Cloud resource scheduling method and device based on artificial intelligence, equipment and medium

Abstract

The invention relates to a cloud resource scheduling method and device based on artificial intelligence, equipment and a medium. The method comprises the following steps: acquiring cloud resource state data and workload demand data; inputting the workload demand data into a task type identification model, and identifying to obtain a workload task type; setting a resource state prediction time period of a cloud resource state prediction model based on a workload loading reference time period corresponding to the workload task type, and inputting the cloud resource state data into the cloud resource state prediction model to obtain cloud resource reference state prediction information; and performing cloud resource scheduling based on the cloud resource reference state prediction information, the workload task type and the workload demand data, and generating workload cloud resource scheduling result information. By adopting the method, the cloud resource utilization rate and the cloud service quality can be improved, so that a dynamic cloud environment and multi-element service requirements can be adapted.

Description

Technical Field

[0001] This application belongs to the field of cloud resource scheduling, and in particular relates to a cloud resource scheduling method, device, equipment and medium based on artificial intelligence. Background Technology

[0002] With the continuous evolution of cloud computing technology and the deep development of the digital economy, enterprises and individual users are increasingly reliant on cloud services. The business scenarios supported by cloud platforms are becoming increasingly diverse, encompassing various needs such as big data processing, artificial intelligence training, core enterprise business systems, and distributed application deployment. Against this backdrop, cloud resource scheduling, as a core technology for ensuring cloud service quality, improving resource utilization, and controlling operating costs, is becoming increasingly important.

[0003] Traditional cloud resource scheduling methods mainly rely on static scheduling strategies. By pre-setting resource allocation rules and configuring cloud resources based on human experience, they lack flexibility, have low resource utilization, and weak service quality assurance capabilities when facing server load fluctuations, network bandwidth changes, node failures, multi-type resource collaboration, and diverse workloads. They are also difficult to adapt to dynamic cloud environments and diverse business needs. Summary of the Invention

[0004] Therefore, it is necessary to provide an AI-based cloud resource scheduling method, device, equipment, and medium that can adapt to dynamic cloud environments and diverse business needs, addressing the aforementioned technical problems.

[0005] Firstly, this application provides a cloud resource scheduling method based on artificial intelligence, including:

[0006] Obtain cloud resource status data and workload demand data;

[0007] Input the workload requirement data into the task type recognition model to identify the workload task type;

[0008] The cloud resource status prediction time period is set based on the workload loading reference time period corresponding to the workload task type, and the cloud resource status data is input into the cloud resource status prediction model to obtain the cloud resource baseline status prediction information.

[0009] Based on cloud resource baseline status prediction information, workload task types, and workload demand data, cloud resource scheduling is performed to generate workload cloud resource scheduling result information.

[0010] In one embodiment, the resource status prediction time period of the cloud resource status prediction model is set based on the workload loading reference time period corresponding to the workload task type, and cloud resource status data is input into the cloud resource status prediction model to obtain cloud resource baseline status prediction information, including:

[0011] The candidate resource status prediction time period with the smallest difference from the reference time period of the same workload loading in the candidate set of resource status prediction time periods is set as the resource status prediction time period of the cloud resource status prediction model.

[0012] Based on cloud resource status data and the status monitoring time of cloud resource status data, a cloud resource status time series sequence is constructed.

[0013] Input the cloud resource status time series into the cloud resource status prediction model to generate the resource status prediction series within the resource status prediction time period;

[0014] Based on the maximum utilization rate of each cloud resource status dimension in the resource status prediction sequence, the baseline cloud resource status prediction information is obtained.

[0015] In one embodiment, the expression for the cloud resource state prediction loss function of the cloud resource state prediction model is:

[0016]

[0017] In the formula, The loss function for cloud resource status prediction. and These are the time-series trend prediction weights and the maximum utilization rate prediction weights, respectively. This represents the total number of cloud resource state training samples. For the first The resource status prediction time period for each cloud resource status training sample. This represents the total number of cloud resource status dimensions. For the first The training sample of cloud resource status within the resource status prediction time period. The first time step The actual label value of the state dimension of each cloud resource status dimension. The first generation of cloud resource state prediction model The first cloud resource state training sample in the resource state prediction sequence The first time step Predicted values ​​for each cloud resource status dimension. For the first The training sample of cloud resource status within the resource status prediction time period. The maximum utilization rate of each cloud resource status dimension is the true label value of the maximum utilization rate. The first generation of cloud resource state prediction model The first cloud resource state training sample in the resource state prediction sequence The maximum utilization prediction value for each cloud resource status dimension.

