Multi-dimensional performance evaluation method and server terminal for AI computing server
By collecting and analyzing data on fan speed, terminal temperature, and memory utilization, an AI computing server monitoring model was established to monitor and adjust performance in real time. This solved the problems of high energy consumption and data security of AI computing servers and simplified multi-dimensional performance evaluation and management.
Patent Information
- Application Number
- CN202411095489.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-08-12
AI Technical Summary
Existing AI computing server performance evaluation methods cannot provide a comprehensive understanding of server performance, cannot monitor and detect potential problems in real time, and pose risks of high energy consumption, heat management, and data security.
By collecting data on server fan speed, terminal temperature, and memory utilization, the data is preprocessed and then input into the input layer for correlation processing. An AI computing server monitoring model is established to monitor and analyze heterogeneous computing performance, energy efficiency, and security performance in real time. The model is then used for regulation and early warning through backpropagation of error terms.
It enables multi-dimensional performance evaluation of AI computing servers, real-time monitoring of performance indicators, provides an intuitive visualization interface, simplifies management, improves management efficiency, and has continuous learning and optimization capabilities.
Smart Images

Figure CN118626358B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of AI computing power server performance evaluation, and discloses a multi-dimensional performance evaluation method and a server terminal for an AI computing power server. Background Art
[0002] With the rapid development of artificial intelligence technology, the hardware configuration of AI computing servers is also constantly improving. In terms of processors, GPUs (graphics processing units) are still the mainstream choice, but there are also accelerators designed specifically for AI that can significantly accelerate the execution of deep learning tasks. The demand for storage capacity and speed of AI computing servers is also increasing. At the same time, the application of high-speed memory (such as DDR4 and HBM) is also improving the data processing capabilities of servers. At the same time, the multi-dimensional performance evaluation methods of AI computing servers are also multifaceted, including: computing performance evaluation, storage performance evaluation, evaluation of communication performance between servers, including bandwidth, latency and throughput. Commonly used indicators include Gb / s (amount of data transmitted per second) and ms (network latency), evaluation of server energy consumption under the same workload, and the balance between energy consumption and performance. Commonly used indicators include power consumption, power efficiency and energy efficiency ratio.
[0003] However, at present, with the rapid development of artificial intelligence, AI computing servers also have many shortcomings. For example, the construction and maintenance costs of AI computing servers are high, which is often difficult to afford, especially for small-scale enterprises or individual users. The high computing power and storage requirements of AI computing servers lead to higher energy consumption, which puts a certain pressure on energy resources, and energy efficiency issues need to be solved to reduce energy consumption. High-performance AI computing servers generate a lot of heat, and an effective cooling system is required to maintain stable operation, prevent faults and repair them in time. Since AI computing servers process a large amount of sensitive data, higher requirements are placed on data security and privacy protection, and data encryption, access control and privacy protection measures need to be strengthened.
[0004] For example, the existing patent with the publication number CN116382843A discloses an industrial AI computing power PaaS platform based on Kubernetes container technology, including: an infrastructure layer for providing cloud servers, network resources and system support; a platform layer, including a business middle platform for technical operations, data operations, content operations and business operations, and a data middle platform for data source collection, data storage, data calculation and management of image resources based on the Harbor private image repository; a service layer, including AI data services and cloud service SDKs. This invention can provide users with application services, AI training, AI online services, training model management, training environment management, resource management and other functions and services, and realize one-stop AI algorithm computing power task hosting for process industrial production processes; it realizes precise management of cloud resources, and can continuously integrate and continuously deploy AI algorithm models designed according to different production processes. The training service encapsulates the training algorithm based on Docker images and container technology.
