Dynamic load balancing system and method for multi-cluster intelligent expansion

Abstract

The present invention provides a dynamic load balancing system and method for intelligent expansion of multiple clusters, which relates to the field of dynamic load balancing technology and includes: a load feature acquisition module; a load saturation prediction module for performing load saturation prediction to obtain multiple predicted load saturations; an expansion instruction parsing module for inputting a cluster expansion determiner to obtain a cluster expansion instruction; a global expansion module for activating a global expansion parsing channel for global expansion when a global expansion instruction is given, thereby obtaining a first multi-cluster optimized architecture; a local expansion module for activating a local expansion decision model for local expansion when a local expansion instruction is given, thereby obtaining a second multi-cluster optimized architecture; and a load balancing control module for load balancing control. The present invention solves the technical problem in the prior art that multiple clusters cannot be adaptively expanded according to load status, resulting in unbalanced load distribution and delayed service response.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic load balancing, and in particular to a dynamic load balancing system and method for multi-cluster intelligent expansion. Background Art

[0002] In a multi-cluster environment, load distribution is often uneven. Some clusters may be overloaded while other clusters' resources remain idle. This leads to wasted resources and delayed responses for some services, which in turn reduces overall system performance and impacts user experience. However, dynamic load balancing and capacity expansion decisions involve multiple factors, including current load, future load forecasts, and resource constraints. Accurately assessing these factors and making quick decisions is a major challenge. Furthermore, business demands can change very rapidly, requiring rapid detection and response to load changes. Traditional load balancing methods often lack sufficient real-time performance and are unable to respond to sudden traffic bursts, resulting in a poor user experience. Summary of the Invention

[0003] This application provides a dynamic load balancing system and method for intelligent expansion of multiple clusters, aiming to solve the technical problems in the existing technology that are unable to adaptively expand multiple clusters according to load status, resulting in uneven load distribution and service response delays.

[0004] The first aspect disclosed in the present application provides a dynamic load balancing system for multi-cluster intelligent expansion, the system comprising: a load feature acquisition module, the load feature acquisition module being used to collect load features according to a multi-cluster architecture to obtain a load feature data set, wherein the multi-cluster architecture comprises multiple clusters, each cluster comprising a number of computing nodes; a load saturation prediction module, the load saturation prediction module being used to predict load saturation for the multiple clusters according to the load feature data set based on a predetermined future time zone window to obtain multiple predicted load saturations; an expansion instruction parsing module, the expansion instruction parsing module being used to input the multiple predicted load saturations into a cluster expansion determiner to obtain a cluster expansion instruction, wherein the The cluster expansion instruction is a global expansion instruction or a local expansion instruction; the global expansion module is used to activate the global expansion resolution channel when the cluster expansion instruction is a global expansion instruction, and globally expand the multi-cluster architecture according to the global expansion resolution channel to obtain a first multi-cluster optimized architecture; the local expansion module is used to activate the local expansion decision model when the cluster expansion instruction is a local expansion instruction, and locally expand the multi-cluster architecture according to the local expansion decision model to obtain a second multi-cluster optimized architecture; the load balancing control module is used to perform load balancing control according to the first multi-cluster optimized architecture or the second multi-cluster optimized architecture.

[0005] The second aspect disclosed in the present application provides a dynamic load balancing method for multi-cluster intelligent expansion, which is implemented by the above-mentioned dynamic load balancing system for multi-cluster intelligent expansion. The method includes: collecting load characteristics according to a multi-cluster architecture to obtain a load characteristic data set, wherein the multi-cluster architecture includes multiple clusters, each cluster includes a number of computing nodes; based on a predetermined future time zone window, predicting the load saturation of the multiple clusters according to the load characteristic data set to obtain multiple predicted load saturations; inputting the multiple predicted load saturations into a cluster expansion determiner to obtain a cluster expansion determiner. capacity instruction, wherein the cluster capacity expansion instruction is a global capacity expansion instruction or a local capacity expansion instruction; when the cluster capacity expansion instruction is a global capacity expansion instruction, a global capacity expansion resolution channel is activated, and the multi-cluster architecture is globally expanded according to the global capacity expansion resolution channel to obtain a multi-cluster optimized first architecture; when the cluster capacity expansion instruction is a local capacity expansion instruction, a local capacity expansion decision model is activated, and the multi-cluster architecture is locally expanded according to the local capacity expansion decision model to obtain a multi-cluster optimized second architecture; load balancing management and control is performed according to the multi-cluster optimized first architecture or the multi-cluster optimized second architecture.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] By collecting the load characteristics of the multi-cluster architecture in real time, the current load status of each cluster and computing node can be grasped. This real-time monitoring can ensure timely response to load changes, and lay a data foundation for subsequent load forecasting and expansion decisions, thereby improving the accuracy of load management. Based on the load characteristic data set, the load saturation of each cluster in the predetermined future time zone window is predicted, and multiple predicted load saturations are obtained, which improves the responsiveness of the system and enables timely expansion decisions under high load conditions to avoid performance degradation or service interruption caused by sudden load increases. Based on the predicted results of load saturation, the cluster expansion determiner is used to decide whether to perform global expansion. Whether it is local expansion or global expansion, such flexibility allows the system to make appropriate resource adjustments based on actual load requirements; during global expansion, the resource capacity of multiple clusters is comprehensively improved to form a multi-cluster optimization first architecture; during local expansion, the focus is on solving the load problem of a specific cluster to form a multi-cluster optimization second architecture. This dual expansion mechanism improves the adaptability of the system; after the expansion is completed, load balancing management and control are carried out according to the new optimization architecture to ensure that the load of all clusters is reasonably distributed to avoid overload or idleness of a cluster. Through dynamic load balancing, the overall performance and resource utilization can be improved to ensure that each cluster can maintain stable service quality during peak load periods.

