A Distributed Task Load Balancing Scheduling Method and System

Abstract

This invention relates to the field of load balancing technology and discloses a distributed task load balancing scheduling method and system. The method includes: obtaining performance requirement indicators of tasks to be executed; allocating tasks to be executed to server nodes in the server cluster network topology according to the performance requirement indicators of the tasks to be executed; obtaining the historical load change rate of server nodes; determining the load threshold corresponding to each server node according to the historical load change rate of server nodes; obtaining the real-time performance indicators of server nodes; determining the load score value according to the real-time performance indicators of server nodes; filtering out overloaded server nodes whose load score values ​​exceed the load threshold; and performing load balancing scheduling on the tasks in the overloaded server nodes. This invention can accurately calculate the optimal allocation node for each task, provide a refined task scheduling strategy for server cluster networks, and effectively improve load balancing.

Description

Technical Field

[0001] This application relates to the field of load balancing technology, and more specifically, to a distributed task load balancing scheduling method and system. Background Technology

[0002] Cloud platforms are built upon various application formats, storing digital resources within a service platform. Resource users can access the cloud platform to efficiently utilize cloud services and obtain relevant resources, regardless of time or location. Load balancing of cloud platform server nodes refers to the use of specific technologies and strategies to rationally distribute network requests or computing tasks across multiple server nodes, ensuring balanced resource allocation and improving system performance, reliability, and scalability.

[0003] However, due to the high dynamism and heterogeneity of cloud platform environments, and the different resource requirements of different tasks on server nodes, existing technologies cannot meet the diverse needs of tasks in the load balancing scheduling process of server nodes, resulting in unreasonable task allocation and low scheduling efficiency. Summary of the Invention

[0004] This invention provides a distributed task load balancing scheduling method and system to solve the problems of unreasonable task allocation and low scheduling efficiency in the load balancing scheduling of server nodes in the prior art. The method includes: obtaining the performance requirement indicators of the tasks to be executed; allocating the tasks to be executed to the server nodes in the server cluster network topology according to the performance requirement indicators of the tasks to be executed; obtaining the historical load change rate of the server nodes; determining the load threshold corresponding to each server node according to the historical load change rate of the server nodes; obtaining the real-time performance indicators of the server nodes; determining the load score value according to the real-time performance indicators of the server nodes; filtering out overloaded server nodes whose load score values ​​exceed the load threshold; and performing load balancing scheduling on the tasks in the overloaded server nodes.

[0005] Furthermore, the step of allocating tasks to be executed to server nodes in the server cluster network topology based on the performance requirement indicators of the tasks to be executed includes: obtaining the performance requirement indicators of the tasks to be executed, which include CPU requirement indicators, IO requirement indicators, network requirement indicators, and memory requirement indicators; establishing a center point and drawing rays evenly outward from the center point as indicator axes; normalizing the performance requirement indicators of the tasks to be executed and filling the normalized performance requirement indicators into the indicator axes; connecting the normalized performance requirement indicators on the indicator axes sequentially to obtain a performance requirement indicator radar chart of the tasks to be executed, and allocating tasks to be executed to server nodes based on the performance requirement indicator radar chart of the tasks to be executed.

[0006] Furthermore, the step of allocating tasks to be executed to server nodes based on the performance requirement radar chart of the tasks to be executed includes: obtaining the node performance indicators of the current server nodes, wherein the node performance indicators include CPU resource indicators, IO resource indicators, network resource indicators, and memory resource indicators; establishing a node resource indicator radar chart based on the node performance indicators; calculating the matching degree between the node resource indicator radar chart of each server node and the performance requirement radar chart of the tasks to be executed; and allocating tasks to be executed to server nodes based on the matching degree between the node resource indicator radar chart of each server node and the performance requirement radar chart of the tasks to be executed.

[0007] Further, the calculation of the matching degree between the node resource indicator radar map of each server node and the performance requirement indicator radar map of the task to be executed includes: aligning the center points of the node resource indicator radar map and the performance requirement indicator radar map of the task to be executed, and aligning the vertices of the two radar maps to obtain an overlap map of the node resource indicator radar map and the performance requirement indicator radar map; calculating the absolute distance difference between corresponding vertices in the overlap map, and determining the average absolute distance difference based on the absolute distance differences of all vertices in the overlap map; calculating the cosine value of the included angle between corresponding sides in the overlap map, and determining the average cosine value based on all included angle cosine values ​​in the overlap map; and multiplying the average absolute distance difference by the average cosine value to obtain the matching degree between the node resource indicator radar map and the performance requirement indicator radar map of the task to be executed.

