CN116339992A - Resource allocation method, device, electronic device and storage medium - Google Patents

Abstract

The present invention discloses a resource allocation method, device, electronic equipment and storage medium, wherein the resource allocation method includes receiving target calculation types to be resource allocated, wherein different calculation types are used to distinguish calculations in different calculation modes; The target computing type allocates the target resource partition, where the target resource partition represents the set of resources allowed to be used when performing the calculation of the target computing type, and different computing types have their corresponding resource partitions; obtain the resource allocation strategy of the target resource partition, where different The resource partitions have their corresponding resource allocation policies; and allocate resources to the target resource partition according to the resource allocation policy of the target resource partition. Can improve server performance.

Description

Resource allocation method, device, electronic device and storage medium

technical field

The invention relates to the technical field of architectural engineering design, in particular to a resource allocation method, device, electronic equipment and storage medium.

Background technique

With the rapid development of computer information technology and the rapid increase in the amount of data in various information units, the current software application system is rapidly evolving in the direction of online, service and intelligence, which leads to a large number of server computing needs. To play the role of server computing, there are two prerequisites: the server can perform high-concurrency computing and sufficient computing resources. Therefore, when performing server computing, how to efficiently schedule server resources has become an important research topic.

At present, in servers, for different types of calculations, resource allocation is performed according to the utilization rate of resources such as CPU and memory. For data-intensive calculations, this type of calculation takes less time for a single calculation and occupies less computing resources. It is more suitable to allocate resources according to the utilization rate of resources such as CPU and memory. But for computing-intensive computing, this type of computing takes a long time, takes up more computing resources, and often fills up the CPU and memory. It is not special to allocate resources according to the utilization rate of CPU, memory and other resources. Be applicable. Simply put, it is because computing-intensive computing often occupies the CPU and memory. If resources are allocated according to the utilization rate of resources such as CPU and memory, resources will be continuously allocated to the corresponding computing resources, which will easily cause resource pool thrashing. And resource waste, thereby reducing server performance.

Contents of the invention

In view of this, the implementation method of the present invention provides a resource allocation method, a resource allocation device, an electronic device, and a computer-readable storage medium, which can improve server performance.

One aspect of the present invention provides a resource allocation method, the method comprising:

Receive the target computing type for resource allocation, where different computing types are used to distinguish the computing of different computing methods;

Allocating a target resource partition for the target computing type, wherein the target resource partition represents a set of resources allowed to be used when performing the calculation of the target computing type, and different computing types have respective corresponding resource partitions;

Obtaining resource allocation strategies of the target resource partition, wherein different resource partitions have respective corresponding resource allocation strategies; and

Allocate resources to the target resource partition according to the resource allocation policy of the target resource partition.

In the technical solutions of some embodiments of the present application, resources are allocated to corresponding resource partitions by assigning corresponding resource partitions to different computing types, and based on resource allocation policies of each resource partition. In this way, when performing calculations of different calculation types, resource allocation can be performed according to their corresponding resource allocation strategies, which avoids resource waste when calculations of different calculation types share the same resource allocation strategy, thereby improving server performance.

In some embodiments, the resource allocation strategy includes dynamically allocating the first dynamic resource to the target resource partition based on the time period of the current time;

The allocating resources to the target resource partition includes:

If the current time is within the first time period, allocate the first dynamic resource to the target resource partition;

If the current time is within a second time period other than the first time period, releasing the first dynamic resource from the target resource partition.

Based on the time period of the current time, the first dynamic resource is dynamically allocated to the target resource partition, which can reduce waste of resources while meeting business requirements.

In some embodiments, the first time period and the second time period are determined based on the following method:

Obtaining a historical time distribution corresponding to the calculation of the target calculation type based on historically executed calculations of the target calculation type;

Based on the historical time distribution, determine the peak time period and the valley time period when the calculation of the target calculation type is executed, and use the peak time period as the first time period, and use the valley time period as the first time period. the second time period.

Dynamically allocate the first dynamic resource to the target resource partition according to the peak time period and valley time period, which can guarantee the basic computing requirements when the target computing type has many computing requests; and when the target computing type has fewer computing requests , to release the first dynamic resource, so as to achieve the purpose of saving resources.

