Capacity expanding and shrinking method, device and equipment based on load entropy and medium

Abstract

The invention relates to the technical field of cloud computing, and discloses a load entropy-based capacity expansion and contraction method, device, equipment and medium, and the method comprises the steps: recognizing the performance index of each service node according to a resource utilization rate, and calculating the real-time node entropy of each service node, extracting an entropy value measurement index in the real-time node entropy and a historical node entropy of each service node in a historical time range, and identifying a minimum entropy value in the historical node entropy; calculating the service cluster entropy of the target system, and dynamically adjusting the capacity expansion and contraction response strength of the target system according to the priority of the target service; analyzing the capacity expansion number through the target entropy, the service cluster entropy, the service flow and the capacity expansion and contraction response intensity, and analyzing the capacity contraction number through the minimum entropy value, the service flow and the capacity expansion and contraction response intensity; and generating a capacity expansion and contraction instruction based on the capacity expansion number and the capacity contraction number, sending the capacity expansion and contraction instruction to the server management platform, and executing the capacity expansion and contraction operation of the target system based on the capacity expansion and contraction instruction. The resource utilization rate of the system can be improved.

Description

Technical Field

[0001] This invention relates to the field of cloud computing technology, and in particular to a scaling method, apparatus, device, and medium based on load entropy. Background Technology

[0002] With the widespread adoption of cloud computing, microservice architecture, and distributed systems, the load status of backend service instances dynamically changes in high-concurrency, high-traffic-fluctuation business scenarios. If resource supply cannot match actual demand in a timely manner, it can easily lead to performance degradation, response latency, or even service avalanche. Therefore, achieving automated elastic scaling of gateway backend services has become a core requirement for ensuring system stability and resource efficiency.

[0003] Existing scaling technologies mostly rely on operations and maintenance personnel to manually determine and deploy new instances, resulting in delayed scaling responses, difficulty in coping with sudden traffic peaks, increased risk of system overload, and thus low system resource utilization. Summary of the Invention

[0004] This invention provides a method, apparatus, device, and medium for scaling up and down based on load entropy to solve the technical problem of low resource utilization in the system.

[0005] Firstly, a scaling method based on load entropy is provided, including: When the target system is detected to be running, the resource utilization rate of each service node in the target system is collected, and the performance indicators of each service node are identified based on the resource utilization rate. Calculate the real-time node entropy of each service node based on the performance indicators, extract the entropy value metric from the real-time node entropy, and use the entropy value metric as the target entropy of the target system. Extract the historical node entropy of each service node within a preset historical time range, and identify the minimum entropy value among the historical node entropies; The service cluster entropy of the target system is calculated based on the resource utilization rate, and the scaling response intensity of the target system is dynamically adjusted according to the priority of the preset target services. The number of expansion units in the target system is analyzed using the target entropy, the service cluster entropy, the service traffic of the target system, and the scaling-up / scaling response intensity; the number of scaling-down units in the target system is analyzed using the minimum entropy value, the service traffic of the target system, and the scaling-up / scaling response intensity. Based on the number of expansion units and the number of reduction units, the expansion / reduction instructions for the target system are generated and sent to the server management platform. The expansion / reduction operations for the target system are then executed based on the expansion / reduction instructions.

[0006] Secondly, a capacity scaling device based on load entropy is provided, comprising: The performance indicator identification module is used to collect the resource utilization rate of each service node in the target system when the target system is detected to be running, and to identify the performance indicator of each service node based on the resource utilization rate. The entropy metric extraction module is used to calculate the real-time node entropy of each service node based on the performance metric, extract the entropy metric from the real-time node entropy, and use the entropy metric as the target entropy of the target system. The minimum entropy value identification module is used to extract the historical node entropy of each service node within a preset historical time range and identify the minimum entropy value among the historical node entropies. The scaling response intensity adjustment module is used to calculate the service cluster entropy of the target system based on the resource utilization rate, and dynamically adjust the scaling response intensity of the target system according to the priority of the preset target services. The scaling up / down unit count analysis module is used to analyze the number of units that can be expanded in the target system using the target entropy, the service cluster entropy, the service traffic of the target system, and the scaling up / down response intensity, and to analyze the number of units that can be reduced in the target system using the minimum entropy value, the service traffic of the target system, and the scaling up / down response intensity. The scaling up / down operation module is used to generate scaling up / down instructions for the target system based on the number of units to be expanded and the number of units to be scaled down, send the scaling up / down instructions to the server management platform, and execute the scaling up / down operation of the target system based on the scaling up / down instructions.

[0007] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described load entropy-based scaling method.

[0008] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-described expansion and contraction method based on load entropy.

[0009] In this invention, the accuracy of system state assessment is improved by using node entropy and cluster entropy models, effectively suppressing instantaneous fluctuations; precise elastic response is achieved through asymmetric scaling decisions, avoiding resource oscillations; and the service quality of critical services is guaranteed through a business priority mechanism, while optimizing resource utilization. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of an application environment for a scaling method based on load entropy in one embodiment of the present invention; Figure 2 This is a flowchart illustrating a scaling method based on load entropy in one embodiment of the present invention. Figure 3 This is an architecture diagram of a scaling service network based on a load entropy-based scaling method in one embodiment of the present invention. Figure 4 This is a schematic diagram of a capacity expansion / contraction device based on load entropy in one embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 6 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] The scaling method based on load entropy provided in this invention can be applied to applications such as... Figure 1In this application environment, the client communicates with the server via a network. The server can collect resource utilization data of each service node within the target system from the client, identify the performance indicators of each node, calculate node entropy based on the performance indicators, and extract the entropy value as the target entropy of the system. It can also retrieve the historical node entropy of each node within a preset historical period and identify the minimum entropy value. Furthermore, it can calculate the service cluster entropy of the target system and dynamically adjust the response intensity of scaling up and down based on the priority of the target business. Based on the target entropy, service cluster entropy, business traffic, and response intensity, it can analyze the number of nodes to be expanded and determine the number of nodes to be scaled down based on the minimum entropy value, business traffic, and response intensity. Finally, it can generate scaling up / down commands, send them to the server management platform for execution, and feed the results back to the client. By using node entropy and cluster entropy, the accuracy of system status assessment is improved, reducing the interference of instantaneous fluctuations. Precise and elastic responses are achieved through asymmetric scaling up / down decisions, avoiding resource oscillations. The service quality of critical businesses is guaranteed through a business priority mechanism, while resource utilization is optimized. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster composed of multiple servers. The present invention will now be described in detail through specific embodiments.

