Cloud computing resource dynamic allocation method and system based on intelligent algorithm

Abstract

The application relates to the technical field of cloud computing, in particular to a cloud computing resource dynamic allocation method and system based on an intelligent algorithm. Resource state data of a computing node and task request data of a to-be-processed task are acquired, the resource state data is compensated according to a change trend of the resource state data, a constraint relationship between resource dimensions is established and a coupling matching degree is calculated, a resource allocation instruction is generated according to the constraint relationship and the coupling matching degree to perform a scheduling allocation operation, execution effect data and a deviation of the coupling matching degree are calculated to judge an oscillation trend, and when the oscillation exists, a next period allocation instruction is generated by correcting a resource constraint condition. The application effectively avoids resource allocation oscillation, improves resource utilization efficiency and task processing stability.

Description

Technical Field

[0001] This invention relates to the field of cloud computing technology, and in particular to a method and system for dynamic allocation of cloud computing resources based on intelligent algorithms. Background Technology

[0002] In the field of dynamic resource allocation in cloud computing, existing conventional practices typically rely on real-time collection of computing node resource status data and task request data, directly generating allocation decisions through threshold comparisons or static resource matching models. The system monitors the current values ​​of individual resources such as CPU utilization, memory usage, and network bandwidth on each node. When resource utilization exceeds a preset threshold or task requests exceed node capacity, resource migration or elastic scaling operations are immediately triggered. Furthermore, some solutions employ a resource matching degree calculation method based on simple weighted summation, linearly combining values ​​from multiple resource dimensions and comparing them with task requirements to select the optimal target node. These methods generally assume that resource status remains stable over a short period and equate historical data collection timelines with the actual state at the decision-making time.

[0003] The aforementioned conventional practices suffer from unavoidable time delays between data collection and decision execution. Directly using historical resource status data leads to allocation decisions lagging behind actual resource changes, especially in scenarios with drastic load fluctuations. The "current state" upon which the decision is based may have shifted significantly, resulting in resource allocation misalignment or repeated switching between overload and idle periods. Furthermore, when resource demand competition among multiple computing nodes is intense, allocation strategies based on a single threshold or simple weighted matching are prone to frequent switching of resource allocation instructions between different nodes, creating oscillations. These oscillations not only incur additional system migration overhead but also disrupt the stability of the cloud computing environment and reduce overall resource utilization efficiency. Current technologies lack effective mechanisms for identifying and suppressing the reverse adjustment trend of resource allocation instructions, failing to fundamentally eliminate the resource fluctuation problem caused by repeated decision adjustments. Summary of the Invention

[0004] This invention provides a method and system for dynamic allocation of cloud computing resources based on intelligent algorithms, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides a method for dynamic allocation of cloud computing resources based on intelligent algorithms, comprising:

[0006] Obtain resource status data from multiple computing nodes and task request data from tasks to be processed. The resource status data includes the occupied values ​​corresponding to multiple resource dimensions, and the task request data includes the demand values ​​corresponding to multiple resource dimensions.

[0007] Based on the changing trend of resource status data over continuous acquisition time, the resource status offset between acquisition time and decision time is predicted, and the resource status data is compensated using the resource status offset to obtain compensated resource status data.

[0008] Establish constraints between multiple resource dimensions, and calculate the coupling matching degree by jointly calculating the occupied value of compensation resource status data and the demand value of task request data based on the constraints.

[0009] Based on the coupling matching degree and preset resource constraints, resource allocation instructions are generated. Based on the resource allocation instructions, resource scheduling and allocation operations are performed on the computing nodes to obtain execution effect data.

[0010] The deviation between the execution effect data and the coupling matching degree is calculated, and the direction of the deviation change within the continuous decision-making cycle is used to determine whether there is an oscillating trend of repeated reverse adjustment of resource allocation instructions;

[0011] When an oscillation trend is detected, the preset resource constraints are modified to obtain the modified resource constraints. Based on the modified resource constraints, the resource allocation instructions for the next decision cycle are generated.

[0012] Based on the changing trend of resource status data over continuous acquisition time, the resource status offset between the acquisition time and the decision time is predicted. The resource status offset is then used to compensate for the resource status data, resulting in compensated resource status data, including:

[0013] The change trend of resource status data over continuous collection time is decomposed according to multiple resource dimensions to obtain trend change components and fluctuation change components.

[0014] Based on the time interval between the data collection time and the decision time, the fusion weights corresponding to the trend change component and the fluctuation change component are determined respectively. The trend change component and the fluctuation change component are weighted and fused using the fusion weights to obtain the initial offsets corresponding to each of the multiple resource dimensions.

[0015] Based on the compensation errors generated by each of the multiple resource dimensions in multiple historical decision-making cycles, compensation coefficients are set for the corresponding resource dimensions. The initial offsets are corrected using the compensation coefficients to obtain the resource status offsets corresponding to each of the multiple resource dimensions.

[0016] The resource status data is compensated by using resource status offsets to compensate the occupied values ​​of the corresponding resource dimensions in the resource status data for multiple resource dimensions, thus obtaining compensated resource status data.

[0017] Establish constraints between multiple resource dimensions, and based on these constraints, jointly calculate the occupancy value of compensation resource status data and the demand value of task request data to obtain the coupling matching degree, including:

[0018] Obtain historical usage data of the corresponding usage values ​​of multiple resource dimensions over multiple historical decision-making periods. Determine the correlation strength between each resource dimension and other resource dimensions based on the historical usage data, and determine the correlation type between each resource dimension and other resource dimensions based on the direction of change of the historical usage data.

[0019] Based on the association strength and association type, set constraint coefficients between each resource dimension and other resource dimensions, and use the constraint coefficients to construct constraint relationships between multiple resource dimensions;

[0020] Based on the constraint relationship, the occupancy values ​​corresponding to other related resource dimensions are adjusted according to the constraint coefficient to obtain the adjusted occupancy values.

[0021] Calculate the matching degree between the corrected occupied value and the corresponding resource dimension requirement value in the task request data to obtain the single-dimensional matching degree for each resource dimension.

[0022] Based on the proportion of the demand value corresponding to each resource dimension in the task request data to the total demand value, the task sensitivity corresponding to each resource dimension is determined, and the fusion coefficient of the single-dimensional matching degree corresponding to each resource dimension is determined in combination with the value of the constraint coefficient.

[0023] The coupling matching degree is obtained by weighted fusion calculation of multiple single-dimensional matching degrees using fusion coefficients.

[0024] Based on the association strength and association type, constraint coefficients are set between each resource dimension and other resource dimensions. The constraint relationships between multiple resource dimensions are constructed using these constraint coefficients, including:

[0025] Obtain historical usage data of the corresponding usage values ​​of multiple resource dimensions in multiple historical decision-making periods, and calculate the correlation coefficient between each resource dimension and other resource dimensions as the correlation strength.

[0026] The historical occupancy data is differentially processed to obtain a change sequence. The covariance between the change sequences is calculated, and the correlation type (positive or negative) is determined based on the sign of the covariance.

[0027] The comprehensive correlation degree is obtained by fusing the absolute values ​​of the correlation strength and covariance. Positive or negative coefficients are assigned to the comprehensive correlation degree according to the correlation type. The comprehensive correlation degree with coefficients is normalized to obtain the initial constraint coefficients.

[0028] Calculate the total absolute value of the initial constraint coefficients of all other resource dimensions associated with each resource dimension as the cumulative impact value. Based on the cumulative impact value, scale the initial constraint coefficients proportionally to obtain the adjusted constraint coefficients.

[0029] The adjusted constraint coefficients are used to construct constraint relationships between multiple resource dimensions.

[0030] Resource allocation instructions are generated based on the coupling matching degree and preset resource constraints. Resource scheduling and allocation operations are then performed on the computing nodes according to these instructions. The execution effect data includes:

[0031] Obtain the capacity boundary parameters in the preset resource constraints, compare the coupling matching degree with the capacity boundary parameters to identify the over-limit resource dimensions, extract the over-limit quantity of the over-limit resource dimensions and convert it into compression weights;

[0032] The compressed demand is obtained by weighting and reducing the demand of the task to be processed in each resource dimension according to the compression weight. The safety margin parameter in the preset resource constraints is obtained, and the deviation between the compressed demand and the safety margin parameter is calculated.