[0008] Furthermore, determining the load threshold corresponding to each server node based on the historical load change rate of the server nodes includes: clustering each server node according to the historical load change rate of the server nodes to obtain a preset load threshold for the server nodes; determining the cluster center of the cluster corresponding to the server node based on the clustering results; and correcting the preset load threshold based on the cluster center of the cluster corresponding to the server node to obtain the corrected load threshold.

[0009] Furthermore, the step of clustering each server node based on its historical load change rate includes: establishing a sample dataset based on the historical load change rate of all server nodes; randomly selecting k initial cluster centers from the sample dataset; calculating the Manhattan distance from the sample data in the sample dataset to the initial cluster centers; dividing each server node into its corresponding cluster based on the Manhattan distance from the sample data in the sample dataset to the initial cluster centers; calculating the average value of the sample data within each cluster; recalculating the cluster centers based on the average value of the sample data within each cluster; repeating the above steps iteratively until the cluster centers no longer change or the number of iterations reaches the preset maximum number of iterations, thereby obtaining the clustering result of the server nodes.

[0010] Further, determining the load score based on the real-time performance metrics of server nodes includes: acquiring the real-time performance metrics of each server node in the historical server cluster network topology, wherein the real-time performance metrics include CPU operation metrics, IO usage metrics, network operation metrics, and memory usage metrics; establishing a training sample set based on the real-time performance metrics of each server node in the historical server cluster network topology; performing expert manual scoring on the real-time performance metrics in the training sample set to obtain manually labeled load score codes; establishing a deep neural network model to perform load score coding on the real-time performance metrics in the training sample set, calculating the loss value between the load score code output by the deep neural network and the manually labeled load capacity score code; iteratively training the deep neural network parameters by minimizing the loss value to obtain a deep neural network model that can encode the real-time performance metrics of server nodes into corresponding load score values; and inputting the real-time performance metrics of the current server node into the trained deep neural network model to obtain the corresponding load score value.

[0011] Furthermore, the load balancing scheduling of tasks in the overloaded server node includes: detecting the surrounding nodes of the overloaded server node in the server cluster network topology, calculating the replacement matching degree between the surrounding nodes and the task to be unloaded; when the replacement matching degree between the surrounding nodes and the task to be unloaded is greater than a first preset threshold, transferring the task to be unloaded to the corresponding surrounding node.

[0012] Further, calculating the replacement matching degree between surrounding nodes and the task to be uninstalled includes: calculating the replacement matching degree between surrounding nodes and the task to be uninstalled according to the replacement matching degree calculation formula, wherein the replacement matching degree calculation formula is as follows:

[0013] ,

[0014] in, To replace the matching degree, Let be the node load value of the i-th surrounding node. The load threshold is the adjusted value for the i-th surrounding nodes. Let be the network distance value between the i-th surrounding node and the overloaded server node. Let represent the matching degree between the radar chart of node resource indicators of the i-th surrounding nodes and the radar chart of performance requirement indicators of the task to be unloaded. To preset the standard matching degree, This is a range adjustment factor.

[0015] To achieve the above objectives, the present invention also provides a distributed task load balancing scheduling system, comprising: a task module, used to obtain performance requirement indicators of tasks to be executed, and to allocate tasks to be executed to server nodes in the server cluster network topology according to the performance requirement indicators of tasks to be executed; a threshold module, used to obtain the historical load change rate of server nodes, and to determine the load threshold corresponding to each server node according to the historical load change rate of server nodes; and a scheduling module, used to obtain real-time performance indicators of server nodes, to determine the load score value according to the real-time performance indicators of server nodes, to filter out overloaded server nodes whose load score values ​​exceed the load threshold, and to perform load balancing scheduling on the tasks in the overloaded server nodes.

[0016] The beneficial effects of this invention are as follows:

[0017] By applying the above technical solutions, this invention can allocate corresponding and suitable server nodes for tasks with different performance requirements, accurately calculate the optimal allocation node for each task, monitor the performance indicators of each server node in real time, promptly detect overloaded server nodes and allocate overloaded tasks to the optimal suitable node in the server cluster network topology, provide a refined task scheduling strategy for the server cluster network, and effectively improve load balancing. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 The overall flowchart of a distributed task load balancing scheduling method proposed in an embodiment of the present invention is shown;

[0020] Figure 2 A schematic diagram of the structure of a distributed task load balancing scheduling system proposed in an embodiment of the present invention is shown. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] This application provides a distributed task load balancing scheduling method, such as... Figure 1 As shown, it includes:

[0023] S101, Obtain the performance requirements of the tasks to be executed, and allocate the tasks to be executed to the server nodes in the server cluster network topology according to the performance requirements of the tasks to be executed.