[0005] The above patent cannot fully understand the performance of the computing server, cannot collect as much data as possible, and find out potential problems with the server by mining data features. It does not integrate artificial intelligence algorithms and algorithm clusters to monitor the server in real time. Summary of the invention
[0006] In order to solve the above technical problems, the main purpose of the present invention is to provide a multi-dimensional performance evaluation method for AI computing servers, including:
[0007] S1. Collecting initial data of the AI computing server, where the initial data includes fan speed, server terminal temperature, and server memory utilization;
[0008] S2. Preprocess the initial data to obtain a time series set, where the time series set includes time series data, and input the time series data into an input layer for data analysis;
[0009] The data preprocessing of the initial data is to remove redundant and erroneous data;
[0010] The input layer serves as the input port for data analysis and is used to receive pre-processed server memory utilization data, fan speed data, and server terminal temperature data;
[0011] The input layer performs correlation processing on the input preprocessed initial data, obtains a time series set, and converts the preprocessed initial data into time series data. The calculation expression of the time series set is as follows:
[0012] ;
[0013] Where X is a set of time series, is the nth time series data input, where n is 1, 2, 3, ...;
[0014] The time series data includes server memory utilization data, fan speed data and server terminal temperature data, and the calculation expression is as follows:
[0015] ;
[0016] in, is the memory utilization data of the nth group of servers, is the nth group of fan speed data, The temperature data of the nth group of server terminals;
[0017] S3. Integrate the time series data through the input layer to obtain the characteristic values of the server memory utilization data, fan speed data, and server terminal temperature data of the AI computing server, and then monitor the real-time operation status of the AI computing server;
[0018] S4. Monitor the fan speed, server terminal temperature and server memory utilization through the AI computing server monitoring model, and then analyze the real-time operation status of the AI computing server to obtain the heterogeneous computing performance, energy efficiency performance and security performance of the AI computing server;
[0019] Perform weight analysis on the time series set, and obtain the weight of the input vector at the current moment and the weight of the output vector at the previous moment. By establishing an AI computing power server monitoring model, the server memory utilization data, fan speed data, and server terminal temperature data are monitored in real time. The calculation expression of the AI computing power server monitoring model is as follows:
[0020] ;
[0021] in, Monitor the output of the model for the AI computing server. is the weight of the time series set input at the current moment, is the time series set input at the current moment, is the weight of the time series set input at the last moment, The time series set input at the last moment;
[0022] make is the weighted input at time t, the calculation expression is as follows:
[0023] ;
[0024] in, is the weighted input;
[0025] S5. Back-propagating the AI computing server monitoring model through the error term, thereby regulating the operation of the AI computing server, and also regulating the operation of the AI computing server terminal;
[0026] By calculating the error term of the AI computing server monitoring model at time t, the heterogeneous computing performance, energy efficiency performance, and security performance of the AI computing server are obtained. The calculation expression is as follows:
[0027] ;
[0028] in, is the error term, Find the sign of the partial derivative of a function;
[0029] The error term reflects the heterogeneous computing performance mapped by the memory utilization data, the energy efficiency performance mapped by the fan speed data, and the security performance mapped by the server terminal temperature data based on the error value between the preset threshold of the AI computing server and the actual time series data.
[0030] S6. If the automatic control of the AI computing power server and the AI computing power server terminal is abnormal and exceeds the preset threshold, an early warning is issued to the AI computing power server and the AI computing power server terminal, and early warning data is obtained;
[0031] S7. The AI computing server monitoring model monitoring, automatic control and early warning data are displayed through a visual screen.
[0032] As a preferred solution of the multi-dimensional performance evaluation method for AI computing power servers of the present invention, wherein:
[0033] The server memory utilization, fan speed and server terminal temperature are all performance indicators that need to be monitored;
[0034] Collect data on server memory utilization, fan speed, and server terminal temperature through sensors, monitoring software, and data collection tools;
[0035] Establish a central database to store and manage the collected data on server memory utilization, fan speed, and server terminal temperature.
[0036] As a preferred solution of the multi-dimensional performance evaluation method for AI computing power servers of the present invention, wherein:
[0037] The preset thresholds include a preset performance threshold, a preset speed threshold and a preset temperature threshold;
[0038] If the heterogeneous computing performance is less than the preset performance threshold, the memory of the AI computing server is not full and the heterogeneous computing performance can be extended. If the heterogeneous computing performance is greater than the preset performance threshold, the memory of the AI computing server is full and the heterogeneous computing performance cannot be extended. The cache of the AI computing server is cleared.
