Task scheduling method and device for edge node and terminal equipment

Abstract

The invention relates to a task scheduling method and device for edge nodes and terminal equipment. The method comprises the following steps: acquiring state information of a plurality of edge nodes in an edge computing system; inputting the state information of each edge node in the plurality of edge nodes into a node load prediction model to obtain a predicted load value, output by the node load prediction model, of the edge node in a future time period; based on the predicted load value, determining a task type of a predicted calculation task of the edge node in a future time period; determining a task scheduling strategy of the edge node based on the task type of the prediction calculation task and the state information of the edge node; and scheduling the calculation tasks of the edge nodes according to the task scheduling strategy. Through the method and the device, the problem that task processing efficiency of an edge computing system is reduced due to a task scheduling method which only pays attention to hardware indexes of edge nodes in related technologies is solved.

Description

Technical Field

[0001] This application belongs to the field of communication technology, and in particular relates to a task scheduling method, apparatus and terminal equipment for edge nodes. Background Technology

[0002] With the rise of edge computing technology, migrating computing tasks to edge nodes to alleviate the problem of insufficient computing resources on terminal devices has become a hot topic in related fields.

[0003] However, current scheduling of computing tasks for edge nodes often focuses solely on their hardware specifications, neglecting their future load. Consequently, edge nodes with good hardware but high future load (e.g., those scheduled for complex computing tasks within the next 5 minutes) are assigned a large number of computing tasks. This can lead to resource constraints and task congestion on these edge nodes in the future, ultimately reducing the overall task processing efficiency of the edge computing system.

[0004] Currently, no effective solution has been proposed to address the issue that task scheduling methods that only focus on the hardware metrics of edge nodes can reduce the task processing efficiency of edge computing systems. Summary of the Invention

[0005] This application provides a task scheduling method, apparatus, and terminal device for edge nodes, which at least solves the problem in related technologies that task scheduling methods that only focus on the hardware indicators of edge nodes reduce the task processing efficiency of edge computing systems.

[0006] In a first aspect, embodiments of this application provide a task scheduling method for edge nodes, comprising: acquiring state information of multiple edge nodes in an edge computing system; inputting the state information of each of the multiple edge nodes into a node load prediction model to obtain a predicted load value of the edge node in a future time period output by the node load prediction model; determining the task type of the predicted computing task of the edge node in the future time period based on the predicted load value; determining the task scheduling strategy of the edge node based on the task type of the predicted computing task and the state information of the edge node; and scheduling the computing tasks of the edge node according to the task scheduling strategy.

[0007] In some embodiments, the node load prediction model includes a Long Short-Term Memory (LSTM) network module, a Temporal Convolutional Network (TCN) module, and a fusion module. Inputting the state information of each of the plurality of edge nodes into the node load prediction model to obtain the predicted load value of the edge node in a future time period output by the node load prediction model includes: for each edge node, obtaining the historical state information of the edge node; inputting the state information and historical state information of the edge node, as well as the state information of other edge nodes adjacent to the edge node, into the node load prediction model; the LSTM module in the node load prediction model generating a temporal feature vector based on the state information and historical state information of the edge node; the TCN module in the node load prediction model generating a spatial feature vector based on the state information of the edge node and the state information of the other edge nodes; and the fusion module in the node load prediction model fusing the temporal feature vector and the spatial feature vector to generate a fused feature vector, and generating the predicted load value based on the fused feature vector.

[0008] In some embodiments, the fusion module in the node load prediction model fuses the temporal feature vector and the spatial feature vector to generate a fused feature vector, including: the fusion module calculating the correlation between the temporal feature vector and the historical state information; the fusion module calculating the mutual information value between the spatial feature vector and the state information of other edge nodes; the fusion module calculating a fusion weight based on the correlation and the mutual information value; and the fusion module fusing the temporal feature vector and the spatial feature vector according to the fusion weight to generate the fused feature vector.

[0009] In some embodiments, determining the task type of the predicted computing task of the edge node in the future time period based on the predicted load value includes: converting the predicted load value into resource requirement information, the resource requirement information including CPU requirement information, memory requirement information and network requirement information; and determining the task type of the predicted computing task based on the resource requirement information, the task type including latency-sensitive or computationally intensive.

[0010] In some embodiments, determining the task scheduling strategy for an edge node based on the task type of the prediction computation task and the state information of the edge node includes: determining a performance score for each edge node based on the state information of each edge node; for each edge node, when the task type of the prediction computation task of the edge node is latency-sensitive, determining the task scheduling strategy includes: when the performance score of the edge node is less than a first preset threshold, selecting an edge node whose network state meets a first preset condition and whose performance score is greater than a second preset threshold as a target edge node, and scheduling the computation task of the edge node to the target edge node; when the task type of the prediction computation task of the edge node is computationally intensive, determining the task scheduling strategy includes: when the performance score of the edge node is less than the first preset threshold, selecting an edge node whose CPU state meets a second preset condition and whose performance score is greater than the second preset threshold as the target edge node, and scheduling the computation task of the edge node to the target edge node.

[0011] In some embodiments, the status information includes CPU status, memory status, and network status; determining the performance score of each edge node based on the status information of each edge node includes: for each edge node, calculating the CPU utilization of the edge node based on the CPU status of the edge node; calculating the memory occupancy of the edge node based on the memory status of the edge node; calculating the communication latency of the edge node based on the network status of the edge node; and calculating the performance score of the edge node based on the CPU utilization, the memory occupancy, and the communication latency, wherein the performance score is inversely proportional to the CPU utilization, inversely proportional to the memory occupancy, and directly proportional to the reciprocal of the communication latency.

