Calculation power scheduling method and device and electronic equipment

Abstract

The embodiment of the invention provides a computing power scheduling method and device and electronic equipment, and relates to the technical field of cloud computing. The computing power scheduling method comprises the following steps: receiving a task request; performing feature extraction on the task; obtaining an optimal computing power scheduling strategy according to the task characteristics and a preset computing power scheduling strategy; and computing power scheduling is carried out according to the optimal computing power scheduling strategy. And computing power scheduling is carried out according to the optimal computing power scheduling strategy, so that the computing power scheduling adapts to task conditions, efficient and intelligent dynamic allocation of computing power resources is realized, and diversified requirements on the computing resources in different application scenes are met.

Description

Technical Field

[0001] The present application relates to the field of cloud computing technology, and specifically to a computing power scheduling method, device, and electronic device. Background Art

[0002] With the rapid development of computing technology, the amount of tasks that need to be performed is increasing, and the amount of computing is increasing. In addition, the computing power of each computing node is limited. Therefore, computing power scheduling is needed, especially in the fields of cloud computing, edge computing, cloud-edge collaborative computing, etc., computing power scheduling is particularly important.

[0003] However, existing computing power scheduling methods have numerous shortcomings. Traditional computing power scheduling strategies are relatively simple and difficult to adapt to complex and changing business scenarios. This leads to low resource utilization and inefficient task processing. This is particularly true in cloud-edge collaborative computing, where the advantages of cloud-edge collaborative computing cannot be fully utilized. For example, when faced with large-scale data processing and tasks with high real-time requirements, traditional scheduling methods cannot accurately schedule tasks based on factors such as task characteristics and network conditions, resulting in increased task execution delays and reduced system performance. Summary of the Invention

[0004] In view of this, the embodiments of the present application provide a computing power scheduling method, device and electronic device, which aim to achieve efficient, intelligent and dynamic allocation of computing power resources by selecting the optimal computing power scheduling strategy for different tasks, thereby meeting the diversified needs of computing resources in different application scenarios.

[0005] In a first aspect, an embodiment of the present application provides a computing power scheduling method, comprising the following steps: Receive task requests; Extract features for the task; Obtain the optimal computing power scheduling strategy based on task characteristics and pre-set computing power scheduling strategy; Perform computing power scheduling according to the optimal computing power scheduling strategy.

[0006] In some possible implementations, feature extraction for a task includes: Extract feature vectors for tasks; Tasks are weighted according to preset indicators.

[0007] In some possible implementations, the pre-set computing power scheduling strategy includes at least one of: a proximity access strategy, a latency priority strategy, a latency circle strategy, a fixed weight strategy, a computing power balancing strategy, a computing network collaboration strategy, and an electronic computing collaboration strategy; Among them, the nearest access strategy is based on the geographic information location of the task initiator, and the task is preferentially scheduled to the computing node closest to the geographic information location of the task initiator; The delay priority strategy dynamically adjusts task allocation based on the real-time network delay and takes the shortest task execution delay as the goal; The latency circle strategy uses preset scheduling rules to allocate tasks within each latency circle, ensuring that tasks are executed within the preset latency requirements. The latency circles are divided into different latency circles based on the preset network topology and latency characteristics. The fixed weight strategy assigns fixed weights to different computing nodes and allocates tasks to different computing nodes according to the weight ratio; The computing power balancing strategy is to monitor the computing power usage of each computing node in real time and dynamically allocate the tasks performed by each computing node to keep the computing power load of each computing node balanced; The computing network collaboration strategy allocates tasks to each computing node based on the synergy between computing resources and network resources, thus achieving optimal configuration of computing network resources. The computer collaboration strategy is to allocate computing power by combining the computing requirements and data processing requirements of the task, as well as computing resources and data resources.

[0008] In some possible implementations, based on task characteristics and pre-set computing power scheduling strategies, the optimal computing power scheduling strategies include: Get the current status of the system, Select one or more computing power scheduling strategies from the pre-set computing power scheduling strategies based on the task characteristics and the current system status.