[0033] Based on the deviation, the adaptability of allocable resources for multiple computing nodes is evaluated and a node ranking sequence is generated. The first computing node in the node ranking sequence is selected as the target computing node. The target computing node identifier is extracted and combined with the compressed demand to generate a resource allocation instruction.

[0034] Based on the resource allocation instructions, resource scheduling and allocation operations are performed on the target computing node. The target computing node adjusts its resource quota according to the resource allocation instructions and starts the pending tasks. Resource consumption data and task completion time of each resource dimension are collected during the execution of the pending tasks to obtain execution effect data.

[0035] Calculate the deviation between the execution effect data and the coupling matching degree, and judge whether the resource allocation instructions have an oscillating trend of repeated reverse adjustments based on the direction of change of the deviation within a continuous decision-making cycle, including:

[0036] Extract the actual consumption of each resource dimension from the execution effect data, calculate the deviation value by the difference between the actual consumption and the expected resource occupancy corresponding to the coupling matching degree, and determine the sign of the deviation value in multiple consecutive decision cycles and record it as a deviation sign sequence.

[0037] The number of symbol flips is counted by comparing adjacent elements of the biased symbol sequence, and the ratio of the number of symbol flips to the number of decision cycles is used as an indicator of oscillation frequency.

[0038] The oscillation frequency index is compared with the preset oscillation threshold. When the oscillation frequency index exceeds the preset oscillation threshold, it is determined that the resource allocation instruction has an oscillation trend of repeated reverse adjustments.

[0039] When an oscillation trend is identified, the preset resource constraints are modified to produce revised resource constraints. Based on these revised constraints, resource allocation instructions for the next decision cycle are generated, including:

[0040] When it is determined that there is an oscillation trend, the deviation values ​​of each resource dimension in multiple consecutive decision cycles are extracted to form deviation time series data. The extreme values ​​of the deviation time series data are extracted to obtain the peak deviation and the valley deviation. The difference between the peak deviation and the valley deviation is calculated as the oscillation amplitude. The oscillation amplitude is multiplied by the oscillation frequency index to obtain the oscillation intensity index.

[0041] The resource dimensions are sorted in descending order according to the magnitude of the oscillation intensity index to form an oscillation sensitive sequence. The resource dimensions with the highest ranking in the oscillation sensitive sequence are extracted as key control dimensions. The capacity boundary parameters corresponding to the key control dimensions in the preset resource constraints are extracted. The expansion ratio of the capacity boundary parameters is calculated according to the oscillation intensity index, and the capacity boundary parameters are amplified and adjusted to obtain the corrected resource constraints.

[0042] Based on the amplified and adjusted capacity boundary parameters in the revised resource constraints, the allocatable space for each resource dimension is recalculated, and resource allocation instructions for the next decision cycle are generated based on the allocatable space.

[0043] A second aspect of this invention provides a cloud computing resource dynamic allocation system based on intelligent algorithms, comprising:

[0044] The data acquisition unit is used to acquire resource status data of multiple computing nodes and task request data of tasks to be processed. The resource status data includes the occupied values ​​corresponding to multiple resource dimensions, and the task request data includes the demand values ​​corresponding to multiple resource dimensions.

[0045] The offset compensation unit is used to predict the resource status offset between the acquisition time and the decision time based on the changing trend of resource status data during continuous acquisition time, and to compensate the resource status data using the resource status offset to obtain compensated resource status data.

[0046] The coupling matching unit is used to establish constraint relationships between multiple resource dimensions. Based on the constraint relationships, it jointly calculates the occupancy value of the compensation resource status data and the demand value of the task request data to obtain the coupling matching degree.

[0047] The resource allocation unit is used to generate resource allocation instructions based on the coupling matching degree and preset resource constraints, perform resource scheduling and allocation operations on computing nodes according to the resource allocation instructions, and obtain execution effect data.

[0048] The oscillation judgment unit is used to calculate the deviation between the execution effect data and the coupling matching degree, and judge whether the resource allocation instruction has an oscillation trend of repeated reverse adjustment based on the direction of change of the deviation within the continuous decision cycle.

[0049] The constraint correction unit is used to suppressively correct the preset resource constraints when it is determined that there is an oscillation trend, to obtain the corrected resource constraints, and to generate the resource allocation instruction for the next decision cycle based on the corrected resource constraints.

[0050] A third aspect of the present invention provides an electronic device, comprising:

[0051] processor;

[0052] Memory used to store processor-executable instructions;

[0053] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0054] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0055] This method significantly improves the real-time accuracy of resource status data by acquiring multi-dimensional resource status data and task requirement data from multiple computing nodes and introducing resource status offset compensation based on the changing trends at continuous acquisition times. The compensated resource status data effectively eliminates the time delay error between acquisition and decision-making, making resource allocation decisions closer to actual load changes and greatly reducing resource waste or under-allocation problems caused by data lag.

[0056] This invention, based on establishing constraints across multiple resource dimensions, jointly calculates the coupling matching degree using compensation resource status data and task request data. This scientifically quantifies the comprehensive suitability between computing node resources and task requirements. The coupling matching degree, combined with resource allocation instructions generated from preset resource constraints, significantly improves the accuracy of resource scheduling, ensuring full and balanced utilization of computing node resources and effectively avoiding resource overload or idleness in a single dimension.

[0057] By continuously calculating the deviation between execution performance data and coupling matching degree, and analyzing the direction of change of this deviation within consecutive decision-making cycles, this method can accurately identify whether resource allocation instructions exhibit an oscillating trend of repeated reverse adjustments. Once the existence of an oscillation trend is confirmed, the preset resource constraints are immediately modified to suppress it, thereby automatically eliminating system oscillations in subsequent decision-making cycles. This mechanism significantly enhances the stability and convergence of the dynamic allocation process, avoids performance fluctuations caused by resource contention, and ensures the long-term efficient operation of the cloud computing platform.

[0058] The revised resource constraints further optimize the generation logic of resource allocation instructions, enabling the entire allocation system to adaptively adjust to dynamic changes. This method significantly reduces resource management overhead and improves overall resource utilization and system robustness while ensuring task execution quality, making it suitable for large-scale, high-concurrency cloud computing environments. Attached Figure Description

[0059] Figure 1 This is a flowchart illustrating a dynamic resource allocation method for cloud computing based on intelligent algorithms.

[0060] Figure 2 This is a flowchart of task scheduling and allocation under resource constraints. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0063] Figure 1 This is a flowchart illustrating the dynamic allocation method for cloud computing resources based on intelligent algorithms according to an embodiment of the present invention.

[0064] Cloud computing resource dynamic allocation methods based on intelligent algorithms include:

[0065] Obtain resource status data from multiple computing nodes and task request data from tasks to be processed. The resource status data includes the occupied values ​​corresponding to multiple resource dimensions, and the task request data includes the demand values ​​corresponding to multiple resource dimensions.

[0066] Based on the changing trend of resource status data over continuous acquisition time, the resource status offset between acquisition time and decision time is predicted, and the resource status data is compensated using the resource status offset to obtain compensated resource status data.

[0067] Establish constraints between multiple resource dimensions, and calculate the coupling matching degree by jointly calculating the occupied value of compensation resource status data and the demand value of task request data based on the constraints.

[0068] Based on the coupling matching degree and preset resource constraints, resource allocation instructions are generated. Based on the resource allocation instructions, resource scheduling and allocation operations are performed on the computing nodes to obtain execution effect data.

[0069] The deviation between the execution effect data and the coupling matching degree is calculated, and the direction of the deviation change within the continuous decision-making cycle is used to determine whether there is an oscillating trend of repeated reverse adjustment of resource allocation instructions;

[0070] When an oscillation trend is detected, the preset resource constraints are modified to obtain the modified resource constraints. Based on the modified resource constraints, the resource allocation instructions for the next decision cycle are generated.

[0071] During the data acquisition phase, resource status data is collected in real time from multiple computing nodes via a monitoring agent. This data covers multiple resource dimensions, including key resource metrics such as processor utilization, memory usage, storage space usage, and network bandwidth usage. Each resource dimension corresponds to a numerical value reflecting the current resource occupancy. Simultaneously, the system receives task request data from users or the scheduler. This data includes the task's requirements across various resource dimensions, such as the number of processor cores needed, required memory capacity, expected storage space size, and network bandwidth requirements. This data provides the foundation for subsequent resource allocation decisions.

