Cloud resource scheduling method and device, computer equipment and storage medium

Abstract

The invention relates to a cloud resource scheduling method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring at least one device time sequence data of a target device; fusing the time sequence data of different devices to obtain fused time sequence data; determining scheduling reference data of the target equipment according to the fused time sequence data based on the trained demand prediction model; wherein the scheduling reference data comprises at least one of a resource demand value, a resource demand scene and scene confidence; according to the scheduling reference data, determining a target scheduling strategy needing to be executed in the resource demand scene; the target scheduling strategy is used for scheduling the cloud resources in the target equipment to meet the resource demand value. By adopting the method, the cloud resource scheduling accuracy can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cloud computing, and in particular to a cloud resource scheduling method and device, computer equipment and a storage medium. BACKGROUND

[0002] With the rapid development of cloud computing and big data technology, the types and scales of services carried by cloud platforms are increasingly complex and diverse, which puts forward higher requirements for the real-time performance, accuracy and self-adaptive ability of resource scheduling.

[0003] Traditional cloud resource scheduling methods mainly rely on historical performance time series data for prediction and decision-making, and achieve short-term prediction and dynamic adjustment of resource demand by constructing a time series model or introducing a reinforcement learning mechanism.

[0004] However, the traditional technology usually only predicts based on performance time series data when calling cloud resources, which has a single prediction angle, and thus the prediction result lacks a connection with the business, thereby causing the problem of poor rationality of cloud resource scheduling. SUMMARY

[0005] Therefore, it is necessary to provide a cloud resource scheduling method, device, computer equipment and storage medium capable of improving the accuracy of cloud resource scheduling to solve the above technical problems.

[0006] In a first aspect, the present application provides a cloud resource scheduling method applied to a target device, which comprises:

[0007] obtaining at least one device time series data of the target device;

[0008] fusing different device time series data to obtain fused time series data;

[0009] determining scheduling reference data of the target device based on the fused time series data according to a trained demand prediction model; wherein the scheduling reference data comprises at least one of a resource demand value, a resource demand scenario and a scenario confidence;

[0010] determining a target scheduling strategy to be executed under the resource demand scenario according to the scheduling reference data; the target scheduling strategy is used to schedule cloud resources in the target device to meet the resource demand value.

[0011] In one embodiment, determining the scheduling reference data of the target device based on the fused time series data according to the trained demand prediction model comprises:

[0012] performing linear transformation on the fused time series data based on a preset first weight matrix and a preset first bias vector to obtain the resource demand value of the target device; and

[0013] performing classification processing on the fused time-series data based on a preset classification function to determine a resource requirement scenario of the target device; and

[0014] performing linear transformation on the fused time-series data based on a preset second weight matrix and a preset second bias vector to obtain a scenario confidence of the target device; the scenario confidence comprises confidence of the target device being classified into different candidate requirement scenarios; the candidate requirement scenarios comprise the resource requirement scenario.

[0015] In one of the embodiments, the method further comprises:

[0016] performing normalization processing on the scenario confidence according to a number of scenarios of different candidate requirement scenarios to update the scenario confidence.

[0017] In one of the embodiments, the device time-series data comprises performance time-series data and non-performance time-series data; the fused time-series data is obtained by performing fusion processing on different device time-series data, comprising:

[0018] performing fusion processing on different device time-series data to obtain initial time-series data;

[0019] performing mean pooling processing on the initial time-series data, the log time-series data and the event time-series data to obtain the fused time-series data.

[0020] In one of the embodiments, the initial time-series data is obtained by performing fusion processing on different device time-series data, comprising:

[0021] performing fusion processing on the performance time-series data and the non-performance time-series data to obtain initial fused data; and performing fusion processing on the performance time-series data and the initial fused data to obtain the initial time-series data.

[0022] In one of the embodiments, the non-performance time-series data comprises log time-series data and event time-series data; the initial fused data is obtained by performing fusion processing on the performance time-series data and the non-performance time-series data, comprising:

[0023] performing fusion processing on the performance time-series data and the log time-series data to obtain first time-series data; and

[0024] performing fusion processing on the performance time-series data and the event time-series data to obtain second time-series data;

[0025] performing fusion processing on the first time-series data and the second time-series data to obtain the initial fused data.

[0026] In one of the embodiments, the target scheduling strategy to be executed under the resource requirement scenario is determined according to the scheduling reference data, comprising:

[0027] obtaining at least one resource scheduling strategy executable by the target device;

[0028] determining cloud resources required in executing the resource scheduling policy at the target device for each resource scheduling policy;

[0029] determining an over-standard rate of the resource scheduling policy according to the cloud resources, the resource demand value and the scenario confidence;

[0030] selecting a target scheduling policy required to be executed under the resource demand scenario from the resource scheduling policies according to the over-standard rate and an execution cost of the resource scheduling policy.

[0031] In one embodiment, determining the over-standard rate of the resource scheduling policy according to the cloud resources, the resource demand value and the scenario confidence comprises:

[0032] determining a resource error according to a difference between the cloud resources and the resource demand value;

[0033] determining the over-standard rate of the resource scheduling policy according to a proportional relationship between the resource error and the scenario confidence.

