Computing power resource arrangement method and system based on dynamic generation strategy

Abstract

The invention discloses a computing power resource arrangement method and system based on a dynamic generation strategy, and the method comprises the steps: monitoring and periodically collecting the computing power resource data of each virtual node in a target computer cluster in real time, and determining the type data of each business flow based on the computing power resource data; designing a corresponding distribution function according to the arrival rate of each type of data so as to analyze the CPU system efficiency of each service flow to obtain CPU system efficiency data; inputting the type data, the computing power resource data, the arrival rate and the CPU system efficiency data into the trained business flow computing power value prediction model to obtain a computing power demand result of the business flow; when the service flow is deployed to the target computer cluster, arranging and scheduling computing power resources according to a computing power demand result; therefore, reasonable allocation and efficient scheduling of computing power resources can be realized.

Description

Technical Field

[0001] The present invention relates to the field of computing power network technology, and in particular to a computing power resource orchestration method and system based on a dynamic generation strategy. Background Art

[0002] With the development of cloud computing and distributed computing, computing resource orchestration technology has gradually become an important means to ensure the efficient use of computing resources. For example, Kubernetes (K8s for short) is the most mainstream distributed cluster management tool currently used. In Kubernetes, computing resource orchestration technology mainly uses container orchestration technology. Kubernetes is an open source container orchestration system developed by Google, which is designed to automate the deployment, expansion and management of containerized applications. It provides a complete set of tools to enable developers and operators to collaborate efficiently and quickly build and operate complex distributed systems.

[0003] Although the existing container orchestration technology has achieved remarkable results in computing resource management, there are still some deficiencies and challenges in practical applications. The existing computing resource orchestration methods have poor orchestration effects. When faced with dynamically changing business needs and complex application scenarios, they cannot reasonably allocate computing resources according to business status, resulting in low comprehensive utilization of computing resources.

[0004] It can be seen that how to effectively manage computing resources has become a technical problem that needs to be urgently solved by technical personnel in this field. Summary of the invention

[0005] The present invention provides a computing power resource scheduling method and system based on a dynamic generation strategy to solve the problem of how to effectively manage computing power resources to improve the utilization rate of computing power resources.

[0006] In order to solve the above technical problems, an embodiment of the present invention provides a computing resource scheduling method based on a dynamic generation strategy, including:

[0007] Real-time monitoring and periodic collection of computing power resource data of each virtual node in the target computer cluster, and determining type data of each business flow based on the computing power resource data;

[0008] Design a corresponding distribution function according to the arrival rate of each type of data to analyze the CPU system efficiency of each business flow and obtain CPU system efficiency data;

[0009] Input the type data, the computing power resource data, the arrival rate and the CPU system efficiency data into the trained business flow computing power value prediction model to obtain the computing power demand result of the business flow;

[0010] When deploying the business flow to the target computer cluster, the computing resources are scheduled and arranged according to the computing demand result.

[0011] Furthermore, the determining the type data of each service flow based on the computing power resource data includes:

[0012] Performing time series analysis on the computing power resource data to extract attribute features of the computing power resource data;

[0013] Determine a target marking rule according to the attribute characteristics;

[0014] The service flows are marked according to the target marking rule to determine the type data of each service flow.

[0015] Furthermore, the corresponding distribution function is designed according to the arrival rate of each type of data to analyze the CPU system efficiency of each service flow to obtain CPU system efficiency data, including:

[0016] Acquire the burst characteristics of the service flow based on the designed distribution function;

[0017] Performing Laplace transform on the set distribution function to obtain system response characteristics of the service flow;

[0018] According to the burst characteristics and the system response characteristics, the arrival ratio of the business flow and the system efficiency are analyzed and counted to obtain the CPU system efficiency data.

[0019] Furthermore, the process of constructing the business flow computing power value prediction model includes:

[0020] Preprocessing the type data, the computing power resource data, the arrival rate, and the CPU system efficiency data to obtain a first target data set;

[0021] Dividing the first target data set into a training set and a test set;

[0022] The training set is input into the neural network model for training, and the test set is input into the neural network model for testing to obtain the business flow computing power value prediction model.

