A cloud computing platform computing power resource allocation method and system

Abstract

The application provides a cloud computing platform computing resource allocation method and system, which is suitable for the field of data processing technology. The method comprises the following steps: generating initial computing resource allocation decision information according to computing operation node running state information, computing operation node task information and a preset computing resource allocation decision information generation model; and generating target computing resource allocation decision information according to the initial computing resource allocation decision information, historical computing resource allocation decision information, a preset computing resource allocation level threshold information and a preset computing resource scheduling switching priority information. The application realizes accurate judgment and intelligent decision of the computing resource allocation state in the multi-node and multi-task concurrent scene of the cloud computing platform, automatically adapts different node load states and dynamically adjusts the scheduling strategy according to the business computing power demand, and is used for improving the resource allocation accuracy and scheduling effectiveness of the heterogeneous computing power cluster of the cloud computing platform.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data processing, and particularly relates to a cloud computing platform computing resource allocation method and system. BACKGROUND

[0002] With the continuous expansion of artificial intelligence, big data and scientific computing business scale, the types of tasks carried by the cloud computing platform tend to be complex and diverse, the number of heterogeneous computing nodes continues to grow, and computing resource allocation, as a core technology of the cloud computing platform, directly determines the cluster resource utilization, task execution delay and business service quality.

[0003] In the prior art, the running state of the computing nodes CPU, memory and the like is usually collected, the priority parameters of the tasks to be scheduled are obtained, the computing resource allocation decision is output through a fixed weight rule, the task admission, preemption and exit processing are completed according to the priority of the tasks, the computing nodes that can carry the tasks are selected in combination with the real-time load indicators of the nodes, the single computing resource allocation result is output, and the scheduling and execution of the computing resources of the cloud platform are completed. However, in the prior art, the accuracy of the computing resource allocation decision is insufficient, and resource mismatch and scheduling shock phenomenon are prone to occur. SUMMARY

[0004] Therefore, the embodiments of the present application provide a cloud computing platform computing resource allocation method and system, aiming at solving the problem of poor accuracy of computing resource allocation decision and easy occurrence of resource mismatch and scheduling shock in the prior art.

[0005] The first aspect of the embodiments of the present application provides a cloud computing platform computing resource allocation method, comprising:

[0006] obtaining running state information of a plurality of computing nodes, task information of a plurality of computing nodes and a plurality of historical computing resource allocation decision information;

[0007] generating initial computing resource allocation decision information according to the running state information of the plurality of computing nodes, the task information of the plurality of computing nodes and a preset computing resource allocation decision information generation model;

[0008] generating target computing resource allocation decision information according to the initial computing resource allocation decision information, the plurality of historical computing resource allocation decision information, a plurality of preset computing resource allocation level threshold information and preset computing scheduling switching priority information.

[0009] The second aspect of the embodiments of the present application provides a cloud computing platform computing resource allocation system, comprising:

[0010] an information acquisition module configured to acquire running state information of a plurality of computing nodes, task information of a plurality of computing nodes and a plurality of historical computing resource allocation decision information;

[0011] An initial computing power resource allocation decision information generation module is configured to generate initial computing power resource allocation decision information according to the plurality of computing power operation node running state information, the plurality of computing power operation node task information, and a preset computing power resource allocation decision information generation model.

[0012] A target computing power resource allocation decision information generation module is configured to generate target computing power resource allocation decision information according to the initial computing power resource allocation decision information, a plurality of historical computing power resource allocation decision information, a plurality of preset computing power resource allocation level threshold information, and preset computing power scheduling switching priority information.

[0013] A third aspect of the embodiment of the present application provides a terminal device, which comprises a memory and a processor, the memory stores a computer program capable of running on the processor, and the processor implements the steps of the cloud computing platform computing power resource allocation method according to the computer program.

[0014] A fourth aspect of the embodiment of the present application provides a computer readable storage medium, which comprises a computer program, and the computer program is executed by a processor to implement the steps of the cloud computing platform computing power resource allocation method according to the first aspect.

[0015] Compared with the prior art, the embodiment of the present application has the beneficial effects that: the embodiment of the present application realizes accurate judgment and intelligent decision of computing power resource allocation state in a multi-node and multi-task concurrent scenario of a cloud computing platform, automatically adapts different node load states and dynamically adjusts a scheduling strategy according to business computing power demand, is used to improve the resource allocation accuracy and scheduling effectiveness of a heterogeneous computing power cluster of the cloud computing platform, realizes maximum utilization of platform computing power resources, guarantees efficient and stable execution of various computing power business tasks, realizes fine and intelligent control of the cloud computing power scheduling process, and greatly improves the overall quality and operation efficiency of cloud platform computing power services. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Figure 1 is an implementation flow diagram of the cloud computing platform computing power resource allocation method provided by the first embodiment of the present application;

[0018] Figure 2is an implementation flowchart of the cloud computing platform computing resource allocation method provided in Embodiment Two of the present application.

[0019] Figure 3 is an implementation flowchart of the cloud computing platform computing resource allocation method provided in Embodiment Three of the present application.

[0020] Figure 4 is an implementation flowchart of the cloud computing platform computing resource allocation method provided in Embodiment Four of the present application.

[0021] Figure 5 is an implementation flowchart of the cloud computing platform computing resource allocation method provided in Embodiment Five of the present application.

[0022] Figure 6 is an implementation flowchart of the cloud computing platform computing resource allocation method provided in Embodiment Six of the present application.

[0023] Figure 7 is an implementation flowchart of the cloud computing platform computing resource allocation method provided in Embodiment Seven of the present application.

[0024] Figure 8 is a structural schematic diagram of the cloud computing platform computing resource allocation system provided in the present application.

[0025] Figure 9 is a schematic diagram of the terminal device provided in the present application. DETAILED DESCRIPTION

[0026] In the following description, for the purpose of explanation and not limitation, specific details are set forth, such as particular system configurations, techniques, etc., in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

[0027] In order to illustrate the technical solutions described in the present application, the following will be described through specific embodiments.

[0028] Figure 1 An implementation flowchart of the cloud computing platform computing resource allocation method provided in Embodiment One of the present application is shown, and is described in detail as follows:

[0029] In step S101, the running state information of a plurality of computing nodes, the task information of a plurality of computing nodes, and a plurality of historical computing resource allocation decision information are acquired.

[0030] In this embodiment, the computing power operation node running state information can be full-dimensional hardware running and resource occupation data generated by each heterogeneous computing power node in the cloud computing cluster within a business running period. Specifically, it can be state data corresponding to load occupancy rate, memory usage rate, disk read / write rate, network throughput rate, and resource idle time of different computing power operation nodes within a unit collection period. It can be obtained through real-time collection by a cluster monitoring probe during the entire process of platform computing power scheduling and running. Specifically, it can be obtained by continuously collecting node running data within the complete execution time range of the computing power scheduling task. The computing power running node task information can be attribute parameter data corresponding to each to-be-executed and executing business task in the cloud computing platform concurrent scheduling scenario. Specifically, it is priority parameter, admission weight parameter, and exit weight parameter corresponding to multiple batches of computing power scheduling tasks, which can cover the basic attribute data of various computing power business tasks such as data reasoning, model training, and simulation calculation. It can be obtained by automatically reading task configuration parameters through a task analysis interface when the computing power task accesses the platform. Specifically, it can be classified and stored in order according to the time sequence of task access to form multiple sets of computing power running node task information in order. The historical computing power resource allocation decision information can be a set of computing power allocation decision results generated by the system computing power scheduling module in a complete computing power scheduling period, from the beginning of receiving business computing power demand to the final completion of computing power resource allocation scheduling, arranged in time sequence. It can be used to represent the scheduling matching and allocation execution process of the platform's past computing power resources, and is stored in the form of time sequence. It can be understood that the historical computing power resource allocation decision information can include multiple sets of historical computing power allocation decision data completed by the system before the current computing power resource allocation calculation is executed in the current computing power scheduling period, which can be used to iteratively verify the rationality and accuracy of the current scheduling decision.

