Multi-dimensional dynamic perception and intelligent computing power scheduling method and device for computing power network

Abstract

The application discloses a multi-dimensional dynamic perception and intelligent computing power scheduling method and device for a computing power network. The method comprises the following steps: deploying a plurality of target components constituting multi-dimensional dynamic perception at terminals, edge nodes and regional computing power centers, and collecting computing power indexes and network indexes of each node; analyzing indexes in a preset period through a time sequence prediction model to obtain computing power bottleneck risk and time delay attenuation information; dynamically obtaining computing power dimension weight and time delay dimension weight according to a type of a task to be executed, and combining the computing power bottleneck risk and the time delay attenuation information to construct a cost function; determining a node corresponding to a minimum cost function as a target node; obtaining path costs corresponding to all feasible migration paths according to the target node, determining a migration path corresponding to a minimum path cost as a target migration path, and triggering intelligent scheduling of computing power according to the target migration path. The method can realize collaborative perception, intelligent decision and closed-loop control of computing power and network resources.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a method and apparatus for multi-dimensional dynamic perception and intelligent computing power scheduling for computing power networks. Background Technology

[0002] With the rapid development of 5G / 6G, cloud-edge-device collaborative computing, and artificial intelligence, computing resources are exhibiting trends of large-scale, heterogeneous, and distributed nature. To achieve on-demand acquisition and global scheduling, the industry has proposed the concept of a computing power network. This aims to unify and route dispersed computing, storage, and accelerator resources through a network, enabling computing power to be discovered, selected, and dynamically allocated like network bandwidth, thus supporting the low-latency, high-reliability, and high-concurrency intelligent business requirements. In typical scenarios such as intelligent manufacturing and the Internet of Vehicles, tasks have strict constraints on end-to-end latency, jitter, and stable computing power supply. Traditional single-cloud-center processing models are insufficient to meet these demands, requiring tasks to be offloaded and migrated between terminals, edge nodes, and the cloud.

[0003] Existing task offloading and computing power scheduling technologies mainly fall into three categories: local decision-making based on static rules, centralized global scheduling, and service chain orchestration driven by programmable networks. However, these approaches suffer from several problems: In highly dynamic environments, computing resources and network status are coupled across dimensions, and existing technologies collect or evaluate these two types of indicators separately, which can easily lead to misjudgments of overall service availability; for diverse task characteristics, existing scheduling and offloading algorithms use fixed or semi-fixed weight configurations, making it difficult to adapt to differences in task types and sudden environmental changes; execution control is mostly open-loop, lacking the ability to intervene midway through runtime. Furthermore, existing solutions combining network programmability and path redirection technologies have limited support for computing power status prediction, synchronous awareness of network status, and joint decision-making and coordinated execution of the two.

[0004] The current computing power network environment has obvious technical problems in task offloading and computing power scheduling, including perception fragmentation and insufficient prediction, lack of joint modeling and forward-looking prediction of computing power load evolution and network quality fluctuations; insufficient decision-making adaptability, making it difficult to dynamically adjust and optimize weights and constraints according to different task types and environmental changes; and insufficient execution closed-loop capability, lacking fast backtracking, cross-domain switching and path redirection mechanisms based on real-time status during runtime. These defects seriously restrict the application effect of computing power networks in low-latency and high-reliability services. Summary of the Invention

[0005] In view of this, the present disclosure provides a multi-dimensional dynamic perception and intelligent computing power scheduling method and apparatus for computing power networks, which can solve many shortcomings in existing computing power network task offloading and scheduling technologies, including the disconnect between computing power and network perception, insufficient computing power load prediction, rigid decision-making models, lack of closed-loop control in the execution process, and insufficient linkage between programmable network and computing power status, so as to realize the collaborative perception, intelligent decision-making and closed-loop control of computing power and network resources.

[0006] In a first aspect, embodiments of this disclosure provide a multi-dimensional dynamic perception and intelligent computing power scheduling method for computing power networks, including:

[0007] Several target components constituting multi-dimensional dynamic perception are deployed at terminals, edge nodes, and regional computing centers, and computing power indicators and network indicators of each node are collected.

[0008] By analyzing the computing power indicators and network indicators of all nodes within a preset period using a time-series prediction model, information on computing power bottleneck risks and latency decay can be obtained.

[0009] Based on the type of task to be executed, dynamically obtain the weights of computing power and latency.

[0010] A cost function is constructed based on the computing power dimension weight, the latency dimension weight, the computing power bottleneck risk, and the latency attenuation information;

[0011] The node corresponding to the minimum cost function is determined and designated as the target node;

[0012] Based on the target node, obtain the path costs corresponding to all feasible migration paths, determine the migration path corresponding to the minimum path cost, and use it as the target migration path.

