WO2021042373A1

WO2021042373A1 - Data processing and task scheduling method, device and system, and storage medium

Info

Publication number: WO2021042373A1
Application number: PCT/CN2019/104727
Authority: WO
Inventors: 陶原; 卢毅军; 李栈; 宋军; 奉有泉; 赵旭; 陈钢
Original assignee: 阿里巴巴集团控股有限公司
Priority date: 2019-09-06
Filing date: 2019-09-06
Publication date: 2021-03-11
Also published as: CN113748398B; CN113748398A

Abstract

A data processing method, device and system, and a storage medium. A setting operation for kernel parameters related to a power consumption management mechanism is combined with artificial intelligence. A performance-power consumption estimation model obtained through training on the basis of artificial intelligence estimates performance data and power consumption data of a physical device under a plurality of value combinations of kernel parameters; furthermore, on the basis of the obtained performance data and power consumption data, suitable values can be set for the kernel parameters, so that the power consumption and performance of the physical device can both be taken into account, and combining same with artificial intelligence can improve the efficiency of parameter setting and reduce the costs.

Description

Data processing and task scheduling method, equipment, system and storage medium

Technical field

This application relates to the field of data processing technology, and in particular to a data processing and task scheduling method, device, system and storage medium.

Background technique

With the development of electronic technology, power management of physical devices such as servers and computers has become more and more important. In order to reduce power consumption based on the normal operation of the device, a power management mechanism is added to the operating system. Taking the Linux operating system as an example, dynamic voltage frequency technology (Dynamic Voltage Frequency Scaling, DVFS) and C-mode (C-state) are the two main mechanisms used by the processor to reduce power consumption and ensure performance.

These power consumption management mechanisms are related to some device parameters, and some device parameters may have multiple values. The power consumption management mechanism can produce different energy-saving effects under different parameter values. How to reasonably set the parameter values of the power consumption management mechanism in an efficient and low-cost manner so that the power consumption management mechanism can produce a better or optimal energy-saving effect is a problem faced by the existing power consumption management mechanism.

Summary of the invention

Various aspects of the present application provide a data processing and task scheduling method, device, system, and storage medium, which are used to efficiently and reasonably set the parameter values of the power consumption management mechanism and reduce costs.

An embodiment of the application provides a data processing method, including: determining at least one kernel parameter related to the power management mechanism of the kernel mode supported by the physical device from the kernel parameters of the physical device; Describe the performance data and power consumption data under multiple value combinations corresponding to at least one core parameter, where at least some of the performance data and power consumption data under the value combination are estimated based on the performance-power consumption prediction model; Determine the target value combination corresponding to the at least one kernel parameter according to the performance data and power consumption data of the physical device under the multiple value combinations; determine the target value combination for the at least one kernel parameter according to the target value combination Setting is made so that the power consumption management mechanism operates according to the value in the target value combination.

An embodiment of the present application also provides a data processing method, including: determining at least one kernel parameter related to the power management mechanism of the kernel mode supported by the physical device from the kernel parameters of the physical device; Performance data and power consumption data of the physical device under the partial value combination corresponding to the at least one kernel parameter; perform model training according to the performance data and power consumption data of the physical device under the partial value combination to Obtain a performance-power consumption prediction model; use the performance-power consumption prediction model to predict the performance data and power consumption data of the physical device under other value combinations corresponding to the at least one core parameter.

An embodiment of the present application also provides a data processing method, including: determining at least one kernel parameter related to device performance from the kernel parameters of the physical device; and obtaining multiple selections of the physical device corresponding to the at least one kernel parameter. The performance data under the value combination, wherein at least part of the performance data under the value combination is estimated based on the performance prediction model; the performance data of the physical device under the multiple value combinations is determined A target value combination corresponding to at least one kernel parameter; the at least one kernel parameter is set according to the target value combination, so that the physical device operates according to the value in the target value combination.

An embodiment of the present application also provides a data processing method, including: determining at least one kernel parameter related to device performance from the kernel parameters of the physical device; using a load test tool to test that the physical device corresponds to the at least one kernel parameter The performance data under the partial value combination of the; perform model training according to the performance data of the physical device under the partial value combination to obtain the performance prediction model; use the performance prediction model to predict the physical device Performance data under other value combinations corresponding to the at least one kernel parameter.

An embodiment of the present application also provides a data processing method, including: determining at least one kernel parameter related to the power consumption of the device from the kernel parameters of the physical device; and obtaining multiple types of parameters corresponding to the at least one kernel parameter of the physical device. The power data under the value combination, wherein at least part of the power data under the value combination is estimated based on the power estimation model; and the power data of the physical device under the multiple value combinations is determined The target value combination corresponding to the at least one kernel parameter; and the at least one kernel parameter is set according to the target value combination, so that the physical device operates according to the value in the target value combination.

An embodiment of the present application also provides a data processing method, including: determining at least one kernel parameter related to the power consumption of the device from the kernel parameters of the physical device; using a load test tool to test the physical device in the at least one kernel parameter Corresponding power data under the partial value combination; perform model training according to the power data of the physical device under the partial value combination to obtain a power estimation model; use the power estimation model to estimate the physical Power data of the device under other value combinations corresponding to the at least one kernel parameter.

An embodiment of the present application also provides a data processing method, including: determining at least one kernel parameter related to the power management mechanism of the kernel mode supported by the physical device from the kernel parameters of the physical device; Performance data and power consumption data of the physical device under the partial value combination corresponding to the at least one kernel parameter; perform model training according to the performance data and power consumption data of the physical device under the partial value combination to Obtain a performance-power consumption prediction model; use the performance-power consumption prediction model to predict the performance data and power consumption data of the physical device under multiple value combinations corresponding to the at least one core parameter; wherein, The multiple value combinations include the partial value combinations.

An embodiment of the present application also provides a device management system, including: at least one physical device and at least one model computing device; wherein the at least one physical device supports a kernel-mode power management mechanism; the at least one A model computing device for determining at least one kernel parameter related to the power management mechanism of the kernel state supported by the target device from the kernel parameters of the target device, and obtaining the target device in the target device based on the artificial intelligence model Performance data and power consumption data under multiple value combinations corresponding to at least one kernel parameter; the target device is any one of the at least one physical device; the target device is configured to The performance data and power consumption data of the target device under multiple value combinations corresponding to the at least one kernel parameter obtained by the model computing device determine the target value combination corresponding to the at least one kernel parameter, and according to the The target value combination sets the at least one kernel parameter.

An embodiment of the present application further provides a data center system, including: a model computing device and at least one computer room, the at least one computer room includes at least one physical device, and the at least one physical device respectively supports a kernel-mode power management mechanism The model computing device is used to determine from the kernel parameters of the target device at least one kernel parameter related to the power management mechanism of the kernel mode supported by the target device, and obtain the target device in the target device based on the artificial intelligence model Performance data and power consumption data under multiple value combinations corresponding to the at least one kernel parameter; the target device is any one of the at least one physical device; the target device is configured to The performance data and power consumption data of the target device under multiple value combinations corresponding to the at least one kernel parameter obtained by the model computing device are determined, and the target value combination corresponding to the at least one kernel parameter is determined, and the target value combination is determined according to the The target value combination sets the at least one kernel parameter.

An embodiment of the present application also provides a physical device, including: a memory and a processor; the memory is used for storing a computer program; the processor is coupled with the memory and is used for executing the computer program for : From the kernel parameters of the physical device, determine at least one kernel parameter related to the power management mechanism of the kernel mode supported by the physical device; obtain multiple selections of the physical device corresponding to the at least one kernel parameter The performance data and power consumption data under the value combination, wherein at least part of the performance data and power consumption data under the value combination are estimated based on the performance-power consumption prediction model; according to the physical device in the multiple Determine the target value combination corresponding to the at least one core parameter based on the performance data and power consumption data under the value combination; set the at least one core parameter according to the target value combination to enable the power consumption management The mechanism operates according to the value in the target value combination.

An embodiment of the present application also provides a model computing device, including: a memory and a processor; the memory is used to store a computer program; the processor is coupled with the memory and is used to execute the computer program for use In: From the kernel parameters of the physical device, determine at least one kernel parameter related to the power management mechanism of the kernel state supported by the physical device; use a load test tool to test the physical device in the at least one kernel parameter corresponding to the at least one kernel parameter. Perform model training based on the performance data and power consumption data of the physical device under the partial value combination to obtain a performance-power consumption prediction model; use the performance data and power consumption data of the physical device under the partial value combination; The performance-power consumption prediction model estimates the performance data and power consumption data of the physical device under other value combinations corresponding to the at least one core parameter.

The embodiments of the present application also provide a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor is caused to implement the steps in the various data processing methods provided in the embodiments of the present application. .

An embodiment of the present application also provides a task scheduling method, including: obtaining the task to be scheduled and the performance requirements of the task to be scheduled; selecting from at least one resource device that meets the performance requirements and the value of the power consumption parameter in the kernel mode meets the design requirements. A resource device with a fixed power consumption requirement; schedule the task to be scheduled to a resource device that meets the performance requirements and the value of the power consumption parameter in the kernel mode meets the set power consumption requirement; wherein the value of the kernel power consumption parameter refers to A value combination of at least one kernel parameter related to the power management mechanism of the kernel mode supported by the resource device.

