AI computing power resource management method and device, equipment and storage medium

Abstract

The invention discloses an AI computing power resource management method and device, equipment and a storage medium, and relates to the field of artificial intelligence, and the method comprises the steps: recognizing the resource information of a target AI computing power resource, classifying the target AI computing power resource through the resource information of the target AI computing power resource, so as to determine a classification result, and building a target resource directory according to the classification result; abstraction processing is carried out on the target AI computing power resource based on the target resource directory to obtain a target logic resource, then the computing power use request is analyzed, and the target AI computing power resource is allocated according to an analysis result; and monitoring a resource use condition of the target logic resource, if the resource use condition represents that the state of the target AI computing power resource is abnormal, performing a preset evaluation operation on the target AI computing power resource to obtain an evaluation result, and performing a preset repair operation on the target AI computing power resource according to the evaluation result to complete AI computing power resource management. Unified perception and management of AI computing power resources can be realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to an AI computing resource management method, device, equipment and storage medium. Background Art

[0002] With the rapid development of artificial intelligence technology, the demand for underlying computing power has exploded. In this context, AI (Artificial Intelligence) computing power resource pooling management is particularly important. On the one hand, AI computing power resource pooling can improve the utilization rate of computing power resources and avoid resource waste. For example, in 2018, the average utilization rate of GPU (Graphics Processing Unit) products was only 10-30%, and the utilization rate of many users was even less than 10%. On the other hand, resource pooling can achieve flexible resource scheduling and management to meet the needs of different application scenarios.

[0003] AI computing power resource pooling management technology has gone through several stages of evolution. Initially, it was a simple virtualization technology that divided the physical GPU into multiple virtual GPUs in equal proportions, and each virtual GPU had the same computing power and video memory. Later, it developed into the arbitrary virtualization stage, where GPU resources can be customized according to needs to meet the personalized needs of different AI models. Next came the remote call technology, which allowed AI applications to be deployed separately from physical GPU servers and remotely call GPU resources over the network. Finally, it developed into the resource pooling stage, using a unified management interface to achieve resource monitoring, scheduling, and recycling, as well as docking with the data center-level resource scheduling platform.

[0004] To sum up, how to achieve unified perception and management of AI computing resources is an urgent problem to be solved. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide an AI computing resource management method, device, equipment and storage medium, which can realize the unified perception and management of AI computing resources. The specific scheme is as follows:

[0006] In a first aspect, the present application provides an AI computing resource management method, including:

[0007] Identify resource information of the target AI computing power resources, classify the target AI computing power resources using the identified resource information of the target AI computing power resources to determine a classification result corresponding to the target AI computing power resources, and establish a corresponding target resource directory according to the classification result;

[0008] Based on the target resource directory, the target AI computing power resource is abstracted to obtain a corresponding target logical resource, and then the obtained computing power usage request is analyzed, and the target AI computing power resource corresponding to the target logical resource is allocated according to the obtained analysis result;

[0009] Monitor the resource usage of the target logical resources. If the resource usage indicates that the state of the target AI computing power resources is abnormal, perform a preset evaluation operation on the target AI computing power resources to obtain a corresponding evaluation result. Perform a preset repair operation on the target AI computing power resources according to the evaluation result to complete AI computing power resource management.

[0010] Optionally, the classifying the target AI computing power resources by using the identified resource information of the target AI computing power resources includes:

[0011] Obtain resource information of the identified target AI computing resource; wherein the resource information includes any one or more of performance parameters, resource types, and applicable scenarios;

[0012] Using the performance parameters of the target AI computing resources, the target AI computing resources are classified into preset levels to determine AI computing resources that meet various preset level conditions from the target AI computing resources;

[0013] Classifying the target AI computing power resource by using the resource type of the target AI computing power resource to determine a classification result of the target AI computing power resource obtained based on the resource type classification; wherein the resource type includes any one or more of computing resources, storage resources, and network resources;

[0014] The target AI computing power resources are classified according to the applicable scenarios of the target AI computing power resources to determine the classification results of the target AI computing power resources obtained based on the applicable scenario classification; wherein the applicable scenarios include any one or more of the preset data analysis scenarios, the preset data reasoning scenarios, and the preset model training scenarios.

[0015] Optionally, allocating the target AI computing power resource corresponding to the target logical resource according to the obtained analysis result includes:

[0016] Allocate the target AI computing resources corresponding to the target logical resources using a preset priority scheduling algorithm based on the AI ​​computing resources that meet the preset level conditions;

[0017] Based on the classification result corresponding to the resource type of the target AI computing power resource, the target AI computing power resource corresponding to the target logical resource is allocated using a preset load balancing scheduling algorithm;

[0018] The preset predictive scheduling algorithm is used to analyze the historical resource allocation to determine the corresponding predicted task requirements, and the target AI computing power resources corresponding to the target logical resources are allocated based on the predicted task requirements.

