Server cluster software and hardware integrated evaluation system and evaluation method

Abstract

The invention relates to the technical field of computers, and discloses a server cluster software and hardware integrated evaluation system, which comprises a data acquisition module used for acquiring hardware performance data and software performance data in a server cluster, and cleaning and normalizing the acquired data; the collaborative evaluation module is used for evaluating the hardware performance data and the software performance data; the invention also discloses a server cluster software and hardware integrated evaluation method which comprises the following steps: data acquisition: acquiring hardware performance data and software performance data in a server cluster, and cleaning and normalizing the acquired data; modeling is carried out through software and hardware cooperation, a software and hardware non-linear interaction relation is quantified, and evaluation accuracy is improved. A long short-term memory network is introduced to realize dynamic load prediction and resource optimization; a comprehensive scoring model is constructed, multi-dimensional performance indexes are quantified in a unified mode, real-time evaluation and performance bottleneck positioning are achieved, and scientific guidance is provided for resource optimization.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a server cluster software and hardware integrated evaluation system and evaluation method. Background Art

[0002] Driven by cloud computing, big data, and artificial intelligence technologies, server clusters have become the core infrastructure for high-performance computing and data processing. However, as clusters expand in size and complexity, how to comprehensively evaluate the software and hardware performance of server clusters has become a key challenge to improving system performance and efficiency. In traditional evaluation methods, software and hardware performance are often analyzed separately, which makes it impossible to accurately reflect the overall impact of the synergy between the two on system performance. This separate analysis method results in a lack of comprehensiveness and scientificity in the evaluation.

[0003] There is a complex nonlinear interactive relationship between hardware performance (such as CPU, GPU, memory and storage) and software performance (such as task response time, throughput and scheduling efficiency) in a server cluster. For example, when there is a bottleneck in hardware resources, the execution efficiency of software tasks may be seriously affected. Conversely, the reduction in software task scheduling efficiency may lead to a decrease in hardware resource utilization. Traditional evaluation methods usually assume that software and hardware performance are independent of each other, lack quantitative analysis of this synergy, and are difficult to provide effective support for performance optimization and system design.

[0004] In addition, the load of a server cluster usually changes dynamically during actual operation. Especially in cloud computing scenarios, load fluctuations may increase significantly due to changes in task allocation and resource requirements. However, most existing evaluation methods are based on static models and are difficult to adapt to dynamic load conditions. This evaluation method often shows insufficient robustness in the case of load fluctuations or emergencies, and cannot provide reliable support for real-time resource scheduling.

[0005] On the other hand, existing performance evaluation systems mostly analyze a single indicator, such as hardware utilization, software response time, or task throughput. However, the evaluation of a single indicator is often too one-sided and lacks a comprehensive measure of the overall performance of the system. In multi-indicator analysis, due to the different dimensions and importance of different indicators, existing methods also lack a unified quantitative framework to balance the weights between multiple indicators, making it difficult to fully reflect the operating status of the system.

[0006] In view of this, the present invention provides a server cluster hardware and software integrated evaluation system and evaluation method. Summary of the invention

[0007] In view of the deficiencies in the prior art, the present invention provides a server cluster hardware and software integrated evaluation system and evaluation method, which solves the problems of difficulty in quantifying the synergy of software and hardware in server clusters, insufficient robustness of dynamic load evaluation, and lack of a unified performance evaluation framework.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: a server cluster software and hardware integrated evaluation system, comprising: The data collection module is used to collect hardware performance data and software performance data in the server cluster, and clean and normalize the collected data; A collaborative evaluation module, used to construct a performance evaluation model of the collaborative effect of software and hardware based on the hardware performance data and the software performance data, and calculate the collaborative performance of the system; The load prediction module is used to predict the future load change trend of the server cluster based on the collected historical load data; A resource optimization scheduling module, used to dynamically adjust the hardware resource allocation and software task scheduling scheme of the server cluster according to the collaborative performance model and load prediction results; Comprehensive evaluation index module, which is used to construct comprehensive performance scores and output performance evaluation results based on hardware latency, software response time, and resource utilization; The visualization module is used to display hardware performance, software performance, load forecast trends and comprehensive evaluation results in real time.

