Resource scheduling method, system and device and computer readable storage medium

Abstract

The invention discloses a resource scheduling method, system and device and a computer readable storage medium, and relates to the technical field of computers, and the method comprises the steps: collecting hardware performance indexes and operation state data of a hyper-converged architecture; obtaining a to-be-processed AI task; predicting resource demand information of the to-be-processed AI task based on the hardware performance index and the operation state data; acquiring a set resource allocation constraint condition set; generating a resource scheduling instruction meeting a resource allocation constraint condition set based on the resource demand information; and performing resource scheduling on the hyper-converged architecture according to the resource scheduling instruction. Complete and dynamic intelligent resource scheduling from perception, prediction, decision making to execution is formed, refined and dynamic management of AI proxy resources under the heterogeneous hyper-converged architecture is achieved, the AI task execution efficiency and the service quality can be remarkably improved, continuous maximization of the resource utilization efficiency is achieved, and the resource utilization efficiency is improved. And the operation and maintenance complexity and the total cost of ownership can be reduced, and the performance of the hyper-converged architecture is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more specifically, to a resource scheduling method, system, device, and computer-readable storage medium. Background Technology

[0002] With the development of AI (Artificial Intelligence) technology, the demand for high-performance GPUs (Graphics Processing Units), high-bandwidth storage, and elastic scheduling for deep learning tasks has exploded.

[0003] However, current hyper-converged infrastructure (HCI) architectures have significant limitations when dealing with AI workloads: resource scheduling mechanisms lack dynamic adaptability, the integration efficiency of heterogeneous computing resources such as CPUs (Central Processing Units), GPUs, and NPUs (Neural Processing Units) is low, storage I / O performance becomes a critical bottleneck, and security isolation mechanisms are insufficient. These problems directly lead to low average GPU utilization, underoptimized vGPU allocation strategies, persistently high storage access latency, and increased data security risks. Existing solutions are often limited to single-dimensional optimization, resulting in performance constraints for hyper-converged infrastructures.

[0004] In conclusion, improving the performance of hyperconverged architectures is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] The purpose of this application is to provide a resource scheduling method that can, to some extent, address the technical problem of how to improve the performance of hyperconverged architectures. This application also provides a resource scheduling system, an electronic device, and a computer-readable storage medium.

[0006] To achieve the above objectives, this application provides the following technical solution: A resource scheduling method, comprising: Collect hardware performance metrics and operational status data of the hyperconverged architecture; Acquire AI tasks to be processed; Based on the hardware performance indicators and the operating status data, predict the resource requirements of the AI ​​task to be processed; Obtain the set of resource allocation constraints; Based on the resource demand information, generate resource scheduling instructions that satisfy the set of resource allocation constraints; The hyperconverged architecture is scheduled according to the resource scheduling instructions.

[0007] In an exemplary embodiment, predicting the resource requirements of the AI ​​task to be processed based on the hardware performance metrics and the operating status data includes: The hardware performance indicators, the operating status data, and the AI ​​task to be processed are input into the pre-trained machine learning model. Obtain the resource requirement information output by the machine learning model.

[0008] In an exemplary embodiment, after performing resource scheduling on the hyperconverged architecture according to the resource scheduling instruction, the method further includes: Collect the operation result data of the hyperconverged architecture, including task completion time, peak resource utilization, and service level agreement achievement rate; The machine learning model is retrained and fine-tuned based on the aforementioned operational results data; The process of generating the resource scheduling instructions is optimized based on the aforementioned operational result data; The resource scheduling process is optimized based on the aforementioned operational results data.

[0009] In an exemplary embodiment, predicting the resource requirements of the AI ​​task to be processed based on the hardware performance metrics and the operating status data includes: The hardware performance metrics and the running status data are processed to obtain a dynamic global view of the resource pool, and resource profiles are generated for physical nodes and virtual instances. Based on the dynamic global view and the resource profile, the resource requirements of the AI ​​task to be processed are predicted.

[0010] In an exemplary embodiment, the collection of hardware performance metrics and operational status data of the hyperconverged architecture includes: Collect hardware performance metrics of the hyperconverged architecture, including CPU utilization, memory usage, GPU memory and computing power, storage IOPS and latency; Collect runtime status data of virtual machines and / or containers in the hyperconverged architecture.

