Adaptive load balancing method and system based on multi-dimensional intelligent perception
By using multi-dimensional intelligent perception and dynamic weight calculation, combined with historical load data prediction, an adaptive load balancing strategy is generated. This solves the problems of single indicators and prediction lag in existing technologies, realizes intelligent and efficient load balancing, and improves service stability and resource utilization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG INSPUR NEW CENTURY TECH CO LTD
- Filing Date
- 2025-12-25
- Publication Date
- 2026-05-12
AI Technical Summary
Existing load balancing methods are difficult to adapt to complex and ever-changing business scenarios. They suffer from limited metric collection dimensions, fixed weights, and lack of load prediction, leading to misjudgment of load, waste of resources, and poor service stability.
Through multi-dimensional intelligent perception, a monitoring agent is deployed to collect multi-dimensional indicator data. Dynamic weights are calculated using the entropy weight method and the analytic hierarchy process. Combined with historical load data, time-series prediction is performed to generate an adaptive load balancing strategy, accurately classify node load levels, and optimize migration operations.
It achieves intelligent, accurate, and efficient load balancing, reduces the load misjudgment rate, improves service stability and resource utilization, and avoids resource waste and network transmission costs.
Smart Images
Figure CN121411984B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of load balancing technology, specifically to an adaptive load balancing method and system based on multi-dimensional intelligent perception. Background Technology
[0002] With the rapid iteration of cloud computing, big data, and artificial intelligence technologies, the access volume of various business systems has exploded. As the core infrastructure for carrying out these businesses, the load distribution of server clusters directly determines the system's response efficiency, service stability, and resource utilization. Load balancing technology, by rationally distributing user requests or tasks to cluster nodes, has become a key supporting technology for solving single-node overload and improving the overall performance of the cluster.
[0003] However, existing load balancing methods still have many limitations in practical applications and are difficult to adapt to complex and ever-changing business scenarios. First, the metrics collected are limited in scope, leading to one-sided load assessment. Traditional methods often rely on single or a few static metrics such as CPU utilization and memory usage as the basis for load judgment, ignoring the impact of key dimensions such as disk I / O, network bandwidth, service response latency, and request success rate on load status. For example, in I / O-intensive businesses, even if CPU utilization is low, disk read / write bottlenecks can still cause servers to be under high load. A single metric assessment can easily lead to misjudgment of load, thereby causing the balancing strategy to fail.
[0004] Secondly, the fixed weights of the metrics result in poor adaptability to different scenarios. In existing technologies, the weights of various load metrics are mostly static values preset manually, making dynamic adjustments impossible based on business type and load fluctuation patterns. Furthermore, load prediction is lacking, leading to significant strategy lag. Most methods generate balancing strategies solely based on the current load state, failing to incorporate the temporal characteristics of historical load data for future load trend prediction. When faced with sudden traffic surges or load peaks, the lagging strategy adjustments cannot proactively mitigate node overload risks, easily leading to service response timeouts, request loss, and other problems, severely impacting user experience.
[0005] Furthermore, load migration lacks precision and results in significant resource waste. A "one-size-fits-all" approach is often used in load migration, failing to effectively alleviate the load on high-load nodes while wasting resources on low-load nodes. This leads to insufficient or excessive migration, increasing network transmission costs and the risk of service interruption. Summary of the Invention
[0006] In order to solve the above-mentioned technical problems, this application proposes the following technical solution:
[0007] In a first aspect, embodiments of this application provide an adaptive load balancing method based on multi-dimensional intelligent perception, including:
[0008] By deploying monitoring agents on each node of the server cluster, multi-dimensional indicator data of the load status of each node are collected periodically.
[0009] The collected multi-dimensional indicator data is preprocessed to obtain standardized state feature data;
[0010] Based on the standardized state feature data, the dynamic weights of each dimension index are determined using a preset weight calculation model.