[0009] Preferably, the data collected in the data acquisition module includes hardware performance data and software performance data, the hardware performance data includes CPU utilization, GPU utilization, memory usage and storage device read and write rates, and the software performance data includes task response time, task queuing time and throughput.

[0010] Preferably, the collaborative assessment module includes: A synergy model building unit, used to build a mathematical model of software-hardware synergy based on the hardware performance function H(x) and the software performance function S(x); A model parameter optimization unit, used to optimize the parameters of the synergy model based on historical data, including weight parameters α, β, γ of the synergy function; The collaborative performance calculation unit is used to input real-time performance data and calculate the collaborative performance value P(x).

[0011] Preferably, the synergy performance value of the synergy performance calculation unit is calculated by the following formula: ; Among them, P(x) represents the comprehensive performance of the system, H(x) represents the hardware performance function, S(x) represents the software performance function, and f(H(x), S(x)) represents the software and hardware synergy function. represents the cumulative effect of synergy, and x represents the input parameter.

[0012] Preferably, the resource optimization scheduling module includes: An objective function construction unit, used to construct an objective function for resource allocation optimization to maximize the collaborative performance under the constraint of the total resource amount C; Optimization algorithm unit, used to iteratively optimize the resource allocation scheme based on the gradient descent method and update the hardware resource allocation and software task allocation .

[0013] Preferably, the optimization formula of the objective function in the objective function construction unit is: ; Among them, P(x) is the synergy performance, and are the allocation of hardware resources and software tasks respectively, and C is the total amount of resources.

[0014] Preferably, the comprehensive evaluation index module calculates the comprehensive performance score by the following formula: ; in, Represents the normalized score of hardware latency, software response time, and resource utilization, Indicates the weight of each indicator.

[0015] Preferably, the load prediction module optimizes the prediction model through the following loss function: ; in, represents the value of the loss function, N represents the total number of samples of the training data, represents the actual load value at the i-th time point, Represents the predicted load value at the i-th time point.

[0016] A server cluster hardware and software integrated evaluation method comprises the following steps: Data collection: Collect hardware performance data and software performance data from the server cluster, and clean and normalize the collected data; Collaborative evaluation: Construct a performance evaluation model for the synergy of software and hardware, and calculate the collaborative performance of the system based on the model; Dynamic load forecasting: Based on historical load data, the long short-term memory network model is used to predict future load trends; Resource optimization scheduling: dynamically adjust the allocation of hardware resources and software tasks according to the collaborative performance model and load prediction results; Comprehensive performance evaluation: Build a comprehensive performance indicator system to integrate hardware performance, software performance and resource utilization for comprehensive scoring; Output results: Generate a comprehensive performance evaluation report and output performance optimization suggestions.

[0017] The present invention provides a server cluster software and hardware integrated evaluation system and evaluation method. It has the following beneficial effects: 1. In view of the problem that existing evaluation methods assume that software and hardware performance are independent and cannot quantify the synergy between the two, the present invention proposes a mathematical model of software and hardware synergy. Through hardware performance function, software performance function, and synergy function, the present invention can comprehensively describe the nonlinear interactive relationship between hardware performance and software performance. The synergy performance model optimizes parameters α, β, γ through historical data to accurately quantify the synergy effect of software and hardware, which significantly improves the scientificity and accuracy of the evaluation.

[0018] 2. Aiming at the problem that existing methods are difficult to cope with dynamic load changes in server clusters, the present invention introduces a long short-term memory network model to achieve real-time prediction of load change trends. Combined with the gradient descent method, the present invention can dynamically adjust the allocation of hardware resources and software tasks to ensure maximum collaborative performance when the load fluctuates. This solution solves the problem of insufficient robustness of traditional static evaluation and optimization methods, enabling the system to maintain high efficiency in a complex operating environment.