An embodiment of the present application also provides a task scheduling device, including: a memory and a processor; the memory is used to store a computer program; the processor is coupled with the memory and is used to execute the computer program for use In: Obtain the task to be scheduled and the performance requirements of the task to be scheduled; from at least one resource device, select a resource device that meets the performance requirements and the core state power consumption parameter value meets the set power consumption requirements; The scheduling task is scheduled to the resource device that meets the performance requirements and the core-mode power consumption parameter value meets the set power consumption requirement; wherein the core-mode power consumption parameter value refers to the core-mode power management mechanism supported by the resource device The value combination of at least one core parameter that is related.

The embodiment of the present application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor is caused to implement the steps in the task scheduling method provided by the embodiment of the present application.

In the embodiment of the present application, the setting operation of the kernel parameters related to the power management mechanism is combined with artificial intelligence, and the performance-power consumption estimation model trained based on the artificial intelligence obtains a variety of choices corresponding to the kernel parameters of the physical device. The performance data and power consumption data under the value combination, and then, based on the obtained performance data and power consumption data, appropriate values can be set for the kernel parameters, taking into account the power consumption and performance of the physical device, and combining with artificial intelligence , Which can improve the efficiency of parameter setting and reduce costs.

Description of the drawings

The drawings described here are used to provide a further understanding of the application and constitute a part of the application. The exemplary embodiments and descriptions of the application are used to explain the application, and do not constitute an improper limitation of the application. In the attached picture:

FIG. 1 is a schematic structural diagram of a data center system provided by an exemplary embodiment of this application;

FIG. 2 is a schematic diagram of the principle of kernel parameter setting of a data center system provided by an exemplary embodiment of this application;

FIG. 3 is a schematic diagram of the working principle of a performance-power consumption prediction model provided by an exemplary embodiment of this application;

FIG. 4a is a schematic structural diagram of a device management system provided by an exemplary embodiment of this application;

FIG. 4b is a schematic structural diagram of an edge cloud network system provided by an exemplary embodiment of this application;

FIG. 5a is a schematic flowchart of a data processing method provided by an exemplary embodiment of this application;

FIG. 5b is a schematic flowchart of another data processing method provided by an exemplary embodiment of this application;

FIG. 6a is a schematic flowchart of yet another data processing method provided by an exemplary embodiment of this application;

FIG. 6b is a schematic flowchart of yet another data processing method provided by an exemplary embodiment of this application;

FIG. 7a is a schematic flowchart of yet another data processing method provided by an exemplary embodiment of this application;

FIG. 7b is a schematic flowchart of yet another data processing method provided by an exemplary embodiment of this application;

FIG. 7c is a schematic flowchart of yet another data processing method provided by an exemplary embodiment of this application;

FIG. 7d is a schematic flowchart of a task scheduling method provided by an exemplary embodiment of this application;

FIG. 8a is a schematic structural diagram of a physical device provided by an exemplary embodiment of this application;

FIG. 8b is a schematic structural diagram of a model computing device provided by an exemplary embodiment of this application;

Fig. 8c is a schematic structural diagram of a task scheduling device provided by an exemplary embodiment of this application.

detailed description

In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described clearly and completely in conjunction with specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

In view of the problems existing in the prior art, in the embodiments of the present application, the setting operation of the kernel parameters related to the power management mechanism is combined with artificial intelligence, and the performance-power consumption prediction model trained based on artificial intelligence can be obtained The performance data and power consumption data of the physical device under multiple combinations of values corresponding to the kernel parameters. Furthermore, based on the obtained performance data and power consumption data, appropriate values can be set for the kernel parameters, taking into account the functions of the physical device. Consumption and performance, combined with artificial intelligence, can improve the efficiency of parameter setting and reduce costs.

The technical solutions provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

Fig. 1 is a schematic structural diagram of a data center system provided by an exemplary embodiment of this application. As shown in FIG. 1, the data center system 100 includes: a model computing device 101 and at least one computer room 102; the at least one computer room 102 includes at least one physical device 103.

Among them, the machine room 102 refers to a physical place where machinery and equipment are stored, for example, it may be a room or a factory building. This embodiment does not limit the number of physical devices 103 in each computer room 102. Each computer room 102 may include one physical device 103 or multiple physical devices 103. Generally speaking, a computer room 102 will include multiple physical devices 103.

In this embodiment, the physical device 103 refers to a physical device that has an operating system installed and supports a kernel-mode power management mechanism. Of course, in addition to the physical devices 103 that are installed with the operating system and support the power management mechanism in the kernel mode, the computer room 102 may also contain some devices that do not require an operating system, and some devices that do not support the kernel mode even though the operating system is installed. There are no restrictions on devices with power management mechanisms. In this embodiment and other embodiments, the focus is on the physical device 103 that is installed with an operating system and supports a kernel-mode power management mechanism.

In this embodiment, the device form of the physical device 103 is not limited, and any device form that has an operating system installed and supports a kernel-mode power management mechanism is applicable to the embodiment of the present application. Optionally, the physical device 103 may be some IT devices installed with an operating system and supporting a kernel-mode power management mechanism, but it is not limited to this. For example, the physical device 103 may include but is not limited to at least one of the following: various server devices, computer devices, and various network switching devices. The server device may include, but is not limited to, a conventional server, a server array, or a cloud server. Optionally, various applications or services can be run on these physical devices 103, such as cloud computing services, game services, instant messaging services, mail services, or online transaction services. Of course, these physical devices 103 may not run any applications or services. Among them, according to whether the physical device 103 runs applications or services, and the type and quantity of running applications or services, the power consumption management mechanism on the physical device 103 will be triggered and play a corresponding role.

In essence, the operating system is a kind of software that is responsible for controlling the hardware resources of physical devices and providing an environment for upper-level applications to run. The operating system provides two CPU operating states, kernel mode and user mode. User mode is the activity space of upper-level applications. The execution of applications must rely on the resources provided by the operating system, including CPU resources, storage resources, and I/O resources. The power consumption management mechanism of this embodiment is a mechanism provided by the operating system to manage the power consumption of physical devices. It is a power consumption management mechanism at the operating system level and needs to run in the kernel mode, referred to as power consumption in the kernel mode for short. Management mechanism. It is worth noting that the power management mechanisms in the kernel mode supported by different physical devices 103 may be the same or different. The power management mechanism of the kernel mode is related to the operating system. If the operating system of the physical device 103 is different, the power management mechanism of the kernel mode supported by the physical device 103 will also be different. For example, if the physical device 103 adopts the Linux operating system as an example, the power management mechanisms in the kernel state supported by it include, but are not limited to: DVFS and C-state.

Among them, DVFS is a dynamic technology that dynamically adjusts the operating frequency and voltage of the chip according to the different needs of the application program running on the chip (such as the CPU) for computing power, so as to achieve the purpose of energy saving. For the same chip, the higher the operating frequency, the higher the voltage required and the greater the energy consumption. C-state is a low-power mechanism that allows the CPU to enter a low-power state when it is idle. The C-states included in C-states start from C0 to Cn. C0 is the normal operating mode of the CPU, and the CPU is at 100% Running state; the higher the value of n after C, the deeper the CPU sleeps, the lower the power consumption of the CPU, and of course it takes more time to return to the C0 mode; where n is a positive integer.

In this embodiment, the power management mechanism in the kernel mode is related to some kernel parameters. These kernel parameters are also the parameters of the power management mechanism in the kernel mode. The values of these kernel parameters can affect the energy saving of the power management mechanism in the kernel mode. effect. In this embodiment, the kernel parameters generally refer to the parameters in the source code of various operating systems, such as the kernel parameters of the Linux operating system, the kernel parameters of the Windows operating system, the kernel parameters of the UNIX operating system, or the kernel parameters of the MAC operating system.

Taking the Linux operating system as an example, the kernel parameters related to DVFS include but are not limited to: the minimum operating frequency that the CPU can run (denoted as scaling_min_freq), the highest operating frequency that the CPU can run (denoted as scaling_max_freq), and the adjustment mode of the CPU operating frequency (Denoted as scaling_governor); for the physical device 103, the energy saving effect of the DVFS can be changed by adjusting the value of at least one of the three parameters. Taking the Linux operating system as an example, the kernel parameters related to C-states include, but are not limited to: the entry time threshold corresponding to each level of C mode (denoted as target_residency); where the entry time threshold indicates that the physical device 103 has entered the corresponding C mode at least The time required to stay in this mode is the time condition that the physical device 103 needs to meet to enter the corresponding C mode; for the physical device 103, the difficulty of the CPU entering the corresponding C mode can be changed by adjusting the entry time threshold corresponding to the corresponding C mode. Easy to change the energy-saving effect of the C-states mechanism.