[0013] Intelligent scheduling of computing power is triggered based on the target node and the target migration path.

[0014] Secondly, this disclosure also provides a computer device, which adopts the following technical solution:

[0015] The computer device includes:

[0016] At least one processor; and,

[0017] A memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to execute any of the above-described multi-dimensional dynamic perception and intelligent computing power scheduling methods for computing power networks.

[0019] Thirdly, embodiments of this disclosure also provide a computer-readable storage medium storing computer instructions for causing a computer to execute any of the above-described multi-dimensional dynamic perception and intelligent computing power scheduling methods for computing power networks.

[0020] Fourthly, embodiments of this disclosure also provide a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of any of the methods described above.

[0021] This application discloses a multi-dimensional dynamic perception and intelligent computing power scheduling method for computing power networks. It deploys target components at terminals, edge nodes, and regional computing power centers to collect computing power and network indicators from each node. This changes the existing technology's approach of collecting or evaluating computing power resources and network status indicators in a fragmented or isolated manner, coupling them across dimensions. Through unified collection, it can comprehensively and accurately obtain real-time status information of each node in the network, providing a richer and more reliable data foundation for subsequent decision-making. By using a time-series prediction model to analyze computing power and network indicators within a preset period, it obtains information on computing power bottleneck risks and latency degradation. This achieves joint modeling and forward-looking prediction of computing power load evolution and network quality fluctuations, overcoming the shortcomings of existing technologies in prediction. By anticipating potential problems, the system can prepare in advance, avoiding service interruptions or performance degradation due to sudden computing power bottlenecks or network quality declines. Furthermore, it dynamically obtains computing power and latency dimension weights based on the type of task to be executed. Unlike existing scheduling and offloading algorithms that use fixed or semi-fixed weight configurations, this method can flexibly adjust weights according to the characteristics and requirements of the task, making decision-making more efficient. The system is designed to be more realistic. It constructs a cost function based on the weights of computing power and latency, computing power bottleneck risks, and latency decay information, and determines the node corresponding to the minimum cost function as the target node. This cost function-based decision-making method comprehensively considers multiple factors and can dynamically adjust and optimize weights and constraints under different task types and environmental changes, making the decision more scientific and reasonable, and improving its adaptability and accuracy. By calculating the path costs corresponding to all feasible migration paths, the migration path corresponding to the minimum path cost is determined as the target migration path. This provides the optimal path selection for task migration, enabling the selection of the best migration method based on real-time status during runtime, achieving rapid backtracking, cross-domain switching, and path redirection based on real-time status. Intelligent scheduling of computing power is triggered based on the target node and target migration path, forming a closed-loop control link of "perception—prediction—decision—execution." This closed-loop control mechanism allows the system to monitor state changes in real time during operation and adjust the scheduling strategy accordingly, avoiding the problems of existing execution control methods being mostly open-loop execution and lacking mid-process intervention capabilities, thus improving the system's reliability and stability.

[0022] The above description is merely an overview of the technical solution disclosed herein. In order to better understand the technical means of this disclosure and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating the multi-dimensional dynamic perception and intelligent computing power scheduling method for computing power networks provided in this embodiment of the disclosure.

[0025] Figure 2 A flowchart illustrating the method for obtaining computing power bottleneck risk and latency attenuation information provided in this embodiment of the disclosure.

[0026] Figure 3 This is a flowchart illustrating the method for dynamically obtaining computing power dimension weights and latency dimension weights provided in this embodiment of the disclosure.

[0027] Figure 4 A flowchart illustrating the method for constructing a cost function provided in this embodiment of the disclosure.

[0028] Figure 5 This is a flowchart illustrating the method for obtaining the target migration path provided in an embodiment of this disclosure.

[0029] Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure. Detailed Implementation

[0030] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0031] It should be understood that the following specific examples illustrate the implementation of this disclosure, and those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0032] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0033] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The drawings only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0034] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0035] Reference Figure 1 This application discloses a multi-dimensional dynamic perception and intelligent computing power scheduling method for computing power networks, including:

[0036] S100 deploys several lightweight target components that constitute multi-dimensional dynamic perception at terminals, edge nodes, and regional computing centers.

[0037] The computing power and network metrics of each node are collected synchronously based on a lightweight target component.

[0038] The terminal consists of several computing nodes, edge nodes refer to nodes with weaker resources, and regional computing centers consist of several high-performance nodes. Specifically, a regional computing center is a cluster of multiple high-performance servers. Computing power metrics include CPU utilization, GPU utilization, memory pressure, queue depth, etc.; network metrics include latency, packet loss rate, bandwidth, jitter, etc.