[0031] In this embodiment, the cloud computing platform only executes one set of core computing power resource allocation logic in a single scheduling period. Before the current computing power task allocation decision is output and executed, the system does not access new computing power scheduling tasks to avoid multi-task scheduling conflicts.

[0032] In this embodiment, the computing power business tasks accepted by the platform can be various computing tasks supported by computing power resources in the field of cloud computing, which can include offline batch computing, online real-time reasoning, artificial intelligence model training, three-dimensional simulation, and other types. It can be understood that various computing power business tasks can construct a dedicated computing power task database based on computing power node running state data and task attribute data to realize intelligent and accurate allocation of platform computing power resources.

[0033] Step S102: Generate initial computing resource allocation decision information based on the running status information of the multiple computing power nodes, the task information of the multiple computing power nodes, and the preset computing power resource allocation decision information generation model.

[0034] In this embodiment, the preset computing power resource allocation decision information generation model can be pre-set manually. It can employ a ConvNeXt feature fusion deep learning model, capable of fusing, extracting, and classifying the characteristics of computing power node operation status and computing power task attributes. The process can begin by calculating the cross-computing power node task priority transition weight difference information using the task exit weight information of the first computing power node and the task admission weight information of the second computing power node. Then, using a pre-set cross-computing power node task priority transition weight difference threshold (set to 0.25), multiple sets of computing power node task information are parsed and processed to generate multiple processed computing power node task information. Next, task loading and unloading information corresponding to the multiple processed computing power node task information are extracted. Then, based on the task loading and unloading features, the multiple computing power node operation status information is filtered to generate multiple filtered computing power node operation status information. Finally, the filtered computing power node operation status information is compared with the processed computing power node operation status information. The node task information is fused and calculated to generate multiple computing power node task scheduling information. Then, using the pre-set benchmark computing power scheduling task parameters of the cloud computing platform, the computing power task scheduling matching degree is calculated to obtain the computing power running node task scheduling matching degree information. Subsequently, the scheduling matching status is judged by the pre-set computing power running node task scheduling matching degree threshold (with a value of 0.7). When the matching degree does not meet the threshold requirement, idle computing power allocation decision information is generated. When the matching degree meets the threshold requirement, the node load running status is judged by combining the pre-set computing power running node load duration threshold (with a value of 120s). Then, load buffer computing power resource allocation decision information or regular computing power scheduling decision information is generated, thus finally generating complete initial computing power resource allocation decision information.

[0035] Step S103: Generate target computing resource allocation decision information based on the initial computing resource allocation decision information, multiple historical computing resource allocation decision information, multiple preset computing resource allocation level threshold information, and preset computing power scheduling switching priority information.

[0036] In this embodiment, firstly, multiple historical computing resource allocation decision information are statistically analyzed to extract the resource occupancy fluctuation characteristics, allocation result effectiveness records, and scheduling switch execution logs of historical computing resource allocation. The initial computing resource allocation decision information is then compared with each preset computing resource allocation level threshold information to obtain the computing resource allocation level identifier corresponding to the initial allocation scheme. Next, combined with preset computing power scheduling switch priority information, the initial computing resource allocation decision information is constrained and corrected. Specifically, when the initial computing resource allocation decision information triggers a cross-level threshold switch, candidate allocation options are sorted and filtered according to scheduling switch priority. Furthermore, the historical execution stability of candidate allocation options is verified using historical computing resource allocation decision information, and candidate allocation options with a historical scheduling failure frequency exceeding a preset frequency threshold are eliminated. After completing threshold verification, priority sorting, and historical data verification, the corrected allocation scheme is output as the target computing resource allocation decision information.

[0037] The cloud computing platform computing resource allocation method provided in this application embodiment enables accurate judgment and intelligent decision-making on the computing resource allocation status in multi-node, multi-task concurrent scenarios of the cloud computing platform. It automatically adapts to different node load states and business computing power requirements to dynamically adjust scheduling strategies, thereby improving the accuracy of resource allocation and scheduling effectiveness of heterogeneous computing power clusters of the cloud computing platform. This maximizes the utilization of platform computing resources, ensures the efficient and stable execution of various computing power business tasks, and enables refined and intelligent management and control of the cloud computing power scheduling process, significantly improving the overall quality and operating efficiency of cloud platform computing power services.

[0038] Figure 2 The flowchart illustrating the implementation of the cloud computing platform computing resource allocation method provided in Embodiment 2 of this application is shown. The difference between this method and Embodiment 1 described above is that:

[0039] The task information of the multiple computing power operation nodes includes the task information of the first computing power operation node and the task information of the second computing power operation node;

[0040] The task information of the first computing power operation node includes the task priority information, the task admission weight information, and the task exit weight information of the first computing power operation node;

[0041] The second computing power operation node task information includes the second computing power operation node task priority information, the second computing power operation node task admission weight information, and the second computing power operation node task exit weight information;

[0042] Wherein, the task priority information of the first computing power node is lower than the task priority information of the second computing power node;

[0043] Step S102 specifically includes:

[0044] Step S201: Based on the task exit weight information of the first computing power node and the task admission weight information of the second computing power node, calculate the cross-computing power node task priority transition weight difference information.

[0045] In this embodiment, the first computing power running node task information and the second computing power running node task information can be two sets of computing power service task parameter information with different priorities that are accessed within a single computing power scheduling cycle. That is, two sets of node task data are accessed between the start time and the end time of the computing power scheduling task. It is understood that there are at least two sets of computing power running node task information within a single computing power scheduling cycle; this embodiment only lists two sets of computing power running node task information for illustrative purposes. It is understood that the first computing power running node task priority information can be used to represent the execution priority of the first set of computing power tasks within a single computing power scheduling cycle, and the second computing power running node task priority information can be used to represent the execution priority of the second set of computing power tasks within the same computing power scheduling cycle. The fact that the first computing power running node task priority information is lower than the second computing power running node task priority information can be used to indicate that the second set of computing power tasks has the authority to be scheduled and executed first. The cross-computing power node task priority transition weight difference information refers to the numerical difference between the task exit weight information of the first computing power node and the task admission weight information of the second computing power node. It can be used to reflect the degree of scheduling transition adaptation between high and low priority tasks, and then it can be used to determine whether the two sets of computing power tasks have the basic conditions for continuous scheduling and alternating execution.

[0046] Step S202: Based on the cross-computing power node task priority overweight difference information and the preset cross-computing power node task priority overweight difference threshold, the task information of the first computing power node and the task information of the second computing power node are parsed and processed to generate multiple processed computing power node task information.

[0047] In this embodiment, the preset threshold for the cross-computing power node task priority transition weight difference can be manually preset, with a value of 0.25. This threshold can be used to distinguish the scheduling transition adaptation levels of two sets of computing power tasks and determine whether high-priority and low-priority tasks can be directly connected for scheduling. The calculated cross-computing power node task priority transition weight difference information can be compared with the preset threshold. Based on the comparison result, the task information of the first and second computing power nodes is differentiated, analyzed, and normalized to correct the admission and exit weight parameters of the two sets of tasks, thereby generating multiple processed computing power node task information adapted to the cluster scheduling rules.

[0048] Step S203: Generate initial computing resource allocation decision information based on the running status information of the multiple computing power nodes, the task information of the multiple processed computing power nodes, and the preset computing power resource allocation decision information generation model.

[0049] In this embodiment, the preset computing resource allocation decision information generation model can be manually preset and can adopt the ConvNeXt feature fusion deep learning model, which can achieve deep fusion and classification of node running status features and task regularization features. It can perform feature matching between the regularized task information of multiple processed computing nodes and the real-time collected running status information of multiple computing nodes, thereby completing the accurate screening of task loading and unloading features. Then, the model completes multi-dimensional feature fusion and scheduling matching degree calculation, and then combines preset multi-level scheduling thresholds to complete the hierarchical determination of idle, load buffer, and normal scheduling scenarios, thereby generating accurate and reliable initial computing resource allocation decision information.

