Method and apparatus for evaluating server performance

Abstract

The application discloses a kind of server performance evaluation method and device, it is related to electric digital data processing technical field, including the actual running data of target server component in multiple environments is collected;Actual running data is input into pre-trained action model to output the first difference value between actual running data and corresponding theoretical running data;First difference value is input into pre-established reward model to output the performance score of component, and the performance evaluation result of target server is determined, solve the poor dynamic adaptability of static threshold method, higher misjudgment rate;While benchmark test tool resource consumption is large, scene generalization ability is weak, real-time is poor, test cycle is long, cannot support online dynamic evaluation technical problem, reaches through multidimensional integration, simplifies test procedure, reduces artificial intervention, more extensive applicable scene, higher accuracy, in addition, shorten test time by simplifying model expression, improve the technical effect of test efficiency.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing technology, and in particular to a method and apparatus for evaluating server performance. Background Technology

[0002] In related technologies, server performance evaluation mainly relies on two types of methods: static thresholding and benchmark testing tools. Static thresholding uses preset fixed thresholds (such as triggering an alarm when processor utilization > 80%), combined with hardware metrics collected by Linux performance tools, to achieve performance evaluation. Benchmark testing tools, on the other hand, use testing tools such as JMeter and LoadRunner to simulate stress scenarios and generate performance reports using predefined test cases.

[0003] However, among related technologies, the static threshold method has poor dynamic adaptability, cannot detect load fluctuations, and has a single indicator, resulting in a high misjudgment rate. In addition, it relies on manual labor, which makes it inefficient and highly subjective. On the other hand, benchmark testing tools consume a lot of resources, have weak scenario generalization ability, poor real-time performance, and long testing cycles. They cannot support online dynamic evaluation, which affects business operations and makes it difficult to meet the real-time evaluation needs of large-scale server clusters. Therefore, improvements are urgently needed. Summary of the Invention

[0004] This invention provides a method and apparatus for evaluating server performance, which at least solves the problems in related technologies, such as poor dynamic adaptability, single indicators, and high misjudgment rate of static threshold method; and high resource consumption, weak scenario generalization ability, poor real-time performance, long test cycle, and inability to support online dynamic evaluation of benchmark testing tools.

[0005] This invention provides a method for evaluating server performance, comprising the following steps: collecting actual operating data of at least one component of a target server in multiple environments; inputting the actual operating data into a pre-trained action model to output a first difference value between the actual operating data and the corresponding theoretical operating data; inputting the first difference value into a pre-established reward model to output a performance score of the at least one component in the corresponding environment, and determining the performance evaluation result of the target server based on the performance score.

[0006] The present invention provides a server performance evaluation device, comprising: a first acquisition module for acquiring actual operating data of at least one component of a target server in multiple environments; a first output module for inputting the actual operating data into a pre-trained action model to output a first difference value between the actual operating data and corresponding theoretical operating data; and a first determination module for inputting the first difference value into a pre-established reward model to output a performance score of the at least one component in the corresponding environment, and determining the performance evaluation result of the target server based on the performance score.

[0007] The present invention also provides a server, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described server performance evaluation methods.

[0008] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of any of the above-described server performance evaluation methods.

[0009] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described server performance evaluation methods.

[0010] This invention allows for the input of actual operational data of at least one component of a target server in multiple environments into a pre-trained action model. This model then outputs a first difference value between the actual operational data and the corresponding theoretical operational data. This first difference value is input into a pre-established reward model, which outputs the performance score of at least one component in the corresponding environment. This determines the performance evaluation result of the target server. Therefore, it addresses the problems of static threshold methods (poor dynamic adaptability, single indicator, high misjudgment rate) and benchmark testing tools (high resource consumption, weak scenario generalization ability, poor real-time performance, long testing cycles, and inability to support online dynamic evaluation). This invention achieves unified judgment of overall server performance through multi-dimensional integration, simplifies the testing process, reduces human intervention, improves the stability and reliability of server performance testing, has a wider range of applicable scenarios, and higher accuracy. Furthermore, designing different models for evaluation calculations simplifies model expression, makes models easier to train, shortens testing time, and improves testing efficiency. Attached Figure Description

[0011] To more clearly illustrate the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart illustrating a server performance evaluation method according to an embodiment of the present invention;

[0013] Figure 2 A block diagram of a first action network provided according to an embodiment of the present invention;

[0014] Figure 3 A block diagram of a second action network provided according to an embodiment of the present invention;

[0015] Figure 4 This is an execution flowchart of an action network model provided according to an embodiment of the present invention;

[0016] Figure 5 A flowchart for training a pre-established action model according to an embodiment of the present invention;

[0017] Figure 6 A flowchart illustrating the working principle of a server performance evaluation method according to an embodiment of the present invention;

[0018] Figure 7 This is a block diagram of a server performance evaluation apparatus provided according to an embodiment of the present invention.

[0019] Figure label:

[0020] Among them, 10 is a server performance evaluation device; 100 is a first acquisition module; 200 is a first output module; and 300 is a first determination module. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0022] It should be noted that, in the description of this invention, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0023] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] The embodiments of the present invention provide a method for evaluating server performance, and the method is described in detail in conjunction with the execution flow of the server performance evaluation method.

[0025] Specifically, Figure 1 A flowchart illustrating a server performance evaluation method according to an embodiment of the present invention.

[0026] like Figure 1 As shown, the method for evaluating server performance includes the following steps:

[0027] In step S101, actual operating data of at least one component of the target server in multiple environments is collected.

[0028] It is understood that, in the embodiments of the present invention, at least one component of the target server may include, but is not limited to, a processor (Central Processing Unit, or CPU), memory, hard disk, network, computing card, etc., and the present invention does not impose specific limitations.

[0029] Additionally, it should be noted that the embodiments of the present invention can utilize relevant tools, such as Zabbix, Nagios, etc., and the present invention does not impose specific limitations on them, to collect actual operating data; actual operating data can also be collected through performance monitors or command-line tools, and the present invention does not impose specific limitations on them.

[0030] In actual implementation, embodiments of the present invention can collect real-time operational data of the target server's processor, memory, hard disk, network, computing card, etc., in multiple environments. These multiple environments can be understood as the target server operating at different temperatures.

[0031] For example, embodiments of the present invention can adjust the ambient temperature to perform performance tests on the target server under different ambient temperatures, thereby obtaining different actual operating data.

[0032] Optionally, in one embodiment of the present invention, collecting actual operating data of at least one component of the target server in multiple environments includes: acquiring at least one of the following: first data of the processor in the target server, second data of the memory, third data of the hard disk, fourth data of the network, fifth data of the computing card, and sixth data of different environments; and determining the actual operating data based on at least one of the first data, second data, third data, fourth data, fifth data, and sixth data.