In some embodiments, the resource allocation strategy includes dynamically allocating the second dynamic resource to the target resource partition based on the job queue of the target resource partition, wherein the job queue includes job requests, and each job request represents a computation to be performed of said target computation type;

The allocating resources to the target resource partition includes:

predicting the total time required when all the job requests in the job queue are executed;

If the total duration exceeds a duration threshold, allocate the second dynamic resource for the target resource partition.

In this way, resource waste caused by irrationally allocating resources to the target resource partition when resources in the target resource partition are fully occupied can be prevented.

In some embodiments, the calculation of the target calculation type is performed by a calculation model, and the job request includes the parameter file required for performing the calculation of the target calculation type, and the model required for calling the calculation model call parameters;

The prediction of the total time required for all job requests in the job queue to be executed includes:

The size of the parameter file corresponding to each job request in the job queue and the model calling parameters are input into the predicted model that has been trained, and the predicted model outputs the total duration.

Based on the size of the parameter file corresponding to each job request and the total duration of the model call parameter prediction, the predicted total duration can be more consistent with the actual time, and the prediction accuracy can be improved.

In some embodiments, before the total duration is output by the prediction model, the prediction model is trained based on the following method:

Obtain the historical parameter files, historical model call parameters, and historical execution time corresponding to the executed historical job requests;

The prediction model is obtained through training using the historical parameter file, the calling parameters of the historical model, and the historical execution time as training data.

Using historical data as training data, the way to obtain training data is relatively simple.

In some embodiments, before training the predictive model, the method further includes:

Perform normalization processing on at least part of the training data.

In this way, data with dimensional differences in the training data can be avoided and the training accuracy can be improved.

In some embodiments, the resource allocation strategy includes allocating resources valid within a preset time period to the target resource partition;

The allocating resources to the target resource partition includes:

Allocating a preset number of target resources to the target resource partition, and setting the target resources to be released after the preset duration.

In this way, within a preset period of time, based on the target resources, basic calculations of the target computing type can be performed to prevent calculation requests of other computing types from taking up all the resources and making it impossible to perform the calculation of the target computing type.

Another aspect of the present invention also provides a resource allocation device, which includes:

A receiving module, configured to receive target computing types to be allocated for resources, wherein different computing types are used to distinguish computing in different computing modes;

A partition allocation module, configured to allocate a target resource partition for the target computing type, wherein the target resource partition represents a set of resources allowed to be used when performing the calculation of the target computing type, and different computing types have respective corresponding resource partitions ;

a policy acquisition module, configured to acquire the resource allocation policy of the target resource partition, wherein different resource partitions have their own corresponding resource allocation policies; and

A resource allocation module, configured to allocate resources to the target resource partition according to the resource allocation policy of the target resource partition.

Another aspect of the present invention also provides an electronic device, the electronic device includes a processor and a memory, the memory is used to store a computer program, and when the computer program is executed by the processor, the above-mentioned method is realized .

Another aspect of the present invention also provides a computer-readable storage medium, the computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the above-mentioned method is realized.

Description of drawings

The features and advantages of the present invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the invention in any way. In the accompanying drawings:

FIG. 1 shows a schematic flowchart of a resource allocation method provided by an embodiment of the present application;

FIG. 2 shows a schematic module diagram of a resource allocation device provided by an embodiment of the present application;

Fig. 3 shows a schematic diagram of an electronic device provided by an embodiment of the present application.

Detailed ways

In order to make the purpose, technical solution and advantages of the implementation method of the present invention clearer, the technical solutions in the implementation method of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the implementation method of the present invention. Obviously, the described implementation method It is a part of the implementation method of the present invention, but not all of the implementation methods. Based on the implementation method in the present invention, all other implementation methods obtained by those skilled in the art without creative work belong to the protection scope of the present invention.

The resource allocation method provided in this application can be applied to servers with computing capabilities. In this embodiment, the server includes a cloud server. The cloud server can be used to perform calculations related to the field of construction engineering, such as calculating the amount of steel bars and engineering quantities. Specifically, for different types of computing, the cloud server may open different interfaces. Users can make the cloud server perform different types of calculations by calling the interface.