[0050] The cloud computing platform computing resource allocation method provided in this application optimizes the task preprocessing logic in multi-task concurrent scenarios, improves the accuracy of subsequent computing power feature matching and decision generation, thereby enhancing the orderliness and stability of multi-task computing power scheduling of the cloud computing platform and ensuring the smooth and efficient scheduling and execution of multiple batches of computing power tasks.

[0051] Figure 3 The flowchart illustrating the implementation of the cloud computing platform computing resource allocation method provided in Embodiment 3 of this application is shown. The difference between this method and Embodiment 2 is that step S203 specifically includes:

[0052] Step S301: Based on the multiple processed computing power running node task information, extract multiple computing power running node task loading information and multiple computing power running node task unloading information.

[0053] In this embodiment, the task information of multiple processed computing power nodes consists of standardized task parameter data after weight parsing and normalization. Each group of task information corresponds to independent task loading requirements and resource unloading rules. It is possible to traverse all processed computing power node task information and then extract the core parameters corresponding to each group of tasks, such as resource loading level, loading time period, unloading conditions, and unloading priority. This generates structured and standardized task loading and unloading information for multiple computing power nodes, providing accurate task feature basis for subsequent node status screening.

[0054] Step S302: Based on the task loading information and task unloading information of the multiple computing power running nodes, the running status information of the multiple computing power running nodes is filtered to obtain multiple filtered running status information of the computing power running nodes.

[0055] In this embodiment, the extracted task loading information and task unloading information of the computing power running nodes can be matched and compared one by one with the real-time load, resource balance, idle time period and other operating status parameters of each computing power running node. Then, computing power nodes with resource balance that meets the task loading requirements and operating status that matches the task unloading rules are selected, and abnormal nodes with excessive load, insufficient resources and inability to meet task scheduling requirements are eliminated. This generates multiple sets of filtered computing power running node operating status information with higher matching degree, so as to achieve accurate matching between task characteristics and node status.

[0056] Step S303: Generate initial computing resource allocation decision information based on the running status information of the multiple filtered computing power nodes, the task information of the multiple processed computing power nodes, and the preset computing power resource allocation decision information generation model.

[0057] In this embodiment, the preset computing power resource allocation decision information generation model can be manually preset and can adopt the ConvNeXt feature fusion deep learning model. It can involve fusing the running status information of multiple filtered computing power nodes after matching and filtering with the regularized task information of multiple processed computing power nodes to calculate accurate node task scheduling matching information. Then, using a manually preset computing power node task scheduling matching degree threshold of 0.7 and a computing power node load duration threshold of 120s, hierarchical status discrimination is completed, thereby distinguishing and generating idle computing power allocation decision information, load buffer computing power resource allocation decision information, and regular computing power scheduling decision information, thus generating complete and accurate initial computing power resource allocation decision information.

[0058] The cloud computing platform computing power resource allocation method provided in this application embodiment achieves accurate extraction of computing power task loading and unloading features and targeted screening of node running status, enabling accurate matching between computing power node resource status and business task requirements, avoiding interference from invalid node data on scheduling decisions, improving the generation accuracy of initial computing power resource allocation decision information, thereby effectively improving the problems of cloud platform computing power resource mismatch and scheduling redundancy, and improving the resource utilization and overall scheduling efficiency of heterogeneous computing power clusters.

[0059] Figure 4 The flowchart illustrating the implementation of the cloud computing platform computing resource allocation method provided in Embodiment 4 of this application is shown. The difference between this method and Embodiment 1 above is that step S102 specifically includes:

[0060] Step S401: Based on the running status information of the multiple computing power nodes and the task information of the multiple computing power nodes, perform fusion calculation to obtain the task scheduling information of the multiple computing power nodes.

[0061] In this embodiment, the task information of each computing power running node and the corresponding computing power operation node running status information can be fused and calculated according to the order of task access within the computing power scheduling cycle, so as to use the fused feature data as the task scheduling information of multiple computing power operation nodes.

[0062] In this embodiment, the running status information of the computing nodes in the m-th task scheduling window within the computing power scheduling period can be determined first. Then, the cluster monitoring probe interface is called to obtain the full-dimensional resource usage data corresponding to the running status information of the computing nodes in real time. This allows the determination of the task information of the computing nodes in the m-th task scheduling window within the computing power scheduling period. Finally, the task parsing interface is called to read the priority parameters, admission weight parameters, and exit weight parameters corresponding to the task information of the computing nodes in real time. Then, the multi-dimensional feature data corresponding to the running status information and the task information of the computing nodes can be fused to obtain the task scheduling information of the computing nodes in the m-th task scheduling window.

[0063] Step S402: Calculate the matching degree based on the task scheduling information of the multiple computing power nodes and the preset benchmark computing power scheduling task parameter information of the cloud computing platform to obtain the task scheduling matching degree information of the computing power running nodes.

[0064] In this embodiment, the preset benchmark computing power scheduling task parameter information of the cloud computing platform can be manually preset. It can refer to the standard computing power resource requirement parameters pre-defined by the cloud computing platform, such as the standard task resource occupancy level, standard task runtime, standard node resource reserve threshold, etc., which can be used to indicate the benchmark scheduling reference conditions that need to be clarified before carrying out specific computing power allocation. The matching degree calculation can refer to calculating the matching degree between the feature data corresponding to the computing power operation node task scheduling information and the preset benchmark computing power scheduling task parameter information of the cloud computing platform. The calculation result is used as the computing power operation node task scheduling matching degree information.

[0065] In this embodiment, it is understood that matching degree calculation cannot be performed in scenarios with completely unavailable nodes. It is possible to first exclude scenarios with completely unavailable nodes before performing matching degree calculation, in order to further identify the unavailable scheduling state or other computing power scheduling states.

[0066] Step S403: Determine whether the task scheduling matching degree information of the computing power running node is less than the preset task scheduling matching degree threshold of the computing power running node; if yes, proceed to step S404; if no, proceed to step S405.

[0067] In this embodiment, the preset task scheduling matching degree threshold for computing power running nodes can be preset manually, with a value of 0.7. By judging the numerical relationship between the task scheduling matching degree information of computing power running nodes and the preset task scheduling matching degree threshold, it can be used to accurately determine whether the task scheduling information of computing power running nodes meets the baseline scheduling conditions. It can also be used to determine whether the task scheduling information of computing power running nodes revolves around the preset baseline computing power scheduling task parameter information of the cloud computing platform, and thus determine whether the current node and task combination has entered the effective computing power resource allocation stage.

[0068] In this embodiment, the initial value of the number of idle scheduling segments can be manually set to 0 to achieve recording of the number of idle scheduling segments. The duration of the initial computing resource allocation decision information corresponding to the m-th task scheduling window in the computing power scheduling cycle can be recorded and stored.

[0069] Step S404: Generate idle computing power allocation decision information as initial computing power resource allocation decision information.

[0070] In this embodiment, when the task scheduling matching degree information of the computing power running node is less than the preset task scheduling matching degree threshold of the computing power running node, it indicates that the task scheduling information of the computing power running node does not meet the scheduling conditions of the preset cloud computing platform benchmark computing power scheduling task parameter information. That is, the resource conditions of the current computing power running node cannot undertake the corresponding computing power running node task, thereby generating idle computing power allocation decision information as the initial computing power resource allocation decision information.

[0071] Step S405: Generate initial computing resource allocation decision information based on the task scheduling information of the multiple computing power nodes, the preset load duration threshold of the computing power running nodes, and the preset computing power resource allocation decision information generation model.

