A cloud platform capacity expansion and contraction method, device, equipment and medium

Abstract

The application discloses a cloud platform capacity expansion and reduction method, device and equipment and a medium, and relates to the technical field of computers.The method comprises the following steps: determining a trend coefficient and a load fluctuation rate according to target resource data of each virtual machine in a cloud platform and a sliding window size; determining a prediction error rate and a service priority; adjusting the sliding window size according to the trend coefficient, the load fluctuation rate and the prediction error rate to obtain an adjusted sliding window size; the prediction error rate represents a deviation percentage between a historical prediction value and an actual value of the target resource data; determining each resource demand prediction value according to the target resource data and the adjusted sliding window size; judging a to-be-executed operation of the cloud platform by using a preset threshold value determined according to the service priority and the resource demand prediction value; the to-be-executed operation comprises a capacity expansion operation and a capacity reduction operation; and performing corresponding capacity expansion and reduction operations on the cloud platform according to the resource demand prediction value based on a judgment result.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, device, and medium for scaling up and down a cloud platform. Background Technology

[0002] With the rapid development of cloud computing technology, more and more enterprises and individuals are deploying applications on cloud platforms. Elastic scaling methods for cloud platforms are mainly divided into two categories: threshold-based methods and prediction-based methods. Threshold-based methods set a threshold for resource usage; when the actual resource utilization exceeds the threshold, scaling up is triggered, and when it falls below the threshold, scaling down is triggered. This method is simple to implement, but it suffers from a lag problem, often resulting in performance bottlenecks or resource waste by the time scaling up or down is triggered.

[0003] Predictive-based methods analyze historical resource usage data to predict future resource demand and perform scaling up or down operations in advance. Commonly used predictive algorithms include time series analysis and machine learning. However, they have the following limitations:

[0004] The parameter adjustment dimension is singular: the sliding window size (N) is adjusted only based on the volatility of resource data, which cannot adapt to the trend changes in load (such as continuous growth or decline); the scaling thresholds (α, β) are fixed ranges, which cannot distinguish the differentiated needs of core and non-core businesses; the margin coefficient (γ) lacks a correlation mechanism with the prediction error, which can easily lead to insufficient scaling or waste of resources.

[0005] The scaling-up and scaling-down strategies lack refinement: scaling up is based solely on static margin coefficients, without considering the acceleration of load growth, making it difficult to cope with sudden traffic peaks; scaling down does not distinguish between core and non-core resource nodes, which may affect the stability of critical services; the frequency limiting mechanism is fixed and cannot be dynamically adjusted according to the intensity of load fluctuations, resulting in response delays or excessive system disturbances.

[0006] It is evident that how to integrate dynamic parameter adjustment with refined strategies to achieve flexible scaling is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a cloud platform scaling method, apparatus, device, and medium that integrates dynamic parameter adjustment and refined strategies to achieve elastic scaling, thereby improving resource utilization and reducing costs. The specific solution is as follows:

[0008] In a first aspect, the present invention discloses a method for scaling up and down a cloud platform, comprising:

[0009] The trend coefficient and load volatility are determined based on the target resource data and sliding window size of each virtual machine in the cloud platform; the target resource data includes CPU utilization, memory usage, disk I / O rate, and network bandwidth utilization; the trend coefficient is a coefficient that characterizes the changing trend of the simple moving average value.

[0010] Determine the prediction error rate and business priority, and adjust the sliding window size based on the trend coefficient, load volatility, and prediction error rate to obtain the adjusted sliding window size; the prediction error rate represents the percentage deviation between the historical predicted values ​​and actual values ​​of the target resource data;

[0011] Based on the target resource data and the adjusted sliding window size, the predicted value of each resource demand is determined. The cloud platform's pending operations are determined by the preset threshold determined based on business priorities and the predicted value of each resource demand. The pending operations include expansion operations and shrinkage operations.

[0012] Based on the corresponding judgment results, the cloud platform is subjected to corresponding scaling up or down operations according to the predicted resource demand.

[0013] Optionally, the trend coefficient and load volatility are determined based on the target resource data of each virtual machine in the cloud platform and the sliding window size, including:

[0014] The simple moving average value of each target resource data is determined based on the sliding window size;

[0015] Determine the first difference between the first and last simple moving average values ​​in a target number of consecutive simple moving average values;

[0016] Determine the sampling period, the sliding window size, and the first product between the target value and the target value; the target value is 2.

[0017] The trend coefficient corresponding to each target resource data is determined based on the ratio between the first difference and the first product.

[0018] The load volatility is determined based on the ratio between the standard deviation and the mean of each target resource data.

[0019] Accordingly, business priorities are determined, including:

[0020] Business priorities are determined based on business priority labels pre-configured via the application programming interface.

[0021] Optionally, the sliding window size can be adjusted based on the trend coefficient, load volatility, and prediction error rate to obtain the adjusted sliding window size, including:

[0022] Calculate the average trend coefficient based on the trend coefficient corresponding to each target resource data;

[0023] Calculate the average load volatility based on the load volatility corresponding to each target resource data;

[0024] Calculate the average prediction error rate based on the prediction error rate corresponding to each target resource data;

[0025] If the average load volatility is greater than the first threshold, the sliding window size is reduced; if the average load volatility is less than the second threshold, the sliding window size is increased to obtain the first sliding window size.

[0026] If the average trend coefficient is greater than the third threshold, the size of the first sliding window is reduced; if the average trend coefficient is less than the fourth threshold, the size of the first sliding window is increased to obtain the size of the second sliding window.

[0027] If the average prediction error rate for a consecutive preset number of times is greater than the fifth threshold, the size of the second sliding window is reduced; if the average prediction error rate for a consecutive preset number of times is less than the sixth threshold, the size of the second sliding window is increased to obtain the adjusted sliding window size.