For example, the cloud server may open interface A and interface B. Among them, interface A is used to realize the logical calculation related to the amount of reinforcement, and interface B is used to realize the logical calculation related to the engineering quantity. When it is necessary to calculate the amount of reinforcement in a construction project, the user can call the interface A through the local computer, and send the parameters related to the calculation of the amount of reinforcement in the construction project (such as building height) to the cloud server, and the cloud server will calculate the amount of reinforcement. calculation, and returns the result of the calculation. Similarly, when the engineering quantity calculation is required in the construction project, the user can call the interface B through the local computer, and send the parameters related to the engineering quantity calculation in the construction project to the cloud server, and the cloud server will calculate the engineering quantity , and returns the result of the calculation. In this way, the relevant calculation logic in the field of construction engineering can be placed in the cloud server, and the cloud server provides the resources required for calculation (such as CPU, memory, etc.), and then it can also be used when the resource configuration of the user's local computer is low. Completion of calculations with high computing power requirements.

Wherein, the process of invoking a cloud server to perform calculation may also be referred to as cloud computing. In the above related description about cloud computing, computing related to the field of construction engineering is taken as an example for illustration. It can be understood that the computing that the cloud server can perform includes but is not limited to the field of construction engineering.

Generally, cloud servers can perform multiple calculations in parallel, and each calculation can be called a computing task. The cloud server can allocate corresponding resources for each computing task to ensure the normal execution of the computing. Wherein, the resource includes but not limited to CPU, memory, hard disk and so on. Under normal circumstances, more resources can be allocated for computing tasks that require a large amount of calculation or higher computing speed to ensure that these computing tasks can be completed in time; For low computing tasks, resource allocation can be appropriately reduced to reduce resource waste. In some technologies, the resource usage of each computing task is monitored to allocate resources to the computing tasks. For example, when computing task A is executed, if the resources of computing task A are in a high-load state, then resources can be added to computing task A to ensure the normal execution of computing task A. But in some scenarios, this resource allocation method is not particularly suitable. For example, for computing-intensive computing, this type of computing takes a long time, takes up more computing resources, and often fills up the CPU and memory. If resources are allocated according to the utilization rate of resources such as CPU and memory, It is obviously unreasonable to continuously allocate new resources to these computing tasks, and it may also cause resource pool thrashing and resource waste, thereby reducing server performance.

In view of this, please refer to FIG. 1 , which is a schematic flowchart of a resource allocation method provided by an embodiment of the present application. In Fig. 1, the resource allocation method includes the following steps:

Step S11, receiving target computing types to be allocated resources, where different computing types are used to distinguish computing in different computing modes.

Computational methods may represent computational logic. Calculations of different calculation types can have different calculation logics. For example, take the field of construction engineering as an example. The calculation of reinforcement quantity and engineering quantity has different calculation logic. Then the calculation of the amount of reinforcement can be used as a calculation type. Calculations of quantities can be considered as another calculation type.

In this embodiment, each calculation type may have a corresponding calculation model. Each calculation model is used to realize the calculation logic of the corresponding calculation type. For example, the calculation model A includes the calculation logic of the steel bar quantity, and is used to realize the calculation of the steel bar quantity, and the calculation model B includes the calculation logic of the engineering quantity, and is used to realize the calculation of the engineering quantity. When it is necessary to calculate the amount of reinforcement, you can run the calculation model A to perform the calculation; when you need to perform the calculation of the engineering quantity, you can run the calculation model B to perform the calculation.

Specifically, before invoking the cloud server to perform a calculation of a calculation type, a calculation model corresponding to the calculation type may be created first. In this way, when the cloud server receives a calculation request of this calculation type, it can run the corresponding calculation model for calculation. A computational model can have input parameters and output parameters. The input parameter may be a parameter required in the calculation process, and the output parameter may be the calculation result. The values of the input parameters are different, and the calculation results obtained are different. For example, taking the field of construction engineering as an example, it is assumed that the number of floors F and the building area S are input parameters when calculating the engineering quantity. When calculating the quantity of project A, the value of the number of floors F is 20, and the value of building area S is 400, then the result obtained after running the calculation model may be 30; when calculating the quantity of project B, The value of the number of floors F is 10, and the value of the building area S is 100, then the result obtained after running the calculation model may be 10. When it is necessary to call the cloud server to perform a calculation of a calculation type, information such as the calculation model to be called and input parameters can be carried in the calculation request, so that the cloud server can run the corresponding calculation model for calculation based on these information.

