AI accelerator card operation and maintenance method, device and equipment based on performance index acquisition and medium

Abstract

The invention discloses an AI accelerator card operation and maintenance method and device based on performance index collection, equipment and a medium, and relates to the technical field of cloud computing, and the method comprises the steps: obtaining a performance index collection request sent by a target host after the communication connection with the target host is established, and determining target peripheral interface equipment information; determining the number of currently mounted accelerator cards and the type of the accelerator cards through a preset accelerator card manufacturer dictionary table and target peripheral interface equipment information, and initializing a performance index acquisition thread according to the number of the accelerator cards and the type of the accelerator cards; verifying a local accelerator card driver and an equipment management library of the AI accelerator card, and after verification is passed, calling the equipment management library to perform performance index acquisition of the AI accelerator card based on the performance index acquisition request and the performance index acquisition thread; and storing the performance index data in a preset database, performing judgment according to a preset performance index alarm rule, and performing display at a target front end so as to perform operation and maintenance on the AI accelerator card.

Description

Technical Field

[0001] The present invention relates to the field of cloud computing technology, and in particular to an AI accelerator card operation and maintenance method, device, equipment, and medium based on performance indicator collection. Background Art

[0002] In today's era of widespread cloud computing, AI (artificial intelligence) computing tasks are growing rapidly. To efficiently utilize resources, cloud computing platforms typically use OpenStack (an open-source cloud computing management platform project) for overall resource management. For specialized hardware resources like AI accelerator cards, they rely on Cyborg (a framework for managing hardware and software acceleration resources). Currently, cloud computing platforms for AI accelerator cards use manual scheduling and management, requiring staff to monitor the cloud computing platform in real time. This is labor-intensive and difficult to promptly detect AI accelerator card anomalies, such as hardware performance degradation or damage, resulting in inefficient management. Furthermore, because staff members need to spend time analyzing AI accelerator card performance metrics, they lack real-time visibility, making it difficult to accurately schedule computing power and allocate tasks based on the actual operating status of AI accelerator cards. This can lead to some AI accelerator cards being overloaded while others remain idle, severely impacting the overall operational efficiency and resource utilization of the cloud computing platform.

[0003] In summary, how to improve the management efficiency of AI accelerator cards to ensure the overall operating efficiency and resource utilization of cloud computing platforms is a technical problem that needs to be solved urgently. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide an AI accelerator card operation and maintenance method, device, equipment, and medium based on performance indicator collection, which can improve the management efficiency of AI accelerator cards to ensure the overall operating efficiency and resource utilization of the cloud computing platform. The specific solution is as follows:

[0005] In a first aspect, the present application provides an AI accelerator card operation and maintenance method based on performance indicator collection, which is applied to a target virtual machine, including:

[0006] After establishing a communication connection with a target host, obtaining a performance indicator collection request sent by the target host and determining target peripheral interface device information of the target virtual machine; the performance indicator collection request is a request initiated by the target host based on a preset performance indicator collection frequency;

[0007] Determine the number of currently mounted AI accelerator cards and the corresponding accelerator card types using a preset accelerator card manufacturer dictionary table and the target peripheral interface device information, and initialize a performance indicator collection thread based on the number of accelerator cards and the accelerator card types.

[0008] Verify the local accelerator card driver and the device management library of the AI ​​accelerator card. After both verifications pass, call the device management library to collect performance indicators of the AI ​​accelerator card based on the performance indicator collection request and the performance indicator collection thread to obtain corresponding performance indicator data.

[0009] The performance indicator data is saved in a preset database, and the performance indicator data is judged according to the preset performance indicator alarm rules, and the performance indicator data is displayed on the target front end, so that the AI ​​accelerator card can be operated and maintained according to the performance indicator data.

[0010] Optionally, after establishing a communication connection with the target host, obtaining a performance indicator collection request sent by the target host and determining target peripheral interface device information of the target virtual machine includes:

[0011] After establishing a communication connection with the target host based on a preset communication protocol, continuously monitoring the established communication channel corresponding to the preset communication protocol, so that the target host can initiate a performance indicator collection request to the target virtual machine through the communication channel based on a preset performance indicator collection frequency;

[0012] Obtaining the performance indicator collection request sent by the target host through the communication channel;

[0013] A target file located locally in a preset path is searched to obtain target peripheral interface device information of the target virtual machine from the target file.

[0014] Optionally, determining the number of accelerator cards and corresponding accelerator card types of the currently mounted AI accelerator cards by using a preset accelerator card manufacturer dictionary table and the target peripheral interface device information includes:

[0015] Parsing the target peripheral interface device information to extract the accelerator card manufacturer code and the accelerator card type code recorded in the target peripheral interface device information;

[0016] Searching a preset accelerator card manufacturer dictionary table to determine an accelerator card type corresponding to both the accelerator card manufacturer code and the accelerator card type code in the preset accelerator card manufacturer dictionary table;

[0017] The number of the extracted accelerator card type codes is determined as the number of accelerator cards of the currently mounted AI accelerator card.

[0018] Optionally, initializing the performance indicator collection thread according to the number and type of the accelerator cards includes:

[0019] Starting a preset number of performance indicator collection threads according to the number of accelerator cards; the preset number is consistent with the number of accelerator cards;

[0020] Determine the accelerator card manufacturer code of the accelerator card type corresponding to the performance indicator collection thread;

[0021] The device management library of the corresponding accelerator card manufacturer is determined based on the accelerator card manufacturer code, so as to initialize the performance indicator collection thread using the device management library.

[0022] Optionally, verifying the local accelerator card driver and the device management library of the AI ​​accelerator card includes:

[0023] Determine whether the local accelerator card driver is successfully identified and loaded normally to obtain the corresponding judgment result;

[0024] If the judgment result indicates that the local accelerator card driver is successfully identified and loaded normally, then the accelerator card driver verification is determined to be passed, and an import operation is performed on the device management library of the AI ​​accelerator card to detect whether the device management library has been installed;

[0025] If the import operation of the device management library is successfully implemented, it is determined that the device management library has been installed, and a verification result is obtained that both the accelerator card driver and the device management library have passed verification.