[0033] It is understood that, in the embodiments of the present invention, the actual operating data may include, but is not limited to, the processor's first data, the memory's second data, the hard disk's third data, the network's fourth data, the computing card's fifth data, and the sixth data for different environments, etc., and the present invention does not impose specific limitations.

[0034] Furthermore, in this embodiment of the invention, the first data may include, but is not limited to, the number of processor cores, frequency, number of threads, and thread frequency, etc., and the invention does not impose specific limitations; the second data may include, but is not limited to, memory capacity, transmission rate, and bandwidth, etc., and the invention does not impose specific limitations; the third data may include, but is not limited to, hard disk capacity, transmission rate, and bandwidth, etc., and the invention does not impose specific limitations; the fourth data may include, but is not limited to, network transmission rate, etc., and the invention does not impose specific limitations; the fifth data may include, but is not limited to, computing power and transmission rate of the computing card, etc., and the invention does not impose specific limitations; the sixth data may include, but is not limited to, ambient temperature and ambient humidity, etc., and the invention does not impose specific limitations.

[0035] In some embodiments, the present invention can determine the actual operating data by acquiring first data from the processor in the target server, second data from the memory, third data from the hard disk, fourth data from the network, fifth data from the computing card, and sixth data from different environments.

[0036] In some embodiments, the present invention can determine the actual operating data by acquiring first data from the processor in the target server, second data from the memory, third data from the hard disk, fourth data from the network, and fifth data from the computing card.

[0037] In some embodiments, the present invention can determine the actual running data by acquiring first data from the processor, second data from the memory, and third data from the hard disk in the target server.

[0038] For example, embodiments of the present invention can determine actual operating data by obtaining the number of processor cores and frequency; memory capacity and transmission rate; hard disk capacity and transmission rate; network transmission rate; computing power and transmission rate of computing card; ambient temperature, etc. The specific settings can be made by those skilled in the art according to the actual situation, and the present invention does not impose specific limitations.

[0039] The actual operating data in this embodiment of the invention includes various types of data, such as first data of the processor, second data of the memory, third data of the hard disk, fourth data of the network, fifth data of the computing card, and sixth data of different environments. This enables multi-dimensional data correlation analysis, constructs a full-dimensional performance profile, and thus improves environmental adaptability and predictive maintenance capabilities.

[0040] Optionally, in one embodiment of the present invention, before inputting the actual running data into the pre-trained action model, the method further includes: determining at least one of the following based on the actual running data: a first structural information of a fully connected neural network, a second structural information of batch normalization, and functional information of an activation function in the action model; concatenating the fully connected neural network, batch normalization, and activation function based on at least one of the first structural information, second structural information, and functional information to obtain a first action network in the action model; merging the concatenated at least two first action networks with the first action network to obtain a first feature of the actual running data; obtaining a second action network in the action model based on the first feature and the first action network; and constructing an action model based on the first action network and the second action network.

[0041] It is understood that, in the embodiments of the present invention, the action model may include, but is not limited to, a first action network and a second action network, and can be specifically set by those skilled in the art according to the actual situation. The present invention does not impose any specific limitations.

[0042] Among them, the first action network FBR network is combined Figure 2 As shown, the FBR network may include, but is not limited to, a fully connected neural network (FNN), batch normalization (BN), and the ReLU activation function. The FBR network is obtained by concatenating the fully connected neural network, batch normalization, and the ReLU activation function.

[0043] Second Action Network FC Network Combination Figure 3 As shown, the FC network can merge the features obtained from two cascaded FBR networks and one FBR network to obtain the first feature. The first feature is then processed by the FBR network to obtain the FC network.

[0044] Furthermore, embodiments of the present invention can be combined with Figure 4As shown, the execution flow of the action network model is introduced, and its main contents are as follows: Acquire actual running data; pass the actual running data through the FBR network to obtain initial features; pass the initial features through the FBR network to obtain second features; pass the second features through the FC network to obtain third features; pass the third features through the FBR network to obtain fourth features; pass the fourth features through the FC network to obtain fifth features; pass the fifth features through the FBR network to obtain feature F1; pass the fifth features through the FBR network to obtain feature F2; pass the actual running data through the FBR network to obtain feature F3; pass feature F3 through the FBR network to obtain feature F4; pass feature F4... Multiply feature F2 bitwise to obtain feature F5; add feature F5 and feature F1 bitwise to obtain feature F6; pass feature F6 through the FBR network to obtain feature F7; merge feature F7 and feature F3 to obtain the sixth feature; pass the sixth feature through the FBR network to obtain the seventh feature; pass the seventh feature through the FBR network to obtain feature F8; pass the seventh feature through the FBR network to obtain feature F9; multiply feature F9 and feature F10 bitwise to obtain feature F10; add feature F10 and feature F9 bitwise to obtain the eighth feature; pass the input value of the eighth feature through the FBR network to obtain the output of the action network model.

[0045] In some embodiments, the present invention can determine the first structural information of the fully connected neural network, the second structural information of the batch normalization, and the functional information of the activation function in the action model based on actual running data. Then, the fully connected neural network, the batch normalization, and the activation function are concatenated to obtain a first action network. At least two concatenated first action networks are merged with the first action network to obtain the first feature of the actual running data, thereby obtaining a second action network, and then constructing an action model.

[0046] For example, embodiments of the present invention can be combined with Figure 2 As shown, the first action network is constructed, combined with... Figure 3 As shown, a second action network is constructed, combining... Figure 4 As shown, construct the action model.

[0047] This invention constructs action models through network modularization, enabling independent optimization or replacement, thereby reducing maintenance costs. It enhances feature fusion through concatenation, meeting low-latency inference requirements and ensuring real-time performance. Furthermore, it adjusts the model structure based on actual operational data, achieving dynamic structural adjustment and improving decision-making quality.

[0048] Optionally, in one embodiment of the present invention, before inputting the actual running data into the pre-trained action model, the method further includes: collecting training running data of at least one component of the server in multiple environments; inputting the training running data into the pre-established action model to output a second difference value between the training running data and the corresponding theoretical running data; calculating the training performance score of at least one component in the corresponding environment based on the second difference value using a pre-established reward model; detecting whether at least one component meets preset performance conditions based on the training performance score; and if at least one component does not meet the preset performance conditions, using discrete estimation to train the parameter information of the pre-established action model to obtain the trained action model, and generating a component that meets the preset performance conditions based on the trained action model.

[0049] In some embodiments, before inputting actual runtime data into the pre-trained action model, the present invention can train the pre-established action model to obtain the trained action model. The present invention can be combined with... Figure 5 As shown, the content of training a pre-established action model is introduced, and its main contents can be summarized as follows:

[0050] Step S501: Collect training run data of at least one component of the server in multiple environments.