[0012] In some embodiments, scheduling the computing tasks of the edge node according to the task scheduling strategy includes: obtaining the network packet loss rate and network latency jitter value of the edge computing system; when the network packet loss rate is less than a third preset threshold and the network latency jitter value is less than a fourth preset threshold, scheduling the computing tasks of the edge node according to the task scheduling strategy at a preset first task distribution rate; when the network packet loss rate is greater than or equal to the third preset threshold, or the network latency jitter value is greater than or equal to the fourth preset threshold, calculating a second task distribution rate based on the first task distribution rate, the network packet loss rate, and the network latency jitter value, and scheduling the computing tasks of the edge node according to the task scheduling strategy at the second task distribution rate, wherein the second task distribution rate is less than the first task distribution rate.

[0013] Secondly, embodiments of this application provide a task scheduling device for edge nodes, comprising: an acquisition module for acquiring state information of multiple edge nodes in an edge computing system; a prediction module for inputting the state information of each of the multiple edge nodes into a node load prediction model to obtain a predicted load value of the edge node in a future time period output by the node load prediction model; a task type determination module for determining the task type of the predicted computing task of the edge node in the future time period based on the predicted load value; a scheduling strategy determination module for determining the task scheduling strategy of the edge node based on the task type of the predicted computing task and the state information of the edge node; and a scheduling module for scheduling the computing tasks of the edge node according to the task scheduling strategy.

[0014] Thirdly, embodiments of this application provide a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the task scheduling method for edge nodes described in any of the first aspects above.

[0015] Fourthly, embodiments of this application provide a computer program product, including a computer program, which, when run, causes the task scheduling method for edge nodes described in any of the first aspects above to be executed.

[0016] Compared to related technologies, the task scheduling method, apparatus, and terminal device for edge nodes provided in this application can acquire the status information of multiple edge nodes in an edge computing system; then, input the status information of each edge node into a node load prediction model to obtain the predicted load value of the edge node in a future time period output by the node load prediction model; subsequently, based on the predicted load value, determine the task type of the predicted computing task of the edge node in the future time period; and based on the task type of the predicted computing task and the status information of the edge node, determine the task scheduling strategy of the edge node; finally, schedule the computing tasks of the edge node according to the task scheduling strategy. In this way, the load of edge nodes can be predicted in the future using a node load prediction model. By determining the task scheduling strategy for edge nodes based on the predicted load value, precise management of edge node computing resources can be achieved. For example, it avoids scheduling too many computing tasks on edge nodes with high predicted load values, or too few computing tasks on edge nodes with low predicted load values, thus fully utilizing the computing resources of edge nodes and improving the overall task processing efficiency of the edge computing system. Furthermore, by determining the task type of the predicted computing tasks for edge nodes in the future time period based on the predicted load value, and determining the task scheduling strategy based on this task type, precise matching of computing resources based on multi-dimensional real-time indicators (edge ​​node status information) and differentiated task type requirements is achieved, further improving the overall task processing efficiency of the edge computing system. This application solves the problem in related technologies where task scheduling methods that only focus on the hardware indicators of edge nodes reduce the task processing efficiency of edge computing systems, achieving the technical effect of improving the overall task processing efficiency of edge computing systems.

[0017] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

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

[0019] Figure 1 This is a flowchart of a task scheduling method for edge nodes according to an embodiment of this application; Figure 2 This is a flowchart illustrating the process of obtaining the predicted load value of an edge node within a future time period, according to one embodiment of this application. Figure 3This is a flowchart illustrating a task scheduling strategy for determining edge nodes according to an embodiment of this application; Figure 4 This is a schematic diagram of the structure of a task scheduling device for an edge node according to an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a terminal device according to an embodiment of this application. Detailed Implementation

[0020] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0021] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0022] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0023] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0024] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0025] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0026] With the rise of edge computing technology, migrating computing tasks to edge nodes to alleviate the problem of insufficient computing resources on terminal devices has become a hot topic in related fields.

[0027] However, current scheduling of computing tasks for edge nodes often focuses solely on their hardware specifications, neglecting their future load. Consequently, edge nodes with good hardware but high future load (e.g., those scheduled for complex computing tasks within the next 5 minutes) are assigned a large number of computing tasks. This can lead to resource constraints and task congestion on these edge nodes in the future, ultimately reducing the overall task processing efficiency of the edge computing system.

[0028] Furthermore, existing task scheduling strategies ignore the heterogeneous requirements of tasks and often use static rules to schedule tasks on edge nodes. This leads to a serious mismatch between the supply of computing resources and task requirements, further reducing the overall task processing efficiency of the edge computing system.

[0029] Currently, no effective solution has been proposed to address the issue that task scheduling methods that only focus on the hardware metrics of edge nodes can reduce the task processing efficiency of edge computing systems.

[0030] In view of this, embodiments of this application provide a task scheduling method for edge nodes, which can obtain the status information of multiple edge nodes in an edge computing system; then, input the status information of each edge node into a node load prediction model to obtain the predicted load value of the edge node in a future time period output by the node load prediction model; subsequently, based on the predicted load value, determine the task type of the predicted computing task of the edge node in the future time period; and based on the task type of the predicted computing task and the status information of the edge node, determine the task scheduling strategy of the edge node; finally, schedule the computing tasks of the edge node according to the task scheduling strategy. In this way, the load of edge nodes can be predicted in the future using a node load prediction model. By determining the task scheduling strategy for edge nodes based on the predicted load value, precise management of edge node computing resources can be achieved. For example, it avoids scheduling too many computing tasks on edge nodes with high predicted load values, or too few computing tasks on edge nodes with low predicted load values, thus fully utilizing the computing resources of edge nodes and improving the overall task processing efficiency of the edge computing system. Furthermore, by determining the task type of the predicted computing tasks for edge nodes in the future time period based on the predicted load value, and determining the task scheduling strategy based on this task type, precise matching of computing resources based on multi-dimensional real-time indicators (edge ​​node status information) and differentiated task type requirements is achieved, further improving the overall task processing efficiency of the edge computing system. This application solves the problem in related technologies where task scheduling methods that only focus on the hardware indicators of edge nodes reduce the task processing efficiency of edge computing systems, achieving the technical effect of improving the overall task processing efficiency of edge computing systems.

