Computing power scheduling processing method and system for cloud resources

Abstract

The invention provides a computing power scheduling processing method and system for cloud resources, and belongs to the technical field of cloud scheduling, and the method specifically comprises the steps: determining a matching training demand of an available workspace based on the available training resources of the available workspace, according to historical training data corresponding to different matching training requirements, training frequency coefficients and frequency training requirements of different matching training requirements are determined, the number of available working spaces corresponding to different frequency training requirements is determined, and the number of available working spaces corresponding to different frequency training requirements is determined in combination with the frequency training coefficients of different matching training requirements. According to the method, the training demand weight coefficients of different frequent training demands are determined, the matched workspace in the available workspaces is determined on the basis of the training demand weight coefficients of different available workspaces in different frequent training demands, and the matched workspace is pushed to the client, so that the reliability of training processing is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of cloud scheduling, and in particular, relates to a computing power scheduling processing method and system for cloud resources. Background Art

[0002] In order to achieve computing power scheduling of cloud resources and meet customers' personalized model training needs, the invention patent application CN202010715540.3 "A resource scheduling method for hybrid cloud" calculates the delay time of different scheduling schemes and compares it with the longest processing time submitted by the user. After multiple rounds of optimization, cloud resource scheduling schemes that meet time constraints but are cheaper are continuously selected, thereby maximizing the utilization of private cloud resources to a great extent. However, there are the following technical problems: When performing computing power scheduling of cloud resources, existing technical solutions ignore the determination of personalized cloud resource scheduling strategies based on customer needs. Since customer needs vary greatly, if personalized cloud resource scheduling strategies cannot be determined based on customer needs, the reliability of cloud resource scheduling cannot be improved.

[0003] In response to the above technical problems, the present application specifically provides a method and system for scheduling and processing computing power for cloud resources. Summary of the invention

[0004] To achieve the purpose of the present invention, the present invention adopts the following technical solutions: According to one aspect of the present invention, a computing power scheduling and processing method for cloud resources is provided.

[0005] A computing power scheduling processing method for cloud resources, specifically comprising: S1 determines the training demand feature of the customer based on the training demand of the customer, determines similar training states in historical training states of the existing idle workspaces based on the training demand feature of the customer and existing training data of the existing idle workspaces, and determines available workspaces in the idle workspaces according to training problem data of different idle workspaces under similar training states; S2: determining the matching training requirements of the available workspace based on the available training resources of the available workspace, and determining the training frequency coefficients and frequent training requirements of different matching training requirements according to the historical training data corresponding to the different matching training requirements; S3 determines the amount of available workspace corresponding to different frequent training requirements, and determines the training requirement weight coefficients of different frequent training requirements in combination with the frequent training coefficients of different matching training requirements; S4 determines a matching workspace in the available workspace based on training demand weight coefficients of different available workspaces under different frequent training demands, and pushes the matching workspace to the client.

[0006] The beneficial effects of the present invention are: Based on the training problem data of different idle workspaces under similar training conditions, the available workspace in the idle workspace is determined, thereby realizing the determination of the available workspace in the idle workspace from the problem situations under similar training conditions in different idle workspaces, avoiding the occurrence of training problems due to mismatch of hardware devices in the idle workspace, and improving the reliability of customer training usage processing.

[0007] The matching workspace in the available workspace is determined based on the training demand weight coefficients of different available workspaces for different frequent training demands, thereby avoiding the influence on other frequent training demands caused by the use of the available workspace due to a large number of matched frequent training demands or a large number of frequent training demands with high training demand levels, thereby satisfying the use needs of other frequent training demands as well as the use needs of customers.

[0008] A further technical solution is that the training demand characteristic quantity of the customer includes a training model and a training data volume.

[0009] A further technical solution is that the existing training data of the idle workspace includes the number of GPUs in use, existing training models, and the amount of training data of different existing training models.