[0018] In one embodiment, cloud resource scheduling is performed based on cloud resource baseline state prediction information, workload task types, and workload demand data to generate workload cloud resource scheduling result information, including:

[0019] Based on cloud resource baseline status prediction information, workload task types and workload demand data, workload virtual machine-level resource scheduling and allocation are performed to generate workload virtual machine-level resource scheduling information.

[0020] Based on workload virtual machine-level resource scheduling information and workload demand data, workload container-level resource scheduling and allocation are performed to generate workload container-level resource scheduling information.

[0021] Based on workload container-level resource scheduling information and workload demand data, workload task-level resource scheduling and allocation are performed to generate workload task-level resource scheduling information.

[0022] By aggregating workload virtual machine-level resource scheduling information, workload container-level resource scheduling information, and workload task-level resource scheduling information, the workload cloud resource scheduling result information is obtained.

[0023] In one embodiment, the workload requirement data includes cloud resource requirement data and task service type labels. The workload requirement data is input into a task type recognition model to identify the workload task type, including:

[0024] A cloud resource demand data feature vector is constructed based on cloud resource demand data, and a task business code feature vector is constructed based on task business type labels. The cloud resource demand data feature vector and the task business code feature vector are then concatenated to obtain the workload demand feature vector.

[0025] Input the workload requirement feature vector into the task type recognition model to generate the workload task type corresponding to the workload requirement data.

[0026] In one embodiment, the task type recognition model is a multilayer perceptron model. The task type recognition loss function of the task type recognition model includes a cloud resource feature regularization loss function and a task type similarity correction loss function. The expressions for the cloud resource feature regularization loss function and the task type similarity correction loss function are as follows:

[0027]

[0028]

[0029] In the formula, and These are the cloud resource feature regularization loss function and the task type similarity correction loss function, respectively. To identify the total number of training samples for task type identification, the training samples should be identified. The weight matrix of the last fully connected layer of the task type identification model. For the first The workload requirement feature vector of the training samples for each task type identification is used. The total number of workload task types. For the task type identification model, the first The training sample for task type recognition belongs to the first task type recognition training sample. Predicted probability values ​​for workload task types. For the task type identification model, the first The training sample for task type recognition belongs to the first task type recognition training sample. Predicted probability values ​​for workload task types. For the first The workload task type is the same as the first Resource requirement similarity coefficient for similar workload task types.

[0030] Secondly, this application also provides a cloud resource scheduling device based on artificial intelligence, comprising:

[0031] The basic data acquisition module is used to acquire cloud resource status data and workload demand data;

[0032] The task type identification module is used to input workload requirement data into the task type identification model and identify the workload task type;

[0033] The baseline state prediction module is used to set the resource state prediction time period of the cloud resource state prediction model based on the workload loading reference time period corresponding to the workload task type, and input the cloud resource state data into the cloud resource state prediction model to obtain the cloud resource baseline state prediction information.

[0034] The load resource scheduling module is used to perform cloud resource scheduling based on cloud resource baseline status prediction information, workload task types, and workload demand data, and generate workload cloud resource scheduling result information.

[0035] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method as described in any of the first aspects of this application.

[0036] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the first aspects of this application.

