A method and system for scheduling computing resources in a computing power network

By building target computing power network and collaborative control module in computing power network, evaluating task execution computing power and selecting appropriate execution mode, the problem of low resource scheduling efficiency in the existing technology is solved, and efficient and flexible resource utilization is achieved. Task execution.

CN118626276BActive Publication Date: 2024-11-05NANJING BOYU IOT TECH CO LTD +1
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411088077.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-09
Publication Date
2024-11-05
Estimated Expiration
2044-08-09

AI Technical Summary

Technical Problem

The existing computing power network is inefficient and flexible in computing power resource scheduling, and cannot flexibly adjust resource allocation according to the actual needs of computing tasks, resulting in waste of resources and low task execution efficiency.

Method used

Build a target computing power network, including multiple computing power agent centers and collaborative control modules, flexibly select a single-center independent execution mode or a multi-center collaborative execution mode, dynamically schedule resources to ensure resource utilization efficiency and task execution efficiency.

Benefits of technology

It realizes flexible resource scheduling according to the requirements of computing tasks, improves computing resource utilization efficiency and task execution efficiency, avoids resource waste, and adapts to computing tasks of different scales and types.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118626276B_ABST
    Figure CN118626276B_ABST
Patent Text Reader

Abstract

The present invention discloses a computing power resource scheduling method and system for a computing power network, which belongs to the field of computing power resource scheduling, wherein the method includes: building a target computing power network, receiving a target computing task, evaluating the required task execution computing power, and comparing it with the single-center computing power; if it is less than or equal to the single-center computing power, selecting an independent computing power proxy center, and processing the target computing task through a single-center independent execution mode; if it is greater than the single-center computing power, selecting an idle computing power proxy center according to the number of computing power proxy centers, forming a temporary collaborative computing unit, and processing the target computing task through a multi-center collaborative execution mode. The present application solves the technical problems of low efficiency and poor flexibility in computing power resource scheduling in a computing power network in the prior art, and achieves the technical effect of flexibly selecting a single-center independent execution mode or a multi-center collaborative execution mode according to the computing power requirements of the computing task, thereby improving the utilization efficiency of computing power resources and the efficiency of task execution.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computing power resource scheduling, and in particular to a computing power resource scheduling method and system for a computing power network. Background Art

[0002] With the development of technologies such as big data and artificial intelligence, the demand for computing resources is growing. As a distributed computing architecture, the computing network connects multiple computing nodes to provide powerful computing power. However, the computing network in the existing technology has some shortcomings in the scheduling of computing resources. The existing computing network adopts a fixed resource allocation strategy and cannot be flexibly adjusted according to the actual needs of computing tasks, resulting in low efficiency in the utilization of computing resources. For example, for small-scale computing tasks, too many computing resources will be allocated, resulting in waste; while for large-scale computing tasks, the allocated computing resources may be insufficient, affecting the efficiency of task execution. Summary of the invention

[0003] The present application provides a computing power resource scheduling method and system for a computing power network, aiming to solve the technical problems of low efficiency and poor flexibility in computing power resource scheduling in a computing power network in the prior art.

[0004] In view of the above problems, the present application provides a computing resource scheduling method and system for a computing network.

[0005] The first aspect disclosed in the present application provides a computing resource scheduling method for a computing network, the method comprising: building a target computing network, the target computing network comprising multiple computing proxy centers and a collaborative control module, wherein the multiple computing proxy centers have the same single-center computing power, each computing proxy center is composed of multiple computing execution nodes, and the multiple computing execution nodes have the same single-node computing power, and the collaborative control module is used to coordinate and manage task allocation and result aggregation between multiple computing proxy centers; receiving a target computing task sent by a task submitter, evaluating the task execution computing power required for the target computing task, and comparing the task execution computing power with the single-center computing power; if the task execution computing power is smaller than the target computing task, the task execution computing power is calculated, and the task execution computing power is compared with the single-center computing power; if the task execution computing power is smaller than the target computing task, the task execution computing power is calculated, and the task execution computing power is compared with the single-center computing power; if the task execution computing power is smaller than the target computing task, the task execution computing power is calculated, and the task execution computing power is calculated ... If the task execution computing power is greater than or equal to the single-center computing power, an idle computing power proxy center is selected from the idle computing power proxy center set as an independent computing power proxy center; the target computing task is assigned to the independent computing power proxy center, and the target computing task is processed through the single-center independent execution mode; if the task execution computing power is greater than the single-center computing power, the number of computing power proxy centers is determined according to the task execution computing power, and an idle computing power proxy center is selected from the idle computing power proxy center set according to the number of computing power proxy centers to form a temporary collaborative computing unit; the target computing task is submitted to the collaborative control module, and the collaborative control module assigns the target computing task to the temporary collaborative computing unit, and the target computing task is processed through the multi-center collaborative execution mode.

[0006] Another aspect disclosed in the present application provides a computing resource scheduling system for a computing network, the system comprising: a network construction unit for constructing a target computing network, the target computing network comprising multiple computing proxy centers and a collaborative control module, wherein the multiple computing proxy centers have the same single-center computing power, each computing proxy center is composed of multiple computing execution nodes, and the multiple computing execution nodes have the same single-node computing power, and the collaborative control module is used to coordinate and manage task allocation and result aggregation between multiple computing proxy centers; a task evaluation unit for receiving a target computing task sent by a task submitter, evaluating the task execution computing power required for the target computing task, and comparing the task execution computing power with the single-center computing power; an independent center selection unit for selecting a task execution center if the task execution computing power is small. is greater than or equal to the single-center computing power, an idle computing power proxy center is selected from the idle computing power proxy center set as an independent computing power proxy center; a single-center execution unit is used to allocate the target computing task to the independent computing power proxy center, and process the target computing task through the single-center independent execution mode; a collaborative unit selection unit is used to determine the number of computing power proxy centers according to the task execution computing power if the task execution computing power is greater than the single-center computing power, and select an idle computing power proxy center from the idle computing power proxy center set according to the number of computing power proxy centers to form a temporary collaborative computing unit; a collaborative execution unit is used to submit the target computing task to the collaborative control module, and the collaborative control module allocates the target computing task to the temporary collaborative computing unit, and processes the target computing task through the multi-center collaborative execution mode.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] Since the target computing power network is built, the target computing power network includes multiple computing power proxy centers and a collaborative control module. Among them, multiple computing power proxy centers have the same single-center computing power, each computing power proxy center is composed of multiple computing power execution nodes, and multiple computing power execution nodes have the same single-node computing power. The collaborative control module is used to coordinate and manage the task allocation and result aggregation among multiple computing power proxy centers. By building a target computing power network, it provides an infrastructure for computing power resource scheduling and realizes flexible and efficient task allocation and processing; receives the target computing task sent by the task submitter, evaluates the task execution computing power required for the target computing task, compares the task execution computing power with the single-center computing power, determines the computing power requirement of the target computing task, and provides a basis for the subsequent selection of an appropriate execution mode; if the task execution computing power is less than or equal to the single-center computing power, selects an idle computing power proxy center from the idle computing power proxy center as an independent computing power proxy center, and selects a single computing power proxy center to independently process tasks with smaller computing power requirements to avoid resource waste; allocates the target computing task to the independent computing power proxy center, processes the target computing task through the single-center independent execution mode, and uses the computing power of a single computing power proxy center to efficiently complete tasks with smaller computing power requirements; if the task execution computing power is less than or equal to the single-center computing power, selects an idle computing power proxy center from the idle computing power proxy center as an independent computing power proxy center, and selects a single computing power proxy center to independently process tasks with smaller computing power requirements to avoid resource waste; allocates the target computing task to the independent computing power proxy center, processes the target computing task through the single-center independent execution mode, and uses the computing power of a single computing power proxy center to efficiently complete tasks with smaller computing power requirements; When the computing power of a row is greater than the computing power of a single center, the number of computing power proxy centers is determined according to the computing power of task execution, and an idle computing power proxy center is selected from the idle computing power proxy center set according to the number of computing power proxy centers to form a temporary collaborative computing unit. For tasks with large computing power requirements, multiple computing power proxy centers are used to collaboratively process them to improve the efficiency of task execution; the target computing task is submitted to the collaborative control module, and the collaborative control module assigns the target computing task to the temporary collaborative computing unit, and the target computing task is processed through the multi-center collaborative execution mode. The scheduling and management functions of the collaborative control module are used to realize the collaborative computing of multiple computing power proxy centers, and the technical solution for efficiently completing tasks with large computing power requirements solves the technical problems of low efficiency and poor flexibility in scheduling computing power resources in the computing power network in the prior art, and achieves the technical effect of flexibly selecting the single-center independent execution mode or the multi-center collaborative execution mode according to the computing power requirements of the computing task, thereby improving the utilization efficiency of computing power resources and the efficiency of task execution.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A flowchart of a method for scheduling computing resources of a computing network is provided for an embodiment of the present application;