[0039] In this embodiment, the lightweight target component is preferably a lightweight probe, used to periodically collect computing power and key network indicators. Specifically, the probe is implemented in both containerized and binary modes: the network probe is based on an improved version of iperf3, with unnecessary logic removed and a multi-threaded mechanism used to collect link performance data such as bandwidth, latency, jitter, and packet loss rate; the resource probe is implemented as a single static binary program in Go, directly calling the Linux cgroups interface and the NVIDIA DCGM library to obtain underlying resource indicators such as CPU, GPU, memory, and I / O. The collected data is uploaded via a gRPC streaming interface and combined with Snappy and Delta incremental compression to reduce transmission load, before being input into a time-series modeling engine to predict bottleneck risks and quality degradation trends in future time windows.

[0040] This step enables unified collection and analysis of network metrics, overcoming the limitations of existing technologies that process the two types of metrics separately.

[0041] S200 uses a time-series prediction model to analyze computing power and network metrics for all nodes within a preset period, thereby obtaining information on computing power bottleneck risks and latency degradation.

[0042] This step not only allows for real-time monitoring of computing power and network operation status, but also enables early identification of potential computing power overload and link degradation. This provides forward-looking input to the decision-making module before task unloading, avoiding scheduling failures caused by relying solely on instantaneous states.

[0043] S300 dynamically obtains the computing power dimension weight and latency dimension weight based on the type of task to be executed;

[0044] A cost function is constructed based on the computing power status, network status, and computing power bottleneck risks.

[0045] This step establishes a task-adaptive decision-making framework that can dynamically adjust and optimize weights and constraints according to different task types and environmental changes, achieving differentiated and refined resource optimization and avoiding the misallocation of computing resources and reduced efficiency caused by fixed-weight models.

[0046] S400: Determine the node corresponding to the minimum cost function and use it as the target node.

[0047] S500: Obtain the path costs corresponding to all feasible migration paths based on the target node, determine the migration path corresponding to the minimum path cost, and use it as the target migration path.

[0048] S600 triggers intelligent scheduling of computing power based on the target node and target migration path.

[0049] Furthermore, this application uses SRv6 technology as the underlying carrier. Through flexible operation of the Segment Identifier (SID), it tightly integrates the computing power status with the network routing process, thereby realizing the construction of dynamic service chains and rapid redirection for computing power networks. In the execution plane, it combines SRv6 and other programmable network technologies to realize path redirection, service chain orchestration and cross-domain computing power migration, thus forming a closed-loop control chain of "perception-prediction-decision-execution".

[0050] This method can perform fast migration, cross-domain switching and path redirection mechanisms based on real-time status during runtime, and realize the collaborative perception, decision-making and execution control of computing power and network resources under complex time-varying conditions, effectively solving the problem of lack of runtime intervention under the existing open-loop mechanism.

[0051] The multi-dimensional dynamic perception and intelligent computing power scheduling method for computing power networks disclosed in this embodiment provides an offloading method that combines multi-dimensional perception, intelligent prediction, adaptive decision-making, and closed-loop control. It can effectively solve the technical problems existing in task offloading and computing power scheduling in the current computing power network environment, and provide strong support for the intelligent business requirements of low latency, high reliability, and high concurrency. In typical scenarios such as intelligent manufacturing and vehicle networking, it can better meet the strict constraints of end-to-end latency, jitter, and stable computing power supply for tasks, improve the application effect of computing power networks in these scenarios, and promote the development of related industries.

[0052] In this embodiment, in terms of computing power, this application focuses on monitoring the status of GPU and memory. Specifically, GPU utilization is calculated by dividing the number of active stream processors by the total number of processors. : ,in, Indicates time The number of processing units in operation. This represents the total number of GPU processing units. This metric accurately reflects the utilization of computing resources.

[0053] Memory pressure is assessed by comprehensively considering cache and swap space usage; memory pressure is... : ,in, Indicates cache usage. Indicates the amount of swap space occupied. Total memory capacity This is a weighting factor used to adjust the impact of the swap area. This metric reveals the level of strain on node memory resources.

[0054] At the network level, the system collects core metrics such as bandwidth utilization, latency, and packet loss rate. Bandwidth utilization is... : ,in, For actual throughput, This is the rated bandwidth of the link, and this metric reflects the link congestion situation.

