Dynamic scheduling method for cloud computing resource pool

Abstract

The invention discloses a dynamic scheduling method for a cloud computing resource pool. The method comprises the following steps: acquiring a time sequence of load data of each resource node in the cloud computing resource pool in a preset time period; performing data preprocessing on the time sequence of the load data of each resource node in the preset time period to obtain a sequence of load time sequence input vectors of the resource nodes; performing time sequence analysis on the sequence of the resource node load time sequence input vectors to obtain a sequence of resource node load time sequence feature vectors; and determining the task allocation priority of each resource node in the cloud computing resource pool based on the node load semantic metric value of each resource node load time sequence feature vector in the sequence of the resource node load time sequence feature vectors. Therefore, the load state of each resource node can be evaluated more accurately, so that the task allocation priority of each resource node in the resource pool is dynamically determined, and the task execution efficiency and the overall resource utilization rate are improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent cloud computing technology, and more specifically, to a cloud computing resource pool dynamic scheduling method. Background Art

[0002] Cloud computing is an Internet-based computing model that provides computing resources as a service to users, allowing them to use resources on demand without having to worry about the physical location and configuration of the resources. Cloud computing resource pool is a technology that organizes multiple physical or virtual computing resources together to form a unified, dynamically adjustable resource supply platform.

[0003] The scheduling of cloud computing resource pool refers to the reasonable allocation and adjustment of tasks of each resource node in the resource pool according to user needs and resource status, so as to improve resource utilization and service quality and reduce operation and maintenance costs and energy consumption. However, due to the large number of resource nodes in the cloud computing resource pool and the heterogeneity, dynamicity and uncertainty between nodes, the dynamic scheduling problem of cloud computing resource pool has the characteristics of high dimension, nonlinearity and complexity, which is difficult to be effectively solved by traditional scheduling methods.

[0004] Therefore, an optimized dynamic scheduling method for cloud computing resource pool is expected. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The present application provides a method for dynamic scheduling of a cloud computing resource pool, which obtains the time series of load data of each resource node in the cloud computing resource pool in a predetermined time period; performs data preprocessing on the time series of load data of each resource node in the predetermined time period to obtain a sequence of resource node load timing input vectors; performs timing analysis on the sequence of resource node load timing input vectors to obtain a sequence of resource node load timing feature vectors; and determines the task allocation priority of each resource node in the cloud computing resource pool based on the node load semantic metric value of each resource node load timing feature vector in the sequence of resource node load timing feature vectors. In this way, the load status of each resource node can be evaluated more accurately, thereby realizing the dynamic determination of the task allocation priority of each resource node in the resource pool, and improving the execution efficiency of tasks and the overall resource utilization.

[0006] In a first aspect, a method for dynamic scheduling of a cloud computing resource pool is provided, which includes:

[0007] Obtaining a time series of load data of each resource node in a cloud computing resource pool in a predetermined time period;

[0008] Performing data preprocessing on the time series of the load data of each resource node in a predetermined time period to obtain a sequence of resource node load time series input vectors;

[0009] Performing a timing analysis on the sequence of the resource node load timing input vectors to obtain a sequence of resource node load timing characteristic vectors;

[0010] Based on the node load semantic metric value of each resource node load time series feature vector in the sequence of the resource node load time series feature vector, the task allocation priority of each resource node in the cloud computing resource pool is determined.

[0011] Compared with the prior art, the cloud computing resource pool dynamic scheduling method provided by the present application obtains the time series of the load data of each resource node in the cloud computing resource pool in a predetermined time period; performs data preprocessing on the time series of the load data of each resource node in the predetermined time period to obtain a sequence of resource node load timing input vectors; performs timing analysis on the sequence of resource node load timing input vectors to obtain a sequence of resource node load timing feature vectors; and determines the task allocation priority of each resource node in the cloud computing resource pool based on the node load semantic metric value of each resource node load timing feature vector in the sequence of resource node load timing feature vectors. In this way, the load status of each resource node can be evaluated more accurately, thereby realizing the dynamic determination of the task allocation priority of each resource node in the resource pool, and improving the execution efficiency of tasks and the overall resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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.