[0011] The comprehensive load index of each node is calculated based on the dynamic weights of the aforementioned indicators, and then each node is divided into different load levels.
[0012] Based on the current load distribution and load level, and combined with the time-series prediction results of historical load data, a load balancing strategy is dynamically generated.
[0013] Based on the load balancing strategy, calculate the load to be migrated, select the optimal target node to migrate to, and execute the load migration operation.
[0014] In one possible implementation, the monitoring agent deployed on each node of the server cluster periodically collects multi-dimensional indicator data on the load status of each node, including:
[0015] The CPU utilization, memory utilization, network bandwidth utilization, disk I / O load, response time, and throughput of each node are periodically collected through the system call interface. The calculation formulas are as follows:
[0016]
[0017]
[0018]
[0019]
[0020]
[0021]
[0022] in, Let t be the total time the CPU is in a busy state. Let t be the total CPU runtime at time t, and Δt be the sampling period. Let t be the amount of memory used. Total memory size Let t be the current network traffic. This is the maximum bandwidth capacity of the network interface. Let t be the number of disk I / O operations. This represents the maximum I / O processing capacity of the disk. Let be the response time of the i-th request at time t. This represents the total number of requests within the time interval Δt.
[0023] In one possible implementation, the collected multi-dimensional indicator data is preprocessed to obtain standardized state feature data, including:
[0024] After deleting null and duplicate values generated during the data collection process, the min-max normalization method is used to map index data of different dimensions to the same numerical range.
[0025] use The principle is to detect outliers and correct them using linear interpolation.
[0026] The data is smoothed using the time window moving average method to obtain standardized state characteristic data.
[0027] In one possible implementation, based on the standardized state feature data, the calculation formula for determining the dynamic weights of each dimension index using a preset weight calculation model is as follows:
[0028]
[0029]
[0030] in, For the first The final dynamic weights of each dimension indicator The objective weights calculated using the entropy weight method. The subjective weights calculated by the analytic hierarchy process (AHP). For weighting coefficients, For the first In the first dimension, the first The weight of indicators for each node This represents the total number of cluster server nodes. This represents the total number of dimensions for the load metric.
[0031] In one possible implementation, a comprehensive load index for each node is calculated based on the dynamic weights of the aforementioned dimensional indicators, thereby classifying each node into different load levels, including:
[0032] The comprehensive load index of each node is calculated based on the dynamic weights of the aforementioned indicators. The calculation formula is as follows:
[0033]
[0034] in, For the first The final dynamic weights of each dimension indicator For the first The overall load index of each server node For the first The node at the th Standardized feature values in each dimension;
[0035] The K-means clustering algorithm is used to cluster the comprehensive load index of all nodes in the cluster, dividing the nodes into low-load nodes, medium-load nodes, and high-load nodes.
[0036] In one possible implementation, a load balancing strategy is dynamically generated based on the current load distribution and load level, combined with time-series prediction results of historical load data, including:
[0037] Predict the overall load index of nodes within a preset time window based on historical load data;
[0038] If a node is currently a high-load node and the predicted overall node load index is greater than the high-load threshold, then the load migration strategy is triggered.
[0039] If a node is currently a low-load node and the predicted overall load index of the node is less than the low-load threshold, it is marked as a candidate node for load migration.
[0040] If the current node is a medium-load node, then maintain the current load status and continuously monitor the current node.
[0041] In one possible implementation, the formula for predicting the overall node load index within a preset future time window based on historical load data is as follows:
[0042]
[0043] in, , These are the autoregressive coefficients. for The overall load index of nodes within the time window. The overall load index of nodes within the time window. , The moving average coefficient is... This is the residual term.
[0044] In one possible implementation, the formula for calculating the amount of load to be migrated is:
[0045]
[0046] in, For nodes The amount of load to be migrated For nodes The current overall load index, For high load threshold, For nodes Total load capacity.