[0072] Based on the changing trend of resource status data over continuous acquisition time, the resource status offset between the acquisition time and the decision time is predicted. The resource status offset is then used to compensate for the resource status data, resulting in compensated resource status data, including:

[0073] The change trend of resource status data over continuous collection time is decomposed according to multiple resource dimensions to obtain trend change components and fluctuation change components.

[0074] Based on the time interval between the data collection time and the decision time, the fusion weights corresponding to the trend change component and the fluctuation change component are determined respectively. The trend change component and the fluctuation change component are weighted and fused using the fusion weights to obtain the initial offsets corresponding to each of the multiple resource dimensions.

[0075] Based on the compensation errors generated by each of the multiple resource dimensions in multiple historical decision-making cycles, compensation coefficients are set for the corresponding resource dimensions. The initial offsets are corrected using the compensation coefficients to obtain the resource status offsets corresponding to each of the multiple resource dimensions.

[0076] The resource status data is compensated by using resource status offsets to compensate the occupied values ​​of the corresponding resource dimensions in the resource status data for multiple resource dimensions, thus obtaining compensated resource status data.

[0077] In scenarios involving dynamic allocation of cloud computing resources, there is often a significant time lag between the collection of resource status data and the actual execution decision. During this time lag, the resource occupancy status of computing nodes may have changed. If the raw data collected is used directly for allocation decisions, a discrepancy will arise between the decision and the actual resource status, thus affecting the accuracy of task scheduling. Therefore, predictive compensation of resource status data is needed to mitigate the error caused by this time lag.

[0078] For each resource dimension, the changing trend of resource status data over continuous acquisition time is decomposed. Specifically, the numerical sequence of resource occupancy for the same resource dimension at continuous acquisition time includes both trend-driven components and fluctuation-driven components caused by short-term random fluctuations. Trend-driven components reflect the monotonic increase or decrease of resource occupancy over a longer time scale, such as the continuously rising CPU utilization rate with the arrival of peak business periods; fluctuation-driven components reflect the random fluctuations in resource occupancy caused by factors such as sudden task bursts and scheduling jitter in a short period of time. When decomposing the two types of components, methods such as moving average filtering can be used to extract the trend component from the original change sequence, and the difference between the original change amount and the trend component is taken as the fluctuation component, thereby decoupling the change patterns of different natures.

[0079] After the decomposition is completed, the time interval between the data collection time and the decision-making time is used as the basis for the decomposition. Determine the fusion weights for the trend-based change component and the volatility-based change component separately. Time interval. The larger the value, the further the decision-making time is from the most recent data collection time. In this case, the resource status is more likely to be dominated by trend-based patterns, while the impact of short-term fluctuations will tend to average out over a longer period; conversely, when... When the value is small, the volatility component has a stronger ability to describe the immediate state and should be given a higher fusion weight. Based on this logic, the fusion weight corresponding to the trend change component should be... Fusion weights corresponding to volatility variation components satisfy ,and Follow It increases with the increase of [something]. Follow The value decreases as the resource increases. Using the aforementioned fusion weights, the two types of components are weighted and fused to obtain the initial offset for each resource dimension. The calculation method is as follows:

[0080]

[0081] in, Indicates the first Trend changes in each resource dimension Indicates the first The initial offset is calculated by weighted fusion of the volatility components of each resource dimension, achieving a balance between trend prediction and volatility correction, thus avoiding excessive prediction bias caused by relying solely on a single component.

[0082] The compensation effects of different resource dimensions vary across historical decision-making cycles. For some resource dimensions (such as memory usage), the changes are relatively stable, resulting in a small estimation error between the initial offset and the actual offset. However, for other resource dimensions (such as network bandwidth), which are significantly affected by external traffic, there may be a more significant systematic deviation between the initial offset and the actual offset. To eliminate this estimation error caused by differences in dimensional characteristics, a compensation coefficient needs to be introduced to correct the initial offset.

[0083] Specifically, the compensation error generated by each resource dimension over multiple historical decision-making periods is statistically analyzed. The compensation error is defined as the difference between the actual resource state and the predicted state after compensation in a given decision-making period. The compensation errors from multiple historical periods are summarized and analyzed to extract the overall bias characteristics of the error for that dimension, and compensation coefficients are then set for the corresponding resource dimensions based on these characteristics. The principle for setting the compensation coefficient is: if the historical compensation error shows a pattern of continuous overestimation, then... A correction factor less than 1 is set to suppress overestimation of the offset; if the historical compensation error shows a pattern of continuous underestimation, then... A correction factor greater than 1 is set to amplify the estimated magnitude of the offset. The initial offset is corrected using a compensation coefficient to obtain the... The final resource state offset for each resource dimension : ,in, The system is dynamically updated based on the statistical characteristics of historical compensation errors. After each complete decision-making cycle, the system is adjusted according to the newly generated compensation errors. Rolling corrections are performed to enable the compensation coefficient to adapt to changes in resource behavior patterns, thus preventing the compensation coefficient from becoming ineffective due to drift in resource usage patterns.

[0084] After obtaining the resource status offsets corresponding to each resource dimension Then, compensation is performed on the occupancy values ​​of the corresponding dimensions in the resource status data according to resource dimensions. Let's assume a computing node is at the collection time... The resource dimension usage value is The occupied value in the compensation resource status data of this dimension for: By independently performing the above compensation operations on each resource dimension, compensated resource status data covering all resource dimensions is ultimately generated. Compared to the original collected data, the compensated resource status data better reflects the actual resource occupancy of computing nodes at the time of decision-making, thus providing a more accurate input basis for subsequent coupling matching degree calculation.

[0085] It is worth noting that the compensation operation is performed independently for each resource dimension, rather than applying the same offset to all dimensions. This is because the changing patterns, fluctuation ranges, and historical error characteristics of different resource dimensions vary, and uniform compensation would lead to overcompensation in some dimensions and undercompensation in others. By establishing trend decomposition models for each dimension, independently determining the fusion weights, and setting individual compensation coefficients, refined offset predictions can be made based on the characteristics of each resource dimension. This improves the accuracy of the overall compensated resource status data and provides reliable data support for subsequent resource allocation decisions.

[0086] Establish constraints between multiple resource dimensions, and based on these constraints, jointly calculate the occupancy value of compensation resource status data and the demand value of task request data to obtain the coupling matching degree, including:

[0087] Obtain historical usage data of the corresponding usage values ​​of multiple resource dimensions over multiple historical decision-making periods. Determine the correlation strength between each resource dimension and other resource dimensions based on the historical usage data, and determine the correlation type between each resource dimension and other resource dimensions based on the direction of change of the historical usage data.

[0088] Based on the association strength and association type, set constraint coefficients between each resource dimension and other resource dimensions, and use the constraint coefficients to construct constraint relationships between multiple resource dimensions;

[0089] Based on the constraint relationship, the occupancy values ​​corresponding to other related resource dimensions are adjusted according to the constraint coefficient to obtain the adjusted occupancy values.

[0090] Calculate the matching degree between the corrected occupied value and the corresponding resource dimension requirement value in the task request data to obtain the single-dimensional matching degree for each resource dimension.

[0091] Based on the proportion of the demand value corresponding to each resource dimension in the task request data to the total demand value, the task sensitivity corresponding to each resource dimension is determined, and the fusion coefficient of the single-dimensional matching degree corresponding to each resource dimension is determined in combination with the value of the constraint coefficient.

[0092] The coupling matching degree is obtained by weighted fusion calculation of multiple single-dimensional matching degrees using fusion coefficients.

[0093] After obtaining the compensation resource status data, it is necessary to further establish constraints between multiple resource dimensions to jointly evaluate the resource status of each dimension with task requirements, ultimately obtaining a coupling matching degree that reflects the overall adaptability of multi-dimensional resources. In a cloud computing environment, resource dimensions such as CPU utilization, memory usage, network bandwidth, and disk I / O are not independent but have significant interrelationships. For example, when CPU utilization is high, memory access frequency often increases; when network bandwidth usage increases, disk I / O buffer usage also increases accordingly. Ignoring these coupling relationships between dimensions will lead to biases when evaluating each dimension individually, thus affecting the accuracy of resource allocation decisions.

[0094] Obtaining historical usage data for multiple resource dimensions across various decision-making periods is fundamental to establishing constraints. This historical usage data encompasses the actual usage sequence of each resource dimension on each computing node over several past decision-making periods. Based on this historical data, for each pair of resource dimensions, the correlation strength between their usage sequences is calculated, serving as a quantitative representation of the association strength. Let resource dimensions... With resource dimension The correlation strength between them is It is calculated using the Pearson correlation coefficient of historical occupancy numerical series. The larger the absolute value, the closer the resource consumption changes between the two dimensions.