[0031] The following will combine Figure 1 This application describes a task scheduling method for edge nodes according to one embodiment. Please refer to [link to relevant documentation]. Figure 1 , Figure 1 This is a flowchart of a task scheduling method for edge nodes according to an embodiment of this application, as follows: Figure 1 As shown, the method includes: Step S101: Obtain the status information of multiple edge nodes in the edge computing system.

[0032] In this embodiment, distributed probes can be deployed at each edge node in the edge computing system to collect the edge node's status information in real time using a high-frequency sampling method. The edge node's status information can include CPU status, memory status, network status, and task status. For example, CPU status can include Central Processing Unit (CPU) utilization, memory status can include memory occupancy, network status can include network throughput and the number of active connections, and task status can include the edge node's task queue information (e.g., task queue length). Additionally, the distributed probes can also collect the address location information of the terminal devices.

[0033] As an example, distributed probes can be implemented based on a trimmed-down operating system kernel module (e.g., a trimmed-down Linux kernel). For instance, predefined callback hooks can be used to interact with the resource scheduler to obtain real-time performance counter data, thereby collecting status information of edge nodes.

[0034] Step S102: Input the status information of each edge node among multiple edge nodes into the node load prediction model to obtain the predicted load value of the edge node in the future time period output by the node load prediction model.

[0035] In this embodiment, the predicted load value output by the node load prediction model may include a multi-dimensional dataset. Specifically, this multi-dimensional dataset may include the following: (1) Basic load forecast: The predicted load of the edge node in the future time period (e.g., the next 5 minutes, the next 10 minutes or the next 15 minutes), represented by normalized data between 0 and 1: the closer the basic load forecast is to 1, the higher the predicted load of the edge node in the future time period.

[0036] (2) Prediction confidence: a scalar value used to characterize the reliability of the prediction result. The lower the prediction confidence, the higher the uncertainty of the prediction, and the more computing resources should be considered when scheduling tasks for edge nodes.

[0037] Furthermore, this multidimensional dataset can also include the distribution of task types for predictive computing tasks at edge nodes over future time periods. For example, the proportion of computationally intensive tasks and the proportion of latency-sensitive tasks in predictive computing tasks. This can improve the accuracy of determining the resource requirements of predictive computing tasks in subsequent steps.

[0038] In addition, this multidimensional dataset can also include indicators of load change trends for edge nodes over future time periods. For example, load change trend indicators could include "slowly increasing," "rapidly increasing," "stable," "slowly decreasing," and "rapidly decreasing." This also improves the accuracy of determining the resource requirements for predictive computing tasks in subsequent steps.

[0039] As an example, if the predicted load value output by the node load prediction model indicates a low prediction confidence or the load change trend indicator is "rapidly rising", the subsequent steps can use a more conservative strategy to convert the predicted load value into resource demand information. For example, the predicted load value can be mapped to higher resource demand information to build a buffer.

[0040] By using a node load prediction model, the load of edge nodes can be predicted in the future. In subsequent steps, the task scheduling strategy for edge nodes can be determined based on the predicted load value, which can achieve precise management of edge node computing resources. For example, it can prevent edge nodes with high predicted load values ​​from being scheduled for too many computing tasks, or prevent edge nodes with low predicted load values ​​from being scheduled for too few computing tasks, thereby making full use of the computing resources of edge nodes and improving the overall task processing efficiency of the edge computing system.

[0041] In this embodiment, the state information of multiple edge nodes can be used as multi-dimensional time series data. After being integrated by a sliding window and marked with metadata, the data is sent to the node load prediction model via a dedicated data transmission link using an adaptive compression algorithm.

[0042] In one embodiment, the node load prediction model can adopt a dual-stream prediction architecture that integrates temporal and spatial features, and output the predicted load value of edge nodes in future time periods through a dynamic weight allocation mechanism.

[0043] As an example, the node load prediction model includes a Long Short-Term Memory (LSTM) module, a Temporal Convolutional Network (TCN) module, and a fusion module.

[0044] The workflow of the node load prediction model can be found in [link to documentation]. Figure 2 , Figure 2 This is a flowchart illustrating the process of obtaining the predicted load value of an edge node over a future time period according to one embodiment of this application, such as... Figure 2 As shown, step S102 above may include the following steps: Step S201: For each edge node, obtain the historical state information of the edge node.

[0045] In this embodiment, the historical state information of the edge node may include the load information of the edge node over a period of time. For example, the load information of the edge node over the past 10 minutes, past 15 minutes, past half hour, or past 1 hour (load value (represented by normalized data between 0 and 1, with the closer to 1 indicating a higher load) and task type distribution, etc.).

[0046] Step S202: Input the state information and historical state information of the edge node, as well as the state information of other edge nodes adjacent to the edge node, into the node load prediction model.

[0047] Step S203: The LSTM module in the node load prediction model generates a time feature vector based on the state information and historical state information of the edge nodes.

[0048] In this embodiment, the LSTM module can extract temporal feature vectors from the historical state information of edge nodes. Specifically, the LSTM module can adopt a multi-layer gated recurrent unit structure. The input of the LSTM module includes the load information (time series data) of the edge nodes over multiple historical periods, and the output of the LSTM module includes a high-dimensional temporal feature vector.

[0049] Step S204: The TCN module in the node load prediction model generates a spatial feature vector based on the state information of the edge nodes and the state information of other edge nodes.

[0050] In this embodiment, the TCN module can extract spatial feature vectors of edge nodes based on the topological relationship graph between edge nodes. Specifically, the TCN module can be constructed based on dilated convolution and residual connections. The input of the TCN module includes the state information (e.g., load information) of other edge nodes adjacent to the edge node, and the output of the TCN module includes spatial feature vectors.

[0051] Step S205: The fusion module in the node load prediction model fuses the time feature vector and the spatial feature vector to generate a fused feature vector, and generates the predicted load value based on the fused feature vector.