[0028] Optionally, the predicted value of each resource demand is determined based on the target resource data and the adjusted sliding window size, including:

[0029] Determine the sum of the target resource data at each time point;

[0030] The ratio between the sum and the adjusted sliding window size is determined as the resource demand forecast.

[0031] Optionally, the pending operations on the cloud platform can be determined based on preset thresholds set according to business priorities and predicted values ​​of various resource demands, including:

[0032] The overall load is obtained by weighting and summing the predicted resource demand values ​​and weight values ​​corresponding to each target resource data.

[0033] If the predicted resource demand or overall load of any target resource data exceeds the first preset threshold determined according to business priority, the operation to be executed by the cloud platform is determined to be an expansion operation.

[0034] If the predicted resource demand and overall load of each target resource data are both less than the second preset threshold determined according to business priority, then the operation to be executed by the cloud platform is determined to be a scaling-down operation.

[0035] Optionally, based on the corresponding judgment results and the predicted resource demand, the cloud platform may be scaled up or down accordingly, including:

[0036] If the operation to be performed on the cloud platform is a capacity expansion operation, the load growth acceleration is determined based on the ratio between the second difference and the sampling period; the second difference is the difference between the simple moving average values ​​of any two consecutive target resource data.

[0037] Determine the margin coefficient based on the load growth acceleration;

[0038] The pre-increment is determined by multiplying the resource demand forecast, the load growth acceleration, and the preset proportional coefficient.

[0039] Determine the second product between the resource demand forecast and the margin factor;

[0040] Determine the third difference between the second product and the current capacity, and determine the target sum between the third difference and the pre-increment;

[0041] The expansion capacity is determined by the maximum value between the target and the preset minimum expansion unit corresponding to the target resource data;

[0042] The largest expansion capacity among the expansion capacities corresponding to each target resource data is determined as the target expansion capacity;

[0043] Perform corresponding capacity expansion operations on the cloud platform based on the target capacity expansion;

[0044] If the operation to be performed on the cloud platform is a scaling-down operation, the scaling-down capacity is determined based on the predicted resource demand.

[0045] The smallest reduction capacity among the reduction capacities corresponding to each target resource data is determined as the target reduction capacity;

[0046] Core nodes are marked according to the service call chain; core nodes are those that carry database, gateway, and authentication services.

[0047] Perform corresponding scaling-down operations on non-core nodes in the cloud platform according to the target scaling-down.

[0048] Optionally, before performing corresponding scaling operations on the cloud platform based on the resource demand forecast based on the corresponding judgment results, the following steps are also included:

[0049] The maximum number of scaling up and down operations per unit time is determined based on the load volatility, so that the cloud platform can perform corresponding scaling up and down operations based on the resource demand forecast value according to the corresponding judgment result based on the maximum number of scaling up and down operations.

[0050] Secondly, the present invention discloses a cloud platform scaling device, comprising:

[0051] The coefficient and volatility determination module is used to determine the trend coefficient and load volatility based on the target resource data and sliding window size of each virtual machine in the cloud platform; the target resource data includes CPU utilization, memory usage, disk I / O rate, and network bandwidth utilization; the trend coefficient is a coefficient that characterizes the changing trend of the simple moving average value;

[0052] The sliding window size adjustment module is used to determine the prediction error rate and business priority. It adjusts the sliding window size based on the trend coefficient, load volatility, and prediction error rate to obtain the adjusted sliding window size. The prediction error rate represents the percentage deviation between the historical predicted value and the actual value of the target resource data.

[0053] The judgment module is used to determine the predicted value of each resource demand based on the target resource data and the adjusted sliding window size. It determines the pending operations of the cloud platform by using a preset threshold determined by business priority and the predicted value of each resource demand. The pending operations include expansion operations and shrinkage operations.

[0054] The scaling up / down operation execution module is used to perform corresponding scaling up / down operations on the cloud platform based on the corresponding judgment results and the predicted value of resource demand.

[0055] Thirdly, the present invention discloses an electronic device, comprising:

[0056] Memory, used to store computer programs;

[0057] A processor is used to execute computer programs to implement cloud platform scaling methods as described above.

[0058] Fourthly, the present invention discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the cloud platform scaling method as described above.

[0059] In this invention, firstly, a trend coefficient and load volatility are determined based on the target resource data of each virtual machine in the cloud platform and the sliding window size. The target resource data includes CPU utilization, memory usage, disk I / O rate, and network bandwidth utilization. The trend coefficient is a coefficient representing the changing trend of a simple moving average value. The prediction error rate and business priority are determined, and the sliding window size is adjusted based on the trend coefficient, load volatility, and prediction error rate to obtain the adjusted sliding window size. The prediction error rate represents the percentage deviation between the historical predicted values ​​and actual values ​​of the target resource data. Based on the target resource data and the adjusted sliding window size, the predicted values ​​of each resource demand are determined. The pending operations of the cloud platform are judged by using a preset threshold determined based on business priority and the predicted values ​​of each resource demand. The pending operations include expansion and contraction operations. Based on the corresponding judgment results, the cloud platform is subjected to corresponding expansion and contraction operations according to the predicted resource demand values.

[0060] Beneficial Effects: This invention achieves intelligent adaptation of the sliding window size by adjusting it from multiple dimensions, including volatility, trend coefficient, and error rate, thus solving the technical challenge of balancing prediction accuracy and real-time performance with fixed windows. Based on business priorities, a differentiated threshold strategy is constructed, achieving precise matching between resource management strategies and business characteristics, overcoming the shortcomings of traditional fixed thresholds that cannot meet diverse business needs. Furthermore, through multi-dimensional parameter optimization, load-characteristic-driven decision logic, and business-aware resource scheduling, the intelligent management level of the cloud platform is improved. It can integrate dynamic parameter adjustment and refined strategies to achieve elastic scaling, thereby improving resource utilization and reducing costs. Attached Figure Description

[0061] To more clearly illustrate the embodiments of the present invention, the accompanying drawings used in the embodiments 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.