[0047] In one possible implementation, the formula for selecting the optimal target inflow node is:
[0048]
[0049] in, Candidate migration nodes Evaluation score, For nodes The load idle time, For nodes Resource redundancy, For the migration node With candidate nodes distance, , and These are the weighting coefficients.
[0050] Secondly, embodiments of this application provide an adaptive load balancing system based on multi-dimensional intelligent perception, comprising:
[0051] The acquisition module is used to periodically collect multi-dimensional indicator data of the load status of each node through monitoring agents deployed on each node in the server cluster.
[0052] The preprocessing module is used to preprocess the collected multi-dimensional indicator data to obtain standardized state feature data;
[0053] The dynamic weight determination module is used to determine the dynamic weights of each dimension index based on the standardized state feature data and using a preset weight calculation model.
[0054] The load classification module is used to calculate the comprehensive load index of each node based on the dynamic weights of the indicators of each dimension, and then classify each node into different load levels.
[0055] The strategy generation module is used to dynamically generate load balancing strategies based on the current load distribution and load level, combined with the time-series prediction results of historical load data.
[0056] The migration operation module is used to calculate the load to be migrated according to the load balancing strategy, select the optimal target migration node, and execute the load migration operation.
[0057] Compared with the prior art, the beneficial effects of this application are as follows:
[0058] This application achieves intelligent, accurate, and efficient load balancing through core designs combining multi-dimensional data collection, dynamic weight calculation, and time-series prediction. By deploying monitoring agents on each node, it constructs a multi-dimensional indicator system covering computation, memory, storage, network, and application dimensions, enabling comprehensive perception of server load status. Compared to traditional single-indicator evaluation, this method avoids one-sided interpretations of load status, accurately captures load bottlenecks in different business scenarios, provides reliable data support for subsequent balancing strategy formulation, and effectively reduces the load misjudgment rate.
[0059] Secondly, this application employs a combined weighting model that integrates entropy weighting and analytic hierarchy process (AHP) to calculate the dynamic weights of indicators across various dimensions. Specifically, entropy weighting mines objective weights based on the dispersion of indicator data, reflecting the inherent value of the data; AHP combines domain expert experience to determine subjective weights, adapting to business scenario requirements. By integrating these two types of weights through weight allocation coefficients, the rigidity of static weights is avoided, while also balancing data objectivity with business subjectivity. This makes load assessment more aligned with actual needs and enhances the scenario adaptability of the load balancing strategy.
[0060] Furthermore, this application predicts node load trends within future time windows based on historical load data and generates a load balancing strategy by combining the current load level with the prediction results. When a node is predicted to enter a high-load state in the future, a load migration operation is triggered in advance; for nodes with consistently low load in the future, they are marked as candidate nodes for migration in advance, breaking the lag of the traditional method's "post-event adjustment". This forward-looking strategy formulation can effectively avoid the risk of node overload caused by sudden loads and ensure the continuity and stability of services. Attached Figure Description
[0061] Figure 1 A flowchart illustrating an adaptive load balancing method based on multi-dimensional intelligent perception provided in an embodiment of this application;
[0062] Figure 2 This is a schematic diagram of an adaptive load balancing system based on multi-dimensional intelligent perception, provided as an embodiment of this application. Detailed Implementation
[0063] The present solution will now be described in conjunction with the accompanying drawings and specific embodiments.
[0064] Figure 1 A flowchart illustrating an adaptive load balancing method based on multi-dimensional intelligent perception, provided in an embodiment of this application, is shown below. Figure 1 An adaptive load balancing method based on multi-dimensional intelligent perception in this embodiment includes:
[0065] S101 periodically collects multi-dimensional indicator data on the load status of each node through monitoring agents deployed on each node in the server cluster.