[0034] In one embodiment, selecting the target scheduling policy from the resource scheduling policies according to the over-standard rate and the execution cost of the resource scheduling policy comprises:

[0035] selecting a resource scheduling policy with an over-standard rate less than a preset over-standard threshold as a reference scheduling policy;

[0036] selecting the reference scheduling policy with the minimum execution cost as the target scheduling policy.

[0037] In one embodiment, the method further comprises:

[0038] determining a demand adjustment strategy matched with the resource demand scenario;

[0039] updating the resource demand value according to the demand adjustment strategy.

[0040] In one embodiment, the demand prediction model is obtained by training in the following manner:

[0041] obtaining sample fusion time series data, a standard resource demand value, a standard resource demand scenario and a standard scenario confidence of a sample device;

[0042] determining a sample resource demand value, a sample resource demand scenario and a sample scenario confidence of the sample device according to the sample fusion time series data based on the demand prediction model to be trained;

[0043] determining a demand loss value according to the sample resource demand value and the standard resource demand value; and,

[0044] determining a scenario loss value according to the sample resource demand scenario and the standard resource demand scenario; and,

[0045] determine a confidence loss value according to the sample scenario confidence and the standard scenario confidence;

[0046] take a sum value among the demand loss value, the scenario loss value and the confidence loss value as a model loss value of the demand prediction model; the model loss value is used for adjusting a model parameter of the demand prediction model.

[0047] In a second aspect, the present application further provides a cloud resource scheduling device, comprising:

[0048] an acquisition module, configured to acquire at least one device time sequence data of a target device;

[0049] a fusion module, configured to perform fusion processing on different device time sequence data to obtain fused time sequence data;

[0050] a data module, configured to determine, based on a demand prediction model trained, scheduling reference data of the target device according to the fused time sequence data; wherein the scheduling reference data comprises at least one of a resource demand value, a resource demand scenario and a scenario confidence;

[0051] an execution module, configured to determine, according to the scheduling reference data, a target scheduling strategy to be executed under the resource demand scenario; the target scheduling strategy is used for scheduling cloud resources in the target device to meet the resource demand value.

[0052] In a third aspect, the present application further provides a computer device, comprising a memory and a processor, the memory stores a computer program, and the processor realizes the following steps when executing the computer program:

[0053] acquiring at least one device time sequence data of a target device;

[0054] performing fusion processing on different device time sequence data to obtain fused time sequence data;

[0055] determining, based on a demand prediction model trained, scheduling reference data of the target device according to the fused time sequence data; wherein the scheduling reference data comprises at least one of a resource demand value, a resource demand scenario and a scenario confidence;

[0056] determining, according to the scheduling reference data, a target scheduling strategy to be executed under the resource demand scenario; the target scheduling strategy is used for scheduling cloud resources in the target device to meet the resource demand value.

[0057] In a fourth aspect, the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the following steps:

[0058] obtaining at least one device time series data of a target device;

[0059] fusing different device time series data to obtain fused time series data;

[0060] based on the trained demand prediction model, determining scheduling reference data of the target device according to the fused time series data; wherein the scheduling reference data comprises at least one of a resource demand value, a resource demand scenario and a scenario confidence level;

[0061] determining a target scheduling strategy to be executed under the resource demand scenario according to the scheduling reference data; the target scheduling strategy is used to schedule cloud resources in the target device to meet the resource demand value.

[0062] In a fifth aspect, the present application also provides a computer program product comprising a computer program which, when executed by a processor, implements the following steps:

[0063] obtaining at least one device time series data of a target device;

[0064] fusing different device time series data to obtain fused time series data;

[0065] based on the trained demand prediction model, determining scheduling reference data of the target device according to the fused time series data; wherein the scheduling reference data comprises at least one of a resource demand value, a resource demand scenario and a scenario confidence level;

[0066] determining a target scheduling strategy to be executed under the resource demand scenario according to the scheduling reference data; the target scheduling strategy is used to schedule cloud resources in the target device to meet the resource demand value.

[0067] The cloud resource scheduling method, device, computer device and storage medium described above, by fusing multi-source device time series data, build a feature representation that comprehensively reflects the system running state, lay a data foundation for accurate prediction; based on the fused time series data, the system uses the trained demand prediction model to synchronously generate multi-dimensional scheduling reference data such as resource demand value, resource demand scenario and scenario confidence level, realizes accurate quantification of system resource demand and intelligent identification of running scenario; on this basis, combined with the prediction reliability reflected by the scenario confidence level, the optimal scheduling strategy is selected under the guidance of the resource demand scenario, which not only ensures the accurate matching of cloud resource allocation and real-time demand, but also effectively balances the short-term resource efficiency and long-term running stability through the scenario-based strategy, further improves the rationality of cloud resource scheduling. BRIEF DESCRIPTION OF DRAWINGS

[0068] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the embodiments or the related art description will be briefly introduced. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0069] Figure 1 An application environment diagram of the cloud resource scheduling method provided by the present embodiment is shown in the following figure.

[0070] Figure 2 A flowchart of the first cloud resource scheduling method provided by the present embodiment is shown in the following figure.