[0051] In this embodiment of the invention, the training running data is: ,in, The number of processor cores; The processor frequency; This refers to the capacity of the memory. This refers to the memory transfer rate. This refers to the hard drive's capacity. This refers to the hard drive's transfer rate. This refers to the transmission rate of the network port. This refers to the computing power value of the computing power card; This is the current temperature value.

[0052] Step S502: Input the training data into the pre-established action model to output the second difference value between the training data and the corresponding theoretical data.

[0053] In this embodiment of the invention, the pre-established action model can be utilized. Calculate the difference between the processor's training data and the corresponding theoretical running data; utilize Calculate the difference between the in-memory training data and the corresponding theoretical running data; utilize Calculate the difference between the hard disk training data and the corresponding theoretical running data; utilize Calculate the difference between the network training data and the corresponding theoretical running data; utilize Calculate the difference between the training data and the corresponding theoretical operating data of the computing power card.

[0054] Furthermore, in embodiments of the present invention, The main tests are the processor's maximum number of threads and frequency (such as turbo boost, which are not specifically limited in this invention). The main tests are on memory capacity and transfer rate (such as between memory and hard drive, between memory and accelerator card, etc., this invention does not impose specific limitations). The main tests focus on the capacity and transfer rate of the hardware (such as between hardware and hard drive, hard drive and memory, etc., which are not specifically limited in this invention). The main test is the transmission rate of the network of the testing equipment; The main tests are the computing power and data transfer rate of the computing card (such as between memory and accelerator card, between accelerator cards, etc., which are not specifically limited in this invention).

[0055] Step S503: Calculate the training performance score using the pre-established reward model.

[0056] In this embodiment of the invention, the training performance scores of the processor, memory, hard disk, network, computing card and energy consumption in the corresponding environment can be calculated based on the second difference value using a pre-established reward model.

[0057] Step S504: Determine whether the server's performance meets certain performance conditions.

[0058] In this embodiment of the invention, if certain performance conditions are met, step S506 is executed; otherwise, step S505 is executed. These performance conditions can be set by those skilled in the art according to actual circumstances, and the present invention does not impose specific limitations.

[0059] Step S505: Retrain the pre-established action model using discrete estimation.

[0060] In this embodiment of the invention, the parameter information of a pre-established action model can be obtained by discrete estimation training, thereby obtaining the trained action model, and generating components that meet certain performance conditions based on the trained action model.

[0061] Step S506: Determine the server's performance evaluation results.

[0062] The embodiments of the present invention can improve scene coverage by collecting training data under different environments, quantify the accuracy of model prediction by comparing actual and theoretical data, facilitate evaluation and optimization, set certain performance conditions to ensure that the performance of components reaches the expected standard, and trigger the optimization process when the standard is not met, thereby improving the model effect.

[0063] Optionally, in one embodiment of the present invention, the parameter information of the pre-established action model is trained using discrete estimation, including: gradient information based on discrete estimation, the probability distribution of action selection in the training parameter information; learning rate information based on discrete estimation, the parameter update magnitude in the training parameter information; logarithmic information based on discrete estimation, the parameter adjustment sensitivity in the training parameter information; and the action value in the training parameter information based on the advantage function based on discrete estimation.

[0064] In some embodiments, the present invention can use unbiased discrete estimation to feed back the values ​​of a pre-established reward model to a pre-trained action model, thereby retraining the parameter information of the pre-established action model. The expression for discrete estimation can be, but is not limited to, the following:

[0065] ,

[0066] in, The parameters are discretely estimated, which determine the probability distribution of action selection. The policy is updated through gradient ascent, so that it tends to favor actions with high advantage values. The learning rate (step size) is used for discrete estimation, which controls the magnitude of parameter updates. Too high a learning rate will cause policy oscillations, while too low a learning rate will result in slow convergence. The gradient of the logarithmic probability of the policy function reflects the sensitivity of the current action to parameter adjustments. The advantage signal is backpropagated to the policy network through the chain rule. The dominance function measures the action. In state The relative value under the given conditions can be expressed, but is not limited to, as:

[0067] ,

[0068] ,

[0069] ,

[0070] in, Let be the action value function, representing the state. Next action The expected cumulative gain For state Value function, in state The average expectation of the strategy followed is The time-series difference error is the instantaneous difference between the estimated and actual values ​​of the function. This is the current reward value. For hyperparameters, This indicates the current time.

[0071] The embodiments of the present invention can more accurately control the action selection process by discretely estimating the training probability distribution, improve the balance between exploration and utilization, provide more stable learning rate information, improve the stability of the model during the training process, help the model adjust parameters more finely, and thus help the model make better decisions.

[0072] In step S102, the actual running data is input into the pre-trained action model to output the first difference value between the actual running data and the corresponding theoretical running data.

[0073] In some embodiments of the present invention, actual running data can be input into a pre-trained action model to obtain a first difference value between the actual running data and the corresponding theoretical running data.

[0074] For example, embodiments of the present invention can calculate a first difference value between actual running data and corresponding theoretical running data using a pre-trained action model.

[0075] Optionally, in one embodiment of the present invention, inputting actual operating data into a pre-trained action model to output a first difference value between the actual operating data and the corresponding theoretical operating data includes: outputting a processor difference value between the actual operating data and the theoretical operating data of the processor based on at least one of the processor's thread information and frequency information in the actual operating data using the pre-trained action model; outputting a memory difference value between the actual operating data and the theoretical operating data of the memory based on at least one of the memory's capacity information and transfer rate information in the actual operating data using the pre-trained action model; and outputting a memory difference value between the actual operating data and the theoretical operating data of the memory based on the hard disk's capacity information and transfer rate information in the actual operating data. Based on at least one of the rate information, the pre-trained action model outputs the hard disk difference value between the actual operating data and the theoretical operating data of the hard disk; based on the network transmission rate information in the actual operating data, the pre-trained action model outputs the network difference value between the actual operating data and the theoretical operating data of the network; based on at least one of the computing power data and transmission rate information of the computing power card in the actual operating data, the pre-trained action model outputs the computing power card difference value between the actual operating data and the theoretical operating data of the computing power card; based on at least one of the processor difference value, memory difference value, hard disk difference value, network difference value, and computing power card difference value, a first difference value is obtained.

[0076] It is understood that the pre-trained action model in the embodiments of the present invention has added action perturbations, such as processor thread step jumps, memory transfer surges, and hard disk read / write speeds. The present invention does not impose specific limitations.

[0077] Furthermore, in some embodiments, the present invention can output the processor difference value between the actual running data and the theoretical running data based on the processor's thread information and frequency information in the actual running data, using processor thread step action perturbation in the pre-trained action model; it can also be based on other action perturbations, and the present invention does not impose specific limitations.

[0078] In some embodiments, the present invention can output the memory difference value between the actual running data and the theoretical running data by using memory transfer surge motion perturbation in the pre-trained action model based on the memory capacity information and transmission rate information in the actual running data; it can also be based on other motion perturbations, and the present invention does not impose specific limitations.