[0009] In some possible implementations, selecting one or more computing power scheduling strategies from pre-set computing power scheduling strategies based on the task classification result and the current system state includes: Based on the task characteristics, determine whether there are tasks containing hard constraints. If so, first select the preset computing power scheduling strategy based on the hard constraints, and then select the corresponding computing power scheduling strategy based on the remaining task characteristics and the computing capabilities of the computing nodes; where the hard constraints are the preset computing power scheduling strategies that must be selected to meet the task characteristics; If it does not exist, the corresponding computing power scheduling strategy is selected based on the task characteristics and the computing capabilities of the computing nodes.

[0010] In some possible implementations, the method further includes: Obtain real-time feedback on the performance of computing power scheduling; Adjust task characteristics or pre-set computing power scheduling strategies based on real-time feedback information.

[0011] In some possible implementations, the method further includes: When a temporary failure occurs on a computing node executing a task, the task is migrated to a neighboring computing node for execution.

[0012] In a second aspect, an embodiment of the present application provides a computing power scheduling device, including: A receiving unit, configured to receive a task request; The scheduling unit is used to extract features from tasks; classify tasks based on the extracted features; obtain the optimal computing power scheduling strategy based on the task classification results and the pre-set computing power scheduling strategy; and schedule computing power based on the optimal computing power scheduling strategy.

[0013] In a third aspect, an embodiment of the present application provides an electronic device, which includes a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the electronic device is triggered to execute the method provided in the first aspect of the embodiment of the present application.

[0014] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the method provided in the first aspect of the embodiment of the present application.

[0015] By adopting the solution provided in the embodiment of the present application, features of tasks are extracted and classified, and then the optimal computing power scheduling strategy is obtained based on the task classification results and the pre-set computing power scheduling strategy; computing power scheduling is performed according to the optimal computing power scheduling strategy, so that the computing power scheduling is adapted to the task situation, thereby realizing efficient, intelligent and dynamic allocation of computing power resources, and meeting the diversified needs of computing resources in different application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0017] Figure 1 A flowchart of a computing power scheduling method provided in an embodiment of the present application; Figure 2 A schematic diagram of the technical architecture of each module provided in the embodiments of the present application; Figure 3 A schematic diagram of the coordination relationship between the modules provided in the embodiment of the present application; Figure 4 A schematic diagram of a computing power scheduling device provided in an embodiment of the present application; Figure 5 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to better understand the technical solution of the present application, the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0019] It should be clear that the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0020] The terms used in the embodiments of the present application are for the purpose of describing specific embodiments only and are not intended to limit the present application. The singular forms "a", "an", "the" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.

[0021] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. Furthermore, the character " / " in this document generally indicates an "or" relationship between the associated objects.

[0022] With the rapid development of computing technology, the amount of tasks that need to be performed is getting larger and larger, and the amount of computing is getting larger and larger. In addition, the computing power of each computing node is limited. Therefore, it is necessary to schedule multiple tasks to different computing nodes to achieve computing power scheduling.

[0023] At present, the computing power scheduling strategy is relatively simple and difficult to adapt to complex and changing business scenarios, resulting in low resource utilization and low task processing efficiency. Especially in cloud-edge collaborative computing, the advantages of cloud-edge collaborative computing cannot be fully utilized. For example, when faced with large-scale data processing and tasks with high real-time requirements, traditional scheduling methods cannot perform accurate scheduling based on task characteristics, network conditions and other factors, resulting in increased task execution delays and decreased system performance.

[0024] Specifically, the existing technology has the following key problems: The separation of latency and computing power: Traditional computing power scheduling methods rely solely on fixed rules or single objectives for computing power scheduling and task allocation, and are unable to dynamically adapt to complex network topologies and power peak and valley fluctuations, resulting in high-latency tasks being unable to meet real-time requirements.

[0025] Insufficient resource coordination: Computing power, network, and power resources are independently scheduled, lacking cross-domain coordination mechanisms, and energy consumption costs remain high.

[0026] Weak dynamic perception capability: Existing methods (such as the Dijkstra algorithm) have difficulty responding to changes in network link status (such as packet loss rate and bandwidth fluctuations) in real time, which affects scheduling stability.

[0027] Therefore, this application proposes a computing power scheduling method that uses multiple strategies to coordinate computing power. Based on task characteristics, system status, and user needs, different scheduling strategies are dynamically selected and combined to achieve flexible switching and complementary advantages among scheduling strategies, thereby improving scheduling accuracy and adaptability. This method can meet the efficient, intelligent, and dynamic allocation of cloud-edge computing resources and the diverse computing resource requirements in different application scenarios. It is applicable to scenarios such as edge computing, the Internet of Things, and autonomous driving that are sensitive to latency and require multi-objective optimization. A detailed description is provided below.