[0011] Figure 2A structural schematic diagram of a computing resource scheduling system of a computing network is provided for an embodiment of the present application.

[0012] Explanation of the accompanying drawings: network building unit 11, task evaluation unit 12, independent center selection unit 13, single center execution unit 14, collaborative unit selection unit 15, collaborative execution unit 16. DETAILED DESCRIPTION

[0013] The overall idea of ​​the technical solution provided by this application is as follows:

[0014] The embodiments of the present application provide a computing resource scheduling method and system for a computing network. By building a target computing network including multiple computing proxy centers and collaborative control modules, a single-center independent execution mode or a multi-center collaborative execution mode can be flexibly selected according to the computing power requirements of the computing tasks, thereby achieving efficient utilization of computing resources and improving task execution efficiency.

[0015] First, build a target computing power network, which includes multiple computing power proxy centers with the same single-center computing power and a collaborative control module for coordinating and managing task allocation and result aggregation. When receiving the target computing task sent by the task submitter, evaluate the execution computing power required for the task and compare it with the single-center computing power. If the task execution computing power is less than or equal to the single-center computing power, select one from the idle computing power proxy center set as an independent computing power proxy center, assign the task to the center and process it through the single-center independent execution mode. If the task execution computing power is greater than the single-center computing power, determine the number of computing power proxy centers required based on the execution computing power, select the corresponding number of centers from the idle computing power proxy center set to form a temporary collaborative computing unit, submit the task to the collaborative control module, and assign it to the temporary collaborative computing unit for processing through the multi-center collaborative execution mode.

[0016] In summary, this application fully considers the differences in computing power requirements of computing tasks, and realizes the reasonable scheduling and efficient use of computing power resources by flexibly selecting the mode of independent execution by a single center or collaborative execution by multiple centers. Compared with the existing technology, it can significantly improve the computing power resource utilization efficiency and task execution efficiency of the computing power network, and at the same time has good flexibility and scalability to adapt to the computing task requirements of different scales and types.

[0017] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically described below in conjunction with the drawings in the specification.

[0018] Embodiment 1, as Figure 1 As shown, an embodiment of the present application provides a computing resource scheduling method for a computing network, the method comprising:

[0019] S1: Build a target computing power network, which includes multiple computing power proxy centers and a collaborative control module, wherein the multiple computing power proxy centers have the same single-center computing power, each computing power proxy center is composed of multiple computing power execution nodes, and the multiple computing power execution nodes have the same single-node computing power. The collaborative control module is used to coordinate and manage task allocation and result aggregation between multiple computing power proxy centers.

[0020] Specifically, in order to achieve efficient scheduling of computing resources and rapid execution of tasks, the first step is to build a target computing network, which includes multiple computing agent centers and a collaborative control module.

[0021] Multiple computing power proxy centers are the middle-layer management units of the computing power network, responsible for managing and scheduling the computing power execution node resources within them, and are used to receive and process various computing tasks. Each computing power proxy center has equal single-center computing power, that is, the computing processing capacity of a single center, which ensures that the processing capacity between computing power proxy centers is balanced and consistent, which is conducive to the reasonable allocation of tasks. In actual deployment, the size of the single-center computing power is flexibly set according to specific application scenarios and performance requirements, such as high-performance computing clusters, server clusters, etc. Multiple computing power execution nodes are deployed inside each computing power proxy center. The computing power execution node is the basic execution unit and is responsible for the actual computing task processing work. All computing power execution nodes have the same single-node computing power, representing the computing processing capacity of a single node. Through the consistency of internal node computing power, tasks can be easily split into several subtasks with similar computing amounts and assigned to each node for parallel processing, thereby effectively improving the efficiency of task execution. At the same time, load balancing between nodes is also easier to achieve.

[0022] In addition to multiple computing power proxy centers, there is also a collaborative control module in the target computing power network. This module is the core control unit of the entire computing power network, responsible for task coordination and scheduling management between computing power proxy centers. Specifically, the collaborative control module receives the target computing tasks from the task submitter, and flexibly distributes them to multiple computing power proxy centers for processing based on the size of the task's computational workload, degree of parallelism and other characteristics. At the same time, the collaborative control module is responsible for tracking and summarizing the task execution progress and results of each computing power proxy center, and returning the complete calculation results to the task submitter. Through centralized management and collaborative scheduling, the collaborative control module can give full play to the distributed computing advantages of the computing power proxy center and significantly improve the flexibility and efficiency of task processing.

[0023] S2: Receive the target computing task sent by the task submitter, evaluate the task execution computing power required for the target computing task, and compare the task execution computing power with the single-center computing power.

[0024] Specifically, when a computing task needs to be executed, the task submitter sends the target computing task to the computing network. The target computing task is received through the task receiving interface of the computing network and enters the task processing flow.

[0025] After receiving the target computing task, the computing network first evaluates its computing power requirements, and estimates the task execution computing power required for the task by comprehensively considering various characteristic parameters of the target computing task, such as computing complexity, data size, task type, priority, etc. Among them, task execution computing power refers to the amount of computing resources required to complete the target computing task, which is measured by indicators such as the number of CPU cores, memory size, and the number of GPUs.

[0026] After obtaining the estimated value of the computing power required to execute the target computing task, the computing power network compares it with the computing power of a single computing power proxy center. The computing power of a single center refers to the sum of the computing power of all computing power execution nodes within a computing power proxy center, representing the overall computing processing power of the center. By comparing the size of the task execution computing power and the computing power of a single center, it is determined whether the task can be completed within a single computing power proxy center or requires collaborative processing by multiple centers, thereby providing a decision-making basis for subsequent task allocation.