[0008] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 This is a schematic diagram of the structure of a dynamic load balancing system with multi-cluster intelligent expansion provided by an embodiment of the present application.

[0010] Figure 2 A flow chart of a dynamic load balancing method for multi-cluster intelligent expansion provided in an embodiment of the present application.

[0011] Explanation of the accompanying symbols: load feature collection module 10, load saturation prediction module 20, expansion instruction parsing module 30, global expansion module 40, local expansion module 50, load balancing control module 60. DETAILED DESCRIPTION

[0012] The embodiments of the present application provide a dynamic load balancing system and method for intelligent multi-cluster expansion, thereby solving the technical problem in the prior art of being unable to adaptively expand multiple clusters according to load status, resulting in unbalanced load distribution and service response delays.

[0013] After introducing the basic principles of this application, various non-limiting embodiments of this application will be specifically described below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described here are only used to explain this application and are not used to limit this application.

[0014] Example 1, as Figure 1 As shown, an embodiment of the present application provides a dynamic load balancing system with multi-cluster intelligent expansion, the system comprising:

[0015] A load feature acquisition module 10 is configured to acquire load features based on a multi-cluster architecture to obtain a load feature data set, wherein the multi-cluster architecture includes multiple clusters, and each cluster includes a number of computing nodes; a load saturation prediction module 20 is configured to predict load saturation for the multiple clusters based on the load feature data set based on a predetermined future time zone window to obtain multiple predicted load saturations; an expansion instruction parsing module 30 is configured to input the multiple predicted load saturations into a cluster expansion determiner to obtain a cluster expansion instruction, wherein the cluster expansion instruction is a global expansion instruction or a local expansion instruction. Command; a global expansion module 40, the global expansion module 40 is used to activate the global expansion resolution channel when the cluster expansion instruction is a global expansion instruction, and globally expand the multi-cluster architecture according to the global expansion resolution channel to obtain a multi-cluster optimized first architecture; a local expansion module 50, the local expansion module 50 is used to activate the local expansion decision model when the cluster expansion instruction is a local expansion instruction, and locally expand the multi-cluster architecture according to the local expansion decision model to obtain a multi-cluster optimized second architecture; a load balancing control module 60, the load balancing control module 60 is used to perform load balancing control according to the multi-cluster optimized first architecture or the multi-cluster optimized second architecture.

[0016] Furthermore, the load saturation prediction module is configured to perform load saturation prediction on the plurality of clusters based on the load characteristic dataset in a predetermined future time zone window to obtain a plurality of predicted load saturations, including:

[0017] Based on the load feature data set, feature identification is performed to obtain multiple current load amounts and multiple constrained load amounts corresponding to the multiple clusters; based on the predetermined future time zone window, load change trends of the multiple clusters are predicted according to the multiple current load amounts to obtain multiple predicted load amounts; based on the multiple predicted load amounts and the multiple constrained load amounts, the multiple predicted load saturations are calculated.

[0018] Furthermore, the expansion instruction parsing module is configured to input the multiple predicted load saturations into the cluster expansion determiner to obtain a cluster expansion instruction, wherein the cluster expansion instruction is a global expansion instruction or a local expansion instruction, including:

[0019] Activate a local expansion determiner within the cluster expansion determiner; input the multiple predicted load saturations into the local expansion determiner, wherein the local expansion determiner includes a load saturation threshold; determine whether the multiple predicted load saturations are greater than the load saturation threshold; if any one of the multiple predicted load saturations is greater than the load saturation threshold, the local expansion determiner outputs the local expansion instruction.

[0020] Furthermore, the local capacity expansion determiner outputs the local capacity expansion instruction, including:

[0021] The predicted load saturation that is greater than the load saturation threshold is recorded as abnormal load saturation; the abnormal load saturation is calculated with the multiple predicted load saturations to obtain the abnormal saturation quantity ratio; it is determined whether the abnormal saturation quantity ratio is greater than a predetermined proportion threshold; if the abnormal saturation quantity ratio is greater than the predetermined proportion threshold, the global expansion determiner in the cluster expansion determiner is activated; the centralized value of the multiple predicted load saturations is calculated to obtain the global load saturation, and the global load saturation is input into the global expansion determiner, wherein the global expansion determiner includes a global load saturation threshold; it is determined whether the global load saturation is greater than the global load saturation threshold; if the global load saturation is greater than the global load saturation threshold, the global expansion determiner outputs the global expansion instruction and prohibits the local expansion instruction.