[0023] Furthermore, when deploying the service flow to the target computer cluster, scheduling computing resources according to the computing power demand result includes:

[0024] Determine the corresponding application according to the attributes of the business flow, and group the business flows in the same application;

[0025] Prioritizing the sets according to the priority attributes of the applications and assigning weights to each of the sets, and modifying the weights based on the real-time characteristics of the traffic flows;

[0026] Mapping the business flows in the set to the application of the corresponding container based on the modified weights, and statistically calculating the total computing power value of the application corresponding to the same type of business flows according to the computing power demand results;

[0027] The application programs are grouped into corresponding containers according to the total computing power value, so as to configure corresponding computing power resources for scheduling.

[0028] Another embodiment of the present invention provides a computing resource scheduling system based on a dynamic generation strategy, including:

[0029] A data acquisition module, used to monitor in real time and periodically collect computing power resource data of each virtual node in the target computer cluster, and determine the type data of each business flow based on the computing power resource data;

[0030] A service flow analysis module is used to design a corresponding distribution function according to the arrival rate of each type of data, so as to analyze the CPU system efficiency of each service flow and obtain CPU system efficiency data;

[0031] A computing power prediction module, used to input the type data, the computing power resource data, the arrival rate and the CPU system efficiency data into a trained business flow computing power value prediction model to obtain a computing power demand result of the business flow;

[0032] The resource scheduling module is used to schedule computing resources according to the computing power demand result when deploying the business flow to the target computer cluster.

[0033] Furthermore, the data acquisition module is specifically used to:

[0034] Performing time series analysis on the computing power resource data to extract attribute features of the computing power resource data;

[0035] Determine a target marking rule according to the attribute characteristics;

[0036] The service flows are marked according to the target marking rule to determine the type data of each service flow.

[0037] Furthermore, the business flow analysis module is specifically used to:

[0038] Acquire the burst characteristics of the service flow based on the designed distribution function;

[0039] Performing Laplace transform on the set distribution function to obtain system response characteristics of the service flow;

[0040] According to the burst characteristics and the system response characteristics, the arrival ratio of the business flow and the system efficiency are analyzed and counted to obtain the CPU system efficiency data.

[0041] Furthermore, the process of constructing the business flow computing power value prediction model includes:

[0042] Preprocessing the type data, the computing power resource data, the arrival rate, and the CPU system efficiency data to obtain a first target data set;

[0043] Dividing the first target data set into a training set and a test set;

[0044] The training set is input into the neural network model for training, and the test set is input into the neural network model for testing to obtain the business flow computing power value prediction model.

[0045] Furthermore, the resource scheduling module is specifically used to:

[0046] Determine the corresponding application according to the attributes of the business flow, and group the business flows in the same application;

[0047] Prioritizing the sets according to the priority attributes of the applications and assigning weights to each of the sets, and modifying the weights according to the real-time characteristics of the traffic flows;

[0048] Mapping the business flows in the set to the application of the corresponding container according to the modified weights, and statistically calculating the total computing power value of the application corresponding to the same type of business flows according to the computing power demand results;

[0049] The application programs are grouped into corresponding containers according to the total computing power value, so as to configure corresponding computing power resources for scheduling.

[0050] Compared with the prior art, the embodiments of the present invention have the following advantages:

[0051] (1) Real-time monitoring and periodic collection of computing resource data of each virtual node in the target computer cluster to obtain business flow data; and through mathematical distribution functions and Laplace transforms, the arrival distribution of business flows is calculated and analyzed to obtain the response characteristics of the system in order to accurately evaluate the system efficiency of business flows; the computing power value of business flows is accurately predicted using a neural network model so as to achieve a reasonable allocation of computing resources.

[0052] (2) By predicting the computing power required for each type of business flow, the business flows are mapped to the corresponding container applications, and they are sorted and weighted according to priority attributes to achieve efficient scheduling and allocation of computing resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flowchart of a computing resource arrangement method based on a dynamic generation strategy in one embodiment of the present invention;

[0054] Figure 2 is a schematic diagram of a mathematical representation of a Laplace transform in one embodiment of the present invention;

[0055] Figure 3 It is a schematic diagram of the computing power scheduling process in one embodiment of the present invention;

[0056] Figure 4 It is a structural block diagram of a computing resource orchestration system based on a dynamic generation strategy in one embodiment of the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] In the description of this application, the terms "first", "second", "third", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the feature. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0059] In the description of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. The terms "vertical", "horizontal", "left", "right", "upper", "lower" and similar expressions used in this article are only for illustrative purposes, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used in this article includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0060] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood by specific circumstances.

