A computing resource scheduling method, apparatus, device, medium and product

Abstract

The application discloses a kind of computing resource scheduling method, device, equipment, medium and product.The computing resource scheduling method includes: current task state data in task processing cluster is collected;Based on current task state data and historical task state data contemporaneous with current task state data, task load state prediction is carried out, and predicted task state data is obtained, including predicted task arrival quantity in future preset period, predicted queue backlog of each type of task and predicted risk level of task processing timeout;According to preset scheduling period, based on current task state data, predicted task state data, pre-constructed task resource matching knowledge base, build computing resource allocation strategy;Task resource matching knowledge base is obtained by cross matching to task type knowledge base and computing node capability knowledge base.The technical scheme of the embodiment of the application can realize dynamic task and computing resource matching, reduce waiting time, and improve resource utilization.

Description

Technical Field

[0001] This invention relates to the field of financial technology, and in particular to a method, apparatus, equipment, medium, and product for scheduling computing resources. Background Technology

[0002] Resource allocation in bank branches is prone to mismatch issues. For example, during peak periods, tasks may pile up while computing resources are idle during off-peak periods; complex tasks may be piling up, but computing nodes with the corresponding capabilities may be abnormally offline; and uneven task distribution among computing nodes may lead to excessively high node resource utilization. Achieving precise and dynamic matching of computing resources and tasks within a single branch is key to improving task processing efficiency.

[0003] Existing technologies propose computing resource scheduling based on time-series analysis or graph neural networks to predict the total number of computing resources required by a network node on a given day, assisting manual formulation of daily task execution plans. This method can only determine how many computing resources are needed online, but it cannot achieve precise matching of computing resources and tasks, cannot dynamically control task waiting time, and does not cover scenarios where the number of tasks fluctuates within a day. Summary of the Invention

[0004] This invention provides a computing resource scheduling method, apparatus, device, medium, and product to solve the problems of resource mismatch and task waiting timeout caused by dynamic changes in the number of tasks and fluctuations in task demand.

[0005] According to one aspect of the present invention, a computing resource scheduling method is provided, comprising: Collect current task status data within the task processing cluster; the current task status data includes the number of current tasks arriving, the backlog of task queues for each type of task, the resource utilization rate of computing nodes, and the task waiting time; Based on the current task status data and the historical task status data from the same period, task load status is predicted to obtain predicted task status data. The predicted task status data includes the predicted number of tasks arriving in the future within a preset time period, the predicted queue backlog of each type of task, and the predicted risk level of task processing timeout. Based on the preset scheduling cycle, a computing resource allocation strategy is constructed using the current task status data, the predicted task status data, and a pre-built task resource matching knowledge base. The task resource matching knowledge base is obtained by cross-matching the task type knowledge base and the computing node capability knowledge base.

[0006] According to another aspect of the present invention, a computing resource scheduling apparatus is provided, comprising: The task status acquisition module is used to collect current task status data within the task processing cluster. The current task status data includes the number of current tasks arriving, the backlog of task queues for each type of task, the resource utilization rate of computing nodes, and the task waiting time. The task status prediction module is used to predict the task load status based on the current task status data and historical task status data from the same period as the current task status data, and to obtain predicted task status data. The predicted task status data includes the predicted number of tasks arriving in the future within a preset time period, the predicted queue backlog of each type of task, and the predicted risk level of task processing timeout. The computing resource scheduling module is used to construct a computing resource allocation strategy based on the current task status data, the predicted task status data, and a pre-built task resource matching knowledge base according to a preset scheduling period; the task resource matching knowledge base is obtained by cross-matching the task type knowledge base and the computing node capability knowledge base.

[0007] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the computing resource scheduling method according to any embodiment of the present invention.

[0008] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the computing resource scheduling method according to any embodiment of the present invention.

[0009] According to another aspect of the present invention, a computer program product is provided, including a computer program that, when executed by a processor, implements the computing resource scheduling method of any embodiment of the present disclosure.

[0010] The technical solution of this invention collects current task status data within a task processing cluster. Based on the current task status data and historical task status data from the same period, task load status is predicted to obtain predicted task status data. The predicted task status data includes the predicted number of tasks arriving within a preset future time period, the predicted queue backlog for each type of task, and the predicted risk level of task processing timeout. According to a preset scheduling cycle, a computing resource allocation strategy is constructed based on the current task status data, the predicted task status data, and a pre-built task resource matching knowledge base. By collecting task status and predicted task status in real time, periodic computing resource scheduling is performed to achieve dynamic and accurate matching, reduce task waiting time, and improve computing resource utilization.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of a computing resource scheduling method provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of a computing resource scheduling method provided in Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of a computing resource scheduling device according to Embodiment 3 of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device that implements the computing resource scheduling method of the present invention. Detailed Implementation

[0014] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0015] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention 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 the invention 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 a 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.

[0016] Example 1 Figure 1 This invention provides a flowchart of a computing resource scheduling method according to Embodiment 1. This embodiment is applicable to situations where periodic resource allocation is performed based on the current task status, predicted task status, and the matching of tasks and resources. The method can be executed by a computing resource scheduling system, including edge computing nodes, a scheduling terminal, a task processing cluster, sensing devices, and a feedback terminal. The edge computing nodes are deployed within the network and are used to preprocess data collected by the sensing devices, current and historical task status data obtained from the task processing cluster, and historical feedback data from the task queue obtained from the feedback terminal, and then send these data to the scheduling terminal. The scheduling terminal is used to perform computing resource scheduling operations based on the data sent by the edge computing nodes.