[0022] Furthermore, the global expansion module is configured to activate a global expansion resolution channel when the cluster expansion instruction is a global expansion instruction, and perform global expansion on the multi-cluster architecture according to the global expansion resolution channel to obtain a multi-cluster optimized first architecture, including:

[0023] A global expansion feature analysis is performed based on the multi-cluster architecture to obtain a global expansion capacity; the global expansion analysis channel includes a global expansion decision model and a global expansion fitness analysis model; based on the global expansion capacity, multiple global expansion decisions are obtained according to the global expansion decision model; based on the global expansion fitness analysis model, the multiple global expansion decisions are iteratively optimized to obtain a global expansion strategy; the multi-cluster architecture is globally expanded according to the global expansion strategy to obtain the multi-cluster optimized first architecture.

[0024] Furthermore, the global expansion fitness analytical model includes a global expansion fitness analytical function, and the global expansion fitness analytical function is:

[0025]

[0026] Among them, GSA represents the global expansion fitness, exp represents the exponential function with the natural constant e as the base, csw and ccw represent the predetermined fitness analysis weight conditions, scx represents the predicted global load balancing degree, sxo represents the expected global load balancing degree, scc represents the predicted global expansion cost, and sco represents the expected global expansion cost.

[0027] Furthermore, the local expansion module is configured to activate a local expansion decision model when the cluster expansion instruction is a local expansion instruction, and perform local expansion on the multi-cluster architecture according to the local expansion decision model to obtain a multi-cluster optimized second architecture, including:

[0028] Perform local expansion feature analysis based on the multi-cluster architecture to determine a local expansion cluster and a local expansion capacity; collect configuration parameters of the local expansion cluster to obtain cluster configuration feature data; input the cluster configuration feature data and the local expansion capacity into the local expansion decision model to obtain a local expansion decision; and based on the multi-cluster architecture, expand the local expansion cluster according to the local expansion decision to obtain the multi-cluster optimized second architecture.

[0029] The following detailed description of the dynamic load balancing method for multi-cluster intelligent expansion will provide those skilled in the art with a clear understanding of the dynamic load balancing system for multi-cluster intelligent expansion in this embodiment. Since this system corresponds to the method disclosed in the embodiment, its description is relatively simple. For relevant details, please refer to the description of the method.

[0030] Example 2 is based on the same inventive concept as the multi-cluster intelligent expansion dynamic load balancing system in the previous embodiment. Figure 2 As shown, the embodiment of the present application provides a dynamic load balancing method for multi-cluster intelligent expansion, the method comprising:

[0031] Load characteristics are collected according to a multi-cluster architecture to obtain a load characteristic data set, wherein the multi-cluster architecture includes multiple clusters, and each cluster includes a number of computing nodes.

[0032] The multi-cluster architecture includes multiple clusters, each of which is composed of several computing nodes. These clusters can be distributed in different physical or virtual environments and collaborate with each other to handle the overall load of the application. Different clusters have different hardware configurations, resource limitations, and load characteristics. The computing nodes in each cluster can be physical servers, virtual machines, containers, etc. They perform specific computing tasks, and the computing nodes communicate and coordinate with each other through the network.

[0033] Load characteristics are key performance indicators during system operation, including but not limited to CPU utilization, memory utilization, disk I / O, network bandwidth, and process queue length. Each cluster and node exhibits different load characteristics at different points in time. The monitoring system automatically collects cluster and node load data. The specific sampling frequency should be determined based on the actual needs and real-time requirements of the system. For multi-cluster architectures, load data from different clusters is aggregated and integrated to obtain a load characteristic dataset. This load characteristic data is used for subsequent load forecasting and capacity expansion decisions.

[0034] Based on a predetermined future time zone window, load saturation prediction is performed on the multiple clusters according to the load characteristic dataset to obtain multiple predicted load saturations.

[0035] The scheduled future time zone window is a time period prediction window used to estimate load changes within a certain period of time. This window can be flexibly set according to business needs, such as 5 minutes, 30 minutes, or 1 hour. The choice of prediction window affects the real-time and accuracy of expansion decisions. A window that is too large may lead to response delays, while a window that is too small may lead to too frequent expansion.

[0036] Load saturation is the ratio of system load to resource limits, expressed as a percentage. For example, when CPU utilization exceeds 80%, the system may be saturated, meaning the load is approaching the upper limit of the system's capacity. The current load (such as CPU utilization and memory usage) and constraints (such as node resource limits) are extracted from the load feature dataset. Based on historical data, the load trend within a predetermined future time window is predicted to obtain a predicted load. This predicted load is compared with the node resource limit to calculate the future load saturation. Multiple predicted load saturation levels are generated for multiple clusters.

[0037] Furthermore, the performing load saturation prediction on the plurality of clusters according to the load characteristic dataset based on the predetermined future time zone window to obtain a plurality of predicted load saturations includes:

[0038] Based on the load feature data set, feature identification is performed to obtain multiple current load amounts and multiple constrained load amounts corresponding to the multiple clusters; based on the predetermined future time zone window, load change trends of the multiple clusters are predicted according to the multiple current load amounts to obtain multiple predicted load amounts; based on the multiple predicted load amounts and the multiple constrained load amounts, the multiple predicted load saturations are calculated.