[0028] See also Figure 1 In a first aspect, an embodiment of the present application provides a computing power scheduling method, comprising the following steps: S101: Receive a task request.

[0029] That is, the system receives task requests from different sources and adds them to the task queue.

[0030] In some embodiments, after S101 , the method further includes: performing preliminary screening and classification on the received tasks, removing invalid or duplicate tasks, and placing valid tasks into a task queue to wait for assignment.

[0031] S102: Extract features of the task.

[0032] That is, tasks are taken out from the task queue in sequence, task feature information is extracted, and data analysis technology is used to conduct in-depth analysis of task features to provide accurate data support for strategy selection.

[0033] As a possible implementation method, the characteristics of the task include task type, data volume, computational complexity, latency requirements, etc.

[0034] In some embodiments, in S102, the features include at least one of task type, service level requirements, energy consumption constraints, and network dependencies. Task types (Type) include compute-intensive (such as AI reasoning), IO-intensive (such as video transcoding), and real-time tasks (such as autonomous driving decision-making). Service level requirements (SLA) include maximum tolerable latency (<50ms / 100ms / 500ms), computing power requirements (CPU / GPU core count), and data volume (MB / GB level). Energy consumption constraints (Energy) include the sensitivity of tasks to electricity (such as green computing tasks that need to be allocated to renewable energy nodes first). Network dependencies (Network) include bandwidth requirements (such as VR tasks require >100Mbps) and whether cross-regional collaboration is required (such as distributed training tasks).

[0035] In some embodiments, in S102, extracting features from the task includes: S111: Extract feature vectors for the task.

[0036] Among them, each type of task is encoded as a four-dimensional vector: , For example, when a task involves real-time video analysis, feature vector modeling yields the following: [IO intensive, latency < 100ms, energy sensitive, bandwidth > 50Mbps]. When a task involves offline log processing, feature vector extraction yields the following: [computation intensive, latency < 500ms, no energy constraints, bandwidth < 10Mbps].

[0037] S112: Perform weighted processing on the tasks according to preset indicators.

[0038] The preset indicators include at least one of a delay indicator function, an energy consumption indicator function, and a load balancing indicator function.

[0039] Specifically, the delay index function Expressed as: , Among them, i is the task number, N is the total number of tasks, is the task delay, The maximum tolerable latency of the task. SLA stands for Service Level Agreement Latency.

[0040] Energy consumption index function Expressed as: , in, Indicates resource utilization; Indicates the energy sensitivity flag. When a task is marked as energy sensitive, , otherwise 0.

[0041] Load balancing indicator function Expressed as: , Where m represents the total number of nodes, is the utilization rate of node j, Load_Std is the average value of all node utilizations. Load_Std represents the standard deviation calculation formula of node utilization, which measures the degree to which the actual utilization of nodes in the system deviates from its average utilization. The larger its value, the greater the difference in node utilization and the more unbalanced the load. The smaller its value, the smaller the difference in utilization between nodes and the more balanced the load. The reciprocal of the load balancing index is taken.

[0042] According to the preset indicators, weighted processing of the tasks includes: setting a common weighted indicator function and a resource utilization overload penalty indicator function for the tasks.

[0043] The ordinary weighted index function Expressed as: , Resource utilization overload penalty indicator function Expressed as: , in, represents the delay index function, represents the energy consumption index function, represents the load balancing index function, m represents the total number of nodes, is the utilization rate of node j, It is worth noting that the penalty coefficient is adjustable in the present invention. The default value is 1000.

[0044] In step S112 , priorities may be further set for the tasks according to the weighted processing results, and the tasks may also be further classified.

[0045] S103: Obtain an optimal computing power scheduling strategy based on task characteristics and pre-set computing power scheduling strategies.

[0046] That is, based on task feature analysis and system status information, the intelligent decision-making module is called to select the optimal scheduling strategy or strategy combination. The decision-making process comprehensively considers multiple factors, such as task priority, resource utilization, network status, etc.