[0017] This method can be executed by a computing resource scheduling device, which can be implemented in hardware and / or software and can be configured in various general-purpose computing devices. For example... Figure 1 As shown, the method includes: S110. Collect current task status data within the task processing cluster; current task status data includes the number of current tasks arriving, the backlog of task queues for each type of task, the resource utilization rate of computing nodes, and task waiting time.

[0018] A task processing cluster is a distributed resource system composed of multiple computing nodes, used to process various types of computing tasks in parallel.

[0019] In this embodiment of the invention, in order to schedule computing resources and improve the utilization rate of computing resources, the current task status data in the task processing cluster is first collected. Specifically, the current task status data may include the number of current tasks arriving, the backlog of task queues for each type of task, the resource occupancy rate of computing nodes, and the task waiting time.

[0020] In a specific example, a monitoring agent can be deployed in the task processing cluster, or application programming interface (API) can be used to collect current task status data at a preset sampling period. For example, the cluster ingress gateway can be used to count the number of current tasks arriving; each task queue manager can be queried to obtain the backlog of task queues for each type of task; the node monitoring interface can be called to read the resource utilization of each compute node, such as CPU, memory, disk I / O, and network bandwidth; and the task waiting time for each task can be calculated based on the timestamp field of each task queue manager.

[0021] Among them, the current task arrival count is the number of tasks that have entered the cluster and are waiting to be processed within the current data collection period, which is used to characterize the real-time load pressure; the task queue backlog is the total number of tasks waiting to be processed in the queue for each type of task; the computing node resource utilization rate is the ratio of the computing resources currently used by each computing node to the total resources, which characterizes the busyness of the node; and the task waiting time is the cumulative waiting time of a task from entering the queue to the current moment without being processed, which is used to evaluate the scheduling timeliness.

[0022] S120. Based on the current task status data and the historical task status data from the same period, predict the task load status to obtain the predicted task status data. The predicted task status data includes the predicted number of tasks arriving in the future within a preset time period, the predicted queue backlog for each type of task, and the predicted risk level of task processing timeout.

[0023] Historical task status data refers to data collected over a previously defined period, such as the number of tasks arriving, the backlog in the task queue, the utilization rate of computing nodes, and the task waiting time within each sampling period over the past three months. The sampling period for historical task status data is consistent with that for current task status data; for example, the sampling period is 15 minutes.

[0024] In this embodiment of the invention, based on the current time, the future preset duration is divided into multiple consecutive and equally long time intervals, and the task status is predicted for each time interval. For example, the next 2 hours are divided into 8 time intervals, each time interval being 15 minutes.

[0025] Furthermore, for each unit time period to be predicted, historical task status data contemporaneous with the current task status data is obtained. From the historical database, the task arrival volume, task queue backlog, task waiting time, and resource utilization of computing nodes are extracted for each day within the past preset number of days, corresponding to the unit time period to be predicted. For example, if the unit time period to be predicted is 10:00-10:15 of the current day, then historical task status data from 10:00-10:15 of each day over the past 90 days is obtained as the historical task status data contemporaneous with the current task status data.

[0026] Based on the current task status data and historical task status data from the same period, task load status is predicted to obtain predicted task status data. For example, the current task status data and historical task status data from the same period are input into a pre-trained load prediction model to obtain the predicted task status data output by the model. The load prediction model can be trained based on a fusion model of empty attention mechanism and long short-term memory network. The predicted task status data includes the predicted task arrival volume for each unit time period within a preset future time period, the predicted queue backlog for each type of task, and the predicted risk level of task processing timeout, where the predicted risk level can be high, medium, or low.

[0027] S130. Based on the preset scheduling cycle, and using the current task status data, predicted task status data, and a pre-built task resource matching knowledge base, construct a computing resource allocation strategy; the task resource matching knowledge base is obtained by cross-matching the task type knowledge base and the computing node capability knowledge base.

[0028] The task resource matching knowledge base is a mapping relationship formed by cross-matching the task type knowledge base and the computing node capability knowledge base. It stores the matching degree between a certain type of task and the processing capability of a certain computing node. The task type knowledge base can include the resource types required for each type of task, computing power requirements, average execution time, and waiting time thresholds. The computing node capability knowledge base can include hardware configuration, available resources, installed software environment, and historical processing efficiency. The capability matching degree can be quantified as a value between 0 and 1, representing the degree to which a computing node is compatible with processing that type of task.

[0029] In this embodiment of the invention, load status is predicted based on current task status data and historical task status data to obtain predicted task status data. Then, a computing resource allocation strategy is constructed based on the current task status data, the predicted task status data, and a pre-built task resource matching knowledge base. Specifically, a resource scheduling benefit function can be constructed using the current task status data, the predicted task status data, and the pre-built task resource matching knowledge base. Furthermore, at least one scheduling constraint term is constructed based on actual task execution requirements. Finally, under the condition of satisfying the scheduling constraint term, the resource scheduling benefit function is solved to obtain the resource allocation strategy.

[0030] For example, the resource scheduling revenue function can be determined by the revenue value allocated to each computing node for each task type. This revenue value can be determined based on current task status data, predicted task status data, and a pre-built task resource matching knowledge base. Specifically, the higher the task resource matching degree, the higher the revenue value; the lower the resource utilization rate of the computing node in the current task status data, the higher the revenue value; and the higher the predicted risk level of task processing timeout in the predicted task status data, the higher the revenue value (i.e., prioritizing tasks with high timeout risk).

[0031] Scheduling constraints can be set based on actual task execution requirements. For example, each computing node can be assigned to at most one type of task in the same cycle; the number of computing nodes assigned to the same type of task cannot exceed the predicted queue backlog of that type of task; assignment is prohibited if the capability matching degree is lower than the threshold; at least one computing node is reserved without being assigned a task, i.e., a spare node is reserved to cope with sudden peaks or node failures.