[0062] Figure 1 A flowchart of a cloud platform scaling method provided in an embodiment of the present invention;

[0063] Figure 2 A flowchart illustrating a specific cloud platform scaling method provided in this embodiment of the invention.

[0064] Figure 3 This is a schematic diagram of a cloud platform scaling device provided in an embodiment of the present invention;

[0065] Figure 4 This is a structural diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0067] The terms "comprising" and "having," and any variations thereof, in the specification and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may include steps or units not listed.

[0068] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0069] Predictive methods analyze historical resource usage data to predict future resource demands and perform scaling up or down operations in advance. Commonly used prediction algorithms include time series analysis and machine learning. However, they have the following limitations: Limited parameter adjustment dimensions: The sliding window size (N) is adjusted only based on resource data volatility, failing to adapt to trend changes in load (such as continuous growth or decline); Scaling up / down thresholds (α, β) are fixed ranges, unable to differentiate between the differentiated needs of core and non-core businesses; The margin coefficient (γ) lacks a correlation mechanism with prediction errors, easily leading to insufficient scaling up or resource waste. Insufficient granularity in scaling up / down strategies: Scaling up is based solely on static margin coefficients, without considering load growth acceleration, making it difficult to cope with sudden traffic peaks; Scaling down does not differentiate between core and non-core resource nodes, potentially affecting the stability of critical services; Fixed frequency limiting mechanisms cannot be dynamically adjusted according to load fluctuation intensity, leading to response lag or excessive system disturbance. To address these technical problems, this invention discloses a cloud platform scaling up / down method, apparatus, equipment, and medium that integrates dynamic parameter adjustment and granular strategies to achieve elastic scaling up / down, thereby improving resource utilization and reducing costs.

[0070] See Figure 1 As shown, this embodiment of the invention provides a cloud platform scaling method, including:

[0071] Step S11: Determine the trend coefficient and load volatility based on the target resource data and sliding window size of each virtual machine in the cloud platform; the target resource data includes CPU utilization, memory usage, disk input / output rate, and network bandwidth utilization; the trend coefficient is a coefficient that characterizes the changing trend of the simple moving average value.

[0072] In this embodiment of the invention, key performance indicators of each virtual machine in the cloud platform are first collected, including but not limited to CPU (Central Processing Unit) utilization, memory usage, disk I / O (input / output) rate, and network bandwidth utilization. The sampling period Δt can be dynamically adjusted within the range of 1-60 seconds (default 5-30 seconds). Next, the simple moving average value of each target resource data is determined based on the sliding window size; the first difference between the first and last simple moving average values ​​in a target number of consecutive simple moving average values ​​is determined; the first product between the sampling period, the sliding window size, and the target value is determined; the target value is 2; the trend coefficient corresponding to each target resource data is determined based on the ratio between the first difference and the first product; and the load volatility is determined based on the ratio between the standard deviation and the mean of each target resource data. Specifically, the formula for the slope of the SMA (Simple Moving Average) value change over three consecutive sliding windows is: Trend coefficient k = ( - ) / (2×Δt×N). Where N is the sliding window size; Δt is the sampling period; The value of the first simple moving average in a 3-sliding-window SMA; This represents the last simple moving average value in the three sliding window SMAs. The ratio of the standard deviation to the mean of the data within the current window is the load volatility. It should be noted that the SMA, trend coefficient (k), volatility (σ), and prediction error rate (e) need to be calculated separately for CPU, memory, disk I / O rate, and network bandwidth utilization, respectively.

[0073] Step S12: Determine the prediction error rate and business priority. Adjust the sliding window size based on the trend coefficient, load volatility, and prediction error rate to obtain the adjusted sliding window size. The prediction error rate represents the percentage deviation between the historical predicted value and the actual value of the target resource data.

[0074] In this embodiment of the invention, the prediction error rate is the percentage deviation between historical predicted values ​​and actual values, calculated as: Error rate e = |Predicted value - Actual value| / Actual value × 100%. Furthermore, business priorities are determined based on pre-configured business priority tags via an application programming interface (API). That is, businesses are categorized into core businesses (such as payment systems and transaction platforms) and non-core businesses (such as log analysis and backup services) based on their importance. These tags can be dynamically configured via the API. The collected data and feature values ​​are then stored in a time-series database (such as InfluxDB or TimescaleDB) to provide a basis for subsequent parameter adjustments and decision-making.

[0075] Then, when adjusting the sliding window size, the average trend coefficient is calculated based on the trend coefficient corresponding to each target resource data; the average load volatility is calculated based on the load volatility corresponding to each target resource data; the average prediction error rate is calculated based on the prediction error rate corresponding to each target resource data; if the average load volatility is greater than the first threshold, the sliding window size is reduced; if the average load volatility is less than the second threshold, the sliding window size is increased to obtain the first sliding window size; if the average trend coefficient is greater than the third threshold, the first sliding window size is reduced; if the average trend coefficient is less than the fourth threshold, the first sliding window size is increased to obtain the second sliding window size; if the average prediction error rate for a consecutive preset number of times is greater than the fifth threshold, the second sliding window size is reduced; if the average prediction error rate for a consecutive preset number of times is less than the sixth threshold, the second sliding window size is increased to obtain the adjusted sliding window size. Specifically, the initial value of the sliding window size N is 20 (range 5-100), and it is dynamically adjusted based on the following multi-dimensional conditions: Basic adjustment logic: When volatility σ > 20%, N decreases by 10%-20% (enhancing sensitivity to the latest data); when σ < 10%, N increases by 10%-20% (smoothing noise interference). Trend adaptation adjustment: If the trend coefficient k > 0.05 for 5 consecutive windows (rapid increase in load), N decreases by an additional 10% regardless of whether volatility is reached or not; if k < -0.05 (rapid decrease in load), N increases by an additional 10%. Error feedback adjustment: When the prediction error rate e > 25% for 3 consecutive predictions, N decreases by 5%-10% (improving prediction real-time performance); when e < 5%, N increases by 5%-10% (improving prediction stability). Furthermore, the adjusted N value must satisfy 5 ≤ N ≤ 100, and the single adjustment amplitude should not exceed 30% to avoid drastic parameter fluctuations.