[0019] Optionally, after abstracting the target AI computing power resource based on the target resource directory to obtain the corresponding target logical resource, the method further includes:

[0020] Creating a preset virtual hardware environment for the first target object based on the target AI computing power resources; wherein creating the preset virtual hardware environment includes allocating a virtual CPU, mounting a virtual GPU, dividing memory, setting storage, and any one or more of network resources;

[0021] Based on the target AI computing power resources, a preset container resource space is divided for the second target object; wherein the preset container resource space includes any one or more of CPU quota, memory quota, file system and network namespace.

[0022] Optionally, the computing power usage request includes any one or more of a preset computing complexity usage request, a preset time usage request, and a preset data volume usage request.

[0023] Optionally, the method further includes:

[0024] Using a preset identity authentication system to identify the identity information of the object to be authorized; wherein the object to be authorized includes any one or more of the user terminal to be authorized, the device to be authorized, and the application to be authorized;

[0025] If the obtained identification result indicates that the object to be authorized meets the preset identity security condition, the preset access rights corresponding to the target logical resource are opened to the object to be authorized.

[0026] Optionally, the step of using a preset identity authentication system to identify the identity information of the object to be authorized includes:

[0027] Identify the identity information of the user terminal to be authorized by using a preset multi-factor authentication to determine whether the obtained identity information of the user terminal to be authorized meets a first preset identity security condition;

[0028] Identifying the identity information of the device to be authorized by detecting the target identifier and the security certificate of the device to be authorized, so as to determine whether the acquired identity information of the device to be authorized meets a second preset identity security condition;

[0029] The identity information of the application to be authorized is identified by verifying the digital signature of the application to be authorized and the preset authority, so as to determine whether the obtained identity information of the application to be authorized meets the third preset identity security condition.

[0030] In a second aspect, the present application provides an AI computing resource management device, including:

[0031] A directory establishment module, used to identify resource information of a target AI computing resource, classify the target AI computing resource using the identified resource information of the target AI computing resource to determine a classification result corresponding to the target AI computing resource, and establish a corresponding target resource directory according to the classification result;

[0032] A resource allocation module, configured to abstract the target AI computing power resources based on the target resource directory to obtain corresponding target logical resources, and then analyze the acquired computing power usage request, and allocate the target AI computing power resources corresponding to the target logical resources according to the obtained analysis results;

[0033] The resource management completion module is used to monitor the resource usage of the target logical resources. If the resource usage indicates that the state of the target AI computing power resource is abnormal, a preset evaluation operation is performed on the target AI computing power resource to obtain a corresponding evaluation result. According to the evaluation result, a preset repair operation is performed on the target AI computing power resource to complete the AI ​​computing power resource management.

[0034] In a third aspect, the present application provides an electronic device, including:

[0035] Memory, used to store computer programs;

[0036] A processor is used to execute the computer program to implement the AI ​​computing power resource management method as described above.

[0037] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the AI ​​computing power resource management method as described above.

[0038] In summary, the present application first identifies the resource information of the target AI computing power resources, and uses the identified resource information of the target AI computing power resources to classify the target AI computing power resources to determine the classification results corresponding to the target AI computing power resources, and establishes a corresponding target resource directory based on the classification results; abstracts the target AI computing power resources based on the target resource directory to obtain the corresponding target logical resources, and then analyzes the acquired computing power usage request, and allocates the target AI computing power resources corresponding to the target logical resources according to the obtained analysis results; monitors the resource usage of the target logical resources, and if the resource usage indicates that the state of the target AI computing power resources is abnormal, performs a preset evaluation operation on the target AI computing power resources to obtain the corresponding evaluation results, and performs a preset repair operation on the target AI computing power resources according to the evaluation results to complete AI computing power resource management. As can be seen from the above, the present application first identifies the resource information of the target AI computing power resources, classifies them according to the information to obtain the classification results, and establishes a target resource directory based on this. Then, the target AI computing power resources in the directory are abstracted to obtain the target logical resources, the obtained computing power usage requests are analyzed, and the target AI computing power resources corresponding to the target logical resources are allocated according to the analysis results. Subsequently, the resource usage of the target logical resources is monitored. If the resource usage indicates that the state of the target AI computing power resources is abnormal, a preset evaluation operation is performed on it to obtain an evaluation result, and then a preset repair operation is performed on the target AI computing power resources based on the evaluation result, thereby completing the AI ​​computing power resource management. In this way, this application realizes unified perception and management of AI computing power resources by establishing a target resource directory, thereby improving the flexibility and convenience of managing diverse AI computing power devices, reducing the complexity of adaptation and use of AI computing power devices, and improving the universality of AI applications running on AI computing power devices, providing strong resource support for the development of artificial intelligence applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0040] Figure 1 A schematic diagram of a core module of AI computing resource pooling disclosed in this application;