[0032] Finally, mixed-integer programming or heuristic algorithms, such as genetic algorithms and simulated annealing algorithms, can be used to solve the above resource scheduling benefit function to obtain the decision variable matrix that maximizes the total benefit, i.e., the corresponding resource allocation strategy.

[0033] The technical solution of this invention collects current task status data within a task processing cluster. Based on the current task status data and historical task status data from the same period, task load status is predicted to obtain predicted task status data. The predicted task status data includes the predicted number of tasks arriving within a preset future time period, the predicted queue backlog for each type of task, and the predicted risk level of task processing timeout. According to a preset scheduling cycle, a computing resource allocation strategy is constructed based on the current task status data, the predicted task status data, and a pre-built task resource matching knowledge base. By collecting task status and predicted task status in real time, periodic computing resource scheduling is performed to achieve dynamic and accurate matching, reduce task waiting time, and improve computing resource utilization.

[0034] Example 2 Figure 2 This is a flowchart of a computing resource scheduling method provided in Embodiment 2 of the present invention. This embodiment further refines the above embodiment, providing specific steps for predicting task load status based on the current task status data and historical task status data contemporaneous with the current task status data, to obtain predicted task status data; and specific steps for constructing a computing resource allocation strategy based on the current task status data, the predicted task status data, and a pre-built task resource matching knowledge base according to a preset scheduling period. Figure 2 As shown, the method includes: S210. Collect current task status data within the task processing cluster; current task status data includes the number of current tasks arriving, the backlog of task queues for each type of task, the resource utilization rate of computing nodes, and task waiting time.

[0035] S220. Starting from the current moment, divide the future preset time period into multiple consecutive unit time periods, and for each unit time period, obtain the current task status data, external influencing factors, historical feedback data of the task queue, and target historical task status data of the same period.

[0036] The external influencing factors include time period identifiers and system event identifiers. For example, the event period identifier may include weekdays / rest days, and the system event identifier may include planned maintenance, data backup windows, etc. The historical feedback data of the task queue includes the number of historical timeouts corresponding to a unit of time period (e.g., the number of times the task processing time exceeds the time threshold) and historical task processing scores (the system's score for processing efficiency).

[0037] In this embodiment of the invention, starting from the current moment, the future preset time period is divided into multiple consecutive unit time periods, and for each unit time period, for example, from the current moment to the next 2 hours, it is divided into 8 consecutive 15-minute unit time periods to be predicted, numbered t, t+1, ..., t+7.

[0038] Furthermore, for each unit time period to be predicted, the latest current task status data is obtained, including the most recently sampled current task arrival volume, the task queue backlog for each type of task, the resource utilization rate of computing nodes, and task waiting time. External influencing factors corresponding to the unit time period to be predicted are obtained from external systems, such as calendar services and event management systems, including time period identifiers and system event identifiers. Historical task status data for the time period number is retrieved from historical task status data, specifically the historical task status data for the same period within the past preset number of days. For example, for the unit time period t+1 to be predicted, historical task status data at time t+1 of each day over the past 90 days is extracted. The historical feedback database is retrieved from the historical feedback database, showing the historical timeout counts and historical task processing scores for the same time period each day within the past preset number of days. Finally, the current task status data, external influencing factors, historical feedback data from the task queue, and target historical task status data for the same period are used as input to the subsequent prediction model. By combining external influencing factors and historical feedback data from the task queue, the accuracy of subsequent predictions can be improved.

[0039] S230. Input the target historical task status data, external influencing factors, current task status data, and historical feedback data of the task queue into the load prediction model. Based on the target historical task status data and the historical feedback data of the task queue, the load prediction model uses a spatiotemporal attention mechanism to calculate weighted historical features. Then, input the weighted historical features, external influencing factors, and current task status data into the long short-term memory network to obtain the hidden layer output.

[0040] In this embodiment of the invention, the target historical task status data Ht, external influencing factor Et, current task status data Rt, and historical feedback data Ut of the task queue are aligned in chronological order and combined into a multi-channel input feature vector. Input into the load forecasting model.

[0041] Furthermore, the load prediction model, based on the target historical task state data and the historical feedback data of the task queue, utilizes a spatiotemporal attention mechanism to calculate weighted historical features. The specific formula for calculating the attention score matrix is ​​as follows: , , in, This is the attention weight matrix. It is a weight matrix. It is a bias term. It is the feature dimension, H t It is the target historical task status data, U t It is historical feedback data from the task queue. It is a weighted historical characteristic.

[0042] Furthermore, the weighted historical features, external influencing factors, and current task state data are input into the Long Short-Term Memory network to obtain the hidden layer output, as shown in the following formula: , in, This is the output of the hidden layer of the Long Short-Term Memory network. It is a weighted historical feature. It is an external influencing factor. This is the current task status data.

[0043] Optionally, embodiments of the present invention further include: Read the pre-trained model parameters of the load prediction model from the running cluster, and obtain the initial model parameters by training the initial model based on the historical data of the new cluster; the amount of historical data is less than the set threshold. The pre-trained model parameters are weighted and fused with the initial model parameters to generate the fused model parameters of the new cluster load prediction model. The parameters of the fusion model are synchronized to the initial model of the new cluster to obtain the load prediction model.

[0044] In this optional embodiment, when the newly deployed computing cluster has not accumulated enough historical load data, it can first read the pre-trained model parameters saved by the running cluster of the same type or similar business scenario that is already running stably for the load prediction model.

