CN-121979674-A - Resource scheduling method, device, equipment, medium and product

Abstract

The method comprises the steps of responding to a power computing scheduling request, obtaining driving compatibility information on a server, wherein the driving compatibility information comprises at least one of state information of each GPU in the server, GPU driving version information, CUDA version information and GPU card identification, determining a driving environment of the power computing scheduling request according to the driving compatibility information, and executing a power computing scheduling task corresponding to the power computing scheduling request in the driving environment. According to the method, the matching of the computing power resources and the drive dependency relationship is realized by acquiring the drive compatibility information with the single GPU card as granularity, and the fully compatible isolation drive environment can be constructed or selected according to the information, so that the computing power scheduling request can be processed on the basis of the required exact drive version, the problem of computing power fragmentation is solved, and the GPUs of different drive versions in the server can be efficiently utilized.

Inventors

• Request for anonymity
• Request for anonymity
• Request for anonymity

Assignees

• 北京电子数智科技有限责任公司

Dates

Publication Date

20260505

Application Date

20260105

Claims (10)

1. 1. A method for scheduling resources, the method comprising: Responding to a computing power scheduling request, and acquiring drive compatibility information on a server, wherein the drive compatibility information comprises at least one of state information of each GPU in the server, GPU drive version information, CUDA version information and GPU card identification; and determining a driving environment of the power computing scheduling request according to the driving compatibility information, and executing a power computing scheduling task corresponding to the power computing scheduling request in the driving environment.
2. 2. The method of claim 1, wherein determining a driving environment for the power scheduling request based on the driving compatibility information comprises: Determining GPU driving version information corresponding to the power computing scheduling request from the driving compatibility information according to the power computing scheduling request; under the condition that a compatible driving environment matched with the GPU driving version information exists on the server, determining the compatible driving environment as the driving environment of the computing power dispatching request, wherein the compatible driving environment is a driving environment established based on a historical computing power dispatching request, and the driving environment of the historical computing power dispatching request is matched with the driving environment of the computing power dispatching request; and under the condition that no compatible driving environment matched with the driving version information exists on the server, loading the driving environment based on the GPU driving version information, and determining the established driving environment as the driving environment of the computing power scheduling request.
3. 3. The resource scheduling method according to claim 2, wherein loading a drive environment based on the drive version information in a case where there is no compatible drive environment on the server that matches the drive version information, comprises: Creating a virtual machine according to virtual machine information matched with the GPU driving version information under the condition that a compatible driving environment matched with the driving version information does not exist on the server; And loading the GPU driving version information in the virtual machine to obtain the driving environment.
4. 4. A method according to claim 3, wherein the state information includes an operating state and an idle state, and before the creating a virtual machine according to virtual machine information matched with the GPU-driven version information, comprising: determining whether the GPU in an idle state exists in the server according to the drive compatibility information; the creating a virtual machine according to the virtual machine information matched with the GPU driving version information includes: And under the condition that the idle GPU exists in the server, the GPU drives the virtual machine information matched with the version information, and the virtual machine is created.
5. 5. The method for scheduling resources according to claim 4, wherein loading the GPU driver version information in the virtual machine to obtain the driving environment includes: Determining a target GPU from the GPUs in the idle state on the server; binding the target GPU with the virtual machine, and loading the GPU driving version information in the bound virtual machine to obtain the driving environment.
6. 6. The resource scheduling method according to claim 5, wherein after the driving environment of the power scheduling request is determined according to the driving compatibility information and the power scheduling task corresponding to the power scheduling request is executed in the driving environment, the method further comprises: updating the state information of the GPU in the driving compatibility information into a working state according to the GPU bound with the virtual machine; and establishing a corresponding relation between the computing power dispatching request and the driving environment, and recording the corresponding relation in the driving compatibility information.
7. 7. A resource scheduling apparatus, the apparatus comprising: The information acquisition module is configured to respond to the computational power scheduling request and acquire driving compatibility information on a server, wherein the driving compatibility information comprises at least one of state information, GPU driving version information, CUDA version information and GPU card identification of each GPU in the server; and the power computing scheduling module is configured to determine a driving environment of the power computing scheduling request according to the driving compatibility information and execute a power computing scheduling task corresponding to the power computing scheduling request in the driving environment.
8. 8. A server for a server, which comprises a server and a server, characterized by comprising the following steps: a memory having a computer program stored thereon; A processor for executing the computer program in the memory to implement the steps of the method of any one of claims 1-6.
9. 9. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any one of claims 1-6.
10. 10. A computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, implements the steps of the method according to any of claims 1-6.