Regardless of the power management mechanism in the kernel mode, the kernel parameters related to it may have multiple values. For example, the maximum operating frequency scaling_max_freq at which the processor can run can be set to 2.4GHZ, 3.6GHZ, and so on. The value of the kernel parameter related to the power management mechanism is different, and the energy saving effect that the power management mechanism can produce will be different. In practical applications, the power consumption of the physical device 103 has a certain relationship with performance. Generally speaking, the lower the power consumption, the worse the performance will be. Therefore, we cannot blindly pursue low power consumption. Ideally, it should be based on application requirements. Balance between power consumption and performance. In different application scenarios, the physical device 103 has different requirements for power consumption and performance. How to set the value of the kernel parameter related to the kernel mode power management mechanism supported by the physical device 103 to enable the power management mechanism It is a problem that needs to be solved to produce a better or optimal energy-saving effect and at the same time meet the performance requirements of the physical device 103.

Among them, the kernel parameters related to the power management mechanism are system-level parameters and need to be set in the source code of the operating system. Each time the values of these kernel parameters are reset, the operating system needs to be reinstalled, and only when the operating system is running After a certain period of time, it can be judged whether the current value can meet the requirements of the device for power consumption and performance, and whether it needs to be adjusted again. If there are many kernel parameters related to the power management mechanism, the number of value combinations of these kernel parameters will also be larger. If a trial method is adopted to select the best value combination, it will take a long time.

Based on the above considerations, in this embodiment, the kernel parameter setting operation is combined with artificial intelligence, and based on the artificial intelligence model, the performance data and functions of the physical device 103 under various value combinations corresponding to the relevant kernel parameters are obtained. Based on the performance data and power consumption data of the physical device 103 under various value combinations corresponding to the relevant kernel parameters, the kernel parameter settings can not only set appropriate values for the kernel parameters, but also At the same time, it takes into account the requirements of the physical device 103 for power consumption and performance, and the combination with artificial intelligence can improve the efficiency of parameter setting and reduce the cost.

In order to achieve the above purpose, in the data center system 100 of this embodiment, a model computing device 101 is added, which is mainly responsible for obtaining the performance data and power consumption data of the physical device 103 under various value combinations corresponding to related kernel parameters. For the model computing device 101, for each physical device 103, at least one kernel parameter related to the power management mechanism of the kernel state supported by the physical device 103 can be determined from the kernel parameters of the physical device 103, based on artificial intelligence The model provides the physical device 103 with its performance data and power consumption data under multiple value combinations corresponding to at least one kernel parameter, so that the physical device 103 can complete the setting of related kernel parameters accordingly.

Wherein, the model computing device 101 may be set in a certain computer room 102, or be independent of each computer room 102, and be set separately in a certain place, for example, it may be set in the cloud. In addition, this embodiment does not limit the device form of the model computing device 101, and may be any computing device with certain computing capabilities and communication capabilities. As shown in FIG. 1, the model computing device 101 may be a conventional server, a cloud server, a server with a GPU, a server with a specific Ai chip, or a server array, etc.

Optionally, the model computing device 101 may communicate with each physical device 103. After obtaining the performance data and power consumption data of each physical device 103 under multiple value combinations corresponding to the relevant kernel parameters, it may be based on The communication connection between it and each physical device 103 provides each physical device 103 with its performance data and power consumption data under a variety of value combinations corresponding to related kernel parameters. Wherein, the model computing device 101 and each physical device 103 may be wirelessly or wiredly connected. Optionally, the physical device 103 can communicate with the model computing device 101 via a mobile network. Among them, the network standard of the mobile network can be 2G (GSM), 2.5G (GPRS), 3G (WCDMA, TD-SCDMA, CDMA2000, UTMS), 4G (LTE), 4G+ (LTE+), 5G, WiMax or coming soon in the future Any of the new network standards, etc. Optionally, the physical device 103 can also communicate with the model computing device 101 through Bluetooth, WiFi, infrared, zigbee, or NFC.

It should be noted that the model computing device 101 does not have to establish a communication connection with the physical device 103. In the scenario where a communication connection is not established between the model computing device 101 and the physical device 103, some mobile storage devices such as mobile hard disks and U disks can also be used to combine the physical device 103 under various combinations of values corresponding to the relevant kernel parameters. The performance data and power consumption data are copied from the model computing device 101 to the physical device 103.

For ease of description, this embodiment takes any physical device 103 as an example, and uses the flowchart shown in FIG. 2 to describe the working principle of the system. For ease of description and distinction, the target device is used as an example in the following content to expand the description, and the target device represents any physical device 103.

For example, in the case of a new physical device in a certain computer room 102, the newly added physical device can be used as the target device. Before installing the operating system for the target device, the source code of the operating system can be modified according to the application requirements of the target device. Kernel parameters related to the power management mechanism are set, and the operating system is installed on the target device according to the operating system source code after the parameter setting. For another example, when the load information (such as an application or service) on the target device changes, in order to allow the target device to meet the power consumption and performance requirements of the new load information, the operating system source code and power consumption can be re-modified. The value of the kernel parameter related to the management mechanism, and the operating system is reinstalled for the target device according to the modified operating system source code. For another example, in the case of renting and selling a piece of equipment to a certain customer, the equipment rented and sold to the customer can be used as the target equipment. In order to ensure that the target equipment can successfully provide corresponding services to the customer, the target can be set according to the customer’s requirements. The device installs the operating system and sets the kernel parameters related to the power management mechanism in the operating system source code before installing the operating system.

Regardless of the foregoing requirements, the model computing device 101 can determine at least one kernel parameter related to the power management mechanism of the kernel mode supported by the target device from the kernel parameters of the target device. The number of core parameters can be one or more, depending on the power management mechanism.

Optionally, the model computing device 101 may have human-computer interaction capabilities, and may perform human-computer interaction with the administrator of the data center system 100 to determine at least one kernel parameter related to the power management mechanism of the kernel mode supported by the target device . For example, the model computing device 101 can provide a human-computer interaction interface or a command window, and the administrator can input the target device identification and the kernel mode power management mechanism supported by the target device to the model computing device 101 through the human-computer interaction interface or command window. According to the information; the model computing device 101 determines from the kernel parameters of the target device, at least one kernel parameter related to the power management mechanism of the kernel mode supported by the target device. For another example, if the model computing device 101 is capable of speech recognition, the administrator can input the identification of the target device and the information of the power management mechanism in the kernel mode supported by the target device into the model computing device 101 by voice; the model computing device 101 according to With this information, from the kernel parameters of the target device, at least one kernel parameter related to the power management mechanism of the kernel mode supported by the target device is determined. Of course, in addition to the above information, the administrator can also directly input to the model computing device 101 the identification of at least one kernel parameter related to the power management mechanism of the kernel mode supported by the target device, such as the parameter name.

After determining at least one kernel parameter related to the power management mechanism of the kernel mode supported by the target device, the model computing device 101 may collect the performance data and power consumption data of the target device under the partial value combination corresponding to the at least one kernel parameter, As sample data, and use these sample data for model training to obtain a performance-power consumption prediction model, as shown in Figure 2. Among them, the performance data of the target device will be different according to different services or applications running on the target device. Performance data mainly refers to some data that can reflect the capacity of the target device to undertake the service and guarantee the service. For example, it can be the actually achieved QoS, QPS, TPS, or response time to the request of the service on the target device, and so on.

It is explained here that if the number of at least one kernel parameter is one, then one value combination includes one value; if the number of at least one kernel parameter is more than one, then one value combination includes multiple kernel parameters one by one. Corresponding multiple values, and the multiple values included in different value combinations are not completely the same.

In this embodiment, as shown in FIG. 2, the model computing device 101 can use a load test tool to collect performance data and power consumption data of the target device under a partial value combination corresponding to at least one kernel parameter. Install and run the load test tool on the target device. The load test tool can simulate the performance data and power consumption data of the target device under different load conditions. In detail, from multiple value combinations corresponding to at least one kernel parameter, select some value combinations, and the number of some value combinations only needs to meet the number required for model training; for each of the partial value combinations According to the load, power consumption and/or performance requirements that the target device needs to meet, the load test tool is used to obtain the performance data and power consumption data of the target device when the value combination meets the corresponding requirements. The performance data and power consumption data when the corresponding requirements are met under the partial value combination are provided to the model computing device 101. In order to simplify the description, the performance data and power consumption data of the target device when it meets the corresponding requirements under each value combination are referred to as the performance data and power consumption data of the target device under each value combination for short.

Among them, for each value combination in the partial value combination, the process of using the load test tool to obtain the performance data and power consumption data of the target device under the value combination includes: for the value combination, first operate The value of at least one kernel parameter in the system source code is modified to the value in the value combination, and the operating system is installed on the target device according to the modified operating system source code; after the operating system is successfully installed, on the target device Install the load test tool, and use the load test tool to simulate the corresponding load condition and obtain the performance data and power consumption data of the target device under the corresponding load according to the load, power consumption and/or performance requirements that the target device should meet.