[0019] 3. In view of the problem that existing methods lack a unified quantitative indicator system, the present invention proposes a comprehensive scoring model based on multi-dimensional performance indicators through a comprehensive evaluation indicator module. By normalizing and weighted summing key indicators such as hardware delay, software response time and resource utilization, a unified comprehensive scoring system is formed. It can not only quantify the overall performance of the server cluster in real time, but also accurately locate performance bottlenecks, providing scientific guidance for resource optimization and system upgrades. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a framework diagram of the system of the present invention; Figure 2 A framework diagram of the collaborative evaluation module of the present invention; Figure 3 A framework diagram of the resource optimization scheduling module of the present invention; Figure 4 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Please see attached Figure 1 The embodiment of the present invention provides a server cluster hardware and software integrated evaluation system, including: The data collection module is used to collect hardware performance data and software performance data in the server cluster, and clean and normalize the collected data; A collaborative evaluation module is used to construct a performance evaluation model of the collaborative effect of software and hardware based on the hardware performance data and the software performance data, and calculate the collaborative performance of the system; The load prediction module is used to predict the future load change trend of the server cluster based on the collected historical load data; Resource optimization and scheduling module, which is used to dynamically adjust the hardware resource allocation and software task scheduling scheme of the server cluster according to the collaborative performance model and load prediction results; Comprehensive evaluation index module, which is used to construct comprehensive performance scores and output performance evaluation results based on hardware latency, software response time, and resource utilization; The visualization module is used to display hardware performance, software performance, load forecast trends and comprehensive evaluation results in real time.

[0023] Specifically, the data collection module is the foundation of the entire system and is used to collect hardware and software performance data in the server cluster in real time. Hardware performance data includes CPU and GPU utilization, memory occupancy, and disk read and write rates. These data are obtained through underlying monitoring tools such as IPMI, SNMP, or Redfish. Software performance data is collected through task monitoring tools, mainly recording task response time, throughput, and task queue time.

[0024] To ensure the accuracy of the data, the data is cleaned and normalized after collection. The data cleaning process includes filtering outliers and filling missing values ​​to eliminate interference in the monitoring process. Normalization maps each performance indicator to the [0,1] interval so that subsequent mathematical modeling can directly process heterogeneous data.

[0025] The collected performance data is processed by the collaborative evaluation module to build a performance evaluation model for the synergy of software and hardware. The system performance is quantified by the following formula: ; Among them, H(x) and S(x) represent hardware and software performance respectively, and f(H(x),S(x)) is the synergy function used to capture the complex nonlinear interaction between hardware performance and software performance.

[0026] The synergy function is obtained by fitting the historical performance data and can be expressed as: ; Among them, α, β, and γ are important parameters that reflect the synergistic effect. When optimizing the model, these parameters are fitted by the least squares method to ensure that the model can accurately describe the synergistic effect of software and hardware. Finally, the module inputs the real-time performance data into the model and calculates the comprehensive synergistic performance P(x).

[0027] The load prediction module predicts the load change trend of the server cluster in the future time window by performing time series analysis on historical load data. This system uses long short-term memory network for prediction, and its powerful time-dependent modeling ability can capture complex load change patterns.

[0028] The model input is the load data sequence of the past period, and the output is the predicted load value in the future. In order to optimize the prediction performance of the model, the loss function uses the mean square error: ; Continuously optimize model parameters through gradient descent method to reduce the predicted value With the true value The error between .

[0029] Based on the collaborative performance model and load prediction results, the resource optimization scheduling module dynamically adjusts the allocation of hardware resources and software tasks to ensure that the system performance is maximized under dynamic load conditions. This module constructs the following optimization goals: ; in, and They represent the allocation of hardware resources and software tasks respectively, and C is the total resource capacity.

[0030] The optimization process uses the gradient descent method to iteratively update: ; Through real-time monitoring and optimization, the system can dynamically schedule resources when load fluctuates, balancing hardware utilization and software execution efficiency.

[0031] In order to comprehensively quantify system performance, the comprehensive evaluation index module introduces comprehensive performance scoring : ; in, It is the normalized score of various performance indicators (such as hardware latency, software response time, and resource utilization). is the weight, which is used to reflect the importance of different indicators.

[0032] By calculating the comprehensive score, the module can locate performance bottlenecks. For example, when a score is too low (such as resource utilization is less than 50%), the module will prompt the user to make optimization adjustments.