[0041] Figure 2 This is a functional schematic diagram of a resource management module disclosed in this application;

[0042] Figure 3This is a flow chart of an AI computing resource management method disclosed in this application;

[0043] Figure 4 A flowchart of a specific AI computing resource management method disclosed in this application;

[0044] Figure 5 This is a schematic diagram of the structure of an AI computing resource management device disclosed in this application;

[0045] Figure 6 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] At present, AI computing power resource pooling management technology has developed to the stage of arbitrary virtualization, and GPU resources can be customized and segmented according to needs to meet the personalized needs of different AI models. Next is the remote call technology, which allows AI applications to be deployed separately from physical GPU servers and remotely call GPU resources through the network. Eventually it will develop to the resource pooling stage, and resource monitoring, scheduling and recycling will be achieved through a unified management interface, as well as docking with the data center-level resource scheduling platform. In order to solve the above technical problems, the present application discloses an AI computing power resource management method, device, equipment and storage medium, which can realize unified perception and management of AI computing power resources.

[0048] The embodiment of the present invention discloses an AI computing resource management method, see Figure 1 As shown, the AI ​​computing power resource management method can be divided into a resource management module, a virtualization module, a monitoring module, a security module, and a scheduling module. The combination of the five modules realizes the pooled management of AI computing power infrastructure resources, which meets the needs of improving AI resource utilization, reducing costs, and improving system flexibility and scalability.

[0049] Specifically, the resource management module is as follows Figure 2As shown, it is responsible for the perception registration, resource integration, classification and directory management of heterogeneous AI computing resources, and provides computing resource information to the upper level; the virtualization module is responsible for virtualizing physical computing resources into logical resources, realizing resource abstraction and isolation, and meeting the virtualization usage scenarios of resource types based on virtual machines and container types; the scheduling module is responsible for dynamic resource allocation based on the computing power usage tasks requested by the upper-level business; the monitoring module is responsible for real-time monitoring of resource usage, status and performance indicators; the security module is responsible for the access control of the resource security management mechanism to ensure the security and reliability of the resource pool.

[0050] See also Figure 3 As shown, an embodiment of the present invention discloses an AI computing resource management method, which may include:

[0051] Step S11: Identify the resource information of the target AI computing power resources, classify the target AI computing power resources using the identified resource information of the target AI computing power resources to determine the classification results corresponding to the target AI computing power resources, and establish a corresponding target resource directory according to the classification results.

[0052] In this embodiment, various types of heterogeneous AI computing resources are first detected and identified, including special computing units such as different types of CPU (Central Processing Unit), GPU, NPU (network processor), FPGA (Field Programmable Gate Array), etc., which can discover the computing resources of the device through monitoring technology and algorithms, and register their detailed resource information. This process ensures that the status of all available computing resources can be grasped in a timely manner, providing accurate basic data for subsequent resource allocation and scheduling. At the same time, resource integration of heterogeneous computing resources of different manufacturers and different performance specifications can also effectively integrate these scattered resources. By establishing a unified resource management framework, various types of computing resources are included, the heterogeneity between resources is eliminated, and different types of resources can work together to provide strong support for AI computing tasks. In addition, for devices with both high-performance GPUs and general-purpose CPUs, the resource management module can reasonably allocate the usage ratio of the two resources according to task requirements, give full play to their respective advantages, and improve overall computing efficiency.