[0071] Figure 3 A flowchart of the scheduling reference data determination step provided by the present embodiment is shown in the following figure.

[0072] Figure 4 A flowchart of the fusion time sequence data determination step provided by the present embodiment is shown in the following figure.

[0073] Figure 5 A flowchart of the target scheduling strategy determination step provided by the present embodiment is shown in the following figure.

[0074] Figure 6 A flowchart of the demand prediction model training step provided by the present embodiment is shown in the following figure.

[0075] Figure 7 A structural block diagram of the cloud resource scheduling device provided by the present embodiment is shown in the following figure.

[0076] Figure 8 An internal structure diagram of the computer device provided by the present embodiment is shown in the following figure. DETAILED DESCRIPTION

[0077] In order to make the purpose, technical solutions and advantages of the present application more clear, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0078] The cloud resource scheduling method provided by the present embodiment can be applied to, for example Figure 1The application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data required by the server 104 to process. The data storage system can be integrated on the server 104, or placed on the cloud or other network servers. The computer device obtains at least one device timing data of the target device; fuse different device timing data to obtain fused timing data; based on the trained demand prediction model, determine the scheduling reference data of the target device according to the fused timing data; wherein the scheduling reference data includes at least one of resource demand value, resource demand scenario and scenario confidence; according to the scheduling reference data, determine the target scheduling strategy needed to be executed under the resource demand scenario; the target scheduling strategy is used to schedule the cloud resources in the target device to meet the resource demand value. Among them, the terminal 102 can be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices, Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 104 can be realized by an independent server or a server cluster composed of multiple servers.

[0079] In an exemplary embodiment, as shown in Figure 2 , a cloud resource scheduling method is provided, and the method is applied to Figure 1 The computer device in the example is described, including the following steps S201 to S204. Among them:

[0080] S201 obtains at least one device timing data of the target device.

[0081] Among them, the target device refers to the computing device or system that needs to be scheduled for cloud resources. These devices can be physical servers, virtual machines, containers or any other form of computing resource instances, which run in a cloud computing environment and need to be reasonably allocated and managed to meet different workload requirements.

[0082] Among them, the device timing data refers to a data set of target device running state and performance indicators recorded in time sequence. The device timing data specifically includes but is not limited to: performance timing data and non-performance timing data; non-performance timing data includes log timing data and event timing data.

[0083] Performance time-series data reflects the time-series data of device performance metrics, such as processor utilization, memory utilization, and disk IOPS (input / output operations per second). This data is used to quantify the current load and performance status of the device. Log time-series data is the time-series information in the log files generated by the device, such as system error logs and API call records. This data provides a detailed context and event sequence of the device's operation. Event time-series data is the time-series data of external events related to the device's operation, such as promotional plans on telemarketing platforms and weather conditions on food delivery platforms. This data may affect the device's load and resource requirements.

[0084] In some embodiments, at least one device timing data of the target device within a preset time period is obtained; the device timing data is preprocessed to update the device timing data.

[0085] For example, the performance time series data collected at time i is denoted as ,

[0086] in, Let i be the processor utilization at time i. Let i be the memory utilization at time i. For disk IOPS; construct a performance time series matrix for a specified time period: Use a Transformer encoder with causal masking to process timing data to update performance timing data: .

[0087] For example, log time-series data is embedded in the embedding layer to generate semantic vectors, which are then used to update the log time-series data. Where θ is the pre-trained parameter, This is the log time-series data before the update. This is the updated log time-series data.

[0088] For example, event time-series data is embedded in the embedding layer to generate semantic vectors, which are then used to update the event time-series data, resulting in... Where θ is the pre-trained parameter, This is the event timeline data before the update. This is the updated event timeline data.

[0089] S202 performs fusion processing on timing data from different devices to obtain fused timing data.

[0090] Among them, fused time-series data refers to unified time-series representation data that can comprehensively reflect the operating status and resource demand characteristics of the device by fusing various device time-series data (including performance time-series data, log time-series data, and event time-series data) of the target device.

[0091] It should be noted that before the fusion processing of different device time series data, the device time series data can also be preprocessed, that is, the device time series data is preprocessed through the following formula (1).

[0092] (1)

[0093] Wherein, X is the performance time series data, is the preprocessed performance time series data, is the preset performance matrix, Y is the log time series data, is the preprocessed log time series data, is the preset log matrix, Z is the event time series data, is the preprocessed event time series data, is the preset event matrix, , , is the length of each device time series data, , , is the original dimension of each device time series data.

[0094] In an alternative embodiment, the different device time series data is directly fused to obtain the fusion time series data.

[0095] In another alternative embodiment, the different device time series data is fused to obtain the initial time series data; the initial time series data, the log time series data and the event time series data are mean-pooling processed to obtain the fusion time series data.

[0096] For example, based on the preset fusion algorithm, the different device time series data is fused to obtain the initial time series data; based on the following formula (2), the initial time series data, the log time series data and the event time series data are mean-pooling processed to obtain the fusion time series data.

[0097] (2)

[0098] Wherein, is the fusion time series data, is the initial time series data, is the log time series data, is the event time series data. It should be noted that, may also be the preprocessed log time series data, may also be the preprocessed event time series data.