[0037] The aforementioned AI-based cloud resource scheduling methods, devices, equipment, and media, through dynamic analysis of workload task types based on a task type recognition model, can accurately distinguish different workload task types, thereby ensuring precise alignment between cloud resource status prediction and workload demand characteristics, and improving the targeting and adaptability of cloud resource scheduling. By dynamically matching the cloud resource status prediction cycle with the workload task type, the resource status prediction cycle can be precisely adapted to the task load characteristics, improving the accuracy of cloud resource baseline status prediction information and enhancing the foresight and adaptability of resource status prediction. By combining cloud resource baseline status prediction information, task type, and load demand for collaborative scheduling, it can comprehensively adapt to scheduling decisions, resource status, task characteristics, and load demands, optimize cloud resource allocation efficiency, improve the rationality of cloud resource configuration, and ensure efficient and stable operation of workloads, thereby improving cloud resource utilization, reducing resource waste and scheduling latency, and thus adapting to dynamic cloud environments and diverse business needs. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 A flowchart illustrating an artificial intelligence-based cloud resource scheduling method provided in one embodiment of this application. Figure 1 ;

[0040] Figure 2 A flowchart illustrating an artificial intelligence-based cloud resource scheduling method provided in one embodiment of this application. Figure 2 ;

[0041] Figure 3 This is a schematic diagram of the structure of an artificial intelligence-based cloud resource scheduling device provided in one embodiment of this application. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0043] In one exemplary embodiment of this application, such as Figure 1 As shown, an artificial intelligence-based cloud resource scheduling method is provided. This embodiment illustrates the application of this method to a resource scheduling terminal. It is understood that this method can also be applied to a resource scheduling server, and further to a resource scheduling system including both a resource scheduling terminal and a resource scheduling server, and is implemented through the interaction between the resource scheduling terminal and the resource scheduling server. In this embodiment, the method includes the following steps:

[0044] Step S101: Obtain cloud resource status data and workload demand data.

[0045] Optionally, the resource scheduling terminal can collect and store multi-dimensional status data of cloud resources in real time through the cloud platform's resource monitoring interface and sensor acquisition units deployed on the cloud server, thus obtaining cloud resource status data. This cloud resource status data may include real-time cloud resource status data or historical cloud resource status data; no limitation is made here.

[0046] For example, cloud resource status data may include, but is not limited to, CPU utilization, memory utilization, storage resource utilization, network bandwidth utilization, network latency, read / write speed, read / write port utilization, GPU utilization, GPU memory utilization, cloud resource load balancing status, storage device response time, network packet loss rate, and resource node health status.

[0047] Optionally, the resource scheduling terminal can collect relevant requirement information of the workloads to be executed in the cloud platform to obtain workload requirement data. This workload requirement data may include cloud resource requirement data and task service type tags.

[0048] Optionally, the resource scheduling terminal can remove outliers from the collected cloud resource status data and workload demand data, and supplement missing data to ensure the integrity and accuracy of the cloud resource status data and workload demand data.

[0049] Step S102: Input the workload requirement data into the task type identification model to identify the workload task type.

[0050] Optionally, the resource scheduling terminal can extract resource demand features from the cloud resource demand data in the workload demand data, and construct a cloud resource demand data feature vector based on the resource demand features.

[0051] Optionally, the resource scheduling terminal can encode the task service type labels in the workload demand data to construct a task service encoding feature vector.

[0052] Optionally, the resource scheduling terminal can concatenate the cloud resource demand data feature vector and the task business coding feature vector in a preset order to obtain the workload demand feature vector.

[0053] Optionally, the resource scheduling terminal can input the workload demand feature vector into a pre-trained task type recognition model to obtain the probability distribution of various workload task types. The resource scheduling terminal can then select the candidate workload task type with the highest probability value as the workload task type.

[0054] For example, the task type recognition model can be, but is not limited to, a multilayer perceptron model.

[0055] Optionally, workload task types may include, but are not limited to, large model training tasks, computationally intensive tasks, memory-intensive tasks, read-write intensive tasks, network-intensive tasks, and batch processing tasks.

[0056] Step S103: Set the resource status prediction time period of the cloud resource status prediction model based on the workload loading reference time period corresponding to the workload task type, and input the cloud resource status data into the cloud resource status prediction model to obtain the cloud resource baseline status prediction information.

[0057] Optionally, the resource scheduling terminal can pre-build a candidate set of resource status prediction time periods, which may include multiple candidate resource status prediction time periods of different durations.