In this embodiment, the number and types of load testing tools are not limited, and can be flexibly selected according to application requirements, operating system types, and the like. For example, this embodiment lists several load testing tools: Stream testing tool for memory performance testing; Specjbb testing tool for testing CPU performance; Speccpu testing tool for testing CPU performance; Fio testing tool for testing disks IO performance; Sysbench testing tool for testing mysql database performance, etc.

Optionally, during the testing process of the above-mentioned several load testing tools, a power consumption collection tool such as a fuel gauge can be used to collect power consumption data of the physical device during the testing process. Among them, the relevant performance parameters tested by the load test tool and the power consumption data collected by the power consumption collection tool during the test are the performance data and power consumption data of the physical device 103 when it meets the corresponding requirements under a certain combination of values. .

After obtaining the performance data and power consumption data of the target device under the partial value combination corresponding to at least one kernel parameter, the model computing device 101 uses these data as sample data for model training. In this embodiment, the process of model training performed by the model computing device 101 is not limited. For example, it can be a model training process based on a deep neural network, or a model training process based on regression analysis, where at least one kernel parameter can be analyzed. The model training method of the association relationship between the value combination of and the performance data and power consumption data of the target device is applicable to the embodiments of the present application.

Among them, regression analysis is a statistical analysis method to determine the quantitative relationship between two or more variables, and it is a predictive modeling technique. In an optional embodiment, the model computing device 101 may adopt a modeling method based on regression analysis. Based on this, the process of model training performed by the model computing device 101 is actually a process of performing regression analysis on the performance data and power consumption data of the target device under the partial value combination corresponding to at least one kernel parameter. After regression analysis, at least one The correlation between the value combination corresponding to the kernel parameter and the performance data and power consumption data of the target device is the performance-power consumption prediction model.

Among them, regression analysis includes linear regression analysis and logistic regression analysis. In an optional embodiment of the present application, linear regression analysis is preferred for modeling. Based on this, the process of model training performed by the model computing device 101 includes: taking partial value combinations corresponding to at least one kernel parameter as an independent variable, and taking the performance data and power consumption data of the target device under the partial value combination as the dependent variable for linearization. Regression analysis, the performance-power consumption prediction model is obtained. The performance-power consumption prediction model obtained by this optional embodiment is a linear regression model.

In an optional embodiment, after the performance-power consumption prediction model is obtained, the model computing device 101 may use the performance-power consumption prediction model to predict the performance of the target device under other value combinations corresponding to at least one core parameter. Data and power consumption data are shown in Figure 2. Compared with the method of using the load test tool to test, using the performance-power estimation model to estimate the performance data and power consumption data of the target device under other value combinations is much faster and can save a lot of time and cost. Among them, the performance data and power consumption data of the target device under other value combinations corresponding to at least one core parameter estimated by the performance-power consumption prediction model are combined with the target device tested by the load test tool in at least one core The performance data and power consumption data under the partial value combinations corresponding to the parameters can obtain the performance data and power consumption data of the target device under multiple value combinations corresponding to at least one core parameter. The "multiple value combinations" here can be all value combinations of at least one kernel parameter, or part of all value combinations of at least one kernel parameter. Regardless of whether "multiple value combinations" are all value combinations of at least one kernel parameter or partial value combinations, "multiple value combinations" include the above "partial value combinations" and "other value combinations". or,

In another optional embodiment, after the performance-power consumption prediction model is obtained, the model computing device 101 may use the performance-power consumption prediction model to estimate that the target device is under multiple value combinations corresponding to at least one kernel parameter Performance data and power consumption data. In the same way, the "multiple value combinations" here can be all value combinations of at least one kernel parameter, or part of all value combinations of at least one kernel parameter. Regardless of whether "multiple value combinations" are all value combinations of at least one kernel parameter or partial value combinations, "multiple value combinations" include the above "partial value combinations" and "other value combinations".

After obtaining the performance data and power consumption data of the target device under multiple value combinations corresponding to at least one core parameter, the model computing device 101 can actively place the target device in at least one core based on the communication connection between it and the target device. The performance data and power consumption data under multiple value combinations corresponding to the parameters are sent to the target device, or, according to the request of the target device, the performance data of the target device under multiple value combinations corresponding to at least one kernel parameter And power consumption data are sent to the target device, or the target device can also actively download its performance data and power consumption data under multiple value combinations corresponding to at least one kernel parameter to the model computing device 101; or it can also be related The personnel sends the performance data and power consumption data of the target device under multiple value combinations corresponding to at least one kernel parameter to the target device through a mobile storage device and copies them from the model computing device 101 to the target device. The embodiment of the present application does not limit the specific manner in which the target device obtains its performance data and power consumption data under multiple value combinations corresponding to at least one kernel parameter.

For the target device, the performance data and power consumption data under multiple value combinations corresponding to at least one core parameter can be obtained, and during the parameter setting process, the multiple values corresponding to at least one core parameter can be used Based on the combined performance data and power consumption data, a target value combination corresponding to at least one core parameter is determined, and at least one core parameter is set according to the target value combination. The process of setting at least one kernel parameter according to the target value combination includes: modifying the value of at least one kernel parameter related to the power consumption management mechanism in the operating system source code to the value in the target value combination, and then, The operating system can also be installed on the physical device according to the modified operating system source code. Wherein, installing the operating system on the physical device according to the modified operating system source code includes: compiling the modified operating system source code to obtain an installation file of the operating system, and running the installation file to complete the installation of the operating system.

In the embodiments of the present application, the specific implementation of determining the target value combination corresponding to at least one core parameter based on the performance data and power consumption data of the target device under multiple value combinations corresponding to at least one core parameter is not limited. the way. The following exemplarily lists several optional implementation manners:

In optional implementation A, the target value combination can be directly determined from the multiple value combinations according to the performance data and power consumption data of the target device under multiple value combinations.

In optional implementation B, the power consumption data and/or performance data that the target device actually needs to meet can be obtained; the power consumption data and/or performance data that the target device actually needs to meet can be combined with the target device in multiple values Match the performance data and power consumption data under the following; according to the actual power consumption data and/or performance data that the target device needs to meet with the performance data and power consumption data of the target device under multiple value combinations, from Among the multiple value combinations, a value combination that meets the matching degree requirement is selected as the target value combination.

In optional implementation C, at least one candidate value combination can be selected from multiple value combinations according to the performance data and power consumption data of the target device under multiple value combinations; the load test tool is used to test the target device Performance data and power consumption data under at least one candidate value combination; further, according to the performance data and power consumption data of the target device under at least one candidate value combination tested by the load test tool, from at least one candidate value combination Determine the target value combination in the value combination. In Embodiment C, in addition to the performance data and power consumption data of the target device under multiple value combinations, the load test tool is also combined to improve the accuracy of the final selected target value combination, and is beneficial to Improve the accuracy of parameter settings.

Further, in Embodiment C, in the process of selecting at least one candidate value combination from multiple value combinations according to the performance data and power consumption data of the target device under multiple value combinations, the target device can be directly selected according to the target value combination. For the performance data and power consumption data of the device under multiple value combinations, at least one candidate value combination is selected from the multiple value combinations. Or, you can also combine the power consumption data that the target device actually needs to meet, or combine the performance data that the target device actually needs to meet, or combine the power consumption data and performance data that the target device actually needs to meet, and choose from a variety of value combinations At least one combination of candidate values.

Among them, the power consumption data and/or performance data that the target device actually needs to meet can be analyzed according to the QoS of the service that the target device actually needs to carry. Generally speaking, each service has its own set of standards to measure its own service performance, such as how long it takes for a request to read the database to get a response is acceptable. Alternatively, the power consumption data and/or performance data required by the service deployment user may also be obtained as the power consumption data and/or performance data that the target device actually needs to meet. Service deployment users refer to users who need to deploy services on target devices. Service deployment users have certain requirements for the services they deploy, such as requiring performance to meet a certain standard, or requiring power consumption not to exceed a certain power threshold, etc. Obtain the power consumption data that the target device actually needs to meet from the power consumption requirements, or obtain the performance data that the target device actually needs to meet from the performance requirements.

For example, the target device can obtain the power consumption data that it actually needs to meet, and compare the power consumption data that the target device actually needs to meet with the performance data and power consumption data of the target device under a combination of multiple values corresponding to at least one core parameter. Perform matching; from a variety of value combinations, obtain a value combination that has a matching degree with the power consumption data that the target device actually needs to meet and meets the matching degree requirement as a candidate value combination. In this embodiment, the power consumption requirements of the target device can be given priority according to application requirements, and the kernel parameters related to the power consumption management mechanism can be set to make the target device meet the power consumption requirements.

For another example, the target device can obtain the performance data that it actually needs to meet, and compare the performance data that the target device actually needs to meet with the performance data and power consumption data of the target device under multiple combinations of values corresponding to at least one kernel parameter. Matching: From a variety of value combinations, obtain a value combination that has a matching degree with the performance data that the target device actually needs to meet and meets the matching degree requirement as a candidate value combination. In this embodiment, the performance requirements of the target device can be prioritized according to application requirements. By setting kernel parameters related to the power consumption management mechanism, the target device can meet the power consumption requirements by converting energy consumption to performance.