[0045] At the same time, a small amount of historical task status data of the new cluster is collected locally. The amount of this data is less than a preset threshold, for example, less than 7 days' worth of data. Based on this small amount of data, the initial model is initially trained to obtain the initial model parameters.

[0046] Furthermore, the pre-trained model parameters and the initial model parameters are weighted and fused according to a preset weighting coefficient to generate fused model parameters adapted to the characteristics of the new cluster. The specific formula is as follows: , in, For the fusion model parameters of the new cluster, For the pre-trained model parameters of the already running cluster, These are the initial model parameters for the new cluster. For the transfer weights ( ).

[0047] Through parameter fusion, mature time-series patterns learned from the existing cluster are preserved, while distribution differences are fine-tuned using a small amount of local data. Finally, the fused model parameters are synchronously loaded into the load prediction model of the new cluster, completing model initialization. Through transfer learning, the new cluster can obtain a high-accuracy load prediction model without waiting for a long period of data accumulation, shortening the prediction bias window during the cold start phase, reducing the risk of overfitting due to data sparsity, and improving the resource scheduling accuracy of the newly deployed cluster in the initial stage of operation.

[0048] S240. The output layer maps the output of the hidden layer to the load prediction result as the prediction task status data; including the predicted task arrival amount in each unit time period within the future preset time period, the prediction queue backlog of each type of task, and the prediction risk level of task processing timeout.

[0049] In this embodiment of the invention, the output layer maps the output of the hidden layer to the load prediction result as the final prediction task state data. The specific formula is as follows: in, This is the output layer weight matrix. For output layer bias terms, This is the output of the hidden layer of the Long Short-Term Memory network. It is the predicted task status data (dimension N+2, total number of task types), where the first N elements are the predicted queue backlog of each type of task within a future set unit time period, the N+1th element is the predicted task arrival number, and the N+2th element is the predicted risk level of task processing timeout.

[0050] By dividing the future time period into multiple consecutive time units and simultaneously acquiring the current task status, external influencing factors, historical data from the same period, and historical feedback data for each time unit, a refined capture of load change patterns and multi-dimensional feature alignment are achieved. Based on this, a spatiotemporal attention mechanism is used to weightedly correlate historical task status and historical timeout feedback, enabling the prediction model to automatically focus on highly correlated historical patterns. Combined with deep modeling of temporal dependencies using a long short-term memory network, the final output is a multi-dimensional prediction result including the predicted task arrival rate, predicted queue backlog, and processing timeout risk level. This improves the accuracy and timeliness of load prediction, providing reliable and fine-grained data support for the formulation of subsequent scheduling strategies.

[0051] S250. According to the preset scheduling cycle, read the current resource utilization rate of the computing nodes in the cluster from the current task status data, read the matching degree between the task type and the processing capacity of the computing node from the task resource matching knowledge base, and read the predicted risk level of task processing timeout and the predicted queue backlog of each type of task from the predicted task status data.

[0052] In this embodiment of the invention, the current resource utilization rate L of the computing nodes within the cluster is read from the current task status data according to a preset scheduling period. j Read the matching degree M of the processing capacity between task type i and computing node j from the task resource matching knowledge base. ij And read the predicted risk level R of task processing timeout from the predicted task status data. i And the predicted queue backlog Q for each type of task i .

[0053] S260. Based on the current resource utilization rate, processing capacity matching degree, predicted risk level, and predicted queue backlog, construct a resource scheduling benefit function.

[0054] In this embodiment of the invention, a resource scheduling benefit function can be constructed based on the current resource utilization rate, processing capacity matching degree, predicted risk level, and predicted queue backlog, which is used to transform the computing resource allocation strategy into an optimal value solution. Specifically, the resource scheduling benefit function can be determined by the benefit value allocated to each computing node for each task type. The benefit value allocated to each computing node for each task type can be determined based on the current task status data, predicted task status data, and a pre-built task resource matching knowledge base. Among these, the higher the task resource matching degree, the higher the benefit value; the lower the resource utilization rate of the computing node in the current task status data, the higher the benefit value; and the higher the predicted risk level of task processing timeout in the predicted task status data, the higher the benefit value (i.e., prioritizing the processing of high timeout risk tasks).

[0055] Optionally, based on the current resource utilization rate, processing capacity matching degree, predicted risk level, and predicted queue backlog, a resource scheduling benefit function is constructed, including: The current resource utilization rate, processing capacity matching degree and predicted risk level are weighted and calculated to obtain the revenue value of each computing node allocated to each task type; Based on the revenue value allocated to each computing node for each task type and the predicted queue backlog for each task type, a resource scheduling revenue function is constructed.

[0056] In this optional embodiment, a specific method is provided for constructing a resource scheduling revenue function based on the current resource utilization rate, processing capacity matching degree, predicted risk level, and predicted queue backlog: First, the current resource utilization rate, processing capacity matching degree, and predicted risk level are weighted and calculated to obtain the revenue value allocated to each computing node for each task type. The specific formula is as follows: , in, It is the revenue value allocated to each computing node for each task type. These are the processing capacity matching weight, node load weight, and timeout risk weight. The matching degree between task type i and the processing capacity of computing node j. It calculates the current resource utilization rate of node j. It is the predicted risk level of task type i.

[0057] Furthermore, based on the revenue value allocated to each computing node for each task type and the predicted queue backlog for each task type, a resource scheduling revenue function is constructed. The specific resource scheduling revenue function is as follows: , in, It is the revenue value allocated to each computing node for each task type. It is the predicted queue backlog for task type i. It is a scheduling decision variable ( (distribute), (Not assigned), N is the total number of task types, and K is the total number of online computer nodes.

