluster node scale of the solution of the present invention can be quickly and flexibly expanded online as needed with business development to avoid performance bottlenecks; the storage capacity and performance of the distributed cluster can be linearly expanded with the increase in the number of nodes, and can ensure that the stored data is evenly distributed to all nodes Hit the mark; excellent stability and reliability, avoiding single point failures, improving system robustness, and ensuring business continuity; high operability, and reasonable organization and arrangement of functional modules; the platform adopts a fully symmetrical distributed architecture, deploys and controls the physical nodes of the computing cluster and the storage network cloud host, without the risk of single point failure, and is more flexible in adding and reducing nodes; it has the characteristics of high availability of cluster physical nodes, cloud hosts, systems, and data. If any node has a problem, the cloud host on the corresponding node can be directly migrated to other normal nodes. During the migration, the data reading and writing of the upper-level business system will not be affected. The parallel platform management interface and computing resource scheduling management provide reliable security for the business system; it supports the creation, modification, startup, and suspension of cloud hosts. , recovery, hibernation, restart, shutdown, clone, migration and snapshot, support for creating cloud hosts by image and customizing cloud host operating system by iso; support for adjusting cloud host configuration, including: online adjustment of CPU, memory, adding virtual disk and adding virtual network card; support for remote connection to cloud host through management interface console; cloud platform can not only manage other virtualization platforms in a unified manner, but also support direct management of physical servers, and can realize resource status and scheduling management of physical servers, which is suitable for data center scenarios with various architecture forms; when the cloud hosts of the corresponding resource pool are unevenly distributed on the physical hosts, it may lead to the exhaustion of resources of the physical nodes of the cloud hosts where the main applications are located, thus affecting the performance of the entire business system;The resource balancing algorithm used by the dynamic resource scheduling function can realize automatic cloud host resource allocation and load balancing functions based on the resource usage of each physical node, improve resource utilization efficiency, and ensure the normal operation of key businesses; according to the user's business needs, the business system computing resources can be automatically increased / reduced by setting scaling rules, seamlessly increasing the elasticity of the business system, and significantly improving the efficiency of computing resource utilization. ;

[0099] In one embodiment, the distributed on-demand computing power architecture subsystem includes:

[0100] The computing cluster physical node deployment subsystem deploys and controls the computing cluster physical nodes and storage network through a fully symmetric distributed architecture;

[0101] The on-demand computing power cluster subsystem combines multiple computing cluster physical nodes into a decentralized cluster that can scale computing power on demand.

[0102] The distributed on-demand computing power architecture subsystem builds a distributed on-demand computing power architecture based on the decentralized cluster of on-demand computing power expansion and reduction.

[0103] Deployment and control of computing cluster physical nodes and storage networks include: computing cluster physical nodes use GPUs and multiple chip models to perform active discovery and scheduling of GPUs / FPGAs / XPUs and computing power job scheduling and deployment, and build an AI elastic computing power resource pool storage network;

[0104] Based on the decentralized cluster of computing power that can be expanded or reduced on demand, a distributed computing power architecture that can be expanded or reduced on demand is built, including: providing underlying computing, storage and network resources through each node to form a software-defined infrastructure resource pool; all controlled system services are deployed to each node in a distributed and microservice architecture, and each node provides cloud service capabilities to the outside world; building a distributed computing power architecture that can be expanded or reduced on demand with redundancy, service self-healing and scalability.

[0105] The principle and effect of the above technical solution are as follows: the distributed on-demand computing power architecture subsystem includes: the computing cluster physical node deployment subsystem, which deploys and controls the computing cluster physical nodes and storage network through a fully symmetrical distributed architecture; the on-demand computing power cluster subsystem, which combines multiple computing cluster physical nodes into an on-demand computing power decentralized cluster; the distributed on-demand computing power architecture subsystem, which builds a distributed on-demand computing power architecture based on the on-demand computing power decentralized cluster; the deployment of the computing cluster physical nodes and storage network includes: the computing cluster physical nodes use GPUs and multiple models of chips for GPU / FPGA / XPU active Discovery scheduling and computing power job scheduling deployment, build AI elastic computing power resource pool storage network; based on the decentralized cluster of computing power that can be expanded or reduced on demand, build a distributed computing power architecture that can be expanded or reduced on demand, including: each node provides underlying computing, storage and network resources to form a software-defined infrastructure resource pool; all controlled system services are deployed to each node in a distributed and microservice architecture, and each node provides cloud service capabilities to the outside world; build a distributed computing power architecture that can be expanded or reduced on demand with redundancy, service self-healing and scalability; significantly improve redundancy, service self-healing and scalability, and can expand or reduce computing power architecture on demand, greatly improving flexibility.