[0053] In this embodiment, after obtaining the resource information of the identified target AI computing power resources; wherein the resource information includes any one or more of performance parameters, resource types, and applicable scenarios; the target AI computing power resources are classified into preset levels using the performance parameters of the target AI computing power resources to determine the AI ​​computing power resources that meet the preset level conditions from the target AI computing power resources; the target AI computing power resources are classified using the resource types of the target AI computing power resources to determine the classification results of the target AI computing power resources obtained based on the resource type classification; wherein the resource types include any one or more of computing resources, storage resources, and network resources; the target AI computing power resources are classified using the applicable scenarios of the target AI computing power resources to determine the classification results of the target AI computing power resources obtained based on the applicable scenario classification; wherein the applicable scenarios include any one or more of preset data analysis scenarios, preset data reasoning scenarios, and preset model training scenarios. Specifically, the target AI computing power resources can be classified in detail based on multiple dimensions such as performance parameters of computing power resources, resource types, and applicable scenarios. According to the classification of performance parameters, the target AI computing resources are evaluated with reference to the computing power, storage capacity, bandwidth and other indicators of the target AI computing resources, and the target AI computing resources are divided into high, medium and low performance levels. According to the classification of resource types, the computing resources, storage resources, network resources, etc. of the target AI computing resources are clearly distinguished so that the corresponding computing resources can be called quickly and accurately under different task requirements. At the same time, according to the classification of applicable scenarios, such as preset data analysis scenarios, preset data reasoning scenarios, and preset model training scenarios, resources are further classified and managed. This classification method makes resource management more refined and can better meet the needs of different types of tasks.

[0054] Furthermore, a detailed resource directory is established based on the classification results to record the specific information of each computing resource, including the resource type, performance parameters, location, current status, etc. The resource directory can be used to easily query and understand the resource status of the entire or partial resource pool. At the same time, the resource management module has the ability to dynamically update the directory, and adjust the directory content in a timely manner as the resources change to ensure the accuracy and timeliness of resource information. At the same time, the computing resource information after perception registration, integration, classification and resource directory management provides a computing resource foundation for resource scheduling, task allocation and performance optimization of the entire system.

[0055] Step S12: abstract the target AI computing power resources based on the target resource directory to obtain corresponding target logical resources, then analyze the obtained computing power usage request, and allocate the target AI computing power resources corresponding to the target logical resources according to the obtained analysis results.

[0056] In this embodiment, virtualization technology is used to abstract various types of physical computing resources. For example, the high-performance parallel computing capability brought by GPU, the multi-tasking capability of general-purpose CPU, and other computing units for specific purposes are all converted into logical resources that can be flexibly scheduled and allocated by the system. This conversion makes physical resources with different performance and characteristics present a unified form at the logical level, providing abstract computing units for subsequent resource management and task allocation.

[0057] In this embodiment, the isolation mechanism can be used to ensure that different tasks and applications do not interfere with each other when using virtual resources. For resource types based on virtual machines, the virtualization module creates an independent operating environment for each virtual machine, including virtual CPU, vGPU (virtual graphics processing unit), memory, storage and network resources. In this way, even if a virtual machine fails or has performance problems, it will not affect the normal operation of other virtual machines.

[0058] Further, based on the target AI computing power resources, a preset virtual hardware environment is created for the first target object; wherein the creation of the preset virtual hardware environment includes allocating virtual CPU, mounting virtual GPU, dividing memory, setting storage and any one or more of network resources; based on the target AI computing power resources, a preset container resource space is divided for the second target object; wherein the preset container resource space includes any one or more of CPU quota, memory quota, file system and network namespace. Specifically, for the virtualization use scenario of container type, the virtualization module uses the lightweight characteristics of container technology to provide an independent resource space for each container to achieve efficient resource isolation. For example, for a virtual machine, an independent operating environment can be created for each virtual machine, including virtual CPU, vGPU, memory, storage and network resources. In this way, even if a virtual machine fails or has performance problems, it will not affect the normal operation of other virtual machines. For the virtualization use scenario of container type, the corresponding CPU quota, memory quota, file system and network namespace need to be allocated to the container. It can be understood that the containers can run independently of each other through the isolation mechanism provided by the virtualization module, while sharing the resources of the physical server to improve resource utilization.

[0059] In this embodiment, a computing power usage request is obtained, wherein the computing power usage request includes any one or more of a preset computing complexity usage request, a preset time usage request, and a preset data volume usage request. Analyze the specific requirements of each task, including multi-dimensional factors such as computing complexity, time requirements, and data volume. For different types of tasks, such as large-scale model training tasks, reasoning tasks with high real-time requirements, or complex data analysis tasks, their unique computing power demand characteristics can be accurately identified. Monitor the resource status of the entire resource pool in real time, including the usage, load level, available capacity, etc. of various physical computing power resources. Based on this real-time information, dynamically adjust the resource allocation strategy to ensure that each task can obtain the most suitable computing power resources.