[0058] By organically integrating the three major scheduling objectives of node capability matching, real-time load balancing, and task timeout risk into a unified quantitative indicator, and maximizing the solution based on this benefit function, the optimal balance between processing throughput, average task waiting time, and node resource utilization can be achieved at the overall level. This avoids the bias caused by optimizing a single objective, thereby significantly improving the global optimality of scheduling decisions and the quality of business services.

[0059] S270. Based on the preset computing node allocation capacity, task queue allocation capacity, capability matching constraints, and elastic reserved resources, at least one scheduling constraint term is constructed, and based on the scheduling constraint term, the resource scheduling benefit function is optimized and solved to obtain the computing resource allocation strategy.

[0060] In this embodiment of the invention, at least one scheduling constraint is constructed based on preset computing node allocation capacity, task queue allocation capacity, capability matching constraints, and elastic reserved resources. Based on these scheduling constraints, the resource scheduling benefit function is optimized to obtain a computing resource allocation strategy. Specifically, the computing node allocation capacity is the maximum number of task types that each computing node can be allocated to within a scheduling cycle; the task queue allocation capacity is the maximum number of computing nodes allocated to queues of the same task type; the capability matching constraint prohibits the allocation of tasks of that type to a computing node when the matching degree between the task type and the computing node's processing capacity is below a set threshold; and the elastic reserved resources are the number of computing nodes that are not allocated any type of task, serving as an elastic buffer resource.

[0061] Within each scheduling cycle, by real-time reading of the current resource utilization of cluster nodes, the matching degree between tasks and node capabilities, the predicted queue backlog, and the timeout risk level, a three-dimensional resource scheduling benefit function that considers capability matching, load balancing, and risk priority is constructed. Multiple scheduling constraints, including node allocation capacity, task queue capacity, capability matching threshold, and elastically reserved resources, are introduced, transforming the resource allocation problem into a constrained optimization model for solution. This not only ensures the executability of the scheduling strategy in the physical cluster but also improves the system's robustness and resilience by reserving buffer resources for sudden loads through an elastic reservation mechanism. The introduction of timeout risk levels enables priority scheduling of high-risk tasks, ultimately achieving a better balance between overall throughput, task latency, and resource utilization.

[0062] Optionally, based on preset computing node allocation capacity, task queue allocation capacity, capacity matching constraints, and elastic reserved resources, at least one scheduling constraint term is constructed, including: Based on the number of task types processed by computing nodes within the same scheduling cycle, a scheduling constraint term based on the allocated capacity of computing nodes is constructed. Based on the number of compute nodes assigned to the same task type and the predicted queue backlog for that task type, a scheduling constraint term based on compute node allocation capacity is constructed. Based on the matching degree between task type and computing node processing capacity and a preset matching degree threshold, a scheduling constraint term based on capacity matching constraint is constructed. Based on the preset scheduling requirements of the task processing cluster, a scheduling constraint based on elastic reserved resources is constructed.

[0063] In this optional embodiment, a specific method is provided for constructing at least one scheduling constraint term based on preset computing node allocation capacity, task queue allocation capacity, capacity matching constraints, and elastic reserved resources: A scheduling constraint term based on the number of task types processed by a computing node within the same scheduling period is constructed. Specifically, the scheduling constraint term corresponding to the computing node allocation capacity ensures that, in solving the resource scheduling benefit function, any computing node can process at most one task type within the same scheduling period. The specific constraint formula is as follows: , ,in, is the scheduling decision variable, and N is the total number of task types.

[0064] Based on the number of compute nodes allocated to the same task type and the predicted queue backlog for that task type, a scheduling constraint term based on compute node allocation capacity is constructed. Specifically, the scheduling constraint term corresponding to the task queue allocation capacity ensures that, in solving the resource scheduling benefit function, the number of compute nodes allocated to queues of the same task type does not exceed the predicted queue backlog for that task type. The specific constraint formula is as follows: , ,in, K is the scheduling decision variable, and K is the total number of online computer nodes.

[0065] Based on the matching degree between task type and computing node processing capacity and a preset matching degree threshold, a scheduling constraint term based on capacity matching constraints is constructed. Specifically, the scheduling constraint term corresponding to the capacity matching constraint prohibits the allocation of tasks of that type to the computing node when the matching degree between the task type and the computing node processing capacity is lower than the preset threshold during the solution of the resource scheduling benefit function. The specific constraint formula is: If (No corresponding skill) ,in, It is a scheduling decision variable. It represents the matching degree between task type i and the processing capabilities of computing node j.

[0066] Based on the preset scheduling requirements of the task processing cluster, scheduling constraints based on elastic reserved resources are constructed. Specifically, the scheduling constraint corresponding to the elastic reserved resources is that, in solving the resource scheduling benefit function, a set number of computing nodes are reserved without being assigned any type of task in each scheduling cycle, serving as elastic buffer resources. The specific constraint formula is as follows: (At least one compute node should be reserved as elastic reserved resources), where, These are the scheduling decision variables, where N is the total number of task types and K is the total number of online computer nodes.

[0067] In the process of calculating the optimal solution of the resource scheduling benefit function, multiple scheduling constraints such as node allocation capacity, task queue capacity, capacity matching threshold, and elastic reserved resources are introduced. This can ensure the executability of the scheduling strategy in the physical cluster. Moreover, the elastic reservation mechanism reserves buffer resources for sudden loads, which can improve the robustness and shock resistance of the system.