[0099] S203, based on the demand prediction model trained, determines the scheduling reference data of the target device according to the fused time series data.

[0100] The demand prediction model can be a model constructed based on machine learning or deep learning technology, which is used to predict the future resource demand of the target device according to the fused time series data. The model is trained by historical data and can output scheduling reference data such as resource demand value, resource demand scenario and scenario confidence.

[0101] The scheduling reference data is comprehensive information output by the demand prediction model, including at least one of resource demand value (specific value of future resource demand), resource demand scenario (such as promotion activity, normal load, etc. Load type) and scenario confidence (reliability of model demand scenario classification).

[0102] The resource demand value is a specific quantitative value of the required computing resource (such as CPU, memory, IOPS, etc.) of the target device in a future time period, which is used to guide resource allocation. The resource demand scenario is the classification of device load type identified by the demand prediction model, such as "double 11 promotion", "burst traffic" or "daily load", etc., which reflects the resource demand characteristics under different business scenarios. Optionally, the resource demand scenario can be understood as the probability distribution of different business scenario classification. The scenario confidence is a reliability quantitative index of the demand prediction model for resource demand scenario classification result, which represents the confidence level of the model classifying the target device into a certain candidate demand scenario, which is usually reflected by a probability value.

[0103] In some embodiments, the fused time series data is input into the demand prediction model to obtain the scheduling reference data of the target device.

[0104] S204, according to the scheduling reference data, determines the target scheduling strategy to be executed under the resource demand scenario.

[0105] The target scheduling strategy is used to guide the specific operation scheme of cloud resource allocation and adjustment of the target device under a specific resource demand scenario, aiming to meet the resource demand while optimizing cost, performance and stability. That is, the target scheduling strategy is used to schedule the cloud resources in the target device to meet the resource demand value.

[0106] In some embodiments, based on the preset strategy determination model, the target scheduling strategy to be executed under the resource demand scenario is determined according to the scheduling reference data.

[0107] The aforementioned cloud resource scheduling method, by integrating time-series data from multiple devices, constructs a feature representation that comprehensively reflects the system's operating status, laying a data foundation for accurate prediction. Based on the integrated time-series data, the system utilizes a trained demand prediction model to synchronously generate multi-dimensional scheduling reference data, including resource demand values, resource demand scenarios, and scenario confidence levels. This enables precise quantification of system resource demands and intelligent identification of operating scenarios. Furthermore, by combining the prediction reliability reflected by scenario confidence levels, the optimal scheduling strategy is selected under the guidance of resource demand scenarios. This ensures accurate matching between cloud resource allocation and real-time demands, while effectively balancing short-term resource efficiency and long-term operational stability through scenario-based strategies, further improving the rationality of cloud resource scheduling.

[0108] Figure 3 This is a flowchart illustrating the steps for determining scheduling reference data in one embodiment. This embodiment refines the steps in the above embodiment where, based on the trained demand prediction model and fused time-series data, the scheduling reference data for the target device is determined. The steps include the following:

[0109] S301 performs a linear transformation on the fused time-series data based on a preset first weight matrix and a preset first bias vector to obtain the resource requirement value of the target device.

[0110] In some embodiments, based on the following formula (3), a linear transformation is performed on the fused time series data based on a preset first weight matrix and a preset first bias vector to obtain the resource requirement value of the target device.

[0111] (3)

[0112] in, It is the value of resource demand, that is, in the future. Resource demand forecasts after time , It is the fusion of time series data. It is a preset first weight matrix. The first bias vector is a preset value, which can be learned through training.

[0113] S302 classifies the fused time-series data based on a preset classification function to determine the resource requirement scenarios of the target device.

[0114] In some embodiments, based on the following formula (4), the fused time series data is classified according to a preset classification function to determine the probability distribution of the target device under different candidate demand scenarios in the future; the candidate demand scenario with the highest probability is selected as the resource demand scenario of the target device.

[0115] (4)

[0116] wherein, is a probability distribution of the target device under different candidate demand scenarios in the future, is a preset third weight matrix, is an average pooling in the time dimension, is the fused time-series data.

[0117] S303 performs linear transformation on the fused time-series data based on a preset second weight matrix and a preset second bias vector to obtain a scene confidence of the target device.

[0118] wherein, the scene confidence comprises a confidence of the target device being classified into different candidate demand scenarios; the candidate demand scenario comprises a resource demand scenario.

[0119] In some embodiments, the scene confidence of the target device is obtained by performing linear transformation on the fused time-series data based on a preset second weight matrix and a preset second bias vector based on the following formula (5).

[0120] (5)

[0121] wherein, is the scene confidence of the target device (at this time, the scene confidence can be a parameter vector of Dirichlet distribution), is the fused time-series data, is a preset second weight matrix, is a preset second bias vector, and both can be obtained by training and learning.

[0122] In some embodiments, the scene confidence is normalized to update the scene confidence according to the number of scenes of different candidate demand scenarios.

[0123] For example, the scene confidence is normalized to update the scene confidence according to the number of scenes of different candidate demand scenarios based on the following formula (6).

[0124] (6)

[0125] wherein, is the updated scene confidence, is the number of scenes of different candidate demand scenarios, is the scene confidence of the target device.