[0060] In addition, the target AI computing resources corresponding to the target logical resources can be allocated using a preset priority scheduling algorithm based on the AI ​​computing resources that meet the preset level conditions; the target AI computing resources corresponding to the target logical resources can be allocated using a preset load balancing scheduling algorithm based on the classification results corresponding to the resource types of the target AI computing resources; the historical resource allocation situation is analyzed using a preset predictive scheduling algorithm to determine the corresponding predicted task requirements, and the target AI computing resources corresponding to the target logical resources are allocated based on the predicted task requirements. Specifically, according to different scenarios and requirements, a suitable scheduling algorithm is flexibly selected. A priority-based scheduling algorithm is adopted to ensure that high-priority tasks can obtain resources first and ensure the timely processing of key services; a load balancing-based scheduling algorithm is used to evenly distribute tasks to each node in the resource pool to avoid the situation where some nodes are overloaded and other nodes are idle, thereby achieving balanced utilization of resources. In addition, it also has a predictive scheduling algorithm, which predicts possible future task requirements through analysis and learning of historical task data, allocates resources in advance, and improves the response speed and resource utilization of the system. Through the synergy of multiple scheduling algorithms, the scheduling module can effectively improve resource utilization and avoid waste and idleness of resources. At the same time, it has the ability to allocate resources and evaluate load balancing, avoiding crashes or performance degradation caused by excessive load on a single node. It can quickly make adjustments when faced with sudden high-load tasks.

[0061] Step S13: monitor the resource usage of the target logical resources. If the resource usage indicates that the state of the target AI computing power resources is abnormal, perform a preset evaluation operation on the target AI computing power resources to obtain a corresponding evaluation result. Perform a preset repair operation on the target AI computing power resources according to the evaluation result to complete AI computing power resource management.

[0062] In this embodiment, the target logical resources are monitored, the utilization of the CPU and GPU is closely tracked, and the resource usage after executing different tasks is updated in real time, for example, the storage capacity and the occupancy of the video memory, the read and write speed of the data, and the health status of the storage device are monitored; the network bandwidth usage, network delay, and the transmission of data packets are monitored in real time; the temperature, voltage, fan speed and other parameters of the monitoring device are monitored; the running status of the program, the activity of the process, and the locking of resources are monitored. Through the careful monitoring of resource usage, accurate data basis can be provided for the reasonable allocation and scheduling of resources. Once the resource status is found to be abnormal, the monitoring module will immediately issue an alarm to avoid the impact of potential failures on the system. In terms of monitoring performance indicators, the performance of the resources is comprehensively evaluated, including computing performance, storage performance, network performance, and other aspects. For computing performance, the processor's computing speed, floating-point computing capability, and parallel processing capability are measured. For storage performance, the storage device's read and write speed, random access time, and data transmission rate are evaluated. For network performance, the network's bandwidth, delay, packet loss rate, and other parameters are monitored. By monitoring these performance indicators, resource performance bottlenecks and problems can be discovered in a timely manner, and strong support can be provided for the optimal configuration and upgrade of target computing resources. Based on the real-time monitoring of resource usage, status, and performance indicators, the monitoring module can provide accurate suggestions for the optimal configuration of resources. Since AI computing scenarios have the characteristics of large parallel computing volume, high bandwidth, and low latency, data collection granularity at the second or millisecond level can be used to ensure the timeliness and accuracy of monitoring. In terms of fault handling, based on the second-level data collection mechanism, the fault point can be quickly located and detailed fault information can be provided.

[0063] As can be seen from the above, the embodiment of the present application first identifies the resource information of the target AI computing power resources, classifies them according to the information to obtain the classification results, and establishes a target resource directory accordingly. Then, the target AI computing power resources in the directory are abstracted to obtain the target logical resources, the obtained computing power usage requests are analyzed, and the target AI computing power resources corresponding to the target logical resources are allocated according to the analysis results. Subsequently, the resource usage of the target logical resources is monitored. If the resource usage indicates that the state of the target AI computing power resources is abnormal, a preset evaluation operation is performed on it to obtain the evaluation result, and then a preset repair operation is performed on the target AI computing power resources based on the evaluation result, thereby completing the AI ​​computing power resource management. In this way, the embodiment of the present application realizes unified perception and management of AI computing power resources by establishing a target resource directory, improves the flexibility and convenience of diverse AI computing power equipment management, reduces the complexity of AI computing power equipment adaptation and use, and improves the universality of AI applications in the operation of AI computing power equipment, providing strong resource support for the development of artificial intelligence applications.

[0064] See also Figure 4As shown, for the embodiment of the present invention, a specific AI computing resource management method is disclosed, which may include:

[0065] Step S21, using a preset identity authentication system to identify the identity information of the object to be authorized; wherein the object to be authorized includes any one or more of a user terminal to be authorized, a device to be authorized, and an application to be authorized.