[0027] By receiving the target computing tasks and evaluating the computing power required to execute them, and comparing them with the computing power of a single center, the foundation is laid for the computing power network to rationally allocate computing resources and efficiently execute tasks. This enables the computing power network to dynamically evaluate the computing power requirements based on the characteristics of the tasks and has the intelligent scheduling capabilities, which helps to give full play to the advantages of networked collaborative computing and improve the overall task execution efficiency.

[0028] S3: If the task execution computing power is less than or equal to the single center computing power, select an idle computing power proxy center from the idle computing power proxy center set as an independent computing power proxy center.

[0029] Specifically, after completing the computing power requirement assessment of the target computing task, the computing power network will select appropriate computing power resources to execute the task based on the task execution computing power.

[0030] First, the computing power network determines whether the task execution computing power of the target computing task is less than or equal to the single-center computing power of a single computing power proxy center. If the execution computing power of the target computing task meets this condition, it means that the target computing task can be completed independently in a single computing power proxy center without the need for cross-center collaboration. At this time, the computing power network selects an idle computing power proxy center from the current set of idle computing power proxy centers and designates it as the computing power proxy center that independently processes the target computing task. Among them, the idle computing power proxy center set refers to the set of all computing power proxy centers that are currently not assigned tasks and are in a waiting state for scheduling. By maintaining this set in real time, the computing power network can dynamically grasp the operating status and load conditions of each computing power proxy center, thereby rationally allocating idle computing power resources and improving efficiency.

[0031] After selecting an independent computing power proxy center, the computing power network assigns the target computing task to the center and instructs it to independently complete the processing of the task through the internal computing power execution node. This single-center independent execution mode is suitable for small-scale, low-computing power tasks, avoiding unnecessary cross-center communication overhead and reducing task execution latency.

[0032] By flexibly selecting the appropriate computing power proxy center to undertake the task based on the task execution computing power of the target computing task, when the task execution computing power is less than or equal to the computing power of a single center, a center is selected from the set of idle computing power proxy centers to independently execute the task. Through independent execution by a single center, the execution delay and communication cost of the task are effectively reduced, and the utilization rate of computing power resources is improved.

[0033] S4: Allocate the target computing task to the independent computing power proxy center, and process the target computing task through a single-center independent execution mode.

[0034] Specifically, once a computing power proxy center that independently processes the target computing task is selected, the computing power network will perform the task allocation and calculation process.

[0035] First, the computing power network packages and sends the complete task description information, input data, and required computing environment configuration of the target computing task to the selected independent computing power proxy center. After receiving the task information package, the independent computing power proxy center further decomposes and analyzes the target computing task. Through task decomposition, the independent computing power proxy center divides the target computing task into multiple subtasks that can be executed in parallel. The granularity and number of these subtasks are determined based on the parallelism, computing amount, data dependency and other characteristics of the task to achieve load balancing and maximize efficiency of task execution.

[0036] After completing the task decomposition, the independent computing power proxy center will assign subtasks to multiple computing power execution nodes within it. The independent computing power proxy center will take over the subtasks through idle computing power execution nodes and execute them in parallel. After processing the assigned subtasks, each computing power execution node returns the calculation results of the subtasks to the independent computing power proxy center. The independent computing power proxy center is responsible for monitoring and managing the task execution progress of each computing power execution node and summarizing the calculation results of the subtasks. When all subtasks are completed, the independent computing power proxy center returns the execution results of the complete target computing task to the task submitter, releases the computing power resources occupied during the task execution, and re-enters the idle waiting state.

[0037] S5: If the task execution computing power is greater than the single center computing power, determine the number of computing power proxy centers according to the task execution computing power, and select an idle computing power proxy center from the idle computing power proxy center set according to the number of computing power proxy centers to form a temporary collaborative computing unit.

[0038] Specifically, when it is assessed that the computing power required for the target computing task is greater than the single-center computing power of a single computing power proxy center, the single-center independent execution mode can no longer meet the computing requirements of the task, and a cross-center collaborative computing method is required to process the task.

[0039] First, the computing power network determines the number of computing power proxy centers that need to be coordinated based on the computing power requirements of the target computing task. The determination of this number follows the following principles: the sum of the computing power of the selected computing power proxy centers should be greater than or equal to the task execution computing power, and the minimum number of computing power proxy centers should be selected as much as possible to reduce the cross-center communication overhead. After determining the required number of computing power proxy centers, the computing power network selects the corresponding number of idle computing power proxy centers from the current set of idle computing power proxy centers to build a temporary collaborative computing unit. This temporary collaborative computing unit is specifically used to process the current target computing task. Its life cycle is synchronized with the task execution process and it is automatically disbanded after the task is completed.

[0040] The computing power proxy centers in the temporary collaborative computing unit are physically distributed in different geographical locations. They are interconnected through high-speed networks to form a unified computing resource pool logically, and accept the unified scheduling and management of the collaborative control module. The collaborative control module is responsible for the reasonable splitting and allocation of the target computing tasks within the temporary collaborative computing unit, coordinating the parallel computing and result aggregation of each computing power proxy center, and finally completing the execution of the entire task.

[0041] By dynamically selecting the required computing power proxy center according to the computing power requirements of task execution and organizing it into a temporary collaborative computing unit with unified scheduling, the distributed computing power resources can be flexibly aggregated to efficiently execute large-scale computing tasks, giving full play to the horizontal expansion and collaborative computing capabilities of the computing power network, and providing strong support for handling complex computing problems. At the same time, the dynamic construction and release of temporary collaborative computing units also help to improve the resource utilization of the entire network and avoid waste of computing power.

[0042] S6: Submit the target computing task to the collaborative control module, and the collaborative control module allocates the target computing task to the temporary collaborative computing unit, and processes the target computing task through a multi-center collaborative execution mode.

[0043] Specifically, after a temporary collaborative computing unit is built to execute the target computing task, the computing power network initiates a cross-center collaborative computing process to achieve efficient processing of large-scale tasks.

[0044] First, the collaborative control module of the computing power network takes over all computing power proxy centers within the temporary collaborative computing unit to form a unified resource scheduling view. Then, the collaborative control module distributes the complete task description information, input data, and required computing environment configuration of the target computing task to each computing power proxy center within the temporary collaborative computing unit. After receiving the task information, each computing power proxy center within the temporary collaborative computing unit participates in the execution of the target computing task under the unified coordination of the collaborative control module. Specifically, the collaborative control module is responsible for further splitting the target computing task, dividing the task into multiple tasks that can be executed in parallel according to factors such as the parallelism of the task, the number and performance of the computing power proxy centers, and assigning them to different computing power proxy centers for processing.