[0058] Optionally, the resource scheduling terminal can query the workload loading reference time period corresponding to the currently identified workload task type. The resource scheduling terminal can match the candidate resource status prediction time period with the smallest difference from the workload loading reference time period in the candidate set of resource status prediction time periods, and set the candidate resource status prediction time period with the smallest difference from the workload loading reference time period as the resource status prediction time period of the cloud resource status prediction model.

[0059] Optionally, the resource scheduling terminal can sort the cloud resource status data according to the status monitoring time corresponding to the cloud resource status data, and construct a cloud resource status time series sequence. The resource scheduling terminal can input the cloud resource status time series sequence into a pre-trained cloud resource status prediction model to generate a resource status prediction sequence within the resource status prediction time period. The resource scheduling terminal can parse the resource status prediction sequence to extract the cloud resource baseline status prediction information.

[0060] Optionally, the cloud resource baseline status prediction information may include, but is not limited to, the maximum utilization, average utilization, and minimum utilization of each cloud resource status dimension within the prediction time period.

[0061] Optionally, cloud resource status prediction models can be built based on, but are not limited to, Transformer or LSTM models.

[0062] Step S104: Based on the cloud resource baseline status prediction information, workload task type and workload demand data, perform cloud resource scheduling and generate workload cloud resource scheduling result information.

[0063] Optionally, the resource scheduling terminal can combine cloud resource baseline status prediction information to determine the future supply capacity of cloud resources. At the same time, it can refer to the resource adaptation characteristics of workload task types and the specific requirements of workload demand data to perform workload virtual machine-level resource scheduling and allocation, determine the number of virtual machines, virtual machine specifications and virtual machine deployment locations, and generate workload virtual machine-level resource scheduling information.

[0064] Optionally, the resource scheduling terminal can further refine the resource allocation scheme based on the workload virtual machine-level resource scheduling information and the relevant requirements for containerized deployment in the workload demand data, perform workload container-level resource scheduling and allocation, determine the number of containers, resource quotas and network configurations in each virtual machine, and generate workload container-level resource scheduling information.

[0065] Optionally, the resource scheduling terminal can perform workload task-level resource scheduling allocation based on workload container-level resource scheduling information, combined with the priority, dependency and resource requirements of each subtask in the workload demand data, clarify the execution order, resource usage ratio and scheduling strategy of each task in each container, and generate workload task-level resource scheduling information.

[0066] Optionally, the resource scheduling terminal can aggregate workload virtual machine-level resource scheduling information, workload container-level resource scheduling information, and workload task-level resource scheduling information to obtain workload cloud resource scheduling result information. The workload cloud resource scheduling result information can accurately match the workload's execution requirements with the cloud resource supply status.

[0067] In the aforementioned AI-based cloud resource scheduling method, the dynamic parsing of workload task types based on a task type identification model can accurately distinguish different workload task types, thereby ensuring precise alignment between cloud resource status prediction and workload demand characteristics, and improving the targeting and adaptability of cloud resource scheduling. By dynamically matching the cloud resource status prediction cycle with the workload task type, the resource status prediction cycle can be precisely adapted to the task load characteristics, improving the accuracy of cloud resource baseline status prediction information and enhancing the foresight and adaptability of resource status prediction. Through collaborative scheduling combining cloud resource baseline status prediction information, task type, and load demand, scheduling decisions, resource status, task characteristics, and load demand can be comprehensively adapted, optimizing cloud resource allocation efficiency, improving the rationality of cloud resource configuration, and ensuring efficient and stable operation of workloads. This can improve cloud resource utilization, reduce resource waste and scheduling latency, and ultimately adapt to dynamic cloud environments and diverse business needs.

[0068] In an optional embodiment of this application, please refer to Figure 1 and Figure 2 Step S103: Based on the workload loading reference time period corresponding to the workload task type, set the resource status prediction time period of the cloud resource status prediction model, and input the cloud resource status data into the cloud resource status prediction model to obtain the cloud resource baseline status prediction information, which may include:

[0069] Step S203: Set the candidate resource status prediction time period with the smallest difference from the same workload loading reference time period in the candidate resource status prediction time period set as the resource status prediction time period of the cloud resource status prediction model.

[0070] Step S204: Based on the cloud resource status data and the status monitoring time of the cloud resource status data, a cloud resource status time series sequence is constructed.