[0106] In one embodiment, the heterogeneous computing resource management control subsystem includes:

[0107] The image management cloud host subsystem, based on the distributed on-demand computing architecture, supports the coexistence of traditional centralized architecture and distributed architecture through image management, and quickly creates cloud hosts and obtains image cloud hosts;

[0108] The cloud host operating subsystem defines the cloud host operating system through the ISO image file according to the image cloud host;

[0109] The lifecycle management control subsystem manages the computing resources of the cloud host throughout their lifecycle according to the cloud host operating system, and performs visual control of heterogeneous computing resource management and scheduling through a unified self-resetting scheduling interface.

[0110] The principles and effects of the above technical solution are as follows: the heterogeneous computing resource management and control subsystem includes: an image management cloud host subsystem, which supports the coexistence of traditional centralized architecture and distributed architecture through image management according to the distributed on-demand computing power architecture, and quickly creates cloud hosts and obtains image cloud hosts; a cloud host operation subsystem, which defines the cloud host operating system through ISO image files according to the image cloud host; a lifecycle management and control subsystem, which manages the cloud host computing resources throughout their life cycle according to the cloud host operating system, and performs visual control of heterogeneous computing resource management and scheduling through a unified self-resetting scheduling interface control; the level of intelligent control is significantly improved.

[0111] In one embodiment, the computing resource management emergency response subsystem includes:

[0112] The City Brain AI Computing Center construction subsystem is based on a distributed on-demand computing architecture and heterogeneous computing resource management and scheduling to build the City Brain AI Computing Center;

[0113] The city data computing power resource management subsystem manages massive city data computing power resources based on the city brain AI computing center;

[0114] The urban governance emergency response subsystem handles urban governance emergency events based on massive urban data computing resource management;

[0115] Building the City Brain AI Computing Center includes: using the City Brain AI algorithm model to perform City Brain intelligent image recognition and build a cloud-edge-end city AI retina platform;

[0116] The intelligent image recognition of the city brain includes: face recognition, license plate recognition, cross-camera tracking of people, abnormal behavior detection, specific scene detection, and intelligent analysis of fireworks and drowning alarms; building a cloud-edge-end city AI retina platform includes: through a full-area intelligent camera monitoring network, deploying edge data centers, and using the computing power sharing platform as the AI ​​algorithm warehouse and computing power support base of the city brain's urban visual cloud platform to build a cloud-edge-end city AI retina platform; using the cloud-edge-end city AI retina platform to perform high-concurrency streaming media services and encoding and decoding, support massive video storage, national standard protocols and mainstream video protocols, and ten-thousand-channel video aggregation management.

[0117] The principles and effects of the above technical solutions are as follows: The computing power resource management emergency response subsystem includes: the city brain AI computing center construction subsystem, which builds the city brain AI computing center based on the distributed on-demand expansion and reduction computing power architecture and heterogeneous computing resource management and scheduling; the city data computing power resource management subsystem, which manages massive city data computing power resources according to the city brain AI computing center; the city governance emergency response subsystem, which handles urban governance emergency events according to the massive city data computing power resource management; the construction of the city brain AI computing center includes: using the city brain AI algorithm model to perform city brain intelligent image recognition and build a cloud-edge-end city AI retina platform; the city brain Intelligent image recognition includes: face recognition, license plate recognition, cross-camera tracking of people, abnormal behavior detection, specific scene detection, and intelligent analysis of fireworks and drowning alarms; building a cloud-edge-end city AI retina platform includes: through a full-area intelligent camera monitoring network, deploying edge data centers, and using a computing power sharing platform as the AI ​​algorithm warehouse and computing power support base of the urban visual cloud platform of the city brain, building a cloud-edge-end city AI retina platform; using the cloud-edge-end city AI retina platform to perform high-concurrency streaming media services and encoding and decoding, support massive video storage, national standard protocols and mainstream video protocols, and ten-thousand-channel video aggregation management; providing reliable security protection for business systems.