[0013] Figure 1 The present invention is a flowchart of a method for dynamic scheduling of a cloud computing resource pool according to an embodiment of the present application.

[0014] Figure 2 Schematic diagram of the architecture of a cloud computing resource pool dynamic scheduling method according to an embodiment of the present application.

[0015] Figure 3 4 is a block diagram of a cloud computing resource pool dynamic scheduling system according to an embodiment of the present application.

[0016] Figure 4 The present invention is a schematic diagram of a scenario of a method for dynamic scheduling of a cloud computing resource pool according to an embodiment of the present application. DETAILED DESCRIPTION

[0017] The following will describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0018] Unless otherwise specified, all technical and scientific terms used in the embodiments of the present application have the same meaning as those commonly understood by those skilled in the art of the present application. The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit the scope of this application.

[0019] In the embodiments of the present application, it should be noted that, unless otherwise specified and limited, the term "connection" should be understood in a broad sense. For example, it can be an electrical connection or a connection between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.

[0020] It should be noted that the terms "first\second\third" involved in the embodiments of the present application are only used to distinguish similar objects, and do not represent a specific order for the objects. It is understandable that the specific order or sequence of "first\second\third" can be interchanged where permitted. It should be understood that the objects distinguished by "first\second\third" can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0021] Cloud computing is an Internet-based computing model that provides computing resources (such as storage, computing power, databases, networks, etc.) as services to users, allowing them to obtain and use these resources on demand without having to worry about the specific physical location and configuration of the resources. This model is usually implemented through virtualization technology, and users can access resources on the cloud platform through the Internet, thereby achieving flexible and scalable computing capabilities.

[0022] Cloud computing usually includes the following service models: Infrastructure as a Service, which provides virtualized computing resources, storage and network infrastructure, on which users can deploy and run operating systems and applications. Platform as a Service, which provides the platform environment required for developing, running and managing applications based on the infrastructure, including development tools, database management systems, middleware, etc. Software as a Service, which provides software applications through the Internet. Users do not need to install and maintain the software, but only need to access it through the Internet to use it.

[0023] The advantages of cloud computing include: users can quickly acquire and release computing resources as needed, achieving elastic expansion and contraction; users can pay based on actual usage, avoiding investment in expensive hardware equipment and facilities; cloud platforms are usually highly available and fault-tolerant, and can provide continuous services; users can access cloud services anytime and anywhere via the Internet without having to worry about physical location.

[0024] Cloud computing resource pool is a technology that organizes multiple physical or virtual computing resources together to form a unified, dynamically adjustable resource supply platform. This technology allows users to dynamically obtain and use computing resources when needed without having to worry about the specific physical location and configuration of the resources. Cloud computing resource pools are usually composed of a variety of hardware devices and software components to provide various types of resources such as computing, storage, and networking.

[0025] Cloud computing resource pools usually use virtualization technology to abstract the computing, storage, and network resources of physical devices into virtual resources, thereby achieving flexible resource allocation and management. Cloud computing resource pools are usually equipped with automated management systems that can automatically adjust resource allocation according to demand, and achieve dynamic scaling and load balancing of resources. Cloud computing resource pools provide a unified management interface through which users can manage and configure resources to achieve unified management and monitoring of resources. One of the design goals of cloud computing resource pools is to achieve elastic scaling and scalability of resources, and to quickly expand or shrink the scale of resources according to demand. Cloud computing resource pools usually use redundant backup and fault-tolerant mechanisms to ensure high availability and stability of resources.

[0026] The scheduling of cloud computing resource pool refers to the reasonable allocation and adjustment of tasks of each resource node in the resource pool according to user needs and resource status, so as to improve resource utilization and service quality and reduce operation and maintenance costs and energy consumption. The scheduling of resource pool is a key management and optimization process, which involves task allocation, load balancing, resource optimization and fault handling.

