Intelligent computing center resource allocation method and system based on dynamic load adjustment

Abstract

The invention relates to the technical field of dynamic resource allocation, in particular to an intelligent computing center resource allocation method and system based on dynamic load adjustment, and the method comprises the following steps: collecting resource use data and dynamically adjusting parameters of a support vector machine in a cloud computing environment so as to optimize resource allocation, and carrying out the dynamic load adjustment through simulation operation and efficiency evaluation. And updating the model to improve performance and resource use efficiency, adjusting a data set by using a sliding window technology, synchronously training data and a running state, continuously optimizing resource configuration, monitoring resource use and service performance, and continuously adjusting resource allocation. According to the method, resource allocation optimization is achieved by dynamically adjusting performance parameters, load changes are effectively matched, resource waste is reduced, the influence of an allocation scheme on services is comprehensively analyzed through performance index simulation evaluation, the resource utilization efficiency and the task execution stability are improved, data set dynamic adjustment is optimized through the sliding window technology, and the service performance is improved. Data updating and operation synchronism is enhanced, and service requirements are accurately adapted.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic resource allocation, and particularly to a resource allocation method and system for an intelligent computing center based on dynamic load adjustment. Background Art

[0002] The technical field of dynamic resource allocation involves the efficient utilization and reasonable allocation of resources to meet the dynamically changing load requirements. The core content of this technical field is to conduct real-time monitoring and analysis of resource usage conditions, and perform targeted dynamic adjustments of resources, thereby achieving efficient allocation of resources among different times, locations, or tasks. This technical field systematically covers related contents such as load prediction, resource scheduling, and distributed computing, and is widely applied to scenarios such as cloud computing, data centers, and intelligent computing centers to address the dynamic changes in computing resource requirements and the imbalance in load distribution.

[0003] Among them, the resource allocation method for an intelligent computing center based on dynamic load adjustment refers to, for the dynamic load situation within the intelligent computing center, by real-time monitoring and analyzing the resource requirements of computing tasks, adopting specific resource allocation strategies to dynamically adjust the computing resources. This patent theme covers the classification and analysis of task load characteristics, combines the availability of actual resources, optimizes the resource allocation path, and realizes flexible allocation of resources through methods such as task priority division, application of load balancing algorithms, and dynamic resource migration.

[0004] The prior art lacks the ability to adjust the resource usage status in real time, making it difficult to quickly adapt to load changes, resulting in rigid resource allocation and low efficiency. There is insufficient analysis of load characteristics and task matching, and the resource allocation is out of touch with the actual demand, leading to resource waste and performance bottlenecks. There is a lack of a dynamic feedback mechanism, the optimization scheme cannot accurately reflect the operating status, the data update is out of sync with the business requirements, the resource utilization efficiency is limited, and the task delay problem is prominent in a multi-task environment. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies existing in the prior art, and to propose a resource allocation method and system for an intelligent computing center based on dynamic load adjustment.

[0006] To achieve the above purpose, the present invention adopts the following technical scheme: A resource allocation method for an intelligent computing center based on dynamic load adjustment, including the following steps:

[0007] S1: Collect the GPU usage rate, CPU usage rate, and storage consumption rate in the cloud computing environment, calculate the average value of performance indicators within a specified time window, and dynamically adjust the γ parameter and penalty coefficient C of the support vector machine until the change in performance indicators remains stable within the target threshold range, and generate the adjusted parameter values;

[0008] S2: Use the adjusted parameter values to update the weights and bias terms in the support vector machine model, recalculate the resource allocation scheme, simulate and run to evaluate the impact of the current resource allocation scheme on the running performance, and generate the current resource allocation efficiency index;

[0009] S3: Based on the current resource allocation efficiency index, analyze the resource usage situation, match the resource usage with various business requirements, adjust the resource allocation according to the business priority, optimize the resource configuration and usage efficiency, and generate the adjusted resource allocation model;

[0010] S4: According to the adjusted resource allocation model, operate the sliding window technology to dynamically adjust the size and step length of the data set, match the business changes, resample and organize the performance indicators, verify that the update of the training data set is synchronized with the running state, and establish the updated training data set;

[0011] S5: Use the updated training data set to continuously monitor and analyze the resource usage and business running performance in the cloud computing environment, optimize the resource allocation according to the continuously obtained data, evaluate the efficiency of the optimized resource configuration and the performance of the business, and generate the resource optimization evaluation result.

[0012] The adjusted parameter values include the γ parameter of the radial basis function and the penalty coefficient C. The current resource allocation efficiency index includes the resource usage situation evaluation value and the performance impact evaluation value. The adjusted resource allocation model includes resource configuration optimization and usage efficiency. The updated training data set includes data size, step length, and performance indicators. The resource optimization evaluation result includes resource configuration efficiency and business performance.

