Soft switch resource flexible scheduling method and device of cloud native architecture, equipment and medium

Abstract

The invention relates to a flexible scheduling method and device for soft switch resources of a cloud native architecture, equipment and a medium. The method comprises the following steps: performing feature extraction and marking on computing nodes, storage resources and network resources by using a resource classification model to generate a resource tag library representing initial classification of each resource, extracting resource feature data from a resource tag library based on each resource initial classification, generating current supply and demand data in combination with resource use state data collected in real time, calculating a resource allocation priority sequence by using a multi-objective optimization algorithm fusing service response time and cost balance optimization parameters, and generating a preliminary sorting result; and detecting a resource supply and demand imbalance state based on the preliminary sorting result and the resource supply and demand change data flow monitored in real time, if the imbalance state is detected, executing resource allocation adjustment calculation based on the current supply and demand data, and generating a preliminary resource allocation scheme, thereby realizing flexible scheduling and optimal allocation of soft switch resources under the cloud native architecture.

Description

Technical Field

[0001] The present invention belongs to the field of communication technology, and in particular relates to a method, device, equipment and medium for elastic scheduling of soft switch resources in a cloud native architecture. Background Art

[0002] With the development of cloud-native architecture in the field of communications technology, there has been a trend toward migrating softswitch systems to cloud-native environments. This technology offers features such as dynamic resource management and elastic service expansion, leading to the development of softswitch resource scheduling methods within cloud-native architectures. Traditionally, softswitch resource scheduling typically utilizes a static or semi-static allocation model, allocating compute nodes, storage resources, and network resources based on preset rules. This approach lacks refined classification of resource characteristics and dynamic awareness of real-time supply and demand status. However, this current resource scheduling approach, or traditional method, suffers from inefficient resource allocation, difficulty balancing service response time and cost consumption, and an inability to respond to resource supply and demand imbalances in real time. These factors lead to insufficient resource utilization and limited elastic scheduling capabilities within cloud-native architectures. Summary of the Invention

[0003] Based on this, it is necessary to provide a method, device, equipment and medium for elastic scheduling of soft switch resources in a cloud native architecture that can solve the above technical problems.

[0004] In a first aspect, the present application provides a method for elastically scheduling soft switch resources in a cloud-native architecture, comprising:

[0005] The resource classification model is used to extract and label the features of computing nodes, storage resources, and network resources to generate a resource tag library. The resource tag library is used to represent the initial classification of each resource.

[0006] Based on the initial classification of each resource, the corresponding resource feature data is extracted from the resource tag library, and combined with the real-time collected resource usage status data, the current supply and demand data of each resource is generated;

[0007] Based on current supply and demand data, a multi-objective optimization algorithm that integrates business response time and cost balance optimization parameters is used to calculate the priority order of resource allocation and generate preliminary ranking results for resource scheduling;

[0008] Detect resource supply and demand imbalances based on preliminary sorting results and real-time monitoring of resource supply and demand change data streams;

[0009] If a supply-demand imbalance is detected, resource allocation adjustment calculations are performed based on the current supply-demand data to generate a preliminary resource allocation plan.

[0010] In one embodiment, after generating the preliminary resource allocation plan, the method further includes:

[0011] Obtain load data of target resources according to the preliminary resource allocation plan;

[0012] Extract features from the load data of target resources and construct a multi-dimensional feature vector that characterizes the dynamic characteristics of load distribution;

[0013] Based on the multidimensional feature vector, the reinforcement learning algorithm is used to predict the future load demand trend of the target resource and generate resource demand trend prediction information.

[0014] In one embodiment, after generating the resource demand trend forecast information, the method further includes:

[0015] Based on resource demand trend forecast information, obtain a multi-dimensional operating status indicator set for each computing node;

[0016] Using the pre-trained performance bottleneck classification model, based on a multi-dimensional set of operating status indicators, we generate the identifiers of computing nodes with resource bottlenecks and their corresponding resource bottleneck category feature vectors.

[0017] If the resource bottleneck category feature vector exceeds the preset threshold range, load balancing is performed on the computing nodes with resource bottlenecks to generate a node resource allocation plan.

[0018] In one embodiment, after generating the node resource allocation plan, the method further includes:

[0019] Obtain real-time availability data of storage resources and network resources associated with the node resource allocation plan;

[0020] Through the distributed scheduling framework, storage resources and network resources are coordinated, matched and allocated based on the preset storage and network resource allocation strategies to generate an optimized resource allocation plan.

