Task scheduling method and device, nonvolatile storage medium and electronic equipment

Abstract

The invention discloses a task scheduling method and device, a nonvolatile storage medium and electronic equipment. The method comprises the steps of determining an emergency degree index of a target task according to submission time and expected completion time of the target task to be scheduled in a cloud platform; determining a resource adaptability index of the resource node according to a calculation performance index of the resource node in the cloud platform and a load rate of the cloud platform, determining a target income for scheduling the target task, and determining a node energy efficiency ratio of the resource node according to power consumption of the resource node and the calculation performance index; and based on the emergency index, the resource suitability index, the target income and the node energy efficiency ratio, evaluating the target task, and scheduling the target task according to an evaluation result. According to the method and the device, the technical problem of unreasonable resource allocation caused by single and static task priority judgment of a related scheduling method is solved.

Description

Technical Field

[0001] This application relates to the field of computer resource management and scheduling technology, and more specifically, to a task scheduling method, apparatus, non-volatile storage medium, and electronic device. Background Technology

[0002] In the context of the rapid development of cloud computing, big data, and artificial intelligence technologies, the efficient management and intelligent scheduling of heterogeneous computing resources have become key factors in improving the performance and service quality of computing platforms. Heterogeneous computing resources refer to data center infrastructure composed of different types of computing units, such as central processing units, graphics processing units, field-programmable gate arrays, and application-specific integrated circuits. Each type of computing resource has irreplaceable advantages in performing specific types of tasks due to its unique architecture and performance characteristics.

[0003] However, the relevant task scheduling methods have exposed a series of technical challenges and limitations when facing heterogeneous resource pools. Specifically, the relevant scheduling algorithms determine the execution priority of tasks based on preset rules or static attributes (such as task submission time and resource demand size), lacking a comprehensive consideration of the urgency of tasks, business value and dynamic adaptability of resources, which may result in critical tasks not being responded to in a timely manner.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides a task scheduling method, apparatus, non-volatile storage medium, and electronic device to at least solve the technical problem of unreasonable resource allocation caused by the single and static judgment of task priority in related scheduling methods.

[0006] According to one aspect of this application, a task scheduling method is provided, comprising: determining an urgency index of a target task based on its submission time and expected completion time in a cloud platform; determining a resource adaptability index of a resource node based on its computing performance index and load rate in the cloud platform; determining the target benefit of scheduling the target task, and determining the node energy efficiency ratio of the resource node based on its power consumption and computing performance index; evaluating the target task based on the urgency index, resource adaptability index, target benefit, and node energy efficiency ratio, and scheduling the target task based on the evaluation results.

[0007] Optionally, determining the target benefit of scheduling the target task includes: identifying the source resource node where the target task is located; determining the load improvement of the source resource node and the target candidate resource node after scheduling the target task from the source resource node to the target candidate resource node, wherein the load improvement is used to characterize the change in node resource utilization caused by scheduling the target task, and the change in node resource utilization is the sum of the decrease in resource utilization of the source resource node and the increase in resource utilization of the target candidate resource node; determining the scheduling overhead incurred in scheduling the target task from the source resource node to the target candidate resource node, wherein the scheduling overhead is jointly determined by the amount of data to be scheduled, network transmission latency, and the overhead of task state serialization and recovery; and subtracting the scheduling overhead from the sum of the load improvement of the source resource node and the load improvement of the target candidate resource node to obtain the target benefit.

[0008] Optionally, determining the target benefit of scheduling the target task includes: obtaining a historical task sequence, wherein the historical task sequence is a sequence of tasks that have been scheduled before the current scheduling time and within a preset time window; concatenating the historical task sequence with additional features of the target task to obtain a conditional sequence, wherein the additional features include at least one of the following: task submission timestamp, requested computation unit type, expected computation duration, workflow identifier, and data storage location information; encoding the conditional sequence using a pre-trained temporal attention encoder to capture the dependency relationship between historical scheduling decisions and the target task, thereby obtaining a sequence encoding vector; processing the sequence encoding vector and the resource adaptability index of candidate resource nodes using a pre-trained decoder, and obtaining an initial target benefit vector corresponding to the candidate resource nodes through cross-attention calculation, wherein the resource adaptability index is used to characterize the comprehensive ability of resource nodes to meet the computational needs and load conditions of the target task; and weightedly fusing the initial target benefit vector and the strategy index to obtain the target benefit, wherein the strategy index includes at least one of the following: the overall load rate of the cloud platform and the unit time pricing cost.

[0009] Optionally, the target task is evaluated based on urgency indicators, resource adaptability indicators, target benefits, and node energy efficiency ratios, including: determining a first weighting coefficient based on the current load status, where the current load status characterizes the overall utilization level of the cloud platform's computing resources; determining a second weighting coefficient based on a resource scarcity index, where the resource scarcity index characterizes the supply-demand ratio of predefined specific types of computing units in the current resource pool; determining a third weighting coefficient based on the scheduling strategy configuration, where the scheduling strategy configuration characterizes the cloud platform's strategy preference among different objectives such as performance optimization, cost control, or energy efficiency priority; determining a fourth weighting coefficient based on an energy efficiency optimization intensity index, where the energy efficiency optimization intensity index characterizes the intensity of the cloud platform's optimization requirements for energy consumption efficiency; and evaluating the target task based on the product of the urgency indicator and the first weighting coefficient, the resource adaptability indicator and the second weighting coefficient, the target benefits and the third weighting coefficient, and the node energy efficiency ratio and the fourth weighting coefficient.

[0010] Optionally, the target task is scheduled based on the evaluation results, including: normalizing the evaluation results to obtain an evaluation score; when the evaluation score of a candidate resource node meets a preset scheduling threshold and the candidate resource node is idle, scheduling the target task to the candidate resource node with the highest evaluation score; when all candidate resource nodes that meet the preset scheduling threshold are occupied, comparing the evaluation score of the target task with the evaluation score of the task currently running on the occupied node, and initiating a task replacement process if the evaluation score of the target task is higher; in the task replacement process, determining the target benefit generated by scheduling the target task from its current node to the occupied node, and when the target benefit is positive, performing a cross-node transfer operation to schedule the target task from its current node to the occupied node; when the evaluation scores of all candidate resource nodes are lower than the preset scheduling threshold, placing the target task in a deferred scheduling queue, waiting for the resource status to change before re-triggering the evaluation process.

[0011] Optionally, after scheduling the target task based on the evaluation results, the task scheduling method further includes: obtaining the scheduling decision corresponding to the target task; performing gradient sensitivity analysis on the evaluation model or decision function on which the scheduling decision is based to obtain the quantitative contribution of urgency index, resource adaptability index, target benefit, and node energy efficiency ratio to the scheduling decision; generating structured explanatory information to explain the scheduling decision based on the quantitative contribution of the features, wherein the structured explanatory information is used to explain the contribution weight of urgency index, resource adaptability index, target benefit, and node energy efficiency ratio; and displaying the scheduling decision and the structured explanatory information in association through a user interface.