Step S12, assigning target resource partitions to the target computing type, where the target resource partition represents a set of resources allowed to be used when performing the calculation of the target computing type, and different computing types have respective corresponding resource partitions.

In this embodiment, the target resource partition initially allocated for the target computing type may be empty. The resources in the target resource partition are allocated according to the resource allocation strategy. For details, please refer to the relevant descriptions of step S13 and step S14, which will not be repeated here.

Resource partitions corresponding to different computing types can be independent of each other. For any calculation type, only the resources in the resource partition corresponding to the calculation type are allowed to be used when performing the calculation of this calculation type. In this way, calculations of different calculation types will not affect each other. For example, suppose computing type A corresponds to resource partition A, and computing type B corresponds to resource partition B. In the case of a large increase in computing requests of computing type A, since computing of computing type A can only use resources in resource partition A and will not occupy resources in resource partition B, it can be guaranteed that computing of computing type B will not be affected , improve system stability.

Step S13 , acquiring the resource allocation policy of the target resource partition, wherein different resource partitions have respective corresponding resource allocation policies.

Specifically, the resource allocation strategy of the target resource partition can be specified according to the computing characteristics of the target computing type. Calculation features include but are not limited to calculation speed, computing power requirements during the calculation process, etc. For example, assuming that a faster calculation speed is required when performing the calculation of the target calculation type, more resources can be allocated to the target resource partition to ensure the calculation speed.

It can be understood that, based on computing characteristics of different computing types, different resource partitions may have corresponding resource allocation policies. For example, suppose computing type A corresponds to resource partition A, and computing type B corresponds to resource partition B. For the calculation of calculation type A, it is required to be able to quickly calculate the result, then more resources can be allocated to resource partition A to ensure that the calculation of calculation type A can have a faster calculation speed; and for the calculation type For the calculation of B, the requirement for calculation speed is not high, so the resource allocation of resource partition B can be appropriately reduced. For another example, assuming that when the calculation of calculation type A is performed, the amount of data processing is large and the requirements for computing power are high, then more resources can be allocated to resource partition A to ensure that the calculation of calculation type A can be executed smoothly; When calculating type A, the amount of data processing is small and the requirements for computing power are low, so the resource allocation of resource partition B can be reduced. For another example, when performing calculation of type A, if the resource usage rate in resource partition A exceeds the threshold, new resources can be allocated to resource partition A; when performing calculation of type B, resources in resource partition B After the usage rate exceeds the threshold for a certain period of time, new resources are allocated to resource partition B. For a detailed description of the resource allocation strategy, you can refer to the subsequent description of the parameters, which will not be repeated here.

Step S14, according to the resource allocation policy of the target resource partition, allocate resources to the target resource partition, so as to execute the calculation of the target computing type based on the resources in the target resource partition.

Specifically, based on the resource allocation strategy, resources can be allocated to the target resource partition in advance before the calculation of the target computing type is executed, or resources can be allocated to the target resource partition in real time during the execution of the calculation of the target computing type. For the specific resource allocation process, refer to subsequent related descriptions, and details are not repeated here.

In the technical solutions of some embodiments of the present application, resources are allocated to corresponding resource partitions by assigning corresponding resource partitions to different computing types, and based on resource allocation policies of each resource partition. In this way, when performing calculations of different calculation types, resource allocation can be performed according to their corresponding resource allocation strategies, which avoids resource waste when calculations of different calculation types share the same resource allocation strategy, thereby improving server performance.

The resource allocation strategy and resource allocation method of the present application will be further described in detail below in conjunction with some specific embodiments.

In some embodiments, the resource allocation strategy includes allocating resources valid within a preset time period to the target resource partition. The above allocation of resources to the target resource partition includes:

Allocate a preset number of target resources to the target resource partition, and set the target resource to be released after a preset period of time.

Specifically, the target resource may also be called a resident resource of the target resource partition. After the target resource is allocated to the target resource partition, the resource will not be released from the target resource partition within a preset period of time, no matter whether a computing request of the target computing type is received or not. In this way, within a preset period of time, based on the target resources, basic calculations of the target computing type can be performed to prevent calculation requests of other computing types from taking up all the resources and making it impossible to perform the calculation of the target computing type.