[0052] In this embodiment, the temporal feature vector and spatial feature vector can be fused through a dynamic weight allocation mechanism to provide a basis for generating the predicted load value. Specifically, the dynamic weight allocation mechanism includes: generating fusion weights based on the ratio of temporal contribution to spatial contribution; obtaining the temporal contribution by calculating the Pearson correlation coefficient between the temporal feature vector and historical state information; calculating the mutual information value between the spatial feature vector and the state information of other edge nodes; and then inputting the aforementioned Pearson correlation coefficient and mutual information value into a normalization function to generate the fusion weights. Finally, the node load prediction model can output a predicted load value including a prediction confidence score through a fully connected neural network. This fully connected neural network can include multiple hidden layers, each of which uses a linear rectified function as the activation function. The final output layer uses a linear activation function to generate the predicted load value and additionally outputs a prediction confidence score to characterize the degree of uncertainty of the prediction result.

[0053] As an example, step S205 above, "the fusion module in the node load prediction model fuses the time feature vector and the spatial feature vector to generate a fused feature vector," includes: the fusion module calculating the correlation between the time feature vector and historical state information; the fusion module calculating the mutual information value between the spatial feature vector and the state information of other edge nodes; the fusion module calculating the fusion weight based on the correlation and mutual information value; and the fusion module fusing the time feature vector and the spatial feature vector according to the fusion weight to generate a fused feature vector.

[0054] In this embodiment, the Pearson correlation coefficient can be used to define the correlation between the temporal feature vector and historical state information. The fusion weights include temporal feature fusion weights and spatial feature fusion weights. The temporal feature fusion weights can be expressed as: The spatial feature fusion weights can be expressed as: .

[0055] in, The function representing the calculation of the Pearson correlation coefficient. The function representing the calculation of mutual information values. Represents the time feature vector. This represents historical status information (time-series data of historical load information). Represents spatial eigenvectors. It represents the status information (load information) of other edge nodes adjacent to the edge node.

[0056] Specifically, historical state information can be standardized to eliminate the influence of dimensions; then, the ratio of the covariance to the standard deviation of the time feature vector and the standardized historical state information can be calculated to obtain the Pearson correlation coefficient between the time feature vector and the historical state information.

[0057] Alternatively, a nonparametric method based on kernel density estimation can be used to calculate the joint probability distribution and edge probability distribution of the spatial feature vector and the state information of adjacent nodes (other edge nodes adjacent to the edge node) through the Gaussian kernel function, and then solve for the mutual information value between the two.

[0058] After obtaining the Pearson correlation coefficients of the temporal feature vectors and historical state information, as well as the mutual information values ​​between the spatial feature vectors and the state information of adjacent nodes, the temporal feature fusion weights and spatial feature fusion weights can be calculated. Furthermore, the temporal and spatial feature fusion weights can be mapped using an activation function to determine their specific values, ensuring that the weights fall within a preset range.

[0059] By fusing the temporal and spatial feature vectors according to the temporal and spatial feature fusion weights, a fused feature vector is obtained. A nonlinear transformation of the fused feature vector using a fully connected neural network then outputs the predicted load and prediction confidence.

[0060] In this way, by constructing a node load prediction model that integrates LSTM and TCN dual streams, and by integrating time feature vectors (time-dependent features) and spatial feature vectors (spatial correlation features) through a dynamic weight allocation mechanism, it is possible to accurately predict the load information of edge nodes in the future time period. This overcomes the limitations of spatiotemporal feature fragmentation in traditional load prediction methods, and significantly reduces the problem of excessive or insufficient reserved computing resources due to prediction deviations, thereby improving the overall computing resource utilization of the edge node cluster.

[0061] Step S103: Based on the predicted load value, determine the task type of the prediction calculation task of the edge node in the future time period.

[0062] In this embodiment, the predicted load value can be mapped to a virtual resource demand vector, and the task type of the predicted computing task of the edge node in the future time period can be determined based on the virtual resource demand vector.

[0063] Mapping strategies can be determined based on the prediction confidence level or load change trend indicator included in the predicted load value. For example, if the predicted load value indicates a low prediction confidence level or the load change trend indicator is "rapidly rising," a more conservative mapping strategy can be used to convert the predicted load value into a virtual resource demand vector. For instance, the predicted load value can be mapped to a higher virtual resource demand vector to build a buffer.

[0064] In one embodiment, step S103 includes: converting the predicted load value into resource requirement information, which includes CPU requirement information, memory requirement information, and network requirement information; and determining the task type of the predicted computing task based on the resource requirement information, which includes latency-sensitive or computationally intensive tasks.

[0065] In this embodiment, CPU requirement information may include the number of virtual cores, memory requirement information may include memory capacity, and network requirement information may include bandwidth requirements. Resource requirement information may also include peak CPU utilization of edge nodes in the future time period, expected task execution time, predicted network latency, and predicted network latency jitter.

[0066] After obtaining resource requirement information, a lightweight classifier can be used to determine the task type of the prediction computation task. This lightweight classifier can adopt a decision model based on multi-feature weighting, and its determination formula can be expressed as: Task type score = ω1 × computational intensity + ω2 × latency sensitivity, ω1 + ω2 = 1; The computational intensity is calculated as follows: (historical average CPU utilization ÷ peak CPU utilization) × (expected task execution time ÷ baseline execution time); the latency sensitivity is calculated as follows: (maximum allowable latency - predicted network latency) ÷ maximum allowable latency + (predicted network latency jitter value ÷ baseline network latency jitter value tolerance).

[0067] It can calculate the historical average CPU utilization and the baseline execution time (i.e., the average execution time of historical computing tasks of edge nodes) based on the historical status information of edge nodes, and obtain the maximum allowable latency and the baseline network latency jitter tolerance from the configuration information of edge nodes.