For another example, the target device can obtain the power consumption data and performance data that it actually needs to meet, and combine the performance data and power consumption data that the target device actually needs to meet with the target device under a combination of multiple values corresponding to at least one kernel parameter The performance data and power consumption data are matched; from a variety of value combinations, the value combination that meets the matching degree requirements of the power consumption data and the performance data that the target device actually needs to meet is obtained as the candidate value combination. In this embodiment, according to application requirements, the power consumption and performance requirements of the target device can be considered at the same time, and the power consumption and performance can be taken into consideration by setting the kernel parameters related to the power consumption management mechanism.

Further, in Embodiment C, using a load test tool to test the performance data and power consumption data of the target device under at least one candidate value combination is to use the load test tool to test the performance data of the target device under each candidate value combination. And the process of power consumption data. Taking the first candidate value combination as an example, the process of using a load test tool to test the performance data and power consumption data of the target device under the first candidate value combination includes: modifying the value of at least one kernel parameter in the source code of the operating system For the value in the first candidate value combination, install the operating system on the target device according to the modified operating system source code; run the load test tool on the target device to test the target device under the first candidate value combination Performance data and power consumption data. Wherein, the first candidate value combination is any one of at least one candidate value combination.

Further, as shown in FIG. 2, in the implementation C, after the target value combination is selected, the performance data and power consumption data of the target device under the target value combination tested by the load test tool can be used to determine the performance -The power consumption estimation model is revised. Among them, the performance data and power consumption data of the target device tested by the load test tool under the target value combination are more in line with the actual requirements, which are used as sample data to modify the performance-power consumption prediction model, which is conducive to improving the performance-power The accuracy of the consumption estimation model makes the estimated performance data and power consumption data more in line with actual requirements.

In an exemplary embodiment, taking the target device as a server or computer device using the Linux operating system as an example, the kernel mode power management mechanism supported by the target device may include: DVFS, then at least one kernel related to DVFS The parameters include: at least one of the lowest operating frequency at which the CPU can run, the highest operating frequency at which the CPU can run, and an adjustment mode of the CPU operating frequency. Taking these three parameters as an example, the model computing device 101 can use the load test tool to test the performance data and power consumption data of the target device under the partial value combination corresponding to these three parameters, and use the target device to test the performance data and power consumption data of the three parameters. Perform model training on the performance data and power consumption data under the corresponding partial value combination to obtain the performance-power consumption estimation model, and then estimate the target device's other values corresponding to these 3 parameters based on the performance-power consumption estimation model Combine the performance data and power consumption data to obtain the performance data and power consumption data of the target device under various value combinations corresponding to these 3 parameters; the target device uses its various value combinations corresponding to these 3 parameters Set these 3 parameters based on the performance data and power consumption data below.

Further, taking the target device as a server or computer device using the Linux operating system as an example, the kernel mode power management mechanism supported by the target device includes in addition to DVFS, it can also include: C-state, C-state contains 3 -11 levels of C mode, each level of C mode has its own corresponding entry time threshold. Based on this, at least one kernel parameter not only includes the above three parameters related to DVFS, but also includes the entry time threshold corresponding to the C mode of each level under the C-state. If the C-state contains 6 levels of C mode, denoted as C1-C6, at least one kernel parameter includes the entry time threshold corresponding to the 6 levels of C mode, plus the above 3 parameters required by DVFS, for a total of 9 parameters , Specifically: scaling_min_freq, scaling_max_freq, scaling_governor, Target_residency: C1, Target_residency: C2, Target_residency: C3, Target_residency: C4, Target_residency: C5, and Target_residency: C6. Taking these 9 parameters as an example, the model computing device 101 can use the load test tool to test the performance data and power consumption data of the target device under the partial value combination corresponding to these 9 parameters, and use the target device to test the performance data and power consumption data of the 9 parameters. Perform model training on the performance data and power consumption data under the corresponding partial value combination to obtain the performance-power consumption estimation model, and then estimate the target device's other values corresponding to these 9 parameters based on the performance-power consumption estimation model Combine the performance data and power consumption data to obtain the performance data and power consumption data of the target device under various value combinations corresponding to these 9 parameters, as shown in Figure 3. After that, the target device can set these 9 parameters based on its performance data and power consumption data under various value combinations corresponding to these 9 parameters.

Of course, the core state power management mechanism supported by the target device may include DVFS alone, or both DVFS and C-state, or C-state alone. In the case that the power management mechanism of the kernel state supported by the target device separately includes the C-state, at least one kernel parameter related to the power management mechanism includes: the entry time threshold corresponding to each level of the C mode. If the C-state includes 11 levels of C mode, at least one kernel parameter includes the entry time threshold corresponding to the 11 levels of C mode, for a total of 11 parameters. The model computing device 101 can use the load test tool to test the performance data and power consumption data of the target device under the partial value combinations corresponding to these 11 parameters, and use the performance data of the target device under the partial value combinations corresponding to these 11 parameters. Perform model training with power consumption data to obtain a performance-power consumption estimation model, and then estimate the performance data and power consumption data of the target device under other combinations of values corresponding to these 11 parameters based on the performance-power estimation model. Obtain the performance data and power consumption data of the target device under various value combinations corresponding to these 11 parameters; the target device is based on its performance data and power consumption data under various value combinations corresponding to these 11 parameters , To set these 11 parameters.

In the embodiment of the present application, for the power management mechanism in the kernel mode, the model computing device 101 can combine the setting operation of the kernel parameters related to the power management mechanism with artificial intelligence, and train the performance-power consumption prediction based on artificial intelligence. Based on the performance-power estimation model, the performance data and power consumption data of the physical device under multiple value combinations corresponding to the relevant kernel parameters are obtained, so that the physical device can use its performance under multiple value combinations Based on the data and power consumption data, the relevant kernel parameters can be set, and appropriate values can be set for the relevant kernel parameters, taking into account the power consumption and performance of the physical device, and the combination of artificial intelligence can improve the efficiency of parameter setting and reduce the cost .

It is worth noting that, in the above-mentioned embodiment, in the setting operation of the kernel parameters related to the power management mechanism, the performance-power estimation model is trained by considering the two aspects of power consumption and performance, but not Limited to this.

For example, in some application scenarios, you may focus on device performance, and have low or no requirements for device power consumption. It is necessary to set reasonable values for the kernel parameters related to device performance; for this, the model computing device 101 can also use In a similar way, the performance prediction model is trained, and the physical device 103 is provided with its performance data under multiple value combinations corresponding to the device performance-related kernel parameters, so that the physical device 103 can use the device performance-related kernel accordingly. The parameters are set.

For the model computing device 101, at least one kernel parameter related to the device performance can be determined from the kernel parameters of the physical device; the load test tool is used to test the performance data of the target device under the partial value combination corresponding to the at least one kernel parameter ; Perform model training according to the performance data of the target device under some value combinations to obtain a performance prediction model; use the performance prediction model to predict the performance data of the target device under other value combinations corresponding to at least one kernel parameter. For the target device, at least one kernel parameter related to device performance can be determined from its kernel parameters; its performance data under multiple value combinations corresponding to at least one kernel parameter can be obtained from the model computing device 101; Determine the target value combination corresponding to at least one kernel parameter from the performance data under multiple value combinations; set at least one kernel parameter according to the target value combination to operate according to the value in the target value combination. Among them, the detailed implementation manners or alternative implementation manners of related descriptions can be referred to the foregoing embodiments, and details are not described herein again.

For another example, in some application scenarios, the power consumption of the device may be focused on, and the performance requirements of the device are not high or not required. It is necessary to set reasonable values for the kernel parameters related to the power consumption of the device; for this reason, the model computing device 101 also The power consumption estimation model can be trained in a similar manner, and the physical device 103 can be provided with its power consumption data under various combinations of values corresponding to the kernel parameters related to the power consumption of the device, so that the physical device 103 can respond accordingly. Kernel parameters related to device performance are set.

For the model computing device 101, at least one kernel parameter related to the power consumption of the device can be determined from the kernel parameters of the physical device; the load test tool is used to test the target device under the partial value combination corresponding to the at least one kernel parameter. The power consumption data of the target device; perform model training according to the power consumption data of the target device under partial value combinations to obtain the power consumption estimation model; use the power consumption estimation model to estimate the target device's other corresponding to the at least one core Power consumption data under the combination of values. For the target device, at least one kernel parameter related to the power consumption of the device can be determined from its kernel parameters; the power consumption of the at least one kernel parameter corresponding to multiple value combinations can be obtained from the model computing device 101 Data; determine the target value combination corresponding to at least one core parameter according to its power consumption data under multiple value combinations; set at least one core parameter according to the target value combination to follow the target value combination Value running. Among them, the detailed implementation manners or alternative implementation manners of related descriptions can be referred to the foregoing embodiments, and details are not described herein again.

In the foregoing embodiments of the present application, a data center system is taken as an example for description, but the solution of combining the kernel parameter setting operation with artificial intelligence provided in the embodiments of the present application is not limited to the data center system. The scheme of combining the kernel parameter setting operation with artificial intelligence provided in the embodiment of the application can be extended to any system or device that needs to set the kernel parameter of the device.