[0061] An embodiment of the present invention provides a computing resource scheduling method based on a dynamic generation strategy. For details, see Figure 1 , Figure 1 The figure shows a flow chart of a computing resource arrangement method based on a dynamic generation strategy in one embodiment of the present invention, which includes the following steps:

[0062] S1. Real-time monitoring and periodic collection of computing power resource data of each virtual node in the target computer cluster, and determining type data of each business flow based on the computing power resource data.

[0063] As you can understand, cluster refers to a group of physical or virtual computer resources (nodes) connected through a network to form a unified computing resource pool. These nodes are considered as a whole at the software level and provide services to the outside world together. Cluster technology improves computing power, availability and scalability through parallel processing and resource sharing.

[0064] Therefore, in this embodiment, the resource usage of each node is collected in real time from the cluster environment, and the acquired monitoring data is summarized and stored.

[0065] In some embodiments of the present invention, Prometheus and node Exporter are used as data collection tools to ensure that the resource usage of each node in the cluster can be fully and accurately obtained. The collected computing power resource indicators include but are not limited to CPU usage, GPU usage, memory occupancy, disk I / O, network throughput, etc. These indicators can fully reflect the resource usage of the system.

[0066] It is worth noting that the collection period is configurable. Preferably, the default setting is to collect data every 10 seconds to balance the system performance overhead while ensuring the real-time nature of the data. After the collected data is aggregated and stored, periodic and random services will also be marked.

[0067] Specifically, a time series analysis is performed on the computing power resource data to extract the attribute characteristics of the computing power resource data. Correspondingly, the extracted CPU usage characteristics include: overall CPU usage, which reflects the overall CPU usage of the system or application within a specific time; user-state CPU usage: the percentage of CPU time consumed by the user process, reflecting the CPU occupancy by the user process; and CPU load characteristics, that is, the number of processes waiting for CPU resources in the system queue within a specific time, reflecting the current CPU work pressure. The extracted memory occupancy characteristics include overall memory usage, available memory or active memory percentage, etc. The corresponding network bandwidth characteristics include sending and receiving byte rates, network delays and jitter, etc., as well as disk I / O characteristics, such as read and write rates, response time, and I / O waiting time.

[0068] Determine the target marking rule according to the above attribute characteristics, mark the service flow according to the target marking rule, and determine the type data of each service flow. Specifically:

[0069] Establish marking rules for periodic and random services. If the data characteristics of computing resources (such as CPU usage, memory usage, etc.) show obvious periodic fluctuations, and the fluctuation pattern matches the periodic activities of the service flow, the service flow will be marked as periodic service. Unmarked services will be uniformly considered as random services, providing a basis for subsequent resource prediction and allocation.

[0070] This embodiment aims to achieve intelligent collection of computing resources in a cluster environment through dynamic monitoring and prediction technology, fully considering the real-time and flexibility of business flow data and computing power scheduling, so that the acquired data can meet the needs of different application scenarios.

[0071] S2. Design a corresponding distribution function according to the arrival rate of each type of data to analyze the CPU system efficiency of each business flow and obtain CPU system efficiency data.

[0072] The embodiment of the present invention models and converts the arrival distribution of business flows, analyzes the system efficiency of business flows, and can understand the distribution of various business flows in the system by counting the arrival rates of different business flows.

[0073] The distribution function is set according to the arrival rate of different types of business flows, and the impulse function is used to represent the burst characteristics of the business flow. Subsequently, the distribution function of the business flow is Laplace transformed, and the transfer function is used to describe the system characteristics. By counting the arrival proportion and system efficiency of different business flows, the system efficiency proportion of the business flow is calculated, so as to accurately evaluate the overall performance of the system. Specifically, the following steps are included:

[0074] According to the arrival frequency of different types of service flows, the distribution function of service flow arrival is set, and the burst characteristics of the service flow are obtained based on the designed distribution function. The distribution function G(t) is expressed by the following formula:

[0075]

[0076] Among them, λ1 is the periodic service arrival rate, λ2 is the random service arrival rate, δ is the impulse function, and t is the time.