[0047] In some possible implementations, in S103, the pre-set computing power scheduling strategy includes at least one of: proximity access strategy, latency priority strategy, latency circle strategy, fixed weight strategy, computing power balancing strategy, computing network collaboration strategy, and electronic computing collaboration strategy.

[0048] The proximity access strategy prioritizes scheduling tasks to the nearest compute node based on the geographic location of the task initiator. Scheduling computing power based on proximity access reduces data transmission distances, thereby lowering network latency and improving task response speed. This strategy is applicable to a variety of scenarios. For example, in security monitoring scenarios, when a video analysis task is generated at a monitoring point, the proximity access strategy prioritizes processing at the edge computing node closest to that monitoring point.

[0049] The latency-priority strategy dynamically adjusts task allocation based on real-time network latency, aiming to minimize task execution latency. This latency-priority strategy schedules computing power, enabling tasks to execute in a low-latency environment and improving user experience, such as in real-time voice interaction and online gaming.

[0050] The latency circle strategy uses preset scheduling rules to allocate tasks within each latency circle, ensuring that tasks are executed within the preset latency requirements. Latency circles are divided into different latency circles based on the preset network topology and latency characteristics. For example, in a city-level IoT application, multiple latency circles can be divided based on network coverage and latency in different areas, and the appropriate latency circle can be selected for different tasks.

[0051] The fixed weight strategy assigns fixed weights to different compute nodes and distributes tasks to them based on the weight ratio. This strategy is suitable for scenarios where compute node performance is relatively stable and task types are relatively fixed, achieving more balanced resource utilization.

[0052] The computing power balancing strategy is to monitor the computing power usage of each computing node in real time, and keep the computing power load of each computing node balanced by dynamically allocating the tasks performed by each computing node; the computing power balancing strategy keeps the computing power load of each node balanced, avoiding the situation where some nodes are overloaded while other nodes are idle as much as possible, and improving overall resource utilization.

[0053] The computing-network collaboration strategy allocates tasks to each computing node based on the synergy between computing and network resources, achieving optimal allocation of computing network resources. This strategy not only considers the computing power of the computing nodes, but also factors such as network bandwidth and stability when scheduling and allocating computing power. For example, in large data transmission tasks, the computing-network collaboration strategy selects appropriate computing nodes based on network bandwidth availability, improving data transmission speed and stability.

[0054] The computer coordination strategy allocates computing power based on the task's computing and data processing requirements, as well as computing and data resources. For tasks requiring large amounts of data processing, the computer coordination strategy prioritizes nodes with rich data storage and strong computing power, thereby improving task processing efficiency.

[0055] In some embodiments, the applicable scenarios and triggering conditions of each preset computing power scheduling strategy are shown in Table 1.

[0056] Table 1 Policy Name Applicable Scenarios Trigger Condition Example Proximity access strategy High data locality requirements (such as AR navigation) The task data volume is >1GB and is latency sensitive Latency priority strategy Real-time control tasks (such as industrial control robots) SLA latency requirement: <50ms Latency Circle Strategy Regional services (such as CDN content delivery) Tasks are allowed to be processed within a specified geographical radius Fixed weight strategy Mixed load scenario (default policy) Mixing of multiple types of tasks without special constraints Computing power balancing strategy Uneven node load (e.g., a GPU node utilization rate > 90%) The difference in computing power between nodes exceeds 30% Computing network collaboration strategy Distributed tasks with limited network bandwidth Cross-region tasks with bandwidth utilization > 80% Computer collaboration strategy Green energy priority scenario Tasks are marked as "low carbon priority" In some possible implementations, based on task characteristics and pre-set computing power scheduling strategies, the optimal computing power scheduling strategies include: Get the current status of the system, Select one or more computing power scheduling strategies from the pre-set computing power scheduling strategies based on the task characteristics and the current system status.

[0057] It's important to note that intelligent decision-making algorithms must be used to select the appropriate scheduling strategy or strategy combination based on task characteristics and the current system state. For example, for tasks with high real-time requirements and small data volumes, a latency-first strategy is preferred. For tasks with high computational complexity and high computing power requirements, a combination of computing power balancing and computing-network collaboration strategies is used for scheduling.