[0079] In some embodiments, the present invention can output the hard disk difference value between the actual operating data and the theoretical operating data by using hard disk read / write rate action perturbations in a pre-trained action model based on the hard disk capacity information and transmission rate information in the actual operating data; it can also be based on other action perturbations, and the present invention does not impose specific limitations.

[0080] In some embodiments, the present invention can output the network difference value between the actual network operation data and the theoretical network operation data based on the network transmission rate information in the actual operation data and using a pre-trained action model.

[0081] In some embodiments, the present invention can output the difference between the actual operating data and the theoretical operating data of the computing card based on the computing power data and transmission rate information of the computing card in the actual operating data, using a pre-trained action model.

[0082] Furthermore, in embodiments of the present invention, a first difference value can be obtained based on processor difference values, memory difference values, hard disk difference values, network difference values, and computing card difference values, etc.

[0083] This invention can achieve refined performance diagnosis capabilities, improve the accuracy of resource optimization, enhance predictive maintenance, and improve model stability by analyzing processor differences, memory differences, hard disk differences, network differences, and computing card differences.

[0084] Optionally, in one embodiment of the present invention, before inputting the first difference value into the pre-established reward model, the method further includes: establishing a processor reward model in the reward model based on the processor difference value in the first difference value, using task information, thread information, and instruction information in the processor difference value; establishing a memory reward model in the reward model based on the memory difference value in the first difference value, using data information and bandwidth information in the memory difference value; establishing a hard disk reward model in the reward model based on the hard disk difference value in the first difference value, using bandwidth information in the hard disk difference value; establishing a network reward model in the reward model based on the network difference value in the first difference value, using transmission data information in the network difference value; establishing a computing power card reward model in the reward model based on the computing power card difference value in the first difference value, using computing power information in the computing power card difference value; establishing an energy consumption reward model in the reward model based on the energy consumption information in the first difference value; and establishing a reward model based on at least one of the processor reward model, memory reward model, hard disk reward model, network reward model, computing power card reward model, and energy consumption reward model. The expression for the processor reward model may be, but is not limited to, as follows:

[0085] ,

[0086] in, This represents the total number of instructions for the processor's tasks. The theoretical number of threads for the processor ， This represents the processor's theoretical thread frequency. This represents the actual number of instructions that the chip can process per cycle. This refers to the actual completion time of the processor; The number of threads required by the processor;

[0087] The expression for the memory reward model can be, but is not limited to, as follows:

[0088] ,

[0089] in, This represents the total amount of data in memory. This represents the theoretical bandwidth of the memory. This refers to the actual bandwidth of the memory. This represents the theoretical capacity of the memory.

[0090] The expression for the hard disk reward model can be, but is not limited to, as follows:

[0091] ,

[0092] in, This refers to the total amount of data on the hard drive. This represents the theoretical bandwidth of the hard drive. This refers to the actual bandwidth of the hard drive. This represents the theoretical capacity of the hard drive.

[0093] The expression for the network reward model can be, but is not limited to, as follows:

[0094] ,

[0095] in, This refers to the size of the data before transmission. This refers to the actual size of the data received. This refers to the actual bandwidth of the network. This represents the theoretical capacity of the network.

[0096] The expression for the computing power card reward model can be, but is not limited to, as follows:

[0097] ,

[0098] in, This refers to the computational cost of the model. This represents the actual computing power value. This refers to the actual bandwidth of the computing card; This represents the theoretical capacity of the computing card.

[0099] The expression for the energy consumption reward model can be, but is not limited to, as follows:

[0100] ,

[0101] in, Theoretical energy consumption; This represents the actual energy consumption value.

[0102] In some embodiments, the reward models that can be established in the present invention may include, but are not limited to, processor reward models, memory reward models, hard disk reward models, network reward models, computing card reward models, and energy consumption reward models, etc., and the present invention does not impose specific limitations.

[0103] In this embodiment of the invention, task information, thread information, and instruction information from the processor difference value can be used to establish a processor reward model, the expression of which may be, but is not limited to:

[0104] ,

[0105] in, This represents the total number of instructions for the processor's tasks. This represents the theoretical number of threads for the processor. This represents the processor's theoretical thread frequency. This represents the actual number of instructions that the chip can process per cycle. This refers to the actual completion time of the processor; This represents the number of threads required by the processor.

[0106] In some embodiments of the present invention, a memory reward model can be established using data information and bandwidth information in the memory difference value, and its expression can be, but is not limited to, as follows:

[0107] ,

[0108] in, This represents the total amount of data in memory. This represents the theoretical bandwidth of the memory. This refers to the actual bandwidth of the memory. This represents the theoretical capacity of the memory.

[0109] In some embodiments of the present invention, the bandwidth information in the hard disk difference value can be used to establish a hard disk reward model, the expression of which may be, but is not limited to:

[0110] ,

[0111] in, This refers to the total amount of data on the hard drive. This represents the theoretical bandwidth of the hard drive. This represents the actual bandwidth of the hard drive, from the start to the end of data transfer, and is a data group. This represents the theoretical capacity of the hard drive.

[0112] In some embodiments of the present invention, a network reward model can be established using transmission data information in the network difference value, and its expression can be, but is not limited to, as follows:

[0113] ,

[0114] in, This refers to the size of the data before transmission. This represents the actual size of the received data, primarily to prevent packet loss. This refers to the actual bandwidth of the network. This represents the theoretical capacity of the network.

[0115] In some embodiments of the present invention, a computing power card reward model can be established using computing power information in the computing power card difference value, and its expression can be, but is not limited to, as follows:

[0116] ,

[0117] in, This refers to the computational cost of the model. This represents the actual computing power value. This refers to the actual bandwidth of the computing card; This represents the theoretical capacity of the computing card.

[0118] In some embodiments, the present invention can establish an energy consumption reward model based on the energy consumption information in the first difference value, and its expression can be, but is not limited to, as follows:

[0119] ,

[0120] in, Theoretical energy consumption; This represents the actual energy consumption value.

[0121] This invention can optimize processor task scheduling and thread management through a processor reward model to improve processing efficiency, optimize memory usage and allocation through a memory reward model to reduce memory bottlenecks, optimize hard disk read and write operations through a hard disk reward model to improve data access speed, optimize network resource allocation through a network reward model to reduce network latency, optimize computing card usage through a computing card reward model to improve computing efficiency, and reduce energy consumption while ensuring performance through an energy consumption reward model to achieve green computing. By combining the advantages of the reward models of each component, the overall performance, resource management and energy efficiency of the server can be comprehensively optimized.

[0122] In step S103, the first difference value is input into the pre-established reward model to output the performance score of at least one component in the corresponding environment, and the performance evaluation result of the target server is determined based on the performance score.