[0076] This approach breaks through the limitations of traditional methods that adjust window size solely based on volatility, creatively integrating multiple dimensions such as load trends and prediction error rates to achieve intelligent adaptation of the sliding window size (N). When the load increases rapidly (k>0.05) or the prediction error is too large (e>25%), the window size is automatically reduced to enhance response sensitivity; when the load is stable or the prediction is accurate, the window size is increased to smooth noise, solving the technical challenge of balancing prediction accuracy and real-time performance with a fixed window.

[0077] Step S13: Determine the predicted value of each resource demand based on the target resource data and the adjusted sliding window size. Determine the cloud platform's pending operations by using the preset threshold determined based on business priority and the predicted value of each resource demand. The pending operations include expansion operations and shrinkage operations.

[0078] In this embodiment of the invention, when determining the predicted value of each resource demand based on the target resource data and the adjusted sliding window size, the sum of the target resource data at each time point is determined; the ratio between the sum and the adjusted sliding window size is determined as the predicted value of resource demand. In this process, the simple moving average is used as the prediction basis, with the formula: SMA(t) = (x(t-N+1) + x(t-N+2) + ... + x(t)) / N; where x(ti) is the resource data at time ti, SMA(t) is the predicted value of resource demand for the time period T (T = N × Δt) after time t; and N is the adjusted sliding window size.

[0079] Then, by determining preset thresholds based on business priorities, core businesses use α=60%-70% (proactive scaling up to ensure high availability) and β=40%-50% (conservative scaling down to reduce service risks); non-core businesses use α=80%-85% (improving resource utilization) and β=20%-30% (aggressive scaling down to reduce waste). Dynamic adjustments are made based on time periods: core businesses reduce α by 5%-10% during peak hours (e.g., 9:00-18:00), while non-core businesses reduce β by 5%-10% during off-peak hours (e.g., 0:00-6:00). This differentiated threshold (α, β) strategy based on business priorities ensures high availability with a conservative threshold (α=60%-70%) for core businesses and enhances resource utilization with an aggressive threshold (α=80%-85%) for non-core businesses. Simultaneously, by dynamically adjusting thresholds based on time period characteristics, a precise match between resource management strategies and business characteristics is achieved, overcoming the shortcomings of traditional fixed thresholds that cannot meet diverse business needs.

[0080] Subsequently, a weighted sum is performed based on the predicted resource demand values ​​and weight values ​​corresponding to each target resource data to obtain the comprehensive load. If the predicted resource demand value or comprehensive load of any target resource data is greater than a first preset threshold determined according to business priority, the operation to be executed by the cloud platform is determined to be an expansion operation. If the predicted resource demand value and comprehensive load of each target resource data are both less than a second preset threshold determined according to business priority, the operation to be executed by the cloud platform is determined to be a shrinking operation. Otherwise, no expansion or shrinking operation is performed. In this embodiment of the invention, the SMA, trend coefficient (k), volatility (σ), and prediction error rate (e) of CPU, memory, etc., are calculated in parallel. Then, according to the weighting strategy read from the configuration center, the predicted values ​​of multiple indicators are aggregated into a comprehensive load score S.

[0081] In one specific embodiment, it is assumed that within the same sampling period, not only CPU data but also memory data (in %) are collected. The current window CPU data is [20%, 21%, 22%, 23%, 24%, 25%, 26%, 27%, 28%, 29%, 31%, 31%, 32%, 33%, 34%, 35%, 36%, 37%, 38%, 39%]; the SMA of CPU utilization itself is: =(20%+21%+22%+...+39%) / 20=35%. Current window memory usage sequence: [65%, 66%, 67%, 65%, 68%, 69%, 70%, 68%, 71%, 70%, 72%, 73%, 71%, 74%, 75%, 73%, 76%, 75%, 77%, 78%]; Calculated memory... =71%. Trend coefficient k_m = 0.03% / s (assuming a stronger upward trend than CPU). Load volatility σ_m = 3.5% (memory usage grows very steadily). CPU metrics: =35%, k_c=0.02% / s, σ_c=8.26%; Memory metrics: =71%, k_m=0.03% / s, σ_m=3.5%. Weights are defined for different services in the configuration center. For example: Core service (virtual machine operation): CPU intensive, weight W_c=0.7, memory weight W_m=0.3; Non-core service (log analysis): Memory intensive, weight W_c=0.4, memory weight W_m=0.6. Calculation formula: S=(SMA_c×W_c+SMA_m×W_m)×100%. Finally, S=(35%*0.7)+(71%*0.3)=24.5%+21.3%=45.8%. The overall load is 45.8%, which has not yet reached the core service expansion threshold α (60%). Decisions are made based on the overall load score S and the independent threshold of any indicator. This is an "OR" logic, ensuring that any shortage of critical resources triggers expansion. Arbitration logic:

[0082] If S>=α OR SMA_c>=α OR SMA_m>=α_m (memory-independent threshold) — trigger expansion; (If any of S, CPU utilization, or memory utilization reaches the expansion threshold, expansion is triggered).