[0058] In some possible implementations, selecting one or more computing power scheduling strategies from pre-set computing power scheduling strategies based on task characteristics and the current state of the system includes: Based on the task characteristics, determine whether there are any tasks with hard constraints. If so, first select the preset computing power scheduling strategy based on the hard constraints. Then, based on the remaining task characteristics and the computing capabilities of the computing nodes, select the corresponding computing power scheduling strategy. The hard constraints are the preset computing power scheduling strategies that must be selected to meet the task characteristics. If it does not exist, the corresponding computing power scheduling strategy is selected based on the task characteristics and the computing power of the computing nodes.

[0059] That is, hard constraints serve as the first level of screening. In some embodiments, hard constraints include: latency-sensitive tasks prioritize the latency-priority strategy, eliminating compute nodes that cannot meet latency requirements; and high-data-volume tasks enforce the proximity-based access strategy to reduce transmission overhead.

[0060] In some embodiments, based on (other) task characteristics and the computing capabilities of the computing nodes, selecting a corresponding computing power scheduling strategy includes: Evaluate the matching degree between the network bandwidth and computing power of each computing node through the computing network collaboration strategy; Schedule the task to a computing node that matches the computing node's network bandwidth and computing power.

[0061] For example, a specific implementation example of computing power scheduling for a city cloud-edge system is provided below.

[0062] The following tasks exist simultaneously in a city's cloud-edge system: Task A: Real-time traffic light control (latency < 30ms, low computational effort).

[0063] Task B: Intersection surveillance video analysis (data volume 2GB, requires GPU computing power).

[0064] Task C: Historical traffic flow prediction (allowing a delay of 500ms, requiring big data processing).

[0065] Receiving a task request means receiving Task A, Task B, and Task C.

[0066] Perform feature extraction on the above tasks ABC: a. Task A is labeled as [real-time control, delay-sensitive, no energy consumption constraint, bandwidth-sensitive].

[0067] b. Task B is marked as [computation-intensive, latency tolerance 100ms, high energy consumption, high bandwidth].

[0068] c. Task C is labeled [Batch, Latency Relaxed, Green Computing, Low Bandwidth].

[0069] According to the task characteristics and the preset computing power scheduling strategy, the optimal computing power scheduling strategy is obtained, and computing power scheduling is performed according to the optimal computing power scheduling strategy.

[0070] Specifically, Task A triggers a latency-prioritized strategy, selecting the edge node closest to the intersection (with a latency radius of 1 km) and allocating redundant copies to two nodes using a fixed-weight strategy. Task B prioritizes proximity (transmitting a 2GB video to the nearest GPU node) and simultaneously checks bandwidth utilization using a computing-network collaboration strategy. If bandwidth utilization exceeds a threshold, the task is diverted to a neighboring area. Task C activates a power-computing collaboration strategy, dispatching to a solar-powered cloud data center and dynamically allocating computing resources using a computing power balancing strategy.

[0071] S104: Perform computing power scheduling according to the optimal computing power scheduling strategy.

[0072] That is, according to the selected optimal computing power scheduling strategy, tasks are assigned to appropriate computing nodes for execution, thereby realizing computing power scheduling.

[0073] It should be noted that during the task execution process, the task execution status and system resource usage are monitored in real time. If an abnormal situation occurs, the scheduling strategy is adjusted in a timely manner. Specifically, in some possible implementations, the method further includes: Obtain real-time feedback on the performance of computing power scheduling; Adjust task characteristics or pre-set computing power scheduling strategies based on real-time feedback information.

[0074] That is, after a computing node receives a task and begins executing it, the system monitors the task execution progress and resource usage in real time. If any abnormalities in task execution or unexpected resource usage are detected, appropriate measures are taken promptly.

[0075] It should be noted that adjusting task characteristics includes adjusting task priority, adjusting task classification results, etc.

[0076] It should also be noted that adjusting the preset computing power scheduling strategy can, on the one hand, adjust a specific preset computing power scheduling strategy, such as readjusting the weights of different computing nodes in a fixed weight strategy. On the other hand, all preset computing power scheduling strategies can be adjusted. For example, based on real-time feedback on the execution effect of task scheduling (such as actual delay and energy consumption), the weight of each computing power scheduling strategy is updated based on the feedback data, and dynamic strategy switching and parameter tuning are achieved based on reinforcement learning (such as the DQN algorithm).

[0077] In some possible implementations, the method further includes: When a temporary failure occurs on a computing node executing a task, the task is migrated to a neighboring computing node for execution.