[0123] In some embodiments, the present invention can input a first difference value into a pre-established reward model, thereby determining the performance scores of different components in the corresponding environment, and thus determining the performance evaluation result of the server.

[0124] In this embodiment of the invention, a processor performance score can be obtained. Memory performance score Hard drive performance score Computing power card performance score Network performance score and energy efficiency score This invention does not impose specific limitations.

[0125] Furthermore, embodiments of the present invention can determine performance evaluation results based on processor performance scores, memory performance scores, hard disk performance scores, network performance scores, and computing card performance scores. The calculation formula can be, but is not limited to, the following:

[0126] ,

[0127] in, , , , , , For hyperparameters; This refers to the server's performance evaluation results; The processor performance score; This represents the memory performance score. This represents the hard drive performance score. Network performance score; This refers to the performance score of the computing card. This represents the energy efficiency performance score.

[0128] The embodiments of the present invention can comprehensively consider processor performance score, memory performance score, hard disk performance score, network performance score, computing card performance score and energy consumption performance score, resulting in more accurate evaluation results and a wider range of applicable scenarios.

[0129] Optionally, in one embodiment of the present invention, the first difference value is input into a pre-established reward model to output the performance score of at least one component in the corresponding environment, including: calculating the processor performance score of the processor in the server using the processor reward model in the pre-established reward model based on the processor difference value in the first difference value; calculating the memory performance score of the memory in the server using the memory reward model in the pre-established reward model based on the memory difference value in the first difference value; calculating the hard disk performance score of the hard disk in the server using the hard disk reward model in the pre-established reward model based on the hard disk difference value in the first difference value; calculating the network performance score of the network in the server using the network reward model in the pre-established reward model based on the network difference value in the first difference value; calculating the computing power card performance score of the computing power card in the server using the computing power card reward model in the pre-established reward model based on the computing power card difference value in the first difference value; and calculating the energy consumption performance score of the server using the energy consumption reward model in the pre-established reward model based on the energy consumption difference value in the first difference value.

[0130] In some embodiments, the present invention can calculate a processor performance score based on the processor difference value in the first difference value using a processor reward model in a pre-established reward model.

[0131] In some embodiments, the present invention can calculate a memory performance score based on the memory difference value in the first difference value using a memory reward model in a pre-established reward model.

[0132] In some embodiments, the present invention can calculate a hard disk performance score based on the hard disk difference value in the first difference value using a hard disk reward model in a pre-established reward model.

[0133] In some embodiments, the present invention can calculate a network performance score based on the network difference value in the first difference value using a network reward model in a pre-established reward model.

[0134] In some embodiments, the present invention can calculate the computing card performance score based on the computing card difference value in the first difference value using the computing card reward model in the pre-established reward model.

[0135] In some embodiments, the present invention can calculate an energy performance score based on the energy consumption difference value in the first difference value using an energy consumption reward model in a pre-established reward model.

[0136] The embodiments of the present invention can evaluate each component individually, more accurately locate performance bottlenecks, facilitate maintenance and expansion, enhance the interpretability of results, comprehensively consider multiple components and energy consumption, help find a balance between performance and energy efficiency, adapt to different loads and environmental changes, and make smarter decisions.

[0137] Optionally, in one embodiment of the present invention, determining the performance evaluation result of the target server based on the performance score includes: counting the total number of performance scores based on the performance scores; calculating the initial performance evaluation result of the target server based on the performance scores; and determining the performance evaluation result based on the total number and the initial performance evaluation result.

[0138] It is understood that, in order to ensure that the performance of the processor, memory, hardware and other components meets the requirements, the embodiments of the present invention may perform multiple evaluations, such as two or three times. The present invention does not impose specific limitations, thereby obtaining different performance scores.

[0139] As one possible implementation, embodiments of the present invention can count the total number of performance scores, then calculate the initial performance evaluation result of the target server, and determine the performance evaluation result based on the total number and the initial performance evaluation result.

[0140] For example, in embodiments of the present invention, the initial performance evaluation result can be calculated using the calculation formula of the performance evaluation result, and the horizontal distance value of the initial performance evaluation result can be taken as the performance evaluation result.

[0141] This invention, by statistically calculating the total performance score, comprehensively considers the performance of multiple components, avoiding the one-sidedness of evaluating a single component, and more accurately reflects the overall performance of the server. This makes the evaluation results more in line with the needs of actual application scenarios, and can quickly locate performance bottlenecks, thereby formulating targeted optimization strategies, reducing operation and maintenance costs, and improving resource utilization.

[0142] The working principle of the server performance evaluation method proposed in this invention will be introduced below with reference to a specific embodiment.

[0143] in, Figure 6 A flowchart illustrating the working principle of a server performance evaluation method according to an embodiment of the present invention.

[0144] Step S601: Collect actual operating data of the target server in different environments.

[0145] In this embodiment of the invention, different environments can be different ambient temperatures.

[0146] Step S602: Input the actual running data into the pre-trained action model to obtain the first difference value between the actual running data and the corresponding theoretical running data.

[0147] In this embodiment of the invention, a pre-trained action model can be used to calculate the first difference value between actual operating data and theoretical operating data of processor, memory, hardware, network, computing card, energy consumption, etc.

[0148] Step S603: Input the first difference value into the pre-established reward model to obtain the performance score under the corresponding environment.

[0149] In this embodiment of the invention, a pre-established model can be used to calculate processor performance score, memory performance score, hard disk performance score, computing card performance score, network performance score, and energy consumption performance score, etc. This invention does not impose specific limitations.

[0150] Step S604: Determine the initial performance evaluation results of the server.

[0151] In this embodiment of the invention, processor performance score, memory performance score, hard disk performance score, computing card performance score, network performance score, and energy consumption performance score can be substituted into the calculation formula of the performance evaluation result to calculate the initial performance evaluation result.

[0152] Step S605: Determine the server's performance evaluation results.

[0153] In this embodiment of the invention, the average value can be calculated based on the total number of initial performance evaluation results, thereby determining the performance evaluation result.

[0154] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0155] The server performance evaluation method proposed in this embodiment of the invention can input the actual operating data of at least one component of the target server in multiple environments into a pre-trained action model, and then output a first difference value between the actual operating data and the corresponding theoretical operating data. The first difference value is then input into a pre-established reward model, and then outputs the performance score of at least one component in the corresponding environment, thereby determining the performance evaluation result of the target server. Therefore, it can solve the problems of poor dynamic adaptability, single index, and high misjudgment rate of static threshold method; and the problems of high resource consumption, weak scenario generalization ability, poor real-time performance, long test cycle, and inability to support online dynamic evaluation technology of benchmark testing tool. It achieves the technical effect of unifying the judgment of the overall server performance through multi-dimensional integration, simplifying the test process, reducing human intervention, improving the stability and reliability of server performance testing, and having a wider range of applicable scenarios and higher accuracy. In addition, designing different models for evaluation calculation simplifies model expression, makes the model easier to train, shortens the test time, and improves the test efficiency.