In some embodiments, the resource allocation strategy includes dynamically allocating the first dynamic resource to the target resource partition based on the time period of the current time. The above allocation of resources to the target resource partition includes:

If the current time is within the first time period, allocate the first dynamic resource to the target resource partition;

If the current time is within a second time period other than the first time period, release the first dynamic resource from the target resource partition.

Specifically, the first time period and the second time period may be configured in the resource allocation policy according to actual conditions. For example, the first time period may be from 0:00 to 12:00 in a day; the second time period may be from 12:00 to 24:00 in a day. For another example, the first time period may be Monday to Friday in a week, and the second time period may be Saturday and Sunday in a week. Based on the time period of the current time, the first dynamic resource is dynamically allocated to the target resource partition, which can reduce waste of resources while meeting business requirements. For example, based on business needs, there is a high requirement for computing speed in the first time period, so the first dynamic resource can be allocated to the target resource partition in the first time period to improve the computing speed; If the requirement for computing speed is low, then the first dynamic resource may be released from the target resource partition to avoid resource waste.

In this embodiment, the first time period and the second time period are determined based on the following method:

Based on the calculation of the target calculation type that has been executed in history, the historical time distribution corresponding to the calculation of the target calculation type is obtained;

Based on the historical time distribution, determine the peak time period and the valley time period when the calculation of the target calculation type is executed, and use the peak time period as the first time period and the valley time period as the second time period.

Wherein, the peak time period is a time period in which calculations of the target calculation type are frequently performed, and the valley time period is a time period in which calculations of the target calculation type are relatively seldom performed. Specifically, a times threshold may be set, and if within a time period, the calculation times of the target calculation type exceeds the times threshold, then this time period is a peak time period. Correspondingly, if within a time period, the number of calculations of the target calculation type does not exceed the number threshold, then this time period is a trough time period.

In some embodiments, the initial values of the first time period and the second time period may be artificially set according to experience. Based on the first time period and the second time period initially set, when the cloud server executes the calculation of the target calculation type, the actual time when each calculation is executed is collected to obtain the historical time distribution corresponding to the calculation of the target calculation type, based on the historical time distribution , you can adjust and modify the first time period and the second time period. For example, take one day as an example. Assume that the first time period is initially set from 0:00 to 10:00, and the second time period is initially set from 10:00 to 24:00. In the actual calculation process of the cloud server, it is found that the peak time period corresponding to the calculation of the target calculation type is usually from 0:00 to 8:00, then the first time period can be modified to 0:00 to 8:00, and the second time period Revised to 8:00 to 24:00.

Dynamically allocate the first dynamic resource to the target resource partition according to the peak time period and valley time period, which can guarantee the basic computing requirements when the target computing type has many computing requests; and when the target computing type has fewer computing requests , to release the first dynamic resource, so as to achieve the purpose of saving resources.

In addition, from the above related descriptions, it can be known that when performing some types of calculations (such as calculation-intensive calculations), the resources of the resource partition will often be fully occupied, but this is a normal situation. If we simply use the resource usage in the resource partition It is unreasonable to allocate resources. In view of this, in some embodiments, the resource allocation strategy of the present application includes a job queue based on the target resource partition, and dynamically allocates the second dynamic resource to the target resource partition, wherein the job queue includes job requests, and each job request represents a pending Computation of the target calculation type performed. The above allocation of resources to the target resource partition includes:

Predict the total time required for all job requests in the job queue to be executed;

When the total duration exceeds the duration threshold, allocate the second dynamic resource for the target resource partition.

Specifically, the duration threshold may be set according to actual conditions. For example, if the calculation speed of the target calculation type is high, the duration threshold can be set to a smaller value. In this way, when there are fewer job requests in the job queue of the target resource partition, the second dynamic resource will be assigned to the target resource partition in time, so as to ensure that the job requests in the job queue can be processed in time.

Based on the total time required to complete the execution of all job requests in the job queue, it is judged whether to assign the second dynamic resource to the target resource partition. This resource allocation method does not consider the resource usage in the target resource partition, which can prevent the target resource from When the resources in the partition are fully occupied, resources are wasted caused by unreasonable allocation of resources to the target resource partition.

Further, in some embodiments, the above-mentioned job request includes a parameter file required when performing the calculation of the target calculation type, and a model calling parameter required when calling the calculation model. The total time required for the execution of all job requests in the above forecast job queue to be completed, including:

Input the parameter file size corresponding to each job request in the job queue and the model calling parameters into the prediction model, and the prediction model outputs the total duration.