[0126] It should be noted that the greater the value of the scene confidence, the more uncertain the model is in predicting the resource demand scenario and the resource demand value.

[0127] In the above embodiment, the linear transformation of the fused time series data by the preset weight matrix and bias vector can efficiently and accurately extract the resource demand features, thereby generating accurate resource demand values. Meanwhile, the classification processing of the fused time series data based on the classification function can dynamically identify the resource demand scenarios of the target device, thereby enhancing the semantic understanding capability. Further, the calculation of the scenario confidence by the second weight matrix and bias vector can quantify the reliability of the demand scenario classification by the model, thereby effectively distinguishing the uncertainty sources. This process realizes the joint optimization of the resource demand values, scenario classification and confidence, thereby significantly improving the prediction accuracy and decision robustness of the cloud resource scheduling, and providing a reliable basis for the adaptive resource allocation.

[0128] Figure 4 A flowchart for determining the steps of the fused time series data in an embodiment. In this embodiment, the device time series data includes performance time series data and non-performance time series data. On this basis, the steps of fusing the different device time series data in the above embodiment to obtain the initial time series data are refined, including the following steps:

[0129] S401 fusing the performance time series data and the non-performance time series data to obtain initial fused data.

[0130] In an optional embodiment, the performance time series data and the non-performance time series data are directly fused to obtain the initial fused data.

[0131] In another optional embodiment, the performance time series data and the log time series data are fused to obtain first time series data, and the performance time series data and the event time series data are fused to obtain second time series data. The first time series data and the second time series data are fused to obtain the initial fused data.

[0132] For example, the performance time series data and the log time series data are fused to obtain the first time series data based on the following formula (7).

[0133] (7)

[0134] wherein, is the first time series data, is the first query vector, is the first key vector, is the first value vector, X is the performance time series data, is the first query matrix, Y is the log time series data, is the first key matrix, is the first value matrix, d x is the feature dimension of the performance time series data, d Y is the feature dimension of the log time series data.

[0135] Exemplarily, the performance time series data and the event time series data are fused to obtain second time series data based on the following formula (8).

[0136] (8)

[0137] wherein, is the second time series data, is a second query vector, is a second key vector, is a second value vector, X is the performance time series data, is a second query matrix, Z is the event time series data, is a second key matrix, is a second value matrix, d x is a performance time series data feature dimension, d Z is an event time series data feature dimension.

[0138] S402 fuses the performance time series data and the initial fusion data to obtain initial time series data.

[0139] In some embodiments, the performance time series data and the initial fusion data are spliced to obtain the initial time series data based on the following formula (9).

[0140] (9)

[0141] wherein, is the initial fusion data, is the performance time series data, is a fourth matrix, is the first time series data, is the second time series data.

[0142] On the basis of the above embodiments, through a phased fusion strategy, the performance time series data and the non-performance time series data (log, event) are initially fused to form initial fusion data containing multi-dimensional features; and then the performance index and the initial fusion data are deeply integrated through secondary fusion to generate initial time series data that takes into account numerical features and semantic context. This hierarchical fusion method can effectively capture the dynamic correlation between performance fluctuations and business events (such as the correlation between CPU surge and promotion activities), significantly improve the completeness of feature representation and the accuracy of time series prediction, and provide a more reliable data foundation for subsequent resource demand prediction and scenario classification. Figure 5 FIG. 1 is a flowchart of a method for determining a target scheduling strategy according to an embodiment of the present application. The method comprises the following steps:

[0143] S501 obtaining at least one resource scheduling strategy executable by the target device.

[0144] The resource scheduling strategy refers to a specific resource allocation and adjustment operation scheme formulated for the resource demand scenario and scheduling reference data of the target device in a cloud computing environment, including resource expansion, migration, release, etc. actions, aiming to meet the business load demand while optimizing resource utilization, cost and service performance.

[0145] S502 determining the cloud resources required in the execution of the resource scheduling strategy by the target device for each resource scheduling strategy.

[0146] In some embodiments, the cloud resources required in the execution of the resource scheduling strategy by the target device for each resource scheduling strategy are determined based on the following formula (10).

[0147] (10)

[0148] wherein, is the resource scheduling strategy.

[0149] S503 determining the over-standard rate of the resource scheduling strategy according to the cloud resources, the resource demand value and the scenario confidence.

[0150] In an optional embodiment, the over-standard rate of the resource scheduling strategy is determined according to the cloud resources, the resource demand value and the scenario confidence based on a preset over-standard function.

[0151] In another optional embodiment, the resource error is determined according to the difference between the cloud resources and the resource demand value; the over-standard rate of the resource scheduling strategy is determined according to the proportional relationship between the resource error and the scenario confidence.

[0152] For example, the difference between the cloud resources and the resource demand value is taken as the resource error based on the following formula (11); the quotient value between the resource error and the scenario confidence is taken as the over-standard rate of the resource scheduling strategy.

[0153] (11)

[0154] wherein, is the over-standard rate, is the cloud resource, is the resource demand value, is the scenario confidence.

[0155] S504 selecting the target scheduling strategy required to be executed under the resource demand scenario from the resource scheduling strategies according to the over-standard rate and the execution cost of the resource scheduling strategy.