[0027] The cloud computing resource pool needs to allocate tasks to appropriate resource nodes for execution based on user requests and the characteristics of tasks, which involves comprehensive consideration of factors such as task priority, resource requirements, and node load. The cloud computing resource pool needs to achieve load balancing to ensure that the loads of various resource nodes are relatively balanced and to avoid excessive or insufficient utilization of resources to improve overall resource utilization. The cloud computing resource pool needs to dynamically adjust the allocation and configuration of resources based on the characteristics of tasks and the status of resources to optimize resource utilization and performance. The cloud computing resource pool needs to implement a fault detection and processing mechanism to promptly detect and handle resource node failures to ensure service continuity and reliability. The cloud computing resource pool needs to consider energy consumption management, reduce energy consumption, and improve energy efficiency through effective resource scheduling and management. The cloud computing resource pool needs to support elastic scaling, dynamically expand or shrink resource scale according to demand to cope with different workload requirements.

[0028] There are a large number of resource nodes in the cloud computing resource pool, and there are characteristics such as heterogeneity, dynamics and uncertainty between the nodes, which makes the dynamic scheduling problem of the cloud computing resource pool difficult and challenging. Among them, the number of resource nodes in the cloud computing resource pool is huge, and the state and characteristics of the resources have high-dimensional and nonlinear characteristics, which makes the search space of the scheduling problem huge and complex. The state and load of the resource nodes in the cloud computing resource pool are dynamic and uncertain. With the change of time and task load, the resource state may change continuously, resulting in increased uncertainty in scheduling decisions. The resource nodes in the cloud computing resource pool may have different computing power, storage capacity, network bandwidth and other characteristics, making task scheduling and resource allocation more complicated. Due to the large number of resource nodes, traditional scheduling methods may find it difficult to effectively handle the scheduling problem of large-scale resource pools, resulting in challenges to the efficiency and scalability of scheduling algorithms. The dynamic scheduling of cloud computing resource pools usually involves multiple optimization objectives, such as resource utilization, task response time, energy consumption, etc., which makes the scheduling problem a multi-objective optimization problem and increases the complexity of scheduling decisions.

[0029] Faced with these difficulties and challenges, traditional scheduling methods may not be able to effectively solve the dynamic scheduling problem of cloud computing resource pools. Therefore, it is necessary to use advanced algorithms and technologies, such as machine learning, deep learning, and reinforcement learning, to deal with the high-dimensional, nonlinear and complex characteristics of resource pool scheduling, as well as challenges such as dynamics, uncertainty and heterogeneity. These advanced technologies can better adapt to the dynamic environment of resource pools, improve the intelligence and adaptability of scheduling decisions, and thus more effectively optimize resource utilization, reduce energy consumption, and improve service quality and user experience.

[0030] Figure 1 The present invention is a flowchart of a method for dynamic scheduling of a cloud computing resource pool according to an embodiment of the present application. Figure 2FIG. 1 is a schematic diagram of the architecture of a cloud computing resource pool dynamic scheduling method according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the cloud computing resource pool dynamic scheduling method includes: 110, obtaining the time series of load data of each resource node in the cloud computing resource pool in a predetermined time period; 120, performing data preprocessing on the time series of load data of each resource node in the predetermined time period to obtain a sequence of resource node load timing input vectors; 130, performing timing analysis on the sequence of resource node load timing input vectors to obtain a sequence of resource node load timing feature vectors; 140, determining the task allocation priority of each resource node in the cloud computing resource pool based on the node load semantic measurement value of each resource node load timing feature vector in the sequence of resource node load timing feature vectors.