[0055] Furthermore, S100 is preferably integrated into the multi-dimensional dynamic perception module, S200-S500 are preferably integrated into the task adaptive decision-making module, and S600 is preferably integrated into the programmable control execution module. This invention, by constructing a complete system of multi-dimensional dynamic perception, task adaptive decision-making, and programmable control execution, can effectively improve the shortcomings of the task offloading process in existing computing power network environments. In terms of perception and prediction, this invention can simultaneously acquire multiple key indicators of computing power and network, and model their future trends, thereby avoiding scheduling failures caused by relying solely on instantaneous states. The system possesses forward-looking capabilities for identifying computing power bottlenecks and judging network quality degradation, providing a reliable basis for subsequent decisions. Regarding the decision-making mechanism, this invention can dynamically adjust the weights of indicators such as computing power and latency in the comprehensive cost function according to the differentiated needs of different task types, achieving flexibility and adaptability in resource allocation and avoiding the misallocation of computing power resources caused by traditional fixed-weight models. In terms of execution control, this invention provides closed-loop control capabilities, enabling the system to trigger timely backtracking, migration, or cross-domain switching based on changes in computing power or network status during task execution, ensuring the effectiveness and stability of offloading decisions in dynamic environments.

[0056] Reference Figure 2 The method of S200, which "analyzes the computing power indicators and network indicators of all nodes within a preset period through a time-series prediction model to obtain information on computing power bottleneck risks and latency decay," specifically includes the following:

[0057] S210: Normalize the computing power and network metrics of all nodes within a preset period to obtain a normalized dataset.

[0058] Specifically, the aforementioned multidimensional indicators, after normalization and anomaly correction, are integrated into a node state vector. : ,in, For the corresponding round-trip time, This represents the corresponding packet loss rate.

[0059] At the feature level, node state vectors are constructed through the fusion of multi-dimensional indicators. This unifies the expression of computing power and network characteristics in the same vector space, ensuring that indicators of different dimensions are comparable.

[0060] S220 identifies and corrects outliers in the normalized dataset to obtain the target dataset.

[0061] Specifically, a modified boxplot method (IQR=1.8) is prioritized to identify and smooth out outliers in the normalized dataset, avoiding bias in prediction results caused by extreme samples. This task is accomplished using Python with the NumPy and Pandas libraries, specifically: 1) Importing the NumPy and Pandas libraries into Python. NumPy is primarily used for efficient numerical computation, while Pandas excels at data reading, processing, and analysis, laying the foundation for subsequent operations; 2) Writing a function named `modified_boxplot` that accepts two parameters: the input one-dimensional dataset and the interquartile range (IQR), with a default value of 1.8. The function's internal operations are as follows: 2.1) Calculating quartiles: Using the `percentile` function from the NumPy library, the first quartile (Q1) and third quartile (Q3) of the dataset are calculated. The first quartile represents the 25th percentile position in the dataset, and the third quartile represents the 75th percentile position. 2.2) Calculate the IQR: Subtract the first quartile from the third quartile to obtain the IQR, which reflects the middle 50% of the data distribution. 2.3) Determine the upper and lower limits: Based on the given IQR coefficient (1.8 here), calculate the lower and upper limits respectively. The lower limit is the first quartile minus 1.8 times the IQR, and the upper limit is the third quartile plus 1.8 times the IQR. 2.4) Identify and trim outliers: Use the NumPy `where` function to replace values ​​in the dataset less than the lower limit with the lower limit, and values ​​greater than the upper limit with the upper limit. This completes the identification and trimming of outliers. Finally, return the trimmed dataset. 3) Load normalized data: Use the `read_csv` function from the pandas library to load the normalized dataset. Note that you need to replace the file path with the path to your own normalized dataset file according to your actual situation. 4) Applying the improved boxplot method to process the data: Iterate through each column of the normalized dataset and call the previously defined `modified_boxplot` function for each column. This method identifies and corrects outliers in each column, ensuring that all outliers in the entire dataset are handled. 5) Obtaining the target dataset: After correcting outliers in each column, the resulting dataset is our final target dataset. This dataset has removed extreme samples that might affect the prediction results, making it more suitable for subsequent analysis and prediction tasks.

[0062] S230 inputs the target dataset into an LSTM-based time series prediction model to obtain information on computing power bottleneck risks and latency degradation.

[0063] For S230, the specific steps include: inputting the target dataset into an LSTM-based time series prediction model to obtain... The corresponding computing power risk probability and latency decay information.

[0064] The probability of computing power risk is :

[0065] ;

[0066] .

[0067] in, , For the threshold, The parameter controls the steepness of the function. This represents the total number of processing units in the GPU.

[0068] Delay attenuation information is : ,in, For the predicted round-trip delay, This is the largest delay in history. For prediction Packet loss rate at that time This represents the weight of the packet loss item.

[0069] Furthermore, it can also be determined whether the probability of computing power risk and the latency decay information are greater than the corresponding set thresholds. If so, it is determined that the corresponding node has a potential computing power bottleneck; if not, it is determined that the corresponding node does not have a potential computing power bottleneck. In this application, it is preferable to select from nodes that do not have potential computing power bottlenecks.