Fig. 4a is a schematic structural diagram of a device management system provided by an exemplary embodiment of this application. As shown in FIG. 4a, the device management system 400 includes: at least one physical device 401 and at least one model computing device 402. This embodiment does not limit the number of physical devices 401, and it may be one or multiple. Similarly, this embodiment does not limit the number of model computing devices 402, and it may be one or multiple.

In addition, this embodiment does not limit the device form of the physical device 401 and the model computing device 402. The physical device 401 may include, but is not limited to, at least one of the following device forms: server devices, computer devices, desktop computers, notebook computers, smart phones, tablet computers, network switching devices, and the like. The server device may include, but is not limited to, a conventional server, a server array, or a cloud server. The model computing device 402 may be a server device such as a conventional server, a cloud server, or a server array. The device form and quantity of the physical device 401 and the model computing device 402 shown in FIG. 4a are only examples and are not limited thereto.

In this embodiment, at least one physical device 401 refers to a physical device installed with an operating system and supporting a kernel-mode power management mechanism. It should be noted that, in addition to the physical device 401 that is installed with an operating system and supports the kernel-mode power management mechanism, the device management system 400 may also include some devices that do not require an operating system, and some devices that are installed with an operating system but do not support There is no restriction on the devices with the power management mechanism in the kernel mode. In this embodiment and other embodiments, the focus is on the physical device 401 that is installed with an operating system and supports a kernel-mode power management mechanism.

The power management mechanism in the kernel mode supported by the physical device 401 is related to some kernel parameters, and these kernel parameters generally have multiple values. The different values of the kernel parameters result in different energy-saving effects in the power management mechanism. In order to produce better or optimal energy-saving effects under different application scenarios and application requirements, and meet performance requirements, in this embodiment, the kernel parameter setting operation is combined with artificial intelligence, and the artificial intelligence model is used as the basis to obtain The performance data and power consumption data of the physical device 401 under multiple value combinations corresponding to the relevant kernel parameters, and further, based on the performance data and power consumption data of the physical device 401 under multiple value combinations corresponding to the relevant kernel parameters , The kernel parameter setting can not only set the appropriate value for the kernel parameter, but also take into account the power consumption and performance requirements of the physical device 401 at the same time, and the combination of artificial intelligence can improve the parameter setting efficiency and reduce the cost.

In this embodiment, at least one model computing device 402 is mainly used to: for each physical device 401, determine from the kernel parameters of the physical device 401 related to the power management mechanism of the kernel state supported by the physical device 401 At least one kernel parameter, based on the artificial intelligence model, provides the physical device 401 with its performance data and power consumption data under multiple value combinations corresponding to at least one kernel parameter, so that the physical device 401 can complete the relevant kernel parameter Set up.

For ease of description, this embodiment takes any physical device 401 as an example, and records any physical device 401 as a target device to describe the detailed working principle of the system.

The model computing device 402 may determine at least one kernel parameter related to the power management mechanism of the kernel mode supported by the target device from the kernel parameters of the target device. After that, the model computing device 402 can collect the performance data and power consumption data of the target device under the partial value combination corresponding to at least one core parameter as sample data, and use these sample data for model training to obtain the performance-power consumption estimate model.

Optionally, the model computing device 402 may use a load test tool to collect performance data and power consumption data of the target device under a partial value combination corresponding to at least one kernel parameter. Install and run the load test tool on the target device. The load test tool can simulate the performance data and power consumption data of the target device under different load conditions. In detail, from multiple value combinations corresponding to at least one kernel parameter, select some value combinations, and the number of some value combinations only needs to meet the number required for model training; for each of the partial value combinations According to the load, power consumption and/or performance requirements that the target device needs to meet, the load test tool is used to obtain the performance data and power consumption data of the target device when the value combination meets the corresponding requirements. The performance data and power consumption data when the corresponding requirements are met under the partial value combination are provided to the model computing device 402. In order to simplify the description, the performance data and power consumption data of the target device when it meets the corresponding requirements under each value combination are referred to as the performance data and power consumption data of the target device under each value combination for short.

After obtaining the performance data and power consumption data of the target device under the partial value combination corresponding to at least one kernel parameter, the model computing device 402 uses these data as sample data for model training. In this embodiment, the process of model training performed by the model computing device 402 is not limited. For example, it can be a model training process based on a deep neural network, or a model training process based on regression analysis, where at least one kernel parameter can be analyzed. The model training method of the association relationship between the value combination of and the performance data and power consumption data of the target device is applicable to the embodiments of the present application.

In an optional embodiment, the model computing device 402 may adopt a regression analysis-based modeling method, that is, perform regression analysis on the performance data and power consumption data of the target device under a combination of partial values corresponding to at least one kernel parameter. Regression analysis can obtain the correlation between the value combination corresponding to at least one kernel parameter and the performance data and power consumption data of the target device, that is, the performance-power consumption prediction model.

Further optionally, choose to use linear regression analysis for modeling. Based on this, the process of model training by the model computing device 402 includes: taking the partial value combination corresponding to at least one kernel parameter as an independent variable, and using the performance data and power consumption data of the target device under the partial value combination as the dependent variable to perform linearization. Regression analysis, the performance-power consumption prediction model is obtained.

In an optional embodiment, after the performance-power consumption prediction model is obtained, the model calculation device 402 can use the performance-power consumption prediction model to predict the performance of the target device under other value combinations corresponding to at least one core parameter. Data and power consumption data. Or, after obtaining the performance-power consumption prediction model, the model computing device 101 may use the performance-power consumption prediction model to estimate the performance data and power consumption data of the target device under multiple value combinations corresponding to at least one core parameter .

Optionally, an implementation manner of determining the target value combination corresponding to at least one kernel parameter includes: directly determining from the multiple value combinations according to the performance data and power consumption data of the target device under multiple value combinations Get the target value combination. Alternatively, another implementation manner for determining the target value combination corresponding to at least one kernel parameter includes: obtaining power consumption data and/or performance data that the target device actually needs to meet; and comparing the power consumption data and/or performance data that the target device actually needs to meet Or performance data to match the performance data and power consumption data of the target device under multiple value combinations; according to the power consumption data and/or performance data that the target device actually needs to meet with the target device under multiple value combinations For the matching degree between the performance data and the power consumption data, a value combination that meets the matching degree requirement is selected from the multiple value combinations as the target value combination. Alternatively, another implementation manner of determining the target value combination corresponding to at least one kernel parameter includes: selecting at least one of the multiple value combinations according to the performance data and power consumption data of the target device under multiple value combinations A combination of candidate values; using a load test tool to test the performance data and power consumption data of the target device under at least one combination of candidate values; further, the target device tested according to the load test tool under at least one combination of candidate values Determine the target value combination from at least one candidate value combination based on the performance data and power consumption data. In this embodiment, in addition to the performance data and power consumption data of the target device under a variety of value combinations, the load test tool is also combined to help improve the accuracy of the final selected target value combination, which is beneficial to Improve the accuracy of parameter settings.

Further, in the process of selecting at least one combination of candidate values, the power consumption data that the target device actually needs to meet can be combined, or the performance data that the target device actually needs to meet, or the power consumption data that the target device actually needs to meet at the same time. And performance data. For example, the power consumption data that the target device actually needs to meet can be obtained, and the power consumption data and/or performance data that the target device actually needs to meet can be combined with the performance data of the target device under a combination of multiple values corresponding to at least one kernel parameter. Matching with the power consumption data; from a variety of value combinations, obtain a value combination whose matching degree with the power consumption data and/or performance data that the target device actually needs to meet the matching degree requirement as a candidate value combination.

Further, in the above embodiment, after the target value combination is selected, the performance data and power consumption data of the target device under the target value combination tested by the load test tool can be used to evaluate the performance-power consumption prediction model Make corrections. Among them, the performance data and power consumption data of the target device tested by the load test tool under the target value combination are more in line with the actual requirements, which are used as sample data to modify the performance-power consumption prediction model, which is conducive to improving the performance-power The accuracy of the consumption estimation model makes the estimated performance data and power consumption data more in line with actual requirements.

For the related description of the model computing device 402, please refer to the related description of the model computing device 101 in the foregoing embodiment. Similarly, for the related description of the target device, please refer to the corresponding description in the foregoing embodiment, which will not be repeated here.

It is explained here that the device management system 400 of this embodiment can be implemented as a system of any form or nature, for example, it can be a data center, or a cluster, or a computer room system, or an edge cloud network system, or a central cloud network system, etc. The embodiment does not limit this.

For example, the device management system 400 is implemented as an edge cloud network system as an example for illustration. As shown in Figure 4b, the edge cloud network system includes: edge computing nodes and servers deployed in the cloud or in the client room. The server communicates with the edge computing node through the network. The server can respond to the request of the edge computing node and provide related cloud services for the edge computing node; in addition, the server can also control, operate and maintain the edge computing node. Edge computing nodes include hardware infrastructure, hardware infrastructure drivers, operating systems, and related applications. The hardware infrastructure includes but is not limited to: CPU, network card, memory, etc. Edge computing nodes may include: base stations, home gateways, personal computers, smart phones, street lights, traffic lights, and/or electronic monitoring equipment on buildings that are added to the edge cloud network.

