Cloud desktop heterogeneous GPU resource scheduling management method and system

Abstract

The invention discloses a cloud desktop heterogeneous GPU resource scheduling management method and system, belongs to the technical field of AI artificial intelligence, cloud desktops and cloud computing, and aims to solve the technical problem that an existing cloud computer cannot flexibly use the computing power of a newly-added heterogeneous GPU. Reporting, counting, distributing and scheduling different heterogeneous GPU resources, and dynamically screening resource nodes meeting user requirements and mounting computing power into a user cloud desktop in combination with general resource scheduling of a CPU and a memory; the method specifically comprises the following steps: managing heterogeneous GPU resources; and scheduling heterogeneous GPU resources of the cloud desktop.

Description

Technical Field

[0001] This invention relates to the fields of AI artificial intelligence, cloud desktop and cloud computing technology, specifically a method and system for scheduling and managing heterogeneous GPU resources in cloud desktops. Background Technology

[0002] Currently, with the gradual expansion of the domestic GPU market, many domestic GPU manufacturers have launched GPU cards targeting different computing power scenarios to meet the needs of users in various industries. From the perspective of cloud desktop products, how to efficiently allocate the underlying GPU computing power to meet the computing power needs of different cloud desktop users' businesses has become a relatively pressing pain point.

[0003] Traditional resource scheduling logic involves the following steps for using GPU resources: GPU computing power discovery, GPU computing power statistics, resource matching during business cloud desktop creation, and creation and use of the GPU resource on a suitable node. This traditional process requires the underlying server to have the specified GPU computing power before a user creates a cloud desktop. Furthermore, it's difficult to meet the needs of cloud desktops already in use with subsequently added GPU computing power. Summary of the Invention

[0004] The technical objective of this invention is to provide a method and system for scheduling and managing heterogeneous GPU resources in cloud desktops, in order to solve the problem that existing cloud computers cannot flexibly utilize the computing power of newly added heterogeneous GPUs.

[0005] The technical objective of this invention is achieved as follows: a cloud desktop heterogeneous GPU resource scheduling and management method. This method detects the characteristic information of different vendors, combines it with the vendor's GPU driver, and completes the reporting, statistics, allocation, and scheduling of different heterogeneous GPU resources. Simultaneously, it combines general resource scheduling of CPU and memory to dynamically select resource nodes that meet user needs and mount computing power to the user's cloud desktop; specifically as follows:

[0006] Managing heterogeneous GPU resources;

[0007] Schedule heterogeneous GPU resources for cloud desktops.

[0008] As a preferred approach, the management of heterogeneous GPU resources is as follows:

[0009] The computing service obtains the device type of all GPU computing cards on the current node through the system API, and matches and enables the corresponding vendor management driver based on the GPU card's Vendor ID;

[0010] Different manufacturers' GPU management drivers obtain detailed resource information of the corresponding GPU computing cards from the nodes and report the resource computing power status to the resource management service;

[0011] The resource management service compiles statistics on the GPU computing power consumption of each node and, based on the computing power request, provides feedback on the node information that meets the resource requirements.

[0012] As a preferred option, the scheduling of heterogeneous GPU resources for cloud desktops is as follows:

[0013] Users can select a GPU computing card that matches their business scenario from the management interface and mount it to a designated cloud desktop based on their business needs.

[0014] The scheduling service checks whether the node where the cloud desktop machine is located meets the user's GPU resource requirements:

[0015] If the conditions are met, GPU resources will be allocated based on the current node;

[0016] If not, filter the compute nodes to see if there are any nodes that meet the requirements for GPU resources and migration resources, and return the optimal node;

[0017] After receiving node information from the scheduling service, the computing service first determines whether the current node has a GPU computing card that meets the computing power requirements:

[0018] If it exists, allocate and consume the specified GPU computing resources from the current node;

[0019] If not, a suitable node is found from the resource management service, and the user's virtual machine is migrated to the specified node before the corresponding GPU driver resources are enabled;

[0020] After the computing service migrates the cloud desktop to the optimal node, it mounts the corresponding GPU computing resources into the cloud desktop and restores the cloud desktop to its initial state.

[0021] As a preferred option, the computing service supports multiple methods for detecting GPU devices, including system APIs and system command lines.