[0013] As a further solution of the present invention, the specific steps for obtaining the adjusted parameter values are as follows:

[0014] S111: Collect the GPU usage rate, CPU usage rate, and storage consumption rate data in the cloud computing environment, perform statistical analysis on each performance indicator in the target time window, and calculate the average value of the performance indicators to obtain the average value set of the performance indicators;

[0015] S112: Use the average value set of the performance indicators to preliminarily analyze the parameters of the support vector machine, calculate the influence weights based on the average value of each performance indicator, and set the γ parameter of the initial radial basis function and the penalty coefficient C of the support vector machine to establish the initial parameter set;

[0016] S113: According to the γ parameter of the radial basis function and the penalty coefficient C in the initial parameter set, combine the variance and standard deviation of the average value set of the performance indicators to adjust the parameters, using the formula:

[0017]

[0018] and

[0019]

[0020] Calculate and obtain the adjusted parameter values, and repeat the above process until the change in the performance index remains stable within the target threshold range;

[0021] Among them, α 1 represents the adjustment weight of the penalty coefficient C for the inverse root of the variance, α 2 represents the adjustment weight of the penalty coefficient C for the absolute value of the average value of the performance index, α 3 represents the adjustment weight of the penalty coefficient C for the inverse operation of the standard deviation, β 1 represents the adjustment weight of the γ parameter of the radial basis function for the ratio of the mean value to the variance, β 2 represents the adjustment weight of the γ parameter of the radial basis function for the logarithm of the variance, β 3 represents the adjustment weight of the γ parameter of the radial basis function for the inverse operation of the standard deviation, M avg represents the set of average values of the performance indicators of GPU utilization rate, CPU utilization rate, and storage consumption rate collected from the cloud computing environment, Var(M avg ) represents the variance of the average value, Mean(M avg ) represents the average value, Std(M avg ) represents the standard deviation, |M avg | represents the absolute value of the average value, C new represents the adjusted penalty coefficient, γ new represents the adjusted parameter of the radial basis function.

[0022] As a further solution of the present invention, the obtaining step of the current resource allocation efficiency index is specifically as follows:

[0023] S211: Recalculate the weights in the support vector machine model based on the adjusted parameter values, and correct the bias term to obtain the updated model parameters;

[0024] S212: Based on the updated model parameters, recalculate the resource allocation plan, perform utilization rate analysis for each resource type, use the difference between the performance requirements and the allocation capabilities as the input, calculate the allocation priorities of multiple resources, and generate a resource allocation plan;

[0025] S213: Based on the resource allocation plan, evaluate the actual performance of the plan through simulation operation, analyze the performance parameters and resource utilization rates in the simulation operation, and use the formula:

[0026]

[0027] Calculate and generate the current resource allocation efficiency index;

[0028] Among them, E current represents the current resource allocation efficiency index, P i represents the performance score of the i-th resource, U i represents the utilization rate of the i-th resource, δ is the weight adjustment coefficient affecting the simulation result, R i represents the remaining availability rate of the i-th resource, γ represents the benchmark deviation term, which is used to adjust the overall balance in the calculation, and n represents the total number of resource items participating in the calculation.

[0029] As a further solution of the present invention, the steps for obtaining the adjusted resource allocation model are specifically as follows:

[0030] S311: Based on the current resource allocation efficiency index, analyze the comparison between the resource requirements of different business types and the existing resource configuration, identify the key differences in resource usage, determine the resource requirement status of multiple businesses, and obtain the resource requirement analysis result;

[0031] S312: Use the resource requirement analysis result to evaluate the priority of the resource requirements of each business, reallocate resources using the weighted sorting method, and generate a resource priority adjustment plan;

[0032] S313: Implement the resource priority adjustment plan, combine the actual available resources and business requirements, optimize the resource allocation using the dynamic resource configuration algorithm, and use the formula:

[0033]

[0034] Obtain the adjusted resource allocation model;

[0035] Among them, M adj represents the adjusted resource allocation model, R i represents the available amount of the i-th type of resource, D i represents the demand of the i-th type of business, W i represents the weight of the i-th type of business, max(W) represents the maximum value of all business weights, and n represents the total number of business types.

[0036] As a further solution of the present invention, the steps for obtaining the updated training data set are specifically as follows:

[0037] S411: Based on the adjusted resource allocation model, use the sliding window technology to adjust the size and step of the data set, match the changes in business requirements, determine the optimal parameter settings of the window, and obtain the configured sliding window parameters;

[0038] S412: Resample the performance metric data using the configured sliding window parameters to reflect the status of business activities during the sampling process, and generate a resampled performance metric data set through the data collation process;

[0039] S413: Verify whether the resampled performance metric data set is synchronized with the system operation status, evaluate the timeliness and accuracy of the data set using statistical analysis methods, and use the formula:

[0040]

[0041] Obtain the updated training data set;

[0042] where D updated represents the updated training data set, V i is the value of the i-th data point, μ is the sample mean, σ is the sample standard deviation, C i is the adjustment coefficient used to correct the deviation of each data point, and n represents the total number of data points.

[0043] As a further solution of the present invention, the steps for obtaining the resource optimization evaluation result are specifically as follows:

[0044] S511: Utilize the updated training data set to continuously monitor the resource usage and business operation performance in the cloud computing environment, and obtain real-time monitoring results through real-time data analysis;

[0045] S512: According to the data continuously obtained from the real-time monitoring results, apply resource management algorithms to optimize resource allocation, calculate the resource utilization rate and predict future demands, and generate a resource allocation optimization plan;

[0046] S513: Adopt the resource allocation optimization plan to evaluate the optimized resource configuration efficiency and business performance, and use the formula:

[0047]

[0048] Obtain the resource optimization evaluation result by combining the resource utilization rate and business performance through the weighted average method;

[0049] where E opt represents the resource optimization evaluation result, U i represents the utilization rate of the i-th resource, T i represents the total available amount of the i-th resource, P i represents the performance weight of the i-th business, and n represents the total number of resources and businesses participating in the calculation.