[0024] In some embodiments of this application, the step of allocating tasks to be executed to server nodes in the server cluster network topology based on the performance requirement indicators of the tasks to be executed includes: obtaining the performance requirement indicators of the tasks to be executed, wherein the performance requirement indicators include CPU requirement indicators, IO requirement indicators, network requirement indicators, and memory requirement indicators; establishing a center point, and drawing rays evenly outward from the center point as indicator axes; normalizing the performance requirement indicators of the tasks to be executed, and filling the normalized performance requirement indicators into the indicator axes; connecting the normalized performance requirement indicators on the indicator axes in sequence to obtain a performance requirement indicator radar chart of the tasks to be executed, and allocating tasks to be executed to server nodes based on the performance requirement indicator radar chart of the tasks to be executed.

[0025] In this embodiment, by detecting the CPU requirement, I / O requirement, network requirement, and memory requirement of the task to be executed, the above four indicators are integrated into the performance requirement indicator radar chart, which comprehensively reflects the multi-dimensional requirements of the task to be executed, thereby enabling the allocation of corresponding suitable server nodes through the performance requirement indicator radar chart.

[0026] In some embodiments of this application, the step of allocating tasks to be executed to server nodes based on the performance requirement index radar chart of the tasks to be executed includes: obtaining the node performance index of the current server node, wherein the node performance index includes CPU resource index, IO resource index, network resource index, and memory resource index; establishing a node resource index radar chart based on the node performance index, calculating the matching degree between the node resource index radar chart of each server node and the performance requirement index radar chart of the tasks to be executed; and allocating tasks to be executed to server nodes based on the matching degree between the node resource index radar chart of each server node and the performance requirement index radar chart of the tasks to be executed.

[0027] In this embodiment, the node performance indicators of each server node are detected sequentially, and the node performance indicators are integrated into the node resource indicator radar chart. The matching degree between the node resource indicator radar chart and the performance requirement indicator radar chart of each server node is calculated, and the server is assigned tasks to be executed based on the matching degree.

[0028] In some embodiments of this application, the calculation of the matching degree between the node resource indicator radar map of each server node and the performance requirement indicator radar map of the task to be executed includes: aligning the center points of the node resource indicator radar map and the performance requirement indicator radar map of the task to be executed, and aligning the vertices of the two radar maps to obtain an overlap map of the node resource indicator radar map and the performance requirement indicator radar map; calculating the absolute distance difference between corresponding vertices of the node resource indicator radar map and the performance requirement indicator radar map in the overlap map, and determining the average absolute distance difference based on the absolute distance differences of all vertices in the overlap map; calculating the cosine value of the included angle between corresponding sides of the node resource indicator radar map and the performance requirement indicator radar map in the overlap map, and determining the average cosine value based on all the included angle cosine values ​​in the overlap map; and multiplying the average absolute distance difference by the average cosine value to obtain the matching degree between the node resource indicator radar map and the performance requirement indicator radar map of the task to be executed.

[0029] In this embodiment, the center points of the two radar charts are overlapped, and the vertices of each indicator are aligned according to CPU demand indicator - CPU resource indicator, IO demand indicator - IO resource indicator, network demand indicator - network resource indicator, and memory demand indicator - memory resource indicator to obtain an overlapped chart. The edges connecting two adjacent vertices of the node resource indicator radar chart and the edges connecting the corresponding vertices of the performance demand indicator radar chart are extracted from the overlapped chart. The cosine of the acute angle between the two edges is calculated to obtain the cosine value of the included angle. The average cosine value of all included angle cosine values ​​in the overlapped chart is calculated. The average cosine value is used to more accurately calculate the resource deviation between the node resources and the requirements of the task to be executed. At the same time, the matching degree between the node resource indicator radar chart and the performance demand indicator radar chart of the task to be executed is calculated by combining the average absolute distance difference.

[0030] S102, obtain the historical load change rate of the server node, and determine the load threshold corresponding to each server node based on the historical load change rate of the server node.

[0031] In some embodiments of this application, determining the load threshold corresponding to each server node based on the historical load change rate of the server nodes includes: clustering each server node according to the historical load change rate of the server nodes to obtain a preset load threshold for the server nodes; determining the cluster center of the cluster corresponding to the server node based on the clustering results; and correcting the preset load threshold based on the cluster center of the cluster corresponding to the server node to obtain the corrected load threshold.