[0095] While determining the strength of the association, it is also necessary to determine the type of association based on the direction of change in historical occupancy data. If the dimension... When the occupied value increases, the dimension If the occupancy value also shows an upward trend, then there is a positive correlation between the two; if the dimension When the occupied value increases, the dimension If the occupancy value shows a downward trend, then there is a negative correlation between the two. The correlation type is determined by... The symbols directly reflect this: positive values ​​correspond to positive associations, and negative values ​​correspond to negative associations. Association strength and association type together constitute the basic description of the constraint relationships between resource dimensions.

[0096] Based on the strength and type of association, set constraint coefficients between each resource dimension and other resource dimensions. The values ​​of the constraint coefficients should be considered comprehensively. The absolute value of the sign and magnitude, when A larger value indicates stronger coupling between the two dimensions, and the absolute value of the corresponding constraint coefficient should also be larger; when When the values ​​are small, the coupling between the two dimensions is weaker, and the absolute value of the constraint coefficients is correspondingly smaller. The sign of the constraint coefficients is consistent with the association type: positive associations correspond to positive constraint coefficients, and negative associations correspond to negative constraint coefficients. All constraint coefficients are utilized. Construct a constraint matrix describing the interrelationships between multiple resource dimensions; this matrix represents the constraints between resource dimensions. In the constraint matrix, diagonal elements are set to 1, indicating that each dimension's influence on itself is unit influence; off-diagonal elements represent the corresponding values. .

[0097] Based on the constraints, the compensated occupancy value for each resource dimension is adjusted. Specifically, for resource dimensions... Its corrected value It is obtained by weighting and summing the post-compensation occupancy value of its own resource with the post-compensation occupancy values ​​of other related resource dimensions according to the corresponding constraint coefficients. ,in For the first The value occupied after compensation for each resource dimension For dimension With dimension The constraint coefficients between them. The corrected value. This comprehensively reflects the impact of other resource dimensions on the dimension. The impact of the actual available resource status can more accurately describe the real resource pressure of the current node in this dimension than simply using the compensated usage value.

[0098] After obtaining the corrected resource occupancy values ​​for each resource dimension, the degree of matching between these values ​​and the corresponding resource dimension requirement values ​​in the task request data is calculated to obtain the single-dimensional matching degree for each resource dimension. The calculation method for one-dimensional matching degree is as follows: Let the resource dimension in the task request data be... The corresponding demand value is The computation node is in dimension The available resource value is (in (where the total capacity of this dimension is), then the single-dimensional matching degree This reflects the degree of fit between available resources and demand. When available resources are sufficient and demand can be met... A higher value is preferred when available resources are scarce or when the occupied value after correction is close to the total capacity. The value is relatively low. By calculating the single-dimensional matching degree for each resource dimension separately, a set of evaluation results reflecting the independent adaptation of each dimension is obtained.

[0099] Determine the fusion coefficient for the single-dimensional matching degree corresponding to each resource dimension. In this case, two aspects of information need to be considered. Firstly, based on the proportion of the demand value for each resource dimension in the task request data to the total demand value, the task sensitivity of that dimension to the current task needs to be determined. ,Right now ,in For the first The required values ​​for each resource dimension A larger value indicates a higher dependence of the current task on resources in that dimension. On the other hand, the value of the constraint coefficient reflects the degree of coupling between that dimension and other dimensions. Dimensions with larger absolute values ​​of constraint coefficients are more influenced by other dimensions in the overall resource status, and their single-dimensional matching degree carries more information. Fusion Coefficient Overall task sensitivity The aggregation amount is determined by the absolute value of the constraint coefficients and normalized so that the sum of the fusion coefficients of all dimensions is 1, thereby ensuring the comparability of the weighted fusion calculation results.

[0100] Using fusion coefficient For multiple one-dimensional matching degrees Perform weighted fusion to obtain the coupling matching degree. ,Right now Coupling matching degree This comprehensively reflects the overall adaptability of current computing nodes across multiple resource dimensions, considering both the mutual constraints and influences between these dimensions, and applying differentiated weighting based on the actual resource requirements of each dimension according to the task. A higher value indicates that the current computing node has a strong comprehensive capability in meeting task requirements; when A low coupling matching degree indicates that the current node has resource bottlenecks in certain dimensions, or is affected by coupling in other dimensions, resulting in insufficient overall adaptability. By using coupling matching degree, the most suitable computing node for executing the current task can be selected more accurately in subsequent resource allocation decisions, avoiding unreasonable resource allocation problems caused by ignoring the coupling relationship between dimensions.

[0101] Based on the association strength and association type, constraint coefficients are set between each resource dimension and other resource dimensions. The constraint relationships between multiple resource dimensions are constructed using these constraint coefficients, including:

[0102] Obtain historical usage data of the corresponding usage values ​​of multiple resource dimensions in multiple historical decision-making periods, and calculate the correlation coefficient between each resource dimension and other resource dimensions as the correlation strength.

[0103] The historical occupancy data is differentially processed to obtain a change sequence. The covariance between the change sequences is calculated, and the correlation type (positive or negative) is determined based on the sign of the covariance.

[0104] The comprehensive correlation degree is obtained by fusing the absolute values ​​of the correlation strength and covariance. Positive or negative coefficients are assigned to the comprehensive correlation degree according to the correlation type. The comprehensive correlation degree with coefficients is normalized to obtain the initial constraint coefficients.

[0105] Calculate the total absolute value of the initial constraint coefficients of all other resource dimensions associated with each resource dimension as the cumulative impact value. Based on the cumulative impact value, scale the initial constraint coefficients proportionally to obtain the adjusted constraint coefficients.

[0106] The adjusted constraint coefficients are used to construct constraint relationships between multiple resource dimensions.

[0107] In constructing constraint relationships among multiple resource dimensions, it is necessary to extract statistical patterns from historical operational data to quantify the degree of mutual influence between these dimensions. Historical resource usage data for multiple resource dimensions over several decision-making periods is obtained as the data foundation for subsequent analysis. This historical usage data encompasses measured values ​​of various resource dimensions, such as CPU utilization, memory utilization, network bandwidth utilization, and disk I / O utilization, over several consecutive decision-making periods. Based on this historical data, the Pearson correlation coefficient between each resource dimension and other resource dimensions is calculated, and this coefficient is used as the correlation strength. This is used to measure the degree of linear correlation between two dimensions in terms of their numerical trends. The range of values ​​is The larger the absolute value, the closer the trend of change between the two resource dimensions; the smaller the absolute value, the weaker the correlation between the two.

[0108] Correlation coefficients alone are insufficient to distinguish whether the influence between resource dimensions is increasing or decreasing in the same direction or in opposite directions. First-order differencing is performed on historical occupancy data to obtain the change sequence of each resource dimension between adjacent decision-making periods. Difference eliminates the baseline bias in the original data, making the sequence more reflective of the dynamic fluctuations in resource occupancy. The covariance is calculated for the change sequences of any two resource dimensions, denoted as . The sign of covariance directly reflects the consistency of the two dimensions in the direction of change: when When two resource dimensions tend to increase or decrease simultaneously, the association type is determined to be a positive association; when When two resource dimensions tend to change in opposite directions—that is, when one increases, the other decreases—the relationship is determined to be negative. This judgment mechanism can effectively distinguish between competitive and cooperative relationships between resource dimensions. For example, CPU-intensive tasks and memory-intensive tasks often show a positive correlation in resource consumption, while high network I / O tasks and local disk I / O tasks may show a negative correlation in certain scenarios.

[0109] After obtaining the correlation strength and covariance information, the correlation strength will be... absolute value and covariance The absolute values ​​are combined to obtain the comprehensive correlation degree. The fusion operation employs a weighted linear combination method, specifically as follows: ,in and To integrate the weighting coefficients, satisfy the following conditions: All values ​​are greater than 0. By normalizing the absolute value of the covariance, dimensional differences are eliminated, bringing it within a comparable range to the correlation coefficient. Overall correlation degree. This comprehensively reflects the statistical correlation between the two resource dimensions; a larger value indicates a more significant mutual influence between the two dimensions.

