Computing power scheduling method and device for computing power network and computing power network

Abstract

The present disclosure relates to a computing power scheduling method, a computing power scheduling device and a computing power network for the computing power network. The method comprises: determining the future task load of each computing power node in the computing power network based on the historical computing power resource utilization of the computing power node; selecting a computing power node whose future task load is less than a task load warning threshold as a candidate computing power node to construct a candidate computing power node set; in response to the future task load of at least one computing power node being not less than the task load warning threshold, taking the at least one computing power node as a warning computing power node; and performing a first operation for each warning computing power node: determining at least one urgent task based on the characteristic label of the to-be-processed task of the warning computing power node, for each urgent task, calculating the path comprehensive weight of each candidate computing power node, selecting the candidate computing power node whose path comprehensive weight is the smallest as a target computing power node, and migrating the urgent task to the target computing power node.

Description

Technical Field

[0001] This disclosure relates to the field of network operation and management technology, and more specifically, to a computing power scheduling method for a computing power network, a computing power scheduling device for a computing power network, a computing power network, and related computer-readable storage media and computer program products. Background Technology

[0002] Computing Power Network (CPN) is a new type of infrastructure that deeply integrates computing power with network connectivity. Through technologies such as resource heterogeneity, dynamic scheduling, and cross-domain collaboration, CPN enables unified orchestration, intelligent scheduling, and global optimization of computing, storage, and network resources to support the high reliability, low latency, and high elasticity service requirements of upper-layer intelligent applications. Summary of the Invention

[0003] A brief overview of this disclosure is given below to provide a basic understanding of some aspects of it. However, it should be understood that this overview is not an exhaustive summary of this disclosure. It is not intended to identify key or essential parts of this disclosure, nor is it intended to limit the scope of this disclosure. Its purpose is merely to present certain concepts of this disclosure in a simplified form as a prelude to the more detailed description that follows.

[0004] According to a first aspect of this disclosure, a computing power scheduling method for a computing power network is provided, comprising: determining the future task load of each computing power node within a preset future time window based on the historical computing power resource utilization rate of each computing power node among a plurality of computing power nodes in the computing power network; selecting computing power nodes whose future task load is less than a task load warning threshold from the plurality of computing power nodes as candidate computing power nodes to construct a set of candidate computing power nodes; in response to at least one computing power node among the plurality of computing power nodes having a future task load not less than the task load warning threshold, designating at least one computing power node as a warning computing power node; and performing a first operation for each warning computing power node, the first operation comprising: determining at least one urgent task based on the characteristic label of the unprocessed task of the warning computing power node; calculating the path comprehensive weight of each candidate computing power node in the set of candidate computing power nodes for each of the at least one urgent task, the path comprehensive weight indicating the cost of migrating the urgent task from the warning computing power node to the candidate computing power node; selecting the candidate computing power node with the smallest path comprehensive weight as a target computing power node; and migrating the urgent task to the target computing power node.

[0005] In some embodiments, for each of the at least one emergency task, calculating the path comprehensive weight of each candidate computing node in the candidate computing node set includes: for each candidate computing node in the candidate computing node set: determining the network transmission cost and network latency cost of migrating the emergency task from the warning computing node to the candidate computing node; determining the processing cost of the candidate computing node in handling the emergency task; determining the reliability cost of the candidate computing node; and determining the path comprehensive weight of the candidate computing node based on the determined network transmission cost, network latency cost, processing cost, and reliability cost.

[0006] In some embodiments, the utilization rate of each computing resource is determined by a computing modeling module in the computing network orchestration management layer of the computing network; and / or the utilization rate weighting coefficient of each computing resource is determined based on the importance of that computing resource to the urgent task.

[0007] In some embodiments, determining the reliability cost of the candidate computing node includes: determining the future risk score of the candidate computing node based on its future task load; determining the current risk score of the candidate computing node based on the changes in its task load within the current time window; determining the historical risk score of the candidate computing node based on its failure records within the historical time window; determining the comprehensive risk score of the candidate computing node based on the determined future risk score, current risk score, and historical risk score; and determining the reliability cost of the candidate computing node based on its reliability weighting coefficient and comprehensive risk score.

[0008] In some embodiments, the first operation further includes: reducing the inbound bandwidth weight of the early warning computing node.

[0009] In some embodiments, reducing the inbound bandwidth weight of the early warning computing node includes: determining a reduction ratio coefficient for the inbound bandwidth weight of the early warning computing node based on the degree to which the future task load of the early warning computing node exceeds the task load early warning threshold, wherein the reduction ratio coefficient is positively correlated with the degree.

[0010] In some embodiments, the first operation further includes: determining the current computing resource utilization rate of the early warning computing node and redetermining the future task load of the early warning computing node within a preset future time window based on the current computing resource utilization rate; and in response to the redetermined future task load being less than the task load early warning threshold, gradually increasing the inbound bandwidth weight of the early warning computing node based on a preset recovery step size and observation period.

[0011] In some embodiments, gradually increasing the inbound bandwidth weight of the early warning computing power node based on a preset increase step size and an observation period includes: performing a second operation, the second operation including: increasing the inbound bandwidth weight of the early warning computing power node based on a preset increase step size; determining the current computing power resource utilization rate of the early warning computing power node and determining at least one of the current task load of the early warning computing power node and the future task load of the early warning computing power node within a preset future time window within the observation period; continuing to perform the second operation in response to at least one being less than a task load early warning threshold; or regressing the inbound bandwidth weight of the early warning computing power node by a preset increase step size and waiting for a preset waiting period before continuing to perform the second operation in response to either at least one being not less than a task load early warning threshold.

[0012] In some embodiments, future and current task loads are determined via a prediction model built on a lightweight time series forecasting architecture.

[0013] In some embodiments, the characteristic labels of the task to be processed include at least one of a deadline label and a priority label; and / or computing resources include at least one of the central processing unit computing power, graphics processing unit computing power, memory and throughput capacity of the computing network.

[0014] In some embodiments, migrating the emergency task to the target computing node includes migrating the emergency task and its context state together to the target computing node.

[0015] In some embodiments, in response to the future task load of at least one of the multiple computing power nodes not being less than the task load warning threshold, designating at least one computing power node as a warning computing power node includes: in response to the future task load of at least one of the multiple computing power nodes not being less than the task load warning threshold in a series of consecutive time windows within a preset future time window, designating at least one computing power node as a warning computing power node.