[0039] The energy efficiency performance of the AI computing server is judged by the fan speed. If the fan speed reaches the preset speed threshold, the AI computing server is decompressed. If the fan speed does not reach the preset speed threshold, the AI computing server is not controlled.
[0040] The safety of the AI computing server terminal is judged through the server terminal temperature data. If the temperature does not reach the preset temperature threshold, the AI computing server terminal operates normally. If the temperature exceeds the preset temperature threshold, the AI computing server terminal is overloaded and needs to be regulated.
[0041] As a preferred solution of the multi-dimensional performance evaluation method for AI computing power servers of the present invention, wherein:
[0042] The AI computing server monitoring model is back-propagated through the error term to regulate the operation of the AI computing server, and the operation of the AI computing server terminal is also regulated;
[0043] The error term is output to the AI computing server monitoring model through the back propagation function to optimize the AI computing server and AI computing server terminal. The calculation expression is as follows:
[0044] ;
[0045] in, is the correction value, T is the matrix transpose, and diag[] is the function for constructing a diagonal matrix;
[0046] The correction value is obtained by using the error term at the previous moment and the initial error term to optimize the AI computing power server and AI computing power server terminal at any moment.
[0047] As a preferred solution of the multi-dimensional performance evaluation method for AI computing power servers of the present invention, wherein:
[0048] Early warning is provided for the AI computing server and the AI computing server terminal, abnormal performance or failure of the AI computing server is discovered in time, and an alarm message is sent to the administrator. At the same time, threshold rules are set. When the performance indicators do not meet the threshold rules, an alarm and emergency processing process is triggered;
[0049] Store historical data and optimize and upgrade the AI computing server monitoring model.
[0050] The present invention discloses a multi-dimensional performance evaluation server terminal for an AI computing server, wherein:
[0051] A data collection module is used to collect initial data of the AI computing server, wherein the initial data includes fan speed, server terminal temperature, and server memory utilization;
[0052] A data processing and analysis module, including a data processing unit for preprocessing initial data, an input layer for performing feature analysis and weight calculation on the preprocessed initial data, and an AI computing server monitoring model for performing parameter monitoring on the AI computing server and the AI computing server terminal;
[0053] A multi-dimensional performance evaluation module, including a feedback unit for feeding back the error term of the AI computing power server monitoring model at time t, and a performance evaluation unit for receiving the error term and adjusting and controlling the performance parameters of the AI computing power server and the AI computing power server terminal;
[0054] Visualization module, used to display the performance of AI computing server and AI computing server terminal;
[0055] The alarm and monitoring module is used to warn of abnormal conditions in AI computing servers and AI computing server terminals.
[0056] An electronic device includes a memory for storing instructions; and a processor for executing the instructions, so that the device implements a multi-dimensional performance evaluation method for an AI computing server.
[0057] A computer-readable storage medium stores a computer program, which, when executed, implements a multi-dimensional performance evaluation method for an AI computing power server.
[0058] Beneficial effects of the present invention:
[0059] By collecting and analyzing multi-dimensional performance data, the present invention can help users fully understand the performance of the server in different aspects, which helps to find bottlenecks and problems and perform targeted optimization;
[0060] The present invention can monitor the performance indicators of the server in real time and set threshold rules for alarm;
[0061] The present invention provides an intuitive visual interface, which displays performance data in the form of charts, curves, etc., so that users can quickly understand the performance of the server. This simplifies the work of server management and maintenance. Users can intuitively view performance data through the interface, identify problems and take corresponding measures, thereby improving management efficiency;
[0062] The present invention has the ability of continuous learning and optimization. By continuously accumulating new performance data and historical data, the present invention can continuously improve the algorithm model, improve the accuracy and efficiency of performance evaluation and optimization, and gradually improve the intelligence level of the multi-dimensional performance evaluation server terminal. BRIEF DESCRIPTION OF THE DRAWINGS
[0063] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:
[0064] Figure 1 This is a system flow chart of the multi-dimensional performance evaluation method for AI computing power servers of the present invention;
[0065] Figure 2 This is a diagram showing the composition of a multi-dimensional performance evaluation server terminal for an AI computing server according to the present invention;
[0066] Figure 3 The present invention is a monitoring diagram for optimizing and simulating server memory utilization data, fan speed data, and server terminal temperature data after error term feedback. DETAILED DESCRIPTION
[0067] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.