[0022] As a preferred option, different heterogeneous GPU computing resource management drivers are started simultaneously, and multiple GPU devices can coexist on a single physical node.

[0023] As a preferred option, users can select GPU computing power information that meets their needs. Supported filtering parameters include GPU manufacturer information, GPU card name, GPU memory size, GPU operating mode, and GPU type (physical card or virtual card). Multiple filtering parameters can be selected for precise matching.

[0024] Even better, the GPU resource management driver identifies the detailed device information of the GPU computing card; among which, the detailed device information of the GPU computing card includes GPU manufacturer information, GPU card name, GPU memory, number of GPUs and GPU operating mode.

[0025] A cloud desktop heterogeneous GPU resource scheduling and management system, which implements the cloud desktop heterogeneous GPU resource scheduling and management method described above; the system includes:

[0026] The computing service module is used to detect the GPU device of the current system through the system interface and enable the corresponding vendor GPU driver according to the GPU card's Vendor ID.

[0027] The information acquisition and reporting module is used to obtain detailed resource information of the corresponding GPU card from the node through the GPU management driver of different manufacturers, and report the resource status to the resource management service.

[0028] The mounting module is used to allow users to select and mount suitable GPU cards from the management platform to the cloud desktop according to their business needs.

[0029] The scheduling service module is used to detect whether the node where the cloud desktop machine is located meets the user's GPU resource requirements.

[0030] If the conditions are met, GPU resources will be allocated based on the current node;

[0031] If not, filter the compute nodes to see if there are any nodes that meet the requirements for GPU resources and migration resources, and return the optimal node;

[0032] The judgment module is used to determine the node information received from the scheduling service through the computing service. If it is the current node, the corresponding GPU driver is enabled to perform resource operations. If it is not the current node, the user virtual machine is migrated to the specified node and the corresponding GPU driver resources are enabled.

[0033] The initial state recovery module is used to mount the user's desired computing power to the user's cloud desktop through the computing service and the corresponding GPU driver from the manufacturer, and restore the cloud desktop to its initial state.

[0034] An electronic device includes: a memory and at least one processor;

[0035] The memory contains computer programs;

[0036] The at least one processor executes the computer program stored in the memory, causing the at least one processor to perform the cloud desktop heterogeneous GPU resource scheduling and management method as described above.

[0037] A computer-readable storage medium storing a computer program that can be executed by a processor to implement the cloud desktop heterogeneous GPU resource scheduling and management method described above.

[0038] The cloud desktop heterogeneous GPU resource scheduling and management method and system of the present invention have the following advantages:

[0039] (i) The computing service of this invention detects the GPU devices of the current system through the system interface and enables the corresponding vendor GPU driver according to the vendor ID of the GPU card; the GPU drivers of different vendors obtain detailed resource information of the corresponding GPU cards from the nodes and report the resource status to the resource management service; the user selects the GPU card that meets the requirements from the management platform and mounts it to the cloud desktop according to the business needs; the scheduling service filters the optimal node according to the resource requirements and prioritizes the current node for matching; the computing service migrates the business cloud desktop to the designated node; the computing service mounts the specified computing power to the cloud desktop through the GPU driver and restores it to the initial state, filters the optimal node that meets the requirements according to the user's requirements for computing power, vendor, and GPU card, and synchronously migrates the user's cloud desktop to the designated node and mounts the specified type of GPU card, thus solving the problem that existing cloud computers cannot flexibly use the newly added heterogeneous GPU computing power;

[0040] (II) This invention detects the characteristic information of different manufacturers and combines the manufacturer's GPU driver to complete the reporting, statistics, allocation and scheduling of different heterogeneous GPU resources. At the same time, it combines the scheduling of general resources such as CPU and memory to dynamically select resource nodes that meet user needs and mount computing power to the user's cloud desktop, thus solving the pain point that existing cloud computers cannot flexibly use the newly added heterogeneous computing power.

[0041] (III) This invention realizes the management of heterogeneous GPU resource scheduling in cloud desktops. By detecting the characteristic information of different manufacturers and combining the manufacturer's GPU driver, it completes the reporting, statistics, allocation and scheduling of different heterogeneous GPU resources. At the same time, it combines the scheduling of general resources such as CPU and memory to dynamically select resource nodes that meet the user's needs and mount the computing power to the user's cloud desktop.