[0010] A further technical solution is that the method for determining the similar training state in the historical training state is: Determining the existing training data of the existing idle workspace as a basis for the existing training model of the idle workspace and the amount of training data of different existing training models; Determine a matching training model based on the training model of the customer and the existing training model, and determine a training deviation coefficient with different matching training models based on the deviation rate of the training data volume in different matching training models in the historical training state; A comprehensive deviation coefficient is determined based on an average value of training deviation coefficients of different matching training models, and a similar training state in the historical training state is determined using the comprehensive deviation coefficient.

[0011] A further technical solution is that the similar training states in the historical training states are historical training states whose comprehensive deviation coefficients are within a preset deviation coefficient range.

[0012] A further technical solution is that the method for determining the matching workspace in the available workspace is: Determining the screening training demand of the frequent training demand corresponding to the available workspace based on the training demand weight coefficient of the available workspace under different frequent training demands; It is determined whether the available workspace is a matching workspace based on the number of screening training requirements.

[0013] A further technical solution is that the screening training needs are frequent training needs whose training need weight coefficient is greater than a preset weight coefficient.

[0014] A further technical solution is that when the number of the screened training requirements is greater than a preset training requirement number threshold, the available workspace is determined to be a matching workspace.

[0015] In a second aspect, the present invention provides a computer system comprising: a memory and a processor that are communicatively connected, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the above-mentioned method for scheduling computing power for cloud resources when running the computer program.

[0016] Other features and advantages will be described in the following description. The objects and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description and drawings.

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and other features and advantages of the present invention will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings.

[0019] Figure 1 It is a flow chart of a computing power scheduling processing method for cloud resources; Figure 2 is a flow chart of a method for determining similar training states in historical training states; Figure 3 is a flow chart of a method for determining an available workspace in an idle workspace; Figure 4 is a flow chart of a method for determining a training frequency coefficient that matches training needs; Figure 5 It is a framework diagram of a computer system. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0021] When scheduling the computing power of cloud resources, due to the large differences in customer needs, it is necessary to determine personalized cloud resource scheduling strategies based on customer needs to improve the reliability of cloud resource scheduling.

[0022] Similar training status: Determine the matching training model based on the customer's training model and the existing training model. Determine the training deviation coefficient with different matching training models based on the deviation rate of the training data volume in different matching training models in the historical training status. Determine the comprehensive deviation coefficient based on the average value of the training deviation coefficient with different matching training models. The historical training status with a comprehensive deviation coefficient less than 0.2 is regarded as a similar training status.

[0023] Available workspace in idle workspace: Based on the training problem data of the idle workspace under similar training states, similar training states with training problems are determined and used as problem training states. When the proportion of problem training states in the number of similar training states is less than 0.1, the idle workspace is determined to be an available workspace.

[0024] Training frequency coefficient for matching training needs: Determine the historical training times corresponding to the matching training needs based on the historical training data corresponding to the matching training needs. Determine the training frequency coefficient for matching training needs based on the preset frequency coefficient corresponding to the historical training times. The matching training needs with a training frequency coefficient greater than 0.6 are regarded as frequent training needs.

[0025] Training demand weight coefficient: Based on the number of available workspaces corresponding to frequent training needs, the number of available spaces for the frequent training needs is determined, and the preset training availability coefficient of the frequent training needs is determined using the number of available spaces; based on the frequent training coefficients that match the training needs, the training demand weight coefficient of the frequent training needs is determined based on the ratio of the frequent training coefficient to the preset training availability coefficient.

[0026] Matching workspace: Based on the sum of the training demand weight coefficients of the available workspace under different frequent training demands, the sum of the weight coefficients of the available workspace is determined. When the sum of the weight coefficients of the available workspace is greater than the preset coefficient threshold, it is determined that the available workspace does not belong to the matching workspace. Embodiment 1 To solve the above problem, according to one aspect of the present invention, Figure 1 As shown, a first aspect is provided. The present invention proposes a computing power scheduling processing method for cloud resources, which specifically includes: S1 determines the training demand feature of the customer based on the training demand of the customer, determines similar training states in historical training states of the existing idle workspaces based on the training demand feature of the customer and existing training data of the existing idle workspaces, and determines available workspaces in the idle workspaces according to training problem data of different idle workspaces under similar training states; Furthermore, the training demand characteristics of the customer include a training model and a training data volume.