[0156] In an alternative embodiment, based on a preset constraint condition, a target scheduling strategy required to be executed under the resource demand scenario is selected from the resource scheduling strategies according to the over-standard rate and the execution cost of the resource scheduling strategy based on formula (12).

[0157] (12)

[0158] wherein, is the resource scheduling strategy, A is the execution cost of the resource scheduling strategy; is the resource scheduling strategy is the execution cost of the resource scheduling strategy is the over-standard rate (i.e. the probability of causing service level agreement violation) of the resource scheduling strategy is the preset over-standard threshold.

[0159] In another alternative embodiment, a resource scheduling strategy with an over-standard rate less than the preset over-standard threshold is selected as a reference scheduling strategy; and a reference scheduling strategy with the minimum execution cost is selected as the target scheduling strategy.

[0160] It should be noted that the embodiment can also determine a demand adjustment strategy matched with the resource demand scenario; and update the resource demand value according to the demand adjustment strategy.

[0161] For example, a demand adjustment strategy matched with the resource demand scenario is determined; and a product value between the resource demand value and a demand amplitude in the demand adjustment strategy is taken as the updated resource demand value.

[0162] For example, if the demand adjustment strategy matched with the resource demand scenario is an "aggressive" strategy (e.g. corresponding to a "big promotion" scenario), it means that higher resource cost is willing to be borne to ensure SLA, and the resource demand prediction value will be virtually increased in subsequent calculation (e.g. the resource demand value is increased by 30% as the demand amplitude), which is equivalent to forcibly requiring more redundancy. For another example, if the demand adjustment strategy matched with the resource demand scenario is a "conservative" strategy (e.g. corresponding to a "fault" scenario), it means that stability is preferred, and the cost penalty of action can be increased to avoid high-risk expansion operation.

[0163] ​In the above embodiment, by comprehensively evaluating the available resource scheduling strategy of the target device, combining the required cloud resource amount for execution, the predicted resource demand value and the scene confidence, the over-standard rate (i.e. resource shortage risk) of each strategy is quantitatively calculated, and further the multi-objective optimization is carried out by taking the execution cost into account. This method can effectively balance the resource efficiency and cost constraints, avoid waste caused by excessive configuration or performance degradation caused by resource shortage, and at the same time enhance the adaptability to uncertain environment through scene confidence, finally select the optimal target scheduling strategy considering safety, economy and real-time performance, significantly improve the intelligent level of cloud resource scheduling.

[0164] Figure 6 A flowchart of the demand prediction model training step in one embodiment. The embodiment refines the above embodiment and includes the following steps:

[0165] S601 obtains sample fusion time series data of a sample device, a standard resource demand value, a standard resource demand scene and a standard scene confidence.

[0166] S602 determines a sample resource demand value, a sample resource demand scene and a sample scene confidence of the sample device based on the sample fusion time series data according to the demand prediction model to be trained.

[0167] S603 determines a demand loss value according to the sample resource demand value and the standard resource demand value, a scene loss value according to the sample resource demand scene and the standard resource demand scene, and a confidence loss value according to the sample scene confidence and the standard scene confidence.

[0168] In some embodiments, the demand loss value is determined according to the sample resource demand value and the standard resource demand value, the scene loss value is determined according to the sample resource demand scene and the standard resource demand scene, and the confidence loss value is determined according to the sample scene confidence and the standard scene confidence based on the following formula (13).

[0169] (13)

[0170] wherein, is the demand loss value, is the sample resource demand value, is the standard resource demand value, is the scene loss value, is the sample resource demand scene, is the standard resource demand scene, is the confidence loss value, is the sample scene confidence, is the standard scene confidence.

[0171] S604 takes the sum value between the demand loss value, the scenario loss value and the confidence loss value as the model loss value of the demand prediction model.

[0172] The model loss value is used to adjust the model parameters of the demand prediction model.

[0173] In some embodiments, the sum value between the demand loss value, the scenario loss value and the confidence loss value is taken as the model loss value of the demand prediction model based on the following formula (14).

[0174] (14)

[0175] wherein, the demand loss value is, the scenario loss value is, the confidence loss value is, the demand weight of the demand loss value is, the scenario weight of the scenario loss value is, the confidence weight of the confidence loss value is.

[0176] In the above embodiments, by constructing a multi-task joint loss function that integrates short-term demand prediction, long-term scenario classification and cognitive uncertainty estimation, the model is forced to simultaneously optimize numerical regression accuracy, scenario classification accuracy and uncertainty calibration ability during the model training phase. This training mechanism can effectively avoid the model from making overconfident false predictions, ensuring that the scenario confidence of its output can truly reflect cognitive uncertainty. The model trained in this way can provide risk-aware prediction results in actual deployment, providing accurate and reliable basis for subsequent resource scheduling decisions, thereby significantly improving the overall robustness and decision security of the cloud resource scheduling system.

[0177] It should be understood that although each step in the flowchart involved in each of the above embodiments is shown in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in each of the above embodiments can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0178] Based on the same inventive concept, the embodiments of the present application also provide a cloud resource scheduling device for implementing the cloud resource scheduling method described above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more cloud resource scheduling device embodiments provided below can refer to the limitations of the cloud resource scheduling method described above, which will not be repeated here.