[0066] In this embodiment, the identity information of the object to be authorized is identified through an identity authentication system. First, the identity information of the user terminal to be authorized is identified by using a preset multi-factor authentication to determine whether the identity information of the user terminal to be authorized meets the first preset identity security condition; the identity information of the device to be authorized is identified by detecting the target identifier and security certificate of the device to be authorized to determine whether the identity information of the device to be authorized meets the second preset identity security condition; the identity information of the application to be authorized is identified by verifying the digital signature and preset permissions of the application to be authorized to determine whether the identity information of the application to be authorized meets the third preset identity security condition. Specifically, for the user terminal to be authorized, multi-factor authentication is used to enhance the security of identity authentication. For the device to be authorized, the security module will check the unique identification, security certificate and other information of the device to ensure the legitimacy and security of the device. For the application to be authorized, the security module will verify its digital signature, permission settings, etc. to prevent malicious applications from entering the resource pool.

[0067] Step S22: If the obtained identification result indicates that the object to be authorized meets the preset identity security condition, the preset access rights corresponding to the target logical resource are opened to the object to be authorized.

[0068] In this embodiment, by identifying the identity information of the object to be authorized, it is ensured that only authorized users, devices and applications can obtain access rights to the resource pool, so that the security and reliability of the resource pool can be ensured.

[0069] From the above, we can see that the identity authentication and access control mechanism has built a multi-level security protection system by designing differentiated authentication strategies for different authorized objects. Multi-factor authentication technology is used for the user side, integrating verification factors such as passwords and biometrics to effectively resist the risk of account theft; the unique hardware identification and encryption certificate verification of the device are carried out to ensure the legitimacy of the physical terminal and establish a trusted communication link; for the application, the authenticity of the code source is verified by digital signature, and its behavior boundary is controlled by combining the permission whitelist mechanism. When all three types of objects have passed their respective security condition verification, the corresponding access rights are granted. In this way, the risk of single-point authentication failure is reduced, providing a technical guarantee basis for the secure sharing of digital resources.

[0070] See also Figure 5 As shown, an embodiment of the present invention discloses an AI computing resource management device, including:

[0071] A directory establishment module 11 is used to identify resource information of the target AI computing power resources, classify the target AI computing power resources using the identified resource information of the target AI computing power resources to determine the classification result corresponding to the target AI computing power resources, and establish a corresponding target resource directory according to the classification result;

[0072] A resource allocation module 12, configured to abstract the target AI computing power resources based on the target resource directory to obtain corresponding target logical resources, and then analyze the acquired computing power usage request, and allocate the target AI computing power resources corresponding to the target logical resources according to the obtained analysis results;

[0073] The resource management completion module 13 is used to monitor the resource usage of the target logical resources. If the resource usage indicates that the state of the target AI computing power resource is abnormal, a preset evaluation operation is performed on the target AI computing power resource to obtain a corresponding evaluation result, and a preset repair operation is performed on the target AI computing power resource according to the evaluation result to complete the AI ​​computing power resource management.

[0074] As can be seen from the above, this application first identifies the resource information of the target AI computing power resources, classifies them according to the information to obtain the classification results, and establishes a target resource directory based on this. Then, the target AI computing power resources in the directory are abstracted to obtain the target logical resources, the obtained computing power usage requests are analyzed, and the target AI computing power resources corresponding to the target logical resources are allocated according to the analysis results. Subsequently, the resource usage of the target logical resources is monitored. If the resource usage indicates that the state of the target AI computing power resources is abnormal, a preset evaluation operation is performed on it to obtain the evaluation result, and then a preset repair operation is performed on the target AI computing power resources based on the evaluation result, thereby completing the AI ​​computing power resource management. In this way, this application realizes unified perception and management of AI computing power resources by establishing a target resource directory, improves the flexibility and convenience of diverse AI computing power equipment management, reduces the complexity of AI computing power equipment adaptation and use, and improves the universality of AI applications in the operation of AI computing power equipment, providing strong resource support for the development of artificial intelligence applications.

[0075] In some specific implementations, the directory creation module 11 includes:

[0076] A resource information identification unit, configured to obtain resource information of the identified target AI computing resource; wherein the resource information includes any one or more of performance parameters, resource types, and applicable scenarios;

[0077] a first resource classification unit, configured to perform a preset level classification operation on the target AI computing power resources by using the performance parameters of the target AI computing power resources, so as to determine AI computing power resources that meet various preset level conditions from the target AI computing power resources;

[0078] a second resource classification unit, configured to classify the target AI computing power resource by using the resource type of the target AI computing power resource to determine a classification result of the target AI computing power resource obtained based on the resource type classification; wherein the resource type includes any one or more of computing resources, storage resources, and network resources;

[0079] The third resource classification unit is used to classify the target AI computing power resources using the applicable scenarios of the target AI computing power resources to determine the classification results of the target AI computing power resources obtained based on the applicable scenario classification; wherein the applicable scenarios include any one or more of the preset data analysis scenarios, preset data reasoning scenarios, and preset model training scenarios.