[0027] Specifically, the existing training data of the idle workspace includes the number of GPUs in use, existing training models, and amounts of training data of different existing training models.

[0028] Specifically, Figure 2 As shown, the method for determining the similar training state in the historical training state is: Determining the existing training data of the existing idle workspace as a basis for the existing training model of the idle workspace and the amount of training data of different existing training models; Determine a matching training model based on the training model of the customer and the existing training model, and determine a training deviation coefficient with different matching training models based on the deviation rate of the training data volume in different matching training models in the historical training state; A comprehensive deviation coefficient is determined based on an average value of training deviation coefficients of different matching training models, and a similar training state in the historical training state is determined using the comprehensive deviation coefficient.

[0029] Furthermore, the similar training states in the historical training states are historical training states whose comprehensive deviation coefficients are within a preset deviation coefficient range.

[0030] In another embodiment, the method for determining the similar training state in the historical training state is: Determining the existing training data of the existing idle workspace as a basis for the existing training model of the idle workspace and the amount of training data of different existing training models; Determine a matching training model based on the training model of the customer and the existing training model; A comprehensive deviation coefficient is determined based on the ratio of the number of deviations of the matching training model, and the similar training state in the historical training state is determined using the comprehensive deviation coefficient.

[0031] Furthermore, the comprehensive deviation coefficient is determined based on the ratio of the number of deviations of the matching training models to the number of matching training models.

[0032] In another embodiment, the method for determining the similar training state in the historical training state is: S11 determines the existing training model of the idle workspace and the amount of training data of different existing training models based on the existing training data of the existing idle workspace, determines the matching training model based on the training model of the customer and the existing training model, and determines the model quantity deviation ratio by using the deviation quantity ratio of the matching training model with the historical training state; Optionally, the above step S11 includes the following contents: S111 determines the existing training model of the idle workspace and the amount of training data of different existing training models based on the existing training data of the existing idle workspace, determines the matching training model based on the training model of the customer and the existing training model, and when the matching training model does not exist for the historical training state, determines that the historical training state does not belong to a similar training state, and when the matching training model exists for the historical training state, proceeds to step S112; S112 uses the matching training model that deviates from the historical training state as the deviation training model. When the number of the deviation training models does not meet the requirement, it is determined that the historical training state does not belong to the similar training state. When the number of the deviation training models meets the requirement, it proceeds to step S113; S113 determines the model quantity deviation ratio by the ratio of the number of the deviation training models in the matching training model. When the model quantity deviation ratio is greater than the preset ratio, the process proceeds to step S114. When the model quantity deviation ratio is not greater than the preset ratio, the process proceeds to step S12. S114: When the number of the deviation training models is greater than the preset number of deviation models, it is determined that the historical training state does not belong to a similar training state; when the number of the deviation training models is not greater than the preset number of deviation models, the process proceeds to step S12.

[0033] S12 determines the data volume deviation coefficient under different matching training models according to the deviation of the training data volume under different matching training models; Optionally, the above step S12 includes the following contents: S121 determines the total deviation of the training data amount according to the deviation of the training data amount under different matching training models. When the total deviation of the training data amount does not meet the requirement, it is determined that the historical training state does not belong to the similar training state. When the total deviation of the training data amount meets the requirement, it proceeds to step S122. S122 obtains the amount of training data under the deviation training model. When the ratio of the amount of training data under the deviation training model to the total amount of training data under the matching training model is greater than the preset data amount ratio, it is determined that the historical training state does not belong to the similar training state. When the ratio of the amount of training data under the deviation training model to the total amount of training data under the matching training model is not greater than the preset data amount ratio, proceed to step S123. S123 determines the data volume deviation coefficient under different matching training models according to the deviation of the training data volume under different matching training models. When there is a matching training model whose data volume deviation coefficient does not meet the requirements, the process proceeds to step S124. When there is no matching training model whose data volume deviation coefficient does not meet the requirements, the process proceeds to step S13. S124 When the number of matching training models that do not meet the requirements of the data volume deviation coefficient is greater than the preset model quantity threshold, it is determined that the historical training state does not belong to a similar training state. When the number of matching training models that do not meet the requirements of the data volume deviation coefficient is not greater than the preset model quantity threshold, proceed to step S13.