[0050] Intelligent computing center resource allocation system based on dynamic load adjustment, the intelligent computing center resource allocation system based on dynamic load adjustment is used to execute the above-mentioned intelligent computing center resource allocation method based on dynamic load adjustment, and the system includes:

[0051] The performance monitoring module collects the GPU usage rate, CPU usage rate, and storage consumption rate in the cloud computing environment, calculates the average value within the target time window, dynamically adjusts the γ parameter and penalty coefficient C of the support vector machine, generates the adjusted parameter values, and generates the performance parameter adjustment result;

[0052] The resource optimization module updates the weights and bias terms in the support vector machine model based on the performance parameter adjustment result, recalculates the resource allocation plan, evaluates the running performance of the current resource allocation plan through simulation operation, and generates the resource allocation efficiency index;

[0053] The resource allocation decision module analyzes the resource requirements of multiple types of services and the current resource usage status based on the resource allocation efficiency index, reallocates resources according to the service priority, optimizes the resource configuration and usage efficiency, and generates the resource configuration model;

[0054] The data management module operates the sliding window technology to adjust the dataset size and step length based on the resource configuration model, matches the service changes, resamples and organizes the performance indicators, verifies the synchronization of the update of the training dataset and the running state, continuously monitors and analyzes the resource usage and service running performance in the cloud computing environment, optimizes the resource allocation, evaluates the efficiency of the optimized resource configuration and the service performance, and generates the resource optimization evaluation result.

[0055] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0056] In the present invention, resource allocation optimization is achieved by dynamically adjusting performance parameters, effectively matching load changes, and reducing resource waste. The impact of the allocation plan on services is comprehensively analyzed through simulation evaluation of performance indicators, improving resource utilization efficiency and task execution stability. The sliding window technology is used to optimize the dynamic adjustment of the dataset, enhancing the synchronization of data update and operation, and accurately adapting to service requirements. Continuously monitor resource usage and performance, and optimize the configuration strategy in real time, significantly improving the load adaptation ability in complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic diagram of the working process of the present invention;

[0058] Figure 2 It is a flowchart of the steps for obtaining the adjusted parameter values of the present invention;

[0059] Figure 3 It is a flowchart of the steps for obtaining the current resource allocation efficiency index of the present invention;

[0060] Figure 4 Flow chart of the acquisition steps of the adjusted resource allocation model of the present invention;

[0061] Figure 5 Flow chart of the acquisition steps of the updated training data set of the present invention;

[0062] Figure 6 Flow chart of the acquisition steps of the resource optimization evaluation result of the present invention. Detailed implementation manners

[0063] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0064] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more unless otherwise specifically defined.

[0065] Embodiment 1

[0066] Please refer to Figure 1 , the present invention provides a technical solution: an intelligent computing center resource allocation method based on dynamic load adjustment, including the following steps:

[0067] S1: Collect the GPU usage rate, CPU usage rate, and storage consumption rate in the cloud computing environment, calculate the average value of the performance indicators within a specified time window, dynamically adjust the γ parameter and penalty coefficient C of the support vector machine until the change of the performance indicators remains stable within the target threshold range, and generate the adjusted parameter values;

[0068] S2: Use the adjusted parameter values to update the weights and bias terms in the support vector machine model, recalculate the resource allocation plan, simulate and evaluate the impact of the current resource allocation plan on the operation performance, and generate the current resource allocation efficiency indicators;

[0069] S3: Based on the current resource allocation efficiency indicators, analyze the resource usage situation, match the resource usage with various business requirements, adjust the resource allocation according to the business priority, optimize the resource configuration and usage efficiency, and generate the adjusted resource allocation model;

[0070] S4: According to the adjusted resource allocation model, operate the sliding window technique to dynamically adjust the size and step length of the data set, match the business changes, resample and organize the performance metrics, verify that the update of the training data set is synchronized with the running state, and establish the updated training data set;

[0071] S5: Utilize the updated training data set to continuously monitor and analyze the resource usage and business operation performance in the cloud computing environment, optimize the resource allocation according to the continuously obtained data, evaluate the efficiency of the optimized resource configuration and the business performance, and generate the resource optimization evaluation result.

[0072] The adjusted parameter values include the γ parameter of the radial basis function and the penalty coefficient C. The current resource allocation efficiency metrics include the resource usage evaluation value and the performance impact evaluation value. The adjusted resource allocation model includes resource configuration optimization and usage efficiency. The updated training data set includes data size, step length, and performance metrics. The resource optimization evaluation result includes resource configuration efficiency and business performance.

[0073] Please refer to Figure 2 , and the specific steps for obtaining the adjusted parameter values are as follows:

[0074] S111: Collect the GPU usage rate, CPU usage rate, and storage consumption rate data in the cloud computing environment. Through statistical analysis of each performance metric in the target time window and calculation of the average value of the performance metric, obtain the average value set of the performance metrics;

[0075] Collect the GPU usage rate, CPU usage rate, and storage consumption rate data in the cloud computing environment. By deploying an automated monitoring tool, record the GPU, CPU, and memory usage data of each node in real time. These data are aggregated hourly to ensure representative performance metrics are obtained. Use database queries for data extraction, filter the data within the target time window through SQL statements, and then exclude outliers and extreme values through a data cleaning program to ensure data quality. Through average value calculation, obtain the average value of each performance metric. This average value set is used to evaluate the overall performance level of the cloud environment.