[0070] To predict future computing power and network evolution, this embodiment prioritizes an LSTM-based time-series prediction model. It takes historical state sequences as input and outputs state estimates for future timeframes, i.e., information regarding the correlation between computing power bottleneck risk and network quality degradation trends. : .in, Indicates the forecast time window, Indicating the length of historical observations, the prediction results can not only provide future GPU utilization and network latency, but also estimate potential risks to computing power and the network.

[0071] Specifically, to model temporal dependencies, this invention introduces a two-layer LSTM structure and employs an attention mechanism to enhance the weight recognition capability of key features. In the prediction phase, it not only outputs values ​​such as GPU utilization and network latency at future time steps but also provides a risk probability distribution. At the quantization level, this invention defines the probability of computational bottleneck risks. With network attenuation index The former maps GPU utilization to risk probability using the Sigmoid function, while the latter combines predicted latency and packet loss rate to assess link degradation trends. These analytical methods ensure both the numerical accuracy of the prediction results and provide interpretable risk quantification indicators, offering an actionable basis for the decision-making module to generate the optimal offloading strategy.

[0072] Reference Figure 3 The method in S300 that "dynamically obtains the computing power dimension weight and latency dimension weight based on the type of task to be executed" specifically includes:

[0073] A100 determines the preset network latency weight and preset computing power weight based on the type of task to be executed.

[0074] Specifically, the types of tasks to be executed include real-time rendering, AI inference, and data batch processing. For different task types, the preset weights are dynamically adjusted, which can start from the root of the task requirements and provide a reasonable initial basis for subsequent weight allocation, making the scheduling scheme more in line with the actual characteristics of the task. Preferably, the preset weight corresponding to the indicator that the current task type has a high requirement for is increased to the maximum weight.

[0075] A200 obtains the environmental state sensitivity of each node based on the computing power utilization and network utilization of each node.

[0076] In a computing network, the operating environment of each node is dynamic. The computing power and network utilization of a node will affect its ability to process tasks. Considering the sensitivity of the environment status can make the determination of weights more in line with the actual operating conditions of the nodes and avoid making unreasonable scheduling decisions when node resources are scarce.

[0077] The A300 determines the latency dimension score based on preset network latency weights and environmental state sensitivity.

[0078] A400 determines the computing power dimension score based on preset computing power weights and environmental state sensitivity.

[0079] Steps A300 and A400 determine the latency dimension score and computing power dimension score based on preset weights and environmental state sensitivity, respectively. This scoring method, which comprehensively considers preset weights and environmental factors, can fully measure the task requirements in different dimensions and the actual situation of the nodes.

[0080] A500 determines the weights of the latency dimension and the computing power dimension based on the latency dimension score and the computing power dimension score.

[0081] The weight of the latency dimension is : ;

[0082] The weight of the computing power dimension is : ;in, Scoring is done based on latency. Scoring is based on computing power.

[0083] The dynamic weights obtained through this method comprehensively consider task characteristic parameters and environmental sensitivity. This not only avoids the uncertainty of manual parameter tuning but also allows for automatic adjustment in response to task changes or network fluctuations. Compared to traditional fixed or semi-fixed weight configurations, it better adapts to diverse task characteristics and complex, ever-changing environments, improving the flexibility and accuracy of task offloading and computing power scheduling. This more effectively meets the computing power and latency requirements of different tasks, enhancing the overall performance and service quality of the computing network.

[0084] The A200 method for "obtaining the environmental state sensitivity of each node based on its computing power utilization and network utilization" specifically includes:

[0085] A210, obtain the rate of change of computing power utilization for each node. ;

[0086] ,in Let be the sum of the CPU utilization and GPU utilization of the node at time t. This is the preset time interval.

[0087] A220, obtain the rate of change of network utilization for each node. .

[0088] ;

[0089] ;

[0090] in, for The bandwidth corresponding to each moment for The bandwidth corresponding to each moment This represents the total bandwidth of the corresponding link. From Bandwidth utilization from time t to time t.

[0091] A230 obtains the environmental state sensitivity of each node based on the rate of change of computing power utilization and the rate of change of network utilization. .

[0092] .

[0093] In a computing network, the resource usage of nodes is dynamic. A simple utilization rate can only reflect the current resource occupancy level, while the rate of change can reflect the dynamic trend of resource usage. For example, although a node's current computing power utilization rate is not high, if the rate of change is fast, it indicates that its resource usage is changing rapidly and it may soon face a resource shortage. By paying attention to the rate of change, we can perceive changes in the node's resource status in advance and provide more forward-looking information for subsequent decisions.

[0094] The environmental state sensitivity is obtained by multiplying the rate of change of computing power utilization and the rate of change of network utilization. This calculation method comprehensively considers the two key resource factors of computing power and network. In actual computing networks, computing power and network resources are interconnected and mutually influential. The execution of a task requires both computing power support and a good network environment. Considering the changes in computing power or network alone is incomplete. By multiplying the rates of change of the two, a comprehensive index can be obtained to measure the environmental state sensitivity of a node, which more accurately reflects the overall changes in the node's resource state.