[0068] The weight coefficients ω1 and ω2 can be dynamically adjusted based on resource demand information: when resource demand information indicates that CPU resources will be scarce in the future, ω1 can be increased to make the classifier more inclined to judge computationally intensive tasks; when resource demand information indicates that network latency or network latency jitter value is increasing, ω2 can be increased to make the classifier more inclined to judge latency-sensitive tasks.

[0069] Subsequently, preset thresholds T1 and T2 (T2>T1) can be set: If the task type score is ≥ T2, the prediction calculation task will be classified as a computationally intensive task. If the task type score is ≤ T1, the prediction calculation task will be classified as a latency-sensitive task. If the task type score is between T1 and T2, it can be combined with the task history execution records of the edge nodes for secondary determination by the random forest classifier.

[0070] The classifier described above can achieve adaptive and accurate identification of the task type of the prediction computing task of the edge node in the future time period by combining the predicted load value and the real-time task characteristics, which provides a reliable basis for subsequent differentiated scheduling.

[0071] Step S104: Based on the task type of the predicted computing task and the status information of the edge node, determine the task scheduling strategy for the edge node.

[0072] In this embodiment, after determining the task type of the prediction computing task, a differentiated task scheduling strategy can be formulated. For example, if the task type of the prediction computing task of the edge node is a computationally intensive task, the computing task of the edge node will be preferentially scheduled to other edge nodes with more abundant CPU resources; if the task type of the prediction computing task of the edge node is a latency-sensitive task, the computing task of the edge node will be preferentially scheduled to other edge nodes with better communication conditions.

[0073] For detailed steps, please refer to [link / reference]. Figure 3 , Figure 3 This is a flowchart illustrating a task scheduling strategy for determining edge nodes according to an embodiment of this application, such as... Figure 3 As shown, step S104 above may include the following steps: Step S301: Determine the performance score of each edge node based on the state information of each edge node.

[0074] In this embodiment, step S301 includes: for each edge node, calculating the CPU utilization of the edge node based on the CPU state of the edge node; calculating the memory occupancy of the edge node based on the memory state of the edge node; calculating the communication latency of the edge node based on the network state of the edge node; and calculating the performance score of the edge node based on the CPU utilization, memory occupancy, and communication latency. The performance score is inversely proportional to the CPU utilization, inversely proportional to the memory occupancy, and directly proportional to the reciprocal of the communication latency.

[0075] Specifically, the CPU utilization of an edge node can be calculated based on the number of available CPU cores, clock speed, and cache size in the CPU status of the edge node; the memory utilization of an edge node can be calculated based on the remaining memory and available storage disk space in the memory status of the edge node; and the communication latency of an edge node can be calculated based on the communication round-trip latency, network latency jitter, and packet loss rate in the network status of the edge node.

[0076] The performance score of an edge node can be expressed as: ; in, To score the real-time performance of edge nodes. This indicates the current CPU utilization of the edge node. This indicates the current memory usage of the edge node. Indicates communication delay. , , These are the first, second, and third weights, which can be dynamically changed, and satisfy the following conditions: + + =1.

[0077] The aforementioned first, second, and third weights can be adjusted based on the historical state information of the edge nodes. Specifically, they can be dynamically adjusted based on the historical variance of each resource indicator (CPU utilization, memory usage, or communication latency), that is, adjusted according to the magnitude of the variance of each resource indicator over the most recent few periods of the edge node. For example, resource indicators with larger variances can be assigned higher weights in the performance score calculation formula to reflect the degree of impact of the dynamic changes of that resource indicator on the performance score of the edge node.

[0078] The first, second, and third weights can be adjusted by minimizing the mean square error between the performance score of the edge node and the actual task completion efficiency. The optimal weight allocation can be solved by using the gradient descent algorithm, and the weights can be updated periodically to improve the accuracy of the performance score of the edge node.

[0079] Step S302: For each edge node, determine whether the prediction computation task of the edge node is latency-sensitive or computationally intensive; if the prediction computation task of the edge node is latency-sensitive, proceed to step S303; if the prediction computation task of the edge node is computationally intensive, proceed to step S304.

[0080] Step S303, determining the task scheduling strategy includes: when the performance score of an edge node is less than a first preset threshold, selecting an edge node whose network state meets the first preset condition and whose performance score is greater than a second preset threshold from multiple edge nodes as the target edge node, and scheduling the computing tasks of the edge node to the target edge node.

[0081] In this embodiment, the first preset condition may refer to a communication latency lower than a first specified value. For example, when the performance score of edge node A is less than a first preset threshold, an edge node with a communication latency lower than the first specified value and a performance score greater than a second preset threshold can be selected from all edge nodes in the edge computing system as the target edge node B. When there are multiple candidate edge nodes, the candidate edge node with the highest performance score can be determined as the target edge node B.

[0082] Subsequently, the computing tasks of edge node A can be scheduled to target edge node B. This task scheduling strategy may include scheduling the current computing tasks of edge node A to target edge node B, and may also include scheduling computing tasks received by edge node A in the future to target edge node B, so as to make full use of the computing resources of the edge node cluster.

[0083] Step S304, determining the task scheduling strategy includes: when the performance score of an edge node is less than a first preset threshold, selecting an edge node whose CPU state meets a second preset condition and whose performance score is greater than a second preset threshold from multiple edge nodes as the target edge node, and scheduling the computing tasks of the edge node to the target edge node.

[0084] In this embodiment, the second preset condition may refer to the number of available CPU cores being higher than a second specified value. For example, when the performance score of edge node A is less than a first preset threshold, an edge node with a number of available CPU cores higher than the second specified value and a performance score greater than the second preset threshold can be selected from all edge nodes in the edge computing system as the target edge node B. When there are multiple candidate edge nodes, the candidate edge node with the highest performance score can be determined as the target edge node B.

[0085] Similar to step S303 above, the task scheduling strategy may include scheduling the current computing tasks of edge node A to the target edge node B, and may also include scheduling computing tasks received by edge node A in the future time period to the target edge node B, so as to make full use of the computing resources of the edge node cluster.