[0042] (iv) This invention supports resource management for various GPU devices, including but not limited to NVIDIA, Moore Threads, Jingjia Microelectronics, Hygon, and Cambricon;

[0043] (v) The resource scheduling of the present invention not only supports scheduling and using the GPU computing power resources of the current node of the cloud desktop, but also supports scheduling and using the GPU computing power resources of other physical nodes.

[0044] (vi) The scheduling algorithm of the present invention supports mode configuration. The supported modes include, but are not limited to, forced matching mode (matching will only succeed if all user requirements are met) and optimal matching mode (matching will only succeed if a certain degree of satisfaction is achieved based on the priority of each GPU's requirements).

[0045] (vii) The computing service of the present invention can simultaneously mount different heterogeneous computing power GPU resources to the same cloud desktop, and is compatible with all GPU manufacturers’ devices. Attached Figure Description

[0046] The invention will be further described below with reference to the accompanying drawings.

[0047] Appendix Figure 1 This is a flowchart illustrating the method for scheduling and managing heterogeneous GPU resources in cloud desktops. Detailed Implementation

[0048] The cloud desktop heterogeneous GPU resource scheduling and management method and system of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] Example 1:

[0050] This embodiment provides a method for scheduling and managing heterogeneous GPU resources in cloud desktops. This method detects characteristic information from different vendors and combines it with vendor GPU drivers to report, statistically analyze, allocate, and schedule heterogeneous GPU resources. Simultaneously, it combines general resource scheduling for CPU and memory to dynamically select resource nodes that meet user needs and mount computing power to the user's cloud desktop. Specifically:

[0051] S1, manage heterogeneous GPU resources;

[0052] S2, schedule heterogeneous GPU resources for cloud desktops.

[0053] The specific steps for managing heterogeneous GPU resources in step S1 of this embodiment are as follows:

[0054] S101. The computing service obtains the device type of all GPU computing cards on the current node through the system API, and matches and enables the corresponding vendor management driver according to the Vender ID of the GPU card.

[0055] S102: Different manufacturers' GPU management drivers obtain detailed resource information of the corresponding GPU computing cards from the nodes and report the resource computing power status to the resource management service.

[0056] S103, the resource management service compiles statistics on the GPU computing power consumption of each node, and provides feedback on the node information that meets the resource requirements based on the computing power request.

[0057] The specific steps for scheduling heterogeneous GPU resources for cloud desktops in step S2 of this embodiment are as follows:

[0058] S201. Users select a GPU computing power card that matches their business scenario from the management interface and mount it to the designated cloud desktop according to their business needs.

[0059] S202. The scheduling service checks whether the node where the cloud desktop machine is located meets the user's GPU resource requirements:

[0060] ① If the conditions are met, GPU resources will be allocated based on the current node;

[0061] ②If not, filter the compute nodes to see if there are any nodes that meet the requirements for GPU resources and migration resources, and return the optimal node;

[0062] S203. After receiving the node information from the scheduling service, the computing service first determines whether the current node has a GPU computing card that meets the computing power requirements:

[0063] ①If it exists, allocate and consume the specified GPU computing power resources from the current node;

[0064] ②If not, find a suitable node from the resource management service, migrate the user virtual machine to the specified node, and then enable the corresponding GPU driver resources;

[0065] S204. After the computing service migrates the cloud desktop to the optimal node, it mounts the corresponding GPU computing resources into the cloud desktop and restores the cloud desktop to its initial state.

[0066] The computing service in this embodiment supports multiple methods for detecting GPU devices, including system APIs and system command lines.

[0067] In this embodiment, different heterogeneous GPU computing resource management drivers are started simultaneously, and multiple GPU device cards exist on a single physical node at the same time.

[0068] In this embodiment, the user selects the GPU computing power information that meets the requirements according to their needs. The supported filtering parameters include GPU manufacturer information, GPU card name, GPU memory size, GPU operating mode and GPU type (physical card or virtual card). Multiple filtering parameters can be selected for precise matching.

[0069] In this embodiment, the GPU resource management driver identifies the detailed device information of the GPU computing card; wherein, the detailed device information of the GPU computing card includes GPU manufacturer information, GPU card name, GPU memory, number of GPUs, and GPU operating mode.