[0076] S112: Use the average value set of the performance metrics to conduct a preliminary analysis of the parameters of the support vector machine. Calculate the influence weight based on the average value of each performance metric, and set the γ parameter of the initial radial basis function and the penalty coefficient C of the support vector machine to establish the initial parameter set;

[0077] Using the set of average values of performance metrics, performance evaluation is carried out through a data analysis platform. First, each performance metric is normalized to convert data with different scales and magnitudes into a comparable format. Then, statistical methods, including mean squared error analysis, are applied to the normalized data to determine the influence weights of each metric. These weights are calculated by comparing the contribution degrees of different performance metrics to the overall system performance. Based on these weights, the system automatically adjusts the parameters of the support vector machine model to generate an initial parameter set, providing optimal initial conditions for the support vector machine.

[0078] S113: According to the γ parameter and penalty coefficient C of the radial basis function in the initial parameter set, combined with the variance and standard deviation of the set of average values of performance metrics, parameter adjustment is carried out using the formula:

[0079]

[0080] and

[0081]

[0082] Calculate to obtain the adjusted parameter values, and repeat the above process until the change in performance metrics remains stable within the target threshold range;

[0083] where, α 1 represents the adjustment weight of the penalty coefficient C to the inverse root of the variance, α 2 represents the adjustment weight of the penalty coefficient C to the absolute value of the average value of performance metrics, α 3 represents the adjustment weight of the penalty coefficient C for the inverse operation of the standard deviation, β 1 represents the adjustment weight of the γ parameter of the radial basis function to the ratio of the mean and variance, β 2 represents the adjustment weight of the γ parameter of the radial basis function to the logarithm of the variance, β 3 represents the adjustment weight of the γ parameter of the radial basis function for the inverse operation of the standard deviation, M avg represents the set of average values of performance metrics such as GPU usage rate, CPU usage rate, and storage consumption rate collected from the cloud computing environment, Var(M avg ) represents the variance of the average value, Mean(M avg ) represents the average value, Std(M avg ) represents the standard deviation, |M avg | represents the absolute value of the average value, C new represents the adjusted penalty coefficient, γ new represents the adjusted parameter of the radial basis function.

[0084] Formula:

[0085]

[0086] and

[0087]

[0088] The advantage of the formula is that by comprehensively using the variance, standard deviation and mean of the data set, the model parameters are dynamically adjusted to match the changes in the performance indicators, optimizing the response speed and accuracy of the model.

[0089] Detailed explanation of the formula and the derivation process of the formula calculation:

[0090] Set the value of M avg to the average value of 150, the variance of 20, and the standard deviation of 4.47, and substitute them into the formula to calculate C new and γ new :

[0091]

[0092] The results show that the current C new and γ new parameter values are more suitable for the performance requirements of the current cloud environment. The parameter adjustment ensures the adaptability and precision of the model under different performance changes. Through these parameter updates, the model is expected to process data more effectively and improve the prediction accuracy.

[0093] Please refer to Figure 3 , the specific steps for obtaining the current resource allocation efficiency index are as follows:

[0094] S211: Recalculate the weights in the support vector machine model based on the adjusted parameter values and correct the bias term to obtain the updated model parameters;

[0095] The adjusted parameter values update the weights and bias terms of the support vector machine model. During the process, first, each weight parameter is recalculated. The adjustment of these weight parameters is optimized based on the previous performance data and the current input variables, including evaluating each weight and adjusting it according to its impact on the model prediction accuracy. This process involves batch calculations, including gradient descent method and other optimization algorithms to ensure the optimal setting of the weights. Through this series of calculation processes, an updated model parameter is obtained. This current model parameter not only reflects the current computing requirements but also matches the current data environment.

[0096] S212: Based on the updated model parameters, recalculate the resource allocation plan, analyze the utilization rate for each resource type, use the difference between the performance requirements and the allocation capabilities as the input, calculate the allocation priorities of multiple resources, and generate the resource allocation plan;

[0097] The recalculation of the resource allocation plan is carried out by using the updated model parameters. First, the performance requirements of each resource are analyzed to determine its relative criticality in the total resource pool. Then, using the model parameters, the allocation of each resource is optimized to ensure that the allocation of each resource can not only meet the performance requirements but also maximize the overall efficiency. By simulating various resource allocation scenarios and analyzing their impact on system performance, a resource allocation plan that comprehensively considers multiple factors is obtained. This plan provides the optimal resource utilization efficiency under the current operating conditions.

[0098] S213: Based on the resource allocation plan, evaluate the actual performance of the plan through simulation operation, analyze the performance parameters and resource utilization rates in the simulation operation, and use the formula:

[0099]

[0100] Calculate and generate the current resource allocation efficiency index;

[0101] Among them, E current represents the current resource allocation efficiency index, P i represents the performance score of the i-th resource, U i represents the utilization rate of the i-th resource, δ is the weight adjustment coefficient affecting the simulation result, R i represents the remaining availability rate of the i-th resource, γ represents the benchmark deviation term, which is used to adjust the overall balance in the calculation, and n represents the total number of resource items participating in the calculation.

