Search

CN-121979671-A - Cloud machine resource scheduling method and device, electronic equipment and storage medium

CN121979671ACN 121979671 ACN121979671 ACN 121979671ACN-121979671-A

Abstract

The cloud machine resource scheduling method and device, the electronic equipment and the storage medium relate to the technical field of cloud computing, and target intention labels which accurately represent application scenes and user intentions are generated through acquisition and intelligent analysis of user operation data by an end side, so that basis is provided for cloud machine resource scheduling. The target intention label can effectively distinguish resource demand differences of different application types and dynamically reflect load changes caused by user operation, so that the cloud can implement dynamic resource allocation matched with real demands according to real-time and refined intention information, and the problem of resource supply and demand mismatch caused by a static allocation scheme in the related technology is solved. The dynamic alignment of cloud machine resource supply and actual complex demands is realized, the overall utilization efficiency and scheduling accuracy of resources are obviously improved, and therefore, the resource redundancy and the operation cost are effectively reduced while the service quality is ensured.

Inventors

  • ZHOU HEYU
  • YU WENQING
  • CHEN ZHENXU

Assignees

  • 中移互联网有限公司
  • 中国移动通信集团有限公司

Dates

Publication Date
20260505
Application Date
20251230

Claims (18)

  1. 1. The cloud machine resource scheduling method is characterized by being applied to an end side application layer and comprising the following steps of: Collecting operation data generated in the process of operating the clouding application, wherein the operation data comprises time sequence behavior data reflecting the use habit of the clouding application and static data identifying the application type of the clouding application, and the behavior data and the static data both comprise at least one type of operation content; Classifying and labeling the operation data to obtain intention label labels corresponding to various operation contents in the operation data; Inputting the intention label mark and the operation data into a locally trained intention recognition model for analysis and processing to obtain a target intention label representing current use information of the clouding application; and transmitting the target intention label to a cloud side service layer so that the cloud side service layer performs differentiated resource prediction and resource scheduling based on the target intention label.
  2. 2. The cloud machine resource scheduling method according to claim 1, wherein the collecting operation data generated in the operation clouding application process includes: responding to starting the clouding application, and directly extracting the static data of the clouding application, wherein the static data at least comprises application context information of the clouding application; acquiring gesture coordinates in the cloud application process to obtain gesture track information, and performing coding fuzzy processing on the gesture track information to obtain a gesture track sequence, wherein the time sequence behavior data at least comprises the gesture track sequence and a sight focus area sequence; And acquiring eyeball fixation point data and eyeball movement data in the cloud application process through an image acquisition device for acquiring preset rights, and generating the sight line focus region sequence according to the eyeball fixation point data and the eyeball movement data.
  3. 3. The cloud machine resource scheduling method according to claim 1, wherein the classifying and labeling the operation data to obtain the intent label labels corresponding to each type of operation content in the operation data comprises: performing behavior pattern recognition on the various operation contents in the operation data respectively to obtain behavior patterns corresponding to the various operation contents; And marking the various operation contents based on the behavior mode to obtain the intention label marks corresponding to the various operation contents.
  4. 4. The cloud machine resource scheduling method according to claim 1, wherein the inputting the intent label and the operation data into a locally trained intent recognition model for analysis processing, and obtaining a target intent label representing current usage information of the cloud application comprises: Preprocessing and feature extraction are carried out on the time sequence behavior data to obtain time sequence feature vectors, and feature coding is carried out on the static data to obtain type feature vectors; Inputting the time sequence feature vector into a long-period memory network, updating the hidden state of the time sequence feature vector through a gating mechanism of the long-period memory network to obtain a target hidden state vector, wherein the intention recognition model comprises the long-period memory network; And splicing the type feature vectors of the target hidden state vector to obtain a target feature vector, and performing classification calculation on the target feature vector to obtain the target intention label.
  5. 5. The cloud machine resource scheduling method of claim 1, further comprising: Acquiring training operation data comprising a real intention label, and training an intention recognition model based on the training operation data to obtain a locally trained intention recognition model; The training process of the intention recognition model comprises the steps of using classification cross entropy as a loss function, calculating the difference between a predicted intention label predicted by the intention recognition model and the real intention label, and updating model parameters of the intention recognition model by using a back propagation algorithm.
  6. 6. The cloud machine resource scheduling method of claim 5, wherein after performing training processing on the intent recognition model based on the training operation data to obtain the locally trained intent recognition model, the method further comprises: Acquiring parameter updating information of the intention recognition model in the model training process, and encrypting the parameter updating information to obtain encrypted updating information; uploading the encrypted update information to the cloud side service layer; And receiving aggregated model parameters sent by the cloud side service layer, and updating the locally trained intention recognition model based on the aggregated model parameters, wherein the aggregated model parameters are obtained by the cloud side service layer through aggregation and updating based on the encrypted updating information.
  7. 7. The cloud machine resource scheduling method is characterized by being applied to a cloud side service layer and comprising the following steps of: Receiving target intention labels uploaded by a plurality of end side application layers respectively, and acquiring local hardware state information and network state information; Constructing space-time diagram data by taking the plurality of end-side application layers as nodes and the target intention labels as time sequence change characteristics of the nodes; inputting the space-time diagram data, the hardware state information and the network state information into a locally trained space-time network model to perform space-time joint modeling to obtain predicted resource distribution aiming at a plurality of cloud applications in a preset future time period; And executing a dynamic resource allocation strategy according to the predicted resource distribution, and allocating differentiated cloud machine resources for the plurality of cloud application.