[0068] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0069] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.
[0070] Embodiment 1
[0071] like Figure 1 As shown, a multi-dimensional performance evaluation method and server terminal for AI computing power servers include:
[0072] S1. Collect the initial data of the AI computing server, which includes the fan speed, server terminal temperature, and server memory utilization;
[0073] Server memory utilization, fan speed, and server terminal temperature are all performance indicators that need to be monitored;
[0074] According to the needs and characteristics of the server, choose the appropriate data collection tool, which can be a sensor, monitoring software or a custom data collection tool. For example, you can use sensors to monitor the server's temperature, fan speed and other information; use monitoring software to collect information such as the server's operating status and resource utilization, or develop a custom data collection tool to obtain specific performance indicators;
[0075] Establish a central database to store and manage collected data;
[0076] S2. Preprocess the initial data to obtain a time series set, where the time series set includes time series data, and input the time series data into an input layer for data analysis;
[0077] The input layer serves as the input port for data analysis and is used to receive pre-processed server memory utilization data, fan speed data, and server terminal temperature data;
[0078] Since AI computing servers process algorithmic issues in many aspects, and the performance indicators of server terminals are not just server terminal temperature, server memory utilization data, fan speed data, and server terminal temperature data are only collected as three items of data for AI computing server instances. Users can collect other data according to actual application needs, including computing power, memory utilization, task scheduling efficiency, real-time performance, etc.
[0079] The input layer processes the input preprocessed data into a time series set by correlation. The calculation expression is as follows:
[0080] ;
[0081] Where X is a set of time series, is the nth time series data input, where n is 1, 2, 3, ...;
[0082] The input time series data includes server memory utilization data, fan speed data, and server terminal temperature data. The calculation expression is as follows:
[0083] ;
[0084] in, is the memory utilization data of n groups of servers, is n sets of fan speed data, is n groups of server terminal temperature data;
[0085] S3. By integrating the time series data processed by the input layer, characteristic values of server memory utilization data, fan speed data, and server terminal temperature data of the AI computing server are obtained, thereby monitoring the real-time operating status of the AI computing server.
[0086] Perform weight analysis on the time series set, and obtain the weight of the input vector at the current moment and the weight of the output vector at the previous moment. By establishing an AI computing power server monitoring model, the server memory utilization data, fan speed data, and server terminal temperature data are monitored in real time. The calculation expression of the AI computing power server monitoring model is as follows:
[0087] ;
[0088] in, Monitor the output of the model for the AI computing server. is the weight of the time series set input at the current moment, is the time series set input at the current moment, is the weight of the time series set input at the last moment, The time series set input at the last moment;
[0089] Further, is the weighted input at time t, the calculation expression is as follows:
[0090] ;
[0091] in, is the weighted input.
[0092] S4. Monitor the fan speed, server terminal temperature and server memory utilization through the AI computing server monitoring model, and then analyze the real-time operation status of the AI computing server to obtain the heterogeneous computing performance, energy efficiency performance and security performance of the AI computing server;
[0093] By calculating the error term of the AI computing server monitoring model at time t, the heterogeneous computing performance, energy efficiency performance, and security performance of the AI computing server are obtained. The calculation expression is as follows:
[0094] ;
[0095] in, is the error term, Find the sign of the partial derivative of a function;
[0096] The error term reflects the heterogeneous computing performance mapped by the memory utilization data, the energy efficiency performance mapped by the fan speed data, and the security performance mapped by the server terminal temperature data based on the error value between the preset threshold of the AI computing server and the actual time series data.