[0102] Formula:

[0103]

[0104] The advantage of the formula is that it not only refers to the performance score and utilization rate of each resource but also introduces the weight adjustment of the remaining availability rate, which helps to more finely evaluate the efficiency of resource allocation and thus optimize the resource allocation strategy.

[0105] Detailed explanation of the formula and the derivation process of the formula calculation:

[0106] There are three resources set, and their performance scores are P 1 = 80, P 2 = 95, P 3 = 75, and the utilization rates are U 1 = 0.90, U 2 = 0.85, U 3 = 0.80, and the remaining availability rates are R 1 = 0.10,

[0107] R 2 = 0.15, R 3= 0.20, adjustment coefficient δ = 0.05, reference deviation γ = 10.

[0108]

[0109] The results show that the current resource allocation efficiency index is 0.8206, indicating that under the given model parameters and resource status, the resource utilization efficiency is relatively high, indicating that the resource allocation optimization performed is successful. This efficiency value shows that the resource usage efficiency and model adjustment are reasonable, verifying the effectiveness of the model parameter update and resource allocation scheme.

[0110] Please refer to Figure 4 , and the steps for obtaining the adjusted resource allocation model are specifically as follows:

[0111] S311: Based on the current resource allocation efficiency index, analyze the comparison between the resource requirements of different business types and the existing resource configuration, identify the key differences in resource usage, determine the resource requirement status of multiple businesses, and obtain the resource requirement analysis result;

[0112] Analysis is carried out based on the current resource allocation efficiency index. This process involves the study of the resource requirements of multiple types of businesses and the comparison of the existing resource configuration. First, collect the real-time resource usage data and performance indicators of multiple types of businesses through a data acquisition system, and then use Python and SQL database queries to extract key resource usage and requirement data. Clean and preprocess the data to eliminate noise and outliers in the data. Then, adopt mean, variance, and trend analysis to quantitatively evaluate the resource requirements. Key indicators such as resource utilization rate and demand growth rate will be used during the quantification process. Through these steps, identify the deficiencies in the resource configuration and the key resource requirements of the businesses, providing a basis for subsequent resource priority evaluation and adjustment, thereby determining the resource requirement status of multiple businesses and ensuring the rationality of resource configuration and maximization of efficiency.

[0113] S312: Adopt the resource requirement analysis result, evaluate the priority of the resource requirements for each business, and use the weighted ranking method to reallocate resources to generate a resource priority adjustment plan;

[0114] Based on the results of resource requirement analysis, the assessment and ranking of resource priorities are carried out. This requires comparing the resource requirements of each business with the existing resource allocation to evaluate which business has the most urgent resource requirements and which resources are over-allocated. The decision matrix and weight allocation issues involved in this assessment process include using MATLAB and Excel for resource ranking and priority assessment. Through these software, the criticality of the resource requirements of multiple businesses can be scored and ranked, and resource reallocation can be carried out according to the scores. High-priority businesses will obtain corresponding resource allocations according to their resource requirements and criticality, so as to ensure that key businesses can receive sufficient resource support, optimize the resource utilization efficiency of the entire organization, and thus generate a resource priority adjustment plan. This plan will directly affect subsequent resource allocation and business execution.

[0115] S313: Implement the resource priority adjustment plan. Combining the actual available resources and business requirements, use a dynamic resource allocation algorithm to optimize resource allocation, using the formula:

[0116]

[0117] Obtain the adjusted resource allocation model;

[0118] Among them, M adj represents the adjusted resource allocation model, R i represents the available amount of the i-th type of resource, D i represents the demand of the i-th type of business, W i represents the weight of the i-th type of business, max(W) represents the maximum value of all business weights, and n represents the total number of business types.

[0119] Formula:

[0120]

[0121] The benefit of the formula is that it combines the quantity D i of business demand and the available amount R i of resources, and refers to the weight W i of the business. This method can dynamically adjust resource allocation, improve the flexibility and efficiency of resource use, and is especially suitable for environments with variable resource requirements.

[0122] Detailed explanation of the formula and the derivation process of formula calculation:

[0123] Suppose there are three types of resources, and the available amounts of Resource 1, Resource 2, and Resource 3 are R 1 = 100, R 2 = 150, R 3 = 200 respectively, and their respective business demands are D 1 = 80, D 2 = 120, D3 = 100, the business weight is W 1 = 1.0, W 2 = 1.5, W 3 = 1.2, the maximum weight max(W) = 1.5.

[0124]

[0125] M adj = (26.67 + 66.67) + (60 + 150) + (66.67 + 160)

[0126] M adj = 93.34 + 210 + 226.67 = 530.01

[0127] The results show that according to the adjusted resource allocation model M adj , the organization can effectively utilize various types of resources to meet the needs of differentiated businesses. Through dynamic weight adjustment, it ensures that key businesses with high priorities obtain sufficient resources, thereby improving the efficiency of overall resource allocation.