[0110] Assign a directional coefficient to the overall correlation degree based on the correlation type: when the correlation type is positive, assign a positive coefficient. The signed comprehensive correlation degree is obtained. When the association type is negative association, a negative coefficient is assigned. ,get The signed comprehensive correlation degree between all resource dimension pairs is normalized to map it to... Within the interval, the initial constraint coefficients are obtained. The normalization formula is Initial constraint coefficients The symbol reflects the direction of the association, and the absolute value reflects the strength of the association, providing a directional quantitative basis for the subsequent construction of constraint relationships.

[0111] There's a problem with directly using initial constraint coefficients: a resource dimension might be strongly correlated with multiple other dimensions. Directly aggregating these constraint coefficients could lead to over-constraint of that dimension, resulting in overly conservative or distorted resource allocation. To address this issue, each resource dimension needs to be calculated... The total absolute value of the initial constraint coefficients of all other associated resource dimensions is used as the cumulative impact value of that dimension. The calculation method is as follows Impact on cumulative value Reflects the resource dimension The total amount of associated impact experienced throughout the entire resource-constrained network. When When the value is large, it indicates a strong correlation between this dimension and several other dimensions, requiring an appropriate reduction in its initial constraint coefficients to avoid excessive constraint concentration; when... When the initial constraint coefficient is small, a larger proportion is retained.

[0112] The initial constraint coefficients are scaled proportionally based on the cumulative impact value to obtain the adjusted constraint coefficients. The calculation method is as follows ,in This is a scaling adjustment factor used to control the sensitivity of scaling. It takes the value of a real number greater than 0 and can be empirically set based on the number of resource dimensions and their correlation density in a real cloud computing environment. When the value is large, the denominator increases. It is moderately compressed, thereby limiting the cumulative constraint strength borne by a single dimension; when When it approaches 0, Approaching This retains all the information of the original constraint coefficients. This scaling mechanism effectively balances the influence weight of each resource dimension in the constraint network, avoiding a decrease in the numerical stability of resource allocation calculations due to excessive correlation of a certain dimension.

[0113] Using the adjusted constraint coefficient Establish constraints between multiple resource dimensions. These constraints will be used when subsequently correcting the compensation resource status data. The corrected value occupied By adjusting the constraint coefficients and the compensated occupied values ​​of each associated dimension The constraint relationships are obtained through weighted summation, thus incorporating cross-dimensional interactions into the resource status assessment. This constraint relationship, constructed based on the statistical characteristics of historical data, can adaptively reflect the actual coupling patterns between resource dimensions in different cloud computing environments. Compared to manually preset fixed constraint coefficients, it has stronger data-driven adaptability and higher resource matching accuracy. In large-scale cloud computing cluster scenarios with a large number of resource dimensions, the above constraint coefficient calculation and scaling mechanism can effectively control the sparsity of the constraint matrix, reduce the complexity of subsequent joint calculations, and retain the main constraint information between key dimensions, ensuring the rationality and reliability of the coupling matching degree calculation results.

[0114] like Figure 2 As shown, Figure 2 This is a flowchart illustrating task scheduling and allocation under resource constraints in an embodiment of the present invention.

[0115] Resource allocation instructions are generated based on the coupling matching degree and preset resource constraints. Resource scheduling and allocation operations are then performed on the computing nodes according to these instructions. The execution effect data includes:

[0116] Obtain the capacity boundary parameters in the preset resource constraints, compare the coupling matching degree with the capacity boundary parameters to identify the over-limit resource dimensions, extract the over-limit quantity of the over-limit resource dimensions and convert it into compression weights;

[0117] The compressed demand is obtained by weighting and reducing the demand of the task to be processed in each resource dimension according to the compression weight. The safety margin parameter in the preset resource constraints is obtained, and the deviation between the compressed demand and the safety margin parameter is calculated.

[0118] Based on the deviation, the adaptability of allocable resources for multiple computing nodes is evaluated and a node ranking sequence is generated. The first computing node in the node ranking sequence is selected as the target computing node. The target computing node identifier is extracted and combined with the compressed demand to generate a resource allocation instruction.

[0119] Based on the resource allocation instructions, resource scheduling and allocation operations are performed on the target computing node. The target computing node adjusts its resource quota according to the resource allocation instructions and starts the pending tasks. Resource consumption data and task completion time of each resource dimension are collected during the execution of the pending tasks to obtain execution effect data.

[0120] After obtaining the coupling matching degree, it needs to be combined with preset resource constraints to generate and execute resource allocation instructions. The preset resource constraints include capacity boundary parameters, which describe the upper limit threshold that each resource dimension is allowed to be occupied by tasks in the current computing node cluster. The coupling matching degree is compared dimension by dimension with the capacity boundary parameters to determine which resource dimensions' matching requirements exceed the allowable range of the capacity boundary, and these dimensions are marked as over-limit resource dimensions. Identifying over-limit resource dimensions essentially involves locating constraint conflict points in the multi-dimensional resource space, ensuring that subsequently generated resource allocation instructions will not lead to node overload or resource contention.

[0121] For each resource dimension exceeding its capacity limit, the excess amount beyond the limit is extracted. This excess amount reflects the gap between the current task demand and available capacity. The excess amounts for each resource dimension are normalized and converted into compression weights. The compression weight is proportional to the excess amount; the larger the excess amount, the stronger the compression of the corresponding dimension. Specifically, let the... The excess quantity of each excess resource dimension is The sum of all transcendental dimensions and transcendental quantities is Then the first Compression weights for each super-limit dimension satisfy This normalization process ensures that the sum of the compression weights for all over-limit dimensions is 1, facilitating subsequent proportional reduction of demand. For resource dimensions that are not over-limit, their compression weights are reset to zero, and they do not participate in the reduction operation, thus avoiding unnecessary contraction of non-bottleneck resource dimensions.

[0122] The compressed demand is obtained by weighting the demand for each resource dimension of the task to be processed according to the compression weight. The calculation method for the compressed demand is: for the resource dimension that exceeds the limit, the original demand is multiplied by... The reduced demand is obtained; for non-over-limit dimensions, the demand remains unchanged. The purpose of this reduction process is to compress the resource requirements of the task to within the capacity boundary, while preserving the core resource requirements of the task as much as possible, and avoiding excessive reduction that could lead to a decline in task execution quality. The compressed demand serves as input for subsequent node adaptation evaluation, directly affecting the node ranking results and the final allocation scheme.

[0123] After obtaining the compressed demand, a safety margin parameter is retrieved from the preset resource constraints. The safety margin parameter represents the proportion of resource reserves that a computing node should reserve during normal operation, used to cope with sudden fluctuations in resource demand and prevent task failure or response delays due to full-load operation. The deviation between the compressed demand and the safety margin parameter is calculated; the deviation measures the distance between the compressed task demand and the node's safety margin boundary. Suppose a computing node is at position... The available capacity of each resource dimension is The safety margin parameter is Then the upper limit of safe allocation for this dimension is The compressed demand is denoted as The corresponding deviation satisfy The smaller the deviation, the closer the compression demand is to the safe allocation limit, and the more reasonable the resource utilization of the node.

[0124] Based on the deviation of each computing node across all resource dimensions, a suitability assessment is performed on multiple computing nodes. The core idea of ​​the suitability assessment is to comprehensively measure the degree to which each node matches the compression requirements in a multi-dimensional resource space. The overall deviation of each node is obtained by summing the deviations across dimensions. The lower the overall deviation, the better the resource distribution of the node matches the compression requirements of the task, and the higher the suitability. All computing nodes are sorted from low to high based on their overall deviation, generating a node ranking sequence. The node ranking sequence intuitively reflects the suitability of each node for carrying the current task; nodes ranked higher have better resource surplus distribution and can complete task scheduling with less resource waste.

[0125] The first computing node with the lowest overall deviation from the node sorting sequence is selected as the target computing node. The node identifier of the target computing node is extracted and encapsulated with the compression requirement to generate a resource allocation instruction. This instruction contains the target computing node identifier, the compression requirement values ​​for each resource dimension, and the corresponding resource type label, forming a structured scheduling instruction message for the resource scheduler to parse and execute. This resource allocation instruction generation process ensures that every scheduling decision has a clear target node and precise resource quotas, avoiding resource waste or conflicts caused by fuzzy allocation.

