Heterogeneous intelligent computing equipment pooling and capability supply method based on container cloud

Abstract

The invention discloses a heterogeneous intelligent computing equipment pooling and capability supply method based on container cloud. The supply method comprises the following steps: establishing a container cloud cluster and carrying out compatibility on a heterogeneous product mirror image; arranging base lines and drivers of different types of intelligent computing card equipment of multiple manufacturers; the container cloud manages intelligent computing card nodes and adapts to and integrates intelligent computing card devices of different manufacturers; starting computing power providing modes of different intelligent computing cards according to the intelligent computing card devices of different manufacturers; performing node pooling on the node marks to form intelligent computing equipment node pooling; a user applies for a distribution process through the intelligent computing device, and life cycle management and computing power resource quota binding of the intelligent computing device computing power resource pooling are carried out; and the user performs cloud delivery of the artificial intelligence application according to the computing power resource quota to realize heterogeneous intelligent computing equipment pooling and capability supply. And node pooling and resource pooling of the intelligent computing equipment are carried out, and a basic capability of providing a service base for intelligent computing capability output is provided.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence services in the financial industry, and in particular to a method for pooling and supplying heterogeneous intelligent computing devices based on container cloud. Background Technology

[0002] Artificial intelligence (AI) technology is sweeping the world, even within the fintech landscape. Many banks, having undergone technological innovation through containerization, are now facing the challenge of AI-driven technological reform. Container clouds, built upon declarative cloud computing technologies such as container orchestration, elastic scaling, platformization, and management, present a significant challenge in providing the capabilities for AI service model execution, service management, and operational monitoring within a cloud-native environment. When providing intelligent computing capabilities, key performance indicators for evaluating a container cloud's ability to support AI service operation include: "intelligent computing device integration," "diversified intelligent computing device operation modes," "intelligent computing device pooling," and "intelligent computing device visualization capabilities."

[0003] In the existing cloud-native environment of the fintech industry, some companies that provide self-developed intelligent computing capabilities, in order to quickly "get started" and run models or artificial intelligence services, directly deploy "bare metal" by installing intelligent computing device drivers and code environments on a single host; while others have carried out preliminary containerization, integrating intelligent computing configuration components related to container runtime after installing intelligent computing device drivers on a single host, thereby completing single-node containerized deployment. The lack of cluster-level heterogeneous intelligent computing device integration, intelligent computing device virtualization, intelligent computing device pooling, and resource quota restrictions prevents the provision of cloud-native intelligent computing device capabilities. Due to the inability to achieve this, the application and allocation of intelligent computing devices is also lacking.

[0004] Disadvantages of existing technology: 1. It lacks the ability to integrate diverse heterogeneous intelligent computing devices (domestic and non-domestic) at the container cloud cluster level.

[0005] 2. It lacks the ability to integrate diverse intelligent computing devices (shared and exclusive) in different modes at the container cloud cluster level.

[0006] 3. It lacks resource pool allocation and resource quota restrictions for container cloud cluster-level intelligent computing devices.

[0007] 4. It lacks a resource application and allocation process after the container cloud cluster-level intelligent computing device pooling.

[0008] 5. Lacks the capability to deliver container cloud cluster-level artificial intelligence applications.

[0009] 6. It lacks visualized intelligent computing device pooling management and lifecycle management, resource application and allocation, and artificial intelligence application delivery. Summary of the Invention

[0010] In view of the above problems, the present invention is proposed to provide a method for pooling and supplying heterogeneous intelligent computing devices based on container cloud to overcome or at least partially solve the above problems.

[0011] According to one aspect of the present invention, a method for pooling and supplying heterogeneous intelligent computing devices based on container cloud is provided, the supply method comprising: The construction of container cloud clusters and compatibility with heterogeneous artifact images; Compile baseline and driver installation information for various models of smart computing cards from multiple manufacturers; Container cloud manages smart computing card nodes and adapts and integrates smart computing card devices from different manufacturers; Different computing card computing power provision modes are activated based on the different manufacturers' smart computing card devices; Node marking is used to form a smart computing device node pool; Users apply for allocation through the intelligent computing device application process, and combine the content information of the node pooling read by the container cloud to perform lifecycle management of the intelligent computing device computing power resource pooling and binding of computing power resource quotas; Users deliver AI applications on the cloud based on their applied computing power resource quotas, enabling the pooling of heterogeneous intelligent computing devices and the supply of capabilities.