[0012] Optionally, the task scheduling method further includes: receiving feedback inputs submitted for scheduling decisions and structured interpretation information, wherein the feedback inputs include: approval, rejection, or specific adjustment intentions for scheduling decisions; based on the feedback inputs, adjusting the weight coefficients or penalty coefficients used in the calculation of scheduling benefits during the generation of scheduling decisions, and writing the adjusted weight coefficients or penalty coefficients back into the evaluation model or decision function.

[0013] According to another aspect of this application, a task scheduling device is also provided, comprising: a first determining module, configured to determine the urgency index of a target task based on the submission time and expected completion time of the target task to be scheduled in a cloud platform; a second determining module, configured to determine the resource adaptability index of a resource node based on the computing performance index of the resource node in the cloud platform and the load rate of the cloud platform; a third determining module, configured to determine the target benefit of scheduling the target task, and determine the node energy efficiency ratio of the resource node based on the power consumption and computing performance index of the resource node; and a scheduling module, configured to evaluate the target task based on the urgency index, resource adaptability index, target benefit, and node energy efficiency ratio, and schedule the target task according to the evaluation result.

[0014] According to another aspect of this application, a non-volatile storage medium is also provided, the storage medium including a stored program, wherein the program, when running, controls the device where the storage medium is located to execute the above-mentioned task scheduling method.

[0015] According to another aspect of this application, an electronic device is also provided, comprising: a memory and a processor, the processor being configured to run a program stored in the memory, wherein the program executes the above-described task scheduling method during runtime.

[0016] According to another aspect of this application, a computer program is also provided, wherein the computer program, when executed by a processor, implements the above-described task scheduling method.

[0017] According to another aspect of this application, a computer program product is also provided, comprising a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-described task scheduling method.

[0018] In this application, the urgency index of a target task is determined based on its submission time and expected completion time in the cloud platform; the resource adaptability index of a resource node is determined based on its computing performance index and load rate; the target benefit of scheduling the target task is determined, and the node energy efficiency ratio of the resource node is determined based on its power consumption and computing performance index; the target task is evaluated based on the urgency index, resource adaptability index, target benefit, and node energy efficiency ratio, and then scheduled based on the evaluation results. By dynamically optimizing the scheduling decision, the priority of the task to be scheduled is determined in a multi-dimensional and dynamic manner, thereby achieving the technical effect of reasonable resource allocation. This solves the technical problem of unreasonable resource allocation caused by the single and static judgment of task priority in related scheduling methods. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0020] Figure 1 This is a flowchart of a task scheduling method according to an embodiment of this application;

[0021] Figure 2 This is a flowchart of another task scheduling method according to an embodiment of this application;

[0022] Figure 3 This is a structural diagram of a task scheduling device according to an embodiment of this application;

[0023] Figure 4 This is a hardware structure block diagram of a computer terminal according to an embodiment of the present application for a task scheduling method. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] According to an embodiment of this application, a method embodiment of a task scheduling method is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0027] Figure 1 This is a flowchart of a task scheduling method according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0028] Step S102: Determine the urgency index of the target task based on the submission time and expected completion time of the target task to be scheduled in the cloud platform.

[0029] Specifically, the urgency index of the task The calculation formula is:

[0030]

[0031] in, The submission time of the target task. The expected completion time for the target task. To avoid extremely small positive values ​​where the denominator is zero. The reciprocal of the urgency rating ensures that the closer the expected completion time is, the higher the urgency rating of the task.

[0032] Step S104: Determine the resource adaptability index of the resource nodes based on the computing performance index of the resource nodes in the cloud platform and the load rate of the cloud platform.

[0033] In step S104, the computing performance metrics (such as CPU utilization, GPU load, and memory bandwidth utilization) and overall load rate of each resource node in the cloud platform are acquired in real time. By analyzing the computing performance metrics and the cloud platform's load rate, the resource node's capacity to handle specific tasks is calculated, i.e., the resource adaptability metric. Resource Adaptability Metrics The calculation formula is:

[0034]

[0035] in, The performance coefficient of the resource node. This represents the current load rate of the node. The calculation formula ensures that resource nodes receive a higher score under high availability and low load conditions, thus giving them priority in resource scheduling.

[0036] Step S106: Determine the target benefit of scheduling the target task, and determine the node energy efficiency ratio of the resource node based on the power consumption and computing performance indicators of the resource node.

[0037] In step S106, the target benefits of scheduling the target task are dynamically evaluated by combining business objectives and resource consumption. Simultaneously, the power consumption and computing performance of resource nodes are considered to determine the node energy efficiency ratio. Specifically, the target benefit calculation includes multiple business dimensions, such as cost savings, improved user experience, and processing efficiency, obtained through a weighted average.

[0038] For example, the target benefit of scheduling a target task can be determined using the following method. For tasks that are in an overloaded or inefficient operating state, the migration benefit can be evaluated using the following model:

[0039]

[0040] in, As the source node, For the target node; and These represent the load improvement resulting from the migration; Migration costs; This represents the cost penalty coefficient.

[0041] In addition, the energy efficiency ratio of a node is measured by the energy consumption per unit of computing power. The calculation formula is:

[0042]

[0043] in, The power consumption per unit time of the node. For computing performance, node energy efficiency ratio The lower the value, the higher the energy efficiency of the resource node.

[0044] Step S108: Based on the urgency index, resource adaptability index, target benefit and node energy efficiency ratio, evaluate the target task and schedule the target task according to the evaluation results.

[0045] In step S108, machine learning algorithms can be used to comprehensively evaluate all tasks to be scheduled, obtaining a priority score for each task. It is understood that the above evaluation process quantifies the scheduling value of the target task on a given resource, thereby determining the optimal scheduling strategy. During the evaluation process, the weights of the indicators are dynamically adjusted based on historical scheduling results to ensure the accuracy and timeliness of the scoring results. Based on the evaluation results, the best resource node is automatically selected for task allocation, or a resource preemption and task migration mechanism is triggered, thereby achieving efficient and intelligent scheduling decisions, ensuring optimal resource utilization and rapid response to critical tasks.

[0046] For example, the target task can be scheduled using the following method.

[0047] The main scheduling process uses the priority score of the "task-resource" combination as the driving input, and integrates task allocation, migration triggering, preemptive scheduling, and energy consumption optimization into the scheduling loop. For each task... With each candidate resource node Calculate the overall score Then, scheduling actions are executed based on the ranking results. The scoring model integrates parameters such as time sensitivity, resource carrying capacity, energy efficiency indicators, and migration benefits, and its formula is as follows:

[0048]

[0049] in, Task urgency, measuring time sensitivity; Resource adaptability reflects the schedulable capability of the target node; Node energy efficiency ratio reflects energy consumption costs; : Migration benefits to determine whether migration is worthwhile; : Configurable or dynamic weight parameters.