[0021] In one embodiment, after generating the optimized resource allocation plan, the method further includes:

[0022] Monitor real-time operational data to optimize resource allocation plans;

[0023] If the business response time in the real-time operation data does not reach the preset service level agreement threshold, the allocation parameters in the optimized resource allocation plan are dynamically adjusted to generate a final resource allocation plan.

[0024] In one embodiment, performing a resource allocation adjustment calculation based on current supply and demand data includes calculating a resource allocation adjustment amount using the following formula:

[0025] ΔA=γ·σ(ξ)·sign(D imb )·ln(1+|D imb |)

[0026] Among them, ΔA is the resource allocation adjustment amount, D imb is the supply-demand imbalance, calculated based on the current supply-demand data, ξ is the resource feature vector extracted from the resource tag library, is the resource feature matching function, and γ is the adaptive coefficient calculated based on the stability score in the resource tag library.

[0027] In one embodiment, the method further includes constraining the dynamic range of the resource allocation adjustment amount ΔA using the following formula:

[0028]

[0029] Among them, ζ1 is the basic elastic constraint radius calculated based on the elastic capacity of each resource type in the resource tag library, is the collaborative constraint matrix, H=[h1,h2,…,h n ] T To extract the historical constraint vector formed by the features of the latest n adjustments from the historical adjustment records, is the time decay weight, λ is the decay coefficient, is the historical volatility measure, and μ is the historical mean.

[0030] In a second aspect, the present application further provides a soft switch resource elastic scheduling device for a cloud native architecture, comprising:

[0031] The resource classification and tagging module is used to extract and tag features of computing nodes, storage resources, and network resources using the resource classification model to generate a resource tag library; the resource tag library is used to represent the initial classification of each resource;

[0032] The supply and demand data generation module is used to extract the corresponding resource feature data from the resource tag library based on the initial classification of each resource, and combine it with the real-time collected usage status data of various resources to generate the current supply and demand data of various resources;

[0033] The multi-objective optimization scheduling module is used to calculate the priority order of resource allocation based on current supply and demand data and generate preliminary ranking results for resource scheduling using a multi-objective optimization algorithm that integrates business response time and cost balance optimization parameters;

[0034] An imbalance detection module is used to detect resource supply and demand imbalance based on preliminary sorting results and real-time monitoring of resource supply and demand change data streams;

[0035] The resource allocation adjustment module is used to perform resource allocation adjustment calculations based on current supply and demand data and generate a preliminary resource allocation plan if a supply and demand imbalance is detected.

[0036] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the following steps when executing the computer program.

[0037] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the following steps when the computer program is executed by a processor.

[0038] The above-mentioned method, apparatus, computer equipment, and storage medium for elastic scheduling of soft switch resources in a cloud-native architecture use a resource classification model to extract features and label computing nodes, storage resources, and network resources to generate a resource tag library, solving the problem of lack of refined resource classification in traditional static allocation and laying a classification foundation for elastic scheduling. By combining real-time collected resource usage status data to generate current supply and demand data, dynamic perception of resource status is achieved. Resource allocation priorities are calculated using a multi-objective optimization algorithm that integrates business response time and cost balance, resolving the conflict between business response time and cost consumption. Imbalanced states are detected by real-time monitoring of resource supply and demand change data streams and dynamic adjustment calculations are performed based on current supply and demand data, overcoming the drawback of traditional solutions that make it difficult to respond to resource supply and demand imbalances in real time. Efficient elastic scheduling and optimized allocation of soft switch resources are achieved in a cloud-native architecture, significantly improving resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 This is a flow chart of a method for elastic scheduling of soft switch resources in a cloud native architecture of the present invention;

[0041] Figure 2 This is a structural diagram of a soft switch resource elastic scheduling device for a cloud native architecture of the present invention;

[0042] Figure 3 FIG666 is a diagram of the 555 structure of the present invention; DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0044] In one embodiment, Figure 1 As shown, a method for elastic scheduling of soft switch resources in a cloud native architecture is provided. This embodiment uses the method applied to a cloud platform scheduling management system as an example. It can be understood that the method can also be deployed in an edge computing node cluster, and can also be applied to a collaborative system composed of a business access terminal, a central cloud resource pool, and a network control unit, and is implemented through the following interactions: When the soft switch system encounters a sudden business load, the access terminal reports the request traffic to the scheduling system in real time. The system extracts the characteristics of computing nodes, storage resources, and network resources through a resource classification model and constructs a resource tag library. It generates supply and demand data in combination with the real-time collected resource usage status, and uses a multi-objective algorithm that integrates response time and cost optimization to calculate the scheduling priority; if a supply and demand imbalance is detected where the resource utilization exceeds the threshold, a dynamic adjustment mechanism is triggered - an expansion instruction is sent to the computing resource pool, the network controller is linked to optimize the transmission path, and the storage resource allocation is coordinated to achieve low-latency elastic scheduling. In this embodiment, the method includes the following steps:

[0045] S01: Use the resource classification model to extract and label the features of computing nodes, storage resources, and network resources to generate a resource tag library; the resource tag library is used to represent the initial classification of each resource.