[0045] Within each computing power proxy center, the execution process of tasks is similar to the independent execution mode of a single center. The computing power proxy center dynamically schedules its internal computing power execution nodes according to the computing power requirements of the task to realize parallel computing of the task. When the task is completed, the computing power proxy center returns the calculation results of the task to the collaborative control module. The collaborative control module tracks and records the task execution progress of each computing power proxy center in real time, and dynamically adjusts the task allocation strategy to balance the computing load of each center. When all computing power proxy centers in the temporary collaborative computing unit have completed their respective task executions, the collaborative control module will collect and summarize the calculation results of all tasks to form the execution results of the target computing task. After that, the collaborative control module returns the execution results of the target computing task to the task submitter and issues an instruction to release all computing power resources occupied by the temporary collaborative computing unit. The temporary collaborative computing unit is then disbanded, and the computing power proxy centers participating in the calculation re-enter the idle state, waiting for further scheduling of the computing power network.

[0046] Through the multi-center collaborative execution mode, the computing power network can give full play to the advantages of distributed collaborative computing under the scheduling of the collaborative control module, significantly improve the execution efficiency of large-scale complex tasks, break through the single-center computing power bottleneck, and improve the utilization efficiency of computing power resources and task execution efficiency.

[0047] Furthermore, the embodiment of the present application also includes:

[0048] In the independent computing power proxy center, splitting the target computing task into multiple subtasks;

[0049] Evaluate the subtask execution computing power required for each subtask, and compare the subtask execution computing power with the single node computing power;

[0050] For a subtask whose execution computing power is less than or equal to the computing power of the single node, an idle computing power execution node is selected from the computing power execution nodes of the independent computing power proxy center, and the subtask is assigned to the idle computing power execution node for subtask processing;

[0051] For subtasks whose execution computing power is greater than the computing power of the single node, the number of computing power execution nodes is determined according to the subtask execution computing power, and a corresponding number of idle computing power execution nodes are selected from the computing power execution nodes of the independent computing power proxy center to form a temporary cascade computing unit, and the subtasks are assigned to the temporary cascade computing units for subtask processing.

[0052] In a feasible implementation, when the independent computing power proxy center receives the target computing task, it first further splits the task into multiple subtasks with smaller granularity and can be executed in parallel. Among them, the division of subtasks comprehensively considers the characteristics of the parallelism, computing amount distribution, data dependency, etc. of the target computing task, aiming to achieve load balancing and efficiency maximization in the task processing process. Then, the computing power requirements of each subtask are evaluated, and the estimated execution computing power requirements of each subtask, i.e., the subtask execution computing power, are obtained by comprehensively considering factors such as the computational complexity and data scale of the subtask. Then, the independent computing power proxy center compares the subtask execution computing power of each subtask with the subtask execution computing power of a single computing power execution node. The single-node computing power is the computing processing capacity of a single computing power execution node within the independent computing power proxy center, and is the basic unit of task scheduling. By comparing the size of the subtask execution computing power and the single-node computing power, the processing scale of each subtask is judged, providing a decision basis for subsequent task allocation.

[0053] For subtasks whose execution power is less than or equal to the single-node power, the independent computing power proxy center will select an idle computing power execution node from its internal idle computing power execution node set and assign the subtask to the idle computing power execution node for independent execution. Idle computing power execution nodes refer to computing nodes that are not currently assigned tasks and are in a waiting state for scheduling. Due to the small computing scale of the subtask, a single computing power execution node can independently complete the processing of the subtask without the need to collaborate with other nodes. For subtasks whose execution power is greater than the computing power of a single node, a single computing power execution node can no longer independently complete the task processing, and multiple computing power execution nodes are required for collaborative computing. At this time, the independent computing power proxy center will determine the number of computing power execution nodes that need to participate in the collaborative computing based on the idle computing power execution nodes of the subtask. Subsequently, the independent computing power proxy center selects the required number of idle computing power execution nodes from its internal idle computing power execution node set and organizes them into temporary cascade computing units. The temporary cascade computing unit is a combination of computing resources dynamically constructed within the independent computing power proxy center for large-grained subtasks. Through the close collaboration of multiple computing power execution nodes, it can effectively handle the computing scale that is difficult for a single node to handle. After building the temporary cascade computing unit, the independent computing power proxy center assigns the large-grained subtask to the temporary cascade computing unit for execution. The multiple computing power execution nodes within the temporary cascade computing unit collaborate to complete the processing of subtasks through parallel computing, exchange of intermediate results, and other methods, and return the results to the independent computing power proxy center.

[0054] By performing fine-grained task splitting and computing power evaluation within the independent computing power proxy center, a single computing power execution node is flexibly scheduled or a temporary cascade computing unit is dynamically constructed according to the computing scale of the subtask, and the computing resources within the independent computing power proxy center are fully utilized. While ensuring the efficiency of task execution, load balancing and computing power utilization are maximized. This adaptive task processing mode enables the independent computing power proxy center to efficiently and flexibly respond to computing tasks of different scales, and improve the task processing capabilities under the single-center independent execution mode.

[0055] Furthermore, the embodiment of the present application also includes:

[0056] According to the parallelism of the target computing task, the target computing task is divided into a plurality of parallel computing branches;

[0057] Allocating each parallel computing branch to different computing power proxy centers within the temporary collaborative computing unit;

[0058] In each computing power proxy center, the parallel computing branch is split into multiple subtasks;

[0059] Evaluate the subtask execution computing power required for each subtask, and compare the subtask execution computing power with the single node computing power;

[0060] For subtasks whose subtask execution computing power is less than or equal to the computing power of the single node, an idle computing power execution node is selected from the computing power execution nodes of the current computing power proxy center, and the subtask is assigned to the idle computing power execution node for subtask processing;

[0061] For subtasks whose execution computing power is greater than the computing power of the single node, the number of computing power execution nodes is determined based on the subtask execution computing power, and a corresponding number of idle computing power execution nodes are selected from the computing power execution nodes of the current computing power proxy center to form a temporary cascade computing unit, and the subtasks are assigned to the temporary cascade computing units for subtask processing.

[0062] In a feasible implementation, when the temporary collaborative computing unit receives the target computing task, the collaborative control module first performs a parallel analysis on the target computing task, and divides it into multiple computing branches that can be executed in parallel according to the inherent parallelism of the task. Parallel computing branches refer to task subsets that can be executed simultaneously on different computing power proxy centers and have no data dependencies on each other. The granularity of the division of parallel computing branches depends on the parallelism characteristics of the target computing task and the number of computing power proxy centers in the temporary collaborative computing unit. After dividing the parallel computing branches, the collaborative control module allocates each parallel computing branch to different computing power proxy centers in the temporary collaborative computing unit. Branch allocation follows the load balancing principle, comprehensively considering factors such as the computing power of each computing power proxy center and the network transmission bandwidth, and tries to make each computing power proxy center obtain parallel computing branches of equal difficulty to avoid idle computing resources or task waiting.