[0071] Step S205: Input the cloud resource status time series into the cloud resource status prediction model to generate the resource status prediction sequence within the resource status prediction time period.

[0072] Step S206: Based on the maximum utilization rate of each cloud resource status dimension in the resource status prediction sequence, set the cloud resource baseline status prediction information.

[0073] In an optional embodiment of this application, the expression for the cloud resource state prediction loss function of the cloud resource state prediction model can be:

[0074]

[0075] In the formula, The loss function for cloud resource status prediction. and These are the time-series trend prediction weights and the maximum utilization rate prediction weights, respectively. This represents the total number of cloud resource state training samples. For the first The resource status prediction time period for each cloud resource status training sample. This represents the total number of cloud resource status dimensions. For the first The training sample of cloud resource status within the resource status prediction time period. The first time step The actual label value of the state dimension of each cloud resource status dimension. The first generation of cloud resource state prediction model The first cloud resource state training sample in the resource state prediction sequence The first time step Predicted values ​​for each cloud resource status dimension. For the first The training sample of cloud resource status within the resource status prediction time period. The maximum utilization rate of each cloud resource status dimension is the true label value of the maximum utilization rate. The first generation of cloud resource state prediction model The first cloud resource state training sample in the resource state prediction sequence The maximum utilization prediction value for each cloud resource status dimension.

[0076] In an optional embodiment of this application, please refer to Figure 1 and Figure 2 Step S104: Based on cloud resource baseline state prediction information, workload task types, and workload demand data, perform cloud resource scheduling to generate workload cloud resource scheduling result information, including:

[0077] Step S207: Based on cloud resource baseline status prediction information, workload task type and workload demand data, perform workload virtual machine level resource scheduling and allocation to generate workload virtual machine level resource scheduling information.

[0078] For example, the resource scheduling terminal can perform virtual machine-level resource scheduling and allocation for workloads based on cloud resource baseline status prediction information, workload task types, and workload requirement data, generating virtual machine-level resource scheduling information for workloads. The resource scheduling terminal can parse the required CPU resource specifications, memory resource specifications, storage resource specifications, network bandwidth specifications, and special hardware resource requirements from the workload requirement data, and simultaneously determine the core priority of resource requirements based on the workload task type. Based on cloud resource baseline status prediction information, the resource scheduling terminal can obtain predicted resource status data for all available virtual machines in the cloud platform. The resource scheduling terminal can call the cloud platform's resource management interface to obtain a list of all currently running virtual machines that are not fully occupied, and simultaneously extract predicted resource status data for each virtual machine within the resource status prediction time period. Based on a constructed virtual machine screening index system, the resource scheduling terminal can perform multi-dimensional matching and scoring ranking of all available virtual machines to filter out the target virtual machine cluster. The resource scheduling terminal can compare the inherent resource attributes of each virtual machine with the core resource requirement parameters of the workload to determine whether the minimum resource requirements are met. The resource scheduling terminal can combine the maximum utilization data in the cloud resource baseline status prediction information to determine whether the resource load is within a reasonable range during the prediction time period, thus avoiding task execution anomalies due to resource overload. The resource scheduling terminal can also perform reasonableness verification on the scored and ranked virtual machines, determine the final target virtual machine cluster, and generate workload virtual machine-level resource scheduling information.

[0079] Step S208: Based on the workload virtual machine-level resource scheduling information and workload demand data, perform workload container-level resource scheduling and allocation to generate workload container-level resource scheduling information.

[0080] Step S209: Based on the workload container-level resource scheduling information and workload demand data, perform workload task-level resource scheduling and allocation to generate workload task-level resource scheduling information.

[0081] Step S210: Summarize the workload virtual machine-level resource scheduling information, workload container-level resource scheduling information, and workload task-level resource scheduling information to obtain the workload cloud resource scheduling result information.

[0082] In an optional embodiment of this application, the workload requirement data may include cloud resource requirement data and task service type labels. Inputting the workload requirement data into a task type identification model to identify the workload task type may include:

[0083] Specifically, the resource scheduling terminal can construct a cloud resource demand data feature vector based on cloud resource demand data, construct a task business code feature vector based on task business type labels, and concatenate the cloud resource demand data feature vector and the task business code feature vector to obtain the workload demand feature vector.