[0033] Finally, the visualization module presents the system operation status in real time in the form of charts, including hardware performance, software performance, load forecast trends and comprehensive evaluation results. Through Grafana or similar tools, users can view the following information: Real-time usage of hardware resources (CPU / GPU utilization, memory usage).

[0034] The execution efficiency of software tasks (task throughput, response time).

[0035] Future load change trend, compared with current load.

[0036] Comprehensive performance score and bottleneck prompts.

[0037] In addition, users can customize analysis views and export reports containing performance evaluations and optimization recommendations to provide a scientific basis for cluster management.

[0038] The data collected in the data acquisition module include hardware performance data and software performance data. The hardware performance data includes CPU utilization, GPU utilization, memory usage and storage device read and write rates. The software performance data includes task response time, task queuing time and throughput.

[0039] Please see attached Figure 2 , the collaborative assessment modules include: A synergy model building unit, used to build a mathematical model of software-hardware synergy based on the hardware performance function H(x) and the software performance function S(x); A model parameter optimization unit, used to optimize the parameters of the synergy model based on historical data, including weight parameters α, β, γ of the synergy function; The collaborative performance calculation unit is used to input real-time performance data and calculate the collaborative performance value P(x).

[0040] The synergy performance value of the synergy performance calculation unit is calculated by the following formula: ; Among them, P(x) represents the comprehensive performance of the system, H(x) represents the hardware performance function, S(x) represents the software performance function, and f(H(x), S(x)) represents the software and hardware synergy function. represents the cumulative effect of synergy, and x represents the input parameter.

[0041] Specifically, in view of the above content, the collaborative evaluation module is one of the core modules of the server cluster hardware and software integrated evaluation system. Its main function is to quantify the complex interactive relationship between hardware performance and software performance based on the mathematical model of hardware and software synergy, and calculate the collaborative performance value of the system through real-time performance data, providing a scientific basis for subsequent load prediction and resource optimization scheduling. The collaborative evaluation module consists of a collaborative model construction unit, a model parameter optimization unit, and a collaborative performance calculation unit. Each unit cooperates with each other to complete the modeling of hardware and software performance synergy, parameter optimization, and dynamic calculation of performance values.

[0042] In this embodiment, the synergy model building unit is used to establish a mathematical model of the synergy between software and hardware. The core goal is to quantify the independent effects and synergy effects of hardware performance and software performance through mathematical formulas. The basic formula of the system performance model is: ; Among them: P(x) represents the comprehensive performance of the system, which is the overall performance of hardware performance H(x) and software performance S(x) and their synergy; H(x) is the hardware performance function, which reflects the operating status of hardware resources (such as CPU, GPU, memory, etc.); S(x) is the software performance function, which describes the execution efficiency of software tasks (such as response time, throughput, etc.); It is a hardware-software synergy function, used to quantify the nonlinear interaction between hardware performance and software performance; It is the cumulative effect of synergy, indicating the overall contribution of software and hardware synergy to the comprehensive performance as the input x (such as load, time) changes.

[0043] The first part of the formula, H(x)·S(x), represents the direct product of hardware and software performance, quantifying their independent contributions; the second part It captures the complex synergy between hardware and software performance.

[0044] In this embodiment, the model parameter optimization unit is used to determine the synergy function In order to accurately describe the nonlinear interaction between hardware and software, the synergy function adopts the following form: ; Among them: α, β, γ are weight parameters, which respectively describe the nonlinear contribution of hardware performance, the nonlinear contribution of software performance, and the interactive contribution between hardware and software.

[0045] The optimization process of model parameters is based on historical data, and the parameters are fitted by the least squares method: ; in: Indicates the actual measured system performance value; Represents the performance value predicted by the model.

[0046] Parameter optimization is done by gradient descent, updating the parameters at each iteration: ; Where: η is the learning rate, which is used to control the step size of parameter update; is the loss function and is the sum of squared prediction errors of historical data.

[0047] Through the above optimization process, it is ensured that the synergy function can accurately reflect the interactive effect of hardware and software performance.