[0016] According to a second aspect of this disclosure, a computing power scheduling device for a computing power network is provided, comprising: a processor; and a memory storing computer-executable instructions, which, when executed by the processor, cause the processor to perform the method described according to any embodiment of the first aspect of this disclosure.

[0017] According to a third aspect of this disclosure, a computer-readable storage medium is provided having computer-executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method described according to any embodiment of the first aspect of this disclosure.

[0018] According to a fourth aspect of this disclosure, a computer program product is provided, the computer program product including instructions that, when executed by a processor, implement the method described according to any embodiment of the first aspect of this disclosure.

[0019] According to a fifth aspect of this disclosure, a computing power network is provided, comprising: a computing power network service layer, a computing power network control layer, a computing power network resource layer, and a computing power network orchestration management layer. The computing power network orchestration management layer includes a computing power network security module and a computing power modeling module. The computing power modeling module is configured to determine the future task load of each computing power node within a preset future time window based on the historical computing power resource utilization rate of each of multiple computing power nodes in the computing power network. The computing power network security module is configured to: select computing power nodes whose future task load is less than a task load warning threshold from among the multiple computing power nodes as candidate computing power nodes to construct a candidate computing power node set; and respond to multiple computing power nodes... If the future task load of at least one computing power node in the node is not less than the task load warning threshold, at least one computing power node is designated as a warning computing power node. For each warning computing power node, a first operation is performed, which includes: determining at least one urgent task based on the characteristic label of the unprocessed task of the warning computing power node; for each of the at least one urgent task, calculating the path comprehensive weight of each candidate computing power node in the candidate computing power node set, the path comprehensive weight indicating the cost of migrating the urgent task from the warning computing power node to the candidate computing power node; selecting the candidate computing power node with the smallest path comprehensive weight as the target computing power node; and migrating the urgent task to the target computing power node. Attached Figure Description

[0020] The foregoing and other features and advantages of this disclosure will become clear from the following description of embodiments illustrated in conjunction with the accompanying drawings. The drawings, incorporated herein and forming a part of the specification, are further used to explain the principles of this disclosure and to enable those skilled in the art to make and use it. Wherein: Figure 1 This is a flowchart illustrating a computing power scheduling method for a computing power network according to some embodiments of the present disclosure; Figure 2 This is a flowchart illustrating a non-limiting example process of a computing power scheduling method for a computing power network according to some embodiments of the present disclosure; Figure 3 This is a schematic block diagram illustrating a computing power scheduling device according to some embodiments of the present disclosure; Figure 4 This is a schematic block diagram illustrating a computer system on which embodiments of the present disclosure may be implemented; Figure 5 This is a schematic block diagram illustrating a computing network according to some embodiments of the present disclosure.

[0021] Note that in the embodiments described below, the same reference numerals are sometimes used across different figures to denote the same parts or parts with the same function, and repeated descriptions are omitted. In some cases, similar reference numerals and letters are used to denote similar items, so once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0022] For ease of understanding, the positions, dimensions, and extents of the structures shown in the accompanying drawings and other materials may not represent actual positions, dimensions, and extents. Therefore, this disclosure is not limited to the positions, dimensions, and extents disclosed in the accompanying drawings and other materials. Detailed Implementation

[0023] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present disclosure.

[0024] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the scope of this disclosure or its application or use. That is, the structures and methods herein are shown in an exemplary manner to illustrate different embodiments of the structures and methods in this disclosure. However, those skilled in the art will understand that they merely illustrate exemplary ways that can be used to implement this disclosure, and not exhaustive ways. Furthermore, the drawings are not necessarily drawn to scale, and some features may be enlarged to show details of specific components.

[0025] In addition, techniques, methods and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods and equipment should be considered part of the specification.

[0026] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0027] In this document, the same or similar characters may be used to represent the same or similar variables. Therefore, once a variable is defined in one embodiment, it does not need to be described again in subsequent embodiments.

[0028] During operation, computing power networks can experience service failures due to the gradual degradation of the load on computing nodes (also referred to as "nodes") (such as load ramp-up, resource leaks, etc.). Fault management mechanisms for computing power networks in related technologies typically only trigger actions when a node crashes or the service becomes completely unavailable. This lack of ability to identify symptoms of load degradation means that critical opportunities for proactive intervention before failures occur are missed. Consequently, these fault management mechanisms fail to take preventative measures before node performance completely collapses, leading to uncontrollable and sudden failures.

[0029] In addition, the fault management mechanisms of related technologies usually adopt hard strategies such as "directly cutting off traffic" or "overall task switching" for faulty nodes, which causes healthy nodes to receive too many tasks from faulty nodes in a short period of time and become overloaded (i.e., cascading failures occur). At the same time, all queued or processed tasks in the faulty node fail, causing drastic fluctuations in service interruption and cold start recovery.

[0030] To this end, this disclosure provides a computing power scheduling method for computing power networks, which identifies computing power nodes that will experience load degradation in the future by predicting the future task load of computing power nodes, filters the urgent tasks among the tasks of the computing power nodes with warning, and migrates the urgent tasks to the healthy computing power nodes with the lowest migration cost.

[0031] Therefore, by identifying "early warning states" before a failure occurs, proactive intervention measures can be taken to prevent the collapse of computing nodes due to gradual overload. Among the intervention measures, only urgent tasks are migrated to healthy computing nodes to avoid cascading failures caused by sudden traffic surges and cold start recovery caused by migrating all tasks, achieving smooth scheduling without service awareness. Furthermore, non-urgent tasks are still handled by the early warning computing nodes to avoid wasting the available computing resources of these nodes and improve the resource utilization of the computing network.

[0032] The following will describe in detail, with reference to the accompanying drawings, various embodiments of a computing power scheduling method for a computing power network. It will be understood that actual computing power scheduling methods for computing power networks may include other steps, but to avoid obscuring the key points of this disclosure, these other steps will not be discussed herein and are not shown in the accompanying drawings.

[0033] refer to Figure 1 This is a flowchart illustrating a computing power scheduling method 100 (hereinafter referred to as "method 100") for a computing power network according to some embodiments of the present disclosure. Figure 1 As shown, method 100 may include steps S102 to S108.

[0034] In step S102, the future task load of a computing node within a preset future time window is determined based on the historical computing resource utilization rate of each computing node among multiple computing nodes in the computing power network.

[0035] In some examples, computing resources may include at least one of the following: central processing unit (CPU) computing power, memory, graphics processing unit (GPU) computing power, and throughput (i.e., input / output (I / O)).