[0084] Specifically, the resource scheduling terminal can input the workload demand feature vector into the task type recognition model to generate the workload task type corresponding to the workload demand data.

[0085] In an optional embodiment of this application, the task type recognition model can be a multilayer perceptron model. The task type recognition loss function of the task type recognition model can include a cloud resource feature regularization loss function and a task type similarity correction loss function. The expressions for the cloud resource feature regularization loss function and the task type similarity correction loss function can be:

[0086]

[0087]

[0088] In the formula, and These are the cloud resource feature regularization loss function and the task type similarity correction loss function, respectively. To identify the total number of training samples for task type identification, the training samples should be identified. The weight matrix of the last fully connected layer of the task type identification model. For the first The workload requirement feature vector of the training samples for each task type identification is used. The total number of workload task types. For the task type identification model, the first The training sample for task type recognition belongs to the first task type recognition training sample. Predicted probability values ​​for workload task types. For the task type identification model, the first The training sample for task type recognition belongs to the first task type recognition training sample. Predicted probability values ​​for workload task types. For the first The workload task type is the same as the first Resource requirement similarity coefficient for similar workload task types.

[0089] In one exemplary embodiment of this application, such as Figure 2 As shown, an artificial intelligence-based cloud resource scheduling method is provided, including:

[0090] Step S201: Obtain cloud resource status data and workload demand data.

[0091] Step S202: Input the workload requirement data into the task type identification model to identify the workload task type.

[0092] Step S203: Set the candidate resource status prediction time period with the smallest difference from the same workload loading reference time period in the candidate resource status prediction time period set as the resource status prediction time period of the cloud resource status prediction model.

[0093] Step S204: Based on the cloud resource status data and the status monitoring time of the cloud resource status data, a cloud resource status time series sequence is constructed.

[0094] Step S205: Input the cloud resource status time series into the cloud resource status prediction model to generate the resource status prediction sequence within the resource status prediction time period.

[0095] Step S206: Based on the maximum utilization rate of each cloud resource status dimension in the resource status prediction sequence, set the cloud resource baseline status prediction information.

[0096] Step S207: Based on cloud resource baseline status prediction information, workload task type and workload demand data, perform workload virtual machine level resource scheduling and allocation to generate workload virtual machine level resource scheduling information.

[0097] Step S208: Based on the workload virtual machine-level resource scheduling information and workload demand data, perform workload container-level resource scheduling and allocation to generate workload container-level resource scheduling information.

[0098] Step S209: Based on the workload container-level resource scheduling information and workload demand data, perform workload task-level resource scheduling and allocation to generate workload task-level resource scheduling information.

[0099] Step S210: Summarize the workload virtual machine-level resource scheduling information, workload container-level resource scheduling information, and workload task-level resource scheduling information to obtain the workload cloud resource scheduling result information.

[0100] The aforementioned AI-based cloud resource scheduling method accurately distinguishes workload task types using a task type recognition model, sets resource status prediction cycles to match the workload loading reference time period, and generates benchmark prediction information containing the maximum utilization rate of each status dimension by combining cloud resource status time series. Through hierarchical resource scheduling at the virtual machine, container, and task levels, it can accurately adapt to cloud resource scheduling, workload characteristics, and resource prediction status, improving the rationality and targeting of cloud resource scheduling, ensuring the stability and efficiency of workload execution, thereby improving cloud resource utilization and cloud service quality, and ultimately adapting to dynamic cloud environments and diverse business needs.

[0101] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0102] Based on the same inventive concept, this application also provides an AI-based cloud resource scheduling device for implementing the AI-based cloud resource scheduling method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more AI-based cloud resource scheduling device embodiments provided below can be found in the limitations of the AI-based cloud resource scheduling method described above, and will not be repeated here.

[0103] In one exemplary embodiment, such as Figure 3 As shown, an artificial intelligence-based cloud resource scheduling device 300 is provided, comprising:

[0104] The basic data acquisition module 301 can be used to acquire cloud resource status data and workload demand data.