[0034] S13 determines a comprehensive deviation coefficient based on the data quantity deviation coefficient and the model quantity deviation ratio under different matching training models, and uses the comprehensive deviation coefficient to determine a similar training state in the historical training state.

[0035] Specifically, Figure 3 As shown, the method for determining the available workspace in the idle workspace is: Based on the training problem data of the idle workspace in the similar training state, a similar training state with a training problem is determined and used as the problem training state; Determining whether the idle workspace is an available workspace is based on the number of the problem training states.

[0036] Further, when the number of the problem training states is not greater than a preset number threshold, the idle workspace is determined to be an available workspace.

[0037] In another embodiment, the method for determining the available workspace in the idle workspace is: Based on the training problem data of the idle workspace under similar training states, similar training states with training problems are determined and used as problem training states. When the number of the problem training states is greater than a preset number of states, it is determined that the idle workspace does not belong to the available workspace. When the number of the problem training states is not greater than the preset number of states: Based on the training problem types under different problem training states, the number of training problem types under different problem training states is determined, and when there is a problem training state in which the number of training problem types does not meet the requirements, it is determined that the idle workspace does not belong to the available workspace; When there is no problem training state where the number of training problem types does not meet the requirement: Determine the training problem abnormality coefficients of different problem training states according to the number of training problem types under different problem training states and the number of occurrences of different training problem types, and when there is a problem training state in which the training problem abnormality coefficient does not meet the requirements, determine that the idle workspace does not belong to the available workspace; When there is no training problem and the training coefficient does not meet the requirements: The basic deviation coefficient is determined based on the number of the problem training states and the ratio of the number of the problem training states to the number of the similar training states. When the basic deviation coefficient does not meet the requirements: Obtaining average values ​​of training problem anomaly coefficients of different problem training states, and when the average values ​​of training problem anomaly coefficients of different problem training states do not meet the requirements, determining that the idle workspace does not belong to the available workspace; When the basic deviation coefficient meets the requirement or the average value of the training problem abnormality coefficients of different problem training states meets the requirement: An abnormal coefficient evaluation amount is determined according to the basic deviation coefficient and the average value of the abnormal coefficients of the training problems in different problem training states, and the abnormal coefficient evaluation amount is used to determine whether the idle workspace is an available workspace.

[0038] Furthermore, the training problem types include insufficient GPU video memory space, insufficient memory space, and excessive GPU load rate.

[0039] S2: determining the matching training requirements of the available workspace based on the available training resources of the available workspace, and determining the training frequency coefficients and frequent training requirements of different matching training requirements according to the historical training data corresponding to the different matching training requirements; Specifically, the matching training requirements of the available workspace are based on the training requirements that can be met by the available training resources corresponding to the available workspace.

[0040] Optionally, if Figure 4 As shown, the method for determining the training frequency coefficient matching the training requirements is: Determine the number of historical training times corresponding to the matching training requirement using the historical training data corresponding to the matching training requirement; Based on the preset frequency coefficient corresponding to the historical training times, the training frequency coefficient matching the training requirements is determined.

[0041] Furthermore, the training frequency coefficient of the matching training requirement has a value range between 0 and 1, wherein when the training frequency coefficient of the matching training requirement is greater than a preset frequency coefficient, the matching training requirement is determined to be a frequent training requirement.

[0042] S3 determines the amount of available workspace corresponding to different frequent training requirements, and determines the training requirement weight coefficients of different frequent training requirements in combination with the frequent training coefficients of different matching training requirements; Specifically, the method for determining the training demand weight coefficient of the frequent training demand is: Based on the number of available working spaces corresponding to the frequent training needs, determining the number of available spaces for the frequent training needs, and determining a preset training availability coefficient for the frequent training needs using the number of available spaces; Based on the frequent training coefficient of the matching training demand, the training demand weight coefficient of the frequent training demand is determined based on the ratio of the frequent training coefficient to the preset training availability coefficient.