[0039] The current load and constrained load of each cluster are extracted from the load feature dataset. The current load reflects the cluster's resource usage at the current moment. Common loads include CPU usage, memory usage, and network usage. The constrained load refers to the maximum load capacity that a cluster or node can carry. For example, the constrained load of CPU usage is the total number of CPU cores, and the constrained load of memory is the total memory capacity. Each cluster has its own resource limitations, so the constrained loads of different clusters may be different.

[0040] In order to predict future load change trends, various methods can be used, such as time series analysis, regression models, machine learning models, etc. For example, a linear regression model can be used to perform linear fitting predictions based on the changing trends of historical load data. By analyzing the historical data and trends of the current load, the load of each cluster in the predetermined future time zone window can be predicted to obtain multiple predicted loads.

[0041] The predicted load saturation indicates the load occupancy of a cluster in a future time zone and reflects the intensity of the cluster load. It is calculated by dividing the predicted load by the constrained load. For each cluster, the predicted load is compared with the constrained load of the cluster to calculate the predicted load saturation. If the predicted load saturation of a cluster reaches the preset threshold, it means that the cluster's resources are close to saturation, and capacity expansion or other load management measures need to be considered.

[0042] The multiple predicted load saturations are input into a cluster expansion determiner to obtain a cluster expansion instruction, wherein the cluster expansion instruction is a global expansion instruction or a local expansion instruction.

[0043] The cluster expansion determiner includes a local expansion determiner and a global expansion determiner, which are used to determine whether to perform an expansion operation and the type of expansion, including local expansion or global expansion. The local expansion determiner is used to determine whether the load of certain clusters has reached a state requiring local expansion; when the load saturation of multiple clusters exceeds the threshold at the same time, the global expansion determiner is activated to determine whether the entire multi-cluster architecture needs to be globally expanded.

[0044] Based on the output of the cluster expansion determinator, if the global load saturation exceeds the global threshold, a global expansion command is issued, local expansion commands are disabled, and the entire multi-cluster architecture is expanded to balance the overall load. If the load of only a single cluster exceeds the threshold and the conditions for global expansion are not met, a local expansion command is issued to expand only that specific cluster. This decision-making mechanism ensures that the system's resources can be dynamically expanded to cope with load changes.

[0045] Furthermore, the inputting the multiple predicted load saturations into the cluster expansion determiner to obtain a cluster expansion instruction, wherein the cluster expansion instruction is a global expansion instruction or a local expansion instruction, includes:

[0046] Activate a local expansion determiner within the cluster expansion determiner; input the multiple predicted load saturations into the local expansion determiner, wherein the local expansion determiner includes a load saturation threshold; determine whether the multiple predicted load saturations are greater than the load saturation threshold; if any one of the multiple predicted load saturations is greater than the load saturation threshold, the local expansion determiner outputs the local expansion instruction.

[0047] Activate the local expansion determiner. The local expansion determiner is part of the cluster expansion determiner. When making a cluster expansion determination, the local expansion determiner first analyzes the load data of each cluster and prepares to make a judgment on the local expansion demand.

[0048] The multiple predicted load saturations obtained are input into the local expansion determiner. The local expansion determiner includes a load saturation threshold, which is a pre-set key parameter representing the maximum load that the cluster can withstand. For example, a load saturation threshold can be set to 80%. That is, when the load saturation exceeds 80%, the cluster's resources begin to approach saturation.

[0049] The system checks the predicted load saturation of each cluster one by one to determine whether it exceeds the set load saturation threshold. For example, if the predicted load saturation of a cluster is 85% and the threshold is 80%, the cluster is considered to be overloaded. If the predicted load saturation is greater than the threshold, the cluster is considered to be saturated soon and capacity expansion is necessary to ensure stable system operation. If the predicted load saturation is less than the threshold, the cluster is considered to be within the acceptable range and capacity expansion is not required for the time being.

[0050] When the predicted load saturation of any cluster is greater than the load saturation threshold, the local expansion determiner will make a local expansion decision and issue a local expansion instruction to ensure timely resource expansion of the corresponding cluster.

[0051] Furthermore, the local capacity expansion determiner outputs the local capacity expansion instruction, including:

[0052] The predicted load saturation that is greater than the load saturation threshold is recorded as abnormal load saturation; the abnormal load saturation is calculated with the multiple predicted load saturations to obtain the abnormal saturation quantity ratio; it is determined whether the abnormal saturation quantity ratio is greater than a predetermined proportion threshold; if the abnormal saturation quantity ratio is greater than the predetermined proportion threshold, the global expansion determiner in the cluster expansion determiner is activated; the centralized value of the multiple predicted load saturations is calculated to obtain the global load saturation, and the global load saturation is input into the global expansion determiner, wherein the global expansion determiner includes a global load saturation threshold; it is determined whether the global load saturation is greater than the global load saturation threshold; if the global load saturation is greater than the global load saturation threshold, the global expansion determiner outputs the global expansion instruction and prohibits the local expansion instruction.