[0105] The task type identification module 302 can be used to input workload requirement data into the task type identification model and identify the workload task type.

[0106] The baseline state prediction module 303 can be used to set the resource state prediction time period of the cloud resource state prediction model based on the workload loading reference time period corresponding to the workload task type, and input the cloud resource state data into the cloud resource state prediction model to obtain cloud resource baseline state prediction information.

[0107] The load resource scheduling module 304 can be used to perform cloud resource scheduling based on cloud resource baseline status prediction information, workload task type and workload demand data, and generate workload cloud resource scheduling result information.

[0108] In an optional embodiment of this application, the baseline state prediction module 303 may also be used for:

[0109] The candidate resource status prediction time period with the smallest difference from the reference time period for the same workload loading in the candidate set of resource status prediction time periods is set as the resource status prediction time period of the cloud resource status prediction model.

[0110] Based on cloud resource status data and the status monitoring time of cloud resource status data, a cloud resource status time series is constructed.

[0111] The cloud resource status time series is input into the cloud resource status prediction model to generate the resource status prediction series within the resource status prediction time period.

[0112] Based on the maximum utilization rate of each cloud resource status dimension in the resource status prediction sequence, the baseline cloud resource status prediction information is obtained.

[0113] In an optional embodiment of this application, the load resource scheduling module 304 may also be used for:

[0114] Based on cloud resource baseline status prediction information, workload task types, and workload demand data, workload virtual machine-level resource scheduling and allocation are performed to generate workload virtual machine-level resource scheduling information.

[0115] Based on workload virtual machine-level resource scheduling information and workload demand data, workload container-level resource scheduling and allocation are performed to generate workload container-level resource scheduling information.

[0116] Based on workload container-level resource scheduling information and workload demand data, workload task-level resource scheduling and allocation are performed to generate workload task-level resource scheduling information.

[0117] By aggregating workload virtual machine-level resource scheduling information, workload container-level resource scheduling information, and workload task-level resource scheduling information, the workload cloud resource scheduling result information is obtained.

[0118] In an optional embodiment of this application, the task type identification module 302 can also be used for:

[0119] A cloud resource demand data feature vector is constructed based on cloud resource demand data, and a task business code feature vector is constructed based on task business type labels. The cloud resource demand data feature vector and the task business code feature vector are then concatenated to obtain the workload demand feature vector.

[0120] Input the workload requirement feature vector into the task type recognition model to generate the workload task type corresponding to the workload requirement data.

[0121] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the artificial intelligence-based cloud resource scheduling method described above.

[0122] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0123] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and 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 modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0124] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A cloud resource scheduling method based on artificial intelligence, characterized in that, The method includes: Obtain cloud resource status data and workload demand data; The workload requirement data is input into the task type identification model to identify the workload task type; Based on the workload loading reference time period corresponding to the workload task type, the resource status prediction time period of the cloud resource status prediction model is set, and the cloud resource status data is input into the cloud resource status prediction model to obtain cloud resource baseline status prediction information. Based on the cloud resource baseline state prediction information, the workload task type, and the workload demand data, cloud resource scheduling is performed to generate workload cloud resource scheduling result information.

2. The method according to claim 1, characterized in that, The process involves setting the resource status prediction time period of the cloud resource status prediction model based on the workload loading reference time period corresponding to the workload task type, and inputting the cloud resource status data into the cloud resource status prediction model to obtain cloud resource baseline status prediction information, including: The candidate resource status prediction time period with the smallest difference from the same workload loading reference time period in the candidate set of resource status prediction time periods is set as the resource status prediction time period of the cloud resource status prediction model. Based on the cloud resource status data and the status monitoring time of the cloud resource status data, a cloud resource status time series is constructed. The cloud resource status time series is input into the cloud resource status prediction model to generate a resource status prediction series within the resource status prediction time period; Based on the maximum utilization rate of each cloud resource status dimension in the resource status prediction sequence, the cloud resource baseline status prediction information is obtained.