[0072] In this embodiment, if the task scheduling matching degree information of the computing power running node is greater than or equal to the preset task scheduling matching degree threshold of the computing power running node, it indicates that the combination of the current computing power computing node and the computing power running node task meets the baseline scheduling conditions. Then, the task scheduling information of multiple computing power computing nodes can be input into the preset computing power resource allocation decision information generation model to extract the multi-dimensional features of the computing power computing node task scheduling information and perform scheduling status classification. Thus, the computing power resource allocation decision information generation model outputs one type of scheduling information, either load buffer computing power resource allocation decision information or regular computing power scheduling decision information, as the current initial computing power resource allocation decision information.

[0073] In this embodiment, the preset computing resource allocation decision information generation model can be pre-set manually. Specifically, this model is a ConvNeXt feature fusion deep learning model, which can be trained using a manually collected computing resource scheduling dataset. First, task scheduling information from multiple computing nodes can be input into the preset computing resource allocation decision information generation model. Then, the ConvNeXt feature fusion deep learning model extracts feature vectors from the task scheduling information of each computing node, performs fusion processing of multiple sets of features, and outputs the predicted probabilities of various scheduling results (idle, load buffer, regular scheduling) corresponding to the task scheduling information of each computing node. Simultaneously, it correlates the computing resource allocation decision information before each task scheduling window for comprehensive judgment, thereby generating multiple initial computing resource allocation decision information that reflects the scheduling stage of each task scheduling window.

[0074] In this embodiment, when m is not equal to 1, the initial computing resource allocation decision information of the platform in the m-th group of task scheduling windows can be determined based on the task scheduling information of the computing power nodes through a preset computing resource allocation decision information generation model. This can be combined with obtaining the initial computing resource allocation decision information of the platform in the (m-1)-th group of task scheduling windows. It can be understood that the (m-1)-th group of task scheduling windows here refers to the task scheduling window information corresponding to the last scheduling decision information among multiple historical computing resource allocation decision information; that is, it is neither the task scheduling window information corresponding to idle computing power allocation decision information nor the task scheduling window information corresponding to invalid scheduling records.

[0075] The cloud computing platform computing resource allocation method provided in this application accurately identifies the idle scheduling state, effectively avoids misjudgment in the computing resource allocation process, improves the accuracy of generating initial computing resource allocation decision information, helps optimize the matching and processing process between heterogeneous computing nodes and business tasks of the cloud computing platform, and improves the stability of computing power scheduling and cluster resource utilization of the cloud computing platform.

[0076] Figure 5 The flowchart illustrating the implementation of the cloud computing platform computing resource allocation method provided in Embodiment 5 of this application is shown. The difference between this method and Embodiment 4 above is that step S405 specifically includes:

[0077] Step S501: Calculate the load duration information of the multiple computing nodes based on the task scheduling information of the multiple computing nodes.

[0078] In this embodiment, it can be understood that each computing power node task scheduling information corresponds to a unique node load start time information and node load end time information. The computing power node load duration information can be obtained by subtracting the node load end time information corresponding to the previous computing power node task scheduling information from the node load start time information of each computing power node task scheduling information.

[0079] Step S502: Determine whether the load duration information of the computing power operation node is greater than the preset load duration threshold of the computing power operation node; if yes, proceed to step S503; if no, proceed to step S504.

[0080] In this embodiment, the preset computing power running node load duration threshold can be preset by humans and is set to 120 seconds. It can be used to distinguish whether the computing power running node load duration information corresponding to the continuous occupation of resources to carry out task operation exceeds the normal duration limit, and thus determine whether the computing power running node is in the load buffer scheduling state.

[0081] Step S503: Generate load buffer computing power resource allocation decision information as initial computing power resource allocation decision information.

[0082] In this embodiment, when the load duration information of the computing power node is greater than the preset load duration threshold of the computing power node, it indicates that the computing power node has been running under high load for too long and needs to enter the load buffer operation stage. It cannot directly execute the full preemption scheduling of new tasks, thereby generating load buffer computing power resource allocation decision information as initial computing power resource allocation decision information, which can be used to mark the buffer control status of the current high load operation of the computing power node.

[0083] Step S504: Generate initial computing resource allocation decision information based on the task scheduling information of the multiple computing power nodes and the preset computing power resource allocation decision information generation model.

[0084] In this embodiment, if the load duration information of the computing power node is less than or equal to the preset load duration threshold of the computing power node, it indicates that the continuous load duration of the computing power node is within the normal range, and the node has the operating conditions to directly undertake computing power task scheduling. Then, the task scheduling information of multiple computing power nodes can be input into the preset computing power resource allocation decision information generation model to extract the features of the computing power node task scheduling information and perform scheduling status classification. Thus, the computing power resource allocation decision information generation model outputs conventional computing power scheduling decision information as the user's current initial computing power resource allocation decision information.

[0085] In this embodiment, the preset computing resource allocation decision information generation model can be manually preset. Specifically, this model is a ConvNeXt feature fusion deep learning model, which can be trained using a manually collected computing resource scheduling dataset. First, task scheduling information from multiple computing nodes can be input into the preset computing resource allocation decision information generation model. Then, the ConvNeXt feature fusion deep learning model extracts feature vectors from the task scheduling information of each computing node, performs fusion processing of multiple sets of features, and outputs the predicted probability of the conventional computing resource scheduling decision result corresponding to the task scheduling information of each computing node. Simultaneously, it correlates the computing resource scheduling decision information before each task scheduling window for comprehensive judgment, thereby generating multiple initial computing resource allocation decision information that reflect the scheduling stages of each task scheduling window.

[0086] The cloud computing platform computing resource allocation method provided in this application accurately identifies the load buffer scheduling state, distinguishes between load buffer scheduling scenarios and regular computing resource scheduling scenarios, reduces the probability of misjudgment in the computing resource allocation process, improves the generation accuracy of initial computing resource allocation decision information, helps optimize the load management logic of heterogeneous computing nodes, reduces the occurrence of computing node overload and scheduling oscillation, and ensures the smooth and orderly execution of various computing business tasks.

[0087] Figure 6 The flowchart illustrating the implementation of the cloud computing platform computing resource allocation method provided in Embodiment Six of this application is shown. Its difference from Embodiment Five described above lies in:

[0088] Multiple preset computing resource allocation level thresholds include multiple preset idle computing resource allocation level thresholds, multiple preset load buffer computing resource allocation level thresholds, and multiple preset dynamic computing resource allocation level thresholds.

[0089] Step S103 specifically includes:

[0090] Step S601: Determine whether the initial computing power resource allocation decision information is idle computing power allocation decision information; if yes, proceed to step S602; if no, proceed to step S604.

[0091] In this embodiment, by determining whether the initial computing power resource allocation decision information is idle computing power allocation decision information, it is ensured that the target computing power resource allocation decision information generated subsequently can accurately adapt to the current idle operation of the computing power nodes.

[0092] Step S602: Calculate the idle time information of the computing nodes based on the task scheduling information of multiple computing nodes corresponding to the initial computing resource allocation decision information.

[0093] In this embodiment, the initial computing power resource allocation decision information is the idle computing power allocation decision information. By extracting the node idle start time information and node idle end time information from the computing power computing node task scheduling information corresponding to the initial computing power resource allocation decision information, and subtracting the corresponding node idle start time information from the node idle end time information corresponding to the idle scheduling, the current computing power computing node idle duration information is obtained.

[0094] Step S603: Generate target computing resource allocation decision information based on the idle time information of the computing power node and multiple preset idle computing power resource allocation level threshold information.

[0095] In this embodiment, the multiple preset threshold information for the allocation level of idle computing power resources can be preset manually. These thresholds may include normal thresholds, observation thresholds, abnormal thresholds, and thresholds for the duration of idle computing power status, as well as threshold information for the number of times idle scheduling segments occur. These thresholds can be used to distinguish different operating conditions of idle computing power scheduling.