[0179] In one exemplary embodiment, as shown in Figure 7 a cloud resource scheduling device is provided, comprising: an acquisition module 701, a fusion module 702, a data module 703 and an execution module 704, wherein:

[0180] The acquisition module 701 is configured to acquire at least one device timing data of a target device.

[0181] The fusion module 702 is configured to perform fusion processing on different device timing data to obtain fused timing data.

[0182] The data module 703 is configured to determine scheduling reference data of the target device based on a trained demand prediction model and the fused timing data, wherein the scheduling reference data comprises at least one of a resource demand value, a resource demand scenario and a scenario confidence.

[0183] The execution module 704 is configured to determine a target scheduling strategy to be executed under a resource demand scenario based on the scheduling reference data, wherein the target scheduling strategy is used to schedule cloud resources in the target device to meet the resource demand value.

[0184] In some embodiments, the data module 703 is further configured to perform linear transformation on the fused timing data based on a preset first weight matrix and a preset first bias vector to obtain the resource demand value of the target device, perform classification processing on the fused timing data based on a preset classification function to determine the resource demand scenario of the target device, and perform linear transformation on the fused timing data based on a preset second weight matrix and a preset second bias vector to obtain the scenario confidence of the target device, wherein the scenario confidence comprises a confidence of the target device being classified into different candidate demand scenarios, and the candidate demand scenario comprises the resource demand scenario.

[0185] In some embodiments, the data module 703 is further configured to normalize the scenario confidence based on the number of different candidate demand scenarios to update the scenario confidence.

[0186] In some embodiments, the fusion module 702 is further configured to perform fusion processing on different device timing data to obtain initial timing data, and perform mean pooling processing on the initial timing data, log timing data and event timing data to obtain the fused timing data.

[0187] In some embodiments, the fusion module 702 is further configured to fuse the performance time series data and the non-performance time series data to obtain initial fusion data, and fuse the performance time series data and the initial fusion data to obtain initial time series data.

[0188] In some embodiments, the fusion module 702 is further configured to fuse the performance time series data and the log time series data to obtain first time series data, and fuse the performance time series data and the event time series data to obtain second time series data, and fuse the first time series data and the second time series data to obtain initial fusion data.

[0189] In some embodiments, the execution module 704 is further configured to obtain at least one resource scheduling strategy executable by the target device, determine, for each resource scheduling strategy, cloud resources required in executing the resource scheduling strategy by the target device, determine an over-standard rate of the resource scheduling strategy according to the cloud resources, the resource demand value, and the scenario confidence, and select, according to the over-standard rate and an execution cost of the resource scheduling strategy, a target scheduling strategy required to be executed in the resource demand scenario from the resource scheduling strategies.

[0190] In some embodiments, the execution module 704 is further configured to determine a resource error according to a difference between the cloud resources and the resource demand value, and determine the over-standard rate of the resource scheduling strategy according to a proportional relationship between the resource error and the scenario confidence.

[0191] In some embodiments, the execution module 704 is further configured to select, as a reference scheduling strategy, a resource scheduling strategy whose over-standard rate is less than a preset over-standard threshold, and select, as the target scheduling strategy, a reference scheduling strategy with a minimum execution cost.

[0192] In some embodiments, the execution module 704 is further configured to determine a demand adjustment strategy matched with the resource demand scenario, and update the resource demand value according to the demand adjustment strategy.

[0193] In some embodiments, the data module 703 is further configured to obtain sample fusion time series data of a sample device, a standard resource demand value, a standard resource demand scenario, and a standard scenario confidence.

[0194] Based on the demand prediction model to be trained, the sample resource demand value, the sample resource demand scene and the sample scene confidence of the sample device are determined according to the sample fused time series data; the demand loss value is determined according to the sample resource demand value and the standard resource demand value; the scene loss value is determined according to the sample resource demand scene and the standard resource demand scene; and the confidence loss value is determined according to the sample scene confidence and the standard scene confidence; the sum value between the demand loss value, the scene loss value and the confidence loss value is taken as the model loss value of the demand prediction model; and the model loss value is used to adjust the model parameter of the demand prediction model.

[0195] The modules in the cloud resource scheduling apparatus can be implemented by software, hardware or a combination thereof. The modules can be embedded in or independent of a processor in a computer device in hardware form, or stored in a memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to the modules.

[0196] In an exemplary embodiment, a computer device, which can be a server, is provided, and an internal structure diagram of the computer device can be as shown in Figure 8 The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store data. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with external terminals through a network connection. The computer program is executed by the processor to implement a cloud resource scheduling method.

[0197] Those skilled in the art can understand that Figure 8 The structure shown in the above description is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0198] In an embodiment, a computer device is also provided, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0199] In an embodiment, a computer readable storage medium is provided, having stored thereon a computer program which, when executed by a processor, implements the steps of any of the above method embodiments.

[0200] In an embodiment, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the steps of any of the above method embodiments.