[0086] It should be noted that the first specified value, the second specified value, the first preset threshold and the second preset threshold mentioned above can all be set according to the actual application environment and user expectations, and this application does not impose any restrictions on them.

[0087] Step S105: Schedule the computing tasks of the edge nodes according to the task scheduling strategy.

[0088] In this embodiment, if the task scheduling strategy requires scheduling the current computing tasks of the edge nodes and the computing tasks of the edge nodes in the future time period, the computing tasks of the edge nodes can be scheduled using hot migration or cold migration methods, and the full or incremental synchronization method can be selected according to the network status of the edge nodes.

[0089] Specifically, the task scheduling process can adopt a three-stage hybrid transmission protocol. The first stage performs a full copy of the container memory image and transmits data at high speed through remote direct data access technology. The second stage enters incremental synchronization, transmitting only the dirty pages that have been modified since the previous synchronization point and maintaining data consistency through a copy-on-write mechanism. The third stage is a post-copy mode, which automatically switches to this mode when the dirty page generation rate exceeds a certain proportion of the currently available network bandwidth.

[0090] In one embodiment, step S105 may include: obtaining the network packet loss rate and network latency jitter value of the edge computing system; when the network packet loss rate is less than a third preset threshold and the network latency jitter value is less than a fourth preset threshold, scheduling the computing tasks of the edge nodes according to a task scheduling strategy at a preset first task distribution rate; when the network packet loss rate is greater than or equal to the third preset threshold, or the network latency jitter value is greater than or equal to the fourth preset threshold, calculating a second task distribution rate based on the first task distribution rate, the network packet loss rate, and the network latency jitter value, and scheduling the computing tasks of the edge nodes according to the task scheduling strategy at the second task distribution rate, wherein the second task distribution rate is less than the first task distribution rate.

[0091] In this embodiment, the network packet loss rate and network latency jitter of the edge computing system can be monitored in real time to determine whether network congestion has occurred. Specifically, a third preset threshold and a fourth preset threshold can be set according to the actual application environment and user expectations. If the network packet loss rate is greater than or equal to the third preset threshold, or the network latency jitter is greater than or equal to the fourth preset threshold, it is determined that network congestion has occurred in the edge computing system. At this time, the task distribution rate can be reduced to prevent the network congestion of the edge computing system from worsening.

[0092] As an example, the second task distribution rate can be expressed as: ; in, For the second task distribution rate, This is the first task distribution rate (also known as the baseline task distribution rate, which is the task distribution rate when the edge computing system does not experience network congestion). For the real-time network packet loss rate of the edge computing system, This represents the real-time network latency jitter value (in milliseconds) for the edge computing system. This is the preset packet loss rate adjustment coefficient. It is a natural constant.

[0093] In this way, by monitoring the network layer of the edge computing system in real time, when the real-time network packet loss rate or real-time network latency jitter exceeds a preset threshold, a degradation command can be sent to reduce the task distribution rate of task scheduling. This can alleviate network congestion in the edge computing system and improve its reliability. In addition, the application layer of the edge computing system can also be monitored in real time. When a high-priority urgent computing task is detected, the computing resources of the target edge node can be forcibly reserved, and the migration operation of non-critical containers can be suspended. This prioritizes the scheduling and processing of high-priority urgent computing tasks, ensuring the service quality and continuity of high-priority urgent computing tasks under abnormal conditions. This can further improve the task processing efficiency of the edge computing system.

[0094] In one embodiment, the node load prediction model and task scheduling strategy can be continuously optimized through a reinforcement learning framework. The absolute error between the predicted load value and the actual load value can be calculated in real time through the prediction deviation monitoring unit. When the average absolute error of multiple consecutive prediction cycles exceeds a set value, an online retraining process is triggered. In addition, the strategy optimization unit can construct a model with node state and task type as the state space and scheduling action as the action space based on the reinforcement learning algorithm. The reward function is designed in a two-dimensional form, taking into account task completion latency and energy consumption level. After the strategy is updated, it is constrained to a specific numerical range through a normalization function.

[0095] In this way, by establishing a three-dimensional dynamic resource profile that integrates CPU, memory, and network states, and introducing real-time performance scoring and task type judgment mechanisms for edge nodes, a panoramic awareness of node status can be provided for task scheduling strategies. This enables edge computing systems to perform precise resource matching based on multi-dimensional real-time indicators and differentiated task type requirements, avoiding the suboptimal allocation problem caused by traditional task scheduling methods that rely on a single hardware indicator for scheduling.

[0096] Through the above steps S101 to S105, the status information of multiple edge nodes in the edge computing system can be obtained. Then, the status information of each edge node is input into the node load prediction model to obtain the predicted load value of the edge node in the future time period output by the node load prediction model. Subsequently, based on the predicted load value, the task type of the predicted computing task of the edge node in the future time period can be determined. Based on the task type of the predicted computing task and the status information of the edge node, the task scheduling strategy of the edge node is determined. Finally, the computing tasks of the edge node are scheduled according to the task scheduling strategy. In this way, the load of edge nodes can be predicted in the future using a node load prediction model. By determining the task scheduling strategy for edge nodes based on the predicted load value, precise management of edge node computing resources can be achieved. For example, it avoids scheduling too many computing tasks on edge nodes with high predicted load values, or too few computing tasks on edge nodes with low predicted load values, thus fully utilizing the computing resources of edge nodes and improving the overall task processing efficiency of the edge computing system. Furthermore, by determining the task type of the predicted computing tasks for edge nodes in the future time period based on the predicted load value, and determining the task scheduling strategy based on this task type, precise matching of computing resources based on multi-dimensional real-time indicators (edge ​​node status information) and differentiated task type requirements is achieved, further improving the overall task processing efficiency of the edge computing system. This application solves the problem in related technologies where task scheduling methods that only focus on the hardware indicators of edge nodes reduce the task processing efficiency of edge computing systems, achieving the technical effect of improving the overall task processing efficiency of edge computing systems.