[0048] In this embodiment, the function of the synergy performance calculation unit is based on the real-time collected hardware performance data H(x) and software performance data S(x), combined with the optimized synergy function , dynamically calculate the comprehensive performance value P(x) of the system. The calculation formula is: ; Among them, the hardware performance data H(x) and software performance data S(x) are collected in real time when the system is running and input into the collaborative evaluation module for analysis. The integral part of is calculated numerically, and its cumulative effect quantifies the contribution of the hardware-software synergy under the current load conditions.

[0049] In specific implementation, the synergy performance calculation unit combines the synergy function parameters α, β, γ obtained by the above optimization and dynamically solves the integral part with a numerical solution. The real-time input of hardware performance data and software performance data ensures that the system can quickly respond to load changes and output comprehensive performance P(x) as the key result of performance evaluation.

[0050] Please see attached Figure 3 , the resource optimization scheduling module includes: An objective function construction unit, used to construct an objective function for resource allocation optimization to maximize the collaborative performance under the constraint of the total resource amount C; Optimization algorithm unit, used to iteratively optimize the resource allocation scheme based on the gradient descent method and update the hardware resource allocation and software task allocation .

[0051] The optimization formula of the objective function in the objective function building unit is: ;

[0052] Among them, P(x) is the synergistic performance, and are the allocation of hardware resources and software tasks respectively, and C is the total amount of resources.

[0053] Specifically, in view of the above content, the resource optimization scheduling module is used in the server cluster hardware and software integrated evaluation system to dynamically optimize the allocation of hardware resources and software tasks to improve the overall performance of the system under dynamic load conditions. Through the collaborative work of the objective function construction unit and the optimization algorithm unit, real-time adjustment of the resource allocation plan is achieved. In the optimization of the objective function, the goal is to maximize the collaborative performance, and the gradient descent method is combined for iterative calculation to ensure that the optimal allocation strategy is obtained under the constraint of the total amount of resources.

[0054] In this embodiment, the objective function construction unit is used to define the mathematical objective function of resource allocation optimization, the core purpose of which is to maximize the comprehensive coordination performance P(x) of the system under the constraint of a given total resource amount C. The specific objective function formula is as follows: ;

[0055] Where: P(x) represents the collaborative performance of the system, which is calculated by the collaborative evaluation module; is the amount of hardware resource allocation, reflecting the usage of hardware resources at time t; is the software task allocation amount, reflecting the scheduling ratio of software tasks at time t; C is the total amount of resources, indicating the upper limit of resources currently available in the system.

[0056] In the objective function formula, the resource allocation constraint Ensure that the allocation of hardware resources and software tasks does not exceed the total resource capacity, and the optimization target maxP(x) aims to improve the contribution of software and hardware synergy to system performance through reasonable allocation.

[0057] This unit is also responsible for dynamically introducing the performance calculation results P(x) of the collaborative evaluation module into the objective function, providing performance feedback to the optimization algorithm unit to guide the adjustment of resource allocation.

[0058] In this embodiment, the optimization algorithm unit is responsible for iteratively optimizing resource allocation using the gradient descent method based on the objective function, and finally outputting the hardware resources. and software tasks The optimal allocation strategy.

[0059] 1. Iterative update of gradient descent method The optimization process is based on the following iterative formula: ; in: and are the hardware resources and software task allocation at the kth iteration respectively; η is the learning rate, which controls the step size of each iteration; and They are the partial derivatives of the comprehensive collaborative performance P(x) with respect to the hardware resource allocation and software task allocation, respectively, reflecting the marginal impact of the current resource allocation on system performance.

[0060] 2. Iterative Optimization Process The specific process of the optimization algorithm unit is as follows: Initialization: Setting the initial resource allocation scheme and , ensuring that the initial value satisfies the constraints .

[0061] Performance feedback: Calculate the collaborative performance P(x) under the current allocation scheme and update the partial derivative value based on the performance calculation results.

[0062] Gradient update: adjust resource allocation according to the gradient descent formula to generate a new allocation plan and .

[0063] Iteration termination condition: Check the convergence of the objective function. If the performance improvement is insufficient (i.e. , ϵ is the set accuracy threshold), the iteration stops; otherwise, continue to the next round of updates.

[0064] Output optimal allocation strategy: When the algorithm converges, output the current hardware resources and software task allocation.