[0078] This allows tasks to be completed even when a compute node experiences a temporary failure, improving the user experience.

[0079] In some embodiments, node status is monitored in real time. If the load fluctuation is greater than 20%, the computing power balancing strategy is activated to migrate tasks. Alternatively, when sufficient renewable energy supply is detected, the power coordination strategy is activated for low-carbon tasks. For example, in the above-mentioned urban cloud-edge system computing power scheduling embodiment, if monitoring detects that the load of a GPU node reaches 85%, the computing power balancing strategy is triggered to migrate some subtasks of Task B to idle nodes. Alternatively, when sufficient solar power is supplied at noon, the power coordination strategy is weighted by 20%, giving priority to allocating green tasks to this node.

[0080] As a possible implementation method, the present invention proposes a scheduling strategy model to realize the mechanism of multi-strategy coordinated scheduling. Figure 2 As shown below, the coordination relationship between each module is as follows: Figure 3 shown.

[0081] Task management module: responsible for receiving, storing, classifying and preprocessing tasks, and providing task information for the scheduling execution module.

[0082] Feature extraction and analysis module: Use machine learning and data analysis techniques to extract and analyze task features and provide data support for strategy selection.

[0083] Strategy decision module: Based on task characteristics and system status, it uses intelligent decision-making algorithms to select appropriate scheduling strategies or strategy combinations from pre-set scheduling strategies.

[0084] Scheduling and execution module: According to the selected optimal scheduling strategy, tasks are assigned to appropriate computing nodes for execution, and the task execution status is monitored in real time.

[0085] Resource management module: responsible for the unified management and operation and maintenance of computing resources and network resources, and provides resource information for scheduling decisions.

[0086] Monitoring and detection module: responsible for detecting and monitoring the computing power status of edge computing nodes and central computing nodes, detecting and monitoring the network status between computing nodes and between users and computing nodes, and saving the detection data in the time series database for use by the feature extraction and analysis module.

[0087] Feedback optimization module: collects task execution results and user feedback, and optimizes and adjusts scheduling strategies and processes.

[0088] According to the above description, compared with the prior art, this application has the following technical advantages: 1. A multi-strategy coordination mechanism breaks through the limitations of single-target scheduling: Traditional technologies rely on fixed rules or single-objective optimization (such as task scheduling based on the LOF algorithm), making them difficult to adapt to diverse scenarios. This application innovatively integrates seven types of computing power scheduling strategies (latency priority, latency zone division, computing network collaboration, etc.), and subsequently dynamically assigns strategy weights through reinforcement learning to achieve global optimal scheduling. This mechanism addresses the shortcomings of traditional methods, such as low resource utilization and large latency fluctuations.

[0089] 2. Dynamic resource scheduling and cross-domain collaborative optimization: In existing technologies, computing power, network, and power resources are independently scheduled, resulting in low energy efficiency. This application builds a joint optimization model by analyzing multi-dimensional resource profiles (CPU / GPU utilization, network bandwidth, and electricity price peaks and valleys), achieving cross-domain resource integration for the first time. For example, the computing collaboration strategy combines GPU power consumption models with electricity price peak and valley models to support task execution during off-peak hours, reducing overall energy costs by 20%.

[0090] 3. Intelligent Traffic Scheduling and Load Balancing: Traditional traffic scheduling methods are typically static and unable to adjust traffic distribution in real time based on network status, which can lead to congestion or imbalance in network paths. However, the intelligent routing and dynamic traffic scheduling mechanism of this application can automatically select the optimal path based on real-time network conditions, link load, bandwidth requirements, and other factors, greatly reducing the probability of congestion and improving network performance. This network-state-based intelligent scheduling significantly improves the throughput and efficiency of multi-cloud interconnection.

[0091] 4. Efficient automated management: By combining automated management with fault recovery mechanisms, the system of the present invention can automatically manage network status, monitor performance, perform fault detection, and switch paths without human intervention, reducing reliance on manual operations and improving the efficiency and reliability of network operations and maintenance.