[0097] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0098] Corresponding to the task scheduling method for edge nodes described in the above embodiments, Figure 4 A schematic diagram of a task scheduling device for an edge node according to an embodiment of this application is shown. For ease of explanation, only the parts related to the embodiment of this application are shown.

[0099] Please see Figure 4The task scheduling device 4 for edge nodes includes: an acquisition module 40, used to acquire the status information of multiple edge nodes in the edge computing system; a prediction module 41, used to input the status information of each edge node into a node load prediction model to obtain the predicted load value of the edge node in a future time period output by the node load prediction model; a task type determination module 42, used to determine the task type of the predicted computing task of the edge node in the future time period based on the predicted load value; a scheduling strategy determination module 43, used to determine the task scheduling strategy of the edge node based on the task type of the predicted computing task and the status information of the edge node; and a scheduling module 44, used to schedule the computing tasks of the edge node according to the task scheduling strategy.

[0100] In one embodiment, the node load prediction model includes a Long Short-Term Memory (LSTM) network module, a Temporal Convolutional Network (TCN) module, and a fusion module. The prediction module 41 is further configured to acquire historical state information of each edge node; input the state information and historical state information of the edge node, as well as the state information of other edge nodes adjacent to the edge node, into the node load prediction model; generate a temporal feature vector based on the state information and historical state information of the edge node by the LSTM module in the node load prediction model; generate a spatial feature vector based on the state information of the edge node and the state information of other edge nodes by the TCN module in the node load prediction model; and fuse the temporal feature vector and the spatial feature vector in the node load prediction model by the fusion module to generate a fused feature vector, and generate a predicted load value based on the fused feature vector.

[0101] In one embodiment, the prediction module 41 is further configured to: calculate the correlation between the time feature vector and the historical state information by the fusion module; calculate the mutual information value between the spatial feature vector and the state information of other edge nodes by the fusion module; calculate the fusion weight based on the correlation and mutual information value by the fusion module; and fuse the time feature vector and the spatial feature vector according to the fusion weight to generate a fused feature vector.

[0102] In one embodiment, the task type determination module 42 is further configured to convert the predicted load value into resource requirement information, including CPU requirement information, memory requirement information and network requirement information; and based on the resource requirement information, determine the task type of the predicted computing task, including latency-sensitive or computationally intensive tasks.

[0103] In one embodiment, the scheduling strategy determination module 43 is further configured to determine the performance score of each edge node based on the state information of each edge node; for each edge node, when the task type of the prediction computing task of the edge node is latency-sensitive, the task scheduling strategy is determined as follows: when the performance score of the edge node is less than a first preset threshold, an edge node whose network state meets the first preset condition and whose performance score is greater than a second preset threshold is selected from multiple edge nodes as the target edge node, and the computing task of the edge node is scheduled to the target edge node; when the task type of the prediction computing task of the edge node is computationally intensive, the task scheduling strategy is determined as follows: when the performance score of the edge node is less than the first preset threshold, an edge node whose CPU state meets the second preset condition and whose performance score is greater than the second preset threshold is selected from multiple edge nodes as the target edge node, and the computing task of the edge node is scheduled to the target edge node.

[0104] In one embodiment, the status information includes CPU status, memory status, and network status; the scheduling policy determination module 43 is further configured to, for each edge node, calculate the CPU utilization of the edge node based on the CPU status of the edge node; calculate the memory occupancy of the edge node based on the memory status of the edge node; calculate the communication latency of the edge node based on the network status of the edge node; and calculate the performance score of the edge node based on the CPU utilization, memory occupancy, and communication latency of the edge node, wherein the performance score is inversely proportional to the CPU utilization, inversely proportional to the memory occupancy, and directly proportional to the reciprocal of the communication latency.

[0105] In one embodiment, the scheduling module 44 is further configured to obtain the network packet loss rate and network latency jitter value of the edge computing system; when the network packet loss rate is less than a third preset threshold and the network latency jitter value is less than a fourth preset threshold, the computing tasks of the edge nodes are scheduled according to the task scheduling strategy at a preset first task distribution rate; when the network packet loss rate is greater than or equal to the third preset threshold, or the network latency jitter value is greater than or equal to the fourth preset threshold, a second task distribution rate is calculated based on the first task distribution rate, the network packet loss rate, and the network latency jitter value, and the computing tasks of the edge nodes are scheduled according to the task scheduling strategy at the second task distribution rate, wherein the second task distribution rate is less than the first task distribution rate.

[0106] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0107] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments 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. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0108] Figure 5 This is a schematic diagram of the structure of a terminal device according to an embodiment of this application. Figure 5 As shown, the terminal device 5 includes: at least one processor 50 ( Figure 5 (Only one is shown in the image) a processor, a memory 51, and a computer program 52 stored in the memory 51 and capable of running on at least one processor 50. When the processor 50 executes the computer program 52, it implements the steps in the above-described task scheduling method embodiments for any of the edge nodes.

[0109] Terminal device 5 can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. Terminal device 5 may include, but is not limited to, processor 50 and memory 51. Those skilled in the art will understand that... Figure 5 This is merely an example of terminal device 5 and does not constitute a limitation on terminal device 5. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0110] The processor 50 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0111] In some embodiments, memory 51 may be an internal storage unit of terminal device 5, such as a hard disk or memory of terminal device 5. In other embodiments, memory 51 may be an external disk of terminal device 5, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on terminal device 5. In other embodiments, memory 51 may include both internal storage units and external disks of terminal device 5. Memory 51 is used to store operating system, applications, bootloader, data, and other programs, such as the program code of computer program 52. Memory 51 may also be used to temporarily store data that has been output or will be output.

[0112] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the above-described task scheduling method embodiments for each edge node.