[0063] When a parallel computing branch is assigned to a computing power proxy center, the computing power proxy center will further split the branch task into multiple subtasks with smaller granularity. The subtask splitting process is similar to the subtask division of a single center. Considering the computing characteristics of the parallel computing branch and the computing resource distribution within the computing power proxy center, the fine-grained division of the branch task is achieved through the task splitting algorithm to prepare for the subsequent subtask scheduling. The computing power proxy center evaluates the computing power requirements of each subtask, compares the subtask execution computing power of each subtask with the computing power of a single computing power execution node, and determines the processing method of each subtask based on the comparison results. For subtasks whose subtask execution computing power is less than or equal to the computing power of a single node, the computing power proxy center will select an idle computing power execution node from its internal idle computing power execution node set and assign the subtask to the node for independent execution. For subtasks whose subtask execution computing power is greater than the computing power of a single node, the computing power proxy center will select the required number of idle computing power execution nodes from its internal idle computing power execution node set based on the execution computing power of the subtask, and build a temporary cascade computing unit to process the subtask.

[0064] Through the two-level task processing mode of cross-center collaboration and internal parallelism, the distributed parallel computing advantages of the computing network are maximized, which can efficiently execute large-scale complex computing tasks and significantly improve the overall performance of task processing. At the same time, the adaptive matching of task splitting granularity and computing resource scheduling method also ensures the balanced utilization and load balancing of computing resources, avoiding the waste of computing resources.

[0065] Furthermore, the embodiment of the present application also includes:

[0066] For target computing tasks processed in a single-center independent execution mode, within the independent computing power proxy center, after any computing power execution node completes subtask processing, it immediately returns the subtask processing result to the independent computing power proxy center and sets the computing power execution node as an idle computing power execution node;

[0067] After all computing power execution nodes complete the subtask processing, the independent computing power proxy center summarizes the processing results of each subtask to obtain the processing result of the target computing task;

[0068] Returning the processing result of the target computing task from the independent computing power proxy center to the task submitter;

[0069] After returning the processing result of the target computing task to the task submitter, the computing resources occupied by the target computing task in the independent computing power proxy center are released, and the independent computing power proxy center is set as an idle computing power proxy center.

[0070] In a feasible implementation, in an independent computing power proxy center, when any computing power execution node completes the processing of the assigned subtask, it immediately returns the subtask processing result to the independent computing power proxy center. The real-time result return mechanism provides information about completed subtasks to the independent computing power proxy center as early as possible to facilitate subsequent result aggregation and progress tracking. At the same time, the computing power execution node that completes the subtask processing will be set as an idle computing power execution node to release the computing resources it occupies so that the independent computing power proxy center can promptly allocate it to other subtasks to be processed, thereby improving the utilization of computing resources. The idle computing power execution node will re-enter the resource pool of the independent computing power proxy center and wait for the next task scheduling.

[0071] The independent computing power proxy center tracks the task execution progress of all computing power execution nodes in real time. When all computing power execution nodes have completed their respective subtask processing, the independent computing power proxy center summarizes and assembles all subtask processing results, and performs necessary post-processing to generate the complete processing result of the target computing task, that is, the target computing task processing result. After generating the target computing task processing result, the independent computing power proxy center returns it to the task submitter through the computing power network. Among them, the task submitter can be an end user, an upper-level application, or other external entity that entrusts the execution of the task.

[0072] After the processing result of the target computing task is returned to the task submitter, the independent computing power proxy center will start the process of releasing the task-related resources, releasing the storage space, memory, I / O and other resources occupied by the independent computing power proxy center during the task execution, as well as all computing power execution nodes assigned to the task. Through resource release, the independent computing power proxy center reclaims the computing resources occupied by the task as soon as possible and improves resource utilization efficiency. At the same time, the independent computing power proxy center will switch its own state to idle state and set it to idle computing power proxy center, indicating that there is no task currently being executed and internal resources can be used to receive new tasks.

[0073] In the single-center independent execution mode, the real-time return mechanism of subtask results and the timely release strategy of task-related resources ensure the efficient use and rapid recovery of computing resources within the independent computing power proxy center, avoiding unnecessary resource occupation and idle waste. At the same time, the orderly summary and reliable return of subtask processing results ensure the smooth completion of the target computing tasks and the accurate delivery of results. The automatic reset of the status of the independent computing power proxy center enables it to continuously and efficiently receive and execute new computing tasks, meeting the computing power network's needs for high-concurrency task processing.

[0074] Furthermore, the embodiment of the present application also includes:

[0075] For the target computing task processed in the multi-center collaborative execution mode, in the temporary collaborative computing unit, after any computing power proxy center completes the parallel computing branch it is responsible for, it immediately returns the branch processing result to the collaborative control module and sets the computing power proxy center as an idle computing power proxy center;

[0076] After all computing power proxy centers complete the parallel computing branch processing, the collaborative control module summarizes the processing results of each branch to obtain the processing results of the target computing task;

[0077] Returning the target computing task processing result from the collaborative control module to the task submitter;

[0078] After returning the processing result to the task submitter, the computing resources occupied by the target computing task in the temporary collaborative computing unit are released, and the temporary collaborative computing unit is dissolved.

[0079] In a feasible implementation, in a temporary collaborative computing unit, when any computing power proxy center completes the assigned parallel computing branch, it immediately returns the branch processing result to the collaborative control module, and provides the collaborative control module with information about the completed branch as soon as possible, so that the collaborative control module can timely understand the progress of task execution and prepare for result aggregation in advance. At the same time, the computing power proxy center that has completed the parallel computing branch processing will be set to an idle state, releasing the computing resources it occupies in the temporary collaborative computing unit and reset to an idle computing power proxy center. These idle computing power proxy centers will not be disbanded immediately, but will return to the resource pool of the computing power network, waiting for further scheduling by the collaborative control module in order to participate in the collaborative computing of other tasks.

[0080] The collaborative control module tracks the execution progress of branches of all computing power proxy centers in the temporary collaborative computing unit in real time. When all computing power proxy centers have completed the parallel computing branches they are responsible for, the collaborative control module summarizes and assembles the processing results of all parallel computing branches to generate the processing results of the target computing task. After generating the processing results of the target computing task, the collaborative control module returns them to the task submitter through the computing power network.

[0081] After the processing result of the target computing task is successfully returned to the task submitter, the collaborative control module starts the process of releasing the task-related resources, including releasing the storage space, memory, I / O and other resources occupied by the temporary collaborative computing unit during the task execution, as well as all computing power proxy centers involved in the task calculation. Through resource release, the computing power network can reclaim the computing resources occupied by the task as soon as possible and improve resource utilization efficiency. At the same time, the collaborative control module disbands the temporary collaborative computing unit specially constructed for the target computing task, releasing the logical management resources and network communication resources occupied by it.

[0082] In the multi-center collaborative execution mode, the real-time return mechanism of parallel computing branch results and the timely release strategy of task-related resources ensure the efficient use and rapid recovery of computing resources within the temporary collaborative computing unit, avoiding unnecessary resource occupation and idle waste. At the same time, the orderly aggregation and reliable return of parallel computing branch results ensure the smooth completion of computing tasks and accurate delivery of results. The dynamic dissolution of temporary collaborative computing units enables the computing network to flexibly respond to computing tasks of different scales and types, build and release collaborative computing resources on demand, and improve the flexibility and adaptability of the entire network.