[0096] In this embodiment, specifically, the threshold for the duration of idle time under observation conditions can be set to 90 seconds, and the threshold for the duration of idle time under abnormal conditions can be set to 180 seconds. The number of times idle scheduling segments are identified is statistically recorded. Specifically, when both the number of idle scheduling segment identifications is less than or equal to 2 and the idle time information of the computing power node is less than or equal to 90 seconds, the computing power allocation status under idle scheduling is determined to be normal. When either the number of idle scheduling segment identifications is equal to 3 or the idle time information of the computing power node is greater than 90 seconds and less than or equal to 180 seconds, the computing power allocation status under idle scheduling is determined to be observation. When either the number of idle scheduling segment identifications is greater than 3 or the idle time information of the computing power node is greater than 180 seconds, the computing power allocation status under idle scheduling is determined to be abnormal, i.e., different target computing power resource allocation decision information is generated.

[0097] Step S604: Determine whether the initial computing power resource allocation decision information is load buffer computing power resource allocation decision information; if yes, proceed to step S605; if no, proceed to step S606.

[0098] In this embodiment, after excluding idle computing power allocation decision information, the type of initial computing power resource allocation decision information is further determined to ensure the pertinence of the generation of target computing power resource allocation decision information, which can be used to match specific computing power load control execution strategies.

[0099] Step S605: Generate target computing resource allocation decision information based on the load duration information of the multiple computing power nodes and the multiple preset load buffer computing power resource allocation level threshold information.

[0100] In this embodiment, multiple preset load buffer computing power resource allocation level thresholds can be manually preset and may include normal thresholds, observation thresholds, and abnormal thresholds for the duration of the load buffer state. These thresholds can be used to define different specific situations of load buffer computing power scheduling. The load duration information of multiple computing nodes can be compared with these thresholds, where the normal threshold is set to 120 seconds, the observation threshold to 240 seconds, and the abnormal threshold to 360 seconds. If the load duration information of the computing node is within the normal threshold range, a normal load buffer state is generated as the target computing power resource allocation decision information. If the load duration information of the computing node is between the normal threshold and the observation threshold, an observation load buffer state is generated as the target computing power resource allocation decision information. If the load duration information of the computing node exceeds the abnormal threshold, an abnormal load buffer state is generated as the target computing power resource allocation decision information, thereby achieving accurate determination of the load buffer scheduling state.

[0101] Step S606: Generate target computing resource allocation decision information based on the initial computing resource allocation decision information, multiple historical computing resource allocation decision information, multiple preset dynamic computing resource allocation level threshold information, and preset computing power scheduling switching priority information.

[0102] In this embodiment, the multiple preset computing resource allocation level thresholds can be manually preset and may include multiple preset idle computing resource allocation level thresholds, multiple preset load buffer computing resource allocation level thresholds, and multiple preset dynamic computing resource allocation level thresholds. The multiple preset idle computing resource allocation level thresholds can be manually preset and may include a steady-state adaptation threshold, a verification control threshold, an intervention trigger threshold, and a threshold for the number of occurrences of idle scheduling segments for the duration of the idle computing power state. The verification control threshold for the duration of the idle computing power state is set to 90 seconds, the intervention trigger threshold for the duration of the idle computing power state is set to 180 seconds, and the threshold for the number of occurrences of idle scheduling segments includes occurrences 2 and 3. The multiple preset load buffer computing resource allocation level thresholds can be manually preset and may include a steady-state adaptation threshold, a verification control threshold, and an intervention trigger threshold for the duration of the load buffer state. The steady-state adaptation threshold for the duration of the load buffer state is set to 120 seconds, the verification control threshold for the duration of the load buffer state is set to 240 seconds, and the intervention trigger threshold for the duration of the load buffer state is set to 360 seconds. Multiple preset dynamic computing resource allocation level thresholds can be manually set, and may include thresholds for computing power state iteration magnitude, steady-state repetition iteration count, state backtracking offset, computing power node state completeness, computing power scheduling sequence disorder, and computing power state iteration time length. Preset computing power scheduling switching priority information can also be manually set, and the criteria for determining four types of computing power scheduling state change types—forward state iteration, cross-level state iteration, steady-state state iteration, and backtracking state iteration—can be set according to the logical sequence of idle scheduling, load buffer scheduling, and regular computing power scheduling. Then, a complete computing power scheduling state sequence can be generated by combining the initial computing power resource allocation decision information with multiple historical computing power resource allocation decision information. The initial computing power resource allocation decision information is then compared with the historical computing power resource allocation decision information. The current computing power scheduling state change type is determined by the preset computing power scheduling switching priority information. The current computing power scheduling state change type is one of four types: state forward iteration, state cross-level iteration, state steady-state iteration, and state backtracking iteration. The computing power scheduling state change, sequence composition structure, and computing power state iteration time sequence length are quantitatively determined by multiple preset dynamic computing power resource allocation level threshold information. Then, the scheduling state that is suitable for the current computing power scheduling stage is analyzed based on the quantitative determination results. That is, the steady-state adaptation, the control status to be verified, and the intervention trigger status under the four computing power scheduling state changes of state forward iteration, state cross-level iteration, state steady-state iteration, and state backtracking iteration, and then the target computing power resource allocation decision information is generated.

[0103] In this embodiment, the pre-defined computing power scheduling switching priority information can include the location indexes of different computing power scheduling states, namely, the location indexes of idle scheduling, load buffer scheduling, and regular computing power scheduling. Then, based on the initial computing power resource allocation decision information of the m-th task scheduling window and the initial computing power resource allocation decision information of the m-1-th task scheduling window, the change in computing power scheduling state can be determined, that is, four types of computing power scheduling state changes can be identified: forward iteration, cross-level iteration, steady-state iteration, and backtracking iteration. Then, based on the duration, the forward iteration can be divided into three types: steady-state adaptation, pending verification and control, and intervention trigger. Then, based on the statistical results, cross-level iteration can be divided into three types: steady-state adaptation, pending verification and control, and intervention trigger. Then, based on the statistical results, steady-state iteration can be divided into three types: steady-state adaptation, pending verification and control, and intervention trigger. Then, based on the statistical results, backtracking iteration can be divided into three types: steady-state adaptation, pending verification and control, and intervention trigger. It is understood that the target computing power resource allocation decision information in this embodiment refers only to the target computing power resource allocation decision information corresponding to the initial computing power resource allocation decision information of the regular computing power scheduling, and does not include the case where the initial computing power resource allocation decision information is the idle computing power allocation decision information or the load buffer computing power resource allocation decision information.

[0104] In this embodiment, the computing power scheduling status identification sequence of the m-th task scheduling window can also be determined. When the length of the computing power scheduling status identification sequence is greater than 3, the composition structure of the sequence can be analyzed. The composition structure of the sequence can be analyzed through the completeness of the computing power node status and the disorder of the computing power scheduling sequence. Then, the composition structure of the computing power scheduling status sequence can be divided into three types: steady-state adaptation, control to be verified, and intervention triggering. When the length of the computing power scheduling status identification sequence is less than or equal to 3, the calculation of the completeness of the computing power node status and the disorder of the computing power scheduling sequence can be skipped. Then, the step of generating initial computing power resource allocation decision information based on the task scheduling information of the multiple computing power nodes and the preset computing power resource allocation decision information generation model can be returned to, thereby restarting the initial identification of idle scheduling, load buffer scheduling, and regular computing power scheduling.

[0105] The cloud computing platform computing power resource allocation method provided in this application embodiment can accurately distinguish and identify the idle computing power scheduling state, the load buffer computing power scheduling state, and the regular computing power scheduling process of the cloud computing platform, thereby improving the accuracy and pertinence of the target computing power resource allocation decision information. It can be used to provide reliable data support for the accurate matching of dynamic computing power load control strategies, thereby helping to optimize the resource allocation and management process of heterogeneous computing power nodes, and ensuring and improving the stability and resource utilization efficiency of cloud computing cluster computing power scheduling.