[0043] Furthermore, the training demand weight coefficient of the frequent training demand ranges from 0 to 1, wherein the larger the training demand weight coefficient of the frequent training demand is, the higher the degree of demand of the frequent training demand for the available working interval is.

[0044] S4 determines a matching workspace in the available workspace based on training demand weight coefficients of different available workspaces under different frequent training demands, and pushes the matching workspace to the client.

[0045] Specifically, the method for determining the matching workspace in the available workspace is: Determining a sum of weight coefficients of the available workspace based on the sum of training demand weight coefficients of the available workspace under different frequent training demands; Based on the weight coefficients, it is determined whether the available workspace is a matching workspace.

[0046] Further, when the sum of the weight coefficients of the available workspace is greater than a preset coefficient threshold, it is determined that the available workspace does not belong to a matching workspace.

[0047] In another embodiment, the method for determining the matching workspace in the available workspace is: Determining the screening training demand of the frequent training demand corresponding to the available workspace based on the training demand weight coefficient of the available workspace under different frequent training demands; It is determined whether the available workspace is a matching workspace based on the number of screening training requirements.

[0048] Furthermore, the screening training demand is a frequent training demand whose training demand weight coefficient is greater than a preset weight coefficient.

[0049] Specifically, when the number of the screened training requirements is greater than a preset training requirement number threshold, the available workspace is determined to be a matching workspace.

[0050] In another embodiment, the method for determining the matching workspace in the available workspace is: S41: obtaining the number of matching training requirements of the available workspace, and determining the basic use requirement coefficient of the available workspace in combination with the number of available workspaces corresponding to different matching training requirements; Optionally, the above step S41 includes the following contents: S411 obtains the number of matching training requirements of the available workspace. When the number of matching training requirements of the available workspace is greater than a preset requirement number threshold, the available workspace is determined to be a matching workspace. When the number of matching training requirements of the available workspace is not greater than the preset requirement number threshold, the process proceeds to step S412. S412 obtains the number of available workspaces corresponding to different matching training requirements. When the number of available workspaces corresponding to different matching training requirements is greater than the preset number of workspaces, it is determined that the available workspace does not belong to the matching workspace. When there is a matching training requirement corresponding to which the number of available workspaces is not greater than the preset number of workspaces, the process proceeds to step S413. S413 determines the basic usage demand coefficient of the available workspace based on the number of matching training requirements of the available workspace and the number of available workspaces corresponding to different matching training requirements. When the basic usage demand coefficient of the available workspace is greater than a preset demand coefficient threshold, it is determined that the available workspace belongs to a matching workspace. When the basic usage demand coefficient of the available workspace is not greater than the preset demand coefficient threshold, the process proceeds to step S42.

[0051] S42, based on the training demand weight coefficients of the available workspace under different frequent training demands, determining the sum of the training demand weight coefficients of the available workspace, and using the sum of the weight coefficients; S43: determining a usage requirement coefficient of the available workspace based on the basic usage requirement coefficient and the weight coefficient of the available workspace, and determining whether the available workspace is a matching workspace by using the usage requirement coefficient. Embodiment 2 In the second aspect, as Figure 5 As shown, the present invention provides a computer system, comprising: a memory and a processor that are communicatively connected, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the above-mentioned computing power scheduling processing method for cloud resources when running the computer program.