[0031] In step 110, the frequency and time interval of data collection are determined to ensure that sufficiently detailed load data is obtained; to ensure the accuracy and completeness of the load data, it may be necessary to consider the delay in data collection and data loss. In this way, the time series data of the resource node load provides a basis for subsequent analysis and processing. In step 120, data cleaning is performed to process missing values, outliers and noise data. Data smoothing is performed to smooth the load data to eliminate short-term fluctuations and retain long-term trends. Data standardization is performed to standardize the load data so that the load data of different resource nodes are comparable. In this way, a cleaned and processed load time series input vector is obtained, which provides high-quality input data for subsequent time series analysis. In step 130, feature extraction is performed on the load time series data, which may include features such as mean, variance, and periodicity. Time series patterns such as periodicity, trend or burst in the load data are identified. In this way, a time series feature vector describing the characteristics of the resource node load is obtained, which helps to have a deeper understanding and analysis of the load data. In step 140, a suitable load semantic measurement method is designed to quantify the importance and urgency of the resource node load. According to the load semantic measurement value, the task allocation priority of each resource node is determined. Determining the priority of task allocation by the load semantic measurement value helps to achieve a reasonable allocation of resource node load and improve the overall performance and efficiency of the system.

[0032] In response to the above technical problems, the technical concept of the present application is to utilize and extract the timing characteristic information of the load data of each resource node in a predetermined time period in a cloud computing resource pool, and measure the load similarity relationship of each resource node relative to all resource nodes, so as to more accurately evaluate the load status of each resource node, thereby realizing the dynamic determination of the task allocation priority of each resource node in the resource pool, and improving the task execution efficiency and overall resource utilization.

[0033] By extracting time series feature information, such as average load, load volatility, load periodicity, etc., the load status of resource nodes can be described more comprehensively, providing more information support for subsequent load similarity analysis and determination of task allocation priorities. By measuring the load similarity relationship of each resource node relative to all resource nodes, the patterns and laws of load behavior can be discovered, thereby more accurately evaluating the load status of resource nodes, helping to identify abnormal load behavior or load differences between different resource nodes, and providing a more refined basis for determining task allocation priorities.

[0034] Based on more accurate load status assessment, the task allocation priority of each resource node can be dynamically determined, so that tasks can be more reasonably allocated to resource nodes with lighter loads, thereby improving task execution efficiency and overall resource utilization. By dynamically determining the task allocation priority, the load of resource nodes can be more balanced, avoiding performance bottlenecks and resource waste caused by unbalanced resource node loads, thereby improving overall resource utilization efficiency.

[0035] Based on this, in the technical solution of the present application, firstly, the time series of the load data of each resource node in the cloud computing resource pool in a predetermined time period is obtained; and the time series of the load data of each resource node in the predetermined time period is arranged as a resource stage load time series input vector according to the time dimension to obtain a sequence of resource node load time series input vectors. It should be understood that in the cloud computing resource pool, the load of each resource node changes over time. By obtaining the time series of the load data of each resource node in the predetermined time period, the complete information of the load data changing over time can be obtained. That is, the time series of the load data of each resource node in the predetermined time period contains the characteristics of the dynamic change mode, trend and periodicity of the load of each resource node, and these characteristic information can help the system better understand the change law of resource load. Secondly, the time series of the load data of each resource node in the predetermined time period is arranged as a sequence of resource stage load time series input vectors according to the time dimension, and a set of ordered load time series input vectors can be formed to describe and characterize the dynamic changes of the load data of each resource node. At the same time, by converting the load data of discrete time series distribution into vector representation, it is also convenient for subsequent model reading and processing.

[0036] In a specific embodiment of the present application, data preprocessing is performed on the time series of load data of each resource node in a predetermined time period to obtain a sequence of resource node load timing input vectors, including: arranging the time series of load data of each resource node in a predetermined time period into resource stage load timing input vectors according to the time dimension to obtain a sequence of resource node load timing input vectors.

[0037] Arranging the load data of resource nodes in a predetermined time period into a sequence of time series input vectors according to the time dimension helps to integrate time-related load information into a unified data structure, which is convenient for subsequent time series analysis and processing. By integrating the sequence of resource node load time series input vectors, it is more convenient to extract time series features, such as average load, load volatility, load periodicity, etc., to provide more comprehensive time series feature information for subsequent load similarity analysis and determination of task allocation priority. Integrating the sequence of resource node load time series input vectors helps to more comprehensively understand the load state changes of each resource node at different time points, so as to more accurately evaluate the load status of the resource node. Based on the sequence of integrated load time series input vectors, the task allocation priority of each resource node can be determined more accurately, so that tasks can be more reasonably allocated to resource nodes with lighter loads, improving the execution efficiency of tasks and overall resource utilization.