[0050] The implementation process is carried out in steps:

[0051] 1. Task access and feature analysis. When a task... After submission, extract its submission time. With the completion time of the requirements Calculate task urgency .

[0052] 2. Resource status awareness and adaptive calculation: Real-time acquisition of computing power performance of each resource node. With current load Calculate adaptability ,in, The higher the value, the more suitable the node is for performing the task.

[0053] 3. Energy Efficiency Indicator Modeling. To integrate energy-saving scheduling constraints, the energy efficiency ratio of each node is defined as: in, per unit power consumption This refers to the computing power or processing capacity of a node.

[0054] 4. Migration Benefit Assessment (when scheduling adjustments or overloaded nodes are involved). If resource-constrained, inefficient, or congested nodes exist, calculate the benefit factor for tasks that may be migrated: .

[0055] 5. Standardize priority value calculation and sorting. Substitute the above indicators into the core scoring formula:

[0056]

[0057] For all candidate nodes Sort by value in descending order.

[0058] 6. Scheduling Action Execution. Based on the scoring results, the scheduler executes one of the following strategies for the task: Initial Allocation: If resources are available, the node with the highest score is selected directly; Preemptive Scheduling: If the node is occupied by a low-priority task and... If the benefit is higher, then task replacement will be performed; migration scheduling: if the migration benefit is higher... If positive, cross-node transfer is initiated; delayed scheduling: when all candidate nodes do not meet the scoring threshold, the process is temporarily suspended.

[0059] 7. Dynamic closed-loop update. This process repeats continuously during task execution and updates adaptively through real-time refresh. , , , Dynamically adjust weights Periodically push out scheduling corrections or migration operations.

[0060] In the above steps, firstly, by analyzing the task submission time and expected completion time, the urgency index of the tasks can be quantified. This accurately identifies which tasks require priority processing due to their urgency, thereby improving the response speed to high-priority tasks and reducing the waiting time for critical tasks. For businesses that rely on real-time computing and response speed (such as autonomous driving, financial transaction analysis, and remote medical diagnosis), this can significantly improve service quality and user experience. Secondly, resource suitability indexes are determined based on the computing performance indicators of resource nodes and the load rate of the entire cloud platform. This ensures that target tasks are allocated to the most suitable resources, avoiding efficiency losses and computing bottlenecks caused by improper resource allocation. By accurately matching tasks and resources, not only can the efficiency and success rate of task execution be improved, but resource utilization can also be optimized, reducing the waste of computing resources, thus achieving load balancing at the global level. Then, determining the target benefits and node energy efficiency ratio for scheduling target tasks can enhance the economic and environmental benefits of the strategy. Target benefits focus on the economic benefits brought by task execution, promoting efficient resource utilization and reducing operating costs. At the same time, evaluating node energy efficiency ratio can favor resource nodes that provide better computing performance per unit of power consumption. Finally, the target task is comprehensively evaluated based on urgency indicators, resource adaptability indicators, target benefits, and node energy efficiency ratio. The evaluation results can guide task scheduling decisions. By integrating the above indicators through a unified scoring model, tasks can be intelligently scheduled to the most suitable resources for execution based on real-time data and the dynamic environment. This dynamic scheduling mechanism ensures that tasks can be completed in the shortest time, at the lowest cost, and with minimal environmental impact, improving the resource allocation efficiency and service level of the cloud platform.

[0061] The following are Figure 1 The steps shown are illustrated and explained by way of example.

[0062] According to some optional embodiments of this application, determining the target benefit of scheduling a target task may specifically include the following steps: identifying the source resource node where the target task is located; determining the load improvement of the source resource node and the target candidate resource node after scheduling the target task from the source resource node to the target candidate resource node, wherein the load improvement is used to characterize the change in node resource utilization caused by scheduling the target task, and the change in node resource utilization is the sum of the decrease in resource utilization of the source resource node and the increase in resource utilization of the target candidate resource node; determining the scheduling overhead incurred in scheduling the target task from the source resource node to the target candidate resource node, wherein the scheduling overhead is jointly determined by the amount of data to be scheduled, network transmission latency, and the overhead of task state serialization and recovery; subtracting the scheduling overhead from the sum of the load improvement of the source resource node and the load improvement of the target candidate resource node to obtain the target benefit.

[0063] In this embodiment, firstly, the source resource node where the target task currently being executed or awaiting execution resides is identified. Secondly, the load improvement amount for each of the two nodes after scheduling the target task from the source resource node to the target candidate resource node is determined. The load improvement amount specifically includes the current resource utilization of the source and target nodes. By analyzing the changes in resource allocation before and after the task migration, the decrease in resource utilization of the source node and the increase in resource utilization of the target node can be calculated. The load improvement amount is the sum of these two, which can intuitively reflect the impact of the scheduling operation on the overall resource utilization efficiency, ensuring that the scheduling decision can improve load balancing.

[0064] Next, the scheduling overhead of scheduling the target task from the source resource node to the target candidate resource node is determined. Scheduling overhead is a composite metric composed of multiple factors: the amount of task data to be scheduled, latency and overhead during network transmission, and the additional costs required for task state serialization and resumption of execution on the target node. It should be noted that scheduling overhead considers not only the physical cost of data migration itself but also the software-level costs of task state transitions, providing a comprehensive cost basis for scheduling decisions.

[0065] Finally, the load improvement of the source resource node is added to the load improvement of the target candidate resource node to obtain the total load improvement effect. Then, the previously calculated scheduling overhead is subtracted from this total improvement to obtain the target benefit of the scheduling operation. The target benefit visually represents the net effect of the scheduling operation; a positive benefit indicates that scheduling can optimize resource utilization and save costs, while a negative benefit indicates that the potential losses from scheduling outweigh the benefits. Quantitative analysis ensures the economic efficiency of scheduling decisions, enabling scheduling operations to find the optimal balance between improving efficiency and controlling costs.

[0066] Through the above steps, the target benefits of heterogeneous computing power resource scheduling can be accurately quantified, supporting intelligent scheduling strategies based on data and benefits, and ensuring that scheduling decisions not only consider immediate resource utilization efficiency, but also take into account long-term economic benefits and resource costs.

[0067] According to some optional embodiments of this application, determining the target benefit of scheduling a target task may specifically include the following steps: obtaining a historical task sequence, wherein the historical task sequence is a sequence of tasks that have been scheduled before the current scheduling time and within a preset time window; concatenating the historical task sequence with additional features of the target task to obtain a conditional sequence, wherein the additional features include at least one of the following: task submission timestamp, requested computation unit type, expected computation duration, workflow identifier, and data storage location information; encoding the conditional sequence using a pre-trained temporal attention encoder to capture the dependency relationship between historical scheduling decisions and the target task, thereby obtaining a sequence encoding vector; processing the sequence encoding vector and the resource adaptability index of candidate resource nodes using a pre-trained decoder, and obtaining an initial target benefit vector corresponding to the candidate resource nodes through cross-attention calculation, wherein the resource adaptability index is used to characterize the comprehensive ability of resource nodes to meet the computational needs and load conditions of the target task; and weightedly fusing the initial target benefit vector and the strategy index to obtain the target benefit, wherein the strategy index includes at least one of the following: the overall load rate of the cloud platform and the unit time pricing cost.