[0011] In an exemplary embodiment, obtaining the set of resource allocation constraints includes: Obtain the set of resource allocation constraints, which includes task priority, task deadline, resource isolation requirements, GPU passthrough or virtualization selection strategy, storage performance level, and network security strategy.

[0012] In an exemplary embodiment, the step of scheduling resources for the hyperconverged architecture according to the resource scheduling instruction includes: Create or adjust virtual machine instances in the hyperconverged architecture according to the resource scheduling instructions; According to the resource scheduling instructions, the physical GPU card is allocated to the target virtual machine in either pass-through mode or vGPU mode; According to the resource scheduling instructions, create and mount a distributed storage volume that meets the performance requirements; Configure network security group policies according to the resource scheduling instructions.

[0013] A resource scheduling system, comprising: The resource status awareness module is used to collect hardware performance indicators and operational status data of the hyperconverged architecture. The task acquisition module is used to acquire AI tasks to be processed. The task requirement prediction module is used to predict the resource requirement information of the AI ​​task to be processed based on the hardware performance indicators and the running status data. The constraint acquisition module is used to acquire the set of resource allocation constraints. The resource scheduling instruction generation module is used to generate resource scheduling instructions that satisfy the resource allocation constraint set based on the resource demand information. The resource scheduling module is used to schedule resources for the hyperconverged architecture according to the resource scheduling instructions.

[0014] An electronic device, comprising: Memory, used to store computer programs; A processor, configured to implement the steps of any of the above-described resource scheduling methods when executing the computer program.

[0015] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the resource scheduling methods described above.

[0016] This application provides a resource scheduling method that collects hardware performance indicators and operational status data of a hyperconverged infrastructure (HCI) architecture; acquires AI tasks to be processed; predicts the resource requirements of the AI ​​tasks based on the hardware performance indicators and operational status data; acquires a set of set resource allocation constraints; generates resource scheduling instructions that satisfy the resource allocation constraints based on the resource requirements; and schedules resources for the HCI architecture according to the resource scheduling instructions. In this application, data acquisition enables the perception of hardware performance indicators, operational status data, and AI tasks to be processed within the HCI architecture, as well as the prediction of resource requirements. The resource allocation constraint set enables the decision-making regarding resource scheduling instructions, and finally, resource scheduling enables the execution of these instructions. This transforms static, rule-based resource management into a complete, dynamic, and intelligent resource scheduling process from perception, prediction, decision-making to execution. It achieves refined and dynamic management of AI agent resources under a heterogeneous HCI architecture, significantly improving AI task execution efficiency and service quality, continuously maximizing resource utilization efficiency, and effectively reducing operational complexity and total cost of ownership while ensuring high performance, thus enhancing the performance of the HCI architecture. The resource scheduling system, electronic device, and computer-readable storage medium provided in this application also solve the corresponding technical problems. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a resource scheduling method provided in an embodiment of this application; Figure 2 This is a diagram illustrating the overall architecture of the resource scheduling method. Figure 3 This is a schematic diagram of the structure of a resource scheduling system provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application; Figure 5 This is another structural schematic diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] Please see Figure 1 , Figure 1 This is a flowchart of a resource scheduling method provided in an embodiment of this application.

[0021] This application provides a resource scheduling method that may include the following steps: Step S101: Collect hardware performance metrics and operational status data of the hyperconverged architecture.

[0022] In practical applications, hardware performance metrics and operational status data that reflect the resource status of the hyperconverged infrastructure (HCI) can be collected first, so that resource scheduling can be performed on the HCI accordingly. The type of HCI can be flexibly determined as needed; for example, it can be a homogeneous heterogeneous HCI architecture.

[0023] In the exemplary embodiment, hardware performance metrics and operational status data can be flexibly set as needed. For example, during the process of collecting hardware performance metrics and operational status data of the hyperconverged architecture, monitoring agents deployed on each node (computing, storage, network) of the hyperconverged cluster can collect hardware performance metrics of the hyperconverged architecture in real time. Hardware performance metrics include CPU utilization, memory usage, GPU memory and computing power, storage IOPS and latency, etc.; and collect operational status data of virtual machines and / or containers in the hyperconverged architecture.