[0068] Optionally, embodiments of the present invention further include: During the execution of the computing resource allocation strategy, the health status of computing nodes, the actual queue backlog of each type of task, and the task waiting time are collected according to the set collection period. Calculate the difference between the actual queue backlog and the predicted queue backlog, as well as the difference between the task waiting time and the waiting time threshold associated with the corresponding task type; If the health status of the computing node includes abnormal offline status, the difference between the actual queue backlog and the predicted queue backlog is greater than a set threshold, or the difference between the task waiting time and the waiting time threshold is greater than a set threshold, a resource abnormality event is determined to have occurred. In the event of at least one resource anomaly, trigger an elastic resource scheduling operation that matches the type of the anomaly.

[0069] In this optional embodiment, during the execution of the computing resource allocation strategy, the real-time health status of each computing node in the cluster (e.g., whether the node is online), the actual queue backlog of each type of task (i.e., the number of tasks that have not yet been processed in the queue of each type of task at the current moment), and the actual waiting time of the currently queued task (the cumulative time from when the task was enqueued to the current moment) are continuously acquired according to a preset collection period, such as every 30 seconds or every minute.

[0070] Based on the real-time data collected above, the difference between the actual queue backlog and the predicted queue backlog is calculated as the backlog deviation, and the difference between the actual task waiting time and the waiting time threshold pre-associated with the task type (e.g., 30 seconds for normal tasks and 10 seconds for high-priority tasks) is calculated as the timeout deviation.

[0071] The system determines that a resource anomaly event has occurred when one of the following three situations is detected: First, the health status of the computing node becomes abnormally offline; second, the difference between the actual queue backlog and the predicted queue backlog exceeds the set deviation threshold; third, the proportion of actual task waiting time exceeding the corresponding waiting time threshold reaches the preset alarm line.

[0072] Upon the occurrence of any of the aforementioned resource anomalies, a matching elastic resource scheduling operation is triggered based on the type of the anomaly. Specifically, for a compute node that goes offline abnormally, unprocessed tasks on that node are reassigned to other healthy nodes, and if other healthy nodes lack sufficient compute resources, reserved nodes are used to replace the failed node. For excessive backlog deviations, additional compute resources are requested from the elastic resource pool to alleviate the overloaded task queue. For timeouts, the scheduling priority of timed-out tasks is increased, and they are inserted at the head of the queue. These tasks can also be forcibly assigned to idle nodes or reserved nodes. All of these elastic operations are recorded in the scheduling log for subsequent data analysis.

[0073] Through a closed-loop monitoring and elastic response mechanism, the computing cluster can detect anomalies in real time and proactively correct them during the execution of resource allocation strategies. This effectively avoids business interruptions and long-term task backlogs caused by node failures, prediction errors, or sudden loads, ensuring the satisfaction of service level agreements and the overall operational stability of the cluster.

[0074] The technical solution of this invention collects current task status data within a task processing cluster. Based on the current task status data and historical task status data from the same period, task load status is predicted to obtain predicted task status data. The predicted task status data includes the predicted number of tasks arriving within a preset future time period, the predicted queue backlog for each type of task, and the predicted risk level of task processing timeout. According to a preset scheduling cycle, a computing resource allocation strategy is constructed based on the current task status data, the predicted task status data, and a pre-built task resource matching knowledge base. By collecting task status and predicted task status in real time, periodic computing resource scheduling is performed to achieve dynamic and accurate matching, reduce task waiting time, and improve computing resource utilization.

[0075] Example 3 Figure 3 This is a schematic diagram of a computing resource scheduling device provided in Embodiment 3 of the present invention. Figure 3 As shown, the device includes: The task status acquisition module 310 is used to collect the current task status data within the task processing cluster; the current task status data includes the current number of tasks arriving, the backlog of task queues for each type of task, the resource utilization rate of computing nodes, and the task waiting time. The task status prediction module 320 is used to predict the task load status based on the current task status data and the historical task status data of the same period as the current task status data, and to obtain the predicted task status data. The predicted task status data includes the predicted number of tasks arriving in the future within a preset time period, the predicted queue backlog of each type of task, and the predicted risk level of task processing timeout. The computing resource scheduling module 330 is used to construct a computing resource allocation strategy based on the current task status data, the predicted task status data, and a pre-built task resource matching knowledge base according to a preset scheduling period; the task resource matching knowledge base is obtained by cross-matching the task type knowledge base and the computing node capability knowledge base.

[0076] The technical solution of this invention collects current task status data within a task processing cluster. Based on the current task status data and historical task status data from the same period, task load status is predicted to obtain predicted task status data. The predicted task status data includes the predicted number of tasks arriving within a preset future time period, the predicted queue backlog for each type of task, and the predicted risk level of task processing timeout. According to a preset scheduling cycle, a computing resource allocation strategy is constructed based on the current task status data, the predicted task status data, and a pre-built task resource matching knowledge base. By collecting task status and predicted task status in real time, periodic computing resource scheduling is performed to achieve dynamic and accurate matching, reduce task waiting time, and improve computing resource utilization.

[0077] Optional, the task status prediction module 320 is specifically used for: Starting from the current moment, the future preset time period is divided into multiple consecutive unit time periods. For each unit time period, the current task status data, external influencing factors, historical feedback data of the task queue, and target historical task status data of the same period are obtained. The external influencing factors include time period identifiers and system event identifiers; the historical feedback data of the task queue includes the historical timeout counts and historical task processing scores corresponding to the unit time period. The target historical task status data, external influencing factors, current task status data, and historical feedback data of the task queue are input into the load prediction model. The load prediction model calculates weighted historical features based on the target historical task status data and historical feedback data of the task queue using a spatiotemporal attention mechanism. The weighted historical features, external influencing factors, and current task status data are then input into a long short-term memory network to obtain the hidden layer output. The output layer maps the output of the hidden layer to load prediction results as prediction task status data; the data includes the predicted task arrival volume for each unit time period within a future preset time period, the predicted queue backlog for each type of task, and the predicted risk level of task processing timeout.