[0066] In this embodiment, the CPU utilization, memory utilization, network bandwidth utilization, disk I / O load, response time, and throughput of each node are periodically collected through the system call interface. The calculation formulas are as follows:
[0067]
[0068]
[0069]
[0070]
[0071]
[0072]
[0073] in, Let t be the total time the CPU is in a busy state. Let t be the total CPU runtime at time t, and Δt be the sampling period. Let t be the amount of memory used. Total memory size Let t be the current network traffic. This is the maximum bandwidth capacity of the network interface. Let t be the number of disk I / O operations. This represents the maximum I / O processing capacity of the disk. Let be the response time of the i-th request at time t. This represents the total number of requests within the time interval Δt.
[0074] S102 preprocesses the collected multi-dimensional indicator data to obtain standardized state feature data.
[0075] In this embodiment, after deleting null and duplicate values generated during the data collection process, min-max standardization is used to eliminate dimensional differences, mapping the index values of each dimension to the [0,1] interval. The standardization formula is as follows:
[0076]
[0077] in, For the first The node at the th Standardized feature values of each dimension For the first The node The original values of each dimension, For all nodes in the cluster The minimum value of each dimension indicator. For the i-th node in the cluster The maximum value of the dimensional indicator.
[0078] The extreme value normalization method is used to map index data of different dimensions to the same numerical range;
[0079] use The principle for detecting outliers is to first obtain the original data sequence of a certain dimension indicator, and then calculate the mean and standard deviation of the sequence. ,but These are outliers, where... The mean of the sequence. The standard deviation of the sequence is used to correct outliers using linear interpolation, and the data is smoothed using a time window moving average method to obtain standardized state characteristic data.
[0080] S103, based on standardized state feature data, uses a preset weight calculation model to determine the dynamic weights of each dimension indicator.
[0081] In this embodiment, considering the differences in the importance of various indicators under different load scenarios, a dynamic weighting model is used to calculate the weight values, avoiding the limitations of static weights. A model combining entropy weighting and analytic hierarchy process is employed, balancing data objectivity with the subjectivity of business experience. The calculation formula is as follows:
[0082]
[0083]
[0084] in, For the first The final dynamic weights of each dimension indicator The objective weights calculated using the entropy weight method. The subjective weights calculated by the analytic hierarchy process (AHP). These are the weighting coefficients, which can be adjusted according to the actual scenario. For the first In the first dimension, the first The weight of indicators for each node This represents the total number of cluster server nodes. This represents the total number of dimensions for the load metric.
[0085] When calculating subjective weights using the Analytic Hierarchy Process (AHP), after determining the subjective weights of load metrics such as CPU utilization, memory utilization, network bandwidth utilization, disk I / O load, response time, and throughput, pairwise importance comparisons are performed on these load metrics. By inviting experts in server operation and maintenance and load balancing to conduct pairwise comparisons of the above load metrics based on the 1-9 scale, the impact of load metrics under different business scenarios is assessed, thus constructing a... Order-order judgment matrix, ,in, As an indicator Relative indicators The importance of CPU utilization is determined by factors such as the computational intensity of the business. For example, in computationally intensive business scenarios, experts would determine that CPU utilization is significantly more important than memory utilization.
[0086] Because expert judgment is subjective, consistency checks must be performed to avoid logical contradictions. The check process in this embodiment includes:
[0087] First, calculate the product of each row of the judgment matrix and the initial normalized weight. The calculation formula is as follows:
[0088]
[0089]
[0090] in, This is the initial weight vector for normalization. For the first The product of row elements. is the order of the matrix.
[0091] Then, the consistency index and consistency ratio are calculated using the following formulas:
[0092]
[0093]
[0094] in, The average random consistency index, As a consistency indicator, This represents the consistency ratio.
[0095] like If the value is less than 0.1, then the matrix is considered to meet the consistency requirement, and the weights can continue to be calculated. If the value is greater than or equal to 0.1, the judgment matrix is considered to have a logical contradiction, and experts need to be invited to readjust the scale value until consistency is achieved. Once the judgment matrix passes the consistency check, it is row-normalized to obtain... Furthermore, it satisfies the weight normalization constraint, ensuring that the sum of the weights of all dimensions is 1.