[0036] In some embodiments, the computing power resource utilization rate of a computing power node (including historical computing power resource utilization rate and current computing power resource utilization rate) can be determined via the computing power modelling module in the computing power network orchestration and management layer of the computing power network.

[0037] In some embodiments, future task load is determined via a predictive model built on a lightweight time series forecasting architecture. In some embodiments, current task load may also be determined via a predictive model.

[0038] In some examples, lightweight time-series models such as single-layer Long Short-Term Memory (LSTM) networks or TimeBase can be used. The prediction model can be deployed in the computing power modeling module of the computing power network, taking the historical computing power resource utilization rate of each computing node over several control periods as input, and outputting predicted task load values ​​for multiple future time windows (e.g., the next 5-10 minutes). Furthermore, to adapt to the resource-constrained nature of computing nodes, the prediction model can limit the number of parameters to below 100,000 by restricting the hidden layer dimension (e.g., limiting hidden units to <128) or employing a segment-level prediction strategy, ensuring inference latency is below 100 ms.

[0039] Historical load data can be used to pre-train the prediction model in an offline environment. Furthermore, during runtime, the prediction model can employ a sliding window mechanism to collect the latest real-time data and dynamically adapt to the drift and periodic changes in node load through online lightweight fine-tuning (e.g., gradient updates based on the most recent few time steps) or exponentially weighted moving averages, thereby improving prediction accuracy. This disclosure does not limit the specific architecture and training method of the prediction model.

[0040] In some embodiments, in response to the future task load of at least one of the multiple computing power nodes not being less than the task load warning threshold, designating at least one computing power node as a warning computing power node includes: in response to the future task load of at least one of the multiple computing power nodes not being less than the task load warning threshold in a series of consecutive time windows within a preset future time window, designating at least one computing power node as a warning computing power node.

[0041] For example, a preset future time window might include N future time windows (N being an integer greater than 1) following the current time window. If the prediction model predicts that the task load of a certain computing node will exceed the task load warning threshold in both the (N-1)th and Nth future time windows—that is, if the future task load of this computing node exceeds the task load warning threshold for two consecutive future time windows—then this computing node is considered to be in a warning state, or in other words, it is designated as a warning computing node. This can improve the accuracy of selecting warning computing nodes.

[0042] In step S104, computing power nodes whose future task load is less than the task load warning threshold are selected from multiple computing power nodes as candidate computing power nodes to construct a candidate computing power node set.

[0043] For example, in the operation and maintenance of computing power networks, "standby computing power nodes" can be pre-selected through strategies such as "resource pre-occupancy" or "virtual grouping". Normally, standby computing power nodes can handle "interruptible" tasks (such as data cleaning tasks) normally, but in response to the detection of a warning computing power node, they can immediately preemptively release resources to ensure that the migration process of emergency tasks of the warning computing power node is smooth and controllable.

[0044] In step S106, in response to the fact that the future task load of at least one of the multiple computing power nodes is not less than the task load warning threshold, at least one computing power node is designated as the warning computing power node.

[0045] In some examples, the distribution of historical computing resource utilization can be determined based on the historical computing resource utilization of each computing node, and the task load warning threshold can be determined based on this distribution.

[0046] Combining the utilization rates of multiple computing resources can avoid the problems associated with warning thresholds determined based on the utilization rate of a single computing resource. For example, a computing node might be mistakenly identified as a warning node when it is running a resource-intensive task. A comprehensive utilization rate can more accurately reflect the true load pressure on a node and reduce unnecessary interventions. Of course, the warning threshold can also be fine-tuned based on the experience of technical personnel according to the task type (such as CPU-intensive, GPU-intensive, I / O-intensive, or memory-intensive).

[0047] As a non-restrictive example, suppose the historical average utilization rates of computing resources for a computing node across multiple historical time windows are: CPU 75%, Memory 82%, GPU 68%, and I / O 55%, with weights of 0.3, 0.3, 0.2, and 0.2, respectively. Then, the overall historical utilization rate of computing resources for the computing node is 0.3×0.75 + 0.3×0.82 + 0.2×0.68 + 0.2×0.55, or 0.717. This value can be used as a task load warning threshold.

[0048] In this example, suppose that when a computing node is running video transcoding and artificial intelligence (AI) inference tasks, the prediction model predicts that the future task load of the computing node within a preset future time window will be 0.9, which exceeds the warning threshold of 0.717. Therefore, the computing node is identified as a "warning computing node".

[0049] In step S108, a first operation is performed for each early warning computing node. The first operation includes: determining at least one urgent task based on the characteristic tags of the tasks to be processed by the early warning computing node; for each of the at least one urgent task, calculating the path comprehensive weight of each candidate computing node in the candidate computing node set; selecting the candidate computing node with the smallest path comprehensive weight as the target computing node; and migrating the urgent task to the target computing node. Here, the path comprehensive weight indicates the cost of migrating the urgent task from the early warning computing node to the candidate computing node.

[0050] In some embodiments, the characteristic labels of a task to be processed may include at least one of a due date label and a priority label. Here, the due date label is used to indicate the urgency of the task to be processed, and the priority label is used to indicate the importance of the task to be processed.

[0051] For example, if a certain early warning computing node has 500 tasks in its current queue, 50 of which are labeled with a deadline of "<300ms" and another 50 are labeled with a priority of "highest priority", then these 100 tasks can be identified as urgent tasks.

[0052] In some embodiments, for each of the at least one emergency task, calculating the path comprehensive weight of each candidate computing node in the candidate computing node set includes: for each candidate computing node in the candidate computing node set: determining the network transmission cost and network latency cost of migrating the emergency task from the warning computing node to the candidate computing node; determining the processing cost of the candidate computing node in handling the emergency task; determining the reliability cost of the candidate computing node; and determining the path comprehensive weight of the candidate computing node based on the determined network transmission cost, network latency cost, processing cost, and reliability cost.

[0053] Because different urgent tasks have varying degrees of demand for different types of computing resources and different levels of emphasis on different types of costs, the overall path weight for the same path may differ for different urgent tasks. Therefore, for multiple urgent tasks with warning computing nodes, suitable target computing nodes can be selected from the set of candidate computing nodes for each urgent task, which helps improve scheduling accuracy.

[0054] In some examples, the path weight of each candidate computing node can be expressed as: ,in, This indicates the overall path weight of the candidate computing node. This indicates the network transmission cost of migrating the emergency task from the early warning computing node to the alternative computing node. This indicates the network latency cost of migrating the emergency task from the early warning computing node to the alternative computing node. This indicates the processing cost of the alternative computing node in handling this urgent task. This represents the reliability cost of the alternative computing node.