Description

Resource scheduling method, device, equipment, medium and product Technical Field The disclosure relates to the technical field of cloud computing, in particular to a resource scheduling method, a device, equipment, a medium and a product. Background In an artificial intelligence reasoning scenario, model optimization achievements (such as TensorRT engines) form a strong binding relationship with a specific version of GPU driver and CUDA environment, resulting in limited deployment of models across environments. In the related art, a uniform driving strategy is adopted, the whole server is used as granularity to manage GPU resources, and the driving version differences of different GPU cards in the server cannot be identified. When the model is compatible with only a portion of the GPU, the system can only totally exclude the server, resulting in computational fragmentation and reduced resource utilization. Disclosure of Invention The invention aims to provide a resource scheduling method, a device, equipment, a medium and a product. To achieve the above object, in a first aspect, the present disclosure provides a resource scheduling method, the method including: Responding to a computing power scheduling request, and acquiring drive compatibility information on a server, wherein the drive compatibility information comprises at least one of state information of each GPU in the server, GPU drive version information, CUDA version information and GPU card identification; and determining a driving environment of the power computing scheduling request according to the driving compatibility information, and executing a power computing scheduling task corresponding to the power computing scheduling request in the driving environment. Optionally, the determining the driving environment of the computing power scheduling request according to the driving compatibility information includes: Determining GPU driving version information corresponding to the power computing scheduling request from the driving compatibility information according to the power computing scheduling request; under the condition that a compatible driving environment matched with the GPU driving version information exists on the server, determining the compatible driving environment as the driving environment of the computing power dispatching request, wherein the compatible driving environment is a driving environment established based on a historical computing power dispatching request, and the driving environment of the historical computing power dispatching request is matched with the driving environment of the computing power dispatching request; and under the condition that no compatible driving environment matched with the driving version information exists on the server, loading the driving environment based on the GPU driving version information, and determining the established driving environment as the driving environment of the computing power scheduling request. Optionally, in the case that no compatible driving environment matching the driving version information exists on the server, loading the driving environment based on the driving version information includes: Creating a virtual machine according to virtual machine information matched with the GPU driving version information under the condition that a compatible driving environment matched with the driving version information does not exist on the server; And loading the GPU driving version information in the virtual machine to obtain the driving environment. Optionally, the state information includes an operating state and an idle state, and before the creating a virtual machine according to the virtual machine information matched with the GPU driving version information, the method includes: determining whether the GPU in an idle state exists in the server according to the drive compatibility information; the creating a virtual machine according to the virtual machine information matched with the GPU driving version information includes: And under the condition that the idle GPU exists in the server, the GPU drives the virtual machine information matched with the version information, and the virtual machine is created. Optionally, loading the GPU driver version information in the virtual machine to obtain the driving environment includes: Determining a target GPU from the GPUs in the idle state on the server; binding the target GPU with the virtual machine, and loading the GPU driving version information in the bound virtual machine to obtain the driving environment. Optionally, after determining a driving environment of the power scheduling request according to the driving compatibility information and executing a power scheduling task corresponding to the power scheduling request in the driving environment, the method further includes: updating the state information of the GPU in the driving compatibility information into a working state according to the GPU bound with the