[0026] Optionally, calling the device management library to collect the performance indicators of the AI ​​accelerator card based on the performance indicator collection request and the performance indicator collection thread includes:

[0027] Determine the target accelerator card type and target AI accelerator card corresponding to the performance indicator collection thread;

[0028] Based on the performance indicator collection request, the performance indicators corresponding to the current target accelerator card type are combined to determine the target performance indicator to be collected; the target performance indicator is a performance indicator supported by the target accelerator card type among the performance indicators corresponding to the target accelerator card types;

[0029] Determine the corresponding target device management library according to the target accelerator card type, and call the target device management library through the performance indicator collection thread to collect the target performance indicator of the target AI accelerator card based on the collection tool in the target device management library.

[0030] Optionally, the step of saving the performance indicator data to a preset database, judging the performance indicator data according to preset performance indicator alarm rules, and displaying the performance indicator data on a target front end so as to operate and maintain the AI ​​accelerator card based on the performance indicator data includes:

[0031] Performing unit standardization processing on the obtained performance indicator data to obtain processed performance indicator data with consistent indicator types, and saving the processed performance indicator data in a preset database;

[0032] Using preset performance indicator alarm rules to judge the processed performance indicator data in the preset database;

[0033] If target performance indicator data matching the performance indicator alarm rule exists in the processed performance indicator data, an alarm is issued for the target performance indicator data, and the target performance indicator data is analyzed to obtain corresponding analysis results;

[0034] A corresponding performance indicator curve is generated according to the performance indicator data, and the performance indicator curve and the analysis results of the target performance indicator data are displayed to the target front end, so that the AI ​​accelerator card can be operated and maintained according to the performance indicator curve and the analysis results.

[0035] In a second aspect, the present application provides an AI accelerator card operation and maintenance device based on performance indicator collection, which is applied to a target virtual machine and includes:

[0036] An information determination module is configured to, after establishing a communication connection with a target host, obtain a performance indicator collection request sent by the target host and determine target peripheral interface device information of the target virtual machine; the performance indicator collection request is a request initiated by the target host based on a preset performance indicator collection frequency;

[0037] A thread initialization module is configured to determine the number of accelerator cards and the corresponding accelerator card types of the currently mounted AI accelerator cards using a preset accelerator card manufacturer dictionary table and the target peripheral interface device information, and to initialize a performance indicator collection thread based on the number of accelerator cards and the accelerator card types.

[0038] An indicator collection module is used to verify the local accelerator card driver and the device management library of the AI ​​accelerator card. After both verifications are passed, based on the performance indicator collection request and the performance indicator collection thread, the module calls the device management library to collect the performance indicators of the AI ​​accelerator card to obtain corresponding performance indicator data.

[0039] A data display module is used to save the performance indicator data to a preset database, judge the performance indicator data according to preset performance indicator alarm rules, and display the performance indicator data on the target front end so as to operate and maintain the AI ​​accelerator card based on the performance indicator data.

[0040] In a third aspect, the present application provides an electronic device, comprising:

[0041] Memory, used to store computer programs;

[0042] A processor is used to execute the computer program to implement the aforementioned AI accelerator card operation and maintenance method based on performance indicator collection.

[0043] 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 aforementioned AI acceleration card operation and maintenance method based on performance indicator collection.

[0044] In this application, after establishing a communication connection with the target host, a performance indicator collection request sent by the target host is obtained, and the target peripheral interface device information of the target virtual machine is determined; the performance indicator collection request is a request initiated by the target host based on a preset performance indicator collection frequency; the number of accelerator cards and the corresponding accelerator card types of the currently mounted AI accelerator card are determined through a preset accelerator card manufacturer dictionary table and the target peripheral interface device information, and a performance indicator collection thread is initialized according to the number of accelerator cards and the accelerator card type; the local accelerator card driver and the device management library of the AI ​​accelerator card are verified, and after the verification is passed, the device management library is called to collect the performance indicators of the AI ​​accelerator card based on the performance indicator collection request and the performance indicator collection thread to obtain corresponding performance indicator data; the performance indicator data is saved in a preset database, and the performance indicator data is judged according to the preset performance indicator alarm rules, and the performance indicator data is displayed on the target front end, so that the AI ​​accelerator card can be operated and maintained according to the performance indicator data. As can be seen from the above, the target virtual machine of this application first establishes a communication connection with the target host, receives the performance indicator collection request sent by the target host based on the preset frequency, determines its own target peripheral interface device information, identifies the number and type of currently mounted AI accelerator cards through the preset accelerator card manufacturer dictionary table, and initializes the performance indicator collection thread accordingly. Then, after the accelerator card driver and the device management library are verified, the performance indicator collection request and the performance indicator collection thread call the device management library to collect performance indicator data, and finally the collected data is stored in the preset database, judged according to the preset alarm rules and displayed on the front end, so as to operate and maintain the AI ​​accelerator card. In this way, through the above process of this application, it is possible to automatically identify the accelerator card configuration, verify the validity of the driver and library files, and realize the timed collection and structured storage of performance indicators; combined with the visual display and alarm rules of the target front end, it can help operation and maintenance personnel monitor the operating status of the AI ​​accelerator card in real time, promptly discover performance anomalies or potential faults, improve the refinement and operation and maintenance efficiency of hardware resource management, ensure the stable and efficient operation of the AI ​​accelerator card, and thus improve the management efficiency of the AI ​​accelerator card to ensure the overall operation efficiency and resource utilization of the cloud computing platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0046] Figure 1This is a flow chart of an AI accelerator card operation and maintenance method based on performance indicator collection disclosed in this application;

[0047] Figure 2 This is a process sequence diagram of an AI accelerator card operation and maintenance method based on performance indicator collection disclosed in this application;

[0048] Figure 3 This is a schematic diagram of the structure of an AI accelerator card operation and maintenance device based on performance indicator collection disclosed in this application;

[0049] Figure 4 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION

[0050] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0051] Currently, cloud computing platforms for AI accelerator cards are manually scheduled and managed, requiring staff to monitor the cloud computing platform in real time. This is labor-intensive and difficult to detect AI accelerator card anomalies, such as hardware performance degradation or damage, leading to low management efficiency. Furthermore, staff members spend time analyzing the performance indicators of AI accelerator cards, unable to grasp them in real time. This makes it difficult to accurately schedule computing power and properly allocate tasks based on the actual operating status of AI accelerator cards. This can lead to some AI accelerator cards being overloaded and others being idle, severely impacting the overall operating efficiency and resource utilization of the cloud computing platform.