[0038] Then, the sequence of resource node load timing input vectors is passed through a load timing feature extractor based on a one-dimensional convolutional layer to obtain a sequence of resource node load timing feature vectors. That is, a more representative load timing feature distribution is extracted from the sequence of resource node load timing input vectors. Specifically, each resource node load timing input vector may contain a large amount of noise and redundant information. Through the processing of the one-dimensional convolutional layer, local abstract features of the timing data can be extracted from it, and the local patterns and correlation information of each resource node load timing input vector can be captured, which is helpful to analyze the periodicity, instantaneous changes and long-term trend characteristics of resource load.

[0039] In a specific embodiment of the present application, a timing analysis is performed on the sequence of the resource node load timing input vectors to obtain a sequence of resource node load timing feature vectors, including: using a deep learning network model to perform timing feature extraction on the sequence of the resource node load timing input vectors to obtain a sequence of the resource node load timing feature vectors.

[0040] Furthermore, the deep learning network model is a load time series feature extractor based on a one-dimensional convolutional layer. The load time series feature extractor based on a one-dimensional convolutional layer includes: an input layer, a convolutional layer, a pooling layer, an activation layer and an output layer.

[0041] In a specific embodiment of the present application, a deep learning network model is used to perform timing feature extraction on the sequence of resource node load timing input vectors to obtain the sequence of resource node load timing feature vectors, including: passing the sequence of resource node load timing input vectors through the load timing feature extractor based on the one-dimensional convolutional layer to obtain the sequence of resource node load timing feature vectors.

[0042] The one-dimensional convolution layer can effectively capture local patterns and features in time series data. In this way, key features in load time series data, such as load fluctuations and periodicity, can be extracted to better describe the load status of resource nodes. Through the processing of the convolution layer, the original time series input vector can be converted into a higher-level abstract feature representation, which helps to reduce the dimension and complexity of the data while retaining important load feature information. Feature extraction based on the one-dimensional convolution layer helps to identify patterns and regularities in load time series data, such as periodic changes, abnormal fluctuations, etc., so as to more accurately evaluate the load status of resource nodes. After obtaining the sequence of resource node load time series feature vectors, the task allocation priority of each resource node can be determined more accurately, so that tasks can be more reasonably allocated to resource nodes with lighter loads, improving the execution efficiency of tasks and overall resource utilization.

[0043] Next, the semantic weight value of the overall feature distribution of the load timing feature vectors of each resource node in the sequence of the resource node load timing feature vectors relative to the sequence of the resource node load timing feature vectors is calculated as the node load semantic metric value to obtain the sequence of node load semantic metric values. Here, the semantic weight value is used as the node load semantic metric value to measure and quantify the load correlation degree of each resource node relative to all resource nodes. In the technical solution of the present application, if the load similarity between the resource node load timing feature vector and the overall load feature distribution of all resource nodes is high, it means that the load characteristics of the current resource node and all resource nodes are more consistent, that is, the load of the current resource node is more consistent with the overall load characteristics; on the contrary, if the load similarity between the current resource node and the overall load feature distribution of all resource nodes is low, it means that the load characteristics of the current resource node and all resource nodes are greatly different, and there is a resource space for allocating tasks. It is worth mentioning that the above assumption assumes that there is no situation where a single resource node is overloaded when the overall load demand is low.

[0044] In a specific embodiment of the present application, based on the node load semantic measurement value of each resource node load timing feature vector in the sequence of the resource node load timing feature vector, the task allocation priority of each resource node in the cloud computing resource pool is determined, including: calculating the semantic weight value of the overall feature distribution of each resource node load timing feature vector in the sequence of the resource node load timing feature vector relative to the sequence of the resource node load timing feature vector as the node load semantic measurement value to obtain a sequence of node load semantic measurement values; based on the sorting of the sequence of the node load semantic measurement values, the task allocation priority of each resource node in the cloud computing resource pool is determined.