[0095] Environmental sensitivity provides an important basis for dynamically determining the weights of computing power and latency dimensions. When a node's environmental sensitivity is high, it indicates that the node's resource status is changing drastically. When scheduling tasks, more careful consideration should be given to the node's resource allocation. The weights of computing power and latency dimensions need to be adjusted appropriately to avoid assigning tasks to nodes with unstable resources. This improves the accuracy and reliability of task offloading and computing power scheduling, and ensures the stable operation of the computing network.

[0096] Reference Figure 4 The method in S300 for "constructing a cost function based on computing power status, network status, and computing power bottleneck risk" includes the following methods for constructing the cost function:

[0097] B100, obtains the predicted network latency for each node;

[0098] B200 determines the cost function for the corresponding node based on network latency, computing power bottleneck risk, computing power dimension weight, and latency dimension weight.

[0099] No. The cost function for each node is: : .

[0100] in, As the weight of the latency dimension, Weights are based on computing power. For the first Nodes The corresponding computing power bottleneck risk.

[0101] In step B100, the predicted network latency for each node is obtained. Network latency is a key indicator of network performance, especially for tasks with high real-time requirements (such as real-time rendering and online games), where low network latency is crucial. By obtaining the predicted network latency, we can understand the network transmission delay of each node in the future, rather than relying solely on current latency data. This helps to prioritize nodes with lower network latency during task scheduling, thereby ensuring the real-time performance and smoothness of the task.

[0102] Step B200 determines the cost function for the corresponding node based on network latency, computing power bottleneck risk, computing power dimension weight, and latency dimension weight. This comprehensive approach, which considers multiple factors, can comprehensively and objectively evaluate the overall status of each node. Computing power bottleneck risk reflects the problems that a node may face in terms of computing power resources, while network latency reflects the performance of network transmission. Combining the two can avoid focusing only on network performance while ignoring computing power bottlenecks, or focusing only on computing power while ignoring network latency, ensuring that the selected node can meet the task requirements in terms of both computing power and network.

[0103] The weights for computing power and latency are dynamically determined based on the type of task to be executed, allowing the cost function to be adjusted according to the characteristics of different tasks. This makes the cost function more closely match the actual needs of the task, improving the targeting and effectiveness of task scheduling.

[0104] The cost function provides a quantitative indicator for task scheduling. By comparing the cost function values ​​of each node, it is easy to determine the most suitable node for executing the task and achieve optimal resource allocation. In complex computing network environments, this cost function-based scheduling method can improve resource utilization, reduce task execution costs, and enhance the overall network operating efficiency and service quality.

[0105] Reference Figure 5 The S500 method of "obtaining the path costs corresponding to all feasible migration paths based on the target node, determining the migration path corresponding to the minimum path cost, and using it as the target migration path," specifically includes:

[0106] S510: Obtain the network latency and processing overhead for each feasible migration path and the path resolution operation.

[0107] Network latency reflects the time required for data transmission during migration, while processing overhead reflects the computing resources consumed in performing operations such as path resolution. Different migration paths may exhibit significant differences in these two aspects. For example, some paths may have low network latency but complex path resolution operations and high processing overhead; while other paths may have fast network transmission but time-consuming resolution processing. By comprehensively considering these two factors, the overall cost of each migration path can be fully measured, avoiding the selection of an unoptimal path due to focusing on only a single factor.

[0108] S520 obtains the path cost corresponding to the feasible migration path based on network latency and processing overhead.

[0109] No. The path cost corresponding to each feasible migration path is :

[0110] ; For network latency of the migration path, The processing overhead of SID header parsing (default is 80-120 microseconds). Weighting network latency overhead To handle overhead weighting, this cost model enables the system to consider both network transmission latency and SID processing latency when selecting a path, avoiding the problem of traditional solutions that only focus on link latency while ignoring programmable processing costs.

[0111] This step obtains the path cost of the corresponding feasible migration path based on network latency and processing overhead. This quantitative method integrates the two different dimensions of network latency and processing overhead into a unified path cost, making different paths comparable and providing a clear basis for subsequent path selection.

[0112] S530 Sort the path costs of all feasible migration paths in ascending order, determine the migration path with the lowest path cost, and use it as the target migration path.

[0113] This step sorts the path costs of all feasible migration paths in ascending order and selects the migration path with the lowest path cost as the target migration path. This sorting and selection mechanism quickly and accurately finds the lowest-cost migration path among numerous feasible paths. This helps improve migration efficiency, reduce resource consumption and time costs during the migration process, and enhance the overall performance of the system.