[0014] Please see Figure 2 As shown, Figure 2 A schematic flowchart of a scaling up / down method based on load entropy provided in an embodiment of the present invention includes the following steps: S1. When the target system is detected to be running, the resource utilization rate of each service node in the target system is collected, and the performance indicators of each service node are identified based on the resource utilization rate.

[0015] In this embodiment of the invention, the target system refers to a distributed service cluster running in a cloud computing environment, the service node refers to a computing instance in the cluster that provides specific service functions, and the resource utilization rate refers to the proportion of various computing resources used by the service node during operation.

[0016] In detail, when the target system is detected to be running, the monitoring and acquisition module deployed on each service node is started; the monitoring and acquisition module is controlled to periodically collect the CPU utilization, memory usage and network card bandwidth utilization of each service node; the collected CPU utilization, memory usage and network card bandwidth utilization are used as raw load data; and the raw load data is reported to the general gateway.

[0017] In this embodiment of the invention, the performance index is a quantitative parameter that comprehensively reflects the resource usage status and operating efficiency of the service node.

[0018] In this embodiment of the invention, identifying the performance metrics of each service node based on the resource utilization rate includes: Based on the resource utilization rate, the CPU utilization rate, memory usage rate, and network card bandwidth utilization rate of each service node are extracted; The weights of CPU utilization, memory usage, and network card bandwidth utilization are configured according to the service operation type of each service node. The performance metrics of each service node are obtained by weighting and summing the weights, CPU utilization, memory usage, and network card bandwidth utilization.

[0019] In detail, CPU utilization is the percentage of a service node's CPU resources being used; memory utilization is the percentage of a service node's memory resources being used; and network interface card (NIC) bandwidth utilization is the percentage of a service node's NIC transmission bandwidth being used. This is achieved through a resource monitoring agent built into the service node. The monitoring agent uses the SNMP or Prometheus protocol to collect data and read the service node's system kernel parameters in real time. For CPU utilization, it is calculated by subtracting the percentage of idle time from 1. For memory utilization, it is calculated by the ratio of used memory capacity to total memory capacity. For NIC bandwidth utilization, it is determined by the ratio of the actual amount of data transmitted by the NIC per unit time to the NIC's maximum transmission bandwidth, thus comprehensively obtaining the core resource usage data of the service node.

[0020] Specifically, the weights are importance coefficients assigned to each resource utilization indicator based on the service node's business operation type. The business operation type is identified by the business tags declared during service node registration, such as compute-intensive businesses (e.g., data modeling, algorithm computation), I / O-intensive businesses (e.g., database queries, file transfer), and network-intensive businesses (e.g., video stream distribution, API gateway). For compute-intensive businesses, CPU utilization has the greatest impact on node performance, with a configured weight of 0.6 for CPU utilization, 0.3 for memory utilization, and 0.1 for network interface card (NIC) bandwidth utilization. For I / O-intensive businesses, the weights are 0.5 for memory utilization, 0.3 for CPU utilization, and 0.2 for NIC bandwidth utilization. For network-intensive businesses, the weights are 0.6 for NIC bandwidth utilization, 0.2 for CPU utilization, and 0.2 for memory utilization. The configuration process is automatically completed through a preset business type-weight mapping table, or it can be manually adjusted by the administrator. This ensures that performance indicator calculations better reflect actual business needs in distributed systems with multiple business types.

[0021] Furthermore, the performance indicators are calculated using a weighted summation method, that is... ,in, The performance metrics are as described above. For CPU utilization, For memory usage, For network card bandwidth utilization, As a weight for CPU utilization, As a weight for memory usage, The weight is the network card bandwidth utilization. ,and It can be configured according to the business type (such as compute-intensive applications). , , CPU dominance ensures that high CPU utilization is quickly reflected in overall load. In the process, timely expansion is triggered, thereby generating comprehensive and accurate performance quantification data. The limitation of single resource indicators is eliminated through performance index calculation, providing a foundation for subsequent node entropy calculation.

[0022] S2. Calculate the real-time node entropy of each service node based on the performance indicators, extract the entropy value metric from the real-time node entropy, and use the entropy value metric as the target entropy of the target system.

[0023] In this embodiment of the invention, real-time node entropy is used to quantify the degree of health deviation of a single server.

[0024] In this embodiment of the invention, calculating the real-time node entropy of each service node based on the performance metric includes: Determine the target load threshold for each service node based on the business type of the target system; Configure the sensitivity coefficient of each service node according to the stability status of the target system; The real-time node entropy of each service node is obtained by performing a nonlinear mapping on the performance indicators, the target load threshold, and the sensitivity coefficient.

[0025] Specifically, the target load threshold is a critical value set according to the service node business type to determine whether the node load is within a reasonable range. The business type of the target system is determined by the business attribute identifier configured during system initialization, such as financial transaction business, general data processing business, and log storage business. For the financial transaction business, high load stability is required, and the target load threshold is set to 60%; for the general data processing business, the target load threshold is set to 70%; for the log storage business, the target load threshold is set to 80%. The target load threshold is stored in the system configuration center and can be dynamically updated according to business requirements to match the load threshold with the business characteristics for target systems of different business types. The sensitivity coefficient is a parameter configured according to the stability state of the target system and is used to adjust the response degree of node entropy to changes in performance indicators. The stability state of the target system is judged by real-time monitoring parameters such as system response delay and error rate, and it is divided into stable state, general state, and fluctuating state. When the system response delay is less than 50 milliseconds and the error rate is less than 0.1%, it is judged as the stable state, and the sensitivity coefficient is configured to 0.5; when the response delay is between 50 - 100 milliseconds and the error rate is between 0.1% - 0.5%, it is judged as the general state, and the sensitivity coefficient is configured to 1.0; when the response delay is higher than 100 milliseconds or the error rate is higher than 0.5%, it is judged as the fluctuating state, and the sensitivity coefficient is configured to 2.0. The configuration process is automatically executed by the system stability monitoring module, and the sensitivity coefficient is updated every 30 seconds.