[0070] Example 2:

[0071] This embodiment provides a cloud desktop heterogeneous GPU resource scheduling and management system, which is used to implement the cloud desktop heterogeneous GPU resource scheduling and management method as described in Embodiment 1; the system includes:

[0072] The computing service module is used to detect the GPU device of the current system through the system interface and enable the corresponding vendor GPU driver according to the GPU card's Vendor ID.

[0073] The information acquisition and reporting module is used to obtain detailed resource information of the corresponding GPU card from the node through the GPU management driver of different manufacturers, and report the resource status to the resource management service.

[0074] The mounting module is used to allow users to select and mount suitable GPU cards from the management platform to the cloud desktop according to their business needs.

[0075] The scheduling service module is used to detect whether the node where the cloud desktop machine is located meets the user's GPU resource requirements.

[0076] If the conditions are met, GPU resources will be allocated based on the current node;

[0077] If not, filter the compute nodes to see if there are any nodes that meet the requirements for GPU resources and migration resources, and return the optimal node;

[0078] The judgment module is used to determine the node information received from the scheduling service through the computing service. If it is the current node, the corresponding GPU driver is enabled to perform resource operations. If it is not the current node, the user virtual machine is migrated to the specified node and the corresponding GPU driver resources are enabled.

[0079] The initial state recovery module is used to mount the user's desired computing power to the user's cloud desktop through the computing service and the corresponding GPU driver from the manufacturer, and restore the cloud desktop to its initial state.

[0080] Example 3:

[0081] This embodiment also provides an electronic device, including: a memory and a processor;

[0082] The memory stores the instructions executed by the computer.

[0083] The processor executes computer execution instructions stored in the memory, causing the processor to execute the cloud desktop heterogeneous GPU resource scheduling and management method in any embodiment of the present invention.

[0084] The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor can be a microprocessor or any conventional processor.

[0085] Memory is used to store computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory. Memory can mainly include a program storage area and a data storage area. The program storage area can store the operating system, at least one application program required for a function, etc.; the data storage area can store data created based on the use of the terminal, etc. In addition, memory can also include high-speed random access memory, and can also include non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart memory cards (SMC), secure digital cards (SD cards), flash memory cards, at least one disk storage device, flash memory devices, or other volatile solid-state storage devices.

[0086] Example 4:

[0087] This embodiment also provides a computer-readable storage medium storing multiple instructions, which are loaded by a processor to cause the processor to execute the cloud desktop heterogeneous GPU resource scheduling and management method in any embodiment of the present invention. Specifically, a system or apparatus equipped with a storage medium may be provided, on which software program code implementing the functions of any of the above embodiments is stored, and the computer (or CPU or MPU) of the system or apparatus can read and execute the program code stored in the storage medium.

[0088] In this case, the program code read from the storage medium can itself implement the function of any of the above embodiments, and therefore the program code and the storage medium storing the program code constitute part of the present invention.

[0089] Storage media embodiments for providing program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RYM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer via a communication network.

[0090] Furthermore, it should be clear that not only can the program code read by the computer be executed, but also the operating system or other components operating on the computer can be instructed based on the program code to perform some or all of the actual operations, thereby realizing the function of any of the embodiments described above.

[0091] Furthermore, it is understood that the program code read from the storage medium is written to the memory set in the expansion board inserted into the computer or to the memory set in the expansion unit connected to the computer. Then, based on the instructions of the program code, the CPU or other components installed on the expansion board or expansion unit execute some and all of the actual operations, thereby realizing the function of any of the embodiments described above.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for scheduling and managing heterogeneous GPU resources in cloud desktops, characterized in that, This method detects characteristic information from different manufacturers and combines it with manufacturer GPU drivers to report, statistically analyze, allocate, and schedule heterogeneous GPU resources. Simultaneously, it combines general resource scheduling for CPU and memory to dynamically select resource nodes that meet user needs and mount computing power to the user's cloud desktop; details are as follows: Managing heterogeneous GPU resources; Schedule heterogeneous GPU resources for cloud desktops.