[0096] S104 calculates the comprehensive load index of each node based on the dynamic weights of various indicators, and then classifies each node into different load levels.
[0097] In this embodiment, the comprehensive load index of each node is calculated based on the dynamic weights of each dimension indicator. The calculation formula is as follows:
[0098]
[0099] in, For the first The final dynamic weights of each dimension indicator For the first The overall load index of each server node; a higher value indicates a higher load. For the first The node The standardized feature values of each dimension are used to cluster the comprehensive load index of all nodes in the cluster using the K-means clustering algorithm, dividing the nodes into low-load nodes, medium-load nodes, and high-load nodes.
[0100] In this embodiment, when the first When the overall load index of a server node is less than or equal to the low load threshold, it is determined to be a low load node. When the overall load index of a server node is less than or equal to the high load threshold and greater than the low load threshold, it is determined to be a medium load node. A node is considered a high-load node when its overall load index exceeds the high-load threshold.
[0101] S105 dynamically generates a load balancing strategy based on the current load distribution and load level, combined with the time-series prediction results of historical load data.
[0102] In this embodiment, the comprehensive load index of nodes within a future preset time window is predicted based on historical load data. The calculation formula is as follows:
[0103]
[0104] in, , These are the autoregressive coefficients. for The overall load index of nodes within the time window. The overall load index of nodes within the time window. , The moving average coefficient is... For residual terms;
[0105] If a node is currently a high-load node and the predicted overall node load index is greater than the high-load threshold, the load migration strategy is triggered. If a node is currently a low-load node and the predicted overall node load index is less than the low-load threshold, it is marked as a candidate node for load migration. If the current node is a medium-load node, the current load status is maintained and the current node is continuously monitored.
[0106] S106. Based on the load balancing strategy, calculate the load to be migrated, select the optimal target node to migrate to, and execute the load migration operation.
[0107] To avoid under- or over-migration and to reduce network costs and service interruption probability during load migration, this embodiment classifies node load levels based on a comprehensive load index. Based on the goal of "reducing high-load nodes below the medium-load threshold," the amount of load to be migrated is accurately calculated. Simultaneously, a target node evaluation function is constructed, incorporating load idleness, resource redundancy, and network distance, to select the optimal migration node. The formula for calculating the amount of load to be migrated is:
[0108]
[0109] in, For nodes The amount of load to be migrated For nodes The current overall load index, For high load threshold, For nodes Total load capacity.
[0110] The formula for selecting the optimal target inflow node is as follows:
[0111]
[0112] in, Candidate migration nodes Evaluation score, For nodes The load idle time, For nodes Resource redundancy, For the migration node With candidate nodes distance, , and These are the weighting coefficients.
[0113] This embodiment achieves a dual improvement in service quality and resource utilization through precise load allocation. On the one hand, it effectively alleviates the pressure on high-load nodes and improves service response speed. On the other hand, it makes full use of the idle resources of low-load nodes, reducing the resource waste rate of the cluster.
[0114] In this embodiment, during the migration operation, target nodes are selected from high to low based on evaluation scores, and the load to be migrated is allocated. The load balancing control node sends migration instructions to both the outgoing node and the target incoming node. The outgoing node transfers the specified load to the target node, and service availability is monitored during the migration process. After the migration is complete, the cluster load balancing degree is calculated to verify whether the balancing target has been achieved. The formula for calculating the cluster load balancing degree is:
[0115]
[0116] in, For the first The overall load index of each server node This represents the average load index of the cluster. The cluster load balancing score is used. If the cluster load balancing score is higher than a preset threshold, the weights and clustering parameters are readjusted to achieve iterative optimization. In this embodiment, after the load migration is completed, the balancing effect is evaluated by calculating the cluster load balancing score. If the balancing score does not reach the preset threshold, the weight model parameters and load level classification threshold are adjusted in reverse, forming a closed-loop iterative mechanism of collection, evaluation, strategy, migration, feedback, and optimization. This self-optimization capability enables the method to adapt to dynamic changes such as cluster size expansion and business type changes, maintaining excellent load balancing performance in the long term and improving the robustness and lifecycle value of the method.