[0026] Specifically, use the Sigmoid function to perform a non-linear mapping on it to generate the node entropy value, where is the node entropy, is the performance indicator, is the target load threshold, usually set to 0.7 (i.e., 70%), representing the optimal operating point of the system, is the sensitivity coefficient, controlling the steepness of the function, and the recommended value is k = 5, is a constant. This model transforms the load state from the traditional binary judgment of "whether it exceeds the threshold" to a continuous health score between 0 and 1, which can effectively smooth instantaneous fluctuations, distinguish short-term stress from continuous high load, and significantly improve the accuracy of system state assessment. The core of this model is to utilize the saturation characteristic of the Sigmoid function to respond sensitively when the load approaches the ideal threshold T and tend to be stable in the light load or heavy load intervals to avoid false expansion and contraction. The Sigmoid function has the following characteristics: when L < T, < 0.5, indicating that the load is low; when L = T, = 0.5, the system is in an ideal state; when L > T, The system rapidly approaches 1, demonstrating "overload sensitivity." The nonlinear response smooths out instantaneous fluctuations and avoids false propagation. This nonlinear characteristic effectively suppresses short-term spikes, enabling the system to respond only to sustained high loads and improving decision-making accuracy.

[0027] In this embodiment of the invention, the entropy metric is a core parameter extracted from node entropy that can characterize the overall load balancing state of the target system. It includes the maximum value and weighted average value of node entropy. The maximum value or weighted average value is then used as the target entropy, which is an entropy metric used to evaluate the current overall operating state of the target system.

[0028] In detail, the maximum value in node entropy can prioritize reflecting the health status of the worst-performing node in the system, ensuring a rapid response to overloaded nodes. When the entropy value of any node exceeds a safe threshold, the system can immediately trigger a scaling-up operation to prevent the spread of single-point failures and improve system reliability and response speed. For example, in financial transaction scenarios, an overloaded node may cause transaction delays. Taking the maximum value prioritizes ensuring the continuity of critical business operations, making it suitable for scenarios with extremely high requirements for system stability and high availability, such as critical businesses like finance and payments, where overload of any node could trigger service avalanche or severe performance degradation. It is also suitable for heterogeneous environments with significant differences in node load, where some nodes may overload before others due to specific tasks. The weighted average can comprehensively consider the status of all nodes, smoothing out instantaneous fluctuations in individual nodes, providing a view of the overall cluster health status, and reducing erroneous scaling-up or scaling-down caused by short-term disturbances. This makes decisions more robust, avoids over-responding to single-node anomalies, and improves resource utilization efficiency. For example, in a typical web service cluster, the load distribution is relatively uniform, and the weighted average can reflect the overall trend, preventing unnecessary scaling operations. This is suitable for scenarios where node loads are relatively uniform, or when the system can tolerate temporary overload of individual nodes without immediate scaling, such as for ordinary computing tasks or web services. Furthermore, it is suitable for environments requiring long-term stability rather than rapid response, where resource utilization efficiency takes precedence over immediate fault handling.

[0029] Furthermore, the entropy metric is obtained by calculating the average or weighted average of the node entropy of all service nodes, which can comprehensively reflect the overall load balance status of the target system. Moreover, it smooths out instantaneous fluctuations through nonlinear mapping, significantly reducing the misjudgment rate.

[0030] S3. Extract the historical node entropy of each service node within a preset historical time range, and identify the minimum entropy value among the historical node entropies.

[0031] In this embodiment of the invention, the historical time range is a pre-set time interval used to obtain past operational data of the service node, which can be set to 1 minute, 5 minutes, or 7 minutes, etc., according to system requirements. Historical node entropy is the node entropy data generated by the service node within the preset historical time range.

[0032] In detail, the system reads a preset historical time range parameter from the configuration center, assuming it is set to 5 minutes; it then uses a database query (such as an SQL statement) to filter out all node entropy records for each service node within that time range, with query conditions including the service node ID and timestamp; the retrieved historical node entropy data is then stored in a temporary data structure categorized by service node. This process employs batch querying and asynchronous processing to avoid excessive pressure on the database, with a query response time not exceeding 3 seconds.

[0033] Specifically, the minimum entropy value is the smallest entropy value recorded in the historical node entropy data, reflecting the lowest load fluctuation state in the past operation of the service node. For the historical node entropy data stored in categories for each service node, the data is sorted in ascending order using a sorting algorithm (such as bubble sort or quick sort), and the first value of the sorted data is taken as the minimum entropy value of that service node. Thus, based on the minimum entropy value, the historical lowest load fluctuation level of the service node is accurately located, providing a key reference for subsequent scaling-down decisions. In other words, historical trend analysis provides a basis for scaling-down decisions, ensuring the stability of resource recovery.

[0034] S4. Calculate the service cluster entropy of the target system based on the resource utilization rate, and dynamically adjust the scaling response intensity of the target system according to the priority of the preset target services.

[0035] In this embodiment of the invention, the service cluster entropy is used to assess the overall health status of the entire service cluster.

[0036] In this embodiment of the invention, calculating the service cluster entropy of the target system based on the resource utilization rate includes: Based on the resource utilization rate, calculate the total resource usage and total resource capacity of all service nodes in the target system, and calculate the ratio between the total resource usage and the total resource capacity; Obtain the number of running instances of the target system, and determine the idle sensitivity coefficient of the target system based on the number of running instances; The product of the number of running instances and the instance idle sensitivity coefficient is exponentialized, and the preset benchmark value in the target system is subtracted from the exponentialized result to obtain the target value. The service cluster entropy of the target system is obtained by weighting the ratio with the target value.

[0037] In detail, total resource usage is the total amount of resources occupied by all service nodes in the target system at a certain moment, including the cumulative usage of resources such as CPU, memory, and network card bandwidth. Total resource capacity is the upper limit of the total resources possessed by all service nodes in the target system, obtained by summing the individual resource capacities of each service node. The number of running instances is the number of business instances currently running in the target system, obtained by querying the instance status interface of the server management platform, and the number of currently running instances is returned in real time. The idle sensitivity coefficient is a parameter determined based on the number of running instances, used to adjust the sensitivity of the service cluster entropy to the number of idle instances. The determination of the idle sensitivity coefficient is based on a preset instance count-coefficient mapping relationship; for example, when the number of running instances is less than 5, the coefficient is 0.3; when the number is between 5 and 15, the coefficient is 0.5; and when the number is greater than 15, the coefficient is 0.8.