[0106] Figure 7 The flowchart illustrating the implementation of the cloud computing platform computing resource allocation method provided in Embodiment Seven of this application is shown. The difference between this method and Embodiment One is that, after step S103, the method further includes:

[0107] Step S701: Obtain computing power load demand information from multiple cloud computing platforms.

[0108] In this embodiment, the computing power load demand information of multiple cloud computing platforms can be the computing power resource application content submitted by each business task within the computing power scheduling cycle. It can include structured field information such as priority number, admission weight, exit weight, task resource occupation level, estimated task running time, and node resource reserve requirements corresponding to the computing power task. It can also include computing power load demand content information of different business types such as offline batch computing, online inference, model training, and simulation calculation. In this way, the computing power load demand information of cloud computing platforms can be directly retrieved through a manually constructed computing power task database.

[0109] Step S702: Generate multiple computing resource allocation optimization decision information based on the target computing resource allocation decision information, the computing load demand information of multiple cloud computing platforms, and multiple preset computing resource allocation optimization strategy information.

[0110] In this embodiment, multiple preset computing resource allocation optimization strategies can be manually preset and have a one-to-one matching relationship with multiple target computing resource allocation decisions. Each target computing resource allocation decision, namely, idle scheduling, load buffer scheduling, and steady-state adaptation, verification control, and intervention triggering under four computing power scheduling state changes (state forward iteration, state cross-level iteration, state steady-state iteration, and state backtracking iteration), can be matched with a specific preset computing resource allocation optimization strategy. Subsequently, multiple target computing resource allocation decisions can be matched with multiple preset computing resource allocation optimization strategies, and different preset computing resource allocation optimization strategies can correspond to different computing power control methods, that is, different computing resource allocation optimization decisions are generated based on the computing power load demand information of the cloud computing platform.

[0111] In this embodiment, it can first be determined whether the target computing power resource allocation decision information belongs to any of the following categories of scheduling state changes: idle, load buffer, forward state iteration, cross-level state iteration, steady-state state iteration, or backtracking state iteration. Then, it can be determined whether the scheduling state change is a steady-state adaptation, a control to be verified, or an intervention trigger. Then, it can achieve precise matching through multiple preset computing power resource allocation optimization strategy information. At the same time, it can combine the specific content in the computing power load demand information of multiple cloud computing platforms, such as the task resource occupancy level and the estimated task runtime, to perform detailed adaptation of the control method. In this way, the control method suitable for the current computing power scheduling scenario can be selected from multiple preset computing power resource allocation optimization strategy information, and then multiple computing power resource allocation optimization decision information can be generated.

[0112] In this embodiment, the multiple preset computing resource allocation optimization strategies can be preset manually and may include: a light adjustment strategy for idle state, a high adjustment strategy for idle state, a light adjustment strategy for load buffer, a high adjustment strategy for load buffer, a light adjustment strategy for forward state iteration, a high adjustment strategy for forward state iteration, a light adjustment strategy for cross-order state iteration, a high adjustment strategy for cross-order state iteration, a light adjustment strategy for steady-state state iteration, a high adjustment strategy for steady-state state iteration, a light adjustment strategy for backtracking state iteration, a high adjustment strategy for backtracking state iteration, a light adjustment strategy for scheduling sequence composition structure, and a high adjustment strategy for scheduling sequence composition structure. The control methods corresponding to the light control strategy for idle state include "the system marks the current computing power node as a pending task state" and "the system issues a scheduling prompt suggesting that the business task be connected to the current node"; the control methods corresponding to the heavy control strategy for idle state include "automatically rescanning the pending task queue and prioritizing the allocation of low-latency business tasks to the current node", "the system scheduling alarm prompts that the computing power node has been idle for a long time and prompts to replenish the scheduling task in time", and "the system interface displays the duration of the current node's idle state and reminds the current node of its idle time in text". The control methods corresponding to the light control strategy for load buffering include "highlighting the threshold alarm for the CPU and memory usage indicators of the current computing power node" and "generating corresponding scheduling guidance instructions based on the scheduling state before load buffering, such as using the guidance instruction 'Does it need to limit the resource quota for the newly connected task?' if it was in regular scheduling before buffering". The control methods corresponding to the heavy control strategy for load buffering include "retrieving the task scheduling records of other computing power nodes under the same load state from the historical scheduling database". The following are examples of the control methods for the positive iteration mild control strategy: "Record as reference" and "Add auxiliary load management by adjusting the resource frequency reduction and task sharding rate limiting of the current node"; the control methods corresponding to the positive iteration mild control strategy include "Displaying the current computing power scheduling status above the scheduling panel and indicating the next scheduling status to be entered" and "System text prompts that computing power scheduling is progressing smoothly to the subsequent status, such as prompting to move from load buffer to regular scheduling 'Try to remove node rate limiting and accept new business tasks'"; the control methods corresponding to the positive iteration high control strategy include "System text prompts that the scheduling progress is too fast and the node resources have not been fully released" and "System requires to maintain the current scheduling status, such as 'Continue to maintain load buffer control for 60 seconds'"; the control methods corresponding to the cross-level iteration mild control strategy include "System prompts that computing power scheduling cross-level iteration has occurred, displays the cross-level iteration situation, including the number, magnitude, previous scheduling status and current scheduling status, and explains which scheduling control steps were omitted" and "System text warnings that the scheduling cross-level magnitude is too large and suggests gradually completing the node status switch";The control methods corresponding to the state cross-order iteration high-level control strategy include "the system displays the scheduling control steps omitted in the cross-order iteration and asks the user to complete them, such as 'Please complete the node load buffer verification in 30 seconds'" and "the system provides guidance on the steps to be completed, such as guiding the load buffer verification 'You can consider verifying whether the node memory and network bandwidth margins meet the conditions for new task access'"; the control methods corresponding to the state steady-state iteration light-level control strategy include "the system warns that the current computing power scheduling state has entered an excessive steady-state iteration" and "the system displays a text guiding the current computing power scheduling forward, such as 'When the load...'" The buffer steady-state iteration state prompts the user, "Now try to lift the node rate limiting control based on the load statistics results"; the control methods corresponding to the state steady-state iteration high-level control strategy include "the system prompts that the computing power node has stayed in the current scheduling state and the steady-state iteration has been too long, and it needs to be immediately promoted to the next scheduling state, and explains the control tasks of the next scheduling state" and "the system displays two guiding text messages to help jump out of the steady-state iteration"; the control methods corresponding to the state backtracking iteration light-level control strategy include "the system provides text prompts to guide the computing power scheduling after the backtracking iteration, for example, when backtracking to the load buffer, prompting, "When resources are tight, you can start from..." The system assesses node load pressure based on dimensions such as task priority, resource quota, and concurrency. The highly controlled state backtracking iteration strategy includes methods such as "the system prompts that excessive backtracking iteration has occurred in the computing power scheduling, indicating the current cluster scheduling risk" and "the system suggests that the user stop backtracking iteration, continue to advance the previous scheduling steps, and provides two text instructions to move forward, such as 'It is recommended to continue regular computing power scheduling and select business tasks that match the available node resources for execution'." The lightly controlled scheduling sequence composition structure strategy includes methods such as "the system provides language prompts about supplementing missing scheduling control steps, such as 'Build...'". The suggestion includes supplementing the node load buffer verification / idle status inspection with the following: "Add a prompt below the scheduling panel indicating missing scheduling control steps, such as prompting 'Does the current node need to perform buffer rate limiting after high load verification?' when load buffer verification is missing." The scheduling sequence composition structure and the corresponding control strategies include: "Displaying various scheduling states such as idle scheduling, load buffer scheduling, and regular computing power scheduling, along with their corresponding definitions, and highlighting the currently missing scheduling control steps" and "Providing guidance on the control of the currently missing scheduling control steps, such as providing different evaluation dimensions to consider when performing load buffer verification on nodes."