[0024] Step S102: Obtain the AI ​​task to be processed.

[0025] In practical applications, the resources of a hyperconverged infrastructure (HCI) serve task execution. Therefore, it is necessary to acquire the AI ​​tasks to be processed so that resources can be allocated to the HCI structure accordingly. The types of AI tasks to be processed can be AI training tasks or AI inference tasks, and the content of the AI ​​tasks can include task type, framework, code size, etc.

[0026] Step S103: Based on hardware performance indicators and operating status data, predict the resource requirements of the AI ​​task to be processed.

[0027] In practical applications, when scheduling resources for a hyperconverged infrastructure according to the AI ​​tasks to be processed, the resource requirements of the AI ​​tasks can be predicted based on hardware performance metrics and operational status data. This resource requirement information reflects the resource needs of the AI ​​tasks. Specifically, historical task execution data and the real-time cluster load at the time the AI ​​task is submitted can be obtained. Based on hardware performance metrics, operational status data, historical task execution data, and real-time cluster load, the resource requirements of the AI ​​task throughout its entire lifecycle, especially during peak phases in the training iteration process, can be predicted. The content of the resource requirement information can be flexibly determined according to actual needs; for example, the resource requirement information may include CPU, memory, and especially GPU computing power and storage bandwidth.

[0028] In an exemplary embodiment, during the process of predicting the resource requirements of the AI ​​task to be processed based on hardware performance indicators and operational status data, the hardware performance indicators and operational status data can be processed to obtain a dynamic global view of the resource pool, and resource profiles can be generated for physical nodes and virtual instances. For example, after cleaning and aggregating the hardware performance indicators and operational status data, big data technology can be used to process this multi-dimensional, time-series data to construct a dynamic global view of the entire resource pool, and generate a refined, real-time updated resource profile for each physical node and virtual instance. Based on the dynamic global view and resource profiles, the resource requirements of the AI ​​task to be processed are predicted. In this way, converting hardware performance indicators and operational status data into a centralized and visualized dynamic global view and resource profile makes it easier to predict resource requirements compared to fragmented hardware performance indicators and operational status data, thus improving the efficiency of resource requirements prediction.

[0029] Step S104: Obtain the set of resource allocation constraints.

[0030] Step S105: Based on resource demand information, generate resource scheduling instructions that satisfy the set of resource allocation constraints.

[0031] In practical applications, during the process of scheduling resources for a hyperconverged architecture according to the AI ​​tasks to be processed, a set of resource allocation constraints can be obtained. Based on the resource demand information, resource scheduling instructions that meet the set of resource allocation constraints can be generated, so that the resource scheduling process is restricted by the set of resource allocation constraints.

[0032] In an exemplary embodiment, the content of the resource allocation constraint set can be flexibly determined according to actual needs. For example, during the process of obtaining the set of resource allocation constraints, the set of resource allocation constraints can be obtained, including task priority, task deadline, resource isolation requirements, GPU passthrough or virtualization selection strategy, storage performance level, and network security strategy. Correspondingly, during the process of generating resource scheduling instructions that satisfy the resource allocation constraint set based on resource demand information, various resource allocation requirements, such as task urgency, completion time, whether resources need to be isolated, which GPU mode to use, and which storage to use, can be combined with the resource allocation demand information. Based on this combination, mathematical modeling and optimization are performed to repeatedly calculate and find the optimal solution within the range of satisfying all constraints, thereby generating resource scheduling instructions. This clarifies how to allocate virtual machines, partition GPUs, and connect storage, ensuring that the resource scheduling instructions can be implemented efficiently.

[0033] In specific application scenarios, during the process of generating resource scheduling instructions that satisfy a set of resource allocation constraints based on resource demand information, the following steps can be taken: First, construct decision variables corresponding to the resource demand information and the set of resource allocation constraints. Then, convert the resource demand information and resource allocation constraints into objective functions and constraint functions of equations. Finally, solve the objective functions and constraint functions using the interior-point method to obtain specific resource scheduling instructions that include specific virtual machine / container configurations, GPU card allocations, storage volume mappings, and network access policies. In this way, resource scheduling instructions that satisfy a set of resource allocation constraints can be generated based on resource demand information using only simple variable generation, function generation, and interior-point method solving, without the need for complex operations, thus improving the efficiency and convenience of resource scheduling instruction generation.