[0083] IF S<=β AND SMA_c<=β AND SMA_m<=β_m — Trigger scaling down (scaling down requires all metrics to be met for greater safety); (scaling down is only triggered when S, CPU utilization, and memory utilization all reach the scaling down threshold).

[0084] Therefore, although the overall load S = 45.8% < 60%, the memory metric SMA_m = 71% has exceeded the independent memory expansion threshold for core business applications (assuming α_m = 70%). This ultimately triggers expansion because the condition SMA_m >= α_m is met.

[0085] Step S14: Based on the corresponding judgment results, perform corresponding scaling up and down operations on the cloud platform according to the predicted resource demand.

[0086] In this embodiment of the invention, if the operation to be performed on the cloud platform is a capacity expansion operation, the load growth acceleration is determined based on the ratio between the second difference and the sampling period; the second difference is the difference between the simple moving averages of any two consecutive target resource data; a margin coefficient is determined based on the load growth acceleration; a pre-increment is determined by multiplying the resource demand forecast, the load growth acceleration, and a preset proportional coefficient; a second product is determined between the resource demand forecast and the margin coefficient; a third difference is determined between the second product and the current capacity, and a target sum is determined between the third difference and the pre-increment; the target sum is then compared with the target resource data. The expansion capacity is determined by the maximum value among the corresponding preset minimum expansion units; the largest expansion capacity among the expansion capacities corresponding to each target resource data is determined as the target expansion capacity; the corresponding expansion operation is performed on the cloud platform based on the target expansion capacity; if the operation to be performed on the cloud platform is a shrinkage operation, the shrinkage capacity is determined based on the predicted resource demand; the smallest shrinkage capacity among the shrinkage capacities corresponding to each target resource data is determined as the target shrinkage capacity; core nodes are marked according to the service call chain; core nodes are nodes that carry database, gateway, and authentication services; the corresponding shrinkage operation is performed on non-core nodes in the cloud platform based on the target shrinkage capacity. Specifically, the expansion strategy is as follows: growth acceleration calculation: load growth acceleration a = ( ) / Δt, reflecting the rate of load change; dynamic margin coefficient: γ=1.1+0.05×a (when a≥0), γ [1.1, 1.5], the greater the acceleration, the higher the margin coefficient; Pre-increment addition: if a>0 and increases for two consecutive windows, the pre-increment = predicted value × a × 0.1 (the proportional coefficient 0.1 can be dynamically adjusted); Expansion formula: Expansion resource amount = max((predicted value × γ - current capacity) + pre-increment, minimum expansion unit). CPU requirement: Expansion resource amount _c = max(( ×γ_c-100%)+Pre-increment_c,Minimum expansion unit_c);Memory requirement:Expansion resource amount_m=max(( (×γ_m-100%) + pre-increment_m, minimum expansion unit_m). The final expansion capacity is the greater of the CPU requirement and the memory requirement as the final number of instances to be expanded. Because expansion is performed on a per-virtual machine or Pod basis, all resource demands must be met. In a specific implementation, 2 virtual machines need to be expanded in terms of CPU. In terms of memory: assuming the same logic is used (γ_m=1.1, pre-increment_m=5%), it is found that 3 virtual machines need to be expanded to meet the memory requirement. The system will ultimately take the maximum value and perform the operation of expanding 3 virtual machines, thereby simultaneously meeting the growth requirements of CPU and memory.

[0087] The scaling-down strategy is as follows: Based on service call chain data, nodes carrying databases, gateways, and authentication services are marked as core nodes, and non-core nodes are scaled down first; safety margin setting: the scaled-down capacity ≥ the predicted value × 1.05 (reserving a 5% safety margin); dependency verification: before scaling down, resource dependencies are checked (e.g., whether scaling down a certain node will cause disk I / O overload of related nodes), and if a risk is found, scaling down is paused. During the scaling-down process, scaling down is performed according to the minimum scaling-down capacity. For example, if the CPU calculation shows that 2 virtual machines need to be scaled down, and the memory calculation shows that 2 virtual machines need to be scaled down, then scaling down is performed according to 2 virtual machines to meet business needs.

[0088] In addition, in cloud platform scaling-down scenarios, besides prioritizing the scaling down of non-critical nodes, you can also prioritize scaling down nodes with CPU utilization below 20% and memory idle rate above 60%; or automatically trigger scaling down during business downtime (such as 2-5 am) and predict load cycles through time series analysis; or set podDisruptionBudget for critical Pods to ensure that core services are not affected during scaling down; or adopt batch scaling down (such as scaling down no more than 10% of the total number of nodes each time) in conjunction with health checks to avoid avalanche effects.

[0089] In this embodiment of the invention, before performing corresponding scaling operations on the cloud platform based on the resource demand forecast based on the corresponding judgment results, the maximum number of scaling operations per unit time is first determined based on the load volatility. This allows for the execution of corresponding scaling operations on the cloud platform based on the resource demand forecast based on the corresponding judgment results and the maximum number of scaling operations. In a specific embodiment, the maximum number of scaling operations per unit time (default 10 minutes) is dynamically adjusted according to the volatility σ: σ>30% (violent fluctuation): 5 operations are allowed for rapid response to user needs; 10%≤σ≤30% (moderate fluctuation): 3 operations are allowed; σ<10% (stable state): 1 operation is allowed to reduce system disturbances. Furthermore, this invention can store historical data and operation records, and optimize parameters such as γ and pre-increment ratio through machine learning models. This intelligent frequency limiting mechanism reduces invalid operations by 70% in a stable state and improves response timeliness by 50% during violent fluctuations.