[0012] Optionally, the smart computing card device specifically includes: Nvidia smart computing card and Ascend smart computing card.

[0013] Optionally, the node marker specifically includes: the manufacturer, model, and operating mode of the intelligent computing device.

[0014] Optionally, the compatibility of the heterogeneous article mirror image specifically includes: The image Manifest file is maintained in the artifact repository to ensure that images of different architectures are accessed and used with a unified name, so that nodes of different architectures can pull images of different architectures according to their own node architecture.

[0015] Optionally, the container cloud-managed smart computing card nodes and the adaptation and integration of smart computing card devices from different manufacturers specifically include: Smart computing card device driver baseline determination: When deploying and installing drivers for smart computing card devices from different manufacturers and of different models, the driver version baseline is unified; Node adaptation and management: The container cloud cluster manages X86 non-IT innovation nodes and ARM64 IT innovation nodes equipped with intelligent computing card hardware as working nodes; Intelligent computing mode integration: The container cloud cluster adapts and integrates cluster plug-in components based on the manufacturer, card model, card specifications, and operating mode of the intelligent computing card hardware device.

[0016] Optionally, the intelligent computing device node pooling specifically includes: The intelligent computing nodes are pooled according to the manufacturer, model, and specifications of the intelligent computing cards in the nodes, and the pooling of intelligent computing nodes is achieved by marking the part with node tags.

[0017] Optionally, the pooling of computing power resources for intelligent computing devices specifically includes: Intelligent computing resource pooling occurs after intelligent computing device card machine pooling. Resource pools are created based on machine pools, and a resource pool can only be bound to one machine pool. All modes of intelligent computing cards supported by the machine pool can be selected and used. A resource pool is uniquely bound to a namespace in the cloud platform.

[0018] Optionally, the heterogeneous intelligent computing device pooling and capability provision specifically includes: Cloud-based deployment, delivery, and operation of AI applications: After the project team users complete the application for intelligent computing resources, they can use the intelligent computing resources in the container cloud just like using general computing resources to complete the cloud-based deployment, delivery, and operation of AI applications. The operating platform of an external artificial intelligence platform: The external artificial intelligence platform needs to use the intelligent computing resources of the container cloud platform and connect to the resource pool and default service account provided by the container cloud platform.

[0019] This invention provides a method for pooling and supplying heterogeneous intelligent computing devices based on container cloud. The supply method includes: building a container cloud cluster and ensuring compatibility with heterogeneous artifact images; organizing baselines and driver installations for intelligent computing card devices of different models from multiple manufacturers; managing intelligent computing card nodes in the container cloud and adapting and integrating intelligent computing card devices from different manufacturers; initiating different computing power supply modes for intelligent computing cards based on different manufacturers; pooling nodes by marking nodes to form an intelligent computing device node pool; managing the lifecycle of the intelligent computing device computing power resource pool and binding computing power resource quotas by users through the intelligent computing device application and allocation process, combined with the container cloud reading the content information of the node pool; and delivering artificial intelligence applications on the cloud according to the computing power resource quotas applied for, thus realizing the pooling and supply of heterogeneous intelligent computing devices. The container cloud can manage x86 non-IT-compliant nodes and ARM64 IT-compliant nodes as working nodes, and perform node pooling and resource pooling of intelligent computing devices, providing a service foundation for intelligent computing capability output.

[0020] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating a method for pooling and supplying heterogeneous intelligent computing devices based on container cloud, provided as an embodiment of the present invention; Figure 2 A flowchart illustrating the construction of a container cloud cluster and compatibility with heterogeneous artifact images provided in this embodiment of the invention; Figure 3 This is a schematic diagram illustrating the computing power provision modes of different smart computing cards based on smart computing card devices from different manufacturers, provided as an embodiment of the present invention. Figure 4 This is a schematic diagram of intelligent computing node pooling provided in an embodiment of the present invention; Figure 5 This is a flowchart illustrating how administrators can directly select projects, create resource pools, and bind them through a cloud platform, as provided in this embodiment of the invention. Figure 6 The flowchart provided in this embodiment of the invention shows how project team users can apply for and create resource pools through a resource application process. Figure 7 A flowchart illustrating how resource pools, projects, and node groups are bound together using tags, as provided in this embodiment of the invention. Figure 8 A flowchart illustrating the cloud deployment, delivery, and operation of an artificial intelligence application provided in this embodiment of the invention. Detailed Implementation

[0023] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0024] The terms "comprising" and "having," and any variations thereof, in the specification, embodiments, claims, and drawings of this invention are intended to cover non-exclusive inclusion, such as including a series of steps or units.