[0126] Based on resource allocation instructions, resource scheduling and allocation operations are performed on the target computing node. Upon receiving the resource allocation instructions, the target computing node adjusts its resource quota according to the compression requirements of each resource dimension in the instructions, including binding the number of CPU cores, setting the memory allocation limit, configuring network bandwidth rate limits, and adjusting storage I / O priorities. After completing the resource quota adjustment, the target computing node starts the pending task, which begins execution within the limited resource quota. During task execution, resource consumption data for each resource dimension is continuously collected, including actual CPU utilization, peak memory usage, network traffic, and storage read / write throughput. The complete duration from task start to task completion is also recorded. The above resource consumption data and task completion time together constitute execution effect data. This execution effect data serves as input for subsequent oscillation trend judgment and resource constraint correction, forming a closed-loop dynamic adjustment mechanism that enables the entire cloud computing resource allocation process to continuously and adaptively optimize the allocation strategy.

[0127] Calculate the deviation between the execution effect data and the coupling matching degree, and judge whether the resource allocation instructions have an oscillating trend of repeated reverse adjustments based on the direction of change of the deviation within a continuous decision-making cycle, including:

[0128] Extract the actual consumption of each resource dimension from the execution effect data, calculate the deviation value by the difference between the actual consumption and the expected resource occupancy corresponding to the coupling matching degree, and determine the sign of the deviation value in multiple consecutive decision cycles and record it as a deviation sign sequence.

[0129] The number of symbol flips is counted by comparing adjacent elements of the biased symbol sequence, and the ratio of the number of symbol flips to the number of decision cycles is used as an indicator of oscillation frequency.

[0130] The oscillation frequency index is compared with the preset oscillation threshold. When the oscillation frequency index exceeds the preset oscillation threshold, it is determined that the resource allocation instruction has an oscillation trend of repeated reverse adjustments.

[0131] After executing resource allocation instructions, a systematic feedback evaluation of the actual execution results is required to determine whether the current allocation strategy has fallen into an oscillation state. Oscillation is a typical unstable behavior in cloud computing resource scheduling, manifested as resource allocation instructions repeatedly adjusting in opposite directions within adjacent decision cycles, causing the resource occupancy of computing nodes to fluctuate between overload and idle, severely affecting overall scheduling efficiency.

[0132] Extracting the actual resource consumption for each dimension from the execution performance data is the first step in oscillation detection. Execution performance data is reported by the compute nodes after completing their tasks, including the actual consumption of various resource dimensions such as actual CPU utilization, actual memory usage, actual network bandwidth throughput, and actual storage I / O read / write volume. These actual consumption figures reflect the true resource requirements of the task running on the current compute node, and there is often a certain discrepancy between them and the estimated requirements in the task request data.

[0133] The deviation value for each resource dimension within the current decision-making cycle is calculated by subtracting the actual consumption from the expected resource occupancy corresponding to the coupling matching degree. The expected resource occupancy, derived from the coupling matching degree calculation, reflects the amount of resources the system expects the task to consume under the joint evaluation of the current resource status and task requirements. The sign of the deviation value has a clear physical meaning: a positive deviation value indicates that actual consumption exceeds expectations, suggesting that the current allocation is too low and resources are insufficient; a negative deviation value indicates that actual consumption is lower than expected, suggesting that the current allocation is too high and resources are redundant and wasted. Let the first... The deviation value of a certain resource dimension within a decision cycle is ,but ,in For this resource dimension in the 1st Actual consumption within a decision-making cycle This represents the corresponding expected resource usage.

[0134] The sign of the deviation values ​​over multiple consecutive decision-making periods is determined, and each deviation value is mapped to a symbol. The specific rule is: when At that time, the record symbol is marked as ;when At that time, the record symbol is marked as ;when At that time, the record symbol is marked as This indicates that the expectation and the reality are completely in line. (Continuous) The symbols within each decision cycle are arranged sequentially to form a bias symbol sequence. ,in Indicates the first The deviation sign markings corresponding to each decision cycle This represents the total number of decision cycles involved in oscillation detection. The deviation sign sequence fully records the history of positive and negative fluctuations in resource allocation results over time, serving as the foundational data for subsequent oscillation frequency calculations.

[0135] The biased sign sequence is compared between adjacent elements, and the number of sign flips is counted. A sign flip is defined as a switch between positive and negative directions between two adjacent non-zero sign markers. At that time, it was believed that in the first The and the first A sign flip occurs once between each decision cycle. The cumulative number of sign flips is denoted as (the sum of the pairs of adjacent non-zero elements in the biased sign sequence). Flip the sign a number of times. The number of effective decision cycles (i.e., the number of non-zero symbols in the sequence, denoted as...) The ratio of ) is defined as the oscillation frequency index. The calculation method is as follows The denominator This represents the total number of valid adjacent element pairs. Oscillation frequency index. The range of values ​​is ,when near When this occurs, it indicates that a sign flip has occurred between almost every adjacent decision-making cycle, meaning that the direction of resource allocation is constantly and repeatedly switching, with extremely violent oscillations; when near When the deviation direction remains stable, the resource allocation strategy tends to converge.

[0136] oscillation frequency index With preset oscillation threshold Perform a comparison. Preset oscillation threshold. Configured based on actual cloud computing environment scheduling experience, typically set to... to The specific value can be adjusted based on the cluster size, task type, and service level agreement requirements. When the current resource allocation instruction is determined to have an oscillating trend of repeated reverse adjustments, a subsequent suppressive correction mechanism is triggered; when If the current allocation strategy is determined to be in a normal convergence state, no oscillation suppression is required, and the allocation instructions for the next decision cycle are generated according to the original resource constraints.

[0137] In practical applications, the window length for oscillation detection The selection of [aspect name] has a significant impact on detection sensitivity and stability. If... If the value is too small, the bias sign sequence will be too short, and the statistical results will be easily affected by occasional fluctuations, leading to misjudgments; if... If the value is too large, the detection response will be significantly delayed, and the oscillation trend may persist for a considerable period before it can be identified. Considering both detection timeliness and statistical reliability, it is recommended to... Set as to Each decision cycle. Within this window, if the oscillation frequency index continues to exceed the preset oscillation threshold, the existence of the oscillation trend can be confirmed with high confidence, and a suppressive correction to the preset resource constraints can be initiated. This smooths the resource allocation amplitude in the next decision cycle, avoids repeated oscillations between overload and idle in the scheduling strategy, and ensures the overall stability and efficiency of cloud computing resource allocation.

[0138] Furthermore, in multi-resource-dimensional scenarios, oscillation detection can calculate the oscillation frequency index for each resource dimension separately, and take the weighted average of the oscillation frequency indices of each dimension as the comprehensive oscillation frequency index. The weights can be configured with reference to the task sensitivity corresponding to each resource dimension, thereby more comprehensively reflecting the overall oscillation status of the resource allocation strategy in the multi-dimensional space, and providing a more accurate input basis for subsequent inhibitory correction.

[0139] When an oscillation trend is identified, the preset resource constraints are modified to produce revised resource constraints. Based on these revised constraints, resource allocation instructions for the next decision cycle are generated, including:

[0140] When it is determined that there is an oscillation trend, the deviation values ​​of each resource dimension in multiple consecutive decision cycles are extracted to form deviation time series data. The extreme values ​​of the deviation time series data are extracted to obtain the peak deviation and the valley deviation. The difference between the peak deviation and the valley deviation is calculated as the oscillation amplitude. The oscillation amplitude is multiplied by the oscillation frequency index to obtain the oscillation intensity index.

[0141] The resource dimensions are sorted in descending order according to the magnitude of the oscillation intensity index to form an oscillation sensitive sequence. The resource dimensions with the highest ranking in the oscillation sensitive sequence are extracted as key control dimensions. The capacity boundary parameters corresponding to the key control dimensions in the preset resource constraints are extracted. The expansion ratio of the capacity boundary parameters is calculated according to the oscillation intensity index, and the capacity boundary parameters are amplified and adjusted to obtain the corrected resource constraints.

[0142] Based on the amplified and adjusted capacity boundary parameters in the revised resource constraints, the allocatable space for each resource dimension is recalculated, and resource allocation instructions for the next decision cycle are generated based on the allocatable space.