[0077] In the embodiment of the present invention, the burst characteristics of the service flow are represented by an impulse function, and the expression of the impulse function δ is as follows:

[0078]

[0079] Perform Laplace transform on the set distribution function to obtain the system response characteristics of the service flow. For details, please refer to Figure 2 , Figure 2 Shown is a mathematical representation of the Laplace transform in one embodiment of the present invention.

[0080] After Laplace transform, transfer function can be used instead of differential equation to analyze and describe the characteristics of the system. The transfer function can reflect the response of the system to the input signal (i.e., business flow), thus helping to evaluate the performance of the system.

[0081] According to the burst characteristics and the system response characteristics, the arrival percentage of the business flow and the system efficiency are analyzed and counted to obtain the CPU system efficiency data. It can be understood that based on the arrival percentage and the system efficiency, the percentage of the system efficiency is calculated. This step helps to identify the business flow that contributes the most to the system efficiency and provides a basis for subsequent resource allocation. In this step, the CPU system efficiency data includes the system efficiency data and percentage data generated in the above process.

[0082] S3. Input the type data, the computing power resource data, the arrival rate and the CPU system efficiency data into the trained business flow computing power value prediction model to obtain the computing power demand result of the business flow.

[0083] It can be understood that by modeling and transforming the arrival distribution of business flows, the system efficiency of business flows is analyzed, and the system efficiency ratio of business flows is calculated. These results or related data (such as the type of business flows, arrival frequency, system efficiency, and original computing power resource data) can be used as sample data to input the computing power value prediction model for training. This helps to more accurately predict the computing power requirements of different business flows.

[0084] In some implementations of the embodiments of the present invention, the preprocessing of the above-mentioned data to be processed may include data cleaning and data transformation. Since the time interval set during data collection is 10 seconds, there will be no repeated data, so only missing value filling is required during data cleaning. For example,

[0085] The missing values ​​are filled by the mode, and the mode of the 1-day data of the resource is taken as the interpolation data. Assume that the dth i If there is missing data, the corresponding filling value calculation formula is as follows:

[0086] mode(d1,d2,…d m )

[0087] Among them, mode is the function for finding the mode, d i , is the data of the resource at the i-th moment, m is the one-day data obtained from the corresponding historical monitoring indicator data, and the default value of m is 8640.

[0088] After preprocessing the data to be processed, a first target data set is obtained, and the first target data set is divided into a training set and a test set; the training set is input into the neural network model for training, and the test set is input into the neural network model for testing to obtain the business flow computing power value prediction model.

[0089] It is understandable that the Long Short-Term Memory (LSTM) algorithm is a variant of the traditional recurrent neural network (RNN). LSTM combines short-term memory and long-term memory by adding input, output, and forget gates. At the same time, the control of LSTM gates also solves the problem of gradient disappearance or gradient explosion that may occur in RNN during back propagation.

[0090] Since the LSTM network has advantages in processing time series data and solving long-term dependency problems, it is an ideal choice for predicting computing power requirements. Therefore, the embodiment of the present invention preferably uses LSTM as a neural network model for training, and the first target data set can be divided into a 70% training set and a 30% test set. In the process of inputting the test set into the neural network model for testing respectively, the performance of the model can be evaluated by comparing the predicted results of the model with the actual results. Commonly used evaluation indicators include accuracy, recall rate, F1 score, etc. After training and testing, the business flow computing power value prediction model is obtained, and the type data, arrival rate, CPU system efficiency data, and computing power resource data are input into it to predict the computing power value of the business flow.

[0091] S4. When deploying the service flow to the target computer cluster, the computing resources are scheduled and arranged according to the computing power demand result. Figure 3 , Figure 3 The figure shows a schematic diagram of the computing power scheduling process in one embodiment of the present invention. Specifically, the following steps are included:

[0092] First, the corresponding application is determined according to the attributes of the business flow, and the business flows in the same application are grouped together.

[0093] Since a variety of terminal devices have a plurality of different types of service flows, in this embodiment, the corresponding application programs are determined according to the different types of service flows, and the service flows belonging to the same application program are grouped together.

[0094] Next, the sets are prioritized according to the priority attributes of the applications and a weight is assigned to each set, and the weight is modified according to the real-time characteristics of the service flow.