[0113] This application provides a computer program product that, when run on a mobile terminal, enables the mobile terminal to implement the steps in the above-described task scheduling method embodiments for each edge node.

[0114] This application implements all or part of the processes in the methods of the above embodiments, which can be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate form. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a task scheduling device or terminal device at an edge node, a recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, such as a USB flash drive, portable hard drive, magnetic disk, or optical disk.

[0115] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0116] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0117] In the embodiments provided in this application, it should be understood that the disclosed apparatus / network devices and methods can be implemented in other ways. For example, the apparatus / network device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0118] 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 network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0119] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A task scheduling method for edge nodes, characterized in that, include: Obtain the status information of multiple edge nodes in the edge computing system; The state information of each of the plurality of edge nodes is input into the node load prediction model to obtain the predicted load value of the edge node in the future time period output by the node load prediction model. Based on the predicted load value, determine the task type of the predicted computing task of the edge node in the future time period; Based on the task type of the prediction calculation task and the status information of the edge node, the task scheduling strategy of the edge node is determined. The computing tasks of the edge nodes are scheduled according to the task scheduling strategy.

2. The method according to claim 1, characterized in that, The node load prediction model includes a Long Short-Term Memory (LSTM) network module, a Temporal Convolutional Network (TCN) module, and a fusion module. The state information of each of the multiple edge nodes is input into the node load prediction model to obtain the predicted load value of the edge node in a future time period, output by the node load prediction model. For each edge node, obtain the historical state information of the edge node; The state information and historical state information of the edge node, as well as the state information of other edge nodes adjacent to the edge node, are input into the node load prediction model. The LSTM module in the node load prediction model generates a time feature vector based on the state information and historical state information of the edge nodes; The TCN module in the node load prediction model generates a spatial feature vector based on the state information of the edge node and the state information of other edge nodes; The fusion module in the node load prediction model fuses the time feature vector and the spatial feature vector to generate a fused feature vector, and the predicted load value is generated based on the fused feature vector.

3. The method according to claim 2, characterized in that, The fusion module in the node load prediction model fuses the temporal feature vector and the spatial feature vector to generate a fused feature vector, including: The fusion module calculates the correlation between the time feature vector and the historical state information; The fusion module calculates the mutual information value between the spatial feature vector and the state information of the other edge nodes; The fusion module calculates the fusion weight based on the relevance and the mutual information value; The fusion module fuses the temporal feature vector and the spatial feature vector according to the fusion weight to generate the fused feature vector.

4. The method according to any one of claims 1 to 3, characterized in that, Based on the predicted load value, the task type of the prediction calculation task for the edge node in the future time period includes: The predicted load value is converted into resource demand information, which includes CPU demand information, memory demand information, and network demand information. Based on the resource requirement information, the task type of the prediction calculation task is determined, and the task type includes latency-sensitive or computationally intensive tasks.

5. The method according to claim 4, characterized in that, Based on the task type of the prediction calculation task and the state information of the edge node, determining the task scheduling strategy for the edge node includes: Based on the state information of each edge node, a performance score for each edge node is determined; For each edge node, when the task type of the prediction calculation task of the edge node is the latency-sensitive type, the task scheduling strategy is determined as follows: when the performance score of the edge node is less than a first preset threshold, select an edge node from the plurality of edge nodes whose network state satisfies the first preset condition and whose performance score is greater than a second preset threshold as a target edge node, and schedule the calculation task of the edge node to the target edge node. When the task type of the prediction computation task at the edge node is computationally intensive, determining the task scheduling strategy includes: when the performance score of the edge node is less than the first preset threshold, selecting from the plurality of edge nodes the edge node whose CPU state meets the second preset condition and whose performance score is greater than the second preset threshold as the target edge node, and scheduling the computation task of the edge node to the target edge node.

6. The method according to claim 5, characterized in that, The status information includes CPU status, memory status, and network status; Based on the state information of each edge node, the performance score of each edge node is determined as follows: For each edge node, the CPU utilization of the edge node is calculated based on the CPU state of the edge node; Calculate the memory occupancy rate of the edge node based on its memory status; Calculate the communication latency of the edge node based on the network state of the edge node; Based on the CPU utilization, memory usage, and communication latency of the edge node, a performance score is calculated for the edge node. The performance score is inversely proportional to the CPU utilization, inversely proportional to the memory usage, and directly proportional to the reciprocal of the communication latency.

7. The method according to any one of claims 1 to 3, characterized in that, Scheduling the computing tasks of the edge nodes according to the task scheduling strategy includes: Obtain the network packet loss rate and network latency jitter value of the edge computing system; When the network packet loss rate is less than a third preset threshold and the network latency jitter value is less than a fourth preset threshold, the computing tasks of the edge nodes are scheduled according to the task scheduling strategy at a preset first task distribution rate. If the network packet loss rate is greater than or equal to the third preset threshold, or the network latency jitter value is greater than or equal to the fourth preset threshold, a second task distribution rate is calculated based on the first task distribution rate, the network packet loss rate, and the network latency jitter value. The computing tasks of the edge nodes are then scheduled according to the task scheduling strategy at the second task distribution rate, where the second task distribution rate is less than the first task distribution rate.

8. A task scheduling device for edge nodes, characterized in that, include: The acquisition module is used to acquire the status information of multiple edge nodes in the edge computing system; The prediction module is used to input the state information of each of the plurality of edge nodes into the node load prediction model to obtain the predicted load value of the edge node in a future time period output by the node load prediction model. The task type determination module is used to determine the task type of the prediction calculation task of the edge node in the future time period based on the predicted load value. The scheduling strategy determination module is used to determine the task scheduling strategy of the edge node based on the task type of the prediction calculation task and the status information of the edge node. The scheduling module is used to schedule the computing tasks of the edge nodes according to the task scheduling strategy.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the task scheduling method for edge nodes as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, Includes a computer program, which, when run, causes the task scheduling method for the edge node as described in any one of claims 1 to 7 to be executed.

Images