[0046] Among them, a resource classification model (a classification tool built based on machine learning algorithms such as neural networks, decision trees, or preset rules) can be used to analyze the multi-dimensional attributes of various resources (such as the number of CPU cores, memory capacity, processor model of computing nodes, read and write speeds, capacity types of storage resources, bandwidth peaks, latency indicators of network resources, etc.), extract characteristic parameters that can characterize resource characteristics, and label resources according to preset classification rules or classification logic obtained from model training to form a label library containing initial classification information of resources. The resource label library stores a set of characteristic labels for each resource, which is used to characterize the initial classification attributes of computing nodes, storage resources, and network resources, such as labeling computing nodes as high-performance computing type and general computing type, labeling storage resources as hot data storage and cold data storage, labeling network resources as low-latency network and high-bandwidth network, etc., to lay a classification foundation for feature matching and supply and demand analysis in the subsequent resource scheduling process.

[0047] S02, based on the initial classification of each resource, extract the corresponding resource feature data from the resource tag library, and combine it with the real-time collected usage status data of various resources to generate the current supply and demand data of various resources.

[0048] Among them, based on the initial classification of each resource represented by the resource tag library, the resource feature data under the corresponding classification is extracted from the resource tag library. The usage status of each type of resource can be dynamically monitored through distributed monitoring components or real-time data acquisition interfaces to obtain real-time usage status data (such as CPU utilization, storage occupancy, and real-time network bandwidth throughput); the extracted static resource feature data and the dynamic usage status data collected in real time are correlated and processed through a data fusion algorithm. Based on a quantitative model of resource supply capacity and business needs, current supply and demand data representing the current available amount, demand amount, and load pressure of computing resources, storage resources, and network resources are generated, providing a data basis for priority calculation and imbalance detection in subsequent resource scheduling.

[0049] S03, based on the current supply and demand data, adopts a multi-objective optimization algorithm that integrates business response time and cost balance optimization parameters to calculate the priority order of resource allocation and generate a preliminary ranking result for resource scheduling.

[0050] Among them, a multi-objective optimization model that includes the business response time minimization objective function and the cost consumption balance objective function can be constructed. The current supply and demand data are used as input parameters. Combined with the preset business response time threshold and the cost optimization weight coefficient (the weight coefficient can be dynamically adjusted based on the business type and service level agreement (SLA)), the model is solved using multi-objective optimization algorithms such as the non-dominated sorting genetic algorithm (NSGA-II) and the particle swarm optimization algorithm (PSO) to obtain the Pareto optimal solution set; based on the comprehensive score of each resource allocation scheme in the optimal solution set (which can be obtained by weighted calculation of indicators such as the business response time compliance rate and the cost consumption index), a preliminary ranking result for resource scheduling is generated, and the allocation priority order of computing nodes, storage resources, and network resources is determined, providing a decision-making basis for subsequent resource scheduling.

[0051] S04: Detecting resource supply and demand imbalance based on the preliminary sorting results and the real-time monitored resource supply and demand change data stream.

[0052] Among them, the resource supply and demand change data stream (including real-time usage status change data of computing resources, storage resources and network resources, such as CPU utilization fluctuations, storage read and write rate changes, dynamic adjustment of network bandwidth occupancy and other time series data) is collected in real time through distributed monitoring components, and the data stream is parsed and feature extracted by the real-time stream data processing engine. The resource allocation status corresponding to the preliminary sorting result and the real-time supply and demand changes are jointly analyzed through a preset imbalance detection model (which can be constructed based on threshold comparison, statistical analysis or machine learning algorithm): when it is detected that the deviation value between the actual resource usage and the allocated amount exceeds the preset imbalance threshold, or the supply and demand change rate reaches the trigger condition, it is determined that there is a resource supply and demand imbalance.

[0053] S05: If a supply-demand imbalance is detected, a resource allocation adjustment calculation is performed based on the current supply-demand data to generate a preliminary resource allocation plan.