[0055] In related technologies, migrating tasks from a faulty node to a backup computing node only considers network transmission costs and network latency costs. In other words, these technologies only consider the perspective of the faulty node, aiming to migrate its tasks to the backup computing node as quickly and stably as possible, without considering whether the backup computing node has sufficient computing resources to handle these tasks. Therefore, these technologies often lead to cascading failures. In contrast, the method 100 disclosed in this paper incorporates both the processing costs and reliability costs of the backup computing node, not only achieving the migration of emergency tasks from the early warning computing node but also ensuring that the migrated emergency tasks can be processed efficiently, thereby improving service quality assurance.

[0056] In some examples, the network transmission cost of the emergency task migrating from the early warning computing node to the alternative computing node is determined based on the bandwidth weighting coefficient and the available bandwidth of the path from the early warning computing node to the alternative computing node; the network latency cost of the emergency task migrating from the early warning computing node to the alternative computing node is determined based on the network latency weighting coefficient and the network transmission latency of the emergency task migrating from the early warning computing node to the alternative computing node, wherein the bandwidth weighting coefficient and the network latency weighting coefficient are determined according to the sensitivity of the emergency task to bandwidth and network latency.

[0057] In some examples, ,in, This is the bandwidth weighting coefficient. This represents the available bandwidth of the path from the warning computing node to the alternative computing node. In some examples, ,in, This is the network latency weighting coefficient. This setting represents the network transmission latency (including propagation latency, switching latency, queuing latency, etc.) for migrating the emergency task from the early warning computing node to the backup computing node. When the emergency task is a bandwidth-sensitive task (such as a video distribution task), this setting can be increased. The value can be increased. Alternatively, when the urgent task is one that is sensitive to network latency (such as a real-time interactive task), the value can be increased. The value of .

[0058] In some embodiments, determining the processing cost of the candidate computing node for handling the urgent task includes: determining the processing cost of the candidate computing node for handling the urgent task based on the utilization rate and utilization weight coefficient of each of the candidate computing node's multiple computing resources. In some examples, the utilization weight coefficient of each computing resource is determined based on the importance of that computing resource to the urgent task.

[0059] In some examples, ,in, This indicates the processing cost of the alternative computing node in handling this urgent task. This represents the weighting coefficient for the utilization rate of the k-th computing resource. This represents the normalized utilization rate of the k-th computing resource. The value range of is [0, 1], and R represents the set of computing resources, R = {CPU, Memory, GPU, I / O}. Here, the sum of the utilization weight coefficients of multiple computing resources is 1, and the more important each computing resource is to the urgent task, the larger its utilization weight coefficient.

[0060] For example, Table 1 below shows the utilization weighting coefficients of four computing resources for different urgent tasks.

[0061] Table 1

[0062] As shown in Table 1, for AI model training tasks, GPU is the key computing resource, therefore the GPU utilization weight coefficient is the largest; for general web service tasks, CPU and memory are the key computing resources, but GPU is not required, therefore the CPU utilization weight coefficient and memory utilization weight coefficient are relatively large, while the GPU utilization weight coefficient is 0; for big data offline analysis tasks, throughput is the most critical computing resource, and CPU is the second most critical computing resource, therefore the I / O utilization weight coefficient is the largest, followed by the CPU utilization weight coefficient; for video transcoding tasks, GPU is the most critical computing resource, and CPU is the second most critical computing resource, therefore the GPU utilization weight coefficient is the largest, followed by the CPU utilization weight coefficient; for database tasks, memory is the most critical computing resource, therefore the memory utilization weight coefficient is the largest.

[0063] In some embodiments, determining the reliability cost of the candidate computing node includes: determining the future risk score of the candidate computing node based on its future task load; determining the current risk score of the candidate computing node based on the changes in its task load within the current time window; determining the historical risk score of the candidate computing node based on its failure records within the historical time window; determining the comprehensive risk score of the candidate computing node based on the determined future risk score, current risk score, and historical risk score; and determining the reliability cost of the candidate computing node based on its reliability weighting coefficient and comprehensive risk score.

[0064] In some examples, ,in, This represents the reliability weighting coefficient. The larger the value, the greater the likelihood that the candidate computing node will be selected (or, the greater the likelihood that it will become the target computing node). This is the overall risk score for the candidate computing power node, ranging from 0 to 1, with higher values ​​indicating a less reliable node. This allows the computing network to avoid high-risk alternative computing nodes that appear to be about to fail, and instead direct traffic to more stable alternative computing nodes, thereby completing a smooth migration of traffic before a failure occurs and improving the overall availability of the computing network.

[0065] In some examples, the overall risk score of the candidate computing power node is determined based on the candidate computing power node’s future risk score and future risk weight coefficient, current risk score and current risk weight coefficient, and historical risk score and historical risk weight coefficient.

[0066] For example, ,in, This indicates the future risk score of the candidate computing node. This indicates the current risk score of the candidate computing node. This indicates the historical risk score of the candidate computing node. , , These represent the future risk weighting coefficient, the current risk weighting coefficient, and the historical risk weighting coefficient, respectively. .

[0067] In some examples, This can be obtained through a predictive model. If the predicted future task load continues to approach the task load warning threshold, then... As the predicted future task load approaches 1, if the predicted future task load is far from the task load warning threshold, then... This then approaches 0. For example, if the future task load of a certain computing node will increase from 60% to 85% within 3 windows (warning threshold = 80%), then... The value is assigned to 0.8 (high risk).

[0068] In some examples, the slope of load change and jitter variance of computing nodes within the current time window can be evaluated based on real-time computing resource data within the current time window, thereby determining... If the load increases sharply or fluctuates drastically within a short period of time, it indicates that the computing node is unstable. The corresponding increase. For example, if the CPU utilization of a computing node spikes from 40% to 75% within 5 minutes, then... The value is assigned to 0.6.

[0069] In some examples, This can be determined based on the historical fault records of computing nodes within a past time window (e.g., 24 hours). Historical fault records can include, for example, the number of times the computing node went down within the past time window, the frequency of entering a warning / fault state, and the abnormal task exit rate. The more historical fault records, the better. The higher the frequency, the better. For example, if a computing node has been identified as a warning computing node three times in the past 24 hours, its historical failure frequency is higher than the average. The value is assigned to 0.4.

[0070] In some examples, the values ​​of the future risk weighting coefficient, the current risk weighting coefficient, and the historical risk weighting coefficient can be adjusted depending on the type of emergency task.