[0083] Furthermore, the embodiment of the present application also includes:

[0084] Constructing a task computing power evaluation model, wherein the task computing power evaluation model is trained based on historical task execution data and task characteristic parameters, wherein the task characteristic parameters include the computational complexity, data scale, task type, and priority of the task;

[0085] The task computing power evaluation model is configured at the task receiving port of the computing power network and the task receiving port of each computing power proxy center to evaluate the computing power requirements of the received tasks in real time;

[0086] At the task receiving port of the computing power network, the task execution computing power required for the received target computing task is evaluated by the task computing power evaluation model;

[0087] At the task receiving port of each computing power proxy center, the task computing power evaluation model is used to evaluate the subtask execution computing power required for the split subtasks.

[0088] In a preferred embodiment, first, a task computing power evaluation model for evaluating the computing power requirements of computing tasks is constructed. The model adopts a machine learning method and is trained and learned based on a large amount of historical task execution data and task characteristic parameters, so as to master the computing power requirements and laws of computing tasks. Among them, the historical task execution data includes the input parameters, execution process records, resource consumption statistics and other information of various types of computing tasks that have been executed in the computing power network, reflecting the demand for computing power resources of different types of tasks in the actual execution process; the task characteristic parameters are the key factors affecting the computing power requirements of computing tasks, including the computational complexity, data scale, task type, priority, etc. of the task. Among them, the computational complexity reflects the computational amount and time complexity of the task, the data scale indicates the size and dimension of the task input data, the task type refers to the computing mode and application field of the task, and the priority affects the scheduling order and resource allocation strategy of the task. By comprehensively considering the historical task execution data and task characteristic parameters, the task computing power evaluation model establishes the corresponding relationship between computing tasks and computing power requirements, providing a reference basis for subsequent real-time computing power evaluation.

[0089] The constructed task computing power evaluation model needs to be configured and deployed at the key nodes of the computing power network in order to receive and evaluate newly submitted computing tasks in real time. Specifically, the task computing power evaluation model will be configured at the task receiving port of the computing power network and the task receiving port of each computing power proxy center. Among them, the task receiving port of the computing power network is the unified entrance for the entire network to receive external computing tasks, and is responsible for receiving the target computing tasks submitted by the task submitter, while the task receiving port of each computing power proxy center is the local entrance for receiving subtasks after the task is decomposed. By configuring the task computing power evaluation model at key nodes, the computing power network can evaluate its computing power resource requirements in real time at different granularities of the task, providing an important basis for subsequent task scheduling decisions.

[0090] When the task receiving port of the computing power network receives a new target computing task, it immediately calls the deployed task computing power evaluation model to evaluate the overall computing power requirements of the task. During the evaluation process, the task computing power evaluation model extracts the key characteristic parameters of the target computing task, such as computing complexity, data scale, task type, priority, etc., and inputs these parameters into the model for calculation and analysis to obtain the task execution computing power required for the task, and the total amount of computing power resources required for the entire target computing task, which is the decision-making basis for the computing power network to schedule tasks. After each computing power proxy center decomposes the received task into multiple subtasks, it calls the deployed task computing power evaluation model to further evaluate the computing power requirements of the subtasks, extracts the key characteristic parameters of the subtasks, and inputs them into the task computing power evaluation model for calculation and analysis to obtain the subtask execution computing power required for the subtask, which reflects the amount of computing power resources required to complete the subtask, and is the basis for task scheduling and resource allocation within the computing power proxy center.

[0091] By building a task computing power evaluation model and configuring it in the computing power network and computing power proxy center, we can achieve real-time and accurate evaluation of computing power requirements for computing tasks, make full use of historical task execution data and task characteristic parameters, adaptively grasp the characteristics and laws of computing power requirements for computing tasks, and provide a basis for task scheduling decisions in the computing power network. At the same time, after the task is decomposed, the subtasks are evaluated twice, which further improves the granularity and accuracy of computing power demand evaluation, making the scheduling and allocation of computing power resources more efficient and reasonable.

[0092] Furthermore, the embodiment of the present application also includes:

[0093] In the target computing power network, deploy an independent collaborative control server;

[0094] The collaborative control module is installed and configured on the collaborative control server, and the collaborative control module includes a task distribution program, a task splitting program, a task scheduling program, a result aggregation program and a result returning program.

[0095] In a preferred embodiment, when constructing the target computing power network, an independent collaborative control server is set in the target computing power network to carry and run the collaborative control module. The collaborative control server is physically independent of other computing power proxy centers and is interconnected with each computing power proxy center through a high-speed network. Through independent deployment, the collaborative control module is isolated from the execution of specific computing tasks, which improves the stability and security of the collaborative control process, and also facilitates independent performance optimization and expansion of the collaborative control server.

[0096] Install and configure the collaborative control module on the collaborative control server. The collaborative control module is the core software component for realizing the collaborative work of the computing power network. It consists of a series of dedicated program modules, including task distribution program, task splitting program, task scheduling program, result aggregation program and result return program. Among them, the task distribution program is responsible for receiving the submitted computing tasks and distributing the tasks to each computing power proxy center for execution; the task splitting program is used to split complex large computing tasks into multiple smaller tasks to facilitate parallel computing in each computing power proxy center in the computing power network; the task scheduling program is responsible for real-time scheduling and coordination of all computing resources in the computing power network to execute computing tasks in the best way; the result aggregation program is used to collect and aggregate the task execution results distributed in different computing power proxy centers and different computing power execution nodes, and perform necessary post-processing and format conversion on the results to generate the final target computing task processing results; the result return program is responsible for returning the target computing task processing results of the task to the task submitter to complete the entire task execution process.

[0097] By introducing an independent collaborative control server into the computing network and installing and configuring a collaborative control module on it, the collaborative control capability and intelligent scheduling level of the computing network are enhanced. The independent deployment of the collaborative control server ensures the stability, security and scalability of the collaborative control process. The various dedicated programs in the collaborative control module comprehensively optimize the collaborative workflow of the computing network from multiple aspects such as task distribution, splitting, scheduling, and result aggregation, so that computing resources can collaboratively perform computing tasks in a more efficient, flexible, and controllable manner.

[0098] In summary, the computing resource scheduling method of a computing network provided by the embodiment of the present application has the following technical effects:

[0099] Build a target computing power network, which includes multiple computing power proxy centers and a collaborative control module, wherein multiple computing power proxy centers have the same single-center computing power, each computing power proxy center is composed of multiple computing power execution nodes, and multiple computing power execution nodes have the same single-node computing power. The collaborative control module is used to coordinate and manage the task allocation and result aggregation between multiple computing power proxy centers, build an infrastructure suitable for computing power resource scheduling, and provide the necessary environment for subsequent task allocation and processing. Receive the target computing task sent by the task submitter, evaluate the task execution computing power required for the target computing task, compare the task execution computing power with the single-center computing power, and provide a judgment basis for the subsequent selection of a suitable execution mode. If the task execution computing power is less than or equal to the single-center computing power, select an idle computing power proxy center from the idle computing power proxy center as an independent computing power proxy center, assign the target computing task to the independent computing power proxy center, and process the target computing task through the single-center independent execution mode, avoiding unnecessary resource waste and completing task processing quickly and efficiently. If the task execution computing power is greater than the single-center computing power, the number of computing power proxy centers is determined according to the task execution computing power, and an idle computing power proxy center is selected from the idle computing power proxy center set according to the number of computing power proxy centers to form a temporary collaborative computing unit, and the target computing task is submitted to the collaborative control module, which assigns the target computing task to the temporary collaborative computing unit, and processes the target computing task through the multi-center collaborative execution mode. Through the collaborative calculation of multiple computing power proxy centers, tasks with large computational volume can be completed efficiently, thereby improving task execution efficiency and computing power resource utilization. This method flexibly selects different execution modes, fully utilizes the computing power of the computing power network, efficiently completes various computing tasks, and significantly improves computing power resource utilization efficiency and task execution efficiency.