[0025] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0026] like Figure 1 As shown, the method for pooling and supplying heterogeneous intelligent computing devices according to this invention requires a complete process. The first step is the construction of a container cloud cluster and compatibility with heterogeneous product images; the second step is to organize the baselines and driver installations of intelligent computing card devices from different manufacturers and models; the third step is to manage intelligent computing card nodes in the container cloud and adapt and integrate intelligent computing card devices from different manufacturers, such as Nvidia intelligent computing cards (x86) and Ascend intelligent computing cards (arm64); the fourth step is to activate different computing power supply modes for intelligent computing cards from different manufacturers, such as shared (virtualization) and exclusive; the fifth step is to pool nodes by marking them with node tags (manufacturer, model, and operating mode of the intelligent computing device); the sixth step is for users to apply for allocation of intelligent computing devices, combined with the container cloud reading the content information of the node pool, to perform lifecycle management and computing power quota binding for the intelligent computing device computing power resource pool; the seventh step is for users to deliver artificial intelligence applications on the cloud according to their applied computing power resource quota, thereby realizing the pooling and supply of heterogeneous intelligent computing devices.

[0027] like Figure 2 As shown, the compatibility of heterogeneous artifact mirrors: The image Manifest file is maintained in the artifact repository to ensure that images of different architectures are accessed and used with a unified name, so that nodes with different architectures can pull images of different architectures according to their own node architecture.

[0028] like Figure 3 As shown, the intelligent computing card device driver baseline determination and node adaptation management provide intelligent computing capabilities in different modes: Smart computing card device driver baseline determination: When deploying and installing drivers for smart computing card devices from different manufacturers and of different models, the driver version baseline is unified to prevent differences in service deployment, delivery and runtime in container cloud.

[0029] Node adaptation and management: The container cloud cluster manages X86 non-IT innovation nodes and ARM64 IT innovation nodes equipped with intelligent computing card hardware as working nodes.

[0030] Intelligent computing mode integration: The container cloud cluster adapts and integrates cluster plug-in components based on the manufacturer, card model, card specifications, and operating mode (shared or exclusive mode) of the intelligent computing card hardware device.

[0031] like Figure 4 As shown, intelligent computing node pooling: Because the same type of card can have different specifications (hardware interface, video memory, number of cores), resulting in varying performance, there are differences in how AI applications run in the cloud. Therefore, it is necessary to divide the intelligent computing nodes into pools according to the manufacturer, model, and specifications of the intelligent computing cards in the nodes, and to mark these parts with node tags to achieve intelligent computing node pooling.

[0032] Different card types: Classified according to the rule of "card model" - "card interface" - "card memory". Card types with the same specifications: Classified according to the "card model" rule The meanings and explanations of node labels are as follows: like Figure 5 As shown, administrators can directly select projects to create resource pools and bind them through the cloud platform.

[0033] like Figure 6 As shown, project team users can apply for and create resource pools through the resource application process: Resource application process: Resource pools are bound to projects and node groups via tags: like Figure 7 As shown.

[0034] Supply of intelligent computing resources: like Figure 8 As shown, the cloud deployment, delivery, and operation of artificial intelligence applications: After the project team users complete the application for intelligent computing resources, they can use the intelligent computing resources in the container cloud just like using general computing resources, and complete the cloud deployment, delivery and operation of artificial intelligence applications.

[0035] The operating platform of external artificial intelligence platforms: External AI platforms that need to use the intelligent computing resources of the container cloud platform can do so by connecting to the resource pool and default service account provided by the container cloud platform.

[0036] Beneficial effects: There is no intrusion into business code, and it does not interfere with the operation of business services in any way.