[0078] Optionally, the computing resource scheduling module 330 includes: The parameter reading unit is used to read the current resource utilization rate of the computing nodes in the cluster from the current task status data according to the preset scheduling period, read the matching degree between the task type and the processing capacity of the computing node from the task resource matching knowledge base, and read the predicted risk level of task processing timeout and the predicted queue backlog of each type of task from the predicted task status data. The function construction unit is used to construct a resource scheduling benefit function based on the current resource utilization rate, processing capacity matching degree, predicted risk level, and predicted queue backlog. The computing resource scheduling unit is used to construct at least one scheduling constraint term based on preset computing node allocation capacity, task queue allocation capacity, capability matching constraints, and elastic reserved resources, and to optimize the resource scheduling benefit function based on the scheduling constraint term to obtain a computing resource allocation strategy.

[0079] Optionally, a computing resource scheduling unit is specifically used for: Based on the number of task types processed by computing nodes within the same scheduling cycle, a scheduling constraint term based on the allocated capacity of computing nodes is constructed. Based on the number of compute nodes assigned to the same task type and the predicted queue backlog for that task type, a scheduling constraint term based on compute node allocation capacity is constructed. Based on the matching degree between task type and computing node processing capacity and a preset matching degree threshold, a scheduling constraint term based on capacity matching constraint is constructed. Based on the preset scheduling requirements of the task processing cluster, a scheduling constraint based on elastic reserved resources is constructed.

[0080] Optional, function building blocks, specifically used for: The current resource utilization rate, processing capacity matching degree and predicted risk level are weighted and calculated to obtain the revenue value of each computing node allocated to each task type; Based on the revenue value allocated to each computing node for each task type and the predicted queue backlog for each task type, a resource scheduling revenue function is constructed.

[0081] Optionally, the computing resource scheduling device also includes: The status acquisition module is used to collect the health status of computing nodes, the actual queue backlog of each type of task, and the task waiting time according to a set collection period during the execution of the computing resource allocation strategy. The difference calculation module is used to calculate the difference between the actual queue backlog and the predicted queue backlog, as well as the difference between the task waiting time and the waiting time threshold associated with the corresponding task type. An abnormal event confirmation module is used to determine that a resource abnormal event has occurred when the health status of the computing node includes abnormal offline, the difference between the actual queue backlog and the predicted queue backlog is greater than a set threshold, or the difference between the task waiting time and the waiting time threshold is greater than a set threshold. The exception scheduling module is used to trigger an elastic resource scheduling operation that matches the type of exception event when at least one resource exception event occurs.

[0082] Optionally, the computing resource scheduling device also includes: The parameter reading module is used to read the pre-trained model parameters of the load prediction model of the running cluster, and the initial model parameters obtained by training the initial model based on the historical data of the new cluster; the amount of the historical data is less than a set threshold. The parameter fusion module is used to weight and fuse the pre-trained model parameters with the initial model parameters to generate fused model parameters for the load prediction model of the new cluster. The parameter synchronization module is used to synchronize the parameters of the fusion model to the initial model of the new cluster to obtain the load prediction model.

[0083] The computing resource scheduling device provided in the embodiments of the present invention can execute the computing resource scheduling method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0084] In the technical solution of this invention, the information collected is information and data authorized by the user or fully authorized by all parties. The collection, storage, use, processing, transmission, provision, disclosure and application of related data all comply with the relevant laws, regulations and standards of relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0085] Example 4 According to embodiments of the present invention, the present invention also provides an electronic device, a readable storage medium, and a computer program product.

[0086] Figure 4 A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, application processors, blade application processors, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0087] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory 12 or a random access memory 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the read-only memory 12 or a computer program loaded from storage unit 18 into the random access memory 13. The random access memory 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, read-only memory 12, and random access memory 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0088] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0089] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing units, graphics processing units, various special-purpose artificial intelligence computing chips, various processors running machine learning model algorithms, digital signal processors, and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as computational resource scheduling.

[0090] In some embodiments, computational resource scheduling may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via read-only memory 12 and / or communication unit 19. When the computer program is loaded into random access memory 13 and executed by processor 11, one or more steps of the computational resource scheduling described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform computational resource scheduling by any other suitable means (e.g., by means of firmware).

[0091] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays, application-specific integrated circuits (ASICs), application-specific standard products (ASICs), systems-on-a-chip (SoCs), complex programmable logic devices, computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0092] Computer programs used to implement the methods of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs can be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or application.

[0093] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, optical fibers, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0094] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a cathode ray tube or liquid crystal display monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0095] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data application processors), or computing systems that include middleware components (e.g., application application processors), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0096] A computing system can include clients and applications. Clients and applications are generally geographically separated and typically interact via communication networks. The client-application relationship is established by computer programs running on the respective computers and having a client-application relationship with each other. An application can be a cloud application, also known as a cloud computing application or cloud host, which is a host product within the cloud computing application architecture to address the shortcomings of traditional physical hosts and virtual private services, such as high management difficulty and weak business scalability.

[0097] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0098] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for scheduling computing resources, characterized in that, include: Collect current task status data within the task processing cluster; The current task status data includes the number of current tasks arriving, the backlog of task queues for each type of task, the resource utilization rate of computing nodes, and the task waiting time. Based on the current task status data and the historical task status data from the same period, task load status is predicted to obtain predicted task status data. The predicted task status data includes the predicted number of tasks arriving in the future within a preset time period, the predicted queue backlog of each type of task, and the predicted risk level of task processing timeout. Based on the preset scheduling period, a computing resource allocation strategy is constructed using the current task status data, the predicted task status data, and a pre-built task resource matching knowledge base. The task resource matching knowledge base is obtained by cross-matching the task type knowledge base and the computing node capability knowledge base.