[0053] All predicted load saturations greater than the preset load saturation threshold are defined as abnormal load saturations. These load saturations reflect situations where cluster loads exceed the safe range, meaning that these clusters are under high resource pressure and may result in performance degradation or even service interruption.

[0054] Count the number of abnormal load saturation points and divide it by the number of predicted load saturation points to calculate the abnormal saturation ratio. This ratio reflects how many clusters have exceeded the saturation threshold. If this ratio is too high, it indicates that the overall system load is high. Local capacity expansion may not solve the problem, and global capacity expansion should be considered.

[0055] The predetermined proportion threshold is a pre-set threshold used to determine whether global expansion is needed. This value can be set based on the system's load bearing capacity and business needs, such as 30% or 40%. If the proportion of abnormal saturation is less than the predetermined threshold, it is considered that local expansion can solve the problem; if the proportion is greater than the threshold, it means that the load pressure is too high and global expansion is required.

[0056] When the proportion of abnormal saturation is greater than the predetermined proportion threshold, it is considered that the global load pressure is large and local expansion cannot completely solve the problem. Therefore, the global expansion determiner in the cluster expansion determiner is activated to determine whether the entire system needs to be expanded.

[0057] The predicted load saturation of multiple clusters is calculated as a centralized value. For example, the average of multiple predicted load saturations is calculated to reflect the average level of the overall load to obtain the global load saturation. This is the overall load situation of the entire system and reflects the overall resource utilization of the system. It is the centralized value of the load saturation of multiple clusters and is used to evaluate the load pressure of the entire system in the future time zone.

[0058] The global load saturation is input into the global capacity expansion determiner, where the global capacity expansion determiner includes a global load saturation threshold, which is a preset threshold for evaluating the overall capacity expansion demand. It indicates the load pressure limit that the system can withstand at the global level. For example, the global load saturation threshold is set to 80%. If the global load saturation exceeds this threshold, it means that the overall system resources are insufficient, which may affect performance or cause instability.

[0059] The calculated global load saturation is compared with the global load saturation threshold. If the global load saturation is lower than or equal to the threshold, it indicates that the overall system is still at a safe load level. Local expansion can be used to address the load problem of the local cluster without the need for global expansion.

[0060] On the other hand, if the global load saturation is greater than the threshold, it is judged that the load pressure on the entire multi-cluster architecture is too large and cannot be alleviated by local expansion alone. Global expansion is required. In this case, the global expansion determiner outputs a global expansion instruction. This instruction requires the system to expand resources at the overall architectural level, including increasing computing resources in multiple clusters or adjusting resource allocation between clusters to balance the overall load of the system.

[0061] When the cluster expansion instruction is a global expansion instruction, a global expansion parsing channel is activated, and the multi-cluster architecture is globally expanded according to the global expansion parsing channel to obtain a multi-cluster optimized first architecture.

[0062] When the cluster expansion instruction is a global expansion instruction, the global expansion analysis channel is activated. The global expansion analysis channel is a logical module responsible for performing a global analysis of the load of the entire multi-cluster architecture and generating an expansion strategy, which includes a global expansion decision model and a global expansion fitness analysis model. The global expansion decision model generates a global expansion decision plan based on the current and predicted load conditions; the global expansion fitness analysis model evaluates the fitness of each expansion decision and continuously optimizes the expansion strategy through iterative optimization, ultimately generating the optimal expansion plan.

[0063] According to the optimal expansion plan, the entire multi-cluster architecture is globally expanded, for example, additional clusters are added. After the expansion is completed, the optimized multi-cluster optimized first architecture is obtained, that is, the state after global expansion.

[0064] Furthermore, when the cluster expansion instruction is a global expansion instruction, activating a global expansion resolution channel, and performing global expansion on the multi-cluster architecture according to the global expansion resolution channel to obtain a multi-cluster optimized first architecture, including:

[0065] A global expansion feature analysis is performed based on the multi-cluster architecture to obtain a global expansion capacity; the global expansion analysis channel includes a global expansion decision model and a global expansion fitness analysis model; based on the global expansion capacity, multiple global expansion decisions are obtained according to the global expansion decision model; based on the global expansion fitness analysis model, the multiple global expansion decisions are iteratively optimized to obtain a global expansion strategy; the multi-cluster architecture is globally expanded according to the global expansion strategy to obtain the multi-cluster optimized first architecture.

[0066] Perform a comprehensive resource and load analysis on the multi-cluster architecture to obtain a global capacity expansion. This capacity expansion is calculated by multiplying the sum of multiple constrained loads by the global load saturation threshold, and then subtracting the sum of multiple predicted loads from the product. The global capacity expansion represents the resources that can be scheduled or reallocated under the current architecture.

[0067] The global expansion analysis channel is a logical module responsible for performing a global analysis of the load of the entire multi-cluster architecture and generating an expansion strategy. This includes a global expansion decision model and a global expansion fitness analysis model. The global expansion decision model generates a global expansion decision plan based on current and predicted load conditions. The global expansion fitness analysis model uses a global expansion fitness analysis function to evaluate the effectiveness of the expansion strategy and continuously optimizes the expansion strategy through iterative optimization, ultimately generating the optimal expansion plan.