[0037] Solving needs from the internal layer of container cloud does not require external expansion or modification.

[0038] It meets the needs of domestic IT innovation based on diverse heterogeneous architectures such as X86 and ARM64.

[0039] The pooling of machine pools can be accomplished using container cloud features without the need for secondary encapsulation.

[0040] The pooling of resources is completed using internal and extended resources of the container cloud, without the need for secondary encapsulation or reference to external databases.

[0041] The visualization method enables full lifecycle management of intelligent computing resources in container cloud, including application, modification, and recycling, without the need to reference external data infrastructure.

[0042] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A container cloud-based heterogeneous intelligent computing device pooling and capability provisioning method, characterized in that, The supply method comprises: The construction of the container cloud cluster and the compatibility of the heterogeneous product image; Organizing the baseline and driver installation of different models of intelligent computing card equipment of multiple manufacturers; The container cloud manages the intelligent computing card nodes and adapts and integrates the intelligent computing card equipment of different manufacturers; According to different intelligent computing card equipment of different manufacturers, the computing power supply mode of different intelligent computing cards is started; Node marking is performed to form an intelligent computing device node pool; Users apply for distribution through intelligent computing devices, and the container cloud reads the content information of the node pool to manage the life cycle of the intelligent computing device computing resource pool and bind the computing resource quota; Users apply for artificial intelligence application cloud delivery according to the computing resource quota to realize heterogeneous intelligent computing device pooling and capability supply.

2. The container cloud-based heterogeneous computing device pooling and capacity provisioning method of claim 1, wherein, The intelligent computing card equipment specifically comprises: Nvidia intelligent computing card, Ascend intelligent computing card.

3. The container cloud-based heterogeneous computing device pooling and capacity provisioning method of claim 1, wherein, The node marking specifically comprises: manufacturer, model, and running mode of the intelligent computing device.

4. The container cloud-based heterogeneous computing device pooling and capacity provisioning method of claim 1, wherein, The compatibility of the heterogeneous product image specifically comprises: Maintain the image Manifest file in the product library to ensure that multiple architecture images are accessed and used with a unified name to meet the needs of nodes of different architectures to pull different architecture images according to their node architecture.

5. The container cloud-based heterogeneous computing device pooling and capacity provisioning method of claim 1, wherein, The container cloud manages the intelligent computing card nodes and adapts and integrates the intelligent computing card equipment of different manufacturers specifically comprises: Intelligent computing card equipment driver baseline determination: unify the baseline of the driver version when deploying and installing the driver for different manufacturers and different models of intelligent computing card equipment; Node adaptation management: the container cloud cluster manages X86 non-credence nodes and ARM64 credence nodes with intelligent computing card hardware devices as worker nodes; Intelligent computing mode integration: the container cloud cluster adapts and integrates cluster plug-in components according to the manufacturer, card model, card specification, and running mode of the intelligent computing card hardware device.

6. The container cloud-based heterogeneous computing device pooling and capacity provisioning method of claim 1, wherein, The intelligent computing device node pooling specifically comprises: According to the manufacturer, card model, and card specification of the intelligent computing card in the node, the intelligent computing node pool is divided, and the node label is marked to realize the intelligent computing node pool.

7. The container cloud-based heterogeneous computing device pooling and capacity provisioning method of claim 1, wherein, The intelligent computing device computing resource pooling specifically comprises: Intelligent resource pooling occurs after intelligent computing card machine pooling, and a resource pool can only bind one machine pool. The resource pool is created according to the machine pool, and all supported mode intelligent cards in the machine pool are selected. A resource pool is uniquely bound to a namespace in the cloud platform.

8. The container cloud-based heterogeneous computing device pooling and capacity provisioning method of claim 1, wherein, The heterogeneous intelligent computing device pooling and capability supply specifically comprises: Cloud deployment and delivery of artificial intelligence applications: after the user of the project group completes the intelligent resource application, the intelligent resource is used like general computing resource in the container cloud to complete the cloud deployment and delivery of artificial intelligence applications; Running base of external artificial intelligence platform: the external artificial intelligence platform needs to use the intelligent resource of the container cloud platform, and interfaces with the resource pool and the default service account provided by the container cloud platform.

Images