[0068] In this embodiment, recent historical task sequences are collected as the data foundation for analyzing scheduling patterns and optimizing scheduling strategies. By limiting the time range, the historical data is ensured to be relevant to the current environment, improving the timeliness and accuracy of the analysis. The historical task sequences are concatenated with relevant additional features of the target task to form a conditional sequence. These additional features include: task submission timestamp, requested computational unit type, expected computation duration, workflow identifier, and data storage location information. This information collectively constitutes a rich context for the conditional sequence, providing necessary background knowledge and historical reference for subsequent intelligent decision-making.

[0069] A pre-trained temporal attention encoder is used to process conditional sequences in order to uncover the potential dependencies between historical scheduling decisions and the current target task. The encoder transforms the conditional sequences into sequence encoding vectors, which include information on task attributes, historical scheduling patterns, and their interactions, providing a structured data representation for the next stage of decoding and decision-making.

[0070] A pre-trained decoder is used to further process the sequence encoding vector and the resource suitability index of candidate resource nodes. The resource suitability index represents the comprehensive ability of a resource node to meet the task's computational requirements and the current load condition. The decoder uses cross-attention computation to combine insights from historical scheduling decisions with the characteristics of the current resource node, generating an initial target benefit vector for each candidate resource node. This initial target benefit vector represents the expected benefit of assigning the target task to a specific resource node, considering resource matching and environmental load.

[0071] The initial target revenue vector is weighted and fused with a series of strategy indicators to derive the final target revenue for scheduling each candidate resource node. These strategy indicators are crucial references for evaluating the economic benefits and system stability of scheduling strategies at a macro level. These indicators include, but are not limited to, the overall load rate of the cloud platform and the unit time pricing cost. Through weighted fusion, various factors can be balanced, ensuring that scheduling decisions consider both the needs of the task itself and the overall platform efficiency and cost control. The resulting strategy indicators provide comprehensive guidance for resource allocation, promoting intelligent and efficient scheduling.

[0072] The above steps can effectively integrate historical scheduling experience, target task characteristics, and resource node attributes to generate scheduling decisions that meet task requirements while also taking into account the overall interests of the platform, reflecting the design essence of advanced scheduling algorithms in heterogeneous computing resource environments.

[0073] In some optional embodiments of this application, the target task is evaluated based on urgency indicators, resource adaptability indicators, target benefits, and node energy efficiency ratios. Specifically, this may include the following steps: determining a first weighting coefficient based on the current load state, where the current load state characterizes the overall utilization level of the cloud platform's computing resources; determining a second weighting coefficient based on a resource scarcity index, where the resource scarcity index characterizes the supply-demand ratio of predefined specific types of computing units in the current resource pool; determining a third weighting coefficient based on a scheduling strategy configuration, where the scheduling strategy configuration characterizes the cloud platform's strategy preference among different objectives such as performance optimization, cost control, or energy efficiency priority; determining a fourth weighting coefficient based on an energy efficiency optimization intensity index, where the energy efficiency optimization intensity index characterizes the intensity of the cloud platform's energy efficiency optimization requirements; and evaluating the target task based on the product of the urgency index and the first weighting coefficient, the resource adaptability index and the second weighting coefficient, the target benefits and the third weighting coefficient, and the node energy efficiency ratio and the fourth weighting coefficient.

[0074] In this embodiment, the current load status is first analyzed to determine a first weighting coefficient, which represents the overall urgency of the cloud platform's computing resources. When resource utilization is high, the first weighting coefficient increases, giving higher weight to the urgency index to accelerate the scheduling of critical tasks; conversely, when resources are relatively abundant, the first weighting coefficient decreases accordingly, increasing resource adaptability and energy efficiency.

[0075] Secondly, a second weighting coefficient is determined based on a predefined resource scarcity index. This second weighting coefficient represents the supply-demand ratio of a specific type of computing unit (such as GPU or FPGA). When a certain type of resource is scarce, the resource suitability index of its related tasks will be given a heavier weight to prioritize the allocation of that specific type of resource.

[0076] Then, referring to the scheduling strategy configuration, the third weighting coefficient is determined. The scheduling strategy configuration can be adjusted by operations and maintenance personnel to balance multiple objectives such as performance optimization, cost control, and energy efficiency priority, ensuring that the scheduling results are highly aligned with the enterprise strategy.

[0077] Next, based on the energy efficiency optimization intensity index, a fourth weighting coefficient is determined. The energy efficiency optimization intensity will directly affect the weighting allocation of the node energy efficiency ratio. When energy efficiency requirements are strict, the fourth weighting coefficient increases, which can drive scheduling decisions to tilt towards low energy consumption.

[0078] Finally, a weighted summation method is used to multiply the four sets of indicators by their respective weight coefficients and then sum them up. The calculation formula can be expressed as: Evaluation result = (urgency index × first weight coefficient) + (resource adaptability index × second weight coefficient) + (target benefit × third weight coefficient) + (node ​​energy efficiency ratio × fourth weight coefficient).

[0079] The above formula comprehensively considers the urgency of tasks, the matching degree between resources and tasks, the economic benefits of scheduling actions, and the need for green computing. This ensures that the scheduling priority of each task in a heterogeneous computing environment is both reasonable and efficient. Through multi-dimensional weighting, a more comprehensive understanding of task characteristics and resource status can be achieved, leading to more rational resource allocation decisions and avoiding the limitations of scheduling based on a single indicator. Dynamically adjustable weighting coefficients allow scheduling strategies to change flexibly with variations in load, resource supply and demand, and energy efficiency requirements, ensuring the stability of the scheduling system and rapid response to environmental changes. Simultaneously considering task timeliness, resource efficiency, cost control, and energy efficiency optimization, it enables multi-objective scheduling optimization in complex heterogeneous resource environments, contributing to the overall performance improvement and sustainable development of the cloud platform.

[0080] As some optional embodiments of this application, scheduling the target task based on the evaluation results may specifically include the following steps: normalizing the evaluation results to obtain an evaluation score; when the evaluation score of a candidate resource node meets a preset scheduling threshold and the candidate resource node is idle, scheduling the target task to the candidate resource node with the highest evaluation score; when all candidate resource nodes that meet the preset scheduling threshold are occupied, comparing the evaluation score of the target task with the evaluation score of the task currently running on the occupied node, and initiating a task replacement process if the evaluation score of the target task is higher; in the task replacement process, determining the target benefit generated by scheduling the target task from its current node to the occupied node, and when the target benefit is positive, performing a cross-node transfer operation to schedule the target task from its current node to the occupied node; when the evaluation scores of all candidate resource nodes are lower than the preset scheduling threshold, placing the target task in a deferred scheduling queue, waiting for the resource status to change before re-triggering the evaluation process.