[0045] Furthermore, in a specific embodiment of the present application, the semantic weight value of each resource node load timing feature vector in the sequence of the resource node load timing feature vector relative to the overall feature distribution of the sequence of the resource node load timing feature vector is calculated as the node load semantic measurement value to obtain a sequence of node load semantic measurement values, including: cascading the sequence of the resource node load timing feature vector to obtain a cascaded feature vector; performing feature distribution correction on the cascaded feature vector to obtain a corrected cascaded feature vector; calculating the semantic weight value of each resource node load timing feature vector in the sequence of the resource node load timing feature vector relative to the corrected cascaded feature vector as the node load semantic measurement value to obtain a sequence of point load semantic measurement values.

[0046] In a specific embodiment of the present application, the semantic weight values ​​of each resource node load timing feature vector in the sequence of the resource node load timing feature vector relative to the corrected cascade feature vector are calculated as the node load semantic metric value to obtain the sequence of the point load semantic metric values, including: using the following semantic weight formula to calculate the semantic weight values ​​of each resource node load timing feature vector in the sequence of the resource node load timing feature vector relative to the corrected cascade feature vector to obtain the sequence of the node load semantic metric values; wherein the semantic weight formula is:

[0047] s i =σ(A*h i +B*V′)

[0048] Where A is 1×N w The vector, N w is the dimension of the resource node load time series feature vector, h i is the load timing characteristic vector of the i-th resource node, V' is the corrected cascade characteristic vector, and B is 1×N h The vector, N h is the dimension of the corrected concatenated feature vector, σ is the Sigmoid function, s i is the i-th semantic weight value.

[0049] In the above technical solution, each resource node load timing feature vector in the sequence of resource node load timing feature vectors expresses the local timing correlation feature of the load data of the corresponding resource node, and in calculating the semantic weight value of each resource node load timing feature vector in the sequence of resource node load timing feature vectors relative to the overall feature distribution of the sequence of resource node load timing feature vectors, the feature distribution of multiple resource node load timing feature vectors in the sequence of resource node load timing feature vectors is taken as a whole to calculate the semantic weight value. However, in the case of integrating the local timing correlation features of each resource node load timing feature vector itself, the significance of the feature distribution information of each of the multiple resource node load timing feature vectors based on its local timing correlation feature distribution will be affected, so that when calculating the semantic weight value of each resource node load timing feature vector relative to the overall feature distribution of the sequence of resource node load timing feature vectors, the overall feature distribution of the sequence of resource node load timing feature vectors is difficult to stably focus on the significant local distribution of the feature, which affects the accuracy of the semantic weight value.

[0050] Based on this, the applicant of the present application takes the sequence of the resource node load timing characteristic vector as a whole, first cascades the sequence of the resource node load timing characteristic vector, and then corrects the cascaded characteristic vector obtained after the cascade, for example, denoted as V, which is expressed as: the cascaded characteristic vector is corrected for characteristic distribution using the following optimization formula to obtain the corrected cascaded characteristic vector; wherein the optimization formula is:

[0051]

[0052] in, and are the squares of the 1-norm and 2-norm of the concatenated feature vector V, respectively, L is the length of the concatenated feature vector V, and ω is a weight hyperparameter, V is the concatenated feature vector, v' i is the eigenvalue of the corrected concatenated eigenvector, v i is the eigenvalue of the concatenated eigenvector, and log represents a logarithmic function with base 2.

[0053] Specifically, by performing geometric registration of the high-dimensional feature manifold shape based on the scale and structural parameters of the cascaded feature vector V, we can focus on features with rich feature semantic information in the feature set composed of the eigenvalues ​​of the cascaded feature vector V, that is, distinguishable and stable interest features that represent dissimilarity based on local context information, to achieve significant labeling of the feature information of the cascaded feature vector V in the calculation process, thereby improving the calculation accuracy of the semantic weight value of the overall feature distribution of the sequence of resource node load timing feature vectors relative to the resource node load timing feature vectors.