[0038] Specifically, through Calculate cluster entropy, where, To serve cluster entropy, As the first weighting coefficient, Total resource usage Total resource capacity This is the second weighting coefficient. It is a constant. The instance idle sensitivity coefficient. To determine the number of running instances, and to avoid blind scaling issues caused by relying solely on node load, resource utilization should be comprehensively considered. With the number of instances The impact on system stability. Where μ and σ are weighting coefficients, satisfying... We recommend μ=0.6 and σ=0.4. The instance idle sensitivity coefficient controls the strength of the impact of the number of instances on entropy. Recommended... , The term (exponential decay term) is used to characterize the increase in management overhead and scheduling complexity as the number of instances increases. It effectively reflects the increase in marginal cost brought about by the growth in the number of instances and is superior to linear penalty. The first term... Reflects resource utilization efficiency; the second item Indicates a scale stability penalty – when the number of instances... When there is too much, the item tends to approach This indicates increased management overhead; combining these two aspects can avoid the problem of instance explosion caused by blind expansion, achieving a balance between resource utilization and system stability. This causes the system to increase entropy when resource utilization is high or the number of instances is excessive, thereby triggering optimization actions and achieving a balance between resource utilization efficiency and overall system stability. Furthermore, the construction method of this composite evaluation index and its use in elastic control decisions include, but are not limited to, the function form, the introduction of an exponential decay term, and the fusion logic of multi-dimensional health indicators.

[0039] Furthermore, to support the SLA requirements of different services, a scaling response strength (i.e., service level bonus parameter) is introduced. It allows for dynamic adjustment of scaling up and down response intensity based on business importance.

[0040] In this embodiment of the invention, the expansion / shrinkage response intensity is a parameter of the response speed and adjustment range of the target system when performing expansion or shrinkage operations. The higher the response intensity, the faster the adjustment speed and the greater the adjustment range.

[0041] In this embodiment of the invention, dynamically adjusting the scaling response strength of the target system according to the priority of a preset target service includes: Identify the business type of the target business and query the priority corresponding to the business type in a predefined business priority mapping table; Identify the target level corresponding to the priority; when the target level is a preset first level, determine the expansion response strength of the target system based on the target level. When the target level is a preset second level, the scaling-down response strength of the target system is determined according to the target level.

[0042] In detail, the target business is the core business that the user pre-specifies and that requires priority in operation. Priority is a hierarchical classification that distinguishes the importance of different target businesses, typically divided into first-level (high priority) and second-level (normal priority). The business type of the target business is identified by the identifier field in the business request header; for example, the identifier field "financial" corresponds to financial transaction business, and "data" corresponds to data processing business. A predefined business priority mapping table is stored in the system configuration center, where financial transaction business, core API services, etc., are mapped to first-level (high priority), and ordinary data processing, log analysis, etc., are mapped to second-level (normal priority).

[0043] Specifically, when the queried priority is level one, the expansion response strength is set to high level, with parameters including a response latency of 1 second and a single expansion ratio not exceeding 30% of the current number of instances. For example, if there are 10 instances currently, the maximum expansion in a single instance is 3. This mapping between priority level and response strength parameters ensures rapid resource replenishment for high-priority services. When the queried priority is level two, the shrinkage response strength is set to normal level, with parameters including a response latency of 5 seconds and a single shrinkage ratio not exceeding 20% ​​of the current number of instances. For example, if there are 10 instances currently, the maximum shrinkage in a single instance is 2. Based on this mapping between priority level and parameters, the shrinkage response strength parameters for normal-priority services avoid excessive shrinkage that could impact service operation, ensuring the normal operation of normal services while optimizing resource allocation.

[0044] Furthermore, higher allocations are made to high-priority businesses (such as financial transactions). Higher configurations offer the advantage of faster scaling up and later scaling down; lower configurations are available for general business applications. This saves resources. It enables differentiated and refined resource scheduling, meeting service quality requirements across multiple business scenarios. The way parameters are introduced into the scaling decision formula and their effect on the response intensity, regardless of whether they are implemented as multiplication factors, exponential offsets or lookup table mappings, should be included in any design that affects the scaling scale through business level.

[0045] S5. Analyze the number of expansion units in the target system using the target entropy, the service cluster entropy, the service traffic of the target system, and the scaling-up / scaling response intensity; and analyze the number of scaling-down units in the target system using the minimum entropy value, the service traffic of the target system, and the scaling-up / scaling response intensity.

[0046] In this embodiment of the invention, the number of expansion units refers to the number of new service instances that the target system needs to add, in order to alleviate resource shortages.

[0047] In this embodiment of the invention, the step of analyzing the number of expansion units in the target system using the target entropy, the service cluster entropy, the service traffic of the target system, and the scaling response strength includes: When the target entropy is less than or equal to a preset first security threshold, or the service cluster entropy is greater than or equal to a preset second security threshold, or the trend of the business traffic of the target system is decreasing or stable, the number of running instances of the target system remains unchanged. When the target entropy is greater than a preset first security threshold, the service cluster entropy is less than a preset second security threshold, and the trend of the business traffic of the target system is a continuous upward trend, the risk sensitivity coefficient of the target system is determined according to the expansion response degree in the expansion and contraction response strength. Based on the risk sensitivity coefficient, the target entropy and the safety threshold are exponentially calculated to obtain a first value; The first value, the expansion response strength, and the number of real-time running instances of the target system obtained in advance are added to the business value to obtain the second value; The second value is rounded up to obtain the number of units required to expand the target system.

[0048] Specifically, if the target entropy is less than or equal to a preset first security threshold, the service cluster entropy is greater than or equal to a preset second security threshold, or the trend of the target system's business traffic is decreasing or stable, then the system load is determined to be within a safe and controllable range, and resource utilization and load stability have not exceeded preset boundaries. In this case, expansion would cause resource redundancy, violating the core objective of efficient resource utilization; therefore, expansion is not triggered, and the current number of instances in the target system remains unchanged. When both the target entropy and the service cluster entropy are greater than the preset security thresholds, it is determined that the overall system load fluctuation exceeds the safe range and cluster resource usage is strained, triggering the expansion count calculation process. The expansion count for the target system is then calculated, where: ,in, To expand the number of units, It is a rounding function. The number of real-time running instances of the target system. For the target entropy, As a safety threshold, For risk sensitivity coefficient, This refers to the scaling response intensity within the scaling response strength. If the current maximum node entropy... > (like =0.75 (first safety threshold) and cluster entropy If the load factor is less than 0.85 (the second safety threshold), and business traffic (such as QPS) shows a continuous upward trend (e.g., an increase of more than 20% in the past minute), then the expansion process begins; otherwise, expansion is not initiated. This ensures that expansion is only triggered when the overall system load fluctuates excessively and cluster resources are strained, avoiding unnecessary expansion operations and reducing resource waste. The number of nodes to be expanded is equal to the result after rounding up. This joint judgment mechanism can effectively avoid blind expansion due to a sudden increase in load at a single point or when the cluster as a whole is already in a state of high management overhead (too many instances), preventing "instance explosion." When node load is high but cluster entropy is too high, the system should prioritize alleviating pressure by optimizing load balancing and restarting abnormal nodes, rather than immediately expanding.