[0095] It is understandable that each application may have different priority attributes, such as real-time requirements, business importance, etc. According to the priority attributes of the application, the set of business flows belonging to the application is sorted, and the set is labeled with a weight value. The business flows in each set are mapped to the application of the corresponding container according to the weight value of the set.

[0096] Furthermore, according to the real-time requirements of various types of business flows, the weight values ​​of the corresponding sets are modified. In the actual operation process, the real-time requirements of some business flows may change. For example, some originally low-priority business flows may suddenly become urgent and require a higher processing priority. Therefore, it is necessary to monitor the real-time requirements of business flows in real time and modify the weight values ​​of the sets according to these changes.

[0097] Finally, the business flows in the set are mapped to the applications in the corresponding containers according to the corrected weights, and the total computing power values ​​of the applications corresponding to the same type of business flows are statistically calculated according to the computing power demand results; the applications are merged into the corresponding containers according to the total computing power values ​​to configure the corresponding computing power resources for scheduling.

[0098] It can be understood that in this step, this embodiment matches the total computing power value of each application with the preset container specifications, selects the container specification that best meets its computing power requirements, and allocates computing power resources that match the total computing power value of the application in each container. This includes configuring a specific number of CPU cores, memory size, storage quota, and network bandwidth for the container. Through the above process, computing power resources can be managed and scheduled more efficiently and flexibly to meet the computing power requirements of different types of business flows and improve the overall performance of the system and user experience.

[0099] The following provides a specific computing power scheduling algorithm to describe the above computing power scheduling and resource scheduling process in detail:

[0100] This embodiment is a computing power orchestration algorithm based on Kubernetes container orchestration technology, which specifically includes the following steps:

[0101] 1) Use Kubernetes to orchestrate containers and form corresponding service function paths through service function chains; map business flows to the corresponding service function paths through classifiers.

[0102] It is understandable that the container orchestration capabilities of Kubernetes can be used to dynamically manage and schedule computing resources. Through the service function chain (SFC) technology, a series of network service functions are connected in a specific order to form a service function path (SFP) to meet the needs of different business flows.

[0103] In SFC, if there is a set of service functions {SF1, SF2,…, SF n}, the service function path SFP is the ordered arrangement of these service functions: SFP = [SF i1 ,SF i2 ,…,SF im ] where i1,i2,...,im are the indexes of the service functions, indicating that the traffic passes through these service functions in sequence. For a given flow f, the classifier maps it to a specific service function path:

[0104] Classifier(f)=SFP k ,

[0105] Among them, SFP k is the service function path applicable to flow f.

[0106] In each service function SF j On the other hand, the flow f passes through the processing function P j :f out =P j (f in ),

[0107] Among them, f in Enter SF j The flow rate, f out It is through SF j Processed traffic. Through the above process, the sequential processing and service delivery of traffic in the service function chain are realized.

[0108] 2) Deploy custom resources and plug-ins in Kubernetes to schedule computing resources.

[0109] In an embodiment of the present invention, custom resources and plug-ins are deployed in Kubernetes to adjust the network topology weight, and bandwidth resource components and performance monitoring components are deployed to optimize the scheduling of computing resources.

[0110] It can be understood that the created custom resources (CRDs) are deployed in the Kubernetes cluster. For example, two custom resources are deployed: SvcGroup CRD: used for service chaining, abstracting the service topology, and used to maintain the dependencies of application microservices; NetworkTopology CRD: abstracting the network infrastructure, used to establish network weights between regions and zones in the cluster.

[0111] The deployed components include a bandwidth resource component, which is published as an extended resource so that existing filtering / scoring plug-ins can consider bandwidth allocation to optimize resource utilization, and a performance monitoring component, which is used to monitor network performance in the cluster and dynamically adjust network topology information.

[0112] Among them, the designed and deployed plug-ins include QueueSort plug-in: TopologicalSort and Filter&Score plug-in: NetworkOverhead.

[0113] In summary, the embodiments of the present invention determine the type data of each business flow by real-time monitoring and periodically collecting the computing power resource data of each virtual node in the target computer cluster, and design a corresponding distribution function according to the arrival rate of the business flow to calculate the system efficiency ratio of the business flow, so as to accurately evaluate the overall performance of the system; predict the computing power value of the business flow through the constructed business flow computing power value prediction model, deploy the business flow to the corresponding container in the cluster, and schedule the computing power resources according to the computing power value; manage and schedule computing power resources more efficiently and flexibly, meet the computing power requirements of different types of business flows, and realize the rational use and optimization of system resources.