[0118] In one embodiment, the cloud platform resource prediction computing power scheduling subsystem includes:

[0119] The cloud host configuration subsystem adjusts the CPU, memory, virtual disk, virtual network card and management interface console online, and remotely connects to the cloud host of the City Brain AI Computing Center to adjust the cloud host configuration;

[0120] The server direct management heterogeneous resource monitoring subsystem directly manages the bare metal physical servers of the cloud platform and monitors and manages the heterogeneous resource status of physical servers in data centers with various architectures;

[0121] The resource prediction computing power pre-scheduling subsystem predicts future resource needs, pre-scheduling allocations in advance, and monitors the real-time resource usage of the system; it manages the seamless migration of AI computing resources in the city brain and elastically scales distributed heterogeneous computing power scheduling for cluster physical node computing resources;

[0122] Carry out seamless migration management of AI computing resources in the city brain and elastically scale distributed heterogeneous computing power scheduling of computing resources in cluster physical nodes, including: remotely connecting to the cloud host of the AI ​​computing center of the city brain through online adjustment of CPU, memory, adding virtual disks, adding virtual network cards and management interface console, and adjusting the configuration of the cloud host; scheduling and managing the resource status of physical servers in data centers with various architectures, and directly managing the bare metal of physical servers on the cloud platform; processing AI computing task requests of the AI ​​retina platform of the city brain, matching them with AI algorithms in the preset AI algorithm warehouse, and scheduling them to available computing resources according to the specific needs of the task;

[0123] By comparing historical data with current conditions, we can predict future resource demands, pre-schedule and allocate resources in advance, prevent resource bottlenecks, and monitor the real-time resource usage of the system; use machine learning and big data analysis to detect database changes of physical servers in real time, and create an appropriate amount of edge data computing instances according to preset rules; collect and analyze historical data, count the operation of various AI algorithms on physical nodes in different clusters, and perform intelligent scheduling of computing power; use a zero-trust model to standardize system resource access rights, and perform system resource scheduling verification and authorization; use a microservice architecture to split each physical server into independent microservice clusters; detect and discover new edge data computing instances in real time; when the cloud hosts of the corresponding resource pool are unevenly distributed on the physical host, and the physical node resources of the cloud host where the main application is located are exhausted, use dynamic resource scheduling, through the resource balancing algorithm, according to the resource usage of each physical node, automatic cloud host resource allocation and load balancing resource dynamic scheduling;

[0124] By setting elastic scaling rules for computing resources, the computing resources of the business system can be automatically increased or automatically reduced. When business demand increases, the elastic business system computing resources can be seamlessly increased; when business demand decreases, the elastic computing resources can be automatically reduced. When any node fails, the cloud host on the corresponding node is directly migrated to other normal nodes, and the upper-level business system data is read and written in parallel during the migration. The city brain AI computing resources are seamlessly managed and the distributed heterogeneous computing power scheduling of cluster physical node computing resources is elastically scalable.

[0125] The principles and effects of the above technical solutions are as follows: the cloud platform resource prediction computing power scheduling subsystem includes: the cloud host configuration subsystem, which remotely connects to the cloud host of the city brain AI computing center by online adjustment of CPU, memory, addition of virtual disk, addition of virtual network card and management interface console, and adjusts the cloud host configuration; the server direct management heterogeneous resource monitoring subsystem, which directly manages the bare metal of the cloud platform physical server, and monitors and manages the heterogeneous resource status of physical servers in data centers with various architectures; the resource prediction computing power pre-scheduling subsystem, which predicts future resource requirements, pre-scheduling allocation in advance, and monitors the real-time resource usage of the system; it performs seamless migration management of the city brain AI computing resources and elastic scaling of distributed heterogeneous computing power scheduling of cluster physical node computing resources;