[0049] Specifically, based on the asymmetric control strategy of aggressive expansion and conservative contraction, when expanding capacity, an exponential growth factor and business bonus (1+α) are adopted to ensure rapid response under high load; Risk sensitivity coefficient Control the "aggressiveness" of the capacity expansion response. At that time, the function grows exponentially; the higher the load, the faster the expansion. Recommended value. , ( Reflecting the principle of prioritizing risk, the recommended value range is... To avoid due to Excessive size leads to "expansion explosion" (referring to a large capacity). By expanding the capacity of critical business operations to ensure they receive more sufficient resource guarantees, aggressive expansion was achieved through asymmetric design, ensuring rapid response under high load.

[0050] In this embodiment of the invention, the number of reduced-scale units is the number of service instances that the target system needs to reduce, in order to optimize resource allocation.

[0051] In this embodiment of the invention, the step of analyzing the number of units to be scaled down in the target system using the minimum entropy value, the service traffic of the target system, and the scaling-down response strength includes: When the minimum entropy value is greater than or equal to a preset first security threshold or the business traffic of the target system shows an upward trend, the number of running instances of the target system remains unchanged. When the minimum entropy value is less than the preset first security threshold and the trend of the service traffic of the target system is decreasing or stabilizing, the scaling down idle sensitivity coefficient of the target system is determined according to the scaling down response degree in the scaling up and down response strength. Based on the shrinkage idle sensitivity coefficient, the minimum entropy value and the security threshold are exponentially calculated to obtain a third value; A fourth value is obtained by adding a service bonus to the third value and the scaling-down response strength; Select the minimum value between the second value and the preset maximum scaling ratio, and multiply the minimum value by the number of real-time running instances of the target system obtained in advance to obtain the fifth value; The fifth value is rounded down to obtain the number of units in the reduced-capacity target system.

[0052] Specifically, if the minimum entropy value is greater than or equal to the preset first security threshold, or if the business traffic trend of the target system is upward, it is determined that the historical minimum load fluctuation of the service node is exactly at the security baseline, and the resources are not in an idle state that requires scaling down. In this case, scaling down would lead to insufficient system redundant resources, which could easily cause stability issues when the business load increases subsequently. Therefore, the scaling down operation is not triggered, and the current number of instances of the target system remains unchanged. When the minimum entropy value is less than the preset security threshold, it is determined that the historical minimum load fluctuation of the service node is far below the security baseline, and the resources have idle redundancy. The scaling down calculation process is then triggered. ,in, To reduce the number of units, This is the floor function. The number of real-time running instances of the target system. To be the minimum value, As a safety threshold, The minimum entropy value, This is the shrinkage idle sensitivity coefficient. The degree of expansion response in the expansion / shrinkage response intensity. This represents the maximum shrinkage ratio. It is based on the lowest entropy value within the past 5 minutes. If the cluster entropy Hcluster < 0.4, and the current business traffic is stable or decreasing (e.g., QPS change rate < ±10%, response time not significantly deteriorating), the scaling-down process begins. This ensures that scaling-down is only triggered when there are nodes with historically low load fluctuations, avoiding excessive scaling-down from impacting business operations. The number of nodes to be scaled down is equal to the result after rounding down. This mechanism achieves the dual goals of "accelerated scaling-down" and "security protection": on the one hand, when the overall cluster resource utilization is low and the number of instances is high, resources can be proactively reclaimed to improve efficiency even if the load on individual nodes does not decrease significantly; on the other hand, by verifying QPS and response time, it ensures that excessive scaling-down is not performed before business is about to recover, guaranteeing service stability.

[0053] Specifically, The idle sensitivity coefficient for scaling down controls the response strength of scaling down to the degree of idleness. When η>1, the response is more sensitive; it is recommended that η=1.2 to ensure that resources are gradually recovered when the load is consistently low, and to avoid excessive scaling down at one time. Reflecting the principle of "more conservative downsizing of key business operations", high The scale down of operations is even smaller. (Maximum reduction ratio), limiting the capacity reduction in a single operation to prevent insufficient system redundancy; recommended. , During volume reduction, introduce Attenuation factor and maximum reduction ratio This design prevents service jitter caused by excessively rapid resource reclamation, effectively solving the oscillation problem of "shrinking and expanding" in traditional symmetrical strategies and improving system stability. A linkage control mechanism of "expanding to strengthen and shrinking to weaken" is implemented using the same parameter α, and resource oscillations are avoided through conservative shrinkage design, thus improving system stability.

[0054] S6. Generate expansion / reduction instructions for the target system based on the number of expansion units and the number of reduction units, and send the expansion / reduction instructions to the server management platform, and execute the expansion / reduction operation of the target system based on the expansion / reduction instructions.

[0055] In this embodiment of the invention, the scaling up / down command refers to the operation command that controls the creation or destruction of service instances, and includes key information such as the number of instances to be adjusted, the adjustment type (scaling up / scaling down), and the execution time.

[0056] In this embodiment of the invention, generating the expansion / reduction instructions for the target system based on the number of expansion units and the number of reduction units includes: An initial expansion instruction is generated based on the number of expansion units to adjust the number of instances, and an initial reduction instruction is generated based on the number of reduction units to adjust the number of instances. Logical verification is performed on the initial expansion instruction and the initial reduction instruction respectively; The verified initial expansion command and the initial reduction quality are respectively converted into the command recognition format of the service management platform; The scaling instructions for the target system are generated based on the instruction recognition format.