[0114] An embodiment of the present invention provides a computing resource scheduling system based on a dynamic generation strategy. For details, see Figure 4 , Figure 4 The structure block diagram of a computing resource orchestration system based on a dynamic generation strategy in one embodiment of the present invention is shown, including:

[0115] The data acquisition module M1 is used to monitor in real time and periodically collect the computing power resource data of each virtual node in the target computer cluster, and determine the type data of each business flow based on the computing power resource data;

[0116] The service flow analysis module M2 is used to design a corresponding distribution function according to the arrival rate of each type of data, so as to analyze the CPU system efficiency of each service flow and obtain CPU system efficiency data;

[0117] The computing power prediction module M3 is used to input the type data, the computing power resource data, the arrival rate and the CPU system efficiency data into the trained business flow computing power value prediction model to obtain the computing power demand result of the business flow;

[0118] The resource scheduling module M4 is used to schedule computing resources according to the computing power demand result when deploying the business flow to the target computer cluster.

[0119] As an example, an embodiment of the present invention uses the data acquisition module M1 to perform a time series analysis on the computing power resource data to extract the attribute characteristics of the computing power resource data; and determines a target marking rule based on the attribute characteristics; and marks the business flow according to the target marking rule to determine the type data of each business flow.

[0120] As an example, the embodiment of the present invention uses the service flow analysis module M2 to specifically:

[0121] Acquire the burst characteristics of the service flow based on the designed distribution function;

[0122] Performing Laplace transform on the set distribution function to obtain system response characteristics of the service flow;

[0123] According to the burst characteristics and the system response characteristics, the arrival ratio of the business flow and the system efficiency are analyzed and counted to obtain the CPU system efficiency data.

[0124] As an example, in the computing power prediction module M3, the process of building the business flow computing power value prediction model includes:

[0125] Preprocessing the type data, the computing power resource data, the arrival rate, and the CPU system efficiency data to obtain a first target data set;

[0126] Dividing the first target data set into a training set and a test set;

[0127] The training set is input into the neural network model for training, and the test set is input into the neural network model for testing to obtain the business flow computing power value prediction model.

[0128] As an example, the embodiment of the present invention uses the resource scheduling module M4 to determine the corresponding application according to the attributes of the service flow, and aggregate the service flows in the same application;

[0129] Prioritizing the sets according to the priority attributes of the applications and assigning weights to each of the sets, and modifying the weights based on the real-time characteristics of the traffic flows;

[0130] Mapping the business flows in the set to the application of the corresponding container based on the modified weights, and statistically calculating the total computing power value of the application corresponding to the same type of business flows according to the computing power demand results;

[0131] The application programs are grouped into corresponding containers according to the total computing power value, so as to configure corresponding computing power resources for scheduling.

[0132] The technical features and technical effects of the computing power resource orchestration system based on dynamic generation strategy proposed in the embodiment of the present invention are the same as the technical features and technical effects of the computing power resource orchestration method based on dynamic generation strategy proposed in the embodiment of the present invention, and will not be repeated here.

[0133] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A computing resource scheduling method based on dynamic generation strategy, characterized in that: include: Real-time monitoring and periodic collection of computing power resource data of each virtual node in the target computer cluster, and determining type data of each business flow based on the computing power resource data; Design a corresponding distribution function according to the arrival rate of each type of data to analyze the CPU system efficiency of each business flow and obtain CPU system efficiency data; Input the type data, the computing power resource data, the arrival rate and the CPU system efficiency data into the trained business flow computing power value prediction model to obtain the computing power demand result of the business flow; When deploying the business flow to the target computer cluster, the computing resources are scheduled and arranged according to the computing demand result.

2. The computing resource scheduling method based on dynamic generation strategy as claimed in claim 1, characterized in that: The determining the type data of each service flow based on the computing power resource data includes: Performing time series analysis on the computing power resource data to extract attribute features of the computing power resource data; Determine a target marking rule according to the attribute characteristics; The service flows are marked according to the target marking rule to determine the type data of each service flow.