[0097] The preset thresholds include a preset performance threshold, a preset speed threshold and a preset temperature threshold;
[0098] If the heterogeneous computing performance is less than the preset performance threshold, the AI computing server memory is not full and the heterogeneous computing performance can be extended. If the heterogeneous computing performance is greater than the preset performance threshold, the AI computing server memory is full and the heterogeneous computing performance cannot be extended, and the cache needs to be cleared.
[0099] The energy efficiency performance of the AI computing server is judged by the fan speed. Since the fan will rotate at a variable speed according to the operation of the AI computing server, if the fan speed reaches the preset speed threshold, the operation of the AI computing server also reaches the threshold, and the AI computing server needs to be decompressed. If the fan speed does not reach the preset speed threshold, the AI computing server does not reach the threshold, and there is no need to control the AI computing server.
[0100] The safety of the AI computing server terminal is judged through the temperature data of the server terminal. Since the chip of the AI computing server terminal generates heat when it runs, the temperature monitoring is also a control of the safe operation of the AI computing server terminal. If the temperature does not reach the preset temperature threshold, the AI computing server terminal runs normally. If the temperature exceeds the preset temperature threshold, the AI computing server terminal is overloaded and needs to be regulated.
[0101] S5. Back-propagating the AI computing server monitoring model through the error term, thereby regulating the operation of the AI computing server, and also regulating the operation of the AI computing server terminal;
[0102] Among them, the error term is output to the AI computing power server monitoring model through the back propagation function to optimize the AI computing power server and AI computing power server terminal. The calculation expression is as follows:
[0103] ;
[0104] in, is the correction value, T is the matrix transpose, and diag[] is the function for constructing a diagonal matrix;
[0105] The constructed diagonal matrix expression is as follows:
[0106] ;
[0107] The correction value is calculated through the error term and the initial error term at the previous moment, thereby optimizing the AI computing power server and AI computing power server terminal at any moment.
[0108] like Figure 3 As shown, after the error term feedback of the server memory utilization, fan speed and server terminal temperature is performed, the parameters of the AI computing server and the AI computing server terminal are adjusted according to the fan speed error value, the server terminal temperature error value and the server memory utilization error value; Figure 3 The x-axis is time, and the y-axis is the feedback error value;
[0109] S6. If the automatic control of the AI computing power server and the server terminal is abnormal and exceeds the preset threshold, an early warning is issued to the AI computing power server and the server terminal, and early warning data is obtained;
[0110] Furthermore, a real-time alarm and monitoring mechanism is introduced to promptly detect server performance anomalies or failures and send alarm information to the administrator. Threshold rules can be set, and when a performance indicator does not meet the threshold rule, an alarm and emergency response process is automatically triggered.
[0111] Furthermore, historical data is stored to optimize and upgrade the AI computing server monitoring model;
[0112] S7. The AI computing server monitoring model monitoring, automatic control and early warning data are displayed through a visual screen.
[0113] Embodiment 2
[0114] like Figure 2 As shown in the figure, the multi-dimensional performance evaluation server terminal for AI computing server includes:
[0115] The data collection module establishes a complete data collection system to monitor the performance indicators of the AI computing server in real time, including computing power, memory utilization, power consumption, task response time, data transmission speed, etc. Data collection can be performed using sensors, monitoring software, or custom data collection tools.
[0116] Data processing and analysis module: The collected performance data needs to be processed and analyzed to extract useful information and indicators. Design corresponding data processing algorithms and analysis models to integrate, count and visualize multi-dimensional data to help users gain a deeper understanding of server performance.
[0117] The multi-dimensional performance evaluation module establishes a multi-dimensional performance evaluation index library, including indicators of multi-task performance, heterogeneous computing performance, real-time performance, energy efficiency performance, scalability, security performance and reliability, so as to comprehensively evaluate server performance.
[0118] The visualization module is designed with an intuitive and easy-to-use visualization interface to display performance data in the form of charts, curves, etc., to help users quickly understand the performance of the server. Users can view the changing trends of different performance indicators through the interface, analyze the root causes of problems and take corresponding optimization measures.