[0057] In detail, the initial expansion command is a raw command generated based on the number of instances to be expanded, without verification or format conversion. The initial reduction command is a raw command generated based on the number of instances to be reduced, without verification or format conversion. The expansion number, adjustment type (expand), and execution time (immediate) are encapsulated into a key-value pair structure, for example, {"adjustType": "expand", "count": 12", "executeTime": "immediate"}. The initial reduction command is generated similarly, encapsulating information including the reduction number, adjustment type (reduction), and execution time (immediate), for example, {"adjustType": "shrink", "count": 3", "executeTime": "immediate"}. The key-value pair structure of the command is checked for completeness and data type correctness (e.g., the number of instances is a positive integer); parameter rationality is verified, checking whether the expansion number exceeds the system's maximum instance limit and whether the reduction number causes the number of instances to fall below the minimum operating requirement; operational feasibility is verified, checking whether the server management platform is operating normally and whether there are sufficient physical resources to support the expansion operation. For example, if the initial expansion command has 12 instances and the maximum number of instances in the system is 20, and the server management platform is normal and has sufficient resources, then the verification will pass. If the initial reduction command has 9 instances, the current number of instances is 10, and the number of instances after reduction is 1, which is lower than the minimum limit of 2, then the verification will fail, and the number of instances to be reduced needs to be adjusted. This multi-dimensional verification logic ensures the correctness and feasibility of the command and avoids system failure caused by executing incorrect commands.

[0058] Specifically, assuming the server management platform recognizes commands in JSON format, the verified initial commands are already in JSON format and require no further conversion. If the initial commands are in other formats, they are converted to JSON using a format conversion tool. The converted initial expansion and contraction commands are then merged, and conflicting operations are removed (e.g., when expanding and contracting simultaneously, commands for higher-priority business operations are executed first), generating the final expansion / contraction commands. For example, the final command might be {"expandCount": 12, "shrinkCount": 3, "adjustType": "both", "executeTime": "immediate"}. Through command merging and conflict handling logic, a unified, conflict-free final command is generated.

[0059] Furthermore, the server management platform is a management system used to uniformly manage the running status and resource allocation of all business instances in the target system. It supports operations such as instance creation, deletion, and startup / shutdown. The final scaling instructions are sent to the server management platform's instruction receiving interface via HTTP / HTTPS protocol, using encrypted transmission to ensure instruction security. After sending, it waits for a response confirmation from the platform. If no confirmation is received within 3 seconds, it resends, with a maximum of 3 retries, thus ensuring secure and reliable transmission of scaling instructions. Upon receiving the instruction, the server management platform parses parameters such as adjustment type and number of instances. Based on the number of instances to be expanded, it creates new business instances and allocates corresponding CPU, memory, network card bandwidth, and other resources. In other words, the decision engine sends a scaling request to the service management platform to start the Expand instance. Based on the number of instances to be scaled down, it terminates some idle business instances, releasing their occupied resources; that is, it sends a scaling request to the service management platform to stop the Shrink instance. The instance creation and termination process is implemented through the platform's instance management module. The execution time depends on the number of instances, but the creation / termination time for a single instance does not exceed 10 seconds.

[0060] Furthermore, such as Figure 3The diagram shows the architecture of the scaling service network. Client service requests are transmitted as data streams through the DMZ egress router and DMZ proxy, then to the CORE egress router and general gateway, before being distributed to various service nodes for processing. Simultaneously, the monitoring and acquisition module collects service resource data and transmits it as a signaling stream to the load entropy calculation module for analysis. The analysis results are then passed to the scaling decision engine to generate scaling instructions. These instructions are sent as a signaling stream to the service management platform, which executes the scaling actions to adjust the service nodes, thus forming a closed loop of business processing and dynamic resource management. Specifically: The general gateway receives service traffic, performs load balancing, and integrates the scaling decision engine. The service management platform is responsible for the creation, destruction, and status maintenance of backend service instances. The monitoring and acquisition module is deployed on each backend server, periodically (e.g., every 10 seconds) collecting metrics such as CPU, memory, and network card usage. The load entropy calculation module runs within the general gateway, calculating node entropy and cluster entropy in real time. The scaling decision engine executes scaling logic and issues instructions based on entropy values ​​and configuration parameters. Each module interacts with the other through internal communication protocols (such as gRPC or REST API). The general gateway and the service management platform transmit scaling commands via a secure channel.

[0061] As can be seen, the above scheme improves the accuracy of system state assessment by using node entropy and cluster entropy models, effectively suppressing instantaneous fluctuations; achieves precise elastic response through asymmetric scaling decisions, avoiding resource oscillations; and ensures the service quality of critical businesses through a business priority mechanism, while optimizing resource utilization.

[0062] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0063] In one embodiment, a load entropy-based scaling device 100 is provided, which corresponds one-to-one with the load entropy-based scaling method described in the above embodiments. For example... Figure 4 As shown, the load entropy-based scaling device 100 includes a performance index identification module 101, an entropy value measurement index extraction module 102, a minimum entropy value identification module 103, a scaling response intensity adjustment module 104, a scaling unit count analysis module 105, and a scaling operation module 106. Detailed descriptions of each functional module are as follows: The performance indicator identification module 101 is used to collect the resource utilization rate of each service node in the target system when the target system is detected to be running, and to identify the performance indicator of each service node based on the resource utilization rate. The entropy metric extraction module 102 is used to calculate the real-time node entropy of each service node based on the performance metric, extract the entropy metric from the real-time node entropy, and use the entropy metric as the target entropy of the target system. The minimum entropy value identification module 103 is used to extract the historical node entropy of each service node within a preset historical time range and identify the minimum entropy value among the historical node entropies. The scaling response intensity adjustment module 104 is used to calculate the service cluster entropy of the target system based on the resource utilization rate, and dynamically adjust the scaling response intensity of the target system according to the priority of the preset target services. The scaling up / down unit count analysis module 105 is used to analyze the number of units that can be expanded in the target system through the target entropy, the service cluster entropy, the service traffic of the target system, and the scaling up / down response intensity, and to analyze the number of units that can be reduced in the target system through the minimum entropy value, the service traffic of the target system, and the scaling up / down response intensity. The scaling up / down operation module 106 is used to generate scaling up / down instructions for the target system based on the number of units to be expanded and the number of units to be scaled down, send the scaling up / down instructions to the server management platform, and execute the scaling up / down operation of the target system based on the scaling up / down instructions.

[0064] In one embodiment, the performance metric identification module 101, when performing the identification of the performance metrics of each service node based on the resource utilization rate, is configured to: Based on the resource utilization rate, the CPU utilization rate, memory usage rate, and network card bandwidth utilization rate of each service node are extracted; The weights of CPU utilization, memory usage, and network card bandwidth utilization are configured according to the service operation type of each service node. The performance metrics of each service node are obtained by weighting and summing the weights, CPU utilization, memory usage, and network card bandwidth utilization.