[0052] In order to overcome the above technical problems, this application provides an AI accelerator card operation and maintenance method based on performance indicator collection to improve the management efficiency of AI accelerator cards to ensure the overall operation efficiency and resource utilization of the cloud computing platform.

[0053] See also Figure 1 As shown, an embodiment of the present invention discloses an AI accelerator card operation and maintenance method based on performance indicator collection, which is applied to a target virtual machine and includes:

[0054] Step S11: After establishing a communication connection with the target host, obtain a performance indicator collection request sent by the target host and determine the target peripheral interface device information of the target virtual machine; the performance indicator collection request is a request initiated by the target host based on a preset performance indicator collection frequency.

[0055] In this embodiment, after the target virtual machine to be operated and maintained is started, it first attempts to establish a communication connection with the target host. After establishing the communication connection with the target host, it obtains a performance indicator collection request sent by the target host and determines the target peripheral interface device information of the target virtual machine. The performance indicator collection request is a request initiated by the target host based on a preset performance indicator collection frequency. The preset performance indicator collection frequency can be flexibly configured according to actual needs. For example, it can be set to collect data every 5 minutes to ensure the real-time nature of the data and the rational use of system resources. The target peripheral interface device information is the basic information of the PCI (a bus protocol for connecting and extending peripheral devices) device of the target virtual machine, including but not limited to the accelerator card manufacturer code, accelerator card type code, etc.

[0056] It is understandable that in order to efficiently utilize resources, cloud computing platforms usually use Openstack for overall resource management, and for special hardware resources such as AI accelerator cards, they are managed with the help of Cyborg. The AI ​​accelerator card operation and maintenance method based on performance indicator collection in this application is particularly suitable for cloud computing platforms that use Openstack for cloud platform resource management and Cyborg for AI accelerator card management. By realizing comprehensive monitoring of the operating status of the AI ​​accelerator card by the cloud platform, key performance indicators such as the memory utilization and temperature of the AI ​​accelerator card can be accurately obtained, so that the cloud platform can grasp the operating status of the AI ​​accelerator card in real time, realize accurate computing power scheduling, and improve the overall resource utilization and operating efficiency of the cloud computing platform. Figure 2The figure shows a process sequence diagram of an AI accelerator card operation and maintenance method based on performance indicator collection provided by the present application. The present application consists of four modules: a performance collection module within a virtual machine, a performance collection module for computing nodes, an indicator aggregation and analysis module, and an indicator display and alarm module. Among them, the performance collection module within the virtual machine mainly realizes the identification of the AI ​​accelerator card, so as to load different performance collection instructions and complete the collection of health information such as the usage and temperature of the AI ​​accelerator card video memory; the computing node performance collection module is mainly used to write a performance collection exporter (a program that provides monitoring sample data), complete communication with the virtual machine, and regularly pull performance indicators; the indicator aggregation and analysis module is used to summarize the performance indicators collected by the computing nodes for storage operations, and judge whether the indicators are normal according to the monitoring indicator alarm rules, and initiate an alarm process for abnormal performance indicators; the indicator display and alarm module is used to display the performance curves of various indicators on the page, and can analyze and process the alarm indicators. It should be pointed out that the performance collection module in the virtual machine can be divided into three types according to function: collection controller, accelerator card identification, and indicator collection. The collection controller is used to realize communication between the host and QEMU (an open source computer emulator and virtualizer) based on the QMP protocol (a communication mechanism suitable for QEMU and based on JSON (a lightweight text data exchange format)), receive indicator collection instructions from computing nodes, and complete the scheduling of indicator collection and data return; the accelerator card identification is used to realize the distinction of AI accelerator card types and the statistics of the number of AI accelerator cards by obtaining and analyzing the PCI information of the virtual machine; the indicator collection is used to obtain performance collection instructions according to the instruction set of each AI accelerator card manufacturer, and obtain indicators such as video memory usage, power, and temperature. In the case of multiple cards, the collection speed can be improved through multi-threading.

[0057] Specifically, after establishing a communication connection with the target host based on a preset communication protocol, the communication channel established corresponding to the preset communication protocol is continuously monitored, so that the target host can initiate a performance indicator collection request to the target virtual machine through the communication channel based on a preset performance indicator collection frequency; obtain the performance indicator collection request sent by the target host through the communication channel; search for the target file located locally in the preset path to obtain the target peripheral interface device information of the target virtual machine from the target file. The preset communication protocol can be a QMP protocol. That is, after the target virtual machine is started, the service of the collection controller will automatically start and attempt to establish a communication connection with the target host based on the preset communication protocol. After the connection is successfully established, the communication channel established corresponding to the preset communication protocol, i.e., the virtual serial port device, is continuously monitored, waiting for the request sent by the target host, so as to obtain the performance indicator collection request initiated by the target host through the communication channel, and parse and process the performance indicator collection request. For example, if the performance indicator collection request is to obtain the AI ​​accelerator card usage rate of the target virtual machine, the collection controller can start the GPU (Graphics Processing Unit) monitoring program in the target virtual machine, and encapsulate the result into a response message in the format of the custom protocol, and transmit it through the communication channel established previously. At the same time, a search is performed for the target file located locally in the preset path to obtain the target peripheral interface device information of the target virtual machine. It should be noted that the target file can be the file / proc / bus / pci / devices, in which the first column is a 4-digit hexadecimal number, the first two digits are the bus number (bus_id), the last two digits are ANDed with 7 to form the function number, and the last two digits are right-shifted by 3 to form the bus device number; the second column is an 8-digit hexadecimal number, the first four digits are the accelerator card manufacturer code (vender_id), and the last four digits are the accelerator card type code (device_id). In this way, after receiving the performance indicator collection request sent by the target host, this embodiment first determines its own target peripheral interface device information, so as to subsequently identify the AI ​​accelerator card based on the information and improve the accuracy of performance indicator collection; controlling the target host to initiate a performance indicator collection request based on the preset performance indicator collection frequency can ensure the real-time nature of the data and the rational use of system resources.

[0058] Step S12: Determine the number of accelerator cards and the corresponding accelerator card types of the currently mounted AI accelerator cards through a preset accelerator card manufacturer dictionary table and the target peripheral interface device information, and initialize a performance indicator collection thread according to the number of accelerator cards and the accelerator card type.