[0119] The alarm and monitoring module introduces real-time alarm and monitoring mechanisms to promptly detect server performance anomalies or failures and send alarm information to administrators. Threshold rules can be set, and when a performance indicator does not meet the threshold rule, the system can automatically trigger the alarm and emergency handling process.
[0120] A multi-dimensional performance evaluation server terminal for AI computing servers. Hardware configuration list:
[0121] Data processing unit, including:
[0122] Multi-dimensional performance evaluation for AI computing servers Server terminal processor: Select a high-performance processor, such as Intel Xeon or AMD EPYC, to provide sufficient computing power.
[0123] Multi-dimensional performance evaluation of AI computing servers Memory of server terminals: Configure large-capacity memory to support the processing of large-scale data and models.
[0124] Multi-dimensional performance evaluation of AI computing servers Storage of server terminals: Choose high-speed solid-state drives (SSDs) to store data and models.
[0125] Multi-dimensional performance evaluation of AI computing servers GPU of server terminals: Integrate powerful GPUs that support deep learning tasks, such as NVIDIA's Tesla or RTX series.
[0126] Multi-dimensional performance evaluation of AI computing servers Network connection of server terminals: Provide high-speed network connection for data transmission and remote access with other devices.
[0127] The server terminal also includes a software operating unit, and the software operating unit includes:
[0128] Multi-dimensional performance evaluation for AI computing servers Operating system of server terminals: Select a suitable operating system, such as Linux or Windows Server, to support the operation and management of AI computing servers.
[0129] Multi-dimensional performance evaluation of AI computing servers AI framework for server terminals: Install and configure popular AI frameworks such as TensorFlow, PyTorch, and Keras for deep learning tasks.
[0130] Drivers and libraries for multi-dimensional performance evaluation of AI computing servers: Install and update GPU drivers and related deep learning libraries to ensure the normal operation of the system.
[0131] Multi-dimensional performance evaluation of AI computing servers Remote management tools for server terminals: Use remote management tools such as SSH or remote desktop to facilitate remote access and management by administrators.
[0132] The server terminal also includes a system management unit: the system management unit is used to monitor and diagnose the multi-dimensional performance evaluation server terminal for AI computing power servers: realize the monitoring and diagnosis functions of the server terminal, and check the status and performance of the server in real time; it is also used to perform resource scheduling on the multi-dimensional performance evaluation server terminal for AI computing power servers: realize the resource scheduling function to reasonably allocate computing resources and GPU resources to different tasks; it is also used for task management: provide task management functions, including task queues, priority settings and task scheduling; it is also used to enhance the security of the server terminal: enhance the security of the server, including access control, identity authentication and data encryption.
[0133] The server terminal also includes a system storage unit, which is used to enhance the scalability of the multi-dimensional performance evaluation server terminal for AI computing servers: scalability is taken into consideration when designing the server terminal so that more computing resources and storage space can be added as needed; it is also used to implement redundancy and backup for the multi-dimensional performance evaluation server terminal for AI computing servers: implement redundancy and backup mechanisms to ensure data security and availability; it is also used to increase the fault tolerance of the multi-dimensional performance evaluation server terminal for AI computing servers: fault-tolerant technologies such as RAID (redundant disk array) are used to protect data from hardware failures. The AI computing server terminal and server terminal in this application both represent a multi-dimensional performance evaluation server terminal for AI computing servers.
[0134] Embodiment 3
[0135] An electronic device includes a memory for storing instructions; a processor for executing instructions, so that the device executes a multi-dimensional performance evaluation method for an AI computing server.
[0136] Embodiment 4
[0137] A computer-readable storage medium stores a computer program, which, when executed, implements a multi-dimensional performance evaluation method for an AI computing power server.