[0126] The seamless migration management of the AI ​​computing resources of the city brain and the elastic scaling of the computing resources of the cluster physical nodes include: adjusting the cloud host configuration by remotely connecting to the cloud host of the city brain AI computing center through online adjustment of CPU, memory, adding virtual disks, adding virtual network cards and management interface console; scheduling and managing the resource status of physical servers in data centers with multiple architectures, and directly managing the bare metal physical servers of the cloud platform; processing the AI ​​computing task requests of the AI ​​retina platform of the city brain, matching them with the AI ​​algorithms in the preset AI algorithm warehouse, and scheduling them to the available computing resources according to the specific requirements of the task; predicting future resource requirements by comparing historical data with the current situation, pre-scheduling and allocating in advance, preventing resource bottlenecks, and monitoring the real-time resource usage of the system; using machine learning and big data analysis to detect database changes of physical servers in real time, and create an appropriate amount of edge data computing instances according to preset rules; collecting and analyzing historical data, counting the operation of various AI algorithms on different cluster physical nodes, and performing intelligent scheduling of computing power; using Based on the zero-trust model, the access rights to system resources are standardized, and system resource scheduling verification and authorization are performed; the microservice architecture is used to split each physical server into an independent microservice cluster; new edge data computing instances are detected and discovered in real time; when the cloud hosts of the corresponding resource pool are unevenly distributed on the physical host and the physical node resources of the cloud host where the main application is located are exhausted, dynamic resource scheduling is used to automatically allocate cloud host resources and dynamically schedule load balancing resources based on the resource usage of each physical node through the resource balancing algorithm; by setting elastic scaling rules for computing resources, the computing resources of the business system are automatically increased or automatically reduced, and when business demand increases, the elastic business system computing resources are seamlessly increased; when business demand decreases, the elastic computing resources are automatically reduced; when any node fails, the cloud host on the corresponding node is directly migrated to other normal nodes, and the upper-level business system data is read and written in parallel during the migration; seamless migration management of urban brain AI computing resources and elastic scaling of cluster physical node computing resources and distributed heterogeneous computing power scheduling are performed; it is suitable for data center scenarios with various architectural forms.

[0127] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation methods. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. A method for scheduling distributed heterogeneous computing power in the AI ​​computing center of the city brain. It is characterized in that include: S100, through a fully symmetric distributed architecture, deploys and controls the computing cluster physical nodes and storage network; Combine multiple computing cluster physical nodes into a decentralized cluster that can scale computing power on demand, and build a distributed architecture that can scale computing power on demand; S200, quickly create cloud hosts through image management and define cloud host operating systems; Manage the entire life cycle of cloud host computing resources, and perform visual control of heterogeneous computing resource management and scheduling through a unified self-reset scheduling interface control; S300, based on the distributed on-demand computing architecture and heterogeneous computing resource management and scheduling, builds the city brain AI computing center to manage massive urban data computing resources and handle urban governance emergencies; S400 can directly manage the bare metal physical servers of the cloud platform by remotely connecting to the cloud host of the City Brain AI computing center; predict future resource requirements and make pre-scheduling allocations in advance; manage the seamless migration of City Brain AI computing resources and elastically scale the distributed heterogeneous computing power scheduling of cluster physical node computing resources; S400 includes: S401, remotely connect to the cloud host of the City Brain AI Computing Center and adjust the cloud host configuration by adjusting the CPU, memory, adding virtual disks, adding virtual network cards and the management interface console online; S402: Directly manage bare metal physical servers on the cloud platform and perform heterogeneous resource status monitoring and management on physical servers in data centers with various architectures; S403, predict future resource demand, pre-schedule and allocate in advance, and monitor the real-time resource usage of the system; perform seamless migration management of AI computing resources in the city brain and elastic scaling of distributed heterogeneous computing power scheduling of cluster physical node computing resources; The city brain AI computing resource seamless migration management and cluster physical node computing resource elastic scaling distributed heterogeneous computing power scheduling include: processing the AI ​​computing task requests of the city brain AI retina platform, matching them with the AI ​​algorithms in the preset AI algorithm warehouse, and scheduling them to available computing resources according to the specific needs of the task; predicting future resource needs by comparing historical data and current conditions, pre-scheduling and allocating in advance, preventing resource bottlenecks, and monitoring the real-time resource usage of the system; using machine learning and big data analysis to detect database changes of physical servers in real time, and creating an appropriate amount of edge data computing instances according to preset rules. For example; collect and analyze historical data, count the running status of various AI algorithms on physical nodes in different clusters, and perform intelligent scheduling of computing power; use a zero-trust model to standardize system resource access rights, and perform system resource scheduling verification and authorization; use a microservice architecture to split each physical server into an independent microservice cluster; detect and discover new edge data computing instances in real time; when the cloud hosts of the corresponding resource pool are unevenly distributed on the physical host, and the physical node resources of the cloud host where the main application is located are exhausted, use dynamic resource scheduling, through a resource balancing algorithm, according to the resource usage of each physical node, to automatically allocate cloud host resources and dynamically schedule load balancing resources; By setting elastic scaling rules for computing resources, the computing resources of the business system can be automatically increased or automatically reduced. When business demand increases, the elastic business system computing resources can be seamlessly increased; when business demand decreases, the elastic computing resources can be automatically reduced. When any node fails, the cloud host on the corresponding node is directly migrated to other normal nodes, and the upper-level business system data is read and written in parallel during the migration. The city brain AI computing resources are seamlessly managed and the distributed heterogeneous computing power scheduling of cluster physical node computing resources is elastically scalable.