[0201] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0202] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0203] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0204] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A cloud resource scheduling method, characterized in that, Applied to a target device, the method includes: Acquire at least one device timing data of the target device; The timing data from different devices are fused to obtain fused timing data. Based on the trained demand prediction model, scheduling reference data for the target device is determined according to the fused time-series data; wherein, the scheduling reference data includes at least one of resource demand value, resource demand scenario, and scenario confidence. Based on the scheduling reference data, a target scheduling strategy to be executed under the resource demand scenario is determined; the target scheduling strategy is used to schedule cloud resources in the target device to meet the resource demand value.

2. The method according to claim 1, characterized in that, The demand prediction model based on the completed training determines the scheduling reference data for the target device according to the fused time-series data, including: Based on a preset first weight matrix and a preset first bias vector, a linear transformation is performed on the fused time-series data to obtain the resource requirement value of the target device; and... Based on a preset classification function, the fused time-series data is classified to determine the resource requirement scenarios of the target device; and... Based on a preset second weight matrix and a preset second bias vector, the fused time-series data is linearly transformed to obtain the scene confidence of the target device; the scene confidence includes the confidence of the target device being classified into different candidate demand scenarios; the candidate demand scenarios include resource demand scenarios.

3. The method according to claim 2, characterized in that, The method further includes: Based on the number of different candidate demand scenarios, the scenario confidence is normalized to update the scenario confidence.

4. The method according to claim 1, characterized in that, The device time-series data includes performance time-series data and non-performance time-series data; the process of fusing different device time-series data to obtain fused time-series data includes: The timing data from different devices are fused to obtain the initial timing data; The initial time-series data, log time-series data, and event time-series data are subjected to mean pooling to obtain fused time-series data.

5. The method according to claim 4, characterized in that, The process of fusing time-series data from different devices to obtain initial time-series data includes: The performance time-series data and non-performance time-series data are fused to obtain initial fused data; The performance time-series data and the initial fusion data are fused to obtain the initial time-series data.

6. The method according to claim 5, characterized in that, The non-performance time-series data includes log time-series data and event time-series data; the fusion processing of the performance time-series data and non-performance time-series data to obtain initial fused data includes: The performance time-series data and the log time-series data are fused to obtain first time-series data; and, The performance time-series data and the event time-series data are fused together to obtain second time-series data; The first time series data and the second time series data are fused to obtain initial fused data.

7. The method according to claim 1, characterized in that, The step of determining the target scheduling strategy to be executed under the resource demand scenario based on the scheduling reference data includes: Obtain at least one resource scheduling strategy that the target device can execute; For each resource scheduling policy, determine the cloud resources required during the execution of the resource scheduling policy on the target device; The over-limit rate of the resource scheduling strategy is determined based on the cloud resources, the resource demand value, and the scenario confidence level. Based on the over-limit rate and the execution cost of the resource scheduling strategy, select the target scheduling strategy to be executed under the resource demand scenario from the resource scheduling strategies.

8. The method according to claim 7, characterized in that, The step of determining the overrun rate of the resource scheduling strategy based on the cloud resources, the resource demand value, and the scenario confidence level includes: The resource error is determined based on the difference between the cloud resources and the resource demand value; The over-limit rate of the resource scheduling strategy is determined based on the proportional relationship between the resource error and the scenario confidence level.

9. The method according to claim 7, characterized in that, The step of selecting a target scheduling strategy from the resource scheduling strategies based on the over-limit rate and the execution cost of the resource scheduling strategy includes: The resource scheduling strategy with an exceedance rate less than the preset exceedance threshold is selected as the reference scheduling strategy; The reference scheduling strategy with the lowest execution cost is selected as the target scheduling strategy.

10. The method according to claim 7, characterized in that, The method further includes: Determine a demand adjustment strategy that matches the resource demand scenario; Adjust the strategy according to the requirements and update the resource requirement value.

11. The method according to any one of claims 1-10, characterized in that, The demand forecasting model is trained in the following manner: Acquire sample fusion time-series data, standard resource requirement values, standard resource requirement scenarios, and standard scenario confidence levels of the sample devices; Based on the demand prediction model to be trained, the sample resource demand value, sample resource demand scenario, and sample scenario confidence level of the sample device are determined according to the sample fusion time series data. Based on the sample resource demand value and the standard resource demand value, determine the demand loss value; and, Determine the scenario loss value based on the sample resource demand scenario and the standard resource demand scenario; as well as, The confidence loss value is determined based on the confidence levels of the sample scenarios and the standard scenarios. The sum of the demand loss value, scenario loss value, and confidence loss value is used as the model loss value of the demand forecasting model; the model loss value is used to adjust the model parameters of the demand forecasting model.

12. A cloud resource scheduling device, characterized in that, The device includes: The acquisition module is used to acquire at least one device timing data of the target device; The fusion module is used to fuse time-series data from different devices to obtain fused time-series data; The data module is used to determine the scheduling reference data of the target device based on the trained demand prediction model and the fused time series data; wherein the scheduling reference data includes at least one of resource demand value, resource demand scenario, and scenario confidence. The execution module is used to determine the target scheduling strategy to be executed under the resource demand scenario based on the scheduling reference data; the target scheduling strategy is used to schedule cloud resources in the target device to meet the resource demand value.

13. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 11.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 11.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 11.

Images