[0138] Importantly, it should be noted that the construction and arrangement of the present application shown in a plurality of different exemplary embodiments are only exemplary. Although only two embodiments are described in detail in this disclosure, it should be readily understood by those who refer to this disclosure that many modifications are possible, for example, the size, scale, structure, shape and ratio of various elements, and parameter values (e.g., temperature, pressure, etc.), installation arrangement, use of materials, color, directional changes, etc., without substantially departing from the novel teachings and advantages of the subject matter described in the application. For example, the element shown as integrally formed can be composed of multiple parts or elements, the position of the element can be inverted or otherwise changed, and the nature or number or position of the discrete element can be changed or changed. Therefore, all such modifications are intended to be included in the scope of the present invention. The order or sequence of any process or method steps can be changed or reordered according to alternative embodiments. Any "device plus function" clause is intended to cover the structure of the execution function described in this article, and is not only structurally equivalent but also equivalent structure. Without departing from the scope of the present invention, other replacements, modifications, changes and omissions can be made in the design, operating conditions and arrangement of the exemplary embodiment. Therefore, the invention is not limited to a specific embodiment, but extends to several modifications still falling within the scope of the appended claims.
[0139] Additionally, in order to provide a concise description of exemplary embodiments, all features of an actual embodiment may not be described (ie, those features that are not relevant to the best mode presently contemplated for carrying out the invention or those features that are not relevant to implementing the invention).
[0140] It should be understood that in the development of any actual implementation, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort may be complex and time-consuming, but for those of ordinary skill having the benefit of this disclosure, the development effort will be a routine task of design, fabrication, and production without undue experimentation.
[0141] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. A multi-dimensional performance evaluation method for AI computing servers, characterized in that: include: S1. Collecting initial data of the AI computing server, where the initial data includes fan speed, server terminal temperature, and server memory utilization; S2. Preprocess the initial data to obtain a time series set, where the time series set includes time series data, and input the time series data into an input layer for data analysis; The data preprocessing of the initial data is to remove redundant and erroneous data; The input layer serves as the input port for data analysis and is used to receive pre-processed server memory utilization data, fan speed data, and server terminal temperature data; The input layer performs correlation processing on the input preprocessed initial data, obtains a time series set, and converts the preprocessed initial data into time series data. The calculation expression of the time series set is as follows: ; Where X is a set of time series, is the nth time series data input, where n is 1, 2, 3, ...; The time series data includes server memory utilization data, fan speed data and server terminal temperature data, and the calculation expression is as follows: ; in, is the memory utilization data of the nth group of servers, is the nth group of fan speed data, The temperature data of the nth group of server terminals; S3. Integrate the time series data through the input layer to obtain the characteristic values of the server memory utilization data, fan speed data, and server terminal temperature data of the AI computing server, and then monitor the real-time operation status of the AI computing server; S4. Monitor the fan speed, server terminal temperature and server memory utilization through the AI computing server monitoring model, and then analyze the real-time operation status of the AI computing server to obtain the heterogeneous computing performance, energy efficiency performance and security performance of the AI computing server; Perform weight analysis on the time series set, and obtain the weight of the input vector at the current moment and the weight of the output vector at the previous moment. By establishing an AI computing power server monitoring model, the server memory utilization data, fan speed data, and server terminal temperature data are monitored in real time. The calculation expression of the AI computing power server monitoring model is as follows: ; in, Monitor the output of the model for the AI computing server. is the weight of the time series set input at the current moment, is the time series set input at the current moment, is the weight of the time series set input at the last moment, The time series set input at the last moment; make is the weighted input at time t, the calculation expression is as follows: ; in, is the weighted input; S5. Back-propagating the AI computing server monitoring model through the error term, and then regulating the operation of the AI computing server, and also regulating the operation of the AI computing server terminal; By calculating the error term of the AI computing server monitoring model at time t, the heterogeneous computing performance, energy efficiency performance, and security performance of the AI computing server are obtained. The calculation expression is as follows: ; in, is the error term, Find the sign of the partial derivative of a function; The error term reflects the heterogeneous computing performance mapped by the memory utilization data, the energy efficiency performance mapped by the fan speed data, and the security performance mapped by the server terminal temperature data, based on the error value between the preset threshold of the AI computing server and the actual time series data. S6. If the automatic control of the AI computing power server and the AI computing power server terminal is abnormal and exceeds the preset threshold, an early warning is issued to the AI computing power server and the AI computing power server terminal, and early warning data is obtained; S7. The AI computing server monitoring model monitoring, automatic control and early warning data are displayed through a visual screen.