[0034] Step S106: Perform resource scheduling on the hyperconverged architecture according to the resource scheduling instructions.

[0035] In practical applications, once a resource scheduling instruction is received, resources can be scheduled for the hyperconverged architecture according to the instruction.

[0036] In an exemplary embodiment, a model can be used to predict resource requirements in order to improve prediction efficiency. That is, in the process of predicting the resource requirements of the AI ​​task to be processed based on hardware performance indicators and operating status data, the hardware performance indicators, operating status data and the AI ​​task to be processed can be input into a pre-trained machine learning model; and the resource requirements output by the machine learning model can be obtained.

[0037] In specific application scenarios, the accuracy of resource demand prediction, resource scheduling instruction generation, and the resource scheduling process all affect the operational performance of the hyperconverged infrastructure (HCI). If these processes are inaccurate, the performance of the HCI will deteriorate. Therefore, to ensure the performance of the HCI, these processes need to be optimized. This means that after scheduling resources according to the resource scheduling instructions, the operational results data of the HCI can be collected, including task completion time, peak resource utilization, and service level agreement (SLA) achievement rate. The machine learning model can then be retrained and fine-tuned based on the operational results data. The generation process of resource scheduling instructions can also be optimized based on the operational results data, such as adjusting the weights of constraints in the resource allocation constraint set to change the generated resource scheduling instructions. Finally, the resource scheduling process can be optimized based on the operational results data. In this way, an intelligent closed-loop system of "monitoring-prediction-decision-execution-feedback-optimization" can be constructed. Figure 2 As shown, it achieves efficient, intelligent, and secure scheduling of AI tasks. It can not only accurately predict task requirements and optimize scheduling strategies to significantly improve the execution efficiency and service quality of AI tasks, but also has powerful self-learning and adaptive capabilities. It can continuously iterate and optimize its own model and algorithm through continuous feedback, and can continuously adapt to the ever-changing workload and cluster environment, thereby continuously maximizing the efficiency of resource utilization.

[0038] In the exemplary embodiment, the content of the resource scheduling instructions and the corresponding resource scheduling process can be configured as needed. For example, during the resource scheduling process of the hyperconverged architecture according to the resource scheduling instructions, the standardized API interface of the hyperconverged management platform can be called, or direct interaction with the underlying driver can be performed to create or adjust virtual machine instances in the hyperconverged architecture according to the resource scheduling instructions; allocate physical GPU cards to the target virtual machine through passthrough mode or vGPU mode according to the resource scheduling instructions; create and mount distributed storage volumes that meet performance requirements according to the resource scheduling instructions; configure network security group policies according to the resource scheduling instructions; and perform task migration of the hyperconverged architecture according to the resource scheduling, etc.

[0039] This application provides a resource scheduling method that collects hardware performance indicators and operational status data of a hyperconverged infrastructure (HCI) architecture; acquires AI tasks to be processed; predicts the resource requirements of the AI ​​tasks based on the hardware performance indicators and operational status data; acquires a set of set resource allocation constraints; generates resource scheduling instructions that satisfy the resource allocation constraints based on the resource requirements; and schedules resources for the HCI architecture according to the resource scheduling instructions. In this application, data acquisition enables the perception of hardware performance indicators, operational status data, and AI tasks to be processed within the HCI architecture, as well as the prediction of resource requirements. The resource allocation constraint set enables the decision-making regarding resource scheduling instructions, and finally, resource scheduling enables the execution of these instructions. This transforms static, rule-based resource management into a complete, dynamic, and intelligent resource scheduling process from perception, prediction, decision-making to execution. It achieves refined and dynamic management of AI agent resources under a heterogeneous HCI architecture, significantly improving AI task execution efficiency and service quality, continuously maximizing resource utilization efficiency, and effectively reducing operational complexity and total cost of ownership while ensuring high performance, thus enhancing the performance of the HCI architecture.