[0054] If a resource supply and demand imbalance is detected, the degree of deviation between resource supply and business demand can be analyzed based on the current supply and demand data. Combined with the resource characteristic attributes stored in the resource tag library (such as the performance type of computing nodes, the read and write characteristics of storage resources, the bandwidth level of network resources, etc.), the adjustment direction and magnitude of resource allocation are calculated through a preset dynamic adjustment strategy (based on a rule engine or machine learning model). The adjustment strategy comprehensively considers the business response time requirements and cost consumption balance goals, and dynamically optimizes the allocation of computing nodes, storage resources, and network resources. The feasibility of the adjusted resource allocation plan can be verified through a distributed scheduling framework. Combined with the real-time monitored resource availability status and business load trends, a preliminary resource allocation plan adapted to the current supply and demand imbalance scenario is generated to achieve elastic scheduling and optimized allocation of soft switch resources under the cloud native architecture. The preliminary resource allocation plan includes a combination of adjustment strategies for various types of resources, which are used to guide specific operations such as expansion / contraction of computing nodes, reallocation of storage resources, and dynamic adjustment of network bandwidth.

[0055] The above-mentioned soft switch resource elastic scheduling method for a cloud-native architecture uses a resource classification model to extract and label features of computing nodes, storage resources, and network resources to generate a resource tag library, thereby solving the problem of lack of refined resource classification in traditional static allocation; based on the initial classification of each resource, feature data is extracted from the tag library and combined with real-time collected resource usage status data to generate current supply and demand data, thereby realizing dynamic perception of resource status and making up for the defect of traditional methods lacking dynamic perception of real-time supply and demand status; a multi-objective optimization algorithm that integrates business response time and cost balance optimization parameters is used to calculate the resource allocation priority order and generate preliminary sorting results, thereby resolving the conflict between business response time and cost consumption; based on the preliminary sorting results and the real-time monitored resource supply and demand change data stream, the imbalance of resource supply and demand is detected. If an imbalance is detected, resource allocation adjustment calculation is performed based on the current supply and demand data and a preliminary resource allocation plan is generated, thereby overcoming the deficiency of traditional plans in responding to resource supply and demand imbalance in real time, realizing efficient elastic scheduling and optimized allocation of soft switch resources, and significantly improving resource utilization.

[0056] In one embodiment, after generating the preliminary resource allocation plan, the method further includes:

[0057] S11, obtaining load data of target resources according to the preliminary resource allocation plan;

[0058] S12, extracting features from the load data of the target resource and constructing a multi-dimensional feature vector representing the dynamic characteristics of the load distribution;

[0059] S13, based on the multi-dimensional feature vector, a reinforcement learning algorithm is used to predict the future load demand trend of the target resource and generate resource demand trend prediction information.

[0060] Specifically, according to the preliminary resource allocation plan, the load data of the target resources (including computing nodes, storage resources or network resources) under the current allocation strategy (including but not limited to real-time operating parameters such as CPU utilization, storage read and write rates, and network bandwidth throughput) is obtained through a distributed monitoring component or a real-time data acquisition interface; the acquired load data is subjected to time domain, frequency domain and statistical feature extraction, such as calculating the mean, variance, peak frequency, fluctuation trend, etc., to construct a vector containing multi-dimensional feature parameters to characterize the dynamic characteristics of the load distribution, which can reflect properties such as the periodicity, burstiness and stability of the load; based on the multi-dimensional feature vector, a reinforcement learning algorithm (such as a deep Q network, proximal policy optimization, etc.) is used to construct a prediction model, and the feature vector is used as the state input. By defining a reward function for changes in resource demand, the model learns the historical change pattern of the load data and predicts the load demand trend of the target resource in the future time period. Resource demand trend prediction information containing information such as load growth / decline rate and peak occurrence time can be generated. By combining load data characterization with reinforcement learning dynamic prediction, forward-looking analysis of resource demand is achieved, providing data support for the optimization of resource scheduling strategies.

[0061] In one embodiment, after generating the resource demand trend forecast information, the method further includes:

[0062] S21, based on the resource demand trend forecast information, obtaining a multi-dimensional operating status indicator set of each computing node;

[0063] S22, using a pre-trained performance bottleneck classification model, based on a multi-dimensional operating status indicator set, generates an identifier of a computing node with a resource bottleneck and its corresponding resource bottleneck category feature vector;

[0064] S23: If the resource bottleneck category feature vector exceeds a preset threshold range, load balancing is performed on the computing nodes with resource bottlenecks to generate a node resource allocation plan.