[0090] In this way, the expansion phase innovatively introduces the concept of load growth acceleration (a), and achieves a forward-looking response to the load increase trend through a dynamic margin coefficient (γ=1.1+0.05×a) and a pre-increment mechanism; the shrinkage phase pioneers the identification of core nodes and the setting of safety margin (≥predicted value×1.05), which enables precise release of resources while ensuring service stability, and solves the problems of traditional expansion and shrinkage strategies being slow to respond to load change trends and easily causing service risks.

[0091] Beneficial Effects: This invention achieves intelligent adaptation of the sliding window size by adjusting it from multiple dimensions, including volatility, trend coefficient, and error rate, thus solving the technical challenge of balancing prediction accuracy and real-time performance with fixed windows. Based on business priorities, a differentiated threshold strategy is constructed, achieving precise matching between resource management strategies and business characteristics, overcoming the shortcomings of traditional fixed thresholds that cannot meet diverse business needs. Furthermore, through multi-dimensional parameter optimization, load-characteristic-driven decision logic, and business-aware resource scheduling, the intelligent management level of the cloud platform is improved. It can integrate dynamic parameter adjustment and refined strategies to achieve elastic scaling, thereby improving resource utilization and reducing costs.

[0092] As described in the previous embodiment, this invention discloses a cloud platform scaling method that enables elastic scaling. The cloud platform scaling method will now be described in detail with reference to specific embodiments.

[0093] See Figure 2 As shown, this invention discloses a specific method for scaling up and down a cloud platform. In one specific embodiment, the scenario parameters are:

[0094] Business types: Virtual machine operation (core business), log analysis (non-core business);

[0095] The sampling period Δt = 5 seconds, and the initial window size N = 20;

[0096] Core business α=60%, β=40%; Non-core business α=85%, β=20%;

[0097] Minimum expansion unit = 2 virtual machines, minimum shrink unit = 2GB memory, CPU 2%.

[0098] The data acquisition module collects the memory usage of each virtual machine every 5 seconds, forming time-series data. For example, the time-series CPU utilization of a certain virtual machine is as follows, calculated as follows:

[0099] The current window's CPU usage percentages are [20%, 21%, 22%, 23%, 24%, 25%, 26%, 27%, 28%, 29%, 31%, 31%, 32%, 33%, 34%, 35%, 36%, 37%, 38%, 39%]; the CPU utilization percentages themselves are as follows: =(20%+21%+22%+...+39%) / 20=35%. The data in the first two windows are [10%,10%,11%,11%,12%,12%,13%,13%,14%,14%,15%,16%,16%,17%,17%,18%,18%,19%,19%,20%]; =(10%+10%+11%+...+20%) / 20=31%; Trend coefficient k=(35%-31%) / (2×5×20)=0.02% / s; Volatility μ=34% (mean within the window) Window data: 30%, 31%, 31%, 32%, 33%, 34%, 35%, 36%, 37%, 38%; Prediction error rate e=8%.

[0100] The first three predicted values ​​were 28%, 30%, and 32%; the corresponding actual values ​​were 30%, 31%, and 33%. The error rates are: e1 = |28-30| / 30×100%≈6.67%, e2 = |30-31| / 31×100%≈3.23%, e3 = |32-33| / 33×100%≈3.03%. The current error rate is the average: e = (6.67% + 3.23% + 3.03%) / 3≈4.31%.

[0101] Sliding window adjustment: Trend coefficient k = 0.02% / s > 0.01% / s (increasing load); Volatility σ = 8.26% < 10% (stable fluctuation); Error rate e = 4.31% < 5% (accurate prediction). Because the trend is upward but volatility is low and prediction is accurate, N is increased from 20 to 22 (an increase of 10%) to balance smoothness and trend response.

[0102] The SMA prediction value is 62% of the current capacity (sum of CPU utilization data at 22 time points / N), which exceeds the core business expansion threshold α=60%, triggering expansion. (This is an explanation using a single type of resource data. If there are multiple resource data, the calculation method for other resource data is the same as the above calculation process. When determining whether to trigger expansion, it is necessary to calculate a comprehensive load S based on the predicted values ​​of all resource data. Then, it is necessary to determine the relationship between the predicted values ​​of all resource data, the comprehensive load, and the expansion threshold. If any of the resource data such as S, CPU utilization, or memory utilization reaches the expansion threshold, expansion is triggered.)

[0103] During capacity expansion, the growth acceleration a = (62% - 58%) / 5 = 0.8% / second; the margin coefficient γ = 1.1 + 0.05 × 0.8 = 1.14; the pre-increment = 62% × 0.8 × 0.1 = 4.96%; the expansion resource amount = max((62% × 1.14 - 100%) + 4.96%, 2 virtual machines) = 2 virtual machines. Real-time monitoring is performed during this process. During peak activity periods (volatility σ = 35% > 30%), 5 expansions are allowed within 10 minutes. The system completes 2 expansions based on real-time load to ensure service stability.

[0104] During scaling down, the load decreases (k = -0.015% / s); the current server is identified as a non-core node, and the scaled-down capacity = predicted value × 1.05 = 25% × 1.05 = 26.25%. (Similarly, during scaling down, a comprehensive load S needs to be calculated based on the predicted values ​​of all resource data. Then, the relationship between the predicted values ​​of all resource data, the comprehensive load, and the scaling-up threshold needs to be determined. Scaling down is only triggered if all data in resource data such as S, CPU utilization, and memory utilization reach the scaling-up threshold). Since the load volatility σ = 5% < 10%, scaling down is only performed once every 10 minutes.