[0032] The step of clustering server nodes based on their historical load change rates includes: establishing a sample dataset based on the historical load change rates of all server nodes; randomly selecting k initial cluster centers from the sample dataset; calculating the Manhattan distance from the sample data in the sample dataset to the initial cluster centers; dividing each server node into its corresponding cluster based on the Manhattan distance from the sample data in the sample dataset to the initial cluster centers; calculating the average value of the sample data within each cluster; recalculating the cluster centers based on the average value of the sample data within each cluster; and repeating the above steps iteratively until the cluster centers no longer change or the number of iterations reaches the preset maximum number of iterations, thereby obtaining the clustering results of the server nodes.

[0033] In this embodiment, each server node is manually set with a preset load threshold. Based on the k-means clustering algorithm, each server node is clustered according to its historical load change rate. The cluster center of each server node is extracted, and each cluster center is normalized, with its value limited to [0.5-1.5]. The normalized cluster center is set as a correction coefficient, and the correction coefficient is multiplied by the preset load threshold of the server node to obtain the corrected load threshold.

[0034] S103: Obtain the real-time performance metrics of the server nodes, determine the load score based on the real-time performance metrics of the server nodes, filter out overloaded server nodes whose load scores exceed the load threshold, and perform load balancing scheduling on the tasks in the overloaded server nodes.

[0035] In some embodiments of this application, determining the load score based on the real-time performance metrics of server nodes includes: obtaining the real-time performance metrics of each server node in the historical server cluster network topology, wherein the real-time performance metrics include CPU operation metrics, IO usage metrics, network operation metrics, and memory usage metrics; establishing a training sample set based on the real-time performance metrics of each server node in the historical server cluster network topology; performing expert manual scoring on the real-time performance metrics in the training sample set to obtain manually labeled load score codes; establishing a deep neural network model to perform load score coding on the real-time performance metrics in the training sample set, and calculating the loss value between the load score code output by the deep neural network and the manually labeled load capacity score code; iteratively training the deep neural network parameters by minimizing the loss value to obtain a deep neural network model that can encode the real-time performance metrics of server nodes into corresponding load score values; and inputting the real-time performance metrics of the current server node into the trained deep neural network model to obtain the corresponding load score value.

[0036] In this embodiment, historical operation records of the cloud platform are collected to obtain real-time performance indicators of each server node in the historical server cluster network topology. The real-time performance indicators are vectorized to obtain a training sample set. A deep neural network model is established to perform load scoring encoding on the vectorized real-time performance indicators. The loss value between the load scoring encoding output by the network and the load scoring encoding manually annotated by actual computer experts is calculated. The parameters of the deep neural network are iteratively trained by minimizing the loss value, thereby obtaining a trained deep neural network model and outputting the load score value corresponding to the current server node.

[0037] In some embodiments of this application, the load balancing scheduling of tasks in overloaded server nodes includes: detecting surrounding nodes of the overloaded server node in the server cluster network topology, calculating the replacement matching degree between the surrounding nodes and the task to be unloaded; when the replacement matching degree between the surrounding nodes and the task to be unloaded is greater than a first preset threshold, transferring the task to be unloaded to the corresponding surrounding node.

[0038] In this embodiment, after detecting an overloaded server node, the task with the highest load occupancy is extracted as the task to be unloaded.

[0039] In some embodiments of this application, calculating the replacement matching degree between surrounding nodes and the task to be uninstalled includes: calculating the replacement matching degree between surrounding nodes and the task to be uninstalled according to a replacement matching degree calculation formula, wherein the replacement matching degree calculation formula is as follows:

[0040] ,

[0041] in, To replace the matching degree, Let be the node load value of the i-th surrounding node. The load threshold is the adjusted value for the i-th surrounding nodes. Let be the network distance value between the i-th surrounding node and the overloaded server node. Let represent the matching degree between the radar chart of node resource indicators of the i-th surrounding nodes and the radar chart of performance requirement indicators of the task to be unloaded. To preset the standard matching degree, This is a range adjustment factor. It is a natural exponential function.