[0059] In this embodiment, after receiving the performance indicator collection request and obtaining the target peripheral interface device information, the number of accelerator cards of the currently mounted AI accelerator card and the corresponding accelerator card type can be determined through the preset accelerator card manufacturer dictionary table and the target peripheral interface device information. At the same time, the performance indicator collection thread is initialized according to the determined accelerator card type and the number of accelerator cards, so as to perform the performance indicator collection operation according to the performance indicator collection thread. Among them, the performance indicator collection thread is responsible for loading the device management library of the currently responsible card type to realize the collection of monitoring indicators; the preset accelerator card manufacturer dictionary table is used to maintain the accelerator card manufacturer code and accelerator card type code supported by the method of this application, so as to identify the AI ​​accelerator card. The following Table 1 is a preset accelerator card manufacturer dictionary table provided by this application.

[0060] Table 1

[0061]

[0062] It should be noted that the processing flow for determining the number of AI accelerator cards currently mounted on the target virtual machine and the corresponding accelerator card types is as follows: parsing the target peripheral interface device information to extract the accelerator card manufacturer code and the accelerator card type code recorded in the target peripheral interface device information; searching the preset accelerator card manufacturer dictionary table to determine the accelerator card type corresponding to both the accelerator card manufacturer code and the accelerator card type code in the preset accelerator card manufacturer dictionary table; and determining the number of the extracted accelerator card type codes as the number of accelerator cards of the currently mounted AI accelerator card. That is, the target peripheral interface device information is parsed to extract the accelerator card manufacturer code and the accelerator card type code recorded therein, and then the preset accelerator card manufacturer dictionary table is searched to compare the accelerator card manufacturer code and the accelerator card type code with the values ​​recorded in the preset accelerator card manufacturer dictionary table based on a preset algorithm to determine the accelerator card type corresponding to the above codes. At the same time, since the target peripheral interface device information records the accelerator card type codes of all AI accelerator cards mounted in the target virtual machine, the number of the accelerator card type codes can be determined as the number of accelerator cards of the currently mounted AI accelerator cards.

[0063] It should be further pointed out that the processing flow for initializing the performance indicator collection thread is as follows: start a preset number of performance indicator collection threads based on the number of accelerator cards; the preset number is consistent with the number of accelerator cards; determine the accelerator card manufacturer code of the accelerator card type corresponding to the performance indicator collection thread; determine the device management library of the corresponding accelerator card manufacturer based on the accelerator card manufacturer code, so as to use the device management library to initialize the performance indicator collection thread. That is, start a number of performance indicator collection threads that is consistent with the number of accelerator cards, that is, start a corresponding performance indicator collection thread for each AI accelerator card, and then determine the accelerator card manufacturer code of the accelerator card type corresponding to the performance indicator collection thread. At the same time, since the device management library is provided by each AI accelerator card manufacturer, different types of AI accelerator cards correspond to different device management libraries. Therefore, it is also necessary to determine the corresponding device management library according to the accelerator card manufacturer code, so as to use the device management library to initialize the performance indicator collection thread. In this way, this embodiment determines the number and type of accelerator cards mounted on the target virtual machine based on the preset accelerator card manufacturer dictionary table and the determined target peripheral interface device information, and can accurately identify the AI ​​accelerator card to determine the adapted device management library, facilitating the operation and maintenance of the accelerator card; starting a corresponding performance indicator collection thread for each AI accelerator card can realize multi-threaded indicator collection, improve the performance indicator collection speed, and thereby improve the efficiency of accelerator card operation and maintenance.

[0064] Step S13: Verify the local accelerator card driver and the device management library of the AI ​​accelerator card. After the verification is passed, call the device management library to collect the performance indicators of the AI ​​accelerator card based on the performance indicator collection request and the performance indicator collection thread to obtain corresponding performance indicator data.

[0065] In this embodiment, in order to avoid the situation where the performance indicator collection is abnormal due to abnormalities in the local accelerator card driver and the device management library of the AI ​​accelerator card, the accelerator card driver and the device management library can be verified first, and after the verification is passed, it is determined that the accelerator card driver is correct and the device management library exists, and based on the performance indicator collection request and the performance indicator collection thread, the device management library is called to collect the performance indicators of the AI ​​accelerator card to obtain the corresponding performance indicator data. It should be pointed out that if the verification of the accelerator card driver or the device management library fails, it means that the accelerator card driver is abnormal or the device management library does not exist, and the performance indicator collection operation is not performed. It can be understood that in the process of the performance indicator collection thread collecting indicators, the main thread of this embodiment can identify the collection process status of each performance indicator collection thread, and uniformly summarize the monitored collection results and return them.

[0066] It should be pointed out that the method of this application relies on the relevant hardware management library provided by each AI accelerator card manufacturer. Since different virtual machines have different driver installation conditions in different scenarios, the device management library may not exist. Therefore, this embodiment can use an explicit post-loading method to load the function call and collect indicators, thereby eliminating the strong dependency between the component and the device management dynamic link library, and does not require all device management libraries to be configured in the environment during compilation. In order to avoid the library missing error caused by the scenario where the accelerator card driver and the device management library are installed separately, this embodiment can verify the accelerator card driver and the device management library separately. The processing flow is as follows: determine whether the local accelerator card driver is successfully identified and loaded normally to obtain a corresponding judgment result; if the judgment result indicates that the local accelerator card driver is successfully identified and loaded normally, then determine that the accelerator card driver verification is passed, and perform an import operation on the device management library of the AI ​​accelerator card to detect whether the device management library has been installed; if the import operation on the device management library is successfully implemented, then determine that the device management library has been installed, and obtain a verification result that both the accelerator card driver and the device management library verification have passed. That is, first determine whether the local accelerator card driver is successfully identified and loaded normally. If it is successfully identified and loaded normally, it means that the accelerator card driver is correct, and it is determined that the accelerator card driver verification has passed. Then, the device management library of the AI ​​accelerator card is imported to detect whether the device management library has been installed. If the import operation is successfully implemented, it means that the device management library exists and is determined to have been installed, and a verification result is obtained that both the accelerator card driver and the device management library verification have passed.