[0105] Beneficial Effects: An operation frequency limit rule linked to load volatility (σ) is established. During periods of severe volatility (σ>30%), the limit is relaxed to 5 times / 10 minutes for rapid response, while during periods of stable conditions (σ<10%), it is tightened to 1 time / 10 minutes to reduce disturbances. This mechanism balances system response speed and stability, avoiding the insufficient adaptability of fixed frequency limits in complex load scenarios. Simultaneously, the sliding window size is adjusted across multiple dimensions, including volatility, trend coefficient, and error rate, reducing the prediction error rate by 30%-50%. Dynamic thresholds adapt to business priorities and time-period characteristics, improving core business availability by over 20% and significantly enhancing parameter adaptability. Expansion, combined with growth acceleration and pre-increment mechanisms, improves response speed to sudden load surges by 40%. Downsizing, through core node protection and safety margin settings, reduces service interruption rate by 60%, thereby optimizing expansion and contraction accuracy.

[0106] See Figure 3 As shown, this embodiment of the invention provides a cloud platform scaling device, comprising:

[0107] The coefficient and volatility determination module 11 is used to determine the trend coefficient and load volatility based on the target resource data and sliding window size of each virtual machine in the cloud platform; the target resource data includes CPU utilization, memory usage, disk input / output rate and network bandwidth utilization; the trend coefficient is a coefficient that characterizes the changing trend of the simple moving average value;

[0108] The sliding window size adjustment module 12 is used to determine the prediction error rate and business priority, and adjust the sliding window size according to the trend coefficient, load fluctuation rate and prediction error rate to obtain the adjusted sliding window size; the prediction error rate represents the percentage deviation between the historical predicted value and the actual value of the target resource data;

[0109] The judgment module 13 is used to determine the predicted value of each resource demand based on the target resource data and the adjusted sliding window size, and to determine the pending operations of the cloud platform by using the preset threshold determined by the business priority and the predicted value of each resource demand; the pending operations include expansion operations and shrinkage operations.

[0110] The scaling up / down operation execution module 14 is used to perform corresponding scaling up / down operations on the cloud platform based on the corresponding judgment results and the predicted value of resource demand.

[0111] Since the embodiments of the device part correspond to the embodiments described above, please refer to the embodiments described in the method part for the embodiments of the device part, and will not be repeated here.

[0112] Beneficial Effects: This invention achieves intelligent adaptation of the sliding window size by adjusting it from multiple dimensions, including volatility, trend coefficient, and error rate, thus solving the technical challenge of balancing prediction accuracy and real-time performance with fixed windows. Based on business priorities, a differentiated threshold strategy is constructed, achieving precise matching between resource management strategies and business characteristics, overcoming the shortcomings of traditional fixed thresholds that cannot meet diverse business needs. Furthermore, through multi-dimensional parameter optimization, load-characteristic-driven decision logic, and business-aware resource scheduling, the intelligent management level of the cloud platform is improved. It can integrate dynamic parameter adjustment and refined strategies to achieve elastic scaling, thereby improving resource utilization and reducing costs.

[0113] Furthermore, embodiments of this application also disclose an electronic device, Figure 4 This is a structural diagram of an electronic device according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application. Specifically, the electronic device may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the cloud platform scaling method disclosed in any of the foregoing embodiments. Furthermore, the electronic device in this embodiment may specifically be an electronic computer.

[0114] In this embodiment, the power supply 23 is used to provide operating voltage for various hardware devices on the electronic device; the communication interface 24 can create a data transmission channel between the electronic device and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0115] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0116] The operating system 221 is used to manage and control the various hardware devices on the electronic device and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the cloud platform scaling method executed by the electronic device as disclosed in any of the foregoing embodiments, the computer program 222 may further include a computer program capable of performing other specific tasks.

[0117] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned cloud platform scaling method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0118] Furthermore, this application also discloses a computer program product, including a computer program / instructions; wherein, when the computer program / instructions are executed by a processor, they implement the aforementioned cloud platform scaling method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0119] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0120] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0121] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0122] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0123] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only intended to help understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for scaling up and down a cloud platform, characterized in that, include: The trend coefficient and load volatility are determined based on the target resource data and sliding window size of each virtual machine in the cloud platform. The target resource data includes CPU utilization, memory usage, disk I / O rate, and network bandwidth utilization. The trend coefficient is a coefficient that characterizes the changing trend of the simple moving average line value; Determine the prediction error rate and business priority, and adjust the sliding window size according to the trend coefficient, the load volatility, and the prediction error rate to obtain the adjusted sliding window size; the prediction error rate represents the percentage deviation between the historical predicted value and the actual value of the target resource data; Based on the target resource data and the adjusted sliding window size, the predicted value of each resource demand is determined. The cloud platform's pending operations are determined by a preset threshold determined by the business priority and the predicted value of each resource demand. The pending operations include expansion operations and reduction operations. Based on the corresponding judgment results, the cloud platform is subjected to corresponding scaling up or down operations according to the predicted resource demand values. The step of performing corresponding scaling operations on the cloud platform based on the corresponding judgment results and the predicted resource demand includes: If the operation to be performed on the cloud platform is a capacity expansion operation, the load growth acceleration is determined based on the ratio between the second difference and the sampling period; the second difference is the difference between the simple moving average values ​​of any two consecutive target resource data. The margin coefficient is determined based on the aforementioned load growth acceleration. The pre-increment is determined by multiplying the predicted resource demand, the load growth acceleration, and a preset proportional coefficient. Determine the second product between the predicted resource demand and the margin coefficient; Determine a third difference between the second product and the current capacity, and determine a target sum between the third difference and the pre-increment; The expansion capacity is determined by the maximum value between the target and the preset minimum expansion unit corresponding to the target resource data; The largest expansion capacity among the expansion capacities corresponding to each of the target resource data is determined as the target expansion capacity; Based on the target capacity expansion, the cloud platform is subjected to corresponding capacity expansion operations.