In the edge cloud network system shown in Figure 4b, the server can have the functions of the model computing device 402 in Figure 4a. Of course, another server dedicated to model training can also be deployed (used to implement the model computing device 402 in Figure 4a). The function of the edge computing node can be used as the physical device 401 in Figure 4a. Assume that the edge cloud node supports two mechanisms, DFVS and C-state, and C-state has 6 levels of C modes. Among them, the kernel parameters related to DFVS and C-state include: scaling_min_freq, scaling_max_freq, scaling_governor, C1-C6 levels The corresponding target_residency has 9 parameters.

Before the edge computing node provides services to customers, it is necessary to install the operating system on the edge computing node for the customer, and set the kernel parameters related to DFVS and C-state in the edge computing node, so that the edge computing node can work in the optimal parameter combination To achieve the goal of balancing performance and power consumption.

In order to achieve the above purpose, the server can use the load test tool to test the performance data and power consumption data of the edge computing node under the partial value combination corresponding to the above 9 parameters, and use the edge computing node to select the partial value combination corresponding to the 9 parameters. Perform model training on the performance data and power consumption data below to obtain the performance-power consumption prediction model; further, based on the performance-power consumption prediction model, predict the performance of the edge computing node under other combinations of values corresponding to these 9 parameters Data and power consumption data, so as to obtain the performance data and power consumption data of the edge computing node under the various value combinations corresponding to these 9 parameters; the edge computing node’s performance data and power consumption data under the various value combinations corresponding to these 9 parameters Performance data and power consumption data are delivered to edge computing nodes. The edge computing node receives its performance data and power consumption data under various value combinations corresponding to these 9 parameters issued by the server, and uses its performance data under various value combinations corresponding to these 9 parameters and Based on the power consumption data, these 9 parameters are set so that the edge computing node can work under the optimal parameter combination after startup, and reduce power consumption as much as possible while ensuring performance.

FIG. 5a is a schematic flowchart of a data processing method provided by an exemplary embodiment of this application. This data processing method can be used to set the kernel parameters of the physical device. As shown in Figure 5a, the method includes:

501. From the kernel parameters of the physical device, determine at least one kernel parameter related to the power management mechanism of the kernel mode supported by the physical device.

502. Acquire performance data and power consumption data of the physical device under multiple value combinations corresponding to the at least one core parameter above, where at least some of the performance data and power consumption data under the value combination are based on performance-power consumption estimation Estimated by the model.

503. Determine a target value combination corresponding to at least one kernel parameter according to the performance data and power consumption data of the physical device under the foregoing multiple value combinations.

504. Set the aforementioned at least one kernel parameter according to the target value combination, so that the power consumption management mechanism operates according to the value in the target value combination.

The execution subject of this embodiment may be a physical device, and the physical device runs an operating system and supports a kernel-mode power management mechanism. The physical device can adjust or manage the kernel power management mechanism through some kernel parameters, and these kernel parameters are also the parameters of the kernel power management mechanism. The kernel parameters related to the kernel-mode power management mechanism may have multiple values, and under different values, the energy-saving effect that the power management mechanism can produce will be different.

In this embodiment, the kernel parameter setting operation is combined with artificial intelligence, and based on the artificial intelligence model, the performance data and power consumption data of the physical device under multiple value combinations corresponding to the relevant kernel parameters are obtained, so that the physical device can be Based on the performance data and power consumption data under the multiple value combinations corresponding to the relevant kernel parameters, the relevant kernel parameters can be set, and appropriate values can be set for the relevant kernel parameters, taking into account the power consumption and performance of the physical device. Combining with artificial intelligence can improve the efficiency of parameter setting and reduce costs.

In an optional embodiment, an implementation manner of step 503 includes: selecting at least one candidate value combination from the multiple value combinations according to the performance data and power consumption data of the physical device under multiple value combinations ; Use a load test tool to test the performance data and power consumption data of the physical device under at least one candidate value combination; use the load test tool to test the performance data and power consumption data of the physical device under at least one candidate value combination , Determine the target value combination from at least one candidate value combination.

Further, after determining the target value combination, it also includes: using the performance data and power consumption data of the physical device tested by the load test tool under the target value combination to modify the performance-power consumption prediction model.

Further, the above-mentioned use of the load test tool to test the performance data and power consumption data of the physical device under at least one candidate value combination includes: for the first candidate value combination, the value of at least one kernel parameter in the operating system source code Modify to the value in the first candidate value combination, and install the operating system on the physical device according to the modified operating system source code; run the load test tool on the physical device to test the physical device in the first candidate value combination Performance data and power consumption data below; where the first candidate value combination is any one of at least one candidate value combination.

Further, the foregoing selecting at least one candidate value combination from the multiple value combinations according to the performance data and power consumption data of the physical device under multiple value combinations includes: obtaining power consumption data that the physical device actually needs to meet; Match the power consumption data that the physical device actually needs to meet with the performance data and power consumption data of the physical device under multiple value combinations; obtain the power consumption data that the physical device actually needs to meet from multiple value combinations The matching degree of at least one candidate value combination that meets the matching degree requirement. Alternatively, another implementation manner of step 503 includes: obtaining the performance data that the physical device actually needs to meet, and combining the performance data that the physical device actually needs to meet with the physical device under a combination of multiple values corresponding to at least one kernel parameter. The performance data and the power consumption data are matched; from the multiple value combinations, at least one candidate value combination whose matching degree with the performance data actually needs to be met by the physical device meets the matching degree requirement is obtained. Or, another implementation manner of step 503 includes: obtaining performance data and power consumption data that the physical device actually needs to meet, and corresponding the performance data and power consumption data that the physical device actually needs to meet with the physical device in at least one kernel parameter Match the performance data and power consumption data under multiple value combinations of, from multiple value combinations, obtain the performance data and power consumption data that the physical device actually needs to meet the matching degree that meets at least one of the matching degree requirements A combination of candidate values.

Further optionally, obtaining the power consumption data that the physical device actually needs to meet includes: analyzing the power consumption data that the physical device actually needs to meet according to the QoS of the service that the physical device actually needs to carry; or, obtaining the power required by the service deployment user The power consumption data is the power consumption data that the physical device actually needs to meet.

Further optionally, obtaining the performance data that the physical device actually needs to meet includes: analyzing the performance data that the physical device actually needs to meet according to the QoS of the service that the physical device actually needs to carry; or, obtaining the performance data required by the service deployment user, As the actual performance data that physical equipment needs to meet.

Further optionally, obtaining the power consumption data and performance data that the physical device actually needs to meet includes: analyzing the power consumption data and performance data that the physical device actually needs to meet according to the QoS of the service that the physical device actually needs to carry; or, obtaining The performance data and performance data required by the service deployment user are the power consumption data and performance data that the physical device actually needs to meet.

In an optional embodiment, an implementation manner of step 502 includes: using a load test tool to test the performance data and power consumption data of the physical device under partial value combinations among the above-mentioned multiple value combinations; Perform model training on performance data and power consumption data under partial value combinations to obtain the performance-power consumption prediction model; use the performance-power consumption prediction model to predict the physical device under other value combinations among multiple value combinations Performance data and power consumption data.

Further optionally, model training is performed according to the performance data and power consumption data of the physical device under partial value combinations to obtain the performance-power consumption prediction model, including: performance data and power consumption data of the physical device under partial value combinations The power consumption data is subjected to regression analysis to obtain the performance-power consumption prediction model.

Furthermore, perform regression analysis on the performance data and power consumption data of the physical device under partial value combinations to obtain the performance-power consumption prediction model, including: taking some value combinations as independent variables, and taking the physical device in part The performance data and power consumption data under the value combination are used as dependent variables to perform linear regression analysis to obtain a performance-power consumption prediction model.

In another optional embodiment, acquiring the performance data and power consumption data of the physical device under multiple value combinations corresponding to at least one kernel parameter includes: receiving the physical device sent by the model computing device corresponding to the at least one kernel parameter Performance data and power consumption data under multiple value combinations; among them, the performance data and power consumption data under at least some value combinations are estimated by the model computing device based on the performance-power consumption prediction model. For the detailed implementation of the performance-power consumption prediction model obtained by the model computing device and the performance-power consumption prediction model estimated based on the performance-power consumption prediction model, the detailed implementation of the performance data and power consumption data of the physical device under at least partial value combinations can be found in the foregoing embodiments. This will not be repeated here.

FIG. 5b is a schematic flowchart of another data processing method provided by an exemplary embodiment of this application. As shown in Figure 5b, the method includes:

51. From the kernel parameters of the physical device, determine at least one kernel parameter related to the power management mechanism of the kernel mode supported by the physical device.

52. Use a load test tool to test the performance data and power consumption data of the physical device under the partial value combination corresponding to the at least one kernel parameter.