[0128] Please refer to Figure 5 , the specific steps for obtaining the updated training dataset are as follows:

[0129] S411: Based on the adjusted resource allocation model, use the sliding window technique to adjust the size and step of the dataset, match the changes in business requirements, determine the optimal parameter settings of the window, and obtain the configured sliding window parameters;

[0130] Based on the adjusted resource allocation model, determine the dynamic changes in resource requirements and supplies through real-time monitoring and analysis of data, and adjust the sliding window technique for dataset size and step adjustment to ensure that the dataset can flexibly respond to the immediate needs of the business. The adjustment process of the sliding window parameters includes performing regression analysis on the previous data to identify the trends in business requirement changes, adjusting the window size according to these trends to capture sufficient data without causing the processing speed to slow down due to excessive data volume. After calculation, the adjustment of the window size and step makes data processing more efficient and the response more rapid.

[0131] S412: Apply the configured sliding window parameters to resample the performance metric data, reflect the state of business activities during the sampling process, and generate a resampled performance metric dataset through the data collation process;

[0132] After completing the configuration of the sliding window parameters, the operation of resampling and organizing the performance metrics is based on the existing system performance and business traffic data. Through random sampling or conditional sampling, a representative data subset is obtained. By removing noise, normalizing, and handling missing values, the quality of the data and the accuracy of sampling are ensured. The resampled dataset can more accurately reflect the current business state and performance status.

[0133] S413: Verify whether the resampled performance metric dataset is synchronized with the system operation state. Use statistical analysis methods to evaluate the timeliness and accuracy of the dataset. Use the formula:

[0134]

[0135] Obtain the updated training dataset;

[0136] Among them, D updated represents the updated training dataset, V i is the value of the i-th data point, μ is the sample mean, σ is the sample standard deviation, C i is the adjustment coefficient used to correct the deviation of each data point, and n represents the total number of data points.

[0137] Formula:

[0138]

[0139] The benefit of the formula is that by normalizing the deviation of each data point from the mean of the dataset and multiplying by an adjustment coefficient, the influence of each data point is adjusted. This helps to reduce the impact of outliers and improve the representativeness and synchronization of the entire dataset.

[0140] Detailed explanation of the formula and the derivation process of the formula calculation:

[0141] There is a dataset set, which includes n data points, and the value of each point is V i , the mean of the dataset is μ, and the standard deviation is σ. Select the adjustment coefficient C i as the key coefficient of each point. If a point is collected during the peak period, a higher coefficient will be given. Through formula calculation, if n = 3, V i = [100, 105, 95], μ = 100, σ = 5, C i = [1.0, 1.1, 0.9], then the calculation process is:

[0142]

[0143] The result shows that after normalization and adjustment, the overall deviation of the updated training dataset is small, indicating that the dataset has good synchronization with the system operation state.

[0144] Please refer to Figure 6 , and the steps for obtaining the resource optimization evaluation result are specifically as follows:

[0145] S511: Utilize the updated training dataset to continuously monitor the resource usage and business operation performance in the cloud computing environment, and obtain real-time monitoring results through real-time data analysis;

[0146] Conduct continuous monitoring using the updated training dataset. This process involves real-time analysis of the usage of various types of resources and business operation performance in the cloud computing environment. Collect the GPU usage rate, CPU usage rate, memory usage, and network bandwidth utilization rate of the server through data collection devices, record the data every five minutes, evaluate the utilization efficiency of multiple resources through these data, and determine whether resource reallocation is required through threshold analysis. This continuous monitoring can assist system administrators in promptly identifying resource bottlenecks and making adjustments, thereby improving the overall business processing capacity.

[0147] S512: Based on the data continuously obtained from the real-time monitoring results, apply resource management algorithms to optimize resource allocation, calculate the resource usage rate and predict future demands, and generate a resource allocation optimization plan;

[0148] Optimize the resource allocation strategy according to the real-time monitoring results. This step requires analyzing the collected data, analyzing the resource consumption patterns of different businesses in multiple time periods, using Python Pandas for data cleaning and preprocessing, calculating the average resource usage rate, predicting the future resource demand trend, and then adjusting the allocation of virtual machine resources according to the prediction results and business priorities, increasing the GPU and CPU allocation for high-priority tasks and reducing the memory occupancy of low-priority tasks. Through these refined operations, the resource configuration can be made more refined and the response to business demands can be more rapid.

[0149] S513: Adopt the resource allocation optimization plan, evaluate the resource configuration efficiency and business performance after optimization, and use the formula:

[0150]

[0151] Obtain the resource optimization evaluation result by combining the resource usage rate and business performance through the weighted average method;

[0152] where E opt represents the resource optimization evaluation result, U i represents the usage rate of the i-th resource, T i represents the total available amount of the i-th resource, P i represents the performance weight of the i-th business, and n represents the total number of resources and businesses participating in the calculation.

[0153] Formula:

[0154]

[0155] The advantage of the formula is that it evaluates the efficiency of resource allocation by integrating the resource utilization rate and the business performance weight. We quantify the direct relationship between business performance and resource allocation, providing data support for resource management decisions.

[0156] Detailed explanation of the formula and the derivation process of formula calculation:

[0157] First, determine the utilization rate U of each resource i , which can be obtained through resource monitoring, including a resource utilization rate of 70%. 70% will be used as the value of U i The total resource amount T i is known. The total number of CPU cores of a server is 16 cores, so the value of T i is 16. The performance weight P i is preset according to the business priority. If the priority of business A is very high, P i can be set to 1.5, and for the lower-priority business B, P i is set to 0.5. Then substitute these values into the formula for calculation to get:

[0158]

[0159] The results show that after comprehensively referring to the resource utilization rate and the business priority weight, the resource allocation efficiency score is 7.5. This is a relatively quantitative indicator used to evaluate whether the current resource allocation plan is reasonable and to guide the resource allocation strategy in future time periods.