[0071] For example, Table 2 below shows the future risk weight coefficient, current risk weight coefficient, and historical risk weight coefficient for emergency tasks in different scenarios.

[0072] Table 2

[0073] As shown in Table 2, for emergency tasks under steady-state operation scenarios, the main requirement is to mitigate risks in advance and filter out accidental fluctuations, that is, to rely primarily on prediction and secondarily on historical data. For urgent tasks under sudden traffic / load spikes, real-time indicators are the most sensitive, and the primary requirement is rapid response to sudden changes; that is, the immediate situation is the priority, with prediction as a secondary measure. For urgent tasks in scenarios where nodes are chronically unstable, it is necessary to penalize nodes that repeatedly cause problems; that is, historical and predictive data should be the primary basis, with the current situation as a secondary consideration. .

[0074] In some embodiments, the first operation may further include: reducing the inbound bandwidth weight of the early warning computing node. Here, the inbound bandwidth weight indicates the number of new tasks assigned to the early warning computing node; the larger the inbound bandwidth weight, the larger the number of new tasks assigned to the early warning computing node, and the smaller the inbound bandwidth weight, the smaller the number of new tasks assigned to the early warning computing node.

[0075] In some embodiments, reducing the inbound bandwidth weight of the early warning computing node includes: determining a reduction ratio coefficient for the inbound bandwidth weight of the early warning computing node based on the degree to which the future task load of the early warning computing node exceeds the task load early warning threshold, wherein the reduction ratio coefficient is positively correlated with the degree.

[0076] The greater the future task load of a warning computing node exceeds the task load warning threshold, the larger the reduction ratio, resulting in a greater reduction in inbound bandwidth weight. This further reduces the load on the warning computing node, mitigating its load degradation. Conversely, the smaller the future task load exceeds the task load warning threshold, the smaller the reduction ratio, resulting in a smaller reduction in inbound bandwidth weight. This avoids excessively reducing the load on the warning computing node, preventing unnecessary business disruptions.

[0077] Compared to related technologies that directly switch all incoming traffic to the early warning computing power node, this disclosure reduces the incoming traffic according to the degree of excess, which can accurately control the traffic allocation of the early warning computing power node and reduce the impact of traffic mutation on the computing power network.

[0078] In some embodiments, the first operation may further include: determining the current computing resource utilization rate of the early warning computing node and redetermining the future task load of the early warning computing node within a preset future time window based on the current computing resource utilization rate; in response to the redetermined future task load being less than the task load early warning threshold, gradually increasing the inbound bandwidth weight of the early warning computing node based on a preset recovery step size and observation period.

[0079] When the future task load of the early warning computing node drops below the task load early warning threshold, it indicates that the early warning computing node has gradually returned to normal. Therefore, the inbound traffic to the early warning computing node can be gradually increased, thereby restoring the service capacity of the computing node.

[0080] In some embodiments, gradually increasing the inbound bandwidth weight of the early warning computing power node based on a preset increase step size and an observation period includes: performing a second operation, the second operation including: increasing the inbound bandwidth weight of the early warning computing power node based on a preset increase step size; determining the current computing power resource utilization rate of the early warning computing power node and determining at least one of the current task load of the early warning computing power node and the future task load of the early warning computing power node within a preset future time window within the observation period; continuing to perform the second operation in response to the at least one being less than a task load early warning threshold; or regressing the inbound bandwidth weight of the early warning computing power node by a preset increase step size and waiting for a preset waiting period before continuing to perform the second operation in response to any of the at least one being not less than a task load early warning threshold.

[0081] As a non-restrictive example, assume that the inbound bandwidth weight of the early warning computing node is reduced by X%, with a preset recovery step size of Δw = 10%. X, with an observation period T of 60 seconds. When the recalculated future task load is less than the task load warning threshold, a weight boost can be performed, i.e. Then, it waits for the observation period T. During the observation period T, the prediction model determines the current task load of the early warning computing node in the observation period T and the future task load in the next time window after the observation period T.

[0082] If both the current and future task loads determined by the prediction model are below the task load warning threshold, then the next weight boost is allowed; if either the current or future task load determined by the prediction model is not less than the task load warning threshold, then... Back to Then wait for a preset waiting period (e.g., 30 seconds) until the state of the warning computing node stabilizes again before performing the second operation.

[0083] It is understandable that the warning computing power node may or may not be able to recover to its original inbound bandwidth weight (i.e., 100% inbound bandwidth weight) through the second operation, depending on the actual load of the warning computing power node. Therefore, in response to the warning computing power node's inbound bandwidth weight recovering to its original value, the second operation is stopped; or in response to multiple recoveries of the warning computing power node's inbound bandwidth weight, where each time either the current task load or the future task load of the warning computing power node is not less than the task load warning threshold, i.e., in response to multiple recoveries followed by multiple regressions of the warning computing power node's inbound bandwidth weight, the second operation is stopped.

[0084] Therefore, by gradually and in batches increasing the bandwidth weight, the probability of sudden changes or fluctuations in traffic of the early warning computing nodes can be reduced, which is conducive to restoring the service capabilities of the early warning computing nodes and improving the service capabilities of the computing network.

[0085] In some embodiments, migrating the emergency task to the target computing node may include migrating the emergency task and its context state together to the target computing node.

[0086] In some examples, the context state may include all the environmental information and intermediate data required for the emergency task to execute at a certain point in time, such as all runtime dependencies of the emergency task from its inception to the current interruption node, resource occupancy markers, business process progress, scheduling identifiers, etc.

[0087] Therefore, the emergency task and its context state are migrated and synchronized to the target computing node, so that the target computing node does not need to re-execute the task from the beginning, but can directly continue the operation, avoiding problems such as task timeout, duplicate calculation, and data loss.

[0088] In addition, it is understood that the method 100 disclosed herein is also applicable to computing nodes that are in a fault state or about to enter a fault state, because the future task load of such computing nodes must exceed the task load warning threshold.

[0089] refer to Figure 2 This is a flowchart illustrating a non-limiting example process 200 of a computing power scheduling method for a computing power network according to some embodiments of the present disclosure.

[0090] like Figure 2As shown, the non-limiting example process 200 includes: at step S202, predicting the future task load of the computing power node via a prediction model; at step S204, determining whether the future task load of the computing power node is not less than the task load warning threshold for N consecutive times; if so, at step S206, selecting the computing power node whose future task load is not less than the task load warning threshold as the warning computing power node; if not, at step S208, selecting the computing power node whose future task load is less than the task load warning threshold as the candidate computing power node to construct a candidate computing power node set.