[0065] For example, based on the resource demand trend prediction information, a multi-dimensional operating status indicator set of each computing node (a multi-dimensional parameter set that characterizes the operating status of the computing node, including but not limited to CPU utilization, memory occupancy, disk read and write rate, network throughput and process scheduling delay, etc.) can be collected in real time through a distributed monitoring system; a pre-trained performance bottleneck classification model (based on historically annotated node operation data trained by a deep learning algorithm, which can identify different types of resource bottleneck characteristics) is used to perform feature analysis and classification on the multi-dimensional operating status indicator set, and a unique identifier of the computing node with a resource bottleneck and its corresponding resource bottleneck category feature vector (including bottleneck type) is generated. , such as CPU bottlenecks, memory bottlenecks, I / O bottlenecks, etc. and their severity quantitative indicators feature vectors); if the bottleneck severity indicator in the resource bottleneck category feature vector exceeds the preset threshold range (dynamically set based on the business service level agreement and historical operation data), the load balancing processing mechanism is triggered, and the computing nodes with resource bottlenecks are load balanced through strategies such as task migration, resource reallocation or node expansion, and a node resource allocation plan containing specific resource adjustment strategies and parameters is generated. By combining the predicted resource demand trend with real-time node status analysis, forward-looking identification and dynamic optimization of computing node resource bottlenecks are achieved, thereby improving the efficiency of system resource utilization.

[0066] In one embodiment, after generating the node resource allocation plan, the method further includes:

[0067] S31, obtaining real-time availability data of storage resources and network resources associated with the node resource allocation plan;

[0068] S32, through the distributed scheduling framework based on the preset storage and network resource allocation strategy, coordinate and match storage resources and network resources and generate an optimized resource allocation plan.

[0069] Specifically, the real-time availability data of storage resources and network resources associated with the node resource allocation plan can be obtained through the distributed resource monitoring component (characterizing storage resources, such as available capacity, read and write rates, IOPS performance and network resources, such as bandwidth utilization, delay jitter, and link status, which are a set of quantitative parameters of the current service status; through the distributed scheduling framework (a resource collaborative scheduling system based on a microservice architecture, which can dynamically allocate resources across nodes), based on preset storage and network resource allocation strategies (including load balancing strategies, QoS guarantee strategies, resource affinity strategies, etc.), the block storage / object storage allocation units of storage resources and the bandwidth slices / routing paths of network resources are coordinated and matched and dynamically allocated to generate an optimized resource allocation plan covering storage resource allocation granularity, network bandwidth priority and inter-node resource collaboration strategies. By linking computing node resource allocation with the real-time availability data of storage and network resources, the collaborative scheduling of the three types of resources under the cloud native architecture is realized.

[0070] In one embodiment, after generating the optimized resource allocation plan, the method further includes:

[0071] S41, monitoring real-time operation data of the optimized resource allocation plan;

[0072] S42: If the service response time in the real-time operation data does not reach the preset service level agreement threshold, dynamically adjust the allocation parameters in the optimized resource allocation plan to generate a final resource allocation plan.

[0073] For example, the real-time operating data of the optimized resource allocation plan (a set of multi-dimensional indicators that characterize the execution status of the resource allocation strategy, including but not limited to business response time, computing node load rate, storage IO latency, network throughput, etc.) can be monitored by a real-time monitoring system deployed in a cloud-native architecture; if the business response time in the real-time operating data does not reach the preset service level agreement (SLA) threshold (pre-set based on business type, user needs and industry standards to measure the degree to which service quality meets the standards), the dynamic adjustment mechanism is triggered, and the allocation parameters in the optimized resource allocation plan (various configuration parameters that affect the resource allocation strategy, such as the allocation ratio of computing resource cores, storage resource read and write priority, network bandwidth slice quota, etc.) are adaptively adjusted through the intelligent scheduling engine to generate a final resource allocation plan that meets the SLA requirements. By building a closed-loop feedback mechanism, the real-time operating data is dynamically compared with the preset service standards to achieve iterative optimization of the resource allocation strategy and ensure that the business response time meets the standards.

[0074] In one embodiment, S51 performs resource allocation adjustment calculation based on current supply and demand data, including calculating the resource allocation adjustment amount using the following formula:

[0075] ΔA=γ·σ(ξ)·sign(Dimb )·ln(1+|D imb |)

[0076] Among them, ΔA is the resource allocation adjustment amount, D imb is the supply-demand imbalance, calculated based on the current supply-demand data, ξ is the resource feature vector extracted from the resource tag library, is the resource feature matching function, and γ is the adaptive coefficient calculated based on the stability score in the resource tag library.