[0040] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of a resource scheduling system provided in an embodiment of this application.

[0041] This application provides a resource scheduling system that may include: The resource status awareness module 101 is used to collect hardware performance indicators and operating status data of the hyperconverged architecture. Task acquisition module 102 is used to acquire AI tasks to be processed; The task requirement prediction module 103 is used to predict the resource requirements of the AI ​​task to be processed based on hardware performance indicators and operating status data. The constraint acquisition module 104 is used to acquire the set of resource allocation constraints. The resource scheduling instruction generation module 105 is used to generate resource scheduling instructions that meet the resource allocation constraint set based on resource demand information. The resource scheduling module 106 is used to schedule resources for the hyperconverged architecture according to resource scheduling instructions.

[0042] This application provides a resource scheduling system in which a task demand prediction module may include: The input unit is used to input hardware performance indicators, operating status data, and AI tasks to be processed into a pre-trained machine learning model. The acquisition unit is used to acquire resource requirement information output by the machine learning model.

[0043] The resource scheduling system provided in this application embodiment may further include: The operation result acquisition module is used to collect the operation result data of the hyperconverged architecture after the resource scheduling module performs resource scheduling on the hyperconverged architecture according to the resource scheduling instructions. The operation result data includes task completion time, peak resource utilization, and service level agreement achievement rate. The adjustment module is used to retrain and fine-tune the machine learning model based on the running result data; optimize the generation process of resource scheduling instructions based on the running result data; and optimize the resource scheduling process based on the running result data.

[0044] This application provides a resource scheduling system in which a task demand prediction module may include: The profile generation unit is used to process hardware performance indicators and operating status data to obtain a dynamic global view of the resource pool and generate resource profiles for physical nodes and virtual instances. The task requirement prediction unit is used to predict the resource requirements of the AI ​​tasks to be processed based on a dynamic global view and resource profile.

[0045] This application provides a resource scheduling system in which the task status awareness module may include: The hardware performance metrics acquisition unit is used to collect hardware performance metrics of the hyperconverged architecture, including CPU utilization, memory usage, GPU memory and computing power, storage IOPS and latency. The runtime status data acquisition unit is used to collect runtime status data of virtual machines and / or containers in a hyperconverged architecture.

[0046] This application provides a resource scheduling system in which a constraint acquisition module may include: The constraint acquisition unit is used to acquire the set of resource allocation constraints, which include task priority, task deadline, resource isolation requirements, GPU passthrough or virtualization selection strategy, storage performance level, and network security strategy.

[0047] This application provides a resource scheduling system. The resource scheduling module can be used to: create or adjust virtual machine instances in a hyperconverged architecture according to resource scheduling instructions; allocate physical GPU cards to target virtual machines through passthrough mode or vGPU mode according to resource scheduling instructions; create and mount distributed storage volumes that meet performance requirements according to resource scheduling instructions; and configure network security group policies according to resource scheduling instructions.

[0048] This application also provides an electronic device and a computer-readable storage medium, both of which have the corresponding effects of the resource scheduling method provided in the embodiments of this application. Please refer to... Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0049] An electronic device provided in this application includes a memory 201 and a processor 202. The memory 201 stores a computer program, and the processor 202 executes the computer program to implement the steps of the resource scheduling method described in any of the above embodiments.

[0050] Please see Figure 5 Another electronic device provided in this application embodiment may further include: an input port 203 connected to the processor 202 for transmitting commands input from the outside to the processor 202; a display unit 204 connected to the processor 202 for displaying the processing results of the processor 202 to the outside; and a communication module 205 connected to the processor 202 for enabling communication between the electronic device and the outside. The display unit 204 may be a display panel, a laser scanner, or the like; the communication method used by the communication module 205 includes, but is not limited to, Mobile High-Definition Link (MHL), Universal Serial Bus (USB), High-Definition Multimedia Interface (HDMI), wireless connectivity: Wireless Fidelity (WiFi), Bluetooth communication technology, Bluetooth Low Energy communication technology, and communication technology based on IEEE 802.11s.

[0051] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of the resource scheduling method described in any of the above embodiments.