2. The cloud platform scaling method according to claim 1, characterized in that, The determination of trend coefficients and load volatility based on target resource data and sliding window size of each virtual machine in the cloud platform includes: The simple moving average value of each target resource data is determined based on the sliding window size. Determine the first difference between the first and last simple moving average values ​​in a target number of consecutive simple moving average values; Determine the first product between the sampling period, the sliding window size, and the target value; the target value is 2. The trend coefficient corresponding to each target resource data is determined based on the ratio between the first difference and the first product. The load volatility is determined based on the ratio between the standard deviation and the mean of each of the target resource data. Accordingly, business priorities are determined, including: Business priorities are determined based on business priority labels pre-configured via the application programming interface.

3. The cloud platform scaling method according to claim 1, characterized in that, The step of adjusting the sliding window size based on the trend coefficient, the load volatility, and the prediction error rate to obtain the adjusted sliding window size includes: Calculate the average trend coefficient based on the trend coefficient corresponding to each of the target resource data; Calculate the average load volatility based on the load volatility corresponding to each of the target resource data; Calculate the average prediction error rate based on the prediction error rate corresponding to each of the target resource data; If the average load volatility is greater than a first threshold, the sliding window size is reduced; if the average load volatility is less than a second threshold, the sliding window size is increased to obtain the first sliding window size. If the average trend coefficient is greater than the third threshold, the size of the first sliding window is reduced; if the average trend coefficient is less than the fourth threshold, the size of the first sliding window is increased to obtain the size of the second sliding window. If the average prediction error rate for a consecutive preset number of times is greater than the fifth threshold, then the size of the second sliding window is reduced; if the average prediction error rate for a consecutive preset number of times is less than the sixth threshold, then the size of the second sliding window is increased, so as to obtain the adjusted sliding window size.

4. The cloud platform scaling method according to claim 1, characterized in that, The step of determining the predicted value of each resource demand based on the target resource data and the adjusted sliding window size includes: Determine the sum of the target resource data at each time point; The ratio between the sum and the adjusted sliding window size is determined as the predicted resource demand value.

5. The cloud platform scaling method according to claim 1, characterized in that, The step of determining the pending operations of the cloud platform based on a preset threshold determined by the business priority and the predicted values ​​of each resource demand includes: The comprehensive load is obtained by weighting and summing the predicted resource demand values ​​and weight values ​​corresponding to each of the target resource data. If the predicted resource demand value of any of the target resource data or the overall load is greater than the first preset threshold determined according to the business priority, then the operation to be executed by the cloud platform is determined to be an expansion operation. If the predicted resource demand value of each of the target resource data and the overall load are both less than the second preset threshold determined according to the business priority, then the operation to be executed by the cloud platform is determined to be a scaling-down operation.

6. The cloud platform scaling method according to claim 5, characterized in that, The step of performing corresponding scaling operations on the cloud platform based on the corresponding judgment results and the predicted resource demand includes: If the operation to be performed on the cloud platform is a scaling-down operation, the scaling-down capacity is determined based on the predicted resource demand. The smallest reduction capacity among the reduction capacities corresponding to each of the target resource data is determined as the target reduction capacity; Core nodes are marked according to the service call chain; the core nodes are those that carry database, gateway, and authentication services. Perform corresponding scaling-down operations on non-core nodes in the cloud platform according to the target scaling-down.

7. The cloud platform scaling method according to any one of claims 1 to 6, characterized in that, Before performing corresponding scaling operations on the cloud platform based on the resource demand forecast value according to the corresponding judgment result, the method further includes: The maximum number of scaling up and down operations per unit time is determined based on the load volatility, so that the cloud platform can perform corresponding scaling up and down operations based on the resource demand forecast value according to the corresponding judgment result based on the maximum number of scaling up and down operations.

8. A cloud platform scaling device, characterized in that, include: The coefficient and volatility determination module is used to determine the trend coefficient and load volatility based on the target resource data of each virtual machine in the cloud platform and the sliding window size; The target resource data includes CPU utilization, memory usage, disk I / O rate, and network bandwidth utilization. The trend coefficient is a coefficient that characterizes the changing trend of the simple moving average line value; The sliding window size adjustment module is used to determine the prediction error rate and business priority, and adjust the sliding window size according to the trend coefficient, the load volatility rate and the prediction error rate to obtain the adjusted sliding window size; the prediction error rate represents the percentage deviation between the historical predicted value and the actual value of the target resource data; The judgment module is used to determine the predicted value of each resource demand based on the target resource data and the adjusted sliding window size, and to determine the pending operations of the cloud platform based on the preset threshold determined by the business priority and the predicted value of each resource demand; the pending operations include expansion operations and shrinkage operations; The scaling up / down operation execution module is used to perform corresponding scaling up / down operations on the cloud platform based on the corresponding judgment results and the predicted resource demand values. The scaling up / down operation execution module is used to determine the load growth acceleration based on the ratio between the second difference and the sampling period if the operation to be executed on the cloud platform is a scaling up operation; the second difference is the difference between the simple moving average values ​​of any two consecutive target resource data. The margin coefficient is determined based on the load growth acceleration; the pre-increment is determined by the product of the resource demand forecast, the load growth acceleration, and the preset proportional coefficient. Determine the second product between the predicted resource demand and the margin coefficient; Determine the third difference between the second product and the current capacity, and determine the target sum between the third difference and the pre-increment; determine the expansion capacity by the maximum value between the target sum and the preset minimum expansion unit corresponding to the target resource data; The largest expansion capacity among the expansion capacities corresponding to each of the target resource data is determined as the target expansion capacity; Based on the target capacity expansion, the cloud platform is subjected to corresponding capacity expansion operations.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the cloud platform scaling method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the cloud platform scaling method as described in any one of claims 1 to 7.

Images