2. The cloud desktop heterogeneous GPU resource scheduling and management method according to claim 1, characterized in that, The specific steps for managing heterogeneous GPU resources are as follows: The computing service obtains the device type of all GPU computing cards on the current node through the system API, and matches and enables the corresponding vendor management driver based on the GPU card's Vendor ID; Different manufacturers' GPU management drivers obtain detailed resource information of the corresponding GPU computing cards from the nodes and report the resource computing power status to the resource management service; The resource management service compiles statistics on the GPU computing power consumption of each node and, based on the computing power request, provides feedback on the node information that meets the resource requirements.

3. The cloud desktop heterogeneous GPU resource scheduling and management method according to claim 1, characterized in that, The specific steps for scheduling heterogeneous GPU resources on cloud desktops are as follows: Users can select a GPU computing card that matches their business scenario from the management interface and mount it to a designated cloud desktop based on their business needs. The scheduling service checks whether the node where the cloud desktop machine is located meets the user's GPU resource requirements: If the conditions are met, GPU resources will be allocated based on the current node; If not, filter the compute nodes to see if there are any nodes that meet the requirements for GPU resources and migration resources, and return the optimal node; After receiving node information from the scheduling service, the computing service first determines whether the current node has a GPU computing card that meets the computing power requirements: If it exists, allocate and consume the specified GPU computing resources from the current node; If not, a suitable node is found from the resource management service, and the user's virtual machine is migrated to the specified node before the corresponding GPU driver resources are enabled; After the computing service migrates the cloud desktop to the optimal node, it mounts the corresponding GPU computing resources into the cloud desktop and restores the cloud desktop to its initial state.

4. The cloud desktop heterogeneous GPU resource scheduling and management method according to claim 1, characterized in that, The computing service supports multiple methods for detecting GPU devices, including system APIs and system command lines.

5. The cloud desktop heterogeneous GPU resource scheduling and management method according to claim 1, characterized in that, Different heterogeneous GPU computing resource management drivers start simultaneously, and multiple GPU device cards exist on a single physical node at the same time.

6. The cloud desktop heterogeneous GPU resource scheduling and management method according to claim 1, characterized in that, Users can select the GPU computing power information that meets their needs. Supported filtering parameters include GPU manufacturer information, GPU card name, GPU memory size, GPU operating mode, and GPU type. Multiple filtering parameters can be selected for precise matching.

7. The cloud desktop heterogeneous GPU resource scheduling and management method according to any one of claims 1 to 6, characterized in that, The GPU resource management driver identifies detailed device information for the GPU computing card; this detailed device information includes GPU manufacturer information, GPU card name, GPU memory, number of GPUs, and GPU operating mode.

8. A cloud desktop heterogeneous GPU resource scheduling and management system, characterized in that, This system is used to implement the cloud desktop heterogeneous GPU resource scheduling and management method as described in any one of claims 1 to 7; the system includes: The computing service module is used to detect the GPU device of the current system through the system interface and enable the corresponding vendor GPU driver according to the VenderID of the GPU card. The information acquisition and reporting module is used to obtain detailed resource information of the corresponding GPU card from the node through the GPU management driver of different manufacturers, and report the resource status to the resource management service. The mounting module is used to allow users to select and mount suitable GPU cards from the management platform to the cloud desktop according to their business needs. The scheduling service module is used to detect whether the node where the cloud desktop machine is located meets the user's GPU resource requirements. If the conditions are met, GPU resources will be allocated based on the current node; If not, filter the compute nodes to see if there are any nodes that meet the requirements for GPU resources and migration resources, and return the optimal node; The judgment module is used to determine the node information received from the scheduling service through the computing service. If it is the current node, the corresponding GPU driver is enabled to perform resource operations. If it is not the current node, the user virtual machine is migrated to the specified node and the corresponding GPU driver resources are enabled. The initial state recovery module is used to mount the user's desired computing power to the user's cloud desktop through the computing service and the corresponding GPU driver from the manufacturer, and restore the cloud desktop to its initial state.

9. An electronic device, characterized in that, include: Memory and at least one processor; The memory contains computer programs; The at least one processor executes the computer program stored in the memory, causing the at least one processor to perform the cloud desktop heterogeneous GPU resource scheduling and management method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that can be executed by a processor to implement the cloud desktop heterogeneous GPU resource scheduling and management method as described in any one of claims 1 to 7.

Images