[0100] Embodiment 2 is based on the same inventive concept as the computing resource scheduling method of a computing network in the above embodiment. Figure 2 As shown, an embodiment of the present application provides a computing resource scheduling system for a computing network, the system comprising:

[0101] A network construction unit 11 is used to construct a target computing power network, wherein the target computing power network includes multiple computing power proxy centers and a collaborative control module, wherein the multiple computing power proxy centers have the same single-center computing power, each computing power proxy center is composed of multiple computing power execution nodes, and the multiple computing power execution nodes have the same single-node computing power, and the collaborative control module is used to coordinate and manage task allocation and result aggregation between multiple computing power proxy centers;

[0102] The task evaluation unit 12 is used to receive the target computing task sent by the task submitter, evaluate the task execution computing power required for the target computing task, and compare the task execution computing power with the single center computing power;

[0103] The independent center selection unit 13 is used to select an idle computing power proxy center from the idle computing power proxy center set as the independent computing power proxy center if the task execution computing power is less than or equal to the single center computing power;

[0104] A single-center execution unit 14 is used to distribute the target computing task to the independent computing power proxy center and process the target computing task through a single-center independent execution mode;

[0105] A collaborative unit selection unit 15 is used to determine the number of computing power proxy centers according to the task execution computing power if the task execution computing power is greater than the single center computing power, and select an idle computing power proxy center from the idle computing power proxy center set according to the number of computing power proxy centers to form a temporary collaborative computing unit;

[0106] The collaborative execution unit 16 is used to submit the target computing task to the collaborative control module, and the collaborative control module allocates the target computing task to the temporary collaborative computing unit to process the target computing task through a multi-center collaborative execution mode.

[0107] Furthermore, the single-center execution unit 14 includes the following execution steps:

[0108] In the independent computing power proxy center, splitting the target computing task into multiple subtasks;

[0109] Evaluate the subtask execution computing power required for each subtask, and compare the subtask execution computing power with the single node computing power;

[0110] For a subtask whose execution computing power is less than or equal to the computing power of the single node, an idle computing power execution node is selected from the computing power execution nodes of the independent computing power proxy center, and the subtask is assigned to the idle computing power execution node for subtask processing;

[0111] For subtasks whose execution computing power is greater than the computing power of the single node, the number of computing power execution nodes is determined according to the subtask execution computing power, and a corresponding number of idle computing power execution nodes are selected from the computing power execution nodes of the independent computing power proxy center to form a temporary cascade computing unit, and the subtasks are assigned to the temporary cascade computing units for subtask processing.

[0112] Furthermore, the collaborative execution unit 16 includes the following execution steps:

[0113] According to the parallelism of the target computing task, the target computing task is divided into a plurality of parallel computing branches;

[0114] Allocating each parallel computing branch to different computing power proxy centers within the temporary collaborative computing unit;

[0115] In each computing power proxy center, the parallel computing branch is split into multiple subtasks;

[0116] Evaluate the subtask execution computing power required for each subtask, and compare the subtask execution computing power with the single node computing power;

[0117] For subtasks whose subtask execution computing power is less than or equal to the computing power of the single node, an idle computing power execution node is selected from the computing power execution nodes of the current computing power proxy center, and the subtask is assigned to the idle computing power execution node for subtask processing;

[0118] For subtasks whose execution computing power is greater than the computing power of the single node, the number of computing power execution nodes is determined based on the subtask execution computing power, and a corresponding number of idle computing power execution nodes are selected from the computing power execution nodes of the current computing power proxy center to form a temporary cascade computing unit, and the subtasks are assigned to the temporary cascade computing units for subtask processing.

[0119] Furthermore, the single-center execution unit 14 further includes the following execution steps:

[0120] For target computing tasks processed in a single-center independent execution mode, within the independent computing power proxy center, after any computing power execution node completes subtask processing, it immediately returns the subtask processing result to the independent computing power proxy center and sets the computing power execution node as an idle computing power execution node;

[0121] After all computing power execution nodes complete the subtask processing, the independent computing power proxy center summarizes the processing results of each subtask to obtain the processing result of the target computing task;

[0122] Returning the processing result of the target computing task from the independent computing power proxy center to the task submitter;

[0123] After returning the processing result of the target computing task to the task submitter, the computing resources occupied by the target computing task in the independent computing power proxy center are released, and the independent computing power proxy center is set as an idle computing power proxy center.

[0124] Furthermore, the collaborative execution unit 16 further includes the following execution steps:

[0125] For the target computing task processed in the multi-center collaborative execution mode, in the temporary collaborative computing unit, after any computing power proxy center completes the parallel computing branch it is responsible for, it immediately returns the branch processing result to the collaborative control module and sets the computing power proxy center as an idle computing power proxy center;

[0126] After all computing power proxy centers complete the parallel computing branch processing, the collaborative control module summarizes the processing results of each branch to obtain the processing results of the target computing task;

[0127] Returning the target computing task processing result from the collaborative control module to the task submitter;

[0128] After returning the processing result to the task submitter, the computing resources occupied by the target computing task in the temporary collaborative computing unit are released, and the temporary collaborative computing unit is dissolved.

[0129] Furthermore, the embodiment of the present application further includes a computing power evaluation unit, which performs the following steps:

[0130] Constructing a task computing power evaluation model, wherein the task computing power evaluation model is trained based on historical task execution data and task characteristic parameters, wherein the task characteristic parameters include the computational complexity, data scale, task type, and priority of the task;

[0131] The task computing power evaluation model is configured at the task receiving port of the computing power network and the task receiving port of each computing power proxy center to evaluate the computing power requirements of the received tasks in real time;

[0132] At the task receiving port of the computing power network, the task execution computing power required for the received target computing task is evaluated by the task computing power evaluation model;

[0133] At the task receiving port of each computing power proxy center, the task computing power evaluation model is used to evaluate the subtask execution computing power required for the split subtasks.

[0134] Furthermore, the embodiment of the present application further includes a collaborative control module deployment unit, which includes the following execution steps:

[0135] In the target computing power network, deploy an independent collaborative control server;

[0136] The collaborative control module is installed and configured on the collaborative control server, and the collaborative control module includes a task distribution program, a task splitting program, a task scheduling program, a result aggregation program and a result returning program.