[0065] 3. Numerical stability of optimization algorithms To improve the numerical stability of the optimization process, the optimization algorithm unit will dynamically adjust the learning rate η. During the gradient descent process, when the performance improvement is slow, η is increased to accelerate convergence; when the performance fluctuations are large, η is reduced to avoid oscillations.

[0066] The comprehensive evaluation index module calculates the comprehensive performance score through the following formula: ;

[0067] in, Represents the normalized score of hardware latency, software response time, and resource utilization, Indicates the weight of each indicator.

[0068] Specifically, the comprehensive evaluation index module is used to quantitatively analyze the overall performance of the server cluster, combining multi-dimensional indicators such as hardware performance, software performance, and resource utilization to calculate the comprehensive performance score. Through scientific scoring models and weight distribution, the comprehensive evaluation index module can output performance results that reflect the current state of the system and provide a basis for performance optimization.

[0069] Overall performance rating The calculation formula is as follows: ;

[0070] in: represents the normalized score of the i-th performance indicator, including key performance parameters such as hardware latency, software response time, and resource utilization; It represents the weight of the i-th performance indicator, reflecting the importance of this performance indicator to the overall performance of the system; m represents the total number of performance indicators.

[0071] This formula combines the performance indicators The weighted sum of , where the weight The settings can be flexibly adjusted according to specific scenarios and user needs.

[0072] In this embodiment, the normalized score calculation of the performance index is Since different performance indicators (such as hardware latency, software response time, and resource utilization) have significantly different dimensions and value ranges, the original indicators need to be normalized and mapped to a unified interval (usually [0,1]) for ease of calculation.

[0073] The normalization formula is:

[0074] Among them: the actual value is the currently measured performance indicator value; the minimum value is the best value of the indicator in the historical records; the maximum value is the worst value of the indicator in the historical records.

[0075] After normalization The closer the value is to 1, the closer the performance is to the historical optimal state; the closer the value is to 0, the more the performance deviates from the optimal state.

[0076] In this embodiment, the weight It is a key parameter that affects the comprehensive performance score. Different system scenarios and user needs have different focuses on performance indicators, so it is necessary to reasonably allocate weights. Common weight setting methods include: Even distribution: Give equal weight to all performance indicators: ;

[0077] Applicable to scenarios where indicators are equally important.

[0078] 2. Experience setting: Assign greater weights to metrics with higher importance based on empirical knowledge or actual needs. For example, in a scenario with high response requirements, you can assign a higher weight to software response time.

[0079] Dynamic Adjustment Using historical performance data or user feedback, dynamically adjust the weights to optimize the applicability of the comprehensive score. The specific formula is: ;

[0080] in, Represents the importance factor of the i-th indicator.

[0081] In this embodiment, the calculation process of the comprehensive performance score is as follows: Data input: Collect real-time performance data of the system, including hardware latency, software response time, and resource utilization.

[0082] Index normalization: Normalize the original performance data to generate a normalized score .

[0083] Weighted calculation: According to the formula , calculate the overall performance score .

[0084] Output results: Output the comprehensive performance score of the system and generate a performance evaluation report based on the score results.

[0085] The comprehensive evaluation index module can evaluate the overall performance of the server cluster in real time. Specific applications include: Performance bottleneck analysis: By observing the Quickly identify weaknesses in hardware, software, or resource utilization using a comprehensive score.

[0086] Optimization direction suggestions: Combined with weight allocation, guide users to prioritize the indicators that have the greatest impact on system performance.

[0087] Dynamic performance monitoring: through real-time calculation , helping users track system performance trends and warn of potential performance issues.

[0088] The load forecasting module optimizes the forecasting model through the following loss function: ;

[0089] in, represents the value of the loss function, N represents the total number of samples of the training data, represents the actual load value at the i-th time point, Represents the predicted load value at the i-th time point.

[0090] Specifically, the load prediction module is a key module in the server cluster hardware and software integrated evaluation system. Its core function is to use historical load data to predict future load change trends and provide forward-looking support for dynamic resource scheduling and performance optimization. During the training process of the prediction model, the module uses the mean square error as the loss function and guides the parameter optimization of the prediction model by minimizing the error function to improve the accuracy of load prediction.