[0065] In one embodiment, the entropy metric extraction module 102, when performing the calculation of the real-time node entropy of each service node based on the performance metric, is used to: Determine the target load threshold for each service node based on the business type of the target system; Configure the sensitivity coefficient of each service node according to the stability status of the target system; The real-time node entropy of each service node is obtained by performing a nonlinear mapping on the performance indicators, the target load threshold, and the sensitivity coefficient.

[0066] In one embodiment, the scaling response intensity adjustment module 104, when performing the calculation of the service cluster entropy of the target system based on the resource utilization rate, is used to: Based on the resource utilization rate, calculate the total resource usage and total resource capacity of all service nodes in the target system, and calculate the ratio between the total resource usage and the total resource capacity; Obtain the number of running instances of the target system, and determine the idle sensitivity coefficient of the target system based on the number of running instances; The product of the number of running instances and the instance idle sensitivity coefficient is exponentialized, and the preset benchmark value in the target system is subtracted from the exponentialized result to obtain the target value. The service cluster entropy of the target system is obtained by weighting the ratio with the target value.

[0067] In one embodiment, the scaling response strength adjustment module 104, when dynamically adjusting the scaling response strength of the target system according to the priority of a preset target service, is further configured to: Identify the business type of the target business and query the priority corresponding to the business type in a predefined business priority mapping table; Identify the target level corresponding to the priority; when the target level is a preset first level, determine the expansion response strength of the target system based on the target level. When the target level is a preset second level, the scaling-down response strength of the target system is determined according to the target level.

[0068] In one embodiment, the scaling up / down unit count analysis module 105, when performing the analysis of the scaling up unit count of the target system using the target entropy, the service cluster entropy, the service traffic of the target system, and the scaling up / down response strength, is used to: When the target entropy is less than or equal to a preset first security threshold, or the service cluster entropy is greater than or equal to a preset second security threshold, or the trend of the business traffic of the target system is decreasing or stable, the number of running instances of the target system remains unchanged. When the target entropy is greater than a preset first security threshold, the service cluster entropy is less than a preset second security threshold, and the trend of the business traffic of the target system is continuously increasing, the risk sensitivity coefficient of the target system is determined according to the expansion response degree in the expansion and contraction response strength. Based on the risk sensitivity coefficient, the target entropy and the safety threshold are exponentially calculated to obtain a first value; The first value, the expansion response strength, and the number of real-time running instances of the target system obtained in advance are added to the business value to obtain the second value; The second value is rounded up to obtain the number of units required to expand the target system.

[0069] In one embodiment, the scaling-up / scaling-down unit count analysis module 105, when performing the analysis of the scaling-down unit count of the target system using the minimum entropy value, the service traffic of the target system, and the scaling-up / scaling response strength, is further configured to: When the minimum entropy value is greater than or equal to a preset first security threshold or the business traffic of the target system shows an upward trend, the number of running instances of the target system remains unchanged. When the minimum entropy value is less than the preset first security threshold and the trend of the service traffic of the target system is decreasing or stabilizing, the scaling down idle sensitivity coefficient of the target system is determined according to the scaling down response degree in the scaling up and down response strength. Based on the shrinkage idle sensitivity coefficient, the minimum entropy value and the security threshold are exponentially calculated to obtain a third value; A fourth value is obtained by adding a service bonus to the third value and the scaling-down response strength; Select the minimum value between the second value and the preset maximum scaling ratio, and multiply the minimum value by the number of real-time running instances of the target system obtained in advance to obtain the fifth value; The fifth value is rounded down to obtain the number of units in the reduced-capacity target system.

[0070] This invention provides a load entropy-based scaling device that improves the accuracy of system state assessment through node entropy and cluster entropy models, effectively suppressing instantaneous fluctuations; achieves precise elastic response through asymmetric scaling decisions, avoiding resource oscillations; and ensures the service quality of critical services through a business priority mechanism, while optimizing resource utilization.

[0071] Specific limitations regarding load entropy-based scaling devices can be found in the limitations of load entropy-based scaling methods described above, and will not be repeated here. Each module in the aforementioned load entropy-based scaling device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0072] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 5As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements server-side functions or steps of a load entropy-based scaling method.

[0073] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements client-side functions or steps of a load entropy-based scaling method.

[0074] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: When the target system is detected to be running, the resource utilization rate of each service node in the target system is collected, and the performance indicators of each service node are identified based on the resource utilization rate. Calculate the real-time node entropy of each service node based on the performance indicators, extract the entropy value metric from the real-time node entropy, and use the entropy value metric as the target entropy of the target system. Extract the historical node entropy of each service node within a preset historical time range, and identify the minimum entropy value among the historical node entropies; The service cluster entropy of the target system is calculated based on the resource utilization rate, and the scaling response intensity of the target system is dynamically adjusted according to the priority of the preset target services. The number of expansion units in the target system is analyzed using the target entropy, the service cluster entropy, the service traffic of the target system, and the scaling-up / scaling response intensity; the number of scaling-down units in the target system is analyzed using the minimum entropy value, the service traffic of the target system, and the scaling-up / scaling response intensity. Based on the number of expansion units and the number of reduction units, the expansion / reduction instructions for the target system are generated and sent to the server management platform. The expansion / reduction operations for the target system are then executed based on the expansion / reduction instructions.

[0075] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: When the target system is detected to be running, the resource utilization rate of each service node in the target system is collected, and the performance indicators of each service node are identified based on the resource utilization rate. Calculate the real-time node entropy of each service node based on the performance indicators, extract the entropy value metric from the real-time node entropy, and use the entropy value metric as the target entropy of the target system. Extract the historical node entropy of each service node within a preset historical time range, and identify the minimum entropy value among the historical node entropies; The service cluster entropy of the target system is calculated based on the resource utilization rate, and the scaling response intensity of the target system is dynamically adjusted according to the priority of the preset target services. The number of expansion units in the target system is analyzed using the target entropy, the service cluster entropy, the service traffic of the target system, and the scaling-up / scaling response intensity; the number of scaling-down units in the target system is analyzed using the minimum entropy value, the service traffic of the target system, and the scaling-up / scaling response intensity. Based on the number of expansion units and the number of reduction units, the expansion / reduction instructions for the target system are generated and sent to the server management platform. The expansion / reduction operations for the target system are then executed based on the expansion / reduction instructions.

[0076] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0077] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0078] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0079] It should be noted that if any software tools or components not belonging to our company appear in the embodiments of this application, they are merely for illustrative purposes and do not represent actual use.