[0091] Next, steps S210 to S230 are executed for each early warning computing node.

[0092] In step S210, for the early warning computing power node i, the inbound bandwidth weight of the early warning computing power node i is reduced; in step S212, the emergency task of the early warning computing power node i is determined; in step S214, for each emergency task, the path comprehensive weight of each candidate computing power node is calculated; in step S216, the candidate computing power node with the smallest path comprehensive weight is selected as the target computing power node, and the emergency task and its context are migrated to the target computing power node.

[0093] In step S218, the future task load of the early warning computing power node i is re-predicted through the prediction model; in step S220, it is determined whether the future task load of the early warning computing power node i is less than the task load early warning threshold; if not, it means that the early warning computing power node i is still in the early warning state, so return to step S210 and continue to reduce the task load of the early warning computing power node i; if yes, it means that the early warning computing power node i is released from the early warning state, and step S222 is executed to increase the inbound bandwidth weight of the early warning computing power node i according to the preset increase step size.

[0094] After executing step S222, in step S224, it is determined whether the task load (including at least one of the current task load and the future task load) of the warning computing power node i is less than the task load warning threshold: If yes, then at step S226, it is determined whether the incoming bandwidth weight of the early warning computing power node i has been restored; if it is determined that the incoming bandwidth weight of the early warning computing power node i has been restored (i.e., S226 "Yes"), then the process ends; if it is determined that the incoming bandwidth weight of the early warning computing power node i has not been restored (i.e., S226 "No"), then the process returns to step S222 to continue to increase the incoming bandwidth weight of the early warning computing power node i. If not, then at step S228, the incoming bandwidth weight of the warning computing power node i is rolled back and a preset waiting period is waited for; then at step S230, it is determined whether there have been multiple rollbacks; if it is determined that there have been multiple rollbacks (i.e., S230 "Yes"), it means that the warning computing power node i cannot continue to increase the incoming bandwidth weight, and then the process ends; if it is determined that there have been no multiple rollbacks (i.e., S230 "No"), then the process returns to step S222 to continue to increase the incoming bandwidth weight of the warning computing power node i.

[0095] This disclosure also provides a computing power scheduling device. (See reference...) Figure 3 This is a schematic block diagram illustrating a computing power scheduling device 300 according to some embodiments of the present disclosure. Figure 3 As shown, the computing power scheduling device 300 includes a processor 302 and a memory 304 storing computer-executable instructions, which, when executed by the processor 302, cause the processor 302 to perform a computing power scheduling method for a computing power network according to any of the foregoing embodiments of this disclosure. The processor 302 may be, for example, a central processing unit (CPU) of the computing power scheduling device 300. The processor 302 may be any type of general-purpose processor, or it may be a processor specifically designed for computing power scheduling, such as an application-specific integrated circuit (“ASIC”). The memory 304 may be coupled to the processor 302 and may include various computer-readable media accessible by the processor 302. In various embodiments, the memory 304 described herein may include volatile and non-volatile media, removable and non-removable media. For example, the memory 304 may include any combination of: random access memory (“RAM”), dynamic RAM (“DRAM”), static RAM (“SRAM”), read-only memory (“ROM”), flash memory, cache memory, and / or any other type of non-transient computer-readable media. The memory 304 may store instructions that, when executed by the processor 302, cause the processor 302 to execute the method 100 according to any of the foregoing embodiments of the present disclosure.

[0096] The computing power scheduling device 300 is configured to perform the method 100 described in any of the foregoing embodiments, and therefore can be referred to the descriptions of the various embodiments of method 100 above, which will not be repeated here.

[0097] This disclosure also provides a computer-readable storage medium having computer-executable instructions stored thereon, which, when executed by a processor, cause the processor to perform a computing power scheduling method for a computing power network according to any of the foregoing embodiments.

[0098] This disclosure also provides a computer program product including instructions that, when executed by a processor, implement the computing power scheduling method for a computing power network according to any of the foregoing embodiments. The instructions can be any set of instructions that can be executed directly by one or more processors, such as machine code, or any set of instructions that can be executed indirectly, such as a script. The instructions can be stored in an object code format for direct processing by one or more processors, or stored in any other computer language, including scripts or sets of independent source code modules that are interpreted on demand or compiled in advance.

[0099] Figure 4This is a schematic block diagram illustrating a computer system 400 on which embodiments of the present disclosure may be implemented. The computer system 400 includes a bus 402 or other communication mechanism for transmitting information, and a processing means 404 coupled to the bus 402 for processing information. The computer system 400 also includes a memory coupled to the bus 402 for storing instructions to be executed by the processing means 404; the memory may be random access memory 406 or other dynamic storage device. The memory may also be used to store temporary variables or other intermediate information during the execution of instructions to be executed by the processing means 404. The computer system 400 also includes a read-only memory (ROM) 408 or other static storage device coupled to the bus 402 for storing static information and instructions for the processing means 404. A storage device 410, such as a magnetic disk or optical disk, is provided and coupled to the bus 402 for storing information and instructions. The computer system 400 may be coupled via the bus 402 to an output device 412 for providing output to a user, such as, but not limited to, a display (such as a cathode ray tube (CRT) or liquid crystal display (LCD)), a speaker, etc. Input devices 414, such as a keyboard, mouse, and microphone, are coupled to bus 402 for transmitting information and command selections to processing device 404. Computer system 400 can execute embodiments of this disclosure. Consistent with certain implementations of this disclosure, results are provided by computer system 400 in response to processing device 404 executing one or more sequences of one or more instructions contained in memory. Such instructions may be read into memory from another computer-readable medium, such as storage device 410. Execution of the sequence of instructions contained in memory causes processing device 404 to perform the methods described herein. Alternatively, the teachings may be implemented using hardwired circuitry in place of or in combination with software instructions. Therefore, implementations of this disclosure are not limited to any particular combination of hardware circuitry and software. In various embodiments, computer system 400 may be connected across a network via network interface 416 to one or more other computer systems, such as computer system 400, to form a networked system. This network may include a private network or a public network such as the Internet. In a networked system, one or more computer systems may store data and supply data to other computer systems. As used herein, the term "computer-readable medium" refers to any medium that participates in providing instructions to processing device 404 for execution. Such media can take many forms, including but not limited to non-volatile media, volatile media, and transmission media. Non-volatile media include, for example, optical discs or magnetic disks such as storage device 410. Volatile media include dynamic memory such as memory. Transmission media include coaxial cables, copper wires, and optical fibers, including wiring that includes bus 402.Common forms of computer-readable media or computer program products include, for example, floppy disks, flexible disks, hard disks, magnetic tapes, or any other magnetic media, CD-ROMs, digital video discs (DVDs), Blu-ray discs, any other optical media, thumb drives, memory cards, RAM, PROMs and EPROMs, fast EPROMs, any other memory chips or cartridges, or any other tangible media from which a computer can read. Various forms of computer-readable media may be involved when carrying one or more sequences of one or more instructions to processing device 404 for execution. For example, instructions may initially be carried on a disk of a remote computer. The remote computer may load the instructions into its dynamic memory and transmit the instructions over a telephone line using a modem. A modem local to computer system 400 may receive data over a telephone line and convert the data into an infrared signal using an infrared transmitter. An infrared detector coupled to bus 402 may receive the data carried in the infrared signal and place the data on bus 402. Bus 402 carries the data to memory, and processing device 404 retrieves the instructions from memory and executes the instructions. For example, instructions received from the memory can be stored on the storage device 410 before or after execution by the processing device 404.