[0117] Corresponding to the adaptive load balancing method based on multi-dimensional intelligent perception provided in the above embodiments, this application also provides an embodiment of an adaptive load balancing system based on multi-dimensional intelligent perception.
[0118] See Figure 2 This application provides an adaptive load balancing system 20 based on multi-dimensional intelligent perception, comprising:
[0119] The acquisition module 201 is used to periodically collect multi-dimensional indicator data of the load status of each node through the monitoring agent deployed on each node in the server cluster.
[0120] The preprocessing module 202 is used to preprocess the collected multi-dimensional indicator data to obtain standardized state feature data.
[0121] The dynamic weight determination module 203 is used to determine the dynamic weights of each dimension index based on standardized state feature data and using a preset weight calculation model.
[0122] The load classification module 204 is used to calculate the comprehensive load index of each node based on the dynamic weights of indicators in each dimension, and then classify each node into different load levels.
[0123] The strategy generation module 205 is used to dynamically generate load balancing strategies based on the current load distribution and load level, combined with the time-series prediction results of historical load data.
[0124] The migration operation module 206 is used to calculate the load to be migrated based on the load balancing strategy, select the optimal target migration node, and execute the load migration operation.
[0125] In this embodiment, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, or the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.
[0126] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0127] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. The protection scope of this application should be determined by the protection scope of the claims.
Claims
1. An adaptive load balancing method based on multi-dimensional intelligent perception, characterized in that, include: By deploying monitoring agents on each node in the server cluster, multi-dimensional metrics of the load status of each node are periodically collected, including: periodically collecting CPU utilization, memory utilization, network bandwidth utilization, disk I / O load, response time and throughput of each node through system call interfaces. The collected multi-dimensional indicator data is preprocessed to obtain standardized state feature data; Based on the standardized state feature data, the dynamic weights of each dimension index are determined using a preset weight calculation model. The calculation formula is as follows: in, For the first The final dynamic weights of each dimension indicator The objective weights calculated using the entropy weight method. The subjective weights calculated by the analytic hierarchy process (AHP). For weighting coefficients, For the first In the first dimension, the first The weight of indicators for each node This represents the total number of cluster server nodes. This represents the total number of dimensions for the load metric. The comprehensive load index of each node is calculated based on the dynamic weights of the aforementioned indicators, and then each node is divided into different load levels, including: The comprehensive load index of each node is calculated based on the dynamic weights of the aforementioned indicators. The calculation formula is as follows: in, For the first The final dynamic weights of each dimension indicator For the first The overall load index of each server node For the first The node at the th Standardized feature values in each dimension; The K-means clustering algorithm is used to cluster the comprehensive load index of all nodes in the cluster, dividing the nodes into low-load nodes, medium-load nodes, and high-load nodes; Based on the current load distribution and load level, and combined with the time-series prediction results of historical load data, a load balancing strategy is dynamically generated, including: Predict the overall load index of nodes within a preset time window based on historical load data; If a node is currently a high-load node and the predicted overall node load index is greater than the high-load threshold, then the load migration strategy is triggered. If a node is currently a low-load node and the predicted overall load index of the node is less than the low-load threshold, it is marked as a candidate node for load migration. If the current node is a medium-load node, then maintain the current load status and continuously monitor the current node; Based on the load balancing strategy, calculate the load to be migrated, select the optimal target node to migrate to, and execute the load migration operation.