[0114] Choosing the migration path with the lowest path cost means completing the task with minimal resource consumption during the migration process. Reducing network latency can lower the time cost of data transmission, and reducing processing overhead can save computing resources, thereby improving the resource utilization efficiency of the entire system. An optimized migration path can reduce the probability of problems during the migration process, such as data loss and transmission errors. A stable migration process helps maintain the normal operation of the system, avoids system failures or service interruptions caused by migration problems, and improves the reliability and stability of the system.

[0115] Furthermore, when a significant deterioration in the state of the selected target node is detected, the system can immediately reassess the cost function and trigger a backhaul or cross-domain switch, thereby achieving true closed-loop control. Specifically, when the acquired node executing the task in the future... Probability of computing power risk at time t. When the value exceeds the preset threshold, an immediate reassessment and dynamic adjustment are triggered, so that the unloading decision no longer depends on a single static judgment, but can be continuously optimized during runtime to ensure the stable operation of the task in a dynamic environment.

[0116] Furthermore, to ensure configuration stability, this application also includes a consistency verification mechanism. This mechanism periodically compares the SID configuration issued by the controller with the actual configuration in the forwarding device, and uses differential calculation to detect offsets. When the difference exceeds a threshold, a repair command is immediately issued to ensure strong consistency between configuration and execution. This application can not only adjust task paths within milliseconds but also achieve dynamic migration of computing resources in cross-domain environments, thus providing reliable execution guarantees for intelligent offloading in computing network environments.

[0117] A computer device according to embodiments of the present disclosure includes a memory and a processor. The memory is used to store non-transitory computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0118] The processor may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the computer device to perform desired functions. In one embodiment of this disclosure, the processor is used to run computer-readable instructions stored in the memory, causing the computer device to perform all or part of the steps of the multi-dimensional dynamic perception and intelligent computing power scheduling method for computing power networks described in the foregoing embodiments of this disclosure.

[0119] Those skilled in the art will understand that, in order to solve the technical problem of how to achieve a good user experience, this embodiment may also include well-known structures such as communication buses and interfaces, and these well-known structures should also be included within the protection scope of this disclosure.

[0120] like Figure 6 This is a schematic diagram of a computer device provided for an embodiment of the present disclosure. It illustrates a structural schematic diagram suitable for implementing the computer device in the embodiments of the present disclosure. Figure 6 The computer device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0121] like Figure 6 As shown, a computer device may include a processor (such as a central processing unit, graphics processing unit, etc.), which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) or programs loaded from storage devices into random access memory (RAM). The RAM also stores various programs and data required for the operation of the computer device. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0122] Typically, the following devices can be connected to the I / O interface: input devices, such as sensors or visual information acquisition devices; output devices, such as displays; storage devices, such as magnetic tapes or hard drives; and communication devices. Communication devices allow the computer device to communicate wirelessly or wiredly with other devices (such as edge computing devices) to exchange data. Although Figure 6 A computer apparatus with various devices is shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or included alternatively.

[0123] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processor, all or part of the steps of the multi-dimensional dynamic perception and intelligent computing power scheduling method for computing power networks according to embodiments of this disclosure are performed.

[0124] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0125] A computer-readable storage medium according to embodiments of the present disclosure stores non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the multi-dimensional dynamic perception and intelligent computing power scheduling method for computing power networks described in the foregoing embodiments of the present disclosure are performed.

[0126] The aforementioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or portable hard drive), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).

[0127] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0128] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0129] In this disclosure, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The block diagrams of devices, apparatuses, devices, and systems involved in this disclosure are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as "comprising," "including," "having," etc., are open-ended terms meaning "including but not limited to," and are used interchangeably with them. The terms "or" and "and" as used herein refer to the terms "and / or," and are used interchangeably with them unless the context clearly indicates otherwise. The term "such as" as used herein refers to the phrase "such as but not limited to," and is used interchangeably with it.

[0130] Additionally, as used herein, the "or" used in a list of items beginning with "at least one" indicates a separate list, such that a list of, for example, "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not imply that the described example is preferred or better than other examples.

[0131] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0132] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0133] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure.