[0077] Specifically, the supply-demand imbalance D is calculated based on the current supply and demand data. imb (It can be obtained by difference analysis and normalization of resource availability and demand), extract the resource feature vector ξ corresponding to the target resource from the resource tag library, and use the resource feature matching function Calculate the current resource characteristics and the optimal characteristic vector ξ opt The degree of matching: The Euclidean distance norm ||·|| is used to measure the similarity between feature vectors. The adaptive coefficient γ (obtained by normalizing and weight mapping the stability score) is calculated based on the pre-labeled resource stability score in the resource tag library (reflecting the stability performance of the resource in the historical scheduling, with a value range of [0,1]). The direction of resource allocation adjustment is determined by the sign function sing (positive for expansion and negative for contraction). Combined with the logarithmic function ln(1+|D imb |) The imbalance is nonlinearly amplified and the resource allocation adjustment ΔA is calculated. By integrating real-time supply and demand data, resource feature matching, and stability attributes into the mathematical model, quantitative calculation and dynamic optimization of the resource adjustment amount are achieved.

[0078] In one embodiment, the method further includes S61, using the following formula to impose a dynamic range constraint on the resource allocation adjustment amount ΔA:

[0079]

[0080] Among them, ζ1 is the basic elastic constraint radius calculated based on the elastic capacity of each resource type in the resource tag library, is the collaborative constraint matrix, H=[h1,h2,…,h n ] T To extract the historical constraint vector formed by the features of the latest n adjustments from the historical adjustment records, is the time decay weight, λ is the decay coefficient, is the historical volatility measure, and μ is the historical mean.

[0081] For example, the basic elastic constraint radius ζ1 (the basic threshold that characterizes the maximum adjustment range that the resource can withstand) can be calculated based on the elastic expansion capability, historical adjustment range and other parameters of each resource type (including computing nodes, storage resources, network resources, etc.) in the resource tag library, and the features of the latest n adjustments (such as adjustment range, adjustment direction, resource status during adjustment, etc.) are extracted from the historical adjustment records to form a historical constraint vector H = [h1, h2, ..., h n ] T (a multidimensional parameter set reflecting the historical pattern of resource adjustment), through the time decay weight formula Calculate the weight of each historical adjustment amount, and adjust the influence of historical data through the attenuation coefficient λ (the weight of the adjustment record at a certain time will decrease over time); calculate the historical volatility measure based on the mean μ variance of the historical adjustment amount, Quantify the fluctuation range of historical adjustment amount; through the collaborative constraint matrix Fusion of historical constraints and the correlation between the current adjustment amount, using the hyperbolic tangent function to map the adjustment amount into the constraint range, through Dynamically constrain resource allocation adjustments, using a min function to ensure they do not exceed the base elasticity constraint radius. A collaborative constraint matrix further refines the constraint range based on historical adjustment patterns. By integrating resource elasticity with historical adjustment patterns, a dynamic and adaptive adjustment constraint mechanism is constructed to prevent system volatility caused by excessive adjustments, thereby improving the stability and reliability of resource scheduling.

[0082] The above-mentioned soft switch resource elastic scheduling method of a cloud native architecture extracts and labels the features of computing nodes, storage and network resources through a resource classification model to generate a resource tag library, thereby solving the problem of lack of refined resource classification in traditional static allocation; extracts feature data from the tag library based on the initial classification of each resource and generates current supply and demand data in combination with the real-time collected resource usage status data to achieve dynamic perception of resource status; adopts a multi-objective optimization algorithm that integrates business response time and cost balance optimization parameters to calculate the resource allocation priority order and generate preliminary sorting results, thereby solving the conflict between business response time and cost consumption that is difficult to balance; based on the preliminary sorting results and real-time monitoring The resource supply and demand change data stream detects the imbalance of resource supply and demand. If an imbalance is detected, resource allocation adjustment calculation is performed based on the current supply and demand data and a preliminary resource allocation plan is generated. At the same time, the reinforcement learning algorithm is combined to predict resource load trends, the performance bottleneck classification model is used to identify node bottlenecks, and the storage and network resource allocation is coordinated through a distributed scheduling framework. The allocation parameters are dynamically optimized according to the business response time and other subsequent processing to overcome the shortcomings of traditional solutions that are difficult to respond to resource supply and demand imbalances in real time, realize efficient and elastic scheduling and optimized allocation of soft switching resources, improve resource utilization, balance business response time and cost consumption, and enhance the elastic scheduling capabilities of the soft switching system under the cloud native architecture.