[0156] Embodiments of the present invention also provide a server performance evaluation device.

[0157] Figure 7 This is a block diagram of a server performance evaluation apparatus provided according to an embodiment of the present invention.

[0158] like Figure 7 As shown, the server performance evaluation device 10 includes: a first acquisition module 100, a first output module 200, and a first determination module 300.

[0159] The first acquisition module 100 is used to acquire actual operating data of at least one component of the target server in multiple environments.

[0160] The first output module 200 is used to input actual running data into a pre-trained action model to output a first difference value between the actual running data and the corresponding theoretical running data.

[0161] The first determining module 300 is used to input the first difference value into a pre-established reward model to output the performance score of at least one component in the corresponding environment, and to determine the performance evaluation result of the target server based on the performance score.

[0162] Optionally, in one embodiment of the present invention, it further includes: a second determining module, a first generating module, a second generating module, a third generating module, and a first constructing module.

[0163] The second determining module is used to determine, based on the actual running data, at least one of the first structural information of the fully connected neural network, the second structural information of the batch normalization, and the functional information of the activation function in the action model before inputting the actual running data into the pre-trained action model.

[0164] The first generation module is used to concatenate a fully connected neural network, batch normalization, and activation function based on at least one of the first structural information, the second structural information, and function information to obtain the first action network in the action model.

[0165] The second generation module is used to merge at least two first action networks connected in series to obtain the first feature of the actual running data.

[0166] The third generation module is used to obtain the second action network in the action model based on the first feature and the first action network.

[0167] The first building module is used to construct an action model based on the first action network and the second action network.

[0168] Optionally, in one embodiment of the present invention, it further includes: a second acquisition module, a second output module, a calculation module, a detection module, and a fourth generation module.

[0169] The second acquisition module is used to acquire training and running data of at least one component of the server in multiple environments before inputting the actual running data into the pre-trained action model.

[0170] The second output module is used to input the training data into the pre-established action model to output the second difference value between the training data and the corresponding theoretical data.

[0171] The calculation module is used to calculate the training performance score of at least one component in the corresponding environment based on a second difference value and using a pre-established reward model.

[0172] The detection module is used to detect whether at least one component meets preset performance conditions based on the training performance score.

[0173] The fourth generation module is used to obtain a trained action model by using discrete estimation to train the parameter information of a pre-established action model when at least one component does not meet the preset performance conditions, and to generate a component that meets the preset performance conditions based on the trained action model.

[0174] Optionally, in one embodiment of the present invention, the fourth generation module includes: a first training unit, a second training unit, a third training unit, and a fourth training unit.

[0175] The first training unit is used to train the probability distribution of action selection in the parameter information based on the gradient information of discrete estimation.

[0176] The second training unit is used to update the parameter magnitude based on the learning rate information and training parameter information, which is based on discrete estimation.

[0177] The third training unit is used to train the sensitivity of parameter adjustments in the parameter information based on the logarithmic information of discrete estimation.

[0178] The fourth training unit is used to train the action value in the parameter information based on the advantage function of discrete estimation.

[0179] Optionally, in one embodiment of the present invention, the first acquisition module 100 includes: an acquisition unit and a first determination unit.

[0180] The acquisition unit is used to acquire at least one of the following: first data from the processor in the target server, second data from the memory, third data from the hard disk, fourth data from the network, fifth data from the computing card, and sixth data from different environments.

[0181] The first determining unit is used to determine the actual operating data based on at least one of the first data, second data, third data, fourth data, fifth data, and sixth data.

[0182] Optionally, in one embodiment of the present invention, the first output module 200 includes: a first output unit, a second output unit, a third output unit, a fourth output unit, a fifth output unit, and a generation unit.

[0183] The first output unit is used to output the processor difference value between the processor's actual operating data and the processor's theoretical operating data based on at least one of the processor's thread information and frequency information in the actual operating data, using a pre-trained action model.

[0184] The second output unit is used to output the memory difference value between the actual memory operation data and the theoretical memory operation data based on at least one of the memory capacity information and transmission rate information in the actual operation data, using a pre-trained action model.

[0185] The third output unit is used to output the hard disk difference value between the actual operating data and the theoretical operating data of the hard disk based on at least one of the capacity information and transfer rate information of the hard disk in the actual operating data, using a pre-trained action model.

[0186] The fourth output unit is used to output the network difference value between the actual network operation data and the theoretical network operation data based on the network transmission rate information in the actual operation data and using a pre-trained action model.

[0187] The fifth output unit is used to output the difference value between the actual operating data and the theoretical operating data of the computing card based on at least one of the computing power data and transmission rate information of the computing card in the actual operating data, using a pre-trained action model.

[0188] The generation unit is used to obtain a first difference value based on at least one of the processor difference value, memory difference value, hard disk difference value, network difference value, and computing card difference value.

[0189] Optionally, in one embodiment of the present invention, it further includes: a second building module, a third building module, a fourth building module, a fifth building module, a sixth building module, a seventh building module, and an eighth building module.

[0190] The second construction module is used to build a processor reward model in the reward model based on the processor difference value in the first difference value and using the task information, thread information and instruction information in the processor difference value before inputting the first difference value into the pre-established reward model.

[0191] The third building module is used to build a memory reward model in the reward model based on the memory difference value in the first difference value and the data information and bandwidth information in the memory difference value.

[0192] The fourth construction module is used to build a hard disk reward model in the reward model based on the hard disk difference value in the first difference value and the bandwidth information in the hard disk difference value.

[0193] The fifth construction module is used to build a network reward model in the reward model based on the network difference value in the first difference value and the transmission data information in the network difference value.

[0194] The sixth construction module is used to establish a computing power card reward model in the reward model based on the computing power card difference value in the first difference value and the computing power information in the computing power card difference value.

[0195] The seventh module is used to build an energy consumption reward model in the reward model based on the energy consumption information in the first difference value.

[0196] The eighth building module is used to establish a reward model based on at least one of the following: processor reward model, memory reward model, hard disk reward model, network reward model, computing card reward model, and energy consumption reward model.

[0197] Optionally, in one embodiment of the present invention, the first determining module 300 includes: a first calculation unit, a second calculation unit, a third calculation unit, a fourth calculation unit, a fifth calculation unit, and a sixth calculation unit.

[0198] The first calculation unit is used to calculate the processor performance score of the processor in the server based on the processor difference value in the first difference value and the processor reward model in the pre-established reward model.

[0199] The second calculation unit is used to calculate the memory performance score of the server's memory based on the memory difference value in the first difference value and using the memory reward model in the pre-established reward model.