[0042] In this embodiment, the nodes within a first preset range of the server node are first considered as surrounding nodes, and then... Obtain the node load utilization of the surrounding nodes corresponding to the overloaded server node, through... The distance influence coefficient between the overloaded server node and its surrounding nodes is obtained; the smaller the network distance, the higher this value. Calculate the matching degree between surrounding nodes and the task to be unloaded. Combine the above parameters to comprehensively determine whether the surrounding nodes can accept the scheduling of overloaded server nodes. If there are no surrounding nodes that meet the conditions, expand the scope of surrounding nodes one by one until a surrounding node that meets the conditions appears, so as to achieve load balancing of the server cluster network.

[0043] To achieve the above objectives, the present invention also provides a distributed task load balancing scheduling system, comprising: a task module, used to obtain performance requirement indicators of tasks to be executed, and to allocate tasks to be executed to server nodes in the server cluster network topology according to the performance requirement indicators of tasks to be executed; a threshold module, used to obtain the historical load change rate of server nodes, and to determine the load threshold corresponding to each server node according to the historical load change rate of server nodes; and a scheduling module, used to obtain real-time performance indicators of server nodes, to determine the load score value according to the real-time performance indicators of server nodes, to filter out overloaded server nodes whose load score values ​​exceed the load threshold, and to perform load balancing scheduling on the tasks in the overloaded server nodes.

[0044] By applying the above technical solutions, this invention obtains the performance requirements of the tasks to be executed, and allocates the tasks to the server nodes in the server cluster network topology based on these requirements. It also obtains the historical load change rate of the server nodes and determines the load threshold for each server node based on this rate. Furthermore, it obtains the real-time performance indicators of the server nodes, determines the load score based on these indicators, and filters out overloaded server nodes whose load scores exceed the load threshold. Finally, it performs load balancing scheduling on the tasks within these overloaded server nodes. This invention can accurately calculate the optimal allocation node for each task, providing a refined task scheduling strategy for the server cluster network and effectively improving load balancing.

[0045] Through the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented in hardware or by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A distributed task load balancing scheduling method, characterized in that, The method includes: Obtain the performance requirements of the tasks to be executed, and allocate the tasks to be executed to the server nodes in the server cluster network topology based on the performance requirements of the tasks to be executed. Obtain the historical load change rate of the server nodes, and determine the load threshold corresponding to each server node based on the historical load change rate of the server nodes. Obtain real-time performance metrics of server nodes, determine load scores based on these metrics, filter out overloaded server nodes whose load scores exceed the load threshold, and perform load balancing scheduling on the tasks in the overloaded server nodes. The process of allocating tasks to be executed to server nodes in the server cluster network topology based on the performance requirements of the tasks to be executed includes: Obtain the performance requirements of the task to be executed, including CPU requirements, I / O requirements, network requirements, and memory requirements. Establish a center point, and draw rays evenly outward from the center point as index axes. The performance requirements for the task to be performed are normalized, and the normalized performance requirements are then filled into the indicator axis. By sequentially connecting the normalized performance requirement indicators on the indicator axis, a performance requirement indicator radar chart of the task to be executed is obtained. The task to be executed is then assigned to the server node based on the performance requirement indicator radar chart of the task to be executed.

2. The distributed task load balancing scheduling method according to claim 1, characterized in that, The process of allocating tasks to server nodes based on the performance requirement radar chart of the tasks to be executed includes: Obtain the node performance metrics of the current server node, including CPU resource metrics, IO resource metrics, network resource metrics, and memory resource metrics. Establish a node resource indicator radar chart based on node performance indicators, and calculate the matching degree between the node resource indicator radar chart of each server node and the performance requirement indicator radar chart of the task to be executed. Tasks to be executed are assigned to server nodes based on the matching degree between the node resource index radar chart of each server node and the performance requirement index radar chart of the task to be executed.

3. The distributed task load balancing scheduling method according to claim 2, characterized in that, The calculation of the matching degree between the node resource index radar chart of each server node and the performance requirement index radar chart of the task to be executed includes: By aligning the center point of the node resource indicator radar chart with the performance requirement indicator radar chart of the task to be executed, and aligning the vertices of the two radar charts, an overlap map of the node resource indicator radar chart and the performance requirement indicator radar chart is obtained. Calculate the absolute distance difference between the corresponding vertices of the node resource index radar chart and the performance requirement index radar chart in the overlapping graph, and determine the average absolute distance difference based on the absolute distance difference of all vertices in the overlapping graph. Calculate the cosine of the angle between the corresponding sides of the node resource index radar chart and the performance requirement index radar chart in the overlapping graph, and determine the average cosine value based on all the angle cosine values ​​in the overlapping graph. Multiplying the average absolute distance difference by the average cosine value yields the matching degree between the node resource index radar chart and the performance requirement index radar chart of the task to be executed.