[0068] The global expansion decision model generates different expansion plans based on the global expansion capacity and the resource usage of each cluster. Each decision plan takes into account the cluster's current resource configuration, load pressure, and global load balancing requirements, generating multiple global expansion decisions suitable for different situations.

[0069] In order to select the optimal expansion plan, all generated global expansion decisions are gradually optimized based on the global expansion fitness analytical model. In each round of iteration, the current expansion plan is evaluated by the fitness analytical function, and the decision parameters are adjusted on this basis. Specifically, the multiple global expansion decisions generated are used as the initial solution, and the fitness analytical function is used to evaluate the effect of each initial expansion decision. The global expansion fitness is calculated. Based on the fitness evaluation, adjustments and optimizations are made to each expansion plan. After multiple rounds of iterative optimization, an expansion strategy with the highest fitness is found. This strategy is the global expansion strategy, which achieves the best balance between load balancing, resource utilization efficiency and expansion cost.

[0070] After obtaining the global expansion strategy, the actual global expansion operation is performed according to this optimal strategy, and cluster addition is executed. After the global expansion operation is completed, a new architectural state is entered, namely the multi-cluster optimized first architecture. This architecture represents the new state of the system after the global expansion.

[0071] Furthermore, the global expansion fitness analytical model includes a global expansion fitness analytical function, and the global expansion fitness analytical function is:

[0072]

[0073] Among them, GSA represents the global expansion fitness, exp represents the exponential function with the natural constant e as the base, csw and ccw represent the predetermined fitness analysis weight conditions, scx represents the predicted global load balancing degree, sxo represents the expected global load balancing degree, scc represents the predicted global expansion cost, and sco represents the expected global expansion cost.

[0074] Specifically, the global expansion fitness analytical function is as follows:

[0075]

[0076] The global expansion fitness (GSA) is an indicator for evaluating the effectiveness of a certain expansion plan. The higher its value, the better the adaptability of the plan. exp represents an exponential function with the natural constant e as the base. The reason for using the exponential function is that it can effectively amplify or reduce the evaluation results, especially when the fitness is high or low, making the results more significant. csw and ccw represent the predetermined fitness analysis weight conditions, which are used to adjust the influence of the predicted load balance and expansion cost in the fitness evaluation. scx represents the predicted global load balance. Ideally, the higher the load balance, the better the system performance. scc represents the predicted global expansion cost, which indicates the possible expenses incurred by implementing the expansion plan, usually including hardware costs, software costs and possible maintenance costs. sxo represents the expected global load balance, which is a target value indicating the load balance level that is expected to be achieved under ideal circumstances. sco represents the expected global expansion cost, which is also a target value indicating the expansion cost that is expected to be achieved under ideal circumstances.

[0077] Overall, this global expansion fitness analytical function is an evaluation tool that combines the effectiveness and cost-effectiveness of load balancing. It aims to ensure both improved load balancing and minimize expansion costs. The weight coefficients in the function allow for adjustments to the importance of different factors, making the final decision more consistent with the actual needs and goals of the system. This evaluation method ensures the scientific and rationality of expansion decisions, and helps the system make timely and effective resource expansion when facing high load pressure.

[0078] When the cluster expansion instruction is a local expansion instruction, the local expansion decision model is activated, and the multi-cluster architecture is partially expanded according to the local expansion decision model to obtain a multi-cluster optimized second architecture.

[0079] When a cluster expansion instruction is a local expansion instruction, the local expansion decision model is activated. This model analyzes the load saturation of a specific cluster and generates local expansion decisions. Unlike global expansion, this model is limited to individual clusters with excessive loads. By analyzing the load saturation of each cluster, clusters with loads exceeding the threshold are designated as target clusters for expansion. The expansion scale is determined, and detailed configuration parameters of the target clusters are collected to obtain cluster configuration feature data. This cluster configuration feature data and the local expansion capacity are input into the local expansion decision model to generate local expansion decisions.

[0080] Based on the generated partial expansion decision, the designated partial expansion cluster is expanded. The expansion operation is to add computing nodes to the partial expansion cluster. After the partial expansion, the optimized multi-cluster optimized second architecture is obtained, that is, the state after the partial expansion.

[0081] Furthermore, when the cluster expansion instruction is a local expansion instruction, activating a local expansion decision model, and performing local expansion on the multi-cluster architecture according to the local expansion decision model to obtain a multi-cluster optimized second architecture, including:

[0082] Perform local expansion feature analysis based on the multi-cluster architecture to determine a local expansion cluster and a local expansion capacity; collect configuration parameters of the local expansion cluster to obtain cluster configuration feature data; input the cluster configuration feature data and the local expansion capacity into the local expansion decision model to obtain a local expansion decision; and based on the multi-cluster architecture, expand the local expansion cluster according to the local expansion decision to obtain the multi-cluster optimized second architecture.