2. According to claim 1, a method for scheduling distributed heterogeneous computing power of a city brain AI computing center, It is characterized in that S100 includes: S101, deploys and controls the computing cluster physical nodes and storage network through a fully symmetric distributed architecture; S102, multiple computing cluster physical nodes are combined into a decentralized cluster that can expand and reduce computing power on demand; S103, building a distributed on-demand computing power architecture based on the on-demand computing power decentralized cluster; Deployment and control of computing cluster physical nodes and storage networks include: computing cluster physical nodes use GPUs and multiple chip models to perform active discovery and scheduling of GPUs / FPGAs / XPUs and computing power job scheduling and deployment, and build an AI elastic computing power resource pool storage network; Based on the decentralized cluster of computing power that can be expanded or reduced on demand, a distributed computing power architecture that can be expanded or reduced on demand is built, including: providing underlying computing, storage and network resources through each node to form a software-defined infrastructure resource pool; all controlled system services are deployed to each node in a distributed and microservice architecture, and each node provides cloud service capabilities to the outside world; building a distributed computing power architecture that can be expanded or reduced on demand with redundancy, service self-healing and scalability.

3. According to claim 1, a method for scheduling distributed heterogeneous computing power of a city brain AI computing center, It is characterized in that S200 includes: S201, based on the distributed on-demand computing architecture, through image management, supports the coexistence of traditional centralized architecture and distributed architecture, quickly creates a cloud host, and obtains the image cloud host; S202, defining the cloud host operating system through the ISO image file according to the image cloud host; S203, according to the cloud host operating system, the cloud host computing resources are managed throughout their life cycle, and the heterogeneous computing resource management and scheduling are visually controlled through a unified self-resetting scheduling interface control.

4. According to claim 1, a method for scheduling distributed heterogeneous computing power of a city brain AI computing center, It is characterized in that S300 includes: S301: Build the AI ​​computing center of the city brain based on the distributed on-demand computing architecture and heterogeneous computing resource management and scheduling; S302: Manage massive urban data computing resources based on the city brain AI computing center; S303, handling urban governance emergencies based on massive urban data computing resource management; Building the City Brain AI Computing Center includes: using the City Brain AI algorithm model to perform City Brain intelligent image recognition and build a cloud-edge-end city AI retina platform.

5. A distributed heterogeneous computing power scheduling system for the city brain AI computing center. It is characterized in that include: The distributed on-demand computing power architecture subsystem deploys and controls the computing cluster physical nodes and storage network through a fully symmetrical distributed architecture; multiple computing cluster physical nodes are combined into an on-demand computing power decentralized cluster to build a distributed on-demand computing power architecture; Heterogeneous computing resource management and control subsystem, quickly creating cloud hosts through image management and defining cloud host operating systems; Manage the entire life cycle of cloud host computing resources, and perform visual control of heterogeneous computing resource management and scheduling through a unified self-reset scheduling interface control; The computing power resource management and emergency response subsystem builds the city brain AI computing center based on the distributed on-demand computing power architecture and heterogeneous computing resource management and scheduling to manage massive urban data computing power resources and handle urban governance emergency events; The cloud platform resource prediction computing power scheduling subsystem directly manages the bare metal of the cloud platform physical server by remotely connecting to the cloud host of the City Brain AI computing center; predicts future resource needs and performs pre-scheduling and allocation in advance; performs seamless migration management of the City Brain AI computing resources and elastic scaling of distributed heterogeneous computing power scheduling of cluster physical node computing resources; The cloud platform resource prediction computing power scheduling subsystem includes: The cloud host configuration subsystem adjusts the CPU, memory, virtual disk, virtual network card and management interface console online, and remotely connects to the cloud host of the City Brain AI Computing Center to adjust the cloud host configuration; The server direct management heterogeneous resource monitoring subsystem directly manages the bare metal physical servers of the cloud platform and monitors and manages the heterogeneous resource status of physical servers in data centers with various architectures; The resource prediction computing power pre-scheduling subsystem predicts future resource needs, pre-scheduling allocations in advance, and monitors the real-time resource usage of the system; it manages the seamless migration of AI computing resources in the city brain and elastically scales distributed heterogeneous computing power scheduling for cluster physical node computing resources; The city brain AI computing resource seamless migration management and cluster physical node computing resource elastic scaling distributed heterogeneous computing power scheduling include: processing the AI ​​computing task requests of the city brain AI retina platform, matching them with the AI ​​algorithms in the preset AI algorithm warehouse, and scheduling them to available computing resources according to the specific needs of the task; predicting future resource needs by comparing historical data and current conditions, pre-scheduling and allocating in advance, preventing resource bottlenecks, and monitoring the real-time resource usage of the system; using machine learning and big data analysis to detect database changes of physical servers in real time, and creating an appropriate amount of edge data computing instances according to preset rules. For example; collect and analyze historical data, count the running status of various AI algorithms on physical nodes in different clusters, and perform intelligent scheduling of computing power; use a zero-trust model to standardize system resource access rights, and perform system resource scheduling verification and authorization; use a microservice architecture to split each physical server into an independent microservice cluster; detect and discover new edge data computing instances in real time; when the cloud hosts of the corresponding resource pool are unevenly distributed on the physical host, and the physical node resources of the cloud host where the main application is located are exhausted, use dynamic resource scheduling, through a resource balancing algorithm, according to the resource usage of each physical node, to automatically allocate cloud host resources and dynamically schedule load balancing resources; By setting elastic scaling rules for computing resources, the computing resources of the business system can be automatically increased or automatically reduced. When business demand increases, the elastic business system computing resources can be seamlessly increased; when business demand decreases, the elastic computing resources can be automatically reduced. When any node fails, the cloud host on the corresponding node is directly migrated to other normal nodes, and the upper-level business system data is read and written in parallel during the migration. The city brain AI computing resources are seamlessly managed and the distributed heterogeneous computing power scheduling of cluster physical node computing resources is elastically scalable.