[0100] According to various embodiments, instructions configured to be executed by a processing device to perform a method are stored on a computer-readable medium. The computer-readable medium may be a device for storing digital information. For example, a computer-readable medium includes a compact disc read-only memory (CD-ROM) as known in the art for storing software. The computer-readable medium is accessed by a processor adapted to execute the instructions configured to be executed.

[0101] This disclosure also provides a computing power network. (See reference) Figure 5 This is a schematic block diagram illustrating a computing network 500 according to some embodiments of the present disclosure. Figure 5As shown, the computing power network 500 may include a computing power network service layer 502, a computing power network control layer 504, a computing power network resource layer 506, and a computing power network orchestration and management layer 508. The architecture of the computing power network 500 can refer to the computing power network architecture in "Y.2501: Computing power network - Framework and architecture" published by the International Telecommunication Union (ITU). Here, the computing power network service layer 502 includes a resource information processing module, a billing module, and a transaction process execution module; the computing power network control layer 504 includes a resource information collection module, a resource allocation module, and a network connection scheduling module; the computing power network resource layer 506 includes a computing resource module, a network resource module, a storage resource module, and a service resource module; and the computing power network orchestration management layer 508 includes a computing power network orchestrator module, a computing power network security module 510, a computing power modeling module 512, and a computing power OAM (Operation, Administration, and Maintenance) module.

[0102] Compared to the computing network architecture of Y.2501, the computing network security module 510 and computing modeling module 512 of this disclosure are specially configured to achieve improved computing scheduling.

[0103] Specifically, the computing power modeling module 512 is configured to determine the future task load of a computing power node within a preset future time window based on the historical computing power resource utilization of each of the multiple computing power nodes in the computing power network. In some examples, the computing power modeling module may be configured with a predictive model.

[0104] The computing power network security module 510 is configured to: select computing power nodes whose future task load is less than the task load warning threshold from multiple computing power nodes as candidate computing power nodes to construct a candidate computing power node set; in response to at least one computing power node among the multiple computing power nodes having a future task load not less than the task load warning threshold, designate at least one computing power node as a warning computing power node; and perform a first operation for each warning computing power node, the first operation including: determining at least one urgent task based on the characteristic label of the unprocessed task of the warning computing power node; for each of the at least one urgent task, calculating the path comprehensive weight of each candidate computing power node in the candidate computing power node set, the path comprehensive weight indicating the cost of migrating the urgent task from the warning computing power node to the candidate computing power node; selecting the candidate computing power node with the smallest path comprehensive weight as the target computing power node; and migrating the urgent task to the target computing power node.

[0105] Various embodiments of the computing network 500 can be found in the description of the various embodiments of method 100 above, and will not be repeated here.

[0106] The foregoing has described one or more exemplary embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0107] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or physical entities, or by products with certain functions. A typical implementation device is a server system. Of course, this disclosure does not exclude the possibility that, with the future development of computer technology, the computer implementing the functions of the above embodiments can be, for example, a personal computer, a laptop computer, an in-vehicle human-machine interaction device, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0108] While one or more embodiments of this disclosure provide the method operation steps as described in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive means. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only execution order. In actual device or terminal product execution, the methods shown in the embodiments or drawings can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment, or even a distributed data processing environment).

[0109] The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or apparatus. Without further limitation, the presence of other identical or equivalent elements in the process, method, product, or apparatus that includes said elements is not excluded. For example, the use of terms such as "first" or "second" to denote names does not indicate any particular order.

[0110] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, when implementing one or more embodiments of this disclosure, the functions of each module can be implemented in one or more software and / or hardware, or a module that performs the same function can be implemented by a combination of multiple sub-modules or sub-units. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division; 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 displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0111] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more blocks of the flowchart illustrations and / or one or more blocks of the block diagrams.

[0112] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more blocks of a block diagram.

[0113] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more blocks of a block diagram.

[0114] Those skilled in the art will understand that one or more embodiments of this disclosure may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, one or more embodiments of this disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0115] One or more embodiments of this disclosure can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a particular task or implement a particular abstract data type. One or more embodiments of this disclosure can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules can reside in local and remote computer storage media, including storage devices.

[0116] The same or similar parts between the various embodiments of this disclosure can be referred to mutually, and each embodiment focuses on describing the differences from other embodiments. In particular, for the apparatus embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and relevant parts can be referred to the description of the method embodiments. In the description of this disclosure, the descriptions of terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., mean that the specific feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of this disclosure. In this disclosure, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this disclosure and the features of the different embodiments or examples.

[0117] Additionally, when used in this disclosure, the terms “here,” “above,” “below,” “below,” “in the following,” “overall,” and similar terms should refer to the entirety of this disclosure and not any particular part thereof. Furthermore, unless expressly stated otherwise or otherwise understood in the context in which they are used, conditional language used herein, such as “may,” “possibly,” “for example,” “like,” etc., is generally intended to express that certain embodiments include, while other embodiments do not, certain features, elements, and / or states. Therefore, such conditional language is not generally intended to imply that one or more embodiments require features, elements, and / or states in any way, or whether such features, elements, and / or states are included or performed in any particular embodiment.

[0118] The above description is merely an embodiment of one or more embodiments of this disclosure and is not intended to limit the scope of the one or more embodiments of this disclosure. Various modifications and variations can be made to the one or more embodiments of this disclosure by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of the claims.