[0083] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0084] Based on the same inventive concept, an embodiment of the present application also provides a cloud-native architecture soft switch resource elastic scheduling device for implementing the above-mentioned cloud-native architecture soft switch resource elastic scheduling method. The implementation solution provided by the device is similar to the implementation solution described in the above-mentioned method. Therefore, the specific limitations of one or more embodiments of a cloud-native architecture soft switch resource elastic scheduling device provided below can be referred to the above limitations of a cloud-native architecture soft switch resource elastic scheduling method, and will not be repeated here.

[0085] In an exemplary embodiment, Figure 2 As shown, a soft switch resource elastic scheduling device with a cloud native architecture is provided, including:

[0086] The resource classification and labeling module 100 is used to extract and label the features of computing nodes, storage resources and network resources using the resource classification model to generate a resource label library; the resource label library is used to represent the initial classification of each resource;

[0087] The supply and demand data generation module 200 is used to extract corresponding resource feature data from the resource tag library based on the initial classification of each resource, and generate current supply and demand data for each resource based on the real-time collected usage status data of each resource;

[0088] The multi-objective optimization scheduling module 300 is used to calculate the priority order of resource allocation based on the current supply and demand data and generate a preliminary ranking result for resource scheduling by using a multi-objective optimization algorithm that integrates the optimization parameters of service response time and cost balance;

[0089] An imbalance detection module 400 is configured to detect a resource supply and demand imbalance based on the preliminary sorting results and the real-time monitored resource supply and demand change data stream;

[0090] The resource allocation adjustment module 500 is configured to perform resource allocation adjustment calculations based on current supply and demand data to generate a preliminary resource allocation plan if a supply and demand imbalance is detected.

[0091] In one embodiment, Figure 3 As shown, the multi-objective optimization scheduling module 300 further includes a prediction optimization unit 310, which is used to:

[0092] Obtain load data of target resources according to the preliminary resource allocation plan;

[0093] Extract features from the load data of target resources and construct a multi-dimensional feature vector that characterizes the dynamic characteristics of load distribution;

[0094] Based on the multidimensional feature vector, the reinforcement learning algorithm is used to predict the future load demand trend of the target resource and generate resource demand trend prediction information.

[0095] In one embodiment, the imbalance detection module 400 is further configured to:

[0096] Based on resource demand trend forecast information, obtain a multi-dimensional operating status indicator set for each computing node;

[0097] Using the pre-trained performance bottleneck classification model, based on a multi-dimensional set of operating status indicators, we generate the identifiers of computing nodes with resource bottlenecks and their corresponding resource bottleneck category feature vectors.

[0098] If the resource bottleneck category feature vector exceeds the preset threshold range, load balancing is performed on the computing nodes with resource bottlenecks to generate a node resource allocation plan.

[0099] In one embodiment, the resource allocation adjustment module 500 is further configured to:

[0100] Obtain real-time availability data of storage resources and network resources associated with the node resource allocation plan;

[0101] Through the distributed scheduling framework, storage resources and network resources are coordinated, matched and allocated based on the preset storage and network resource allocation strategies to generate an optimized resource allocation plan.

[0102] In one embodiment, the resource allocation adjustment module 500 is further configured to:

[0103] Monitor real-time operational data to optimize resource allocation plans;

[0104] If the business response time in the real-time operation data does not reach the preset service level agreement threshold, the allocation parameters in the optimized resource allocation plan are dynamically adjusted to generate a final resource allocation plan.

[0105] In one embodiment, the resource allocation adjustment module 500 is further configured to perform resource allocation adjustment calculations based on current supply and demand data, including calculating a resource allocation adjustment amount using the following formula:

[0106] ΔA=γ·σ(ξ)·sign(D imb )·ln(1+|D imb |)

[0107] Among them, ΔA is the resource allocation adjustment amount, D imb is the supply-demand imbalance, calculated based on the current supply-demand data, ξ is the resource feature vector extracted from the resource tag library, is the resource feature matching function, and γ is the adaptive coefficient calculated based on the stability score in the resource tag library.

[0108] In one embodiment, the resource allocation adjustment module 500 is further configured to use the following formula to impose a dynamic range constraint on the resource allocation adjustment amount ΔA:

[0109]

[0110] Among them, ζ1 is the basic elastic constraint radius calculated based on the elastic capacity of each resource type in the resource tag library, is the collaborative constraint matrix, H=[h1,h2,…,h n ] T To extract the historical constraint vector formed by the features of the latest n adjustments from the historical adjustment records, is the time decay weight, λ is the decay coefficient, is the historical volatility measure, and μ is the historical mean.

[0111] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the soft switch resource elastic scheduling method for a cloud native architecture as described above are implemented.