6. According to claim 5, a distributed heterogeneous computing power scheduling system for the city brain AI computing center, It is characterized in that The distributed on-demand computing architecture subsystem includes: The computing cluster physical node deployment subsystem deploys and controls the computing cluster physical nodes and storage network through a fully symmetric distributed architecture; The on-demand computing power cluster subsystem combines multiple computing cluster physical nodes into a decentralized cluster that can scale computing power on demand. The distributed on-demand computing power architecture subsystem builds a distributed on-demand computing power architecture based on the decentralized cluster of on-demand computing power expansion and reduction. Deployment and control of computing cluster physical nodes and storage networks include: computing cluster physical nodes use GPUs and multiple chip models to perform active discovery and scheduling of GPUs / FPGAs / XPUs and computing power job scheduling and deployment, and build an AI elastic computing power resource pool storage network; Based on the decentralized cluster of computing power that can be expanded or reduced on demand, a distributed computing power architecture that can be expanded or reduced on demand is built, including: providing underlying computing, storage and network resources through each node to form a software-defined infrastructure resource pool; all controlled system services are deployed to each node in a distributed and microservice architecture, and each node provides cloud service capabilities to the outside world; building a distributed computing power architecture that can be expanded or reduced on demand with redundancy, service self-healing and scalability.

7. According to claim 5, a scheduling system for distributed heterogeneous computing power of the city brain AI computing center, It is characterized in that The heterogeneous computing resource management and control subsystem includes: The image management cloud host subsystem, based on the distributed on-demand computing architecture, supports the coexistence of traditional centralized architecture and distributed architecture through image management, and quickly creates cloud hosts and obtains image cloud hosts; The cloud host operating subsystem defines the cloud host operating system through the ISO image file according to the image cloud host; The lifecycle management control subsystem manages the computing resources of the cloud host throughout their lifecycle according to the cloud host operating system, and performs visual control of heterogeneous computing resource management and scheduling through a unified self-resetting scheduling interface.

8. According to claim 5, a scheduling system for distributed heterogeneous computing power of the city brain AI computing center, It is characterized in that The computing resource management emergency response subsystem includes: The City Brain AI Computing Center construction subsystem is based on a distributed on-demand computing architecture and heterogeneous computing resource management and scheduling to build the City Brain AI Computing Center; The city data computing power resource management subsystem manages massive city data computing power resources based on the city brain AI computing center; The urban governance emergency response subsystem handles urban governance emergency events based on massive urban data computing resource management; Building the City Brain AI Computing Center includes: using the City Brain AI algorithm model to perform City Brain intelligent image recognition and build a cloud-edge-end city AI retina platform.

Citation Information

Cited By

  • Heterogeneous computing power full life cycle scheduling method and system

    CN122195612A