[0160] The intelligent computing center resource allocation system based on dynamic load adjustment is used to execute the above-mentioned intelligent computing center resource allocation method based on dynamic load adjustment. The system includes:

[0161] The performance monitoring module collects the GPU utilization rate, CPU utilization rate, and storage consumption rate in the cloud computing environment, calculates the average value within the target time window, dynamically adjusts the γ parameter and penalty coefficient C of the support vector machine, generates the adjusted parameter values, and generates the performance parameter adjustment result;

[0162] The resource optimization module updates the weights and bias terms in the support vector machine model based on the performance parameter adjustment result, recalculates the resource allocation plan, evaluates the running performance of the current resource allocation plan through simulation operation, and generates the resource allocation efficiency indicator;

[0163] Based on the resource allocation efficiency metrics, the resource allocation decision module analyzes the resource requirements of multiple types of services and the current resource usage situation, reallocates resources according to the service priorities, optimizes the resource configuration and usage efficiency, and generates a resource configuration model;

[0164] Based on the resource configuration model, the data management module operates the sliding window technology to adjust the dataset size and step length, matches the service changes, resamples and reorganizes the performance metrics, verifies the synchronization of the update of the training dataset and the running status, continuously monitors and analyzes the resource usage and service running performance in the cloud computing environment, optimizes the resource allocation, evaluates the efficiency of the optimized resource configuration and the service performance, and generates a resource optimization evaluation result.

[0165] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A resource allocation method for an intelligent computing center based on dynamic load adjustment, characterized in that: The following steps are involved: Collect GPU utilization, CPU utilization, and storage consumption rate in the cloud computing environment, calculate the average performance index within the specified time window, dynamically adjust the γ parameter and penalty coefficient C of the support vector machine until the performance index changes remain stable within the target threshold range, and generate the adjusted parameter value; Using the adjusted parameter values, updating the weights and bias terms in the support vector machine model, recalculating the resource allocation plan, simulating and running to evaluate the impact of the current resource allocation plan on the operating performance, and generating a current resource allocation efficiency index; Based on the current resource allocation efficiency index, analyze resource usage, match resource usage with multiple business requirements, adjust resource allocation according to business priorities, optimize resource configuration and usage efficiency, and generate an adjusted resource allocation model; According to the adjusted resource allocation model, the sliding window technology is operated to dynamically adjust the size and step length of the data set to match the business changes, resample and sort the performance indicators, verify the synchronization of the update of the training data set with the running status, and establish the updated training data set; The updated training data set is used to continuously monitor and analyze resource usage and business operation performance in the cloud computing environment, optimize resource allocation based on continuously acquired data, evaluate the efficiency of optimized resource configuration and business performance, and generate resource optimization evaluation results.

2. The intelligent computing center resource allocation method based on dynamic load adjustment according to claim 1 is characterized in that: The adjusted parameter values ​​include the γ parameter and penalty coefficient C of the radial basis function, the current resource allocation efficiency index includes a resource usage status evaluation value and a performance impact evaluation value, the adjusted resource allocation model includes resource configuration optimization and usage efficiency, the updated training data set includes data size, step size, and performance indicators, and the resource optimization evaluation results include resource configuration efficiency and business performance.

3. The intelligent computing center resource allocation method based on dynamic load adjustment according to claim 2 is characterized in that: The steps for obtaining the adjusted parameter value are specifically as follows: Collect GPU usage, CPU usage, and storage consumption rate data in the cloud computing environment, perform statistical analysis on each performance indicator in the target time window, and calculate the average value of the performance indicator to obtain the average value set of performance indicators; Using the average value set of the performance indicators, a preliminary analysis is performed on the parameters of the support vector machine, the influence weight of each performance indicator is calculated based on its average value, and the γ parameter and penalty coefficient C of the initial radial basis function of the support vector machine are set to establish an initial parameter set; According to the γ parameter and penalty coefficient C of the radial basis function in the initial parameter set, the parameters are adjusted in combination with the variance and standard deviation of the mean value set of the performance indicators, using the formula: and Calculate and obtain the adjusted parameter value, and repeat the above process until the change of the performance indicator remains stable within the target threshold range; Among them, α1 represents the adjustment weight of the penalty coefficient C for the inverse root of the variance, α2 represents the adjustment weight of the penalty coefficient C for the absolute value of the mean value of the performance index, α3 represents the adjustment weight of the penalty coefficient C for the inverse operation of the standard deviation, β1 represents the adjustment weight of the γ parameter of the radial basis function for the ratio of the mean and the variance, β2 represents the adjustment weight of the γ parameter of the radial basis function for the logarithm of the variance, β3 represents the adjustment weight of the γ parameter of the radial basis function for the inverse operation of the standard deviation, M avg Represents the average value of the performance indicators of GPU utilization, CPU utilization, and storage consumption rate collected from the cloud computing environment. Var(M avg ) represents the variance of the mean, Mean(M avg ) represents the mean value, Std(M avg ) represents the standard deviation, |M avg | represents the absolute value of the mean, C new represents the adjusted penalty coefficient, γ new Represents the parameters of the adjusted radial basis function.