[0081] In this embodiment, the evaluation results are normalized to convert scores with different dimensions into the same range, facilitating comparison and ranking to obtain an evaluation score. This process ensures that each evaluation result can be understood and used under a unified standard, regardless of the characteristics of the target task or the performance differences between candidate resource nodes; the normalized evaluation score can be directly used for decision-making.

[0082] When the evaluation score of a candidate resource node exceeds a preset scheduling threshold, and that node is currently idle, the operation of scheduling the target task to the candidate resource node with the highest evaluation score is executed. This decision ensures that task scheduling only occurs when resource conditions meet certain standards and resource waste is avoided, thereby improving the efficiency and rationality of resource allocation and preventing unnecessary resource consumption.

[0083] If all candidate resource nodes whose evaluation scores meet the preset scheduling threshold are already occupied, further processing is performed by comparing the evaluation score of the target task with the evaluation score of the task currently executing on the occupied node. In this scenario, if the evaluation score of the target task is found to be higher than any task on the occupied node, a task replacement process is initiated, that is, the high-priority target task replaces the low-priority occupied task in order to preempt resource execution.

[0084] In the task replacement process, the target benefit of scheduling the target task from its current location to the occupying node is determined. The benefit calculation includes factors such as the amount of data to be scheduled, network transmission latency, and the overhead of task state serialization and recovery. If the calculation results show that the target benefit generated after migration is positive, that is, the positive effect of the migration operation outweighs the negative effect, then a cross-node transfer operation is performed to schedule the target task from its current location to the occupying node to achieve efficient resource reallocation.

[0085] If the evaluation scores of all candidate resource nodes fail to reach the preset scheduling threshold, it means that no resource node is suitable for immediate scheduling of the task. In this case, the target task is placed in a dedicated deferred scheduling queue, temporarily suspending the scheduling operation and waiting for subsequent changes in resource status. Once resource conditions improve, the evaluation process is retried, and the evaluation scores are recalculated to provide the latest basis for subsequent scheduling decisions.

[0086] By following the above methods and steps, intelligent scheduling of heterogeneous computing resources can be achieved by fully considering resource efficiency, task urgency, and energy consumption optimization strategies. This ensures that critical tasks can be processed in a timely manner, while also making reasonable use of resources and considering economic efficiency, thus providing strong technical support for the efficient operation of the cloud platform.

[0087] In some optional embodiments of this application, after scheduling the target task based on the evaluation results, the task scheduling method may further include the following steps: obtaining the scheduling decision corresponding to the target task; performing gradient sensitivity analysis on the evaluation model or decision function on which the scheduling decision is based to obtain the quantitative feature contribution of urgency index, resource adaptability index, target benefit, and node energy efficiency ratio to the scheduling decision; generating structured explanatory information for explaining the scheduling decision based on the quantitative feature contribution, wherein the structured explanatory information is used to explain the contribution weight of urgency index, resource adaptability index, target benefit, and node energy efficiency ratio; and displaying the scheduling decision and the structured explanatory information in association through a user interface.

[0088] In this embodiment, a scheduling decision corresponding to the target task is obtained. Within the scheduling process, a detailed resource allocation plan has been developed for the specific task. This decision is based on a comprehensive analysis of task characteristics and resource status, aiming to optimally meet task requirements while considering resource efficiency and energy consumption optimization. Gradient sensitivity analysis is performed on the evaluation model or decision function upon which the scheduling decision is generated. The core purpose is to quantify the influence of each key indicator on the final scheduling decision. Through this analysis, the four dimensions of urgency, resource adaptability, target benefit, and node energy efficiency ratio are mapped to their respective contributions to the scheduling decision. The contribution indicates how the scheduling result changes under model parameter fine-tuning, thus revealing the relative importance of each indicator in the decision. Based on the calculated quantified feature contribution, structured explanatory information is generated, representing the weight of each of the urgency, resource adaptability, target benefit, and node energy efficiency ratio in the decision. Creating structured explanatory information makes the basis and influencing factors of scheduling decisions transparent and explainable, which is of great significance for users and operations personnel to understand scheduling logic, evaluate the rationality of decisions, and optimize subsequent strategies. Scheduling decisions and corresponding structured explanatory information are displayed in conjunction with the user interface. Specifically, on the cloud platform's management interface or scheduling console, not only is it displayed which resource node a task will be scheduled to, but also a detailed analysis and explanation behind this decision is presented. For example, the explanatory information may indicate that the scheduling decision is mainly influenced by the high contribution of node energy efficiency ratio, or that the urgency index is given a high weight due to the urgency of the task. Through this method, users can intuitively see the factors considered in the scheduling process, thereby enhancing the transparency of decisions, increasing user trust in the scheduling mechanism, and facilitating timely adjustments to scheduling strategies by operations personnel based on actual business needs.

[0089] On the other hand, based on the evaluation results, after scheduling the target task, the task scheduling method may further include the following steps:

[0090] 1. The scheduling scoring model embeds an interpretation interface. Based on the unified priority scoring model, the scheduling decision score is set as follows:

[0091]

[0092] in: The above four items correspond to task urgency, resource adaptability, node energy efficiency parameters, and migration benefits, respectively, and are the same set of input variables used in the main scheduling process. Weight Vector Parameters in the unified scoring model Binding or mapping.

[0093] 2. Feature Contribution Calculation Mechanism. To reveal the degree of influence of each scheduling input on the final decision, the system calculates the feature contribution based on gradient sensitivity or Shapley value methods:

[0094]

[0095] in, The influence weights of the four types of scheduling factors in this scheduling decision are used to generate explanatory information.

[0096] 3. Decision Path Visualization. The main driving factors in the scheduling results are directly displayed as structured explanatory information, such as: "Priority is given due to high task urgency (contribution 0.33)"; "This node is selected due to its energy efficiency advantage (contribution 0.24)"; "Migration action is triggered because the migration benefit exceeds the threshold (contribution 0.18)". The explanatory results can be presented as a list of factors, a tree-like path, or a causal chain.

[0097] 4. User Feedback and Strategy Optimization. If a user believes a scheduling result is unreasonable, the system allows for fine-tuning of the corresponding weights based on feedback. This, in turn, affects the scheduling model parameters. or migration cost factor Feedback sources can include: manual annotation; policy rule adjustments; comparison of runtime logs; and weight ratio corrections. These adjustments will be automatically written back to the main scheduling model and will take effect in the next scheduling round.

[0098] 5. Closed-loop adaptive update. By combining interpreted output with user feedback, scheduling parameters are continuously adjusted to achieve a closed-loop process of "interpretation—feedback—optimization—rescheduling". This mechanism applies to the same main scheduling loop and requires no additional modules.