2. The adaptive load balancing method based on multi-dimensional intelligent perception according to claim 1, characterized in that, The CPU utilization, memory utilization, network bandwidth utilization, disk I / O load, response time, and throughput of each node are periodically collected through the system call interface, and the calculation formulas are as follows: in, Let t be the total time the CPU is in a busy state. Let t be the total CPU runtime at time t, and Δt be the sampling period. Let t be the amount of memory used. Total memory size Let t be the current network traffic. This is the maximum bandwidth capacity of the network interface. Let t be the number of disk I / O operations. This represents the maximum I / O processing capacity of the disk. Let be the response time of the i-th request at time t. This represents the total number of requests within the time interval Δt.
3. The adaptive load balancing method based on multi-dimensional intelligent perception according to claim 1, characterized in that, The collected multi-dimensional indicator data is preprocessed to obtain standardized state feature data, including: After deleting null and duplicate values generated during the data collection process, the min-max normalization method is used to map index data of different dimensions to the same numerical range. use The principle is to detect outliers and correct them using linear interpolation. The data is smoothed using the time window moving average method to obtain standardized state characteristic data.
4. The adaptive load balancing method based on multi-dimensional intelligent perception according to claim 1, characterized in that, The formula for predicting the comprehensive load index of nodes within a preset time window based on historical load data is as follows: in, , These are the autoregressive coefficients. for The overall load index of nodes within the time window. The overall load index of nodes within the time window. , The moving average coefficient is... This is the residual term.
5. The adaptive load balancing method based on multi-dimensional intelligent perception according to claim 1, characterized in that, The formula for calculating the load to be migrated is: in, For nodes The amount of load to be migrated For nodes The current overall load index, For high load threshold, For nodes Total load capacity.
6. The adaptive load balancing method based on multi-dimensional intelligent perception according to claim 1, characterized in that, The formula for selecting the optimal target inflow node is as follows: in, Candidate migration nodes Evaluation score, For nodes The load idle time, For nodes Resource redundancy, For the migration node With candidate nodes distance, , and These are the weighting coefficients.
7. An adaptive load balancing system based on multi-dimensional intelligent perception, characterized in that, include: The acquisition module is used to periodically collect multi-dimensional indicator data of the load status of each node through monitoring agents deployed on each node in the server cluster. This includes periodically collecting CPU utilization, memory utilization, network bandwidth utilization, disk I / O load, response time and throughput of each node through system call interfaces. The preprocessing module is used to preprocess the collected multi-dimensional indicator data to obtain standardized state feature data; The dynamic weight determination module is used to determine the dynamic weights of each dimension indicator based on the standardized state feature data and using a preset weight calculation model. The calculation formula is as follows: in, For the first The final dynamic weights of each dimension indicator The objective weights calculated using the entropy weight method. The subjective weights calculated by the analytic hierarchy process (AHP). For weighting coefficients, For the first In the first dimension, the first The weight of indicators for each node This represents the total number of cluster server nodes. This represents the total number of dimensions for the load metric. The load classification module is used to calculate the comprehensive load index of each node based on the dynamic weights of the aforementioned indicators, and then classify each node into different load levels, including: The comprehensive load index of each node is calculated based on the dynamic weights of the aforementioned indicators. The calculation formula is as follows: in, For the first The final dynamic weights of each dimension indicator For the first The overall load index of each server node For the first The node at the th Standardized feature values in each dimension; The K-means clustering algorithm is used to cluster the comprehensive load index of all nodes in the cluster, dividing the nodes into low-load nodes, medium-load nodes, and high-load nodes; The strategy generation module is used to dynamically generate load balancing strategies based on the current load distribution and load level, combined with time-series prediction results of historical load data, including: Predict the overall load index of nodes within a preset time window based on historical load data; If a node is currently a high-load node and the predicted overall node load index is greater than the high-load threshold, then the load migration strategy is triggered. If a node is currently a low-load node and the predicted overall load index of the node is less than the low-load threshold, it is marked as a candidate node for load migration. If the current node is a medium-load node, then maintain the current load status and continuously monitor the current node; The migration operation module is used to calculate the load to be migrated according to the load balancing strategy, select the optimal target migration node, and execute the load migration operation.