4. The intelligent computing center resource allocation method based on dynamic load adjustment according to claim 3 is characterized in that: The steps for obtaining the current resource allocation efficiency index are specifically as follows: Recalculating the weights in the support vector machine model based on the adjusted parameter values ​​and correcting the deviation terms to obtain updated model parameters; Based on the updated model parameters, recalculate the resource allocation plan, perform utilization analysis for each resource type, use the difference between performance requirements and allocation capabilities as input, calculate the allocation priorities of multiple resources, and generate a resource allocation plan; Based on the resource allocation scheme, the actual performance of the scheme is evaluated through simulation operation, and the performance parameters and resource utilization in the simulation operation are analyzed using the formula: Calculate and generate current resource allocation efficiency indicators; Among them, E current Represents the current resource allocation efficiency index, P i represents the performance score of the i-th resource, U i represents the utilization rate of the i-th resource, δ is the weight adjustment coefficient affecting the simulation results, R i represents the remaining availability of the i-th resource, γ represents the benchmark deviation term, which is used to adjust the overall balance in the calculation, and n represents the total number of resource items involved in the calculation.

5. The intelligent computing center resource allocation method based on dynamic load adjustment according to claim 4 is characterized in that: The steps for obtaining the adjusted resource allocation model are specifically as follows: Based on the current resource allocation efficiency index, analyze the comparison between the resource requirements of differentiated business types and the existing resource configuration, identify the key differences in resource usage, determine the resource requirements of multiple businesses, and obtain resource requirement analysis results; Using the resource demand analysis results, the resource demand of each business is prioritized, and resources are reallocated using a weighted ranking method to generate a resource priority adjustment plan; Implement the resource priority adjustment plan, combine the actual available resources with business needs, and use a dynamic resource allocation algorithm to optimize resource allocation, using the formula: Obtaining an adjusted resource allocation model; Among them, M adj represents the adjusted resource allocation model, R i represents the available amount of the i-th resource, D i represents the demand for the i-th type of business, W i represents the weight of the i-th type of business, max(W) represents the maximum value of all business weights, and n represents the total number of business types.

6. The intelligent computing center resource allocation method based on dynamic load adjustment according to claim 5 is characterized in that: The steps for obtaining the updated training data set are specifically as follows: Based on the adjusted resource allocation model, the size and step length of the data set are adjusted using the sliding window technology to match the changes in business requirements, determine the optimal parameter settings of the window, and obtain the configured sliding window parameters; Apply the configured sliding window parameters to resample the performance indicator data to reflect the status of business activities during the sampling process, and generate a resampled performance indicator data set through a data sorting process; Verify whether the resampled performance indicator data set is synchronized with the system operation status, and use statistical analysis methods to evaluate the timeliness and accuracy of the data set, using the formula: Get the updated training dataset; Among them, D updated represents the updated training dataset, V i is the value of the ith data point, μ is the sample mean, σ is the sample standard deviation, and C i is the adjustment factor used to correct the deviation of each data point, and n represents the total number of data points.

7. The intelligent computing center resource allocation method based on dynamic load adjustment according to claim 6 is characterized in that: The steps for obtaining the resource optimization evaluation results are specifically as follows: Using the updated training data set, continuously monitoring resource usage and business operation performance in the cloud computing environment, and obtaining real-time monitoring results through real-time data analysis; Applying a resource management algorithm to optimize resource allocation based on the data continuously obtained from the real-time monitoring results, and calculating resource utilization and predicting future demand to generate a resource allocation optimization plan; The resource allocation optimization scheme is adopted to evaluate the resource allocation efficiency and business performance after optimization, using the formula: The resource optimization evaluation results are obtained by combining resource utilization and business performance through the weighted average method; Among them, E opt Indicates the resource optimization evaluation result, U i represents the utilization rate of the i-th resource, T i represents the total available amount of the i-th resource, P i represents the performance weight of the ith service, and n represents the total number of resources and services involved in the calculation.

8. Intelligent computing center resource allocation system based on dynamic load adjustment, characterized in that: According to the intelligent computing center resource allocation method based on dynamic load adjustment according to any one of claims 1 to 7, the system comprises: The performance monitoring module collects GPU utilization, CPU utilization, and storage consumption rate in the cloud computing environment, calculates the average value within the target time window, dynamically adjusts the γ parameter and penalty coefficient C of the support vector machine, generates adjusted parameter values, and generates performance parameter adjustment results; The resource optimization module updates the weights and bias items in the support vector machine model based on the performance parameter adjustment results, recalculates the resource allocation plan, evaluates the operating performance of the current resource allocation plan through simulation operation, and generates a resource allocation efficiency index; The resource allocation decision module analyzes the resource requirements and current resource usage of multiple types of services based on the resource allocation efficiency index, reallocates resources according to service priorities, optimizes resource allocation and usage efficiency, and generates a resource allocation model; Based on the resource allocation model, the data management module operates the sliding window technology to adjust the data set size and step size, match business changes, resample and organize performance indicators, verify that the update of the training data set is synchronized with the operating status, continuously monitor and analyze resource usage and business operation performance in the cloud computing environment, optimize resource allocation, evaluate the efficiency of the optimized resource allocation and business performance, and generate resource optimization evaluation results.

Images