3. The computing resource scheduling method based on dynamic generation strategy as claimed in claim 1, characterized in that: The corresponding distribution function is designed according to the arrival rate of each type of data to analyze the CPU system efficiency of each business flow to obtain CPU system efficiency data, including: Acquire the burst characteristics of the service flow based on the designed distribution function; Performing Laplace transform on the set distribution function to obtain system response characteristics of the service flow; According to the burst characteristics and the system response characteristics, the arrival ratio of the business flow and the system efficiency are analyzed and counted to obtain the CPU system efficiency data.

4. The computing resource scheduling method based on dynamic generation strategy as claimed in claim 1, characterized in that: The process of building the business flow computing power value prediction model includes: Preprocessing the type data, the computing power resource data, the arrival rate, and the CPU system efficiency data to obtain a first target data set; Dividing the first target data set into a training set and a test set; The training set is input into the neural network model for training, and the test set is input into the neural network model for testing to obtain the business flow computing power value prediction model.

5. The computing resource scheduling method based on dynamic generation strategy as claimed in claim 1, characterized in that: When deploying the service flow to the target computer cluster, scheduling computing resources according to the computing power demand result includes: Determine the corresponding application according to the attributes of the business flow, and group the business flows in the same application; Prioritizing the sets according to the priority attributes of the applications and assigning weights to each of the sets, and modifying the weights according to the real-time characteristics of the traffic flows; Mapping the business flows in the set to the application of the corresponding container according to the modified weights, and statistically calculating the total computing power value of the application corresponding to the same type of business flows according to the computing power demand results; The application programs are grouped into corresponding containers according to the total computing power value, so as to configure corresponding computing power resources for scheduling.

6. A computing resource scheduling system based on dynamic generation strategy, characterized in that: include: A data acquisition module, used to monitor in real time and periodically collect computing power resource data of each virtual node in the target computer cluster, and determine the type data of each business flow based on the computing power resource data; A service flow analysis module is used to design a corresponding distribution function according to the arrival rate of each type of data, so as to analyze the CPU system efficiency of each service flow and obtain CPU system efficiency data; A computing power prediction module, used to input the type data, the computing power resource data, the arrival rate and the CPU system efficiency data into a trained business flow computing power value prediction model to obtain a computing power demand result of the business flow; The resource scheduling module is used to schedule computing resources according to the computing power demand result when deploying the business flow to the target computer cluster.

7. The computing resource scheduling system based on dynamic generation strategy as claimed in claim 6, characterized in that: The data acquisition module is specifically used for: Performing time series analysis on the computing power resource data to extract attribute features of the computing power resource data; Determine a target marking rule according to the attribute characteristics; The service flows are marked according to the target marking rule to determine the type data of each service flow.

8. The computing resource scheduling system based on dynamic generation strategy as claimed in claim 6, characterized in that: The business flow analysis module is specifically used for: Acquire the burst characteristics of the service flow based on the designed distribution function; Performing Laplace transform on the set distribution function to obtain system response characteristics of the service flow; According to the burst characteristics and the system response characteristics, the arrival ratio of the business flow and the system efficiency are analyzed and counted to obtain the CPU system efficiency data.

9. The computing resource scheduling system based on dynamic generation strategy as claimed in claim 6, characterized in that: The process of building the business flow computing power value prediction model includes: Preprocessing the type data, the computing power resource data, the arrival rate, and the CPU system efficiency data to obtain a first target data set; Dividing the first target data set into a training set and a test set; The training set is input into the neural network model for training, and the test set is input into the neural network model for testing to obtain the business flow computing power value prediction model.

10. The computing resource scheduling system based on dynamic generation strategy according to claim 6, characterized in that: The resource scheduling module is specifically used for: Determine the corresponding application according to the attributes of the business flow, and group the business flows in the same application; Prioritizing the sets according to the priority attributes of the applications and assigning weights to each of the sets, and modifying the weights according to the real-time characteristics of the traffic flows; Mapping the business flows in the set to the application of the corresponding container according to the modified weights, and statistically calculating the total computing power value of the application corresponding to the same type of business flows according to the computing power demand results; The application programs are grouped into corresponding containers according to the total computing power value, so as to configure corresponding computing power resources for scheduling.

Images