53. Perform model training according to the performance data and power consumption data of the physical device under the above-mentioned partial value combination to obtain a performance-power consumption prediction model.

54. Use the performance-power consumption estimation model to estimate the performance data and power consumption data of the physical device under other value combinations corresponding to the at least one kernel parameter mentioned above.

In an optional embodiment, an implementation manner of step 53 includes: performing regression analysis on the performance data and power consumption data of the physical device under the above-mentioned partial value combinations to obtain a performance-power consumption prediction model.

Further, the linear regression analysis method can be used for model training. The model training process of obtaining the performance-power consumption prediction model includes: using the above combination of values as independent variables, and using the performance data and power consumption data of the physical device under the above combination of values as dependent variables to perform linear regression analysis to obtain Performance-power consumption prediction model.

In this embodiment, combined with artificial intelligence, only the performance data and power consumption data of the physical device under the partial value combination corresponding to the above at least one kernel parameter are collected, and the performance-power consumption prediction model is trained based on the collected data. Based on the performance-power consumption prediction model, the performance data and power consumption data of the physical device under other combinations of values corresponding to at least one of the above-mentioned core parameters are estimated. This can greatly improve the efficiency of data acquisition, reduce costs, and perform processing for the physical device. Kernel parameter settings provide data conditions.

In addition to the foregoing method embodiments, this application also provides the following method embodiments.

As shown in FIG. 6a, another embodiment of a data processing method includes:

601. Determine at least one kernel parameter related to the performance of the device from the kernel parameters of the physical device.

602. Acquire performance data of the physical device under multiple value combinations corresponding to the at least one kernel parameter, where at least part of the performance data under the value combination is estimated based on a performance prediction model.

603. Determine the target value combination corresponding to at least one core parameter according to the performance data of the physical device under the foregoing multiple value combinations.

604. Set the at least one kernel parameter mentioned above according to the target value combination, so that the physical device operates according to the value in the target value combination.

In this embodiment, the kernel parameter setting operation is combined with artificial intelligence, and based on the artificial intelligence model, the performance data of the physical device under multiple value combinations corresponding to the relevant kernel parameters is obtained, so that the physical device can be used in the relevant kernel. Based on the performance data under the multiple value combinations corresponding to the parameters, the relevant kernel parameters can be set, and appropriate values can be set for the relevant kernel parameters, which is conducive to ensuring the performance of the physical device, and the combination with artificial intelligence can improve Parameter setting efficiency and cost reduction.

As shown in Figure 6b, another embodiment of a data processing method includes:

61. From the kernel parameters of the physical device, determine at least one kernel parameter related to the performance of the device.

62. Use a load test tool to test the performance data of the physical device under the partial value combination corresponding to the at least one kernel parameter.

63. Perform model training according to the performance data of the physical equipment under the above combination of partial values to obtain a performance prediction model.

64. Use the performance prediction model to predict the performance data of the physical device under other value combinations corresponding to at least one kernel parameter.

In this embodiment, combined with artificial intelligence, only the performance data of the physical device under the partial value combination corresponding to the above-mentioned at least one kernel parameter is collected, and the performance prediction model is trained based on the collected data, and the performance prediction model is estimated based on the performance prediction model. The performance data of the physical device under other value combinations corresponding to the above at least one kernel parameter can be obtained, which can greatly improve the efficiency of data acquisition, reduce the cost, and provide data conditions for the physical device to perform the kernel parameter setting.

As shown in FIG. 7a, another embodiment of a data processing method includes:

701. From the kernel parameters of the physical device, determine at least one kernel parameter related to the power consumption of the device.

702. Acquire power consumption data of the physical device under multiple value combinations corresponding to the at least one kernel parameter, where at least part of the power consumption data under the value combination is estimated based on a power consumption estimation model.

703. Determine a target value combination corresponding to at least one kernel parameter according to the power consumption data of the physical device under the foregoing multiple value combinations.

704. Set the at least one kernel parameter mentioned above according to the target value combination, so that the physical device operates according to the value in the target value combination.

In this embodiment, the kernel parameter setting operation is combined with artificial intelligence, and based on the artificial intelligence model, the power consumption data of the physical device under multiple value combinations corresponding to the relevant kernel parameter is obtained, so that the physical device can be Based on the power consumption data under the multiple value combinations corresponding to the kernel parameters, the relevant kernel parameters can be set, and appropriate values can be set for the relevant kernel parameters, which is conducive to reducing the power consumption of physical devices and combining with artificial intelligence , Which can improve the efficiency of parameter setting and reduce costs.

As shown in Figure 7b, another embodiment of a data processing method includes:

71. From the kernel parameters of the physical device, determine at least one kernel parameter related to the power consumption of the device.

72. Use a load test tool to test the power consumption data of the physical device under the partial value combination corresponding to the at least one kernel parameter.

73. Perform model training according to the power consumption data of the physical device under partial value combinations to obtain a power consumption prediction model.

74. Use the power consumption estimation model to estimate the power consumption data of the physical device under other value combinations corresponding to the at least one kernel parameter mentioned above.

In this embodiment, combined with artificial intelligence, only the power consumption data of the physical device under the partial value combination corresponding to the above at least one kernel parameter is collected, and the power consumption estimation model is trained based on the collected data, and the power consumption estimation is based on The model predicts the power consumption data of the physical device under other value combinations corresponding to at least one of the above-mentioned kernel parameters, which can greatly improve data acquisition efficiency, reduce costs, and provide data conditions for the physical device to set the kernel parameters.

As shown in FIG. 7c, another embodiment of a data processing method includes:

711. From the kernel parameters of the physical device, determine at least one kernel parameter related to the power management mechanism of the kernel mode supported by the physical device.

712. Use a load test tool to test the performance data and power consumption data of the physical device under the partial value combination corresponding to the at least one kernel parameter.

713. Perform model training according to the performance data and power consumption data of the physical device under partial value combinations to obtain a performance-power consumption prediction model.

714. Use the performance-power consumption estimation model to estimate the performance data and power consumption data of the physical device under multiple value combinations corresponding to the at least one kernel parameter; wherein, the multiple value combinations include the aforementioned partial values combination.

In addition to the foregoing data processing method, the embodiment of the present application also provides a task scheduling method. As shown in Figure 7d, the task scheduling method includes:

721. Obtain the to-be-scheduled task and the performance requirements of the to-be-scheduled task.

722. From at least one resource device, select a resource device that meets the foregoing performance requirements and that the value of the core state power consumption parameter meets the set power consumption requirement.

723. Schedule the task to be scheduled to a resource device that meets the foregoing performance requirements and the core state power consumption parameter value meets the set power consumption requirement.

In this embodiment, the value of the power consumption parameter in the kernel mode refers to a combination of values of at least one kernel parameter related to the power management mechanism of the kernel mode supported by the resource device. Among them, the power management mechanism in the kernel mode refers to a mechanism provided by the operating system to manage the power consumption of physical devices. It is a power management mechanism at the operating system level and needs to run in the kernel mode.

In this embodiment, the type of operating system is not limited. For example, it may be a Linux operating system, a Windows operating system, a UNIX operating system, or a MAC operating system. The power management mechanism of the kernel mode supported by different operating systems will be different. Among them, DVFS and C-state are examples of two kernel-state power management mechanisms supported by the Linux operating system.

Among them, the power management mechanism in the kernel mode is related to some kernel parameters. These kernel parameters are also the parameters of the power management mechanism in the kernel mode. The value of these kernel parameters can affect the energy saving effect of the power management mechanism in the kernel mode. For any resource device, the prerequisite for the normal use of the power management mechanism in the kernel mode is to set the values of the kernel parameters related to the power management mechanism in the kernel mode in advance. This means that at least one kernel parameter related to the power management mechanism of the kernel mode of the device in the normal working state has a certain value, and the combination of these values is the kernel mode power consumption parameter value of this embodiment. The core state power consumption parameter value of the resource device can reflect the power consumption of the resource device to a certain extent.

Based on the above considerations, in scenarios that require task scheduling such as computer rooms, data centers, or edge cloud networks, task scheduling can be performed in combination with the core state power consumption parameter values of resource devices. Among them, resource devices refer to various physical devices responsible for performing tasks in resource scheduling scenarios. For example, they can be terminal devices such as laptops, tablets, smart phones, or edge computing nodes, or they can be servers, server clusters, and server arrays. Or server-side equipment such as cloud servers. For example, the resource device can be a server or a server cluster, but it is not limited to this.

In a task scheduling scenario, a task scheduling device (or called a task scheduler) is generally deployed, which is mainly used to schedule tasks to be scheduled to appropriate resource devices. The task scheduling device can run program codes such as task scheduling services, programs, or plug-ins to implement the following task scheduling process.

In the task scheduling process, the task to be scheduled and the performance requirements of the task to be scheduled need to be obtained. Optionally, the task scheduling device may provide a human-computer interaction interface for the user, and the user may provide the tasks to be scheduled and their performance requirements through the human-computer interaction interface. The human-computer interaction interface can be an application page, a web page or a command window. Alternatively, the task scheduling device can also support