[0099] As some alternative embodiments of this application, the task scheduling method may further include the following steps: receiving feedback input submitted for scheduling decisions and structured interpretation information, wherein the feedback input includes: approval, rejection or specific adjustment intention of the scheduling decision; based on the feedback input, adjusting the weight coefficients or penalty coefficients used in the calculation of scheduling benefits during the generation of scheduling decisions, and writing the adjusted weight coefficients or penalty coefficients back into the evaluation model or decision function.

[0100] In this embodiment, feedback input is parsed to identify the user's attitude towards the current scheduling decision (approval, rejection, or specific adjustment intention). Based on the feedback input, the weighting coefficients used in generating the scheduling decision are dynamically adjusted, such as the relative importance of urgency indicators, resource adaptability indicators, target benefits, and node energy efficiency ratios, or the cost penalty coefficients involved in the scheduling benefit calculation. For example, dynamic adjustment can be accomplished through an online learning mechanism. The feedback is quantified into fine-tuned values ​​of weights or coefficients, and the adjusted weighting coefficients or penalty coefficients are promptly written back into the evaluation model or decision function to ensure that subsequent scheduling decisions are more aligned with user needs and actual business scenarios, enhancing the adaptability of the scheduling strategy and user satisfaction. This closed-loop adjustment mechanism, through continuous optimization based on user feedback, achieves the continuous evolution of the scheduling logic, ensuring the rationality and efficiency of scheduling decisions.

[0101] Figure 2 This is a flowchart of another task scheduling method according to an embodiment of this application, such as... Figure 2 As shown, the method includes the following steps:

[0102] Step S201, User / Operations Personnel Submit Task Scheduling Request: When a user or operations team wants to execute a computing task on the cloud platform, they initiate a task scheduling request that enters the system. The request contains basic information about the task, such as task ID, type, required resource specifications, submission time, and expected completion time.

[0103] Step S202: The AI ​​scheduling engine generates scheduling results: Upon receiving a task request, the intelligent scheduling engine begins operation. Based on the algorithm model and real-time resource status, it calculates the matching degree and priority of each resource node for the current task, generating preliminary scheduling results. The scheduling results include the task-resource association score and priority ranking.

[0104] Step S203, calculate the contribution of each feature: The AI ​​scheduling engine further refines the scheduling decision by analyzing features such as task urgency, resource adaptability, energy consumption characteristics and migration benefits, and calculates the degree of influence of each feature on the scheduling decision, i.e., the feature contribution.

[0105] Step S204, Generate an Explainability Report: Based on feature contribution analysis, a detailed explainability report is generated. The report clearly shows the specific contribution of each feature to the scheduling decision, and why a specific resource node was selected for task scheduling. This provides transparency in task scheduling for operations and maintenance personnel, facilitating auditing or subsequent optimization.

[0106] Step S205: Users / Operations personnel review the scheduling explanation and provide feedback: Operations personnel or users have the right to review the scheduling report, understand the scheduling decisions, and provide feedback accordingly. If dissatisfied, they can adjust the weights or specify priorities to pass the feedback information to the scheduling system, triggering the next round of optimization.

[0107] Step S206: Determine if the user is satisfied: If the user is satisfied with the current scheduling result, the process will proceed to the next step; otherwise, the user can express dissatisfaction, prompting the scheduling system to re-evaluate or adjust the strategy.

[0108] Step S207, Execute scheduling strategy: After the user is satisfied, the scheduling strategy is executed, that is, according to the previous evaluation and sorting, the task is assigned to the most suitable resource node and the task execution process is started.

[0109] Step S208, Actual Execution of Resource Tasks: After the task is scheduled to the selected resource node, actual computation begins. During this period, the execution status of the task and the resource usage will be continuously monitored to facilitate subsequent performance evaluation and resource status updates.

[0110] Step S209, New Data Updates Model or Optimizes Decisions: As tasks are executed, newly generated performance and resource usage data are collected and fed back to the AI ​​scheduling engine. This data is used to update model parameters or adjust scheduling strategies, forming a closed-loop learning and optimization process. This allows the scheduling system to continuously evolve over time, improving scheduling accuracy and efficiency.

[0111] The above steps clearly illustrate the complete process from task request, intelligent scheduling, decision interpretation, strategy execution, and closed-loop optimization, demonstrating the flexibility, intelligence, and self-learning capabilities of the intelligent priority scheduling system when handling heterogeneous computing resource scheduling. The entire process ensures the timeliness of task scheduling, the rationality of resource allocation, and the transparency of operation and maintenance management, forming a key technical framework for achieving efficient scheduling in heterogeneous resource environments.

[0112] Figure 3 This is a structural diagram of a task scheduling device according to an embodiment of this application, such as... Figure 3 As shown, the device includes:

[0113] The first determining module 32 is used to determine the urgency index of the target task based on the submission time and expected completion time of the target task to be scheduled in the cloud platform.

[0114] The second determining module 34 is used to determine the resource adaptability index of the resource node based on the computing performance index of the resource node in the cloud platform and the load rate of the cloud platform.

[0115] The third determining module 36 is used to determine the target benefit of scheduling the target task, and to determine the node energy efficiency ratio of the resource node based on the power consumption of the resource node and the computing performance index.

[0116] The scheduling module 38 is used to evaluate the target task based on the urgency index, the resource adaptability index, the target benefit, and the node energy efficiency ratio, and to schedule the target task according to the evaluation results.

[0117] Optionally, determining the target benefit of scheduling the target task may specifically include the following steps: identifying the source resource node where the target task is located; determining the load improvement of the source resource node and the target candidate resource node after scheduling the target task from the source resource node to the target candidate resource node, wherein the load improvement is used to characterize the change in node resource utilization caused by scheduling the target task, and the change in node resource utilization is the sum of the decrease in resource utilization of the source resource node and the increase in resource utilization of the target candidate resource node; determining the scheduling overhead incurred in scheduling the target task from the source resource node to the target candidate resource node, wherein the scheduling overhead is jointly determined by the amount of data to be scheduled, network transmission latency, and the overhead of task state serialization and recovery; subtracting the scheduling overhead from the sum of the load improvement of the source resource node and the load improvement of the target candidate resource node to obtain the target benefit.

[0118] Optionally, determining the target benefit of scheduling the target task may specifically include the following steps: obtaining a historical task sequence, wherein the historical task sequence is a sequence of tasks that have been scheduled before the current scheduling time and within a preset time window; concatenating the historical task sequence with additional features of the target task to obtain a conditional sequence, wherein the additional features include at least one of the following: task submission timestamp, requested computation unit type, expected computation duration, workflow identifier, and data storage location information; encoding the conditional sequence using a pre-trained temporal attention encoder to capture the dependency relationship between historical scheduling decisions and the target task, thereby obtaining a sequence encoding vector; processing the sequence encoding vector and the resource adaptability index of candidate resource nodes using a pre-trained decoder, and obtaining an initial target benefit vector corresponding to the candidate resource nodes through cross-attention calculation, wherein the resource adaptability index is used to characterize the comprehensive ability of resource nodes to meet the computational needs and load conditions of the target task; and weightedly fusing the initial target benefit vector and the strategy index to obtain the target benefit, wherein the strategy index includes at least one of the following: the overall load rate of the cloud platform and the unit time pricing cost.