[0137] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0138] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A computing resource scheduling method for a computing network, characterized in that: The method comprises: Build a target computing power network, the target computing power network includes multiple computing power proxy centers and a collaborative control module, wherein the multiple computing power proxy centers have the same single-center computing power, each computing power proxy center is composed of multiple computing power execution nodes, the multiple computing power execution nodes have the same single-node computing power, and the collaborative control module is used to coordinate and manage task allocation and result aggregation between multiple computing power proxy centers; Receive a target computing task sent by a task submitter, evaluate the task execution computing power required for the target computing task, and compare the task execution computing power with the single-center computing power; If the task execution computing power is less than or equal to the single center computing power, an idle computing power proxy center is selected from the idle computing power proxy center set as an independent computing power proxy center; Allocating the target computing task to the independent computing power proxy center, and processing the target computing task through a single-center independent execution mode; If the task execution computing power is greater than the single center computing power, the number of computing power proxy centers is determined according to the task execution computing power, and an idle computing power proxy center is selected from the idle computing power proxy center set according to the number of computing power proxy centers to form a temporary collaborative computing unit; Submitting the target computing task to the collaborative control module, which allocates the target computing task to the temporary collaborative computing unit, and processes the target computing task through a multi-center collaborative execution mode; Allocating the target computing task to the independent computing power proxy center, and processing the target computing task through a single-center independent execution mode, including: In the independent computing power proxy center, splitting the target computing task into multiple subtasks; Evaluate the subtask execution computing power required for each subtask, and compare the subtask execution computing power with the single node computing power; For a subtask whose execution computing power is less than or equal to the computing power of the single node, an idle computing power execution node is selected from the computing power execution nodes of the independent computing power proxy center, and the subtask is assigned to the idle computing power execution node for subtask processing; For a subtask whose execution computing power is greater than the computing power of the single node, the number of computing power execution nodes is determined according to the subtask execution computing power, and a corresponding number of idle computing power execution nodes are selected from the computing power execution nodes of the independent computing power proxy center to form a temporary cascade computing unit, and the subtask is assigned to the temporary cascade computing unit for subtask processing; Allocating the target computing task to the temporary collaborative computing unit, and processing the target computing task through a multi-center collaborative execution mode, including: According to the parallelism of the target computing task, the target computing task is divided into a plurality of parallel computing branches; Allocating each parallel computing branch to different computing power proxy centers within the temporary collaborative computing unit; In each computing power proxy center, the parallel computing branch is split into multiple subtasks; Evaluate the subtask execution computing power required for each subtask, and compare the subtask execution computing power with the single node computing power; For subtasks whose subtask execution computing power is less than or equal to the computing power of the single node, an idle computing power execution node is selected from the computing power execution nodes of the current computing power proxy center, and the subtask is assigned to the idle computing power execution node for subtask processing; For subtasks whose execution computing power is greater than the computing power of the single node, the number of computing power execution nodes is determined based on the subtask execution computing power, and a corresponding number of idle computing power execution nodes are selected from the computing power execution nodes of the current computing power proxy center to form a temporary cascade computing unit, and the subtasks are assigned to the temporary cascade computing units for subtask processing.

2. A computing resource scheduling method for a computing network according to claim 1, characterized in that: The target computing task is processed by a single-center independent execution mode, and the method further includes: For target computing tasks processed in a single-center independent execution mode, within the independent computing power proxy center, after any computing power execution node completes subtask processing, it immediately returns the subtask processing result to the independent computing power proxy center and sets the computing power execution node as an idle computing power execution node; After all computing power execution nodes complete the subtask processing, the independent computing power proxy center summarizes the processing results of each subtask to obtain the processing result of the target computing task; Returning the processing result of the target computing task from the independent computing power proxy center to the task submitter; After returning the processing result of the target computing task to the task submitter, the computing resources occupied by the target computing task in the independent computing power proxy center are released, and the independent computing power proxy center is set as an idle computing power proxy center.

3. The method for scheduling computing resources of a computing network according to claim 1, characterized in that: The target computing task is processed by a multi-center collaborative execution mode, and the method further includes: For the target computing task processed in the multi-center collaborative execution mode, in the temporary collaborative computing unit, after any computing power proxy center completes the parallel computing branch it is responsible for, it immediately returns the branch processing result to the collaborative control module and sets the computing power proxy center as an idle computing power proxy center; After all computing power proxy centers complete the parallel computing branch processing, the collaborative control module summarizes the processing results of each branch to obtain the processing results of the target computing task; Returning the target computing task processing result from the collaborative control module to the task submitter; After returning the processing result to the task submitter, the computing resources occupied by the target computing task in the temporary collaborative computing unit are released, and the temporary collaborative computing unit is dissolved.

4. The method for scheduling computing resources of a computing network according to claim 1, characterized in that: The method further comprises: Constructing a task computing power evaluation model, wherein the task computing power evaluation model is trained based on historical task execution data and task characteristic parameters, wherein the task characteristic parameters include the computational complexity, data scale, task type, and priority of the task; The task computing power evaluation model is configured at the task receiving port of the computing power network and the task receiving port of each computing power proxy center to evaluate the computing power requirements of the received tasks in real time; At the task receiving port of the computing power network, the task execution computing power required for the received target computing task is evaluated by the task computing power evaluation model; At the task receiving port of each computing power proxy center, the task computing power evaluation model is used to evaluate the subtask execution computing power required for the split subtasks.

5. The method for scheduling computing resources of a computing network according to claim 1, characterized in that: The method further comprises: In the target computing power network, deploy an independent collaborative control server; The collaborative control module is installed and configured on the collaborative control server, and the collaborative control module includes a task distribution program, a task splitting program, a task scheduling program, a result aggregation program and a result returning program.

6. A computing resource scheduling system for a computing network, characterized in that: A computing resource scheduling method for a computing network according to any one of claims 1 to 5, the system comprising: A network construction unit, the network construction unit is used to build a target computing power network, the target computing power network includes multiple computing power proxy centers and a collaborative control module, wherein the multiple computing power proxy centers have the same single-center computing power, each computing power proxy center is composed of multiple computing power execution nodes, the multiple computing power execution nodes have the same single-node computing power, and the collaborative control module is used to coordinate and manage task allocation and result aggregation between multiple computing power proxy centers; A task evaluation unit, the task evaluation unit is used to receive the target computing task sent by the task submitter, evaluate the task execution computing power required for the target computing task, and compare the task execution computing power with the single center computing power; An independent center selection unit, the independent center selection unit is used to select an idle computing power proxy center from the idle computing power proxy center set as the independent computing power proxy center if the task execution computing power is less than or equal to the single center computing power; A single-center execution unit, the single-center execution unit is used to distribute the target computing task to the independent computing power proxy center, and process the target computing task through a single-center independent execution mode; A collaborative unit selection unit, wherein the collaborative unit selection unit is used to determine the number of computing power proxy centers according to the task execution computing power if the task execution computing power is greater than the single center computing power, and select an idle computing power proxy center from the idle computing power proxy center set according to the number of computing power proxy centers to form a temporary collaborative computing unit; A collaborative execution unit, wherein the collaborative execution unit is used to submit the target computing task to the collaborative control module, and the collaborative control module allocates the target computing task to the temporary collaborative computing unit, and processes the target computing task through a multi-center collaborative execution mode.

Citation Information

Cited By

  • Computing power scheduling strategy optimization system based on reinforcement learning

    CN119356824A

  • A computing power scheduling strategy optimization system based on reinforcement learning

    CN119356824B