[0134] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A multi-dimensional dynamic perception and intelligent computing power scheduling method for a computing power network, characterized in that, The method comprises the following steps: Deploying several target components constituting multi-dimensional dynamic perception at terminals, edge nodes and regional computing power centers and collecting computing power indicators and network indicators of each node; Analyzing the computing power indicators and the network indicators of all nodes within a preset period through a time series prediction model to obtain computing power bottleneck risk and time delay attenuation information; Dynamically obtaining computing power dimension weight and time delay dimension weight according to the type of a task to be executed; Constructing a cost function according to the computing power dimension weight, the time delay dimension weight, the computing power bottleneck risk and the time delay attenuation information; Determining a node corresponding to the minimum cost function as a target node; Obtaining path costs corresponding to all feasible migration paths according to the target node, determining a migration path corresponding to the minimum path cost as a target migration path; Triggering intelligent scheduling of computing power according to the target node and the target migration path; The step of analyzing the computing power indicators and the network indicators of all nodes within a preset period through a time series prediction model to obtain computing power bottleneck risk and time delay attenuation information comprises the following steps: The step of inputting the target data set into an LSTM-based time series prediction model to obtain computing power bottleneck risk and time delay attenuation information comprises the following steps: inputting the target data set into an LSTM-based time series prediction model, obtaining corresponding computing power risk probability, time delay attenuation information; the computing power risk probability of the said computing power risk probability is : ， ； wherein , is a threshold value, is a parameter for the steepness of the control function, is the total number of processing units of the GPU; the delay decay information of the time delay is : wherein, is the predicted round-trip time delay, is the historical maximum time delay.

2. The method of claim 1, wherein, The step of analyzing the computing power indicators and the network indicators of all nodes within a preset period to obtain a target data set comprises the following steps: Normalizing the computing power indicators and the network indicators of all nodes within a preset period to obtain a normalized data set; Identifying and modifying outliers in the normalized data set to obtain a target data set.

3. The method of claim 1, wherein, The step of dynamically obtaining computing power dimension weight and time delay dimension weight according to the type of a task to be executed comprises the following steps: Determining a preset network delay weight and a preset computing power weight according to the type of a task to be executed; Obtaining an environmental state sensitivity of each node based on the computing power utilization rate and the network utilization rate of each node; Determining a time delay dimension score according to the preset network delay weight and the environmental state sensitivity; Determining a computing power dimension score according to the preset computing power weight and the environmental state sensitivity; Determining a time delay dimension weight and a computing power dimension weight according to the time delay dimension score and the computing power dimension score; The time delay dimension weight is : ; The computing power dimension weight is : ; wherein, is the delay dimension score, is the computing power dimension score.

4. The multi-dimensional dynamic perception and intelligent computing power scheduling method for the computing power network according to claim 3, characterized in that, The step of obtaining an environmental state sensitivity of each node based on the computing power utilization rate and the network utilization rate of each node comprises the following steps: Obtaining a change rate of the computing power utilization rate of each node; The change rate of the computing power utilization rate is : wherein is the sum of CPU utilization and GPU utilization of the node at time t, is a preset time interval; Obtaining a change rate of the network utilization rate of each node; The rate of change of network utilisation is : ； wherein, is the maximum bandwidth utilization from time t to time t+1, is the minimum bandwidth utilization from time t to time t+1, is the maximum bandwidth utilization from time t to time t+1, is the minimum bandwidth utilization from time t to time t+1, Obtaining an environmental state sensitivity of each node according to the change rate of the computing power utilization rate and the change rate of the network utilization rate of each node; The environmental state sensitivity is : .

5. The multi-dimensional dynamic perception and intelligent computing power scheduling method for the computing power network according to claim 3, characterized in that, The step of constructing a cost function according to the computing power dimension weight, the time delay dimension weight, the computing power bottleneck risk and the time delay attenuation information comprises the following steps: Obtaining a predicted network delay corresponding to each node; According to the network delay, the computing power bottleneck risk, the computing power dimension weight, and the delay dimension weight, a cost function of the corresponding node is determined; The cost function for the first node is : wherein, is a latency dimension weight, is a computing power dimension weight, is a node corresponding computing power bottleneck risk.

6. The multi-dimensional dynamic perception and intelligent computing power scheduling method for the computing power network according to claim 5, characterized in that, The method comprises the following steps: Obtaining the network delay corresponding to each feasible migration path, and performing a path analysis operation to obtain the processing overhead corresponding to each feasible migration path; According to the network delay and the processing overhead, the path cost corresponding to each feasible migration path is obtained; In some embodiments, the method further comprises: In some embodiments, the method further comprises: : ; network latency for the migration path, processing overhead for SID header parsing, network latency overhead weight, processing overhead weight; All the path costs corresponding to the feasible migration paths are sorted in ascending order, and the migration path corresponding to the minimum path cost is determined as the target migration path.

7. A computer apparatus, comprising: The computer device comprises: At least one processor; and The memory is in communication connection with the at least one processor; wherein The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the multi-dimensional dynamic perception and intelligent computing power scheduling method for a computing power network according to any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for causing a computer to execute the multi-dimensional dynamic perception and intelligent computing power scheduling method for a computing power network according to any one of claims 1-6.

9. A computer program product comprising computer instructions, characterized in that, The computer instructions are executed by the processor to implement the steps of the method according to any one of claims 1-6.

Images