4. The distributed task load balancing scheduling method according to claim 3, characterized in that, The process of determining the load threshold for each server node based on its historical load change rate includes: Cluster each server node according to its historical load change rate, obtain the preset load threshold of the server node, and determine the cluster center of the corresponding cluster of the server node based on the clustering results. The preset load threshold is corrected based on the cluster center of the cluster corresponding to the server node to obtain the corrected load threshold.

5. The distributed task load balancing scheduling method according to claim 4, characterized in that, The clustering of server nodes based on their historical load change rates includes: A sample dataset is established based on the historical load change rate of all server nodes, and k initial cluster centers are randomly selected from the sample dataset. Calculate the Manhattan distance from the sample data in the sample dataset to the initial cluster center, and divide each server node into the corresponding cluster based on the Manhattan distance from the sample data in the sample dataset to the initial cluster center; Calculate the average value of the sample data within each cluster, and recalculate the cluster centers based on the average value of the sample data within each cluster; Repeat the above steps until the cluster centers no longer change or the number of iterations reaches the preset maximum number of iterations, and obtain the clustering results of the server nodes.

6. The distributed task load balancing scheduling method according to claim 1, characterized in that, The process of determining the load score based on the real-time performance metrics of the server nodes includes: Obtain real-time performance metrics for each server node in the historical server cluster network topology. The real-time performance metrics include CPU operation metrics, IO usage metrics, network operation metrics, and memory usage metrics. Establish a training sample set based on the real-time performance metrics of each server node in the historical server cluster network topology. The real-time performance metrics in the training sample set are manually scored by experts to obtain manually annotated load score codes. A deep neural network model is established to encode the load score of real-time performance indicators in the training sample set, and the loss value between the load score code output by the deep neural network and the manually labeled load capacity score code is calculated. By iteratively training the parameters of a deep neural network by minimizing the loss value, a deep neural network model is obtained that can encode the real-time performance indicators of server nodes into corresponding load scores. Input the real-time performance metrics of the current server node into the trained deep neural network model to obtain the corresponding load score.

7. The distributed task load balancing scheduling method according to claim 4, characterized in that, The load balancing scheduling of tasks in overloaded server nodes includes: Detect the surrounding nodes of overloaded server nodes in the server cluster network topology and calculate the replacement matching degree between the surrounding nodes and the tasks to be unloaded. When the replacement matching degree between the surrounding nodes and the task to be uninstalled is greater than the first preset threshold, the task to be uninstalled is transferred to the corresponding surrounding node.

8. The distributed task load balancing scheduling method according to claim 7, characterized in that, The calculation of the replacement matching degree between surrounding nodes and the task to be unloaded includes: The replacement matching degree between surrounding nodes and the task to be uninstalled is calculated according to the replacement matching degree calculation formula, wherein the replacement matching degree calculation formula is as follows: ， in, To replace the matching degree, Let be the node load value of the i-th surrounding node. The corrected load threshold for the i-th surrounding nodes. Let be the network distance value between the i-th surrounding node and the overloaded server node. Let represent the matching degree between the radar chart of node resource indicators of the i-th surrounding nodes and the radar chart of performance requirement indicators of the task to be unloaded. To preset the standard matching degree, This is a range adjustment factor.

9. A distributed task load balancing scheduling system, characterized in that, include: The task module is used to obtain the performance requirements of the tasks to be executed and to allocate the tasks to be executed to the server nodes in the server cluster network topology based on the performance requirements of the tasks to be executed. The threshold module is used to obtain the historical load change rate of server nodes and determine the load threshold corresponding to each server node based on the historical load change rate of server nodes. The scheduling module is used to obtain real-time performance indicators of server nodes, determine the load score based on the real-time performance indicators of server nodes, filter out overloaded server nodes whose load score exceeds the load threshold, and perform load balancing scheduling on the tasks in the overloaded server nodes. The task module assigns tasks to be executed to server nodes in the server cluster network topology based on the performance requirements of the tasks to be executed, including: Obtain the performance requirements of the task to be executed, including CPU requirements, I / O requirements, network requirements, and memory requirements. Establish a center point, and draw rays evenly outward from the center point as index axes. The performance requirements for the task to be performed are normalized, and the normalized performance requirements are then filled into the indicator axis. By sequentially connecting the normalized performance requirement indicators on the indicator axis, a performance requirement indicator radar chart of the task to be executed is obtained. The task to be executed is then assigned to the server node based on the performance requirement indicator radar chart of the task to be executed.

Images