2. The multi-dimensional performance evaluation method for AI computing server according to claim 1 is characterized in that: The server memory utilization, fan speed and server terminal temperature are all performance indicators that need to be monitored; Collect data on server memory utilization, fan speed, and server terminal temperature through sensors, monitoring software, and data collection tools; Establish a central database to store and manage the collected data on server memory utilization, fan speed, and server terminal temperature.
3. The multi-dimensional performance evaluation method for AI computing server according to claim 2 is characterized in that: The preset thresholds include a preset performance threshold, a preset speed threshold and a preset temperature threshold; If the heterogeneous computing performance is less than the preset performance threshold, the memory of the AI computing server is not full and the heterogeneous computing performance can be extended. If the heterogeneous computing performance is greater than the preset performance threshold, the memory of the AI computing server is full and the heterogeneous computing performance cannot be extended. The cache of the AI computing server is cleared. The energy efficiency performance of the AI computing server is judged by the fan speed. If the fan speed reaches the preset speed threshold, the AI computing server is decompressed. If the fan speed does not reach the preset speed threshold, the AI computing server is not controlled. The safety of the AI computing server terminal is judged through the server terminal temperature data. If the temperature does not reach the preset temperature threshold, the AI computing server terminal operates normally. If the temperature exceeds the preset temperature threshold, the AI computing server terminal is overloaded and needs to be regulated.
4. The multi-dimensional performance evaluation method for AI computing server according to claim 3 is characterized in that: The AI computing server monitoring model is back-propagated through the error term to regulate the operation of the AI computing server, and also regulate the operation of the AI computing server terminal, including: The error term is output to the AI computing server monitoring model through the back propagation function to optimize the AI computing server and AI computing server terminal. The calculation expression is as follows: ; in, is the correction value, T is the matrix transpose, and diag[] is the function for constructing a diagonal matrix; The correction value is obtained by using the error term at the previous moment and the initial error term to optimize the AI computing power server and AI computing power server terminal at any moment.
5. The multi-dimensional performance evaluation method for AI computing server according to claim 4 is characterized in that: Early warning is provided for the AI computing server and the AI computing server terminal, abnormal performance or failure of the AI computing server is discovered in time, and an alarm message is sent to the administrator. At the same time, threshold rules are set. When the performance indicators do not meet the threshold rules, an alarm and emergency processing process is triggered; Store historical data and optimize and upgrade the AI computing server monitoring model.
6. A multi-dimensional performance evaluation server terminal for an AI computing server, used to implement a multi-dimensional performance evaluation method for an AI computing server according to any one of claims 1 to 5, characterized in that: include: A data collection module is used to collect initial data of the AI computing server, wherein the initial data includes fan speed, server terminal temperature, and server memory utilization; A data processing and analysis module, including a data processing unit for preprocessing initial data, an input layer for performing feature analysis and weight calculation on the preprocessed initial data, and an AI computing server monitoring model for performing parameter monitoring on the AI computing server and the AI computing server terminal; A multi-dimensional performance evaluation module, including a feedback unit for feeding back the error term of the AI computing power server monitoring model at time t, and a performance evaluation unit for receiving the error term and adjusting and controlling the performance parameters of the AI computing power server and the AI computing power server terminal; Visualization module, used to display the performance of AI computing server and AI computing server terminal; The alarm and monitoring module is used to warn of abnormal conditions in AI computing servers and AI computing server terminals.
7. An electronic device, characterized in that: include: A memory for storing instructions; A processor, used to execute the instruction so that the device executes the multi-dimensional performance evaluation method for an AI computing server as described in any one of claims 1-5.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, it implements the multi-dimensional performance evaluation method for AI computing power servers as described in any one of claims 1-5.