[0092] The method provided in this application can accurately capture the impact of different links and network topologies on latency. It is applicable to multi-region, complex network environments and can help identify links and paths with high latency. However, the current algorithm's requirements for computing resources must also be considered, as it requires accurate network topology and link latency data, and large-scale networks require extensive computing and data collection. Applicable scenarios are as follows: Smart security: In smart security scenarios, the system receives video surveillance tasks from various monitoring points. Based on the task's location, it uses a proximity-based access strategy to assign the task to the edge computing node closest to the monitoring point for real-time video analysis, such as facial recognition and behavior detection. Furthermore, a latency-prioritized strategy ensures that video analysis results are promptly fed back to the monitoring center, enabling rapid response.

[0093] Industrial Internet: In industrial internet scenarios, the system receives sensor data and equipment control tasks from production equipment. Based on the computational complexity and latency requirements of these tasks, computing power balancing and computing network collaboration strategies are employed to allocate data processing tasks to appropriate computing nodes, ensuring real-time processing of production data and precise equipment control.

[0094] Smart Healthcare: In smart healthcare scenarios, the system receives medical imaging data and diagnostic tasks from medical devices. Based on the data volume and computing requirements, a computer-based collaborative strategy is employed to store the medical imaging data on data-rich nodes and distribute it to high-performance computing nodes for diagnostic analysis, improving the accuracy and efficiency of medical diagnoses.

[0095] See also Figure 4 , is a computing power scheduling device provided in an embodiment of the present application, such as Figure 4 As shown, the device includes: Receiving unit 401, used for receiving a task request; The scheduling unit 402 is used to extract features of the task; obtain an optimal computing power scheduling strategy based on the task features and a pre-set computing power scheduling strategy; and perform computing power scheduling according to the optimal computing power scheduling strategy.

[0096] As a possible implementation manner, the scheduling unit 402 is specifically configured to extract feature vectors of tasks and perform weighted processing on the tasks according to preset indicators.

[0097] As a possible implementation manner, the scheduling unit 402 is specifically configured to mark the priorities of the tasks according to the extracted features.

[0098] As a possible implementation method, the scheduling unit 402 is specifically used to obtain the current state of the system, and select one or more computing power scheduling strategies from pre-set computing power scheduling strategies based on the task classification result and the current state of the system.

[0099] As a possible implementation, the scheduling unit 402 is specifically configured to select one or more computing power scheduling strategies from preset computing power scheduling strategies based on the task classification result and the current state of the system, including: Based on the task characteristics, determine whether there are tasks containing hard constraints. If so, first select the preset computing power scheduling strategy based on the hard constraints, and then select the corresponding computing power scheduling strategy based on the remaining task characteristics and the computing capabilities of the computing nodes; where the hard constraints are the preset computing power scheduling strategies that must be selected to meet the task characteristics; If it does not exist, the corresponding computing power scheduling strategy is selected based on the task characteristics and the computing capabilities of the computing nodes.

[0100] As a possible implementation method, the scheduling unit 402 is specifically used to obtain real-time feedback information on the execution effect of computing power scheduling; and update the preset computing power scheduling strategy based on the real-time feedback information.

[0101] As a possible implementation manner, the scheduling unit 402 is specifically configured to migrate the task to a neighboring computing node for execution when a temporary failure occurs on the computing node executing the task.

[0102] Corresponding to the above embodiments, the present application also provides an electronic device. Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The electronic device 500 may include: a processor 501, a memory 502, and a communication unit 503. These components communicate via one or more buses. Those skilled in the art will appreciate that the structure of the electronic device shown in the figure does not limit the embodiments of the present invention. It may be a bus structure or a star structure, and may include more or fewer components than shown, or combine certain components, or arrange the components differently.

[0103] The communication unit 503 is configured to establish a communication channel so that the electronic device can communicate with other devices, receive user data sent by other devices, or send user data to other devices.

[0104] The processor 501 is the control center of the electronic device. It uses various interfaces and lines to connect various parts of the entire electronic device. It runs or executes software programs and / or modules stored in the memory 502, and calls data stored in the memory to perform various functions of the electronic device and / or process data. The processor can be composed of an integrated circuit (IC), for example, it can be composed of a single packaged IC, or it can be composed of multiple packaged ICs with the same or different functions. For example, the processor 501 can only include a central processing unit (CPU). In an embodiment of the present invention, the CPU can be a single computing core or multiple computing cores.

[0105] The memory 502 is used to store execution instructions of the processor 501. The memory 502 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0106] When the execution instructions in the memory 502 are executed by the processor 501, the electronic device 500 can execute Figure 1 Some or all of the steps in the illustrated embodiments.