[0052] The computer-readable storage media involved in this application include random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs (compact disc read-only memory), or any other form of storage media known in the art.

[0053] This application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the resource scheduling method described in any of the above embodiments.

[0054] For descriptions of relevant parts in the resource scheduling system, electronic device, and computer-readable storage medium provided in this application's embodiments, please refer to the detailed description of the corresponding parts in the resource scheduling method provided in this application's embodiments; they will not be repeated here. Furthermore, parts of the technical solutions provided in this application that are consistent with the implementation principles of corresponding technical solutions in the prior art have not been described in detail to avoid excessive elaboration.

[0055] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0056] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A resource scheduling method, characterized in that, include: Collect hardware performance metrics and operational status data of the hyperconverged architecture; Acquire AI tasks to be processed; Based on the hardware performance indicators and the operating status data, predict the resource requirements of the AI ​​task to be processed; Obtain the set of resource allocation constraints; Based on the resource demand information, generate resource scheduling instructions that satisfy the set of resource allocation constraints; The hyperconverged architecture is scheduled according to the resource scheduling instructions.

2. The resource scheduling method according to claim 1, characterized in that, The process of predicting the resource requirements of the AI ​​task to be processed based on the hardware performance indicators and the operating status data includes: The hardware performance indicators, the operating status data, and the AI ​​task to be processed are input into the pre-trained machine learning model. Obtain the resource requirement information output by the machine learning model.

3. The resource scheduling method according to claim 2, characterized in that, After performing resource scheduling on the hyperconverged architecture according to the resource scheduling instructions, the method further includes: Collect the operation result data of the hyperconverged architecture, including task completion time, peak resource utilization, and service level agreement achievement rate; The machine learning model is retrained and fine-tuned based on the aforementioned operational results data; The process of generating the resource scheduling instructions is optimized based on the aforementioned operational result data; The resource scheduling process is optimized based on the aforementioned operational results data.

4. The resource scheduling method according to claim 1, characterized in that, The process of predicting the resource requirements of the AI ​​task to be processed based on the hardware performance indicators and the operating status data includes: The hardware performance metrics and the running status data are processed to obtain a dynamic global view of the resource pool, and resource profiles are generated for physical nodes and virtual instances. Based on the dynamic global view and the resource profile, the resource requirements of the AI ​​task to be processed are predicted.

5. The resource scheduling method according to claim 1, characterized in that, The collected hardware performance metrics and operational status data of the hyperconverged architecture include: Collect hardware performance metrics of the hyperconverged architecture, including CPU utilization, memory usage, GPU memory and computing power, storage IOPS and latency; Collect runtime status data of virtual machines and / or containers in the hyperconverged architecture.

6. The resource scheduling method according to claim 1, characterized in that, The process of obtaining the set of resource allocation constraints includes: Obtain the set of resource allocation constraints, which includes task priority, task deadline, resource isolation requirements, GPU passthrough or virtualization selection strategy, storage performance level, and network security strategy.

7. The method according to claim 1, characterized in that, The step of scheduling resources for the hyperconverged architecture according to the resource scheduling instructions includes: Create or adjust virtual machine instances in the hyperconverged architecture according to the resource scheduling instructions; According to the resource scheduling instructions, the physical GPU card is allocated to the target virtual machine in either pass-through mode or vGPU mode; According to the resource scheduling instructions, create and mount a distributed storage volume that meets the performance requirements; Configure network security group policies according to the resource scheduling instructions.

8. A resource scheduling system, characterized in that, include: The resource status awareness module is used to collect hardware performance indicators and operational status data of the hyperconverged architecture. The task acquisition module is used to acquire AI tasks to be processed. The task requirement prediction module is used to predict the resource requirement information of the AI ​​task to be processed based on the hardware performance indicators and the running status data. The constraint acquisition module is used to acquire the set of resource allocation constraints. The resource scheduling instruction generation module is used to generate resource scheduling instructions that satisfy the resource allocation constraint set based on the resource demand information. The resource scheduling module is used to schedule resources for the hyperconverged architecture according to the resource scheduling instructions.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the resource scheduling method as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the resource scheduling method as described in any one of claims 1 to 7.

Images