[0080] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A scaling method based on load entropy, characterized in that, include: When the target system is detected to be running, the resource utilization rate of each service node in the target system is collected, and the performance indicators of each service node are identified based on the resource utilization rate. Calculate the real-time node entropy of each service node based on the performance indicators, extract the entropy value metric from the real-time node entropy, and use the entropy value metric as the target entropy of the target system. Extract the historical node entropy of each service node within a preset historical time range, and identify the minimum entropy value among the historical node entropies; The service cluster entropy of the target system is calculated based on the resource utilization rate, and the scaling response intensity of the target system is dynamically adjusted according to the priority of the preset target services. The number of expansion units in the target system is analyzed using the target entropy, the service cluster entropy, the service traffic of the target system, and the scaling-up / scaling response intensity; the number of scaling-down units in the target system is analyzed using the minimum entropy value, the service traffic of the target system, and the scaling-up / scaling response intensity. Based on the number of expansion units and the number of reduction units, the expansion / reduction instructions for the target system are generated and sent to the server management platform. The expansion / reduction operations for the target system are then executed based on the expansion / reduction instructions.

2. The scaling method based on load entropy as described in claim 1, characterized in that, The step of identifying the performance metrics of each service node based on the resource utilization rate includes: Based on the resource utilization rate, the CPU utilization rate, memory usage rate, and network card bandwidth utilization rate of each service node are extracted; The weights of CPU utilization, memory usage, and network card bandwidth utilization are configured according to the service operation type of each service node. The performance metrics of each service node are obtained by weighting and summing the weights, CPU utilization, memory usage, and network card bandwidth utilization.

3. The scaling method based on load entropy as described in claim 1, characterized in that, The step of calculating the real-time node entropy of each service node based on the performance metrics includes: Determine the target load threshold for each service node based on the business type of the target system; Configure the sensitivity coefficient of each service node according to the stability status of the target system; The real-time node entropy of each service node is obtained by performing a nonlinear mapping on the performance indicators, the target load threshold, and the sensitivity coefficient.

4. The scaling method based on load entropy as described in claim 1, characterized in that, The step of calculating the service cluster entropy of the target system based on the resource utilization rate includes: Based on the resource utilization rate, calculate the total resource usage and total resource capacity of all service nodes in the target system, and calculate the ratio between the total resource usage and the total resource capacity; Obtain the number of running instances of the target system, and determine the idle sensitivity coefficient of the target system based on the number of running instances; The product of the number of running instances and the instance idle sensitivity coefficient is exponentialized, and the preset benchmark value in the target system is subtracted from the exponentialized result to obtain the target value. The service cluster entropy of the target system is obtained by weighting the ratio with the target value.

5. The scaling method based on load entropy as described in claim 1, characterized in that, The step of dynamically adjusting the scaling response strength of the target system according to the priority of the preset target services includes: Identify the business type of the target business and query the priority corresponding to the business type in a predefined business priority mapping table; Identify the target level corresponding to the priority; when the target level is a preset first level, determine the expansion response strength of the target system based on the target level. When the target level is a preset second level, the scaling-down response strength of the target system is determined according to the target level.

6. The scaling method based on load entropy as described in claim 1, characterized in that, The analysis of the number of expansion units in the target system using the target entropy, the service cluster entropy, the service traffic of the target system, and the scaling response strength includes: When the target entropy is less than or equal to a preset first security threshold, or the service cluster entropy is greater than or equal to a preset second security threshold, or the trend of the business traffic of the target system is decreasing or stable, the number of running instances of the target system remains unchanged. When the target entropy is greater than a preset first security threshold, the service cluster entropy is less than a preset second security threshold, and the trend of the business traffic of the target system is a continuous upward trend, the risk sensitivity coefficient of the target system is determined according to the expansion response degree in the expansion and contraction response strength. Based on the risk sensitivity coefficient, the target entropy and the safety threshold are exponentially calculated to obtain a first value; The first value, the expansion response strength, and the number of real-time running instances of the target system obtained in advance are added to the business value to obtain the second value; The second value is rounded up to obtain the number of units required to expand the target system.

7. The scaling method based on load entropy as described in claim 1, characterized in that, The step of analyzing the number of units to be scaled down in the target system using the minimum entropy value, the service traffic of the target system, and the scaling-up / scaling response strength includes: When the minimum entropy value is greater than or equal to a preset first security threshold or the business traffic of the target system shows an upward trend, the number of running instances of the target system remains unchanged. When the minimum entropy value is less than the preset first security threshold and the trend of the service traffic of the target system is decreasing or stabilizing, the scaling down idle sensitivity coefficient of the target system is determined according to the scaling down response degree in the scaling up and down response strength. Based on the shrinkage idle sensitivity coefficient, the minimum entropy value and the security threshold are exponentially calculated to obtain a third value; A fourth value is obtained by adding a service bonus to the third value and the scaling-down response strength; Select the minimum value between the second value and the preset maximum scaling ratio, and multiply the minimum value by the number of real-time running instances of the target system obtained in advance to obtain the fifth value; The fifth value is rounded down to obtain the number of units in the reduced-capacity target system.

8. A capacity scaling device based on load entropy, characterized in that, include: The performance indicator identification module is used to collect the resource utilization rate of each service node in the target system when the target system is detected to be running, and to identify the performance indicator of each service node based on the resource utilization rate. The entropy metric extraction module is used to calculate the real-time node entropy of each service node based on the performance metric, extract the entropy metric from the real-time node entropy, and use the entropy metric as the target entropy of the target system. The minimum entropy value identification module is used to extract the historical node entropy of each service node within a preset historical time range and identify the minimum entropy value among the historical node entropies. The scaling response intensity adjustment module is used to calculate the service cluster entropy of the target system based on the resource utilization rate, and dynamically adjust the scaling response intensity of the target system according to the priority of the preset target services. The scaling up / down unit count analysis module is used to analyze the number of units that can be expanded in the target system using the target entropy, the service cluster entropy, the service traffic of the target system, and the scaling up / down response intensity, and to analyze the number of units that can be reduced in the target system using the minimum entropy value, the service traffic of the target system, and the scaling up / down response intensity. The scaling up / down operation module is used to generate scaling up / down instructions for the target system based on the number of units to be expanded and the number of units to be scaled down, send the scaling up / down instructions to the server management platform, and execute the scaling up / down operation of the target system based on the scaling up / down instructions.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the load entropy-based scaling method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the load entropy-based scaling method as described in any one of claims 1 to 7.

Images