[0091] The loss function used by the load prediction module is defined as follows: ;

[0092] in: Represents the value of the loss function, which is used to quantify the overall error between the predicted value and the actual value. N represents the total number of samples of the training data, that is, the number of sampling points of the historical load data. Indicates the actual load value at the i-th time point, which is the real load data of the server cluster obtained by the data acquisition module. It represents the predicted load value at the i-th time point, which is the prediction result output by the prediction model.

[0093] The core idea of ​​the loss function is to measure the overall performance of the prediction model by calculating the square of the error between the predicted value and the actual value and averaging it.

[0094] Squared error term: For each sample point, the square of the i-th error Measures the difference between the predicted value and the actual value: when = When , the square of the error is 0, indicating that the predicted value is completely accurate; When the predicted value deviates from the actual value, the square of the error increases rapidly as the deviation increases, thus applying a higher optimization weight to the predicted points with large deviations.

[0095] Error accumulation term: The squared errors of all sample points are accumulated to calculate the overall error.

[0096] Sample normalization: The sum of errors is normalized to ensure that the value of the loss function is independent of the number of samples, thereby ensuring the comparability of the model on data sets of different sizes.

[0097] Through this loss function, the load forecasting module can quantify the performance of the forecasting model on historical data, and optimize the model parameters based on this to make the forecasting results closer to the actual load value.

[0098] In this embodiment, the optimization process of model training is as follows: The load forecasting module usually uses LSTM (Long Short-Term Memory Network) as the forecasting model, and its optimization process combines the loss function Adjust the parameters. The specific process is as follows: Input data: The historical load sequence [L(tk), L(t-k+1), …, L(t-1)] is input into the LSTM network, and the network outputs the predicted value of the next time step through multi-layer time series modeling .

[0099] Calculate the error: For each sample of the training data, calculate the actual load value With the predicted load value The square of the error.

[0100] Accumulated error: The squared errors of all samples are accumulated and the loss function L is calculated.

[0101] Backpropagation and parameter update: Based on the gradient of the loss function, the model parameters are updated using an optimization algorithm (such as the Adam optimizer or stochastic gradient descent): ;

[0102] in: is the parameter of the model at the kth iteration; η is the learning rate, which is used to control the step size of parameter update; is the gradient of the loss function with respect to the model parameters.

[0103] Iteration termination condition: When the loss function L converges to a certain threshold (i.e. , ϵ is the set accuracy), stop training and output the optimized model parameters.

[0104] In this embodiment, the optimization effect of the loss function is: By using mean square error as the loss function, the load prediction module can achieve the following optimization effects: Error minimization: By minimizing , ensuring the predicted value As close to the actual value as possible , improving the prediction accuracy of the model.

[0105] Focused optimization of large error points: The square term of the loss function gives higher weight to large error points, so that the model pays more attention to sample points with large deviations during the optimization process.

[0106] Enhanced generalization ability: The average calculation method of the loss function globally constrains the model parameters, avoids overfitting of the model to certain samples, and improves the prediction performance of unseen data.

[0107] In actual operation, the load prediction module applies the trained long short-term memory network to real-time load prediction. The specific steps are as follows: Input real-time data: Input the most recent historical load series into the model and output the predicted load value for the next time step.

[0108] Rolling forecast: The model uses the current forecast value as subsequent input to achieve rolling forecasts for multiple time steps and form future load change trends.

[0109] Application of results: prediction results Provided to the resource optimization scheduling module to guide the dynamic allocation of hardware resources and software tasks.

[0110] Please see attached Figure 4 , a server cluster hardware and software integrated evaluation method, comprising the following steps: Data collection: Collect hardware performance data and software performance data from the server cluster, and clean and normalize the collected data; Collaborative evaluation: Build a performance evaluation model for the synergy of software and hardware, and calculate the collaborative performance of the system based on the model; Dynamic load forecasting: Based on historical load data, the long short-term memory network model is used to predict future load trends; Resource optimization scheduling: Dynamically adjust the allocation of hardware resources and software tasks based on the collaborative performance model and load prediction results; Comprehensive performance evaluation: Build a comprehensive performance indicator system to integrate hardware performance, software performance and resource utilization for comprehensive scoring; Output results: Generate a comprehensive performance evaluation report and output performance optimization suggestions.