[0067] It should be further pointed out that the processing flow for collecting the performance indicators of the AI ​​accelerator card is as follows: determine the target accelerator card type and the target AI accelerator card corresponding to the performance indicator collection thread; based on the performance indicator collection request, integrate the performance indicators corresponding to the current target accelerator card type to determine the target performance indicators to be collected; the target performance indicators are the performance indicators supported by the target accelerator card type among the several performance indicators corresponding to the respective target accelerator card types; determine the corresponding target device management library according to the target accelerator card type, and call the target device management library through the performance indicator collection thread to collect the target performance indicators of the target AI accelerator card based on the collection tool in the target device management library. That is, first determine the current corresponding target accelerator card type and the target AI accelerator card, and then integrate the performance indicators corresponding to the current target accelerator card type to determine the common target performance indicators of each AI accelerator card to be collected. The following Table 2 is a schematic table of common performance indicators supported by each AI accelerator card provided in this application.

[0068] Table 2

[0069]

[0070] After determining the target performance indicator, the corresponding target device management library is determined according to the target accelerator card type, and the target device management library is called through the performance indicator collection thread to collect the target performance indicator of the target AI accelerator card based on the collection tool in the target device management library. For example, monitoring of specific AI accelerator cards requires installing the corresponding monitoring library and related dynamic libraries. These libraries then provide function interfaces to collect AI accelerator card operating metrics, including hardware status and performance data. For another type of mainstream graphics processing unit (GPU) accelerator cards, monitoring functionality can be implemented using a dynamic link library (DLL) provided by the GPU manufacturer. This library, part of the GPU management library, allows developers to programmatically access GPU management and monitoring functions. It can directly communicate with the GPU driver to obtain detailed data such as hardware status, performance metrics, and power consumption. This library is widely used in monitoring tools, cluster management systems, and automated scripts to effectively manage and monitor GPU resources. For other high-performance computing accelerator cards, monitoring is implemented using the Data Center Management Interface (DCMI), a standardized API for monitoring and managing data center infrastructure. It enables unified management of hardware such as servers, storage, and network devices, including hardware status, performance metrics, and power consumption data. This helps automate data center operations and improves management efficiency. In this way, before collecting performance indicators, this embodiment first verifies the local accelerator card driver and the corresponding device management library of the target virtual machine to avoid the situation where performance indicator collection is abnormal due to abnormalities in the local accelerator card driver and the device management library of the AI ​​accelerator card; the accelerator card driver and the device management library are verified separately to avoid library missing errors caused by the scenario where the accelerator card driver and the device management library are installed separately; an explicit post-loading method is used to load and call functions and collect indicators, thereby eliminating the strong dependency between components and device management dynamic link libraries. At compile time, it is not necessary to configure all device management libraries in the environment. As long as the AI ​​accelerator card type is in the development adaptation list, the underlying collection logic can be ignored, and the fixed interface collection can be called directly to obtain monitoring data with unified variable types, thereby facilitating subsequent data submission, storage, and display; common performance indicators of AI accelerator cards are selected for collection to avoid abnormalities in indicator collection due to the fact that some performance indicators are not supported by the accelerator card.

[0071] Step S14: Save the performance indicator data to a preset database, judge the performance indicator data according to preset performance indicator alarm rules, and display the performance indicator data on the target front end so that the AI ​​accelerator card can be operated and maintained according to the performance indicator data.

[0072] In this embodiment, after obtaining the performance indicator data, it is saved in a preset database, and the performance indicator data is judged according to the preset performance indicator alarm rules to detect whether the accelerator card has an abnormality through the indicators, and the performance indicator data is displayed on the target front end so that the AI ​​accelerator card can be operated and maintained according to the performance indicator data.

[0073] Specifically, the obtained performance indicator data is subjected to unit standardization processing to obtain processed performance indicator data with consistent indicator types, and the processed performance indicator data is saved in a preset database; the processed performance indicator data in the preset database is judged using preset performance indicator alarm rules; if there is target performance indicator data that matches the performance indicator alarm rules in the processed performance indicator data, an alarm is issued for the target performance indicator data, and the target performance indicator data is analyzed to obtain corresponding analysis results; a corresponding performance indicator curve is generated according to the performance indicator data, and the performance indicator curve and the analysis results of the target performance indicator data are displayed in the target front end, so as to operate and maintain the AI ​​accelerator card according to the performance indicator curve and the analysis results. Among them, the target front end is the front-end page of the cloud platform, which is used to intuitively display the curves of various performance indicators, including the memory utilization curve, temperature change curve, etc. of the AI ​​accelerator card, so that the administrator can intuitively understand the operation trend of the AI ​​accelerator card. That is, the obtained performance indicator data is first subjected to unit standardization processing to obtain processed performance indicator data with consistent indicator types, and the processed performance indicator data is saved in a preset database for subsequent query and analysis, wherein the format of the processed performance indicator data is as follows:

[0074] {

[0075] "return": {

[0076] "monitor": [

[0077] {

[0078] "dimension": {

[0079] "data_type": "GPU",

[0080] "device_type": "10de:2236",

[0081] "device_name": "NVIDIA A10",

[0082] "id": "e924ee80-8fbc-6a29-3794-b912d9c57fe2"

[0083] },

[0084] "metrics": {

[0085] "vm_gpu_power_draw": 24.002,

[0086] "vm_gpu_utilization": 0,

[0087] "vm_gpu_memory_utilization": 0,

[0088] "vm_gpu_health": 1,

[0089] "vm_gpu_memory_used_space": 0.267517

[0090] }

[0091] } ]

[0093] }

[0094] }

[0095] Then, the processed performance indicator data in the preset database is judged according to the pre-set performance indicator alarm rules. If the judgment result indicates that the processed performance indicator data contains target performance indicator data that matches the performance indicator alarm rules, it means that the target performance indicator data exceeds the normal range and an alarm needs to be issued for the target performance indicator data. For example, if the temperature of the AI ​​accelerator card exceeds the safety threshold, an alarm process is immediately initiated to notify relevant personnel to handle it. At the same time, a detailed analysis and processing function is provided for the target performance indicator data, and it is analyzed to obtain corresponding analysis results, helping administrators to quickly locate the root cause of the problem and take corresponding measures based on the analysis results. Subsequently, a corresponding performance indicator curve is generated based on the performance indicator data, and the performance indicator curve and the analysis results are displayed to the target front end, so that the AI ​​accelerator card can be operated and maintained based on the performance indicator curve and the analysis results. In this way, after obtaining the performance indicator data, this embodiment saves it to a preset database for subsequent query and analysis; displays the performance indicator curve of the performance indicator data on the target front end, so that the administrator can intuitively understand the operating trend of the AI ​​accelerator card; provides detailed analysis and processing functions for the target performance indicator data that generates alarms, helping administrators to quickly locate the root cause of the problem based on the analysis results and take corresponding measures, thereby improving the operation and maintenance efficiency of the accelerator card.