[0200] The third calculation unit is used to calculate the hard drive performance score of the hard drives in the server based on the hard drive difference value in the first difference value and using the hard drive reward model in the pre-established reward model.

[0201] The fourth calculation unit is used to calculate the network performance score of the server's network based on the network difference value in the first difference value and the network reward model in the pre-established reward model.

[0202] The fifth calculation unit is used to calculate the computing card performance score of the computing card in the server based on the computing card difference value in the first difference value and using the computing card reward model in the pre-established reward model.

[0203] The sixth calculation unit is used to calculate the server's energy performance score based on the energy consumption difference value in the first difference value and using the energy consumption reward model in the pre-established reward model.

[0204] Optionally, in one embodiment of the present invention, wherein,

[0205] The expression for the processor reward model can be, but is not limited to, as follows:

[0206] ,

[0207] in, This represents the total number of instructions for the processor's tasks. The theoretical number of threads for the processor ， This represents the processor's theoretical thread frequency. This represents the actual number of instructions that the chip can process per cycle. This refers to the actual completion time of the processor; The number of threads required by the processor;

[0208] The expression for the memory reward model can be, but is not limited to, as follows:

[0209] ,

[0210] in, This represents the total amount of data in memory. This represents the theoretical bandwidth of the memory. This refers to the actual bandwidth of the memory. This represents the theoretical capacity of the memory.

[0211] The expression for the hard disk reward model can be, but is not limited to, as follows:

[0212] ,

[0213] in, This refers to the total amount of data on the hard drive. This represents the theoretical bandwidth of the hard drive. This refers to the actual bandwidth of the hard drive. This represents the theoretical capacity of the hard drive.

[0214] The expression for the network reward model can be, but is not limited to, as follows:

[0215] ,

[0216] in, This refers to the size of the data before transmission. This refers to the actual size of the data received. This refers to the actual bandwidth of the network. This represents the theoretical capacity of the network.

[0217] The expression for the computing power card reward model can be, but is not limited to, as follows:

[0218] ,

[0219] in, This refers to the computational cost of the model. This represents the actual computing power value. This refers to the actual bandwidth of the computing card; This represents the theoretical capacity of the computing card.

[0220] The expression for the energy consumption reward model can be, but is not limited to, as follows:

[0221] ,

[0222] in, Theoretical energy consumption; This represents the actual energy consumption value.

[0223] Optionally, in one embodiment of the present invention, the first determining module 300 includes: a statistical unit, a seventh calculation unit, and a second determining unit.

[0224] The statistics unit is used to count the total number of performance scores based on the performance score.

[0225] The seventh calculation unit is used to calculate the initial performance evaluation results of the target server based on the performance score.

[0226] The second determining unit is used to determine the performance evaluation results based on the total number and the initial performance evaluation results.

[0227] For a description of the features in the embodiment corresponding to the server performance evaluation device, please refer to the relevant description in the embodiment corresponding to the server performance evaluation method, which will not be repeated here.

[0228] The server performance evaluation device proposed in this embodiment of the invention can input the actual operating data of at least one component of the target server in multiple environments into a pre-trained action model, and then output a first difference value between the actual operating data and the corresponding theoretical operating data. The first difference value is then input into a pre-established reward model, and then outputs the performance score of at least one component in the corresponding environment, thereby determining the performance evaluation result of the target server. Therefore, it can solve the problems of poor dynamic adaptability, single index, and high misjudgment rate of static threshold method; and the problems of high resource consumption, weak scenario generalization ability, poor real-time performance, long test cycle, and inability to support online dynamic evaluation technology of benchmark testing tool. It achieves the technical effect of unifying the judgment of the overall server performance through multi-dimensional integration, simplifying the test process, reducing human intervention, improving the stability and reliability of server performance testing, and having a wider range of applicable scenarios and higher accuracy. In addition, designing different models for evaluation calculation simplifies model expression, makes the model easier to train, shortens the test time, and improves the test efficiency.

[0229] Embodiments of the present invention also provide a server, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above-described server performance evaluation method embodiments.

[0230] Embodiments of the present invention also provide a computer-readable storage medium storing a computer program configured to execute the steps in any of the above-described server performance evaluation method embodiments when run.

[0231] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0232] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above-described server performance evaluation method embodiments.

[0233] Embodiments of the present invention also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the above-described server performance evaluation method embodiments.

[0234] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0235] The above provides a detailed description of a server performance evaluation method provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of this invention. It should be noted that those skilled in the art can make various improvements and modifications to this invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this invention.

Claims

1. A method for evaluating server performance, characterized in that, Includes the following steps: Collect actual operating data of at least one component of the target server in multiple environments; The actual running data is input into a pre-trained action model to output a first difference value between the actual running data and the corresponding theoretical running data; The first difference value is input into a pre-established reward model to output the performance score of the at least one component in the corresponding environment, and the performance evaluation result of the target server is determined based on the performance score. Before inputting the actual operational data into the pre-trained action model, the process also includes: Based on the actual operating data, at least one of the following is determined in the action model: the first structural information of the fully connected neural network, the second structural information of the batch normalization, and the functional information of the activation function. Based on at least one of the first structural information, the second structural information, and the function information, the fully connected neural network, the batch normalization, and the activation function are concatenated to obtain the first action network in the action model; The first action network is merged into at least two concatenated first action networks to obtain the first feature of the actual operation data; Based on the first feature and the first action network, the second action network in the action model is obtained; The action model is constructed based on the first action network and the second action network.

2. The server performance evaluation method according to claim 1, characterized in that, Before inputting the actual operational data into the pre-trained action model, the following steps are also included: Collect training and running data from at least one component of the server in multiple environments; The training data is input into a pre-established action model to output a second difference value between the training data and the corresponding theoretical data. Based on the second difference value, the training performance score of the at least one component in the corresponding environment is calculated using a pre-established reward model; Based on the training performance score, detect whether the at least one component meets the preset performance conditions; If at least one component does not meet the preset performance conditions, the parameter information of the pre-established motion model is trained by discrete estimation to obtain a trained motion model, and a component that meets the preset performance conditions is generated based on the trained motion model.

3. The server performance evaluation method according to claim 2, characterized in that, The parameter information used to train the pre-established action model using discrete estimation includes: Based on the gradient information of the discrete estimation, train the probability distribution of action selection in the parameter information; Based on the learning rate information of the discrete estimation, the parameter update magnitude in the parameter information is trained; Based on the logarithmic information of the discrete estimate, the sensitivity of parameter adjustment in the parameter information is trained; Based on the advantage function of the discrete estimation, the action value in the parameter information is trained.

4. The server performance evaluation method according to claim 1, characterized in that, The data collected includes actual operational data of at least one component of the target server in multiple environments, including: Acquire at least one of the following in the target server: first data from the processor, second data from the memory, third data from the hard disk, fourth data from the network, fifth data from the computing card, and sixth data from different environments; The actual operating data is determined based on at least one of the first data, the second data, the third data, the fourth data, the fifth data, and the sixth data.