3. The method according to claim 2, characterized in that, The expression for the cloud resource status prediction loss function of the cloud resource status prediction model is as follows: In the formula, The loss function for predicting the cloud resource status is... and These are the time-series trend prediction weights and the maximum utilization rate prediction weights, respectively. This represents the total number of cloud resource state training samples. For the first The resource status prediction time period for each of the cloud resource status training samples. This represents the total number of cloud resource status dimensions. For the first The cloud resource status training sample within the resource status prediction time period is the [number]th [sample]. The first time step The actual label value of the state dimension of the cloud resource state dimension. The first generation of the cloud resource status prediction model The first of the resource state prediction sequences of the cloud resource state training samples The first time step The predicted values ​​for each of the aforementioned cloud resource status dimensions. For the first The cloud resource status training sample within the resource status prediction time period is the [number]th [sample]. The maximum utilization rate true label value of the maximum utilization rate of each of the cloud resource status dimensions. The first generation of the cloud resource status prediction model The first of the resource state prediction sequences of the cloud resource state training samples The maximum utilization prediction value for the maximum utilization of each cloud resource status dimension.

4. The method according to claim 1, characterized in that, The process of scheduling cloud resources based on the cloud resource baseline state prediction information, the workload task type, and the workload demand data, and generating workload cloud resource scheduling result information, includes: Based on the cloud resource baseline state prediction information, the workload task type and the workload demand data, workload virtual machine-level resource scheduling and allocation are performed to generate workload virtual machine-level resource scheduling information. Based on the workload virtual machine-level resource scheduling information and the workload demand data, the workload container-level resource scheduling and allocation are performed to generate workload container-level resource scheduling information. Based on the workload container-level resource scheduling information and the workload demand data, the workload task-level resource scheduling and allocation are performed to generate workload task-level resource scheduling information. By summarizing the workload virtual machine-level resource scheduling information, the workload container-level resource scheduling information, and the workload task-level resource scheduling information, the workload cloud resource scheduling result information is obtained.

5. The method according to any one of claims 1 to 4, characterized in that, The workload requirement data includes cloud resource requirement data and task service type tags. The step of inputting the workload requirement data into the task type identification model to identify the workload task type includes: Based on the cloud resource demand data, a cloud resource demand data feature vector is constructed, and based on the task business type label, a task business encoding feature vector is constructed. The cloud resource demand data feature vector and the task business encoding feature vector are then concatenated to obtain the workload demand feature vector. The workload requirement feature vector is input into the task type identification model to generate the workload task type corresponding to the workload requirement data.

6. The method according to claim 5, characterized in that, The task type recognition model is a multilayer perceptron model. The task type recognition loss function of the model includes a cloud resource feature regularization loss function and a task type similarity correction loss function. The expressions for the cloud resource feature regularization loss function and the task type similarity correction loss function are as follows: In the formula, and These are the cloud resource feature regularization loss function and the task type similarity correction loss function, respectively. To identify the total number of training samples for task type identification, the training samples should be identified. The weight matrix of the last fully connected layer of the model for identifying the task type. For the first The workload requirement feature vector of each task type identification training sample. The total number of workload task types. The task type identification model for the first task The training sample for task type recognition belongs to the first... The predicted probability value for the workload task type described above. The task type identification model for the first task The training sample for task type recognition belongs to the first... The predicted probability value for the workload task type described above. For the first The workload task types described in this class are the same as those in the first class. The resource requirement similarity coefficient of the workload task type described in the class.

7. A cloud resource scheduling device based on artificial intelligence, characterized in that, The device includes: The basic data acquisition module is used to acquire cloud resource status data and workload demand data; The task type identification module is used to input the workload requirement data into the task type identification model and identify the workload task type; The baseline state prediction module is used to set the resource state prediction time period of the cloud resource state prediction model based on the workload loading reference time period corresponding to the workload task type, and input the cloud resource state data into the cloud resource state prediction model to obtain cloud resource baseline state prediction information. The load resource scheduling module is used to perform cloud resource scheduling based on the cloud resource baseline status prediction information, the workload task type, and the workload demand data, and generate workload cloud resource scheduling result information.

8. 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 method of any one of claims 1 to 6.

9. 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 method of any one of claims 1 to 6.

Images