Specifically, the parameter file may include data required for performing calculations of the target calculation type. For example, when calculating the amount of steel bars in a building, it may be necessary to store the length, width, and height information of each wall of the building in a parameter file, and send the parameter file to the cloud server, so that the cloud server can perform reinforcement based on the data in the parameter file. Quantity calculation. It can be understood that, for different calculations of the same calculation type, the size of the parameter file may be different. For example, comparing the calculation of the reinforcement quantity of 10 buildings with the calculation of the reinforcement quantity of 1 building, the parameter file of 10 buildings will obviously be larger than the parameter file of 1 building. Therefore, the parameter file size can reflect the amount of data to be processed. The larger the parameter file, the larger the amount of data to be processed, and the correspondingly longer execution time of the calculation.

Similarly, model call parameters can represent the computing model to be called. It can be understood that since the calculation logic of different calculation models is different, the running time of each calculation model may also be different. That is, the calculation model is related to the execution time of the calculation.

To sum up, the parameter file size corresponding to each job request and the model call parameters are related to the total duration. Based on the parameter file size corresponding to each job request and the model call parameters to predict the total duration, the predicted total duration can be compared with the actual time. , which improves the prediction accuracy.

Further, in some embodiments, the prediction model can be obtained based on the following method:

Obtain the historical parameter files, historical model call parameters, and historical execution time corresponding to the executed historical job requests;

Use historical parameter files, historical model call parameters, and historical execution time as training data to train a predictive model.

Specifically, historical parameter files and historical model call parameters can be used as independent variables, and historical execution time as dependent variables, and the prediction model can be obtained by training based on the method of machine learning linear regression. Using historical data as training data, the way to obtain training data is relatively simple.

Among them, the historical execution time of the job request can be obtained based on the following method: record the submission time of the job request and the end time when the job request execution ends, and subtract the submission time from the end time to obtain the historical execution time of the job request.

Further, in some embodiments, among the above-mentioned historical data used for training the prediction model, some data may have a problem of dimensional differences. For example, if an 8-core processor and a 4-core processor are used to execute the same job request A, obviously, the execution time corresponding to the 8-core processor will be shorter than the execution time corresponding to the 4-core processor, that is, the job request A is executed in different Under resource configuration, there will be different execution times, and there are dimensional differences between these execution times. To avoid this dimensional difference, at least part of the training data can be normalized before training a predictive model on historical data. For example, the execution time corresponding to the 8-core processor and the execution time corresponding to the 4-core processor are normalized to the execution time corresponding to the 1-core processor. In this way, data with dimensional differences in the training data can be avoided and the training accuracy can be improved.

The resource allocation strategy and the resource allocation method of the present application are exemplarily described above. It can be understood that resource allocation strategies and resource allocation methods can be adjusted according to actual conditions, and any resource allocation scheme based on computing type should be within the scope of protection of this application.

Please refer to FIG. 2 , which is a schematic diagram of modules of a resource allocation device provided by an embodiment of the present application. Resource allocation devices include:

A receiving module, configured to receive target computing types to be allocated for resources, wherein different computing types are used to distinguish computing in different computing modes;

A partition allocation module, configured to allocate a target resource partition for the target computing type, wherein the target resource partition represents a set of resources allowed to be used when performing the calculation of the target computing type, and different computing types have respective corresponding resource partitions ;

a policy acquisition module, configured to acquire the resource allocation policy of the target resource partition, wherein different resource partitions have their own corresponding resource allocation policies; and

A resource allocation module, configured to allocate resources to the target resource partition according to the resource allocation policy of the target resource partition.

Please refer to FIG. 3 , which is a schematic diagram of an electronic device provided by an embodiment of the present application. The electronic device includes a processor and a memory, and the memory is used to store a computer program. When the computer program is executed by the processor, the above method is realized.

Wherein, the processor may be a central processing unit (Central Processing Unit, CPU). The processor can also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application-specific integrated circuits (Application Specific Integrated Circuit, ASIC), field-programmable gate array (Field-Programmable Gate Array, FPGA) or other Chips such as programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above-mentioned types of chips.