[0083] Partial expansion feature analysis involves analyzing the predicted load saturation of each cluster in a multi-cluster architecture and identifying clusters whose predicted load saturation exceeds a threshold. These clusters are under heavy load and require additional resources. Clusters with predicted load saturation greater than a preset load saturation threshold are designated as partial expansion clusters. The partial expansion capacity is the amount of resources required to be added to the partial expansion cluster. This is calculated by multiplying the constrained load of the partial expansion cluster by the load saturation threshold, and then subtracting the predicted load of the partial expansion cluster from the product.

[0084] To ensure that capacity expansion meets the actual needs of the cluster, it is necessary to collect detailed configuration parameters for the partially expanded cluster. These parameters include, but are not limited to: the number of nodes, i.e., the current number of compute nodes in the cluster; node resource configuration, i.e., the resource availability of each node, including the number of CPU cores, memory, and storage capacity; network bandwidth, i.e., the speed of network connections between nodes in the cluster and external networks; and load distribution strategy, i.e., how tasks or requests are distributed within the cluster, to determine whether there are any issues such as load imbalance. These configuration parameters are collected through monitoring tools or management systems to form a complete dataset of cluster configuration characteristics.

[0085] The local expansion decision model is used to generate local expansion plans. It comprehensively considers the current cluster configuration and the calculated local expansion capacity to determine how to expand capacity. By analyzing cluster configuration characteristics and local expansion capacity, the local expansion decision model outputs a local expansion decision. While the number of clusters remains unchanged, computing nodes in the local expansion cluster can be added to better handle the cluster's load.

[0086] Based on the local expansion decision, the selected local expansion cluster is actually expanded. After the local expansion is completed, the second multi-cluster optimization architecture is formed. This architecture represents the state after expansion. After expansion, the load pressure of the expansion cluster is relieved, and the resource usage of the entire system is more balanced.

[0087] Load balancing management and control is performed according to the multi-cluster optimization first architecture or the multi-cluster optimization second architecture.

[0088] The first multi-cluster optimization architecture is a global expansion architecture, meaning that clusters are added to multiple clusters. Global load balancing is then used to distribute the load appropriately. The second multi-cluster optimization architecture is a local expansion architecture, meaning that only some clusters are expanded. Therefore, load balancing focuses on these locally expanded clusters to ensure that the expanded resources are fully utilized. This load balancing management and control can achieve optimal resource utilization, reduce latency, and improve overall system performance and reliability.

[0089] In summary, the dynamic load balancing method for multi-cluster intelligent expansion provided by the embodiments of the present application has the following technical effects:

[0090] By collecting the load characteristics of the multi-cluster architecture in real time, the current load status of each cluster and computing node can be grasped. This real-time monitoring can ensure timely response to load changes, and lay a data foundation for subsequent load forecasting and expansion decisions, thereby improving the accuracy of load management. Based on the load characteristic data set, the load saturation of each cluster in the predetermined future time zone window is predicted, and multiple predicted load saturations are obtained, which improves the responsiveness of the system and enables timely expansion decisions under high load conditions to avoid performance degradation or service interruption caused by sudden load increases. Based on the predicted results of load saturation, the cluster expansion determiner is used to decide whether to perform global expansion. Whether it is local expansion or global expansion, such flexibility allows the system to make appropriate resource adjustments based on actual load requirements; during global expansion, the resource capacity of multiple clusters is comprehensively improved to form a multi-cluster optimization first architecture; during local expansion, the focus is on solving the load problem of a specific cluster to form a multi-cluster optimization second architecture. This dual expansion mechanism improves the adaptability of the system; after the expansion is completed, load balancing management and control are carried out according to the new optimization architecture to ensure that the load of all clusters is reasonably distributed to avoid overload or idleness of a cluster. Through dynamic load balancing, the overall performance and resource utilization can be improved to ensure that each cluster can maintain stable service quality during peak load periods.

[0091] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A dynamic load balancing system with multi-cluster intelligent expansion, characterized by: The system comprises: A load signature collection module, configured to collect load signatures based on a multi-cluster architecture to obtain a load signature dataset, wherein the multi-cluster architecture includes a plurality of clusters, each cluster including a plurality of computing nodes; a load saturation prediction module, configured to perform load saturation prediction on the plurality of clusters based on the load characteristic dataset based on a predetermined future time zone window, and obtain a plurality of predicted load saturations; an expansion instruction parsing module, the expansion instruction parsing module being configured to input the plurality of predicted load saturations into a cluster expansion determiner to obtain a cluster expansion instruction, wherein the cluster expansion instruction is a global expansion instruction or a local expansion instruction; a global expansion module, configured to activate a global expansion parsing channel when the cluster expansion instruction is a global expansion instruction, and perform global expansion on the multi-cluster architecture according to the global expansion parsing channel to obtain a multi-cluster optimized first architecture; a local expansion module, configured to activate a local expansion decision model when the cluster expansion instruction is a local expansion instruction, and perform local expansion on the multi-cluster architecture according to the local expansion decision model to obtain a multi-cluster optimized second architecture; A load balancing control module, configured to perform load balancing control according to the multi-cluster optimization first architecture or the multi-cluster optimization second architecture; The global expansion module is configured to activate a global expansion parsing channel when the cluster expansion instruction is a global expansion instruction, and perform global expansion on the multi-cluster architecture according to the global expansion parsing channel to obtain a multi-cluster optimized first architecture, including: Performing global expansion feature analysis based on the multi-cluster architecture to obtain global expansion capacity; The global expansion analysis channel includes a global expansion decision model and a global expansion adaptability analysis model; Based on the global expansion capacity, and according to the global expansion decision model, a plurality of global expansion decisions are obtained; Based on the global expansion fitness analytical model, iteratively optimize the multiple global expansion decisions to obtain a global expansion strategy; Globally expanding the multi-cluster architecture according to the global expansion strategy to obtain the multi-cluster optimized first architecture; The global expansion fitness analytical model includes a global expansion fitness analytical function, and the global expansion fitness analytical function is: ； in, It represents the global expansion fitness, exp represents the exponential function with the natural constant e as the base, csw and ccw represent the predetermined fitness analysis weight conditions, scx represents the predicted global load balancing degree, sxo represents the expected global load balancing degree, scc represents the predicted global expansion cost, and sco represents the expected global expansion cost.