[0111] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A server cluster hardware and software integrated evaluation system, characterized in that: include: The data collection module is used to collect hardware performance data and software performance data in the server cluster, and clean and normalize the collected data; A collaborative evaluation module, used to construct a performance evaluation model of the collaborative effect of software and hardware based on the hardware performance data and the software performance data, and calculate the collaborative performance of the system; The load prediction module is used to predict the future load change trend of the server cluster based on the collected historical load data; A resource optimization scheduling module, used to dynamically adjust the hardware resource allocation and software task scheduling scheme of the server cluster according to the collaborative performance model and load prediction results; Comprehensive evaluation index module, which is used to construct comprehensive performance scores and output performance evaluation results based on hardware latency, software response time, and resource utilization; The visualization module is used to display hardware performance, software performance, load forecast trends and comprehensive evaluation results in real time.

2. A server cluster hardware and software integrated evaluation system according to claim 1, characterized in that: The data collected in the data collection module includes hardware performance data and software performance data. The hardware performance data includes CPU utilization, GPU utilization, memory usage and storage device read and write rates, and the software performance data includes task response time, task queuing time and throughput.

3. A server cluster hardware and software integrated evaluation system according to claim 1, characterized in that: The collaborative assessment module includes: A synergy model building unit, used to build a mathematical model of software-hardware synergy based on the hardware performance function H(x) and the software performance function S(x); A model parameter optimization unit, used to optimize the parameters of the synergy model based on historical data, including weight parameters α, β, γ of the synergy function; The collaborative performance calculation unit is used to input real-time performance data and calculate the collaborative performance value P(x).

4. A server cluster hardware and software integrated evaluation system according to claim 3, characterized in that: The synergy performance value of the synergy performance calculation unit is calculated by the following formula: ； Among them, P(x) represents the comprehensive performance of the system, H(x) represents the hardware performance function, S(x) represents the software performance function, and f(H(x), S(x)) represents the software and hardware synergy function. represents the cumulative effect of synergy, and x represents the input parameter.

5. The server cluster hardware and software integrated evaluation system according to claim 1, characterized in that: The resource optimization scheduling module includes: An objective function construction unit, used to construct an objective function for resource allocation optimization to maximize the collaborative performance under the constraint of the total resource amount C; Optimization algorithm unit, used to iteratively optimize the resource allocation scheme based on the gradient descent method and update the hardware resource allocation and software task allocation .

6. A server cluster hardware and software integrated evaluation system according to claim 5, characterized in that: The optimization formula of the objective function in the objective function construction unit is: ； Among them, P(x) is the synergy performance, and are the allocation of hardware resources and software tasks respectively, and C is the total amount of resources.

7. A server cluster hardware and software integrated evaluation system according to claim 1, characterized in that: The comprehensive evaluation index module calculates the comprehensive performance score through the following formula: ； in, Represents the normalized score of hardware latency, software response time, and resource utilization, Indicates the weight of each indicator.

8. The server cluster hardware and software integrated evaluation system according to claim 1, characterized in that: The load prediction module optimizes the prediction model through the following loss function: ； in, represents the value of the loss function, N represents the total number of samples of the training data, represents the actual load value at the i-th time point, Represents the predicted load value at the i-th time point.

9. A server cluster hardware and software integrated evaluation method, characterized in that: The following steps are involved: Data collection: Collect hardware performance data and software performance data from the server cluster, and clean and normalize the collected data; Collaborative evaluation: Construct a performance evaluation model for the synergy of software and hardware, and calculate the collaborative performance of the system based on the model; Dynamic load forecasting: Based on historical load data, the long short-term memory network model is used to predict future load trends; Resource optimization scheduling: dynamically adjust the allocation of hardware resources and software tasks according to the collaborative performance model and load prediction results; Comprehensive performance evaluation: Build a comprehensive performance indicator system to integrate hardware performance, software performance and resource utilization for comprehensive scoring; Output results: Generate a comprehensive performance evaluation report and output performance optimization suggestions.

Images