5. The server performance evaluation method according to claim 1, characterized in that, The step of inputting the actual running data into a pre-trained action model to output a first difference value between the actual running data and the corresponding theoretical running data includes: Based on at least one of the processor's thread information and frequency information in the actual operating data, the processor difference value between the processor's actual operating data and the processor's theoretical operating data is output using the pre-trained action model. Based on at least one of the memory capacity information and transmission rate information in the actual operating data, the memory difference value between the actual operating data of the memory and the theoretical operating data of the memory is output using the pre-trained action model; Based on at least one of the hard drive capacity information and transfer rate information in the actual operating data, the hard drive difference value between the actual operating data and the theoretical operating data of the hard drive is output using the pre-trained action model. Based on the network transmission rate information in the actual operating data, the network difference value between the actual operating data and the theoretical operating data of the network is output using the pre-trained action model. Based on at least one of the computing power data and transmission rate information of the computing power card in the actual operation data, the computing power card difference value between the actual operation data of the computing power card and the theoretical operation data of the computing power card is output using the pre-trained action model. The first difference value is obtained based on at least one of the processor difference value, the memory difference value, the hard disk difference value, the network difference value, and the computing card difference value.

6. The server performance evaluation method according to claim 1, characterized in that, Before inputting the first difference value into the pre-established reward model, the process also includes: Based on the processor difference value in the first difference value, a processor reward model in the reward model is established using the task information, thread information and instruction information in the processor difference value; Based on the memory difference value in the first difference value, a memory reward model in the reward model is established using the data information and bandwidth information in the memory difference value; Based on the hard disk difference value in the first difference value, the hard disk reward model in the reward model is established using the bandwidth information in the hard disk difference value; Based on the network difference value in the first difference value, the network reward model in the reward model is established using the transmission data information in the network difference value; Based on the computing power card difference value in the first difference value, the computing power card reward model in the reward model is established using the computing power information in the computing power card difference value. Based on the energy consumption information in the first difference value, an energy consumption reward model is established in the reward model; The reward model is established based on at least one of the processor reward model, the memory reward model, the hard disk reward model, the network reward model, the computing card reward model, and the energy consumption reward model.

7. The server performance evaluation method according to claim 6, characterized in that, The step of inputting the first difference value into a pre-established reward model to output the performance score of the at least one component in the corresponding environment includes: Based on the processor difference value in the first difference value, the processor performance score of the processor in the server is calculated using the processor reward model in the pre-established reward model; Based on the memory difference value in the first difference value, the memory performance score of the memory in the server is calculated using the memory reward model in the pre-established reward model; Based on the hard disk difference value in the first difference value, the hard disk performance score of the hard disk in the server is calculated using the hard disk reward model in the pre-established reward model; Based on the network difference value in the first difference value, the network performance score of the network in the server is calculated using the network reward model in the pre-established reward model; Based on the computing card difference value in the first difference value, the computing card performance score of the computing card in the server is calculated using the computing card reward model in the pre-established reward model; Based on the energy consumption difference value in the first difference value, the energy consumption performance score of the server is calculated using the energy consumption reward model in the pre-established reward model.

8. The server performance evaluation method according to claim 6, characterized in that, in, The expression for the processor reward model is: ， in, This represents the total number of instructions for the processor's tasks. The theoretical number of threads for the processor ， This represents the processor's theoretical thread frequency. This represents the actual number of instructions that the chip can process per cycle. This refers to the actual completion time of the processor; The number of threads required by the processor; The expression for the memory reward model is: ， in, This represents the total amount of data in memory. This represents the theoretical bandwidth of the memory. This refers to the actual bandwidth of the memory. This represents the theoretical capacity of the memory. The expression for the hard disk reward model is: ， in, This refers to the total amount of data on the hard drive. This represents the theoretical bandwidth of the hard drive. This refers to the actual bandwidth of the hard drive. This represents the theoretical capacity of the hard drive. The expression for the network reward model is: ， in, This refers to the size of the data before transmission. This refers to the actual size of the data received. This refers to the actual bandwidth of the network. This represents the theoretical capacity of the network. The expression for the computing power card reward model is: ， in, This refers to the computational cost of the model. This represents the actual computing power value. This refers to the actual bandwidth of the computing card; This represents the theoretical capacity of the computing card. The expression for the energy consumption reward model is: ， in, Theoretical energy consumption; This represents the actual energy consumption value.

9. The server performance evaluation method according to claim 1, characterized in that, The process of determining the performance evaluation result of the target server based on the performance score includes: Based on the performance scores, the total number of performance scores is calculated; Based on the performance score, calculate the initial performance evaluation result of the target server; The performance evaluation result is determined based on the total number and the initial performance evaluation result.

10. A server performance evaluation device, characterized in that, include: The first acquisition module is used to collect actual operating data of at least one component of the target server in multiple environments; The first output module is used to input the actual running data into the pre-trained action model to output a first difference value between the actual running data and the corresponding theoretical running data. The first determining module is used to input the first difference value into a pre-established reward model to output the performance score of the at least one component in the corresponding environment, and to determine the performance evaluation result of the target server based on the performance score; This also includes: The second determining module is used to determine, based on the actual running data, at least one of the first structural information of the fully connected neural network, the second structural information of the batch normalization, and the functional information of the activation function in the action model before inputting the actual running data into the pre-trained action model; The first generation module is used to concatenate the fully connected neural network, the batch normalization and the activation function based on at least one of the first structural information, the second structural information and the function information to obtain the first action network in the action model; The second generation module is used to merge at least two concatenated first action networks and the first action network to obtain the first feature of the actual running data; The third generation module is used to obtain the second action network in the action model based on the first feature and the first action network; The first construction module is used to construct the action model based on the first action network and the second action network.

11. The server performance evaluation apparatus according to claim 10, characterized in that, Also includes: The second acquisition module is used to acquire training and running data of at least one component of the server in multiple environments before inputting the actual running data into the pre-trained action model. The second output module is used to input the training running data into a pre-established action model to output a second difference value between the training running data and the corresponding theoretical running data. The calculation module is used to calculate the training performance score of the at least one component in the corresponding environment based on the second difference value using a pre-established reward model; The detection module is used to detect whether the at least one component meets the preset performance conditions based on the training performance score. The fourth generation module is used to, when at least one component does not meet the preset performance conditions, use discrete estimation to train the parameter information of the pre-established motion model to obtain the trained motion model, and generate a component that meets the preset performance conditions based on the trained motion model.

12. A server, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the server performance evaluation method as described in any one of claims 1-9.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the server performance evaluation method as described in any one of claims 1-9.

Images