2. The method according to claim 1, wherein task load status prediction is performed based on the current task status data and historical task status data contemporaneous with the current task status data to obtain predicted task status data, comprising: Starting from the current moment, the future preset time period is divided into multiple consecutive unit time periods. For each unit time period, the current task status data, external influencing factors, historical feedback data of the task queue, and target historical task status data of the same period are obtained. The external influencing factors include time period identifiers and system event identifiers; the historical feedback data of the task queue includes the historical timeout counts and historical task processing scores corresponding to the unit time period. The target historical task status data, external influencing factors, current task status data, and historical feedback data of the task queue are input into the load prediction model. The load prediction model calculates weighted historical features based on the target historical task status data and historical feedback data of the task queue using a spatiotemporal attention mechanism. The weighted historical features, external influencing factors, and current task status data are then input into a long short-term memory network to obtain the hidden layer output. The output layer maps the output of the hidden layer to load prediction results as prediction task status data; the data includes the predicted task arrival volume for each unit time period within a future preset time period, the predicted queue backlog for each type of task, and the predicted risk level of task processing timeout.

3. The method according to claim 1, wherein a computing resource allocation strategy is constructed based on the current task status data, the predicted task status data, and a pre-built task resource matching knowledge base according to a preset scheduling period, comprising: According to the preset scheduling cycle, the current resource utilization rate of the computing nodes in the cluster is read from the current task status data, the matching degree between the task type and the processing capacity of the computing node is read from the task resource matching knowledge base, and the predicted risk level of task processing timeout and the predicted queue backlog of each type of task are read from the predicted task status data. Based on the current resource utilization rate, processing capacity matching degree, predicted risk level, and predicted queue backlog, a resource scheduling benefit function is constructed. Based on the preset computing node allocation capacity, task queue allocation capacity, capability matching constraints, and elastic reserved resources, at least one scheduling constraint term is constructed, and the resource scheduling benefit function is optimized and solved based on the scheduling constraint term to obtain the computing resource allocation strategy.

4. The method according to claim 3, wherein at least one scheduling constraint is constructed based on preset computing node allocation capacity, task queue allocation capacity, capability matching constraints, and elastic reserved resources, including: Based on the number of task types processed by computing nodes within the same scheduling cycle, a scheduling constraint term based on the allocated capacity of computing nodes is constructed. Based on the number of compute nodes assigned to the same task type and the predicted queue backlog for that task type, a scheduling constraint term based on compute node allocation capacity is constructed. Based on the matching degree between task type and computing node processing capacity and a preset matching degree threshold, a scheduling constraint term based on capacity matching constraint is constructed. Based on the preset scheduling requirements of the task processing cluster, a scheduling constraint based on elastic reserved resources is constructed.

5. The method according to claim 3, wherein a resource scheduling benefit function is constructed based on the current resource occupancy rate, processing capacity matching degree, predicted risk level, and predicted queue backlog, comprising: The current resource utilization rate, processing capacity matching degree and predicted risk level are weighted and calculated to obtain the revenue value of each computing node allocated to each task type; Based on the revenue value allocated to each computing node for each task type and the predicted queue backlog for each task type, a resource scheduling revenue function is constructed.

6. The method according to claim 1, characterized in that, Also includes: During the execution of the computing resource allocation strategy, the health status of computing nodes, the actual queue backlog of each type of task, and the task waiting time are collected according to the set collection period. Calculate the difference between the actual queue backlog and the predicted queue backlog, as well as the difference between the task waiting time and the waiting time threshold associated with the corresponding task type; If the health status of the computing node includes abnormal offline status, the difference between the actual queue backlog and the predicted queue backlog is greater than a set threshold, or the difference between the task waiting time and the waiting time threshold is greater than a set threshold, a resource abnormality event is determined to have occurred. In the event of at least one resource anomaly, trigger an elastic resource scheduling operation that matches the type of the anomaly.

7. The method according to claim 2, characterized in that, Also includes: Read the pre-trained model parameters of the load prediction model from the running cluster, and obtain the initial model parameters by training the initial model based on the historical data of the new cluster; the amount of historical data is less than a set threshold. The pre-trained model parameters and the initial model parameters are weighted and fused to generate the fused model parameters of the load prediction model for the new cluster. The parameters of the fusion model are synchronized to the initial model of the new cluster to obtain the load prediction model.

8. A computing resource scheduling device, characterized in that, include: The task status acquisition module is used to collect the current task status data within the task processing cluster; The current task status data includes the number of current tasks arriving, the backlog of task queues for each type of task, the resource utilization rate of computing nodes, and the task waiting time. The task status prediction module is used to predict the task load status based on the current task status data and historical task status data from the same period as the current task status data, and to obtain predicted task status data. The predicted task status data includes the predicted number of tasks arriving in the future within a preset time period, the predicted queue backlog of each type of task, and the predicted risk level of task processing timeout. The computing resource scheduling module is used to construct a computing resource allocation strategy based on the current task status data, the predicted task status data, and a pre-built task resource matching knowledge base according to a preset scheduling period. The task resource matching knowledge base is obtained by cross-matching the task type knowledge base and the computing node capability knowledge base.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the computing resource scheduling method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the computing resource scheduling method according to any one of claims 1-7.

11. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the computing resource scheduling method according to any one of claims 1-7.

Images