[0119] Optionally, the target task is evaluated based on urgency indicators, resource adaptability indicators, target benefits, and node energy efficiency ratios. Specifically, this may include the following steps: determining a first weighting coefficient based on the current load status, where the current load status characterizes the overall utilization level of the cloud platform's computing resources; determining a second weighting coefficient based on a resource scarcity index, where the resource scarcity index characterizes the supply-demand ratio of predefined specific types of computing units in the current resource pool; determining a third weighting coefficient based on the scheduling strategy configuration, where the scheduling strategy configuration characterizes the cloud platform's strategy preference among different objectives such as performance optimization, cost control, or energy efficiency priority; determining a fourth weighting coefficient based on an energy efficiency optimization intensity index, where the energy efficiency optimization intensity index characterizes the intensity of the cloud platform's energy efficiency optimization requirements; and evaluating the target task based on the product of the urgency indicator and the first weighting coefficient, the resource adaptability indicator and the second weighting coefficient, the target benefits and the third weighting coefficient, and the node energy efficiency ratio and the fourth weighting coefficient.

[0120] Optionally, the target task is scheduled based on the evaluation results, which may include the following steps: normalizing the evaluation results to obtain an evaluation score; when the evaluation score of a candidate resource node meets a preset scheduling threshold and the candidate resource node is idle, the target task is scheduled to the candidate resource node with the highest evaluation score; when all candidate resource nodes that meet the preset scheduling threshold are occupied, the evaluation score of the target task is compared with the evaluation score of the task currently running on the occupied node, and if the evaluation score of the target task is higher, a task replacement process is initiated; in the task replacement process, the target benefit generated by scheduling the target task from its current node to the occupied node is determined, and when the target benefit is positive, a cross-node transfer operation is performed to schedule the target task from its current node to the occupied node; when the evaluation scores of all candidate resource nodes are lower than the preset scheduling threshold, the target task is placed in a deferred scheduling queue, and the evaluation process is retried after the resource status changes.

[0121] Optionally, after scheduling the target task based on the evaluation results, the task scheduling module has the following specific functions: obtaining the scheduling decision corresponding to the target task; performing gradient sensitivity analysis on the evaluation model or decision function on which the scheduling decision is based to obtain the quantitative contribution of urgency index, resource adaptability index, target benefit, and node energy efficiency ratio to the scheduling decision; generating structured explanatory information to explain the scheduling decision based on the quantitative contribution of the features, wherein the structured explanatory information is used to explain the contribution weight of urgency index, resource adaptability index, target benefit, and node energy efficiency ratio; and displaying the scheduling decision and the structured explanatory information in association through the user interface.

[0122] Optionally, the task scheduling module also has the following functions: receiving feedback inputs submitted for scheduling decisions and structured interpretation information, wherein the feedback inputs include: approval, rejection or specific adjustment intentions of the scheduling decisions; based on the feedback inputs, adjusting the weight coefficients or penalty coefficients used in the calculation of scheduling benefits during the generation of scheduling decisions, and writing the adjusted weight coefficients or penalty coefficients back to the evaluation model or decision function.

[0123] It should be noted that the above Figure 3 The modules in can be program modules (e.g., a set of program instructions that implements a specific function) or hardware modules. For the latter, they can be represented in the following forms, but are not limited to these: each of the above modules is represented by a processor, or the functions of each of the above modules are implemented by a processor.

[0124] It should be noted that, Figure 3 Preferred embodiments of the shown examples can be found in [reference needed]. Figure 1The relevant descriptions of the embodiments shown will not be repeated here.

[0125] Figure 4 A hardware block diagram of a computer terminal for implementing a task scheduling method is shown. Figure 4 As shown, the computer terminal 40 may include one or more processors 402 (shown as 402a, 402b, ..., 402n in the figure) 402 (processor 402 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 404 for storing data, and a transmission module 406 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 4 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 40 may also include... Figure 4 The more or fewer components shown, or having the same Figure 4 The different configurations shown.

[0126] It should be noted that the aforementioned one or more processors 402 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 40. As involved in the embodiments of this application, the data processing circuits serve as processor control (e.g., selection of a variable resistor termination path connected to an interface).

[0127] The memory 404 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the task scheduling method in this embodiment. The processor 402 executes various functional applications and data processing by running the software programs and modules stored in the memory 404, thereby implementing the aforementioned task scheduling method. The memory 404 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 404 may further include memory remotely located relative to the processor 402, and these remote memories can be connected to the computer terminal 40 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0128] The transmission module 406 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 40. In one example, the transmission module 406 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission module 406 may be a radio frequency (RF) module, used for wireless communication with the Internet.

[0129] The display can be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 40.

[0130] It should be noted here that, in some optional embodiments, the above... Figure 4 The computer terminal shown may include hardware elements (including circuitry), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware and software elements. It should be noted that... Figure 4 This is only one instance of a specific particular instance, and is intended to illustrate the types of components that may exist in the aforementioned computer terminal.

[0131] It should be noted that, Figure 4 The computer terminal shown is used to execute Figure 1 The task scheduling method shown above is also applicable to this electronic device, and will not be repeated here.

[0132] This application also provides a non-volatile storage medium, which includes a stored program, wherein the program, when running, controls the device where the storage medium is located to execute the above-mentioned task scheduling method.

[0133] The non-volatile storage medium performs the following functions: determining the urgency index of the target task based on its submission time and expected completion time in the cloud platform; determining the resource adaptability index of the resource nodes based on their computing performance index and load rate; determining the target benefit of scheduling the target task and the node energy efficiency ratio of the resource nodes based on their power consumption and computing performance index; evaluating the target task based on the urgency index, resource adaptability index, target benefit, and node energy efficiency ratio, and scheduling the target task based on the evaluation results.

[0134] This application also provides an electronic device, including a memory and a processor, wherein the processor is used to run a program stored in the memory, and the program executes the above-described task scheduling method during runtime.

[0135] The processor is used to run programs that perform the following functions: determine the urgency index of the target task based on the submission time and expected completion time of the target task to be scheduled in the cloud platform; determine the resource adaptability index of the resource node based on the computing performance index of the resource node in the cloud platform and the load rate of the cloud platform; determine the target benefit of scheduling the target task, and determine the node energy efficiency ratio of the resource node based on the power consumption and computing performance index of the resource node; evaluate the target task based on the urgency index, resource adaptability index, target benefit and node energy efficiency ratio, and schedule the target task according to the evaluation results.

[0136] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0137] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0138] In the above embodiments of this application, the information collected is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with relevant laws, regulations and standards, take necessary protective measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0139] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another component, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection can be through some interfaces; the indirect coupling or communication connection between units or modules can be electrical or other forms.