[0107] In a specific implementation, the present invention further provides a computer storage medium, wherein the computer storage medium may store a program that, when executed, may include some or all of the steps of each embodiment of the computing power scheduling method provided by the present invention. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0108] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software and a necessary general-purpose hardware platform. Based on this understanding, the technical solutions in the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium such as ROM / RAM, a magnetic disk, or an optical disk, and includes instructions for enabling a computer device (such as a personal computer, server, or network device) to execute the methods described in various embodiments of the present invention, or portions thereof.

[0109] In this specification, reference can be made to the same or similar parts between the various embodiments. In particular, for the device embodiment and the terminal embodiment, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description in the method embodiment.

Claims

1. A computing power scheduling method, characterized in that: The steps include: Receive task requests; performing feature extraction on the task; Obtain the optimal computing power scheduling strategy based on task characteristics and pre-set computing power scheduling strategy; Computing power scheduling is performed according to the optimal computing power scheduling strategy.

2. The computing power scheduling method according to claim 1, characterized in that: The feature extraction for the task includes: Performing feature vector extraction on the task; The tasks are weighted according to preset indicators.

3. The computing power scheduling method according to claim 1, characterized in that: The preset computing power scheduling strategy includes at least one of: a proximity access strategy, a latency priority strategy, a latency circle strategy, a fixed weight strategy, a computing power balancing strategy, a computing network collaboration strategy, and an electronic computing collaboration strategy; The nearest access strategy is based on the geographical information location of the task initiator, and the task is preferentially scheduled to the computing node closest to the geographical information location of the task initiator for execution; The delay priority strategy is to dynamically adjust task allocation based on the real-time network delay and take the shortest task execution delay as the goal; The delay circle strategy is to use preset scheduling rules to allocate tasks within each delay circle to ensure that tasks are executed within the range that meets the preset delay requirements; the delay circles are different delay circles divided according to the preset network topology and delay characteristics; The fixed weight strategy is to assign fixed weights to different computing nodes and assign tasks to different computing nodes according to the weight ratio; The computing power balancing strategy is to monitor the computing power usage of each computing node in real time and dynamically allocate the tasks performed by each computing node to keep the computing power load of each computing node balanced; The computing network collaboration strategy is to allocate tasks to each computing node based on the synergy of computing resources and network resources to achieve optimal configuration of computing network resources; The computer collaboration strategy is to allocate computing power based on the computing requirements and data processing requirements of the task, as well as computing resources and data resources.

4. The computing power scheduling method according to claim 1 or 3, characterized in that: The optimal computing power scheduling strategy obtained based on the task characteristics and the preset computing power scheduling strategy includes: Get the current status of the system, One or more computing power scheduling strategies are selected from preset computing power scheduling strategies based on the task characteristics and the current state of the system.

5. The computing power scheduling method according to claim 1, characterized in that: The selecting one or more computing power scheduling strategies from preset computing power scheduling strategies according to the task classification result and the current state of the system includes: Based on the task characteristics, determine whether there are tasks containing hard constraints. If so, first select the preset computing power scheduling strategy based on the hard constraints, and then select the corresponding computing power scheduling strategy based on the remaining task characteristics and the computing capabilities of the computing nodes; where the hard constraints are the preset computing power scheduling strategies that must be selected to meet the task characteristics; If it does not exist, the corresponding computing power scheduling strategy is selected based on the task characteristics and the computing capabilities of the computing nodes.

6. The computing power scheduling method according to claim 1, characterized in that: The method further comprises: Obtain real-time feedback on the performance of computing power scheduling; Adjust task characteristics or preset computing power scheduling strategies based on the real-time feedback information.

7. The computing power scheduling method according to claim 1, characterized in that: The method further comprises: When a temporary failure occurs on a computing node executing a task, the task is migrated to a neighboring computing node for execution.

8. A computing power scheduling device, characterized in that: include: A receiving unit, configured to receive a task request; A scheduling unit, configured to extract features from the tasks and classify the tasks according to the extracted features; Based on the task classification results and the preset computing power scheduling strategy, an optimal computing power scheduling strategy is obtained; and computing power scheduling is performed according to the optimal computing power scheduling strategy.

9. An electronic device, characterized in that: The electronic device comprises a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the electronic device is triggered to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

Images