[0054] Further, based on the order of the sequence of the node load semantic metric values, the task allocation priority of each resource node in the cloud computing resource pool is determined. Here, the task allocation priority of each resource node represents the priority order of the resource node to which the task is assigned. In this way, the execution efficiency of the task and the overall resource utilization are improved.

[0055] In summary, a dynamic scheduling method for a cloud computing resource pool based on an embodiment of the present application is explained, which utilizes and extracts the timing characteristic information of the load data of each resource node in the cloud computing resource pool within a predetermined time period, and measures the load similarity relationship of each resource node relative to all resource nodes, thereby more accurately evaluating the load status of each resource node, thereby achieving dynamic determination of the task allocation priority of each resource node in the resource pool, and improving the execution efficiency of tasks and overall resource utilization.

[0056] In one embodiment of the present application, Figure 3 FIG. 1 is a block diagram of a cloud computing resource pool dynamic scheduling system according to an embodiment of the present application. Figure 3 As shown, according to the cloud computing resource pool dynamic scheduling system 200 of the embodiment of the present application, it includes: a load data acquisition module 210, which is used to obtain the time series of load data of each resource node in the cloud computing resource pool in a predetermined time period; a data preprocessing module 220, which is used to perform data preprocessing on the time series of load data of each resource node in the predetermined time period to obtain a sequence of resource node load timing input vectors; a timing analysis module 230, which is used to perform timing analysis on the sequence of resource node load timing input vectors to obtain a sequence of resource node load timing feature vectors; a task allocation priority determination module 240, which is used to determine the task allocation priority of each resource node in the cloud computing resource pool based on the node load semantic measurement value of each resource node load timing feature vector in the sequence of resource node load timing feature vectors.

[0057] Here, those skilled in the art can understand that the specific functions and operations of each unit and module in the above-mentioned cloud computing resource pool dynamic scheduling system have been referred to above. Figure 1 to Figure 2The method for dynamic scheduling of cloud computing resource pools has been described in detail, and therefore, its repeated description will be omitted.

[0058] As described above, the cloud computing resource pool dynamic scheduling system 200 according to the embodiment of the present application can be implemented in various terminal devices, such as a server for dynamic scheduling of cloud computing resource pools. In one example, the cloud computing resource pool dynamic scheduling system 200 according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the cloud computing resource pool dynamic scheduling system 200 can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the cloud computing resource pool dynamic scheduling system 200 can also be one of the many hardware modules of the terminal device.

[0059] Alternatively, in another example, the cloud computing resource pool dynamic scheduling system 200 and the terminal device may also be separate devices, and the cloud computing resource pool dynamic scheduling system 200 may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0060] Figure 4 FIG. 1 is a schematic diagram of a scenario of a method for dynamically scheduling a cloud computing resource pool according to an embodiment of the present application. Figure 4 As shown, in this application scenario, first, the time series of load data of each resource node in the cloud computing resource pool in a predetermined time period is obtained (for example, Figure 4 Then, the time series of the acquired load data is input to a server (for example, Figure 4 In the S) shown in FIG. 1 , the server is capable of processing the time series of the load data based on a cloud computing resource pool dynamic scheduling algorithm to determine the task allocation priority of each resource node in the cloud computing resource pool.

[0061] It should also be noted that in the apparatus, device and method of the present application, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present application.

[0062] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0063] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A cloud computing resource pool dynamic scheduling method, characterized in that: include: Obtaining a time series of load data of each resource node in a cloud computing resource pool in a predetermined time period; Performing data preprocessing on the time series of the load data of each resource node in a predetermined time period to obtain a sequence of resource node load time series input vectors; Performing a timing analysis on the sequence of the resource node load timing input vectors to obtain a sequence of resource node load timing characteristic vectors; Based on the node load semantic metric value of each resource node load time series feature vector in the sequence of the resource node load time series feature vector, the task allocation priority of each resource node in the cloud computing resource pool is determined.