[0052] Optionally, the above step S42 includes the following contents: S421: obtaining the frequent training demands of the available workspace. When the number of the frequent training demands of the available workspace is greater than the preset number of frequent demands, determining that the available workspace belongs to the matching workspace. When the number of the frequent training demands of the available workspace is not greater than the preset number of frequent demands, proceeding to step S422. S422: based on the training demand weight coefficients of the available workspace under different frequent training demands, if it is determined that there is a frequent training demand with a training demand weight coefficient greater than a preset weight coefficient, then the process proceeds to step S423; if there is no frequent training demand with a training demand weight coefficient greater than the preset weight coefficient, then the process proceeds to step S424; S423: taking the frequent training demands whose training demand weight coefficient is greater than the preset weight coefficient as screening training demands, and when the number of the screening training demands does not meet the requirement, determining that the available workspace belongs to the matching workspace, and when the number of the screening training demands meets the requirement, proceeding to step S424; S424 determines the sum of the training demand weight coefficients of the available workspace based on the training demand weight coefficients of the available workspace under different frequent training demands, and uses it as the sum of weight coefficients. When the sum of weight coefficients is greater than the preset weight coefficient threshold, it is determined that the available workspace belongs to the matching workspace. When the sum of weight coefficients is not greater than the preset weight coefficient threshold, proceeds to step S43.

[0053] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0054] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0055] The above description is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of this specification.

Claims

1. A computing power scheduling processing method for cloud resources, characterized in that: Specifically include: Based on the training needs of the customer, determine the training needs characteristic quantity of the customer, use the training needs characteristic quantity of the customer and the existing training data of the existing idle workspace as a basis to determine the similar training state in the historical training state of the existing idle workspace, and determine the available workspace in the idle workspace according to the training problem data of different idle workspaces under similar training states; Based on the available training resources of the available workspace, determining the matching training requirements of the available workspace, and determining the training frequency coefficients and frequent training requirements of different matching training requirements according to the historical training data corresponding to the different matching training requirements; Determine the amount of available workspace corresponding to different frequent training needs, and determine the training demand weight coefficients of different frequent training needs in combination with the frequent training coefficients of different matching training needs; Based on the training demand weight coefficients of different available workspaces under different frequent training demands, a matching workspace in the available workspace is determined, and the matching workspace is pushed to the client.

2. The computing power scheduling processing method for cloud resources according to claim 1, characterized in that: The training demand characteristics of the customer include a training model and a training data volume.

3. The computing power scheduling processing method for cloud resources according to claim 1, characterized in that: The existing training data of the idle workspace includes the number of GPUs in use, existing training models, and the amount of training data of different existing training models.

4. The computing power scheduling processing method for cloud resources according to claim 1, characterized in that: The method for determining the similar training state in the historical training state is: Determining the existing training data of the existing idle workspace as a basis for the existing training model of the idle workspace and the amount of training data of different existing training models; Determine a matching training model based on the training model of the customer and the existing training model, and determine a training deviation coefficient with different matching training models based on the deviation rate of the training data volume in different matching training models in the historical training state; A comprehensive deviation coefficient is determined based on an average value of training deviation coefficients of different matching training models, and a similar training state in the historical training state is determined using the comprehensive deviation coefficient.

5. The computing power scheduling processing method for cloud resources according to claim 4, characterized in that: The similar training states in the historical training states are historical training states whose comprehensive deviation coefficients are within a preset deviation coefficient range.

6. The computing power scheduling processing method for cloud resources according to claim 1, characterized in that: The method for determining the available workspace in the idle workspace is: Based on the training problem data of the idle workspace in the similar training state, a similar training state with a training problem is determined and used as the problem training state; Determining whether the idle workspace is an available workspace is based on the number of the problem training states.

7. The computing power scheduling processing method for cloud resources according to claim 6, characterized in that: When the number of the problem training states is not greater than a preset number threshold, the idle workspace is determined to be an available workspace.

8. The computing power scheduling processing method for cloud resources according to claim 1, characterized in that: The training problem types include insufficient GPU video memory space, insufficient memory space, and excessive GPU load rate.

9. The computing power scheduling processing method for cloud resources according to claim 1, characterized in that: The matching training requirements of the available workspace are based on the training requirements that can be met by the available training resources corresponding to the available workspace.

10. A computer system comprising: A memory and a processor connected in communication, and a computer program stored in the memory and capable of running on the processor, characterized in that when the processor runs the computer program, a computing power scheduling and processing method for cloud resources as described in any one of claims 1-9 is executed.

Images