[0112] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0113] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0114] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A method for elastic scheduling of soft switch resources in a cloud native architecture, characterized in that: The method comprises: Using the resource classification model to extract and label features of computing nodes, storage resources, and network resources, a resource tag library is generated; the resource tag library is used to characterize the initial classification of each resource; Based on the initial classification of each resource, corresponding resource feature data is extracted from the resource tag library, and combined with the real-time collected usage status data of each resource, current supply and demand data of each resource is generated; Based on the current supply and demand data, a multi-objective optimization algorithm that integrates business response time and cost balance optimization parameters is used to calculate the priority order of resource allocation and generate a preliminary ranking result for resource scheduling; Detecting resource supply and demand imbalance based on the preliminary sorting results and the real-time monitored resource supply and demand change data stream; If a supply-demand imbalance is detected, a resource allocation adjustment calculation is performed based on the current supply-demand data to generate a preliminary resource allocation plan.

2. The method according to claim 1, characterized in that After generating the preliminary resource allocation plan, the method further includes: According to the preliminary resource allocation plan, obtaining load data of the target resource; Extracting features from the load data of the target resource to construct a multi-dimensional feature vector representing the dynamic characteristics of the load distribution; Based on the multidimensional feature vector, a reinforcement learning algorithm is used to predict the future load demand trend of the target resource to generate resource demand trend prediction information.

3. The method according to claim 2, characterized in that After generating the resource demand trend forecast information, the method further includes: Based on the resource demand trend forecast information, obtaining a multi-dimensional operating status indicator set for each computing node; Using a pre-trained performance bottleneck classification model, based on the multi-dimensional operating status indicator set, generates an identifier of a computing node with a resource bottleneck and a corresponding resource bottleneck category feature vector; If the resource bottleneck category feature vector exceeds a preset threshold range, load balancing processing is performed on the computing node with the resource bottleneck to generate a node resource allocation plan.

4. The method according to claim 3, characterized in that After generating the node resource allocation plan, the method further includes: Acquiring real-time availability data of storage resources and network resources associated with the node resource allocation scheme; The distributed scheduling framework is used to coordinate and match the storage resources and network resources based on the preset storage and network resource allocation strategy to generate an optimized resource allocation plan.

5. The method according to claim 4, characterized in that After generating the optimized resource allocation plan, the method further includes: Monitoring real-time operational data of the optimized resource allocation scheme; If the service response time in the real-time operation data does not reach a preset service level agreement threshold, the allocation parameters in the optimized resource allocation plan are dynamically adjusted to generate a final resource allocation plan.

6. The method according to claim 1, characterized in that The resource allocation adjustment calculation is performed based on the current supply and demand data, including calculating the resource allocation adjustment amount using the following formula: ΔA=γ·σ(ξ)·sign(D imb )·ln(1+|D imb |) Among them, ΔA is the resource allocation adjustment amount, D imb is the supply-demand imbalance degree, calculated based on the current supply-demand data, ξ is the resource feature vector extracted from the resource tag library, is the resource feature matching function, and γ is the adaptive coefficient calculated based on the stability score in the resource tag library.

7. The method according to claim 6, characterized in that The method further includes applying a dynamic range constraint to the resource allocation adjustment amount ΔA using the following formula: Wherein, ζ1 is the basic elastic constraint radius calculated based on the elastic capability of each resource type in the resource tag library, is the collaborative constraint matrix, H=[h1,h2,…,h n ] T To extract the historical constraint vector formed by the features of the latest n adjustments from the historical adjustment records, is the time decay weight, λ is the decay coefficient, is the historical volatility measure, and μ is the historical mean.

8. A soft switch resource elastic scheduling device for cloud native architecture, characterized in that: The device comprises: The resource classification and labeling module is used to extract and label the features of computing nodes, storage resources, and network resources using the resource classification model to generate a resource label library; the resource label library is used to represent the initial classification of each resource; A supply and demand data generation module is used to extract corresponding resource feature data from the resource tag library based on the initial classification of each resource, and generate current supply and demand data of each resource in combination with the real-time collected usage status data of each resource; A multi-objective optimization scheduling module is used to calculate the priority order of resource allocation based on the current supply and demand data and generate a preliminary ranking result for resource scheduling by using a multi-objective optimization algorithm that integrates business response time and cost balance optimization parameters; An imbalance detection module, configured to detect a resource supply and demand imbalance based on the preliminary sorting result and a real-time monitored resource supply and demand change data stream; The resource allocation adjustment module is configured to, if a supply-demand imbalance is detected, perform resource allocation adjustment calculations based on the current supply-demand data to generate a preliminary resource allocation plan.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images