2. The method for dynamic scheduling of cloud computing resource pool according to claim 1, characterized in that: Performing data preprocessing on the time series of the load data of each resource node in a predetermined time period to obtain a sequence of resource node load time series input vectors, including: The time series of the load data of each resource node in a predetermined time period are arranged according to the time dimension as resource stage load timing input vectors to obtain a sequence of the resource node load timing input vectors.

3. The method for dynamic scheduling of cloud computing resource pool according to claim 2, characterized in that: Performing a timing analysis on the sequence of the resource node load timing input vectors to obtain a sequence of resource node load timing feature vectors, including: A deep learning network model is used to extract timing features of the sequence of the resource node load timing input vectors to obtain a sequence of the resource node load timing feature vectors.

4. The method for dynamic scheduling of cloud computing resource pool according to claim 3, characterized in that: The deep learning network model is a load time series feature extractor based on a one-dimensional convolutional layer.

5. The method for dynamic scheduling of cloud computing resource pool according to claim 4, characterized in that: The load time series feature extractor based on one-dimensional convolution layer includes: an input layer, a convolution layer, a pooling layer, an activation layer and an output layer.

6. The method for dynamic scheduling of cloud computing resource pool according to claim 5, characterized in that: Using a deep learning network model to extract time series features from a sequence of the resource node load time series input vectors to obtain a sequence of the resource node load time series feature vectors, including: The sequence of the resource node load timing input vectors is passed through the load timing feature extractor based on the one-dimensional convolution layer to obtain the sequence of the resource node load timing feature vectors.

7. The method for dynamic scheduling of cloud computing resource pool according to claim 6, characterized in that: Determining the task allocation priority of each resource node in the cloud computing resource pool based on the node load semantic metric value of each resource node load time series feature vector in the sequence of the resource node load time series feature vectors includes: Calculating the semantic weight value of each resource node load time series feature vector in the sequence of resource node load time series feature vectors relative to the overall feature distribution of the sequence of resource node load time series feature vectors as a node load semantic metric value to obtain a sequence of node load semantic metric values; Based on the order of the sequence of the node load semantic metric values, a task allocation priority of each resource node in the cloud computing resource pool is determined.

8. The method for dynamic scheduling of cloud computing resource pool according to claim 7, characterized in that: Calculating the semantic weight value of each resource node load time series feature vector in the sequence of resource node load time series feature vectors relative to the overall feature distribution of the sequence of resource node load time series feature vectors as a node load semantic metric value to obtain a sequence of node load semantic metric values, including: Cascading the sequences of the resource node load time series feature vectors to obtain a cascade feature vector; Performing feature distribution correction on the cascade feature vector to obtain a corrected cascade feature vector; The semantic weight value of each resource node load time series feature vector in the sequence of the resource node load time series feature vectors relative to the corrected cascade feature vector is calculated as the node load semantic metric value to obtain the sequence of the point load semantic metric values.

9. The method for dynamic scheduling of cloud computing resource pool according to claim 8, characterized in that: Calculating the semantic weight value of each resource node load time series feature vector in the sequence of resource node load time series feature vectors relative to the corrected cascade feature vector as the node load semantic metric value to obtain the sequence of point load semantic metric values, including: The semantic weight value of each resource node load time series feature vector in the sequence of the resource node load time series feature vector relative to the corrected cascade feature vector is calculated using the following semantic weight formula to obtain a sequence of the node load semantic measurement values; wherein the semantic weight formula is: s i =σ(A*h i +B*V′) Where A is 1×N w The vector, N w is the dimension of the resource node load time series feature vector, h i is the load timing characteristic vector of the i-th resource node, V' is the corrected cascade characteristic vector, and B is 1×N h The vector, N h is the dimension of the corrected concatenated feature vector, σ is the Sigmoid function, s i is the i-th semantic weight value.

Images