As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs, non-transitory computer-executable programs and modules, such as program instructions/modules corresponding to the methods in the embodiments of the present invention. The processor executes various functional applications and data processing of the processor by running non-transitory software programs, instructions, and modules stored in the memory, that is, implements the methods in the above method implementation manners.

The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application program required by at least one function; the data storage area may store data created by the processor, and the like. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory may optionally include memory located remotely from the processor, and such remote memory may be connected to the processor via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

An embodiment of the present application further provides a computer-readable storage medium, where the computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the above method is implemented.

Although the implementation method of the present invention has been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations all fall into the scope of the appended claims. within the limited range.

Claims (11)

1. A resource allocation method, characterized in that the method comprises: Receive the target computing type for resource allocation, where different computing types are used to distinguish the computing of different computing methods; Allocating a target resource partition for the target computing type, wherein the target resource partition represents a set of resources allowed to be used when performing the calculation of the target computing type, and different computing types have respective corresponding resource partitions; Obtaining resource allocation strategies of the target resource partition, wherein different resource partitions have respective corresponding resource allocation strategies; and Allocate resources to the target resource partition according to the resource allocation policy of the target resource partition. 2. The method according to claim 1, wherein the resource allocation strategy comprises dynamically allocating the first dynamic resource to the target resource partition based on the time period of the current time; The allocating resources to the target resource partition includes: If the current time is within the first time period, allocate the first dynamic resource to the target resource partition; If the current time is within a second time period other than the first time period, releasing the first dynamic resource from the target resource partition. 3. The method according to claim 2, wherein the first time period and the second time period are determined based on the following method: Obtaining a historical time distribution corresponding to the calculation of the target calculation type based on historically executed calculations of the target calculation type; Based on the historical time distribution, determine the peak time period and the valley time period when the calculation of the target calculation type is executed, and use the peak time period as the first time period, and use the valley time period as the first time period. the second time period. 4. The method according to claim 1, wherein the resource allocation strategy comprises dynamically assigning a second dynamic resource to the target resource partition based on a job queue of the target resource partition, wherein the job queue including job requests, each job request representing a computation to be performed of the target computation type; The allocating resources to the target resource partition includes: predicting the total time required when all the job requests in the job queue are executed; If the total duration exceeds a duration threshold, allocate the second dynamic resource for the target resource partition. 5. The method according to claim 4, wherein the calculation of the target calculation type is performed through a calculation model, and the job request includes a parameter file required for performing the calculation of the target calculation type, and calling The model calling parameters required when calculating the model; The prediction of the total time required for all job requests in the job queue to be executed includes: The size of the parameter file corresponding to each job request in the job queue and the model calling parameters are input into the predicted model that has been trained, and the predicted model outputs the total duration. 6. The method according to claim 5, wherein, before outputting the total duration by the predictive model, the predictive model is trained based on the following method: Obtain the historical parameter files, historical model call parameters, and historical execution time corresponding to the executed historical job requests; The prediction model is obtained through training using the historical parameter file, the calling parameters of the historical model, and the historical execution time as training data. 7. The method according to claim 6, wherein, before training the predictive model, the method further comprises: Perform normalization processing on at least part of the training data. 8. The method according to claim 1, wherein the resource allocation strategy comprises allocating resources that are valid within a preset duration to the target resource partition; The allocating resources to the target resource partition includes: Allocating a preset number of target resources to the target resource partition, and setting the target resources to be released after the preset duration. 9. A resource allocation device, characterized in that the device comprises: A receiving module, configured to receive target computing types to be allocated for resources, wherein different computing types are used to distinguish computing in different computing modes; A partition allocation module, configured to allocate a target resource partition for the target computing type, wherein the target resource partition represents a set of resources allowed to be used when performing the calculation of the target computing type, and different computing types have respective corresponding resource partitions ; a policy acquisition module, configured to acquire the resource allocation policy of the target resource partition, wherein different resource partitions have their own corresponding resource allocation policies; and A resource allocation module, configured to allocate resources to the target resource partition according to the resource allocation policy of the target resource partition. 10. A computer-readable storage medium, characterized in that the computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the computer program according to any one of claims 1 to 8 can be realized. method. 11. An electronic device, characterized in that the electronic device comprises a processor and a memory, the memory is used to store a computer program, and when the computer program is executed by the processor, the computer program according to claims 1 to 8 can be realized. any of the methods described.

Images