[0080] In some specific implementations, the resource allocation module 12 includes:

[0081] A first resource allocation unit, configured to allocate the target AI computing power resources corresponding to the target logical resources by using a preset priority scheduling algorithm based on the AI ​​computing power resources that meet various preset level conditions;

[0082] A second resource allocation unit, configured to allocate the target AI computing power resource corresponding to the target logical resource by using a preset load balancing scheduling algorithm based on the classification result corresponding to the resource type of the target AI computing power resource;

[0083] The third resource allocation unit is used to analyze the historical resource allocation situation by using a preset predictive scheduling algorithm to determine the corresponding predicted task requirements, and allocate the target AI computing power resources corresponding to the target logical resources based on the predicted task requirements.

[0084] In some specific implementations, the AI ​​computing power resource management device may further include:

[0085] An environment creation module, used to create a preset virtual hardware environment for the first target object based on the target AI computing power resources; wherein the creation of the preset virtual hardware environment includes allocating a virtual CPU, mounting a virtual GPU, dividing memory, setting storage and network resources;

[0086] A space partitioning module is used to partition a preset container resource space for a second target object based on the target AI computing power resources; wherein the preset container resource space includes any one or more of a CPU quota, a memory quota, a file system, and a network namespace.

[0087] In some specific implementations, the computing power usage request includes any one or more of a preset computing complexity usage request, a preset time usage request, and a preset data volume usage request.

[0088] In some specific implementations, the AI ​​computing power resource management device may further include:

[0089] An identity information identification module, used to identify the identity information of the object to be authorized using a preset identity authentication system; wherein the object to be authorized includes any one or more of a user terminal to be authorized, a device to be authorized, and an application to be authorized;

[0090] The permission opening module is used to open the preset access rights corresponding to the target logical resource to the object to be authorized if the obtained identification result indicates that the object to be authorized meets the preset identity security condition.

[0091] In some specific implementations, the AI ​​computing power resource management device may further include:

[0092] A first security condition judgment module, used to identify the identity information of the user terminal to be authorized by using a preset multi-factor authentication to determine whether the acquired identity information of the user terminal to be authorized meets a first preset identity security condition;

[0093] A second security condition judgment module is used to identify the identity information of the device to be authorized by detecting the target identifier of the device to be authorized and the security certificate, so as to determine whether the acquired identity information of the device to be authorized meets a second preset identity security condition;

[0094] The third security condition judgment module is used to identify the identity information of the application to be authorized by verifying the digital signature of the application to be authorized and the preset authority, so as to determine whether the obtained identity information of the application to be authorized meets the third preset identity security condition.

[0095] Furthermore, the present application also discloses an electronic device. Figure 6 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram cannot be regarded as any limitation on the scope of use of the present application.

[0096] Figure 6A schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the AI ​​computing resource management method disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0097] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0098] In addition, the memory 22, as a carrier for storing resources, can be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0099] Among them, the operating system 221 is used to manage and control various hardware devices on the electronic device 20 and the computer program 222, which can be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program that can be used to complete the AI ​​computing resource management method performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include a computer program that can be used to complete other specific tasks.

[0100] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the aforementioned disclosed AI computing resource management method is implemented. The specific steps of the method can be referred to the corresponding contents disclosed in the aforementioned embodiments, and will not be repeated here.

[0101] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0102] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0103] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0104] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0105] The technical solution provided by the present application is introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technicians in this field, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for managing AI computing resources, characterized in that: include: Identify resource information of the target AI computing power resources, classify the target AI computing power resources using the identified resource information of the target AI computing power resources to determine a classification result corresponding to the target AI computing power resources, and establish a corresponding target resource directory according to the classification result; Based on the target resource directory, the target AI computing power resource is abstracted to obtain a corresponding target logical resource, and then the obtained computing power usage request is analyzed, and the target AI computing power resource corresponding to the target logical resource is allocated according to the obtained analysis result; Monitor the resource usage of the target logical resources. If the resource usage indicates that the state of the target AI computing power resources is abnormal, perform a preset evaluation operation on the target AI computing power resources to obtain a corresponding evaluation result. Perform a preset repair operation on the target AI computing power resources according to the evaluation result to complete AI computing power resource management.