  8. 8. The cloud machine resource scheduling method according to claim 7, wherein the inputting the space-time diagram data, the hardware state information and the network state information into a locally trained space-time network model to perform space-time joint modeling, and obtaining the predicted resource distribution for a plurality of cloud applications in a preset future time period comprises: performing graph rolling operation on target time sequence change characteristics of a target node in each time step, and aggregating time sequence change characteristics of neighbor nodes associated with the target node to generate a target characteristic sequence with enhanced spatial context of the target node in a continuous time window, wherein the target node is any node in the space-time graph data; inputting the target feature sequence to a gating circulation unit of the local trained space-time network model, and learning a dynamic evolution rule in a time dimension to obtain a space-time fusion feature; Mapping the space-time fusion characteristic into a resource demand predicted value, and generating a resource thermodynamic diagram representing the space and time distribution of the resource demand based on the resource demand predicted value, wherein the resource thermodynamic diagram comprises the predicted resource distribution.
  9. 9. The cloud machine resource scheduling method according to claim 8, wherein said executing a dynamic resource allocation policy according to the predicted resource distribution, allocating differentiated cloud machine resources for the plurality of cloud applications comprises: a basic resource guarantee area is allocated for the plurality of cloud applications, and the basic resource guarantee area is used for providing fixed quota resources meeting respective minimum operation requirements of the plurality of cloud applications; a thermodynamic pre-allocation resource pool is configured for the plurality of cloud applications, and cloud machine resources in the thermodynamic pre-allocation resource pool are dynamically allocated to the plurality of cloud applications according to the predicted resource distribution; And configuring a backup elastic resource pool for the plurality of cloud applications, wherein the backup elastic resource pool is used for supplementing resources when the predicted resource distribution indicates that cloud machine resources in the thermal pre-allocation resource pool are insufficient or the resource demand of the cloud applications is increased.
  10. 10. The cloud machine resource scheduling method according to claim 9, wherein dynamically allocating cloud machine resources in the thermal preconditioning resource pool to the clouding application according to the predicted resource distribution comprises: Determining the resource allocation priority corresponding to each of the plurality of cloud applications according to the size of the resource demand predicted value corresponding to each of the plurality of cloud applications in the predicted resource distribution, wherein the larger the resource demand predicted value is, the higher the resource allocation priority is; and dynamically distributing cloud machine resources in the thermal pre-distribution resource pool to the cloud application based on the resource distribution priority.
  11. 11. The cloud machine resource scheduling method of claim 7, further comprising: Acquiring training resource data, wherein the training resource data at least comprises a historical intention label, historical hardware state information and historical network state information of a historical period; inputting the training resource data into a space-time network model for space-time joint modeling to obtain training prediction resource distribution of a preset subsequent period; And taking the actual resource distribution of the preset subsequent period as a training target, and optimizing parameters of the space-time network model by minimizing errors between the predicted resource distribution for training and the training target to obtain the local trained space-time network model.
  12. 12. The cloud machine resource scheduling method of claim 7, further comprising: Receiving encryption update information uploaded by each of the plurality of terminal application layers, and performing information aggregation update processing on the plurality of encryption update information to obtain aggregated model parameters; And respectively sending the aggregated model parameters to the plurality of end-side application layers.
  13. 13. The cloud machine resource scheduling device is characterized by comprising: The cloud application processing system comprises an acquisition unit, a cloud application processing unit and a cloud application processing unit, wherein the acquisition unit is used for acquiring operation data generated in the process of operating the cloud application, the operation data comprises time sequence behavior data reflecting the use habit of the cloud application and static data identifying the application type of the cloud application, and the behavior data and the static data comprise at least one type of operation content; The labeling unit is used for carrying out classification labeling processing on the operation data to obtain intention label labels corresponding to various operation contents in the operation data; The analysis unit is used for inputting the intention label mark and the operation data into a locally trained intention recognition model for analysis and processing to obtain a target intention label representing the current use information of the clouding application; and the transmission unit is used for transmitting the target intention label to a cloud side service layer so that the cloud side service layer provides basis for differentiated resource prediction and resource scheduling based on the target intention label.
  14. 14. The cloud machine resource scheduling device is characterized by comprising: The receiving unit is used for receiving target intention labels uploaded by the plurality of end-side application layers respectively; the acquisition unit is used for acquiring local hardware state information and network state information; The construction unit is used for constructing space-time diagram data by taking the plurality of end-side application layers as nodes and the target intention labels as time sequence change characteristics of the nodes; The modeling unit is used for inputting the space-time diagram data, the hardware state information and the network state information into a locally trained space-time network model to perform space-time joint modeling so as to obtain predicted resource distribution aiming at a plurality of cloud applications in a preset future time period; and the allocation unit is used for executing a dynamic resource allocation strategy according to the predicted resource distribution and allocating differentiated cloud machine resources for the cloud application.
  15. 15. A cloud machine resource scheduling system is characterized by comprising at least one end side application layer and a cloud side service layer, Each of the at least one terminal side application layer is configured with a cloud machine resource scheduling device according to claim 13; The cloud side service layer is provided with the cloud machine resource scheduling device according to claim 14; the cloud side service layer is in communication connection with the cloud side application layer.
  16. 16. An electronic device, comprising: at least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6 or the method of any one of claims 7-12.
  17. 17. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-6 or the method of any one of claims 7-12.
  18. 18. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1-6 or the method according to any one of claims 7-12.