[0140] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0141] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0142] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to related technologies, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0143] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A task scheduling method, characterized by, The method comprises the following steps: determining an urgency index of a target task to be scheduled in a cloud platform according to a submission time and an expected completion time of the target task; determining a resource adaptability index of a resource node in the cloud platform according to a computing performance index of the resource node and a load rate of the cloud platform; determining a target benefit of scheduling the target task, and determining a node energy efficiency ratio of the resource node according to a power consumption of the resource node and the computing performance index; evaluating the target task based on the urgency index, the resource adaptability index, the target benefit and the node energy efficiency ratio, and scheduling the target task according to an evaluation result.

2. The method of claim 1, wherein, The method for determining the target benefit of scheduling the target task comprises the following steps: identifying a source resource node where the target task is located; determining a load improvement amount of the source resource node and a target candidate resource node respectively after the target task is scheduled from the source resource node to the target candidate resource node, wherein the load improvement amount is used to represent a change amount of node resource utilization caused by scheduling the target task, and the change amount of node resource utilization is a sum of a decrease amount of resource utilization of the source resource node and an increase amount of resource utilization of the target candidate resource node; determining a scheduling overhead generated by scheduling the target task from the source resource node to the target candidate resource node, wherein the scheduling overhead is determined by a to-be-scheduled data amount, a network transmission delay and an overhead of task state serialization and recovery; subtracting the scheduling overhead from a sum of the load improvement amount of the source resource node and the load improvement amount of the target candidate resource node to obtain the target benefit.

3. The method of claim 1, wherein, The method for determining the target benefit of scheduling the target task comprises the following steps: obtaining a historical task sequence, wherein the historical task sequence is a task sequence that has been completed scheduling within a preset time window before a current scheduling time; splicing the historical task sequence and an additional feature of the target task to obtain a conditional sequence, wherein the additional feature comprises at least one of the following: a task submission timestamp, a requested computing unit type, an expected computing duration, an identifier of a belonging workflow and data storage location information; encoding the conditional sequence by using a pre-trained time sequence attention encoder to capture a dependency relationship between historical scheduling decisions and the target task, to obtain a sequence encoding vector; processing the sequence encoding vector and a resource adaptability index of a candidate resource node by using a pre-trained decoder to obtain an initial target benefit vector corresponding to the candidate resource node through cross-attention calculation, wherein the resource adaptability index is used to represent a comprehensive ability of the resource node to meet the computing demand and load status of the target task; weighting and fusing the initial target benefit vector and a strategy index to obtain the target benefit, wherein the strategy index comprises at least one of the following: an overall load rate of the cloud platform and a unit time pricing cost.

4. The method of claim 1, wherein, evaluating the target task based on the urgency indicator, the resource adaptability indicator, the target revenue, and the node energy efficiency ratio, including: determining a first weight coefficient according to a current load state, wherein the current load state is used to represent a utilization level of overall computing resources of the cloud platform; determining a second weight coefficient according to a resource scarcity degree indicator, wherein the resource scarcity degree indicator is used to represent a supply-demand ratio of a predefined specific type of computing unit in a current resource pool; determining a third weight coefficient according to a scheduling strategy configuration, wherein the scheduling strategy configuration is used to represent a strategy preference of the cloud platform between performance optimization, cost control, or energy efficiency priority different targets; determining a fourth weight coefficient according to an energy efficiency optimization intensity indicator, wherein the energy efficiency optimization intensity indicator is used to represent an optimization requirement intensity of the cloud platform on energy consumption efficiency; evaluating the target task based on a product of the urgency indicator and the first weight coefficient, a product of the resource adaptability indicator and the second weight coefficient, a product of the target revenue and the third weight coefficient, and a product of the node energy efficiency ratio and the fourth weight coefficient.

5. The method of claim 1, wherein, scheduling the target task according to the evaluation result, including: normalizing the evaluation result to obtain an evaluation score; when the evaluation score of a candidate resource node meets a preset scheduling threshold and the candidate resource node is in an idle state, scheduling the target task to the candidate resource node with the highest evaluation score; when all candidate resource nodes meeting the preset scheduling threshold are occupied, comparing the evaluation score of the target task with the evaluation score of a task running on an occupied node, and starting a task replacement process when the evaluation score of the target task is higher; in the task replacement process, determining a target revenue generated by scheduling the target task from a current node where the target task is located to an occupied node, and performing a cross-node transfer operation of scheduling the target task from the current node to the occupied node when the target revenue is positive; when the evaluation score of all candidate resource nodes is lower than the preset scheduling threshold, placing the target task in a temporary scheduling queue and waiting for a resource state change to trigger an evaluation process again.

6. The method of claim 1, wherein, After scheduling the target task according to the evaluation result, the method further includes: obtaining a scheduling decision corresponding to the target task; performing gradient sensitivity analysis on an evaluation model or a decision function on which the scheduling decision is based to obtain a quantitative feature contribution degree of the urgency indicator, the resource adaptability indicator, the target revenue, and the node energy efficiency ratio to the scheduling decision; generating structured explanation information for explaining the scheduling decision based on the quantitative feature contribution degree, wherein the structured explanation information is used to explain a contribution weight of each of the urgency indicator, the resource adaptability indicator, the target revenue, and the node energy efficiency ratio; associating and displaying the scheduling decision and the structured explanation information through a user interface.

7. The method of claim 6, wherein, The method further includes: receiving a feedback input submitted for the scheduling decision and the structured explanation information, wherein the feedback input comprises an approval, a denial or a specific adjustment intention for the scheduling decision; adjusting a weight coefficient or a penalty coefficient in a scheduling benefit calculation according to the feedback input, and writing the adjusted weight coefficient or penalty coefficient back to the evaluation model or the decision function.

8. A task scheduling apparatus characterized by comprising: comprising: a first determining module configured to determine an urgency index of a target task according to a submission time and an expected completion time of the target task in a cloud platform; a second determining module configured to determine a resource adaptability index of a resource node in the cloud platform according to a computing performance index of the resource node and a load rate of the cloud platform; a third determining module configured to determine a target benefit of scheduling the target task, and determine a node energy efficiency ratio of the resource node according to a power consumption and the computing performance index of the resource node; a scheduling module configured to evaluate the target task based on the urgency index, the resource adaptability index, the target benefit and the node energy efficiency ratio, and schedule the target task according to an evaluation result.

9. A non-volatile storage medium, comprising: The non-volatile storage medium comprises a stored program, wherein the program controls a device in which the non-volatile storage medium is located to perform the task scheduling method in any one of claims 1 to 7 when the program is running.

10. An electronic device, comprising: comprising: a memory and a processor, the processor being configured to run a program stored in the memory, wherein the program performs the task scheduling method in any one of claims 1 to 7 when the program is running.

11. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the task scheduling method in any one of claims 1 to 7.

Images