[0113] In this embodiment, when the target computing power resource allocation decision information is a steady-state adaptation under idle computing power scheduling, no intervention is required. When the target computing power resource allocation decision information is a pending verification and control situation under idle computing power scheduling, a mild idle state control strategy can be adopted. When the target computing power resource allocation decision information is an intervention-triggered situation under idle computing power scheduling, a high-level idle state control strategy can be adopted. Similarly, when the target computing power resource allocation decision information is a steady-state adaptation under load-buffered computing power scheduling, no intervention is required. When the target computing power resource allocation decision information is a pending verification and control situation under load-buffered computing power scheduling, a mild load-buffered control strategy can be adopted. When the target computing power resource allocation decision information is an intervention-triggered situation under load-buffered computing power scheduling, a high-level load-buffered control strategy can be adopted. When the target computing resource allocation decision information is a steady-state adaptation under the state of positive iterative computing power scheduling, no intervention is needed. When the target computing resource allocation decision information is a situation requiring verification and control under the state of positive iterative computing power scheduling, a mild state-positive iterative control strategy can be adopted. When the target computing resource allocation decision information is an intervention-triggered situation under the state of positive iterative computing power scheduling, a highly controlled state-positive iterative strategy can be adopted. Similarly, when the target computing resource allocation decision information is a steady-state adaptation under the state of cross-order iterative computing power scheduling, no intervention is needed. When the target computing resource allocation decision information is a situation requiring verification and control under the state of cross-order iterative computing power scheduling, a mild state-cross-order iterative control strategy can be adopted. When the target computing resource allocation decision information is an intervention-triggered situation under the state of cross-order iterative computing power scheduling, a highly controlled state-cross-order iterative strategy can be adopted. When the target computing resource allocation decision information is a steady-state adaptation under a steady-state iterative computing power scheduling state, no intervention is needed. When the target computing resource allocation decision information is a control situation to be verified under a steady-state iterative computing power scheduling state, a mild steady-state iterative control strategy can be adopted. When the target computing resource allocation decision information is an intervention-triggered situation under a steady-state iterative computing power scheduling state, a highly controlled steady-state iterative strategy can be adopted. Similarly, when the target computing resource allocation decision information is a steady-state adaptation under a backtracking iterative computing power scheduling state, no intervention is needed. When the target computing resource allocation decision information is a control situation to be verified under a backtracking iterative computing power scheduling state, a mild backtracking iterative control strategy can be adopted. When the target computing resource allocation decision information is an intervention-triggered situation under a backtracking iterative computing power scheduling state, a highly controlled backtracking iterative strategy can be adopted.Furthermore, multiple preset computing resource allocation optimization strategies can be matched against the composition structure of the computing power scheduling status identification sequence. If the identification indicates a steady-state adaptation, no intervention is required; if it indicates a control-to-be-verified situation, a mild adjustment strategy based on the scheduling sequence composition structure can be adopted; and if it indicates an intervention-triggered situation, a highly controlled strategy based on the scheduling sequence composition structure can be adopted. Subsequently, the computing power task queue, node resource quota, and rate limiting parameters within the current computing power scheduling cycle can be modified, supplemented, or adjusted using the target computing power resource allocation decision information, thereby generating computing power resource allocation optimization decision information. However, such operations are all considered as control and optimization strategies for the same computing power scheduling cycle and do not need to be implemented as new computing power tasks.

[0114] The cloud computing platform computing power resource allocation method provided in this application improves the auxiliary optimization and control effect of the scheduling and allocation process of heterogeneous computing power nodes, effectively realizes the application path of artificial intelligence computing power scheduling technology, and thus efficiently optimizes the resource allocation and management process of cloud computing platform computing power nodes, significantly ensuring and improving the stability of cloud computing cluster computing power scheduling and overall resource utilization efficiency.

[0115] Corresponding to the method in the above embodiments, Figure 8 The diagram shows a structural block diagram of a cloud computing platform computing resource allocation system provided in an embodiment of this application. For ease of explanation, only the parts related to the embodiments of this application are shown. Figure 8 The example cloud computing platform computing resource allocation system can be the execution entity of the cloud computing platform computing resource allocation method provided in the aforementioned embodiment 1.

[0116] Reference Figure 8 The cloud computing platform's computing resource allocation system includes:

[0117] The information acquisition module 810 is used to acquire the running status information of multiple computing power nodes, the task information of multiple computing power nodes, and the historical computing power resource allocation decision information.

[0118] The initial computing power resource allocation decision information generation module 820 is used to generate initial computing power resource allocation decision information based on the running status information of the multiple computing power operation nodes, the task information of the multiple computing power operation nodes, and the preset computing power resource allocation decision information generation model.

[0119] The target computing resource allocation decision information generation module 830 is used to generate target computing resource allocation decision information based on the initial computing resource allocation decision information, multiple historical computing resource allocation decision information, multiple preset computing resource allocation level threshold information, and preset computing power scheduling switching priority information.

[0120] The process by which each module in the cloud computing platform computing resource allocation system provided in this application implements its respective function can be found in the foregoing. Figure 1 The description of Embodiment 1 shown will not be repeated here.

[0121] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0122] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0123] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0124] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0125] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance. It should also be understood that although the terms "first," "second," etc., are used in the text to describe various elements in some embodiments of this application, these elements should not be limited by these terms. These terms are merely used to distinguish one element from another. For example, a first table may be named a second table, and similarly, a second table may be named a first table, without departing from the scope of the various described embodiments. Both the first table and the second table are tables, but they are not the same table.

[0126] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0127] The cloud computing platform computing resource allocation method provided in this application embodiment can be applied to terminal devices such as mobile phones, tablets, wearable devices, in-vehicle devices, augmented reality / virtual reality devices, laptops, super mobile personal computers, netbooks, and personal digital assistants. This application embodiment does not impose any restrictions on the specific type of terminal device.

[0128] For example, the terminal device may be a station in a WLAN, a cellular phone, a cordless phone, a session initiation protocol phone, a wireless local loop station, a personal digital processing device, a handheld device with wireless communication capabilities, a computing device or other processing device connected to a wireless modem, an in-vehicle device, a vehicle networking terminal, a computer, a laptop computer, a handheld communication device, a handheld computing device, a satellite wireless device, a wireless modem card, a set-top box, a user premises equipment, and / or other devices for communication over a wireless system, as well as next-generation communication systems, such as mobile terminals in 5G networks or mobile terminals in future evolved public terrestrial mobile networks, etc.

[0129] Figure 9 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. For example... Figure 9 As shown, the terminal device 9 of this embodiment includes: at least one processor 90 ( Figure 9 (Only one is shown in the image), a memory 91, which stores a computer program 92 that can run on the processor 90. When the processor 90 executes the computer program 92, it implements the steps in the above-described embodiments of the cloud computing platform computing resource allocation methods, for example... Figure 1 Steps S101 to S103 are shown. Alternatively, when the processor 90 executes the computer program 92, it implements the functions of each module / unit in the above system embodiments, for example... Figure 8 The functions of modules 810 to 830 are shown.

[0130] The terminal device 9 can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor 90 and a memory 91. Those skilled in the art will understand that... Figure 9 This is merely an example of terminal device 9 and does not constitute a limitation on terminal device 9. It may include more or fewer components than shown, or combine certain components, or different components. For example, the terminal device may also include input transmission devices, network access devices, buses, etc.