Description

Cloud machine resource scheduling method and device, electronic equipment and storage medium Technical Field The present application relates to the field of cloud computing technologies, and in particular, to a cloud machine resource scheduling method and apparatus, an electronic device, and a storage medium. Background The cloud mobile phone technology provides remote mobile experience for users by virtualizing a mobile phone operating system in the cloud. Clouding applications, as their lightweight form, focus on providing a single application service. Current resource scheduling schemes typically employ a static partitioning strategy, i.e., a pre-partitioning of a physical server into several equal logical resource units. When the cloud application is started, the scheduling system searches for an idle unit which is provided with the corresponding application in the resource pool for overall allocation. However, the current resource scheduling scheme uniformly allocates fixed specification resources to cause idle low-load application resources and possibly insufficient high-load application without regard to the differentiated requirements of the application on resources such as a central processing unit (Central Processing Unit, a CPU), a graphic processor (Graphics Processing Unit, a GPU) and a memory, and meanwhile, ignores load fluctuation caused by user operation, and cannot realize accurate resource supply. The problems of low resource utilization rate and low allocation efficiency are caused, so that the operation cost is increased, and the user experience is possibly influenced. Therefore, how to realize dynamic and accurate scheduling of cloud machine resources to improve the resource utilization rate and the allocation efficiency is a problem to be solved at present. Disclosure of Invention The application provides a cloud machine resource scheduling method and device, electronic equipment and a storage medium. The method mainly aims at solving the problem of how to realize dynamic and accurate scheduling of cloud machine resources so as to improve the utilization rate and the distribution efficiency of the resources. According to a first aspect of the present application, there is provided a cloud machine resource scheduling method, where the method is applied to an end side application layer, and includes: Collecting operation data generated in the process of operating the clouding application, wherein the operation data comprises time sequence behavior data reflecting the use habit of the clouding application and static data identifying the application type of the clouding application, and the behavior data and the static data comprise at least one type of operation content; classifying and labeling the operation data to obtain intention label labels corresponding to various operation contents in the operation data; The intention label marking and the operation data are input into a locally trained intention recognition model for analysis and processing, and a target intention label representing current use information of the cloud application is obtained; and transmitting the target intention label to the cloud side service layer so that the cloud side service layer performs differentiated resource prediction and resource scheduling based on the target intention label. According to a first aspect of the present application, there is provided a cloud machine resource scheduling method, where the method is applied to a cloud side service layer, and includes: Receiving target intention labels uploaded by a plurality of end side application layers respectively, and acquiring local hardware state information and network state information; Constructing time-space diagram data by taking a plurality of end side application layers as nodes and taking a target intention label as a time sequence change characteristic of the nodes; the method comprises the steps of inputting space-time diagram data, hardware state information and network state information into a locally trained space-time network model to carry out space-time joint modeling to obtain predicted resource distribution aiming at a plurality of cloud applications in a preset future time period; And executing a dynamic resource allocation strategy according to the predicted resource distribution, and allocating differentiated cloud machine resources for a plurality of cloud applications. According to a third aspect of the present application, there is provided a cloud machine resource scheduling device, where the device is configured in an end side application layer, and includes: The cloud application processing system comprises an acquisition unit, a cloud application processing unit and a cloud application processing unit, wherein the acquisition unit is used for acquiring operation data generated in the process of operating the cloud application, the operation data comprises time sequence behavior data reflecting the use habit of the clo