[0143] Upon detecting an oscillation trend, instead of directly applying the same constraint corrections to all resource dimensions, the system first quantifies the specific performance of each resource dimension during the oscillation period to implement targeted, differentiated inhibitory corrections. Specifically, deviation values ​​for each resource dimension are extracted over multiple consecutive decision-making periods and arranged chronologically to form independent time-series deviation data for each resource dimension. Extreme values ​​are extracted from the time-series deviation data, identifying the maximum deviation value within the statistical window as the peak deviation and the minimum deviation value as the trough deviation. The difference between the peak and trough deviations yields the oscillation amplitude for that resource dimension during the oscillation period. The oscillation amplitude directly reflects the maximum fluctuation range of resource allocation for that dimension between adjacent decision-making periods. The oscillation amplitude is then compared with the previously calculated oscillation frequency index. The product operation yields the oscillation intensity index. The oscillation intensity index comprehensively reflects both the severity and frequency of oscillations. Resource dimensions with large oscillation amplitudes and high frequencies will obtain higher oscillation intensity indices, indicating that resource scheduling in that dimension is in a more unstable state and requires priority intervention.

[0144] Let the first The peak deviation of each resource dimension within the statistical window is: Valley deviation is Then the oscillation amplitude in that dimension satisfy Oscillation intensity index satisfy ,in For the first The oscillation frequency index corresponding to each resource dimension is calculated separately for each dimension in the preceding oscillation detection step. Different resource dimensions often have different oscillation characteristics. For example, CPU utilization may exhibit high-frequency, low-amplitude oscillations, while memory usage may exhibit low-frequency, high-amplitude fluctuations. By comprehensively quantifying the oscillation intensity index, the degree of instability of each dimension can be compared horizontally on a unified scale.

[0145] After calculating the oscillation intensity index for each resource dimension, all resource dimensions participating in scheduling are sorted in descending order according to the magnitude of the oscillation intensity index, forming an oscillation-sensitive sequence. The resource dimension ranking higher in the oscillation-sensitive sequence has a larger oscillation intensity index, indicating that the resource allocation of that dimension has repeatedly experienced large-scale, high-frequency reverse adjustments within the historical decision window, which is the main source of overall scheduling instability. Several resource dimensions with high rankings are extracted from the oscillation-sensitive sequence as key control dimensions. The number of key control dimensions can be dynamically determined based on the distribution characteristics of the oscillation intensity index; for example, resource dimensions with oscillation intensity indices exceeding the average of all dimensions can be selected, or a fixed number of dimensions from the top few positions in the sequence can be directly selected. The control scope is concentrated on the resource dimensions with the most significant oscillations, avoiding unnecessary interference to stable resource dimensions, thereby ensuring the accuracy and locality of the correction operation.

[0146] For the identified key control dimensions, the capacity boundary parameters corresponding to these dimensions are extracted from the preset resource constraints. These capacity boundary parameters describe the upper limit of resource allocation allowed for that dimension under the current resource constraints and are the core constraint basis for generating resource allocation instructions. The root cause of the oscillation phenomenon lies in the overly tight setting of the capacity boundary parameters, leading to frequent alternating adjustments of over-limit compression and relaxation near the boundary in task resource demands. Therefore, the core idea of ​​the inhibitory correction is to appropriately increase the capacity boundary parameters, providing more adjustable space for resource allocation, thereby reducing the frequency of triggering boundary constraints and eliminating the root cause of repeated reverse adjustments.

[0147] The expansion ratio of the capacity boundary parameters for each key control dimension is calculated based on the oscillation intensity index. The expansion ratio is positively correlated with the oscillation intensity index; the larger the oscillation intensity index, the higher the instability of that dimension, requiring a greater relaxation of the capacity boundary. Let the ... The capacity boundary parameters for each key regulatory dimension are: The expansion ratio is The corrected capacity boundary parameters satisfy Expansion ratio From the oscillation intensity index After normalization, it is compared with the preset expansion coefficient. Multiplying them together yields the result, i.e. ,in for The relative intensity value after normalization of all key control dimensions The preset hyperparameters for controlling the overall expansion range are typically set between 0.05 and 0.3 to ensure a moderate correction range and avoid resource waste due to excessive relaxation of constraints. After the above calculations, the capacity boundary parameters of each key control dimension are amplified and adjusted. The adjusted results replace the original capacity boundary parameters of the corresponding dimensions in the preset resource constraints, resulting in the corrected resource constraints. The capacity boundary parameters of non-key control dimensions remain unchanged to maintain the continuity and stability of the overall scheduling strategy.

[0148] Based on the revised resource constraints, the allocatable space for each resource dimension is recalculated. The allocatable space equals the difference between the revised capacity boundary parameter and the corresponding dimension's occupancy value in the current compensated resource status data, reflecting the actual amount of remaining resources available for new tasks under the current system load. For key control dimensions, since the capacity boundary parameter has been amplified, its allocatable space is larger than before the revision, resulting in more sufficient margin for task resource demands at the boundary, thus reducing the probability of triggering over-limit compression operations. Based on the updated allocatable space for each resource dimension, combined with the demand values ​​for each dimension in the task request data, resource matching and allocation calculations are re-executed to generate resource allocation instructions for the next decision cycle. The generated resource allocation instructions will allocate computing node resources to the tasks to be processed under more relaxed constraint boundaries, allowing resource scheduling to gradually converge from an oscillating state to a stable state.

[0149] In subsequent decision-making cycles, continuously monitor changes in deviation values ​​across various resource dimensions, and if the oscillation frequency index... Reduce to the preset oscillation threshold If the oscillation trend in this dimension has been effectively suppressed, the corrected capacity boundary parameters can be gradually restored to their original settings, achieving a gradual recovery of the constraints and preventing a decline in resource utilization caused by maintaining excessively loose constraints for an extended period. The differentiated capacity boundary correction mechanism driven by the oscillation intensity index can maximize the overall resource utilization efficiency of the system while ensuring the stability of resource scheduling.

[0150] A second aspect of this invention provides a cloud computing resource dynamic allocation system based on intelligent algorithms, comprising:

[0151] The data acquisition unit is used to acquire resource status data of multiple computing nodes and task request data of tasks to be processed. The resource status data includes the occupied values ​​corresponding to multiple resource dimensions, and the task request data includes the demand values ​​corresponding to multiple resource dimensions.

[0152] The offset compensation unit is used to predict the resource status offset between the acquisition time and the decision time based on the changing trend of resource status data during continuous acquisition time, and to compensate the resource status data using the resource status offset to obtain compensated resource status data.

[0153] The coupling matching unit is used to establish constraint relationships between multiple resource dimensions. Based on the constraint relationships, it jointly calculates the occupancy value of the compensation resource status data and the demand value of the task request data to obtain the coupling matching degree.

[0154] The resource allocation unit is used to generate resource allocation instructions based on the coupling matching degree and preset resource constraints, perform resource scheduling and allocation operations on computing nodes according to the resource allocation instructions, and obtain execution effect data.

[0155] The oscillation judgment unit is used to calculate the deviation between the execution effect data and the coupling matching degree, and judge whether the resource allocation instruction has an oscillation trend of repeated reverse adjustment based on the direction of change of the deviation within the continuous decision cycle.

[0156] The constraint correction unit is used to suppressively correct the preset resource constraints when it is determined that there is an oscillation trend, to obtain the corrected resource constraints, and to generate the resource allocation instruction for the next decision cycle based on the corrected resource constraints.

[0157] A third aspect of the present invention provides an electronic device, comprising:

[0158] processor;

[0159] Memory used to store processor-executable instructions;

[0160] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0161] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0162] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic allocation of cloud computing resources based on intelligent algorithms, characterized in that, include: Obtain resource status data from multiple computing nodes and task request data from tasks to be processed. The resource status data includes the occupied values ​​corresponding to multiple resource dimensions, and the task request data includes the demand values ​​corresponding to multiple resource dimensions. Based on the changing trend of resource status data during continuous acquisition, the resource status offset between the acquisition time and the decision time is predicted, and the resource status data is compensated using the resource status offset to obtain compensated resource status data. Establish constraints between multiple resource dimensions, and calculate the coupling matching degree by jointly calculating the occupied value of compensation resource status data and the demand value of task request data based on the constraints. Based on the coupling matching degree and preset resource constraints, resource allocation instructions are generated. Based on the resource allocation instructions, resource scheduling and allocation operations are performed on the computing nodes to obtain execution effect data. The deviation between the execution effect data and the coupling matching degree is calculated, and the direction of the deviation change within the continuous decision-making cycle is used to determine whether there is an oscillating trend of repeated reverse adjustment of resource allocation instructions; When an oscillation trend is identified, the preset resource constraints are modified to obtain the modified resource constraints. Based on the modified resource constraints, resource allocation instructions for the next decision cycle are generated.