2. The AI ​​computing resource management method according to claim 1, characterized in that: The classifying the target AI computing resources by using the identified resource information of the target AI computing resources includes: Obtain resource information of the identified target AI computing resource; wherein the resource information includes any one or more of performance parameters, resource types, and applicable scenarios; Using the performance parameters of the target AI computing resources, the target AI computing resources are classified into preset levels to determine AI computing resources that meet various preset level conditions from the target AI computing resources; Classifying the target AI computing power resource by using the resource type of the target AI computing power resource to determine a classification result of the target AI computing power resource obtained based on the resource type classification; wherein the resource type includes any one or more of computing resources, storage resources, and network resources; The target AI computing power resources are classified according to the applicable scenarios of the target AI computing power resources to determine the classification results of the target AI computing power resources obtained based on the applicable scenario classification; wherein the applicable scenarios include any one or more of the preset data analysis scenarios, the preset data reasoning scenarios, and the preset model training scenarios.

3. The AI ​​computing resource management method according to claim 2, characterized in that: The allocating the target AI computing power resource corresponding to the target logical resource according to the obtained analysis result includes: Allocate the target AI computing resources corresponding to the target logical resources using a preset priority scheduling algorithm based on the AI ​​computing resources that meet the preset level conditions; Based on the classification result corresponding to the resource type of the target AI computing power resource, the target AI computing power resource corresponding to the target logical resource is allocated using a preset load balancing scheduling algorithm; The preset predictive scheduling algorithm is used to analyze the historical resource allocation to determine the corresponding predicted task requirements, and the target AI computing power resources corresponding to the target logical resources are allocated based on the predicted task requirements.

4. The AI ​​computing resource management method according to claim 1, characterized in that: After the target AI computing power resource is abstracted based on the target resource directory to obtain the corresponding target logical resource, the method further includes: Creating a preset virtual hardware environment for the first target object based on the target AI computing power resources; wherein creating the preset virtual hardware environment includes allocating a virtual CPU, mounting a virtual GPU, dividing memory, setting storage, and any one or more of network resources; Based on the target AI computing power resources, a preset container resource space is divided for the second target object; wherein the preset container resource space includes any one or more of CPU quota, memory quota, file system and network namespace.

5. The AI ​​computing resource management method according to claim 1, characterized in that: The computing power usage request includes any one or more of a preset computing complexity usage request, a preset time usage request, and a preset data volume usage request.

6. The AI ​​computing resource management method according to any one of claims 1 to 5, characterized in that: Also includes: Using a preset identity authentication system to identify the identity information of the object to be authorized; wherein the object to be authorized includes any one or more of the user terminal to be authorized, the device to be authorized, and the application to be authorized; If the obtained identification result indicates that the object to be authorized meets the preset identity security condition, the preset access rights corresponding to the target logical resource are opened to the object to be authorized.

7. The AI ​​computing resource management method according to claim 6, characterized in that: The method of using a preset identity authentication system to identify the identity information of the object to be authorized includes: Identify the identity information of the user terminal to be authorized by using a preset multi-factor authentication to determine whether the obtained identity information of the user terminal to be authorized meets a first preset identity security condition; Identifying the identity information of the device to be authorized by detecting the target identifier and the security certificate of the device to be authorized, so as to determine whether the acquired identity information of the device to be authorized meets a second preset identity security condition; The identity information of the application to be authorized is identified by verifying the digital signature of the application to be authorized and the preset authority, so as to determine whether the obtained identity information of the application to be authorized meets the third preset identity security condition.

8. An AI computing resource management device, characterized in that: include: A directory establishment module, used to identify resource information of a target AI computing resource, classify the target AI computing resource using the identified resource information of the target AI computing resource to determine a classification result corresponding to the target AI computing resource, and establish a corresponding target resource directory according to the classification result; A resource allocation module, configured to abstract the target AI computing power resources based on the target resource directory to obtain corresponding target logical resources, and then analyze the acquired computing power usage request, and allocate the target AI computing power resources corresponding to the target logical resources according to the obtained analysis results; The resource management completion module is used to monitor the resource usage of the target logical resources. If the resource usage indicates that the state of the target AI computing power resource is abnormal, a preset evaluation operation is performed on the target AI computing power resource to obtain a corresponding evaluation result. According to the evaluation result, a preset repair operation is performed on the target AI computing power resource to complete the AI ​​computing power resource management.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, used to execute the computer program to implement the AI ​​computing resource management method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: Used to save computer programs; wherein, when the computer program is executed by a processor, it implements the AI ​​computing power resource management method as described in any one of claims 1 to 7.

Images