2. The system according to claim 1, wherein The load saturation prediction module is configured to perform load saturation prediction on the plurality of clusters based on the load characteristic dataset based on a predetermined future time zone window to obtain a plurality of predicted load saturations, including: Performing feature recognition based on the load feature data set to obtain multiple current loads and multiple constrained loads corresponding to the multiple clusters; Based on the predetermined future time zone window, predicting load change trends of the multiple clusters according to the multiple current loads to obtain multiple predicted loads; The plurality of predicted load saturations are calculated based on the plurality of predicted load amounts and the plurality of constrained load amounts.

3. The system according to claim 1, wherein: The expansion instruction parsing module is configured to input the multiple predicted load saturations into the cluster expansion determiner to obtain a cluster expansion instruction, wherein the cluster expansion instruction is a global expansion instruction or a local expansion instruction, including: activating a local capacity expansion determiner within the cluster capacity expansion determiner; inputting the plurality of predicted load saturations into the local expansion determiner, wherein the local expansion determiner includes a load saturation threshold; determining whether the plurality of predicted load saturations are greater than the load saturation threshold; If any one of the multiple predicted load saturations is greater than the load saturation threshold, the local capacity expansion determiner outputs the local capacity expansion instruction.

4. The system according to claim 3, wherein: The local capacity expansion determiner outputs the local capacity expansion instruction, including: Recording the predicted load saturation greater than the load saturation threshold as abnormal load saturation; Calculating the quantity ratio of the abnormal load saturation and the multiple predicted load saturations to obtain the quantity ratio of abnormal saturation; Determine whether the abnormal saturation ratio is greater than a predetermined ratio threshold; If the proportion of the abnormal saturation number is greater than the predetermined proportion threshold, activating the global expansion determiner in the cluster expansion determiner; Calculating a centralized value of the plurality of predicted load saturations to obtain a global load saturation, and inputting the global load saturation into the global capacity expansion determiner, wherein the global capacity expansion determiner includes a global load saturation threshold; Determining whether the global load saturation is greater than the global load saturation threshold; If the global load saturation is greater than the global load saturation threshold, the global capacity expansion determiner outputs the global capacity expansion instruction and prohibits the local capacity expansion instruction.

5. The system according to claim 1, wherein: The local expansion module is configured to activate a local expansion decision model when the cluster expansion instruction is a local expansion instruction, and perform local expansion on the multi-cluster architecture according to the local expansion decision model to obtain a multi-cluster optimized second architecture, including: Analyze local expansion characteristics based on the multi-cluster architecture to determine the local expansion cluster and local expansion capacity; Collecting configuration parameters of the local expansion cluster to obtain cluster configuration feature data; Inputting the cluster configuration feature data and the local expansion capacity into the local expansion decision model to obtain a local expansion decision; Based on the multi-cluster architecture, the local expansion cluster is expanded according to the local expansion decision to obtain the multi-cluster optimized second architecture.

6. A dynamic load balancing method for multi-cluster intelligent expansion, characterized in that: Based on the implementation of the multi-cluster intelligent expansion dynamic load balancing system according to any one of claims 1 to 5, the method includes: Collecting load characteristics according to a multi-cluster architecture to obtain a load characteristic data set, wherein the multi-cluster architecture includes a plurality of clusters, each cluster including a plurality of computing nodes; Based on a predetermined future time zone window, performing load saturation prediction on the plurality of clusters according to the load characteristic dataset to obtain a plurality of predicted load saturations; Inputting the plurality of predicted load saturations into a cluster expansion determiner to obtain a cluster expansion instruction, wherein the cluster expansion instruction is a global expansion instruction or a local expansion instruction; When the cluster expansion instruction is a global expansion instruction, activating a global expansion parsing channel, and performing global expansion on the multi-cluster architecture according to the global expansion parsing channel to obtain a multi-cluster optimized first architecture; When the cluster expansion instruction is a local expansion instruction, activating a local expansion decision model, and performing local expansion on the multi-cluster architecture according to the local expansion decision model to obtain a multi-cluster optimized second architecture; Load balancing management and control is performed according to the multi-cluster optimization first architecture or the multi-cluster optimization second architecture.

Images