2. The method according to claim 1, characterized in that, Based on the changing trend of resource status data over continuous acquisition time, the resource status offset between the acquisition time and the decision time is predicted. The resource status offset is then used to compensate for the resource status data, resulting in compensated resource status data, including: The change trend of resource status data over continuous collection time is decomposed according to multiple resource dimensions to obtain trend change components and fluctuation change components. Based on the time interval between the data collection time and the decision time, the fusion weights corresponding to the trend change component and the fluctuation change component are determined respectively. The trend change component and the fluctuation change component are weighted and fused using the fusion weights to obtain the initial offsets corresponding to each of the multiple resource dimensions. Based on the compensation errors generated by each of the multiple resource dimensions in multiple historical decision-making cycles, compensation coefficients are set for the corresponding resource dimensions. The initial offsets are corrected using the compensation coefficients to obtain the resource status offsets corresponding to each of the multiple resource dimensions. The resource status data is compensated by using resource status offsets to compensate the occupied values ​​of the corresponding resource dimensions in the resource status data for multiple resource dimensions, thus obtaining compensated resource status data.

3. The method according to claim 1, characterized in that, Establish constraints between multiple resource dimensions, and based on these constraints, jointly calculate the occupancy value of compensation resource status data and the demand value of task request data to obtain the coupling matching degree, including: Obtain historical usage data of the corresponding usage values ​​of multiple resource dimensions over multiple historical decision-making periods. Determine the correlation strength between each resource dimension and other resource dimensions based on the historical usage data, and determine the correlation type between each resource dimension and other resource dimensions based on the direction of change of the historical usage data. Based on the association strength and association type, set constraint coefficients between each resource dimension and other resource dimensions, and use the constraint coefficients to construct constraint relationships between multiple resource dimensions; Based on the constraint relationship, the occupancy values ​​corresponding to other related resource dimensions are adjusted according to the constraint coefficient to obtain the adjusted occupancy values. Calculate the matching degree between the corrected occupied value and the corresponding resource dimension requirement value in the task request data to obtain the single-dimensional matching degree for each resource dimension. Based on the proportion of the demand value corresponding to each resource dimension in the task request data to the total demand value, the task sensitivity corresponding to each resource dimension is determined, and the fusion coefficient of the single-dimensional matching degree corresponding to each resource dimension is determined in combination with the value of the constraint coefficient. The coupling matching degree is obtained by weighted fusion calculation of multiple single-dimensional matching degrees using fusion coefficients.

4. The method according to claim 3, characterized in that, Based on the association strength and association type, constraint coefficients are set between each resource dimension and other resource dimensions. The constraint relationships between multiple resource dimensions are constructed using these constraint coefficients, including: Obtain historical usage data of the corresponding usage values ​​of multiple resource dimensions in multiple historical decision-making periods, and calculate the correlation coefficient between each resource dimension and other resource dimensions as the correlation strength. The historical occupancy data is differentially processed to obtain a change sequence. The covariance between the change sequences is calculated, and the correlation type (positive or negative) is determined based on the sign of the covariance. The comprehensive correlation degree is obtained by fusing the absolute values ​​of the correlation strength and covariance. Positive or negative coefficients are assigned to the comprehensive correlation degree according to the correlation type. The comprehensive correlation degree with coefficients is normalized to obtain the initial constraint coefficients. Calculate the total absolute value of the initial constraint coefficients of all other resource dimensions associated with each resource dimension as the cumulative impact value. Based on the cumulative impact value, scale the initial constraint coefficients proportionally to obtain the adjusted constraint coefficients. The adjusted constraint coefficients are used to construct constraint relationships between multiple resource dimensions.

5. The method according to claim 1, characterized in that, Resource allocation instructions are generated based on coupling matching degree and preset resource constraints. Resource scheduling and allocation operations are then performed on computing nodes according to these instructions. The execution effect data includes: Obtain the capacity boundary parameters in the preset resource constraints, compare the coupling matching degree with the capacity boundary parameters to identify the over-limit resource dimensions, extract the over-limit quantity of the over-limit resource dimensions and convert it into compression weights; The compressed demand is obtained by weighting and reducing the demand of the task to be processed in each resource dimension according to the compression weight. The safety margin parameter in the preset resource constraints is obtained, and the deviation between the compressed demand and the safety margin parameter is calculated. Based on the deviation, the adaptability of allocable resources for multiple computing nodes is evaluated and a node ranking sequence is generated. The first computing node in the node ranking sequence is selected as the target computing node. The target computing node identifier is extracted and combined with the compressed demand to generate a resource allocation instruction. Based on the resource allocation instructions, resource scheduling and allocation operations are performed on the target computing node. The target computing node adjusts its resource quota according to the resource allocation instructions and starts the pending tasks. Resource consumption data and task completion time of each resource dimension are collected during the execution of the pending tasks to obtain execution effect data.

6. The method according to claim 1, characterized in that, Calculate the deviation between the execution effect data and the coupling matching degree, and judge whether the resource allocation instructions have an oscillating trend of repeated reverse adjustments based on the direction of change of the deviation within a continuous decision-making cycle, including: Extract the actual consumption of each resource dimension from the execution effect data, calculate the deviation value by the difference between the actual consumption and the expected resource occupancy corresponding to the coupling matching degree, and determine the sign of the deviation value in multiple consecutive decision cycles and record it as a deviation sign sequence. The number of symbol flips is counted by comparing adjacent elements of the biased symbol sequence, and the ratio of the number of symbol flips to the number of decision cycles is used as an indicator of oscillation frequency. The oscillation frequency index is compared with the preset oscillation threshold. When the oscillation frequency index exceeds the preset oscillation threshold, it is determined that the resource allocation instruction has an oscillation trend of repeated reverse adjustments.

7. The method according to claim 1, characterized in that, When an oscillation trend is identified, the preset resource constraints are modified to produce revised resource constraints. Based on these revised constraints, resource allocation instructions for the next decision cycle are generated, including: When it is determined that there is an oscillation trend, the deviation values ​​of each resource dimension in multiple consecutive decision cycles are extracted to form deviation time series data. The extreme values ​​of the deviation time series data are extracted to obtain the peak deviation and the valley deviation. The difference between the peak deviation and the valley deviation is calculated as the oscillation amplitude. The oscillation amplitude is multiplied by the oscillation frequency index to obtain the oscillation intensity index. The resource dimensions are sorted in descending order according to the magnitude of the oscillation intensity index to form an oscillation sensitive sequence. The resource dimensions with the highest ranking in the oscillation sensitive sequence are extracted as key control dimensions. The capacity boundary parameters corresponding to the key control dimensions in the preset resource constraints are extracted. The expansion ratio of the capacity boundary parameters is calculated according to the oscillation intensity index, and the capacity boundary parameters are amplified and adjusted to obtain the corrected resource constraints. Based on the amplified and adjusted capacity boundary parameters in the revised resource constraints, the allocatable space for each resource dimension is recalculated, and resource allocation instructions for the next decision cycle are generated based on the allocatable space.

8. A cloud computing resource dynamic allocation system based on intelligent algorithms, used to implement the method as described in any one of claims 1-7, characterized in that, include: The data acquisition unit is used to acquire resource status data of multiple computing nodes and task request data of tasks to be processed. The resource status data includes the occupied values ​​corresponding to multiple resource dimensions, and the task request data includes the demand values ​​corresponding to multiple resource dimensions. The offset compensation unit is used to predict the resource status offset between the acquisition time and the decision time based on the changing trend of resource status data during continuous acquisition time, and to compensate the resource status data using the resource status offset to obtain compensated resource status data. The coupling matching unit is used to establish constraint relationships between multiple resource dimensions. Based on the constraint relationships, it jointly calculates the occupancy value of the compensation resource status data and the demand value of the task request data to obtain the coupling matching degree. The resource allocation unit is used to generate resource allocation instructions based on the coupling matching degree and preset resource constraints, perform resource scheduling and allocation operations on computing nodes according to the resource allocation instructions, and obtain execution effect data. The oscillation judgment unit is used to calculate the deviation between the execution effect data and the coupling matching degree, and judge whether the resource allocation instruction has an oscillation trend of repeated reverse adjustment based on the direction of change of the deviation within the continuous decision cycle. The constraint correction unit is used to suppressively correct the preset resource constraints when it is determined that there is an oscillation trend, to obtain the corrected resource constraints, and to generate the resource allocation instruction for the next decision cycle based on the corrected resource constraints.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images