[0096] As can be seen from the above, the target virtual machine of the embodiment of the present application first establishes a communication connection with the target host, receives a performance indicator collection request sent by the target host based on a preset frequency, determines its own target peripheral interface device information, identifies the number and type of currently mounted AI accelerator cards through a preset accelerator card manufacturer dictionary table, and initializes the performance indicator collection thread accordingly. Then, after the accelerator card driver and the device management library are verified, the device management library is called based on the performance indicator collection request and the performance indicator collection thread to collect performance indicator data. Finally, the collected data is stored in a preset database, judged according to the preset alarm rules, and displayed on the front end to facilitate operation and maintenance of the AI ​​accelerator card.In this way, through the above process of the embodiment of the present application, on the one hand, after the target virtual machine receives the performance indicator collection request sent by the target host, it first determines its own target peripheral interface device information, so as to subsequently identify the AI ​​accelerator card based on the information, thereby improving the accuracy of performance indicator collection; on the other hand, the target host is controlled to initiate a performance indicator collection request based on the preset performance indicator collection frequency, which can ensure the real-time nature of the data and the rational use of system resources; on the other hand, the number and type of accelerator cards mounted on the target virtual machine are determined based on the preset accelerator card manufacturer dictionary table and the determined target peripheral interface device information, which can accurately identify the AI ​​accelerator card to determine the adapted device management library, thereby facilitating Accelerator card operation and maintenance operations; on the one hand, a corresponding performance indicator collection thread is started for each AI accelerator card, which can realize multi-threaded indicator collection, improve the collection speed of performance indicators, and thus improve the efficiency of accelerator card operation and maintenance; on the one hand, before collecting performance indicators, the local accelerator card driver and the corresponding device management library of the target virtual machine are verified to avoid the situation where the performance indicator collection is abnormal due to abnormalities in the local accelerator card driver and the device management library of the AI ​​accelerator card; on the other hand, the accelerator card driver and the device management library are verified separately to avoid the library missing error caused by the scenario where the accelerator card driver and the device management library are installed separately; on the other hand, explicit post-loading is used The function is loaded and called and indicators are collected in a way to eliminate the strong dependency between the component and the device management dynamic link library. At compile time, it is not necessary to configure all the device management libraries in the environment. As long as the AI ​​accelerator card type is in the development adaptation list, the underlying collection logic can be ignored, and the fixed interface collection can be directly called to obtain monitoring data with unified variable types, which is convenient for subsequent data submission, storage, and display. On the one hand, the common performance indicators of the AI ​​accelerator card are selected for collection to avoid abnormalities in indicator collection due to the fact that some performance indicators are not supported by the accelerator card. On the one hand, after obtaining the performance indicator data, it is saved in the preset database for subsequent query and analysis. On the other hand, the performance indicator data is stored in the preset database for subsequent query and analysis. The performance indicator curve of the energy indicator data can be displayed on the target front end, allowing administrators to intuitively understand the operating trends of the AI ​​accelerator card. On the one hand, it provides detailed analysis and processing functions for the target performance indicator data that generates alarms, helping administrators to quickly locate the root cause of the problem and take corresponding measures based on the analysis results, thereby improving the refinement of hardware resource management and operation and maintenance efficiency, and ensuring the stable and efficient operation of the AI ​​accelerator card. On the other hand, only one adaptation development is required. If the AI ​​accelerator card type needs to be changed, only the acquisition controller and other plug-ins need to be adapted, developed and upgraded, which can reduce the development workload and thereby improve the management efficiency of the AI ​​accelerator card to ensure the overall operation efficiency and resource utilization of the cloud computing platform.

[0097] Accordingly, see Figure 3As shown, the embodiment of the present application also provides an AI accelerator card operation and maintenance device based on performance indicator collection, which is applied to a target virtual machine and includes:

[0098] An information determination module 11 is configured to, after establishing a communication connection with a target host, obtain a performance indicator collection request sent by the target host and determine target peripheral interface device information of the target virtual machine; the performance indicator collection request is a request initiated by the target host based on a preset performance indicator collection frequency;

[0099] A thread initialization module 12 is configured to determine the number of accelerator cards and the corresponding accelerator card types of the currently mounted AI accelerator cards using a preset accelerator card manufacturer dictionary table and the target peripheral interface device information, and to initialize a performance indicator collection thread based on the number of accelerator cards and the accelerator card types.

[0100] The indicator collection module 13 is used to verify the local accelerator card driver and the device management library of the AI ​​accelerator card. After the verification is passed, based on the performance indicator collection request and the performance indicator collection thread, the device management library is called to collect the performance indicators of the AI ​​accelerator card to obtain corresponding performance indicator data;

[0101] The data display module 14 is used to save the performance indicator data to a preset database, judge the performance indicator data according to preset performance indicator alarm rules, and display the performance indicator data on the target front end so as to operate and maintain the AI ​​accelerator card based on the performance indicator data.

[0102] As can be seen from the above, the target virtual machine of the embodiment of the present application first establishes a communication connection with the target host, receives a performance indicator collection request sent by the target host based on a preset frequency, determines its own target peripheral interface device information, identifies the number and type of the currently mounted AI accelerator cards through a preset accelerator card manufacturer dictionary table, and initializes the performance indicator collection thread accordingly. Then, after the accelerator card driver and device management library are verified, the performance indicator collection request and the performance indicator collection thread call the device management library to collect performance indicator data. Finally, the collected data is stored in a preset database, judged according to the preset alarm rules and displayed on the front end to enable operation and maintenance of the AI ​​accelerator card. In this way, through the above process of the embodiment of the present application, it is possible to automatically identify the accelerator card configuration, verify the validity of the driver and library files, and realize the timed collection and structured storage of performance indicators; combined with the visual display and alarm rules of the target front end, it can help operation and maintenance personnel monitor the operating status of the AI ​​accelerator card in real time, promptly discover performance anomalies or potential faults, improve the refinement of hardware resource management and operation and maintenance efficiency, ensure the stable and efficient operation of the AI ​​accelerator card, and thus improve the management efficiency of the AI ​​accelerator card to ensure the overall operation efficiency and resource utilization of the cloud computing platform.