Claims

1. A method for scheduling computing power in a computing power network, comprising: The future task load of a computing node within a preset future time window is determined based on the historical computing resource utilization rate of each computing node in the computing power network. From the plurality of computing power nodes, select computing power nodes whose future task load is less than the task load warning threshold as candidate computing power nodes to construct a candidate computing power node set; In response to the fact that the future task load of at least one of the plurality of computing power nodes is not less than the task load warning threshold, the at least one computing power node is designated as a warning computing power node; as well as For each early warning computing node, a first operation is performed, the first operation including: At least one urgent task is identified based on the characteristic tags of the tasks to be processed by the early warning computing node. For each of the at least one emergency task, calculate the comprehensive path weight for each candidate computing node in the set of candidate computing nodes. The comprehensive path weight indicates the cost of migrating the emergency task from the warning computing node to the candidate computing node. Select the candidate computing node with the smallest path comprehensive weight as the target computing node, and migrate the emergency task to the target computing node.

2. The method according to claim 1, wherein, For each of the at least one urgent task, calculating the comprehensive path weight for each candidate computing node in the set of candidate computing nodes includes: For each candidate computing node in the set of candidate computing nodes: Determine the network transmission cost and network latency cost of migrating the emergency task from the early warning computing node to the alternative computing node; Determine the processing cost of the alternative computing node for handling this urgent task; Determine the reliability cost of the candidate computing node; and The path weight of the candidate computing node is determined based on the determined network transmission cost, network latency cost, processing cost, and reliability cost.

3. The method according to claim 2, wherein, The processing cost of this alternative computing node for handling this urgent task includes: The processing cost of the candidate computing node for handling the urgent task is determined based on the utilization rate and utilization weight coefficient of each computing resource among the multiple computing resources of the candidate computing node.

4. The method according to claim 3, wherein, The utilization rate of each computing resource is determined by the computing power modeling module in the computing power network orchestration management layer of the computing power network; and / or The utilization weighting coefficient for each computing resource is determined based on the importance of that computing resource to the urgent task.

5. The method according to claim 2, wherein, Determining the reliability cost of this alternative computing node includes: The future risk score of the candidate computing power node is determined based on its future task load. The current risk score of the candidate computing power node is determined based on the changes in task load within the current time window. The historical risk score of the candidate computing power node is determined based on the fault records within the historical time window of the candidate computing power node; The comprehensive risk score of the candidate computing power node is determined based on the identified future risk score, current risk score, and historical risk score; and The reliability cost of the candidate computing node is determined based on its reliability weighting coefficient and comprehensive risk score.

6. The method according to claim 1, wherein, The first operation also includes: reducing the inbound bandwidth weight of the early warning computing node.

7. The method according to claim 6, wherein, Reducing the inbound bandwidth weight of this early warning computing node includes: The reduction ratio of the incoming bandwidth weight of the early warning computing node is determined based on the degree to which the future task load of the early warning computing node exceeds the task load early warning threshold, wherein the reduction ratio is positively correlated with the degree.

8. The method according to claim 6, wherein, The first operation also includes: Determine the current computing resource utilization rate of the early warning computing node and, based on the current computing resource utilization rate, re-determine the future task load of the early warning computing node within a preset future time window; and In response to the redefined future task load being less than the task load warning threshold, the inbound bandwidth weight of the warning computing node is gradually increased based on a preset recovery step size and observation period.

9. The method according to claim 8, wherein, The inbound bandwidth weight of the early warning computing power node is gradually increased based on a preset recovery step size and observation period, including: Perform a second operation, the second operation including: The incoming bandwidth weight of the early warning computing node is increased based on the preset recovery step size; Within the observation period, determine the current computing power resource utilization rate of the early warning computing power node and determine at least one of the current task load of the early warning computing power node and the future task load of the early warning computing power node within a preset future time window; In response to at least one of them being less than the task load warning threshold, the second operation continues; or In response to any one of the above-mentioned factors being not less than the task load warning threshold, the incoming bandwidth weight of the warning computing node is rolled back by the preset recovery step size and the second operation is continued after waiting for a preset waiting period.

10. The method according to claim 9, wherein, The future task load and the current task load are determined by a prediction model built on a lightweight time series prediction architecture.

11. The method according to claim 1, wherein, The characteristic tags of the task to be processed include at least one of a due date tag and a priority tag; and / or The computing resources include at least one of the following: central processing unit computing power, graphics processing unit computing power, memory, and throughput capacity of the computing network.

12. The method according to claim 1, wherein, Migrating the emergency task to the target computing node includes: The emergency task and its context state are migrated to the target computing node.

13. The method according to claim 1, wherein, In response to the fact that the future task load of at least one of the plurality of computing power nodes is not less than the task load warning threshold, the at least one computing power node is designated as a warning computing power node, including: In response to the fact that the future task load of at least one of the plurality of computing power nodes is not less than the task load warning threshold in a series of consecutive time windows in the preset future time window, the at least one computing power node is designated as a warning computing power node.

14. A computing power scheduling device for a computing power network, comprising: processor; as well as A memory storing computer-executable instructions, which, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 13.

15. A computer-readable storage medium having computer-executable instructions stored thereon, the computer-executable instructions causing the processor to perform the method according to any one of claims 1 to 13 when executed by a processor.

16. A computer program product comprising instructions that, when executed by a processor, implement the method according to any one of claims 1 to 13.

17. A computing power network, comprising: The computing power network consists of a service layer, a control layer, a resource layer, and an orchestration layer. The orchestration layer includes a network security module and a computing power modeling module. in, The computing power modeling module is configured to determine the future task load of a computing power node within a preset future time window based on the historical computing power resource utilization rate of each computing power node among multiple computing power nodes in the computing power network. The computing power network security module is configured as follows: From the plurality of computing power nodes, select computing power nodes whose future task load is less than the task load warning threshold as candidate computing power nodes to construct a candidate computing power node set; In response to the fact that the future task load of at least one of the plurality of computing power nodes is not less than the task load warning threshold, the at least one computing power node is designated as a warning computing power node; For each early warning computing node, a first operation is performed, the first operation including: At least one urgent task is identified based on the characteristic tags of the tasks to be processed by the early warning computing node. For each of the at least one emergency task, calculate the comprehensive path weight for each candidate computing node in the set of candidate computing nodes. The comprehensive path weight indicates the cost of migrating the emergency task from the warning computing node to the candidate computing node. Select the candidate computing node with the smallest path comprehensive weight as the target computing node, and migrate the emergency task to the target computing node.

Images