[0131] The processor 90 may be a central processing unit, or it may be other general-purpose processors, digital signal processors, application-specific integrated circuits, off-the-shelf programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0132] In some embodiments, the memory 91 may be an internal storage unit of the terminal device 9, such as a hard disk or memory of the terminal device 9. The memory 91 may also be an external storage device of the terminal device 9, such as a plug-in hard disk, smart memory card, secure digital card, flash memory card, etc., equipped on the terminal device 9. Furthermore, the memory 91 may include both internal and external storage units of the terminal device 9. The memory 91 is used to store operating systems, applications, bootloaders, data, and other programs, such as the program code of computer programs. The memory 91 can also be used to temporarily store data that has been sent or will be sent.

[0133] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0134] This application also provides a terminal device, which includes at least one memory, at least one processor, and a computer program stored in the at least one memory and executable on the at least one processor. When the processor executes the computer program, it causes the terminal device to implement the steps in any of the above method embodiments.

[0135] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0136] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0137] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0138] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0139] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

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

[0141] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for allocating computing resources on a cloud computing platform, characterized in that, include: Acquire running status information of multiple computing power nodes, task information of multiple computing power nodes, and historical computing power resource allocation decision information; Based on the running status information of the multiple computing power nodes, the task information of the multiple computing power nodes, and the preset computing power resource allocation decision information generation model, initial computing power resource allocation decision information is generated. Based on the initial computing power resource allocation decision information, multiple historical computing power resource allocation decision information, multiple preset computing power resource allocation level threshold information, and preset computing power scheduling switching priority information, target computing power resource allocation decision information is generated.

2. The cloud computing platform computing resource allocation method as described in claim 1, characterized in that, The task information of the multiple computing power operation nodes includes the task information of the first computing power operation node and the task information of the second computing power operation node; The task information of the first computing power operation node includes the task priority information, the task admission weight information, and the task exit weight information of the first computing power operation node; The second computing power operation node task information includes the second computing power operation node task priority information, the second computing power operation node task admission weight information, and the second computing power operation node task exit weight information; Wherein, the task priority information of the first computing power node is lower than the task priority information of the second computing power node; The step of generating initial computing resource allocation decision information based on the running status information of the multiple computing power nodes, the task information of the multiple computing power nodes, and a preset computing resource allocation decision information generation model specifically includes: Based on the task exit weight information of the first computing power node and the task admission weight information of the second computing power node, the cross-computing power node task priority transition weight difference information is calculated. Based on the cross-computing power node task priority overweight difference information and the preset cross-computing power node task priority overweight difference threshold, the first computing power running node task information and the second computing power running node task information are parsed and processed to generate multiple processed computing power running node task information. Based on the running status information of the multiple computing power nodes, the task information of the multiple processed computing power nodes, and the preset computing power resource allocation decision information generation model, initial computing power resource allocation decision information is generated.

3. The cloud computing platform computing resource allocation method as described in claim 2, characterized in that, The step of generating initial computing resource allocation decision information based on the running status information of the multiple computing power nodes, the task information of the multiple processed computing power nodes, and a preset computing resource allocation decision information generation model specifically includes: Based on the multiple processed computing power running node task information, multiple computing power running node task loading information and multiple computing power running node task unloading information are extracted. Based on the task loading information and task unloading information of the multiple computing power running nodes, the running status information of the multiple computing power running nodes is filtered to obtain multiple filtered running status information of the computing power running nodes. Based on the running status information of multiple filtered computing power nodes, the task information of multiple processed computing power nodes, and the preset computing power resource allocation decision information generation model, initial computing power resource allocation decision information is generated.

4. The cloud computing platform computing resource allocation method as described in claim 1, characterized in that, The step of generating initial computing resource allocation decision information based on the running status information of the multiple computing power nodes, the task information of the multiple computing power nodes, and a preset computing resource allocation decision information generation model specifically includes: Based on the combined calculation of the running status information and task information of the multiple computing power nodes, the task scheduling information of the multiple computing power nodes is obtained. The matching degree is calculated based on the task scheduling information of the multiple computing power nodes and the preset benchmark computing power scheduling task parameter information of the cloud computing platform to obtain the matching degree information of the computing power running node task scheduling. Determine whether the task scheduling matching degree information of the computing power running node is less than the preset task scheduling matching degree threshold of the computing power running node; If so, then generate idle computing power allocation decision information as initial computing power resource allocation decision information; If not, then based on the task scheduling information of the multiple computing power nodes, the preset load duration threshold of the computing power running nodes, and the preset computing power resource allocation decision information generation model, the initial computing power resource allocation decision information is generated.

5. The cloud computing platform computing resource allocation method as described in claim 4, characterized in that, The step of generating initial computing resource allocation decision information based on the task scheduling information of the multiple computing power nodes, a preset computing power running node load duration threshold, and a preset computing power resource allocation decision information generation model specifically includes: Based on the task scheduling information of the multiple computing power nodes, the load duration information of the multiple computing power nodes is calculated; Determine whether the load duration information of the computing power operation node is greater than the preset load duration threshold of the computing power operation node; If so, then load buffer computing power resource allocation decision information is generated as initial computing power resource allocation decision information; If not, then based on the task scheduling information of the multiple computing power nodes and the preset computing power resource allocation decision information generation model, the initial computing power resource allocation decision information is generated.

6. The cloud computing platform computing resource allocation method as described in claim 5, characterized in that, Multiple preset computing resource allocation level thresholds include multiple preset idle computing resource allocation level thresholds, multiple preset load buffer computing resource allocation level thresholds, and multiple preset dynamic computing resource allocation level thresholds. The step of generating target computing resource allocation decision information based on the initial computing resource allocation decision information, multiple historical computing resource allocation decision information, multiple preset computing resource allocation level threshold information, and preset computing power scheduling switching priority information specifically includes: When the initial computing power resource allocation decision information is the idle computing power allocation decision information, the idle time information of the computing power computing nodes is calculated based on the task scheduling information of multiple computing power computing nodes corresponding to the initial computing power resource allocation decision information. Based on the idle time information of the computing power nodes and multiple preset idle computing power resource allocation level thresholds, target computing power resource allocation decision information is generated. When the initial computing power resource allocation decision information is the load buffer computing power resource allocation decision information, the target computing power resource allocation decision information is generated based on the load duration information of the multiple computing power nodes and the multiple preset load buffer computing power resource allocation level threshold information. When the initial computing power resource allocation decision information is not the idle computing power allocation decision information, and the initial computing power resource allocation decision information is not the load buffer computing power resource allocation decision information, the target computing power resource allocation decision information is generated based on the initial computing power resource allocation decision information, multiple historical computing power resource allocation decision information, multiple preset dynamic computing power resource allocation level threshold information, and preset computing power scheduling switching priority information.

7. The cloud computing platform computing resource allocation method as described in claim 1, characterized in that, After the step of generating target computing resource allocation decision information based on the initial computing resource allocation decision information, multiple historical computing resource allocation decision information, multiple preset computing resource allocation level threshold information, and preset computing power scheduling switching priority information, the method further includes: Obtain computing power load requirements from multiple cloud computing platforms; Based on the target computing power resource allocation decision information, the computing power load demand information of multiple cloud computing platforms, and the information of multiple preset computing power resource allocation optimization strategies, multiple computing power resource allocation optimization decision information are generated.

8. A cloud computing platform computing resource allocation system, characterized in that, include: The information acquisition module is used to acquire the running status information of multiple computing power nodes, the task information of multiple computing power nodes, and the historical computing power resource allocation decision information. The initial computing power resource allocation decision information generation module is used to generate initial computing power resource allocation decision information based on the running status information of the multiple computing power operation nodes, the task information of the multiple computing power operation nodes, and the preset computing power resource allocation decision information generation model. The target computing resource allocation decision information generation module is used to generate target computing resource allocation decision information based on the initial computing resource allocation decision information, multiple historical computing resource allocation decision information, multiple preset computing resource allocation level threshold information, and preset computing power scheduling switching priority information.

9. A terminal device, characterized in that, The terminal device includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images