[0103] In some specific implementations, the information determination module 11 may specifically include:

[0104] a channel monitoring unit configured to, after establishing a communication connection with a target host based on a preset communication protocol, continuously monitor the established communication channel corresponding to the preset communication protocol, so that the target host can initiate a performance indicator collection request to the target virtual machine through the communication channel based on a preset performance indicator collection frequency;

[0105] a request acquiring unit, configured to acquire the performance indicator collection request sent by the target host through the communication channel;

[0106] The file search unit is used to search for a target file located locally in a preset path, so as to obtain the target peripheral interface device information of the target virtual machine from the target file.

[0107] In some specific implementations, the thread initialization module 12 may specifically include:

[0108] an information parsing unit, configured to parse the target peripheral interface device information to extract the accelerator card manufacturer code and the accelerator card type code recorded in the target peripheral interface device information;

[0109] a dictionary table search unit, configured to search a preset accelerator card manufacturer dictionary table to determine an accelerator card type corresponding to both the accelerator card manufacturer code and the accelerator card type code in the preset accelerator card manufacturer dictionary table;

[0110] The quantity determining unit is configured to determine the number of the extracted accelerator card type codes as the number of accelerator cards of the currently mounted AI accelerator cards.

[0111] In some specific implementations, the thread initialization module 12 may specifically include:

[0112] A thread starting unit, configured to start a preset number of performance indicator collection threads according to the number of the accelerator cards; the preset number is consistent with the number of the accelerator cards;

[0113] a code name determining unit, configured to determine the accelerator card manufacturer code name of the accelerator card type corresponding to the performance indicator collection thread;

[0114] The management library determining unit is configured to determine the device management library of the corresponding accelerator card manufacturer based on the accelerator card manufacturer code, so as to initialize the performance indicator collection thread using the device management library.

[0115] In some specific implementations, the indicator collection module 13 may specifically include:

[0116] A condition judgment unit is used to judge whether the local accelerator card driver is successfully identified and loaded normally, so as to obtain a corresponding judgment result;

[0117] a management library importing unit, configured to, if the judgment result indicates that the local accelerator card driver is successfully identified and loaded normally, determine that the accelerator card driver verification has passed, and import the device management library of the AI ​​accelerator card to detect whether the device management library has been installed;

[0118] The result determination unit is configured to determine that the device management library has been installed if the import operation on the device management library is successfully implemented, and obtain a verification result that both the accelerator card driver and the device management library have passed verification.

[0119] In some specific implementations, the indicator collection module 13 may specifically include:

[0120] An accelerator card determination unit, configured to determine a target accelerator card type and a target AI accelerator card corresponding to the performance indicator collection thread;

[0121] An indicator synthesis unit is configured to synthesize the performance indicators corresponding to the current target accelerator card type based on the performance indicator collection request to determine a target performance indicator to be collected; the target performance indicator is a performance indicator supported by the target accelerator card type among the performance indicators corresponding to the target accelerator card types;

[0122] A management library calling unit is used to determine a corresponding target device management library according to the target accelerator card type, and call the target device management library through the performance indicator collection thread to collect the target performance indicator of the target AI accelerator card based on the collection tool in the target device management library.

[0123] In some specific implementations, the data display module 14 may specifically include:

[0124] a data storage unit, configured to perform unit standardization processing on the obtained performance indicator data to obtain processed performance indicator data with consistent indicator types, and save the processed performance indicator data into a preset database;

[0125] a data judging unit, configured to judge the processed performance indicator data in the preset database using a preset performance indicator alarm rule;

[0126] a data analysis unit configured to, if target performance indicator data matching the performance indicator alarm rule exists in the processed performance indicator data, generate an alarm for the target performance indicator data, and analyze the target performance indicator data to obtain a corresponding analysis result;

[0127] A result display unit is used to generate a corresponding performance indicator curve based on the performance indicator data, and display the performance indicator curve and the analysis results of the target performance indicator data to the target front end, so as to operate and maintain the AI ​​accelerator card based on the performance indicator curve and the analysis results.

[0128] Furthermore, the embodiment of the present application also discloses an electronic device, Figure 4 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram cannot be considered as any limitation on the scope of use of this 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 ​​accelerator card operation and maintenance method based on performance indicator collection disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0129] In this embodiment, the power supply 23 is used to provide operating 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. The communication protocol it follows is any communication protocol that can be applied to the technical solution of this 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. Its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0130] In addition, the memory 22 as a carrier for resource storage can be a read-only memory, random access memory, disk or 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.

[0131] The operating system 221 is used to manage and control the hardware devices and computer program 222 on the electronic device 20. It can be Windows Server, Netware, Unix, Linux, etc. In addition to including computer programs that can be used to implement the AI ​​accelerator card operation and maintenance method based on performance indicator collection performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include computer programs that can be used to perform other specific tasks.

[0132] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when executed by a processor, the computer program implements the aforementioned disclosed method for operating and maintaining an AI accelerator card based on performance indicator collection. The specific steps of this method can be referred to the corresponding content disclosed in the aforementioned embodiments and will not be repeated here.

[0133] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference can be made to the descriptions of the identical or similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and the relevant parts can be referred to the descriptions of the methods.

[0134] 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 above description has generally described the components and steps of each example according to their functions. 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 beyond the scope of this application.

[0135] The steps of the methods or algorithms 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 random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0136] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0137] The above is a detailed introduction to the technical solution provided by the present application. Specific examples are used herein 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 those skilled in the art, according to the ideas of the present application, there may 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. An AI accelerator card operation and maintenance method based on performance indicator collection, characterized in that: Applied to the target virtual machine, including: After establishing a communication connection with a target host, obtaining a performance indicator collection request sent by the target host and determining target peripheral interface device information of the target virtual machine; the performance indicator collection request is a request initiated by the target host based on a preset performance indicator collection frequency; Determine the number of currently mounted AI accelerator cards and the corresponding accelerator card types using a preset accelerator card manufacturer dictionary table and the target peripheral interface device information, and initialize a performance indicator collection thread based on the number of accelerator cards and the accelerator card types. Verify the local accelerator card driver and the device management library of the AI ​​accelerator card. After both verifications pass, call the device management library to collect performance indicators of the AI ​​accelerator card based on the performance indicator collection request and the performance indicator collection thread to obtain corresponding performance indicator data. The performance indicator data is saved in a preset database, and the performance indicator data is judged according to the preset performance indicator alarm rules, and the performance indicator data is displayed on the target front end, so that the AI ​​accelerator card can be operated and maintained according to the performance indicator data.

2. The AI ​​accelerator card operation and maintenance method based on performance indicator collection according to claim 1 is characterized in that: After establishing a communication connection with the target host, obtaining a performance indicator collection request sent by the target host and determining target peripheral interface device information of the target virtual machine include: After establishing a communication connection with the target host based on a preset communication protocol, continuously monitoring the established communication channel corresponding to the preset communication protocol, so that the target host can initiate a performance indicator collection request to the target virtual machine through the communication channel based on a preset performance indicator collection frequency; Obtaining the performance indicator collection request sent by the target host through the communication channel; A target file located locally in a preset path is searched to obtain target peripheral interface device information of the target virtual machine from the target file.

3. The AI ​​accelerator card operation and maintenance method based on performance indicator collection according to claim 1 is characterized in that: The determining the number of accelerator cards and the corresponding accelerator card types of the currently mounted AI accelerator cards by using the preset accelerator card manufacturer dictionary table and the target peripheral interface device information includes: Parsing the target peripheral interface device information to extract the accelerator card manufacturer code and the accelerator card type code recorded in the target peripheral interface device information; Searching a preset accelerator card manufacturer dictionary table to determine an accelerator card type corresponding to both the accelerator card manufacturer code and the accelerator card type code in the preset accelerator card manufacturer dictionary table; The number of the extracted accelerator card type codes is determined as the number of accelerator cards of the currently mounted AI accelerator card.

4. The AI ​​accelerator card operation and maintenance method based on performance indicator collection according to claim 3 is characterized in that: Initializing the performance indicator collection thread according to the number and type of the accelerator cards includes: Starting a preset number of performance indicator collection threads according to the number of accelerator cards; the preset number is consistent with the number of accelerator cards; Determine the accelerator card manufacturer code of the accelerator card type corresponding to the performance indicator collection thread; The device management library of the corresponding accelerator card manufacturer is determined based on the accelerator card manufacturer code, so as to initialize the performance indicator collection thread using the device management library.

5. The AI ​​accelerator card operation and maintenance method based on performance indicator collection according to claim 1 is characterized in that: Verifying the local accelerator card driver and the device management library of the AI ​​accelerator card includes: Determine whether the local accelerator card driver is successfully identified and loaded normally to obtain the corresponding judgment result; If the judgment result indicates that the local accelerator card driver is successfully identified and loaded normally, then the accelerator card driver verification is determined to be passed, and an import operation is performed on the device management library of the AI ​​accelerator card to detect whether the device management library has been installed; If the import operation of the device management library is successfully implemented, it is determined that the device management library has been installed, and a verification result is obtained that both the accelerator card driver and the device management library have passed verification.

6. The AI ​​accelerator card operation and maintenance method based on performance indicator collection according to claim 1 is characterized in that: The calling the device management library to collect the performance indicators of the AI ​​accelerator card based on the performance indicator collection request and the performance indicator collection thread includes: Determine the target accelerator card type and target AI accelerator card corresponding to the performance indicator collection thread; Based on the performance indicator collection request, the performance indicators corresponding to the current target accelerator card type are combined to determine the target performance indicator to be collected; the target performance indicator is a performance indicator supported by the target accelerator card type among the performance indicators corresponding to the target accelerator card types; Determine the corresponding target device management library according to the target accelerator card type, and call the target device management library through the performance indicator collection thread to collect the target performance indicator of the target AI accelerator card based on the collection tool in the target device management library.

7. The AI ​​accelerator card operation and maintenance method based on performance indicator collection according to any one of claims 1 to 6, characterized in that: Saving the performance indicator data in a preset database, judging the performance indicator data according to preset performance indicator alarm rules, and displaying the performance indicator data on a target front end so as to operate and maintain the AI ​​accelerator card according to the performance indicator data includes: Performing unit standardization processing on the obtained performance indicator data to obtain processed performance indicator data with consistent indicator types, and saving the processed performance indicator data in a preset database; Using preset performance indicator alarm rules to judge the processed performance indicator data in the preset database; If target performance indicator data matching the performance indicator alarm rule exists in the processed performance indicator data, an alarm is issued for the target performance indicator data, and the target performance indicator data is analyzed to obtain corresponding analysis results; A corresponding performance indicator curve is generated according to the performance indicator data, and the performance indicator curve and the analysis results of the target performance indicator data are displayed to the target front end, so that the AI ​​accelerator card can be operated and maintained according to the performance indicator curve and the analysis results.

8. An AI accelerator card operation and maintenance device based on performance indicator collection, characterized in that: Applied to the target virtual machine, including: An information determination module is configured to, after establishing a communication connection with a target host, obtain a performance indicator collection request sent by the target host and determine target peripheral interface device information of the target virtual machine; the performance indicator collection request is a request initiated by the target host based on a preset performance indicator collection frequency; A thread initialization module is configured to determine the number of accelerator cards and the corresponding accelerator card types of the currently mounted AI accelerator cards using a preset accelerator card manufacturer dictionary table and the target peripheral interface device information, and to initialize a performance indicator collection thread based on the number of accelerator cards and the accelerator card types. An indicator collection module is used to verify the local accelerator card driver and the device management library of the AI ​​accelerator card. After both verifications are passed, based on the performance indicator collection request and the performance indicator collection thread, the module calls the device management library to collect the performance indicators of the AI ​​accelerator card to obtain corresponding performance indicator data. A data display module is used to save the performance indicator data to a preset database, judge the performance indicator data according to preset performance indicator alarm rules, and display the performance indicator data on the target front end so as to operate and maintain the AI ​​accelerator card based on the performance indicator data.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the AI ​​accelerator card operation and maintenance method based on performance indicator collection as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that Used to store computer programs; wherein, when the computer program is executed by a processor, it implements the AI ​​accelerator card operation and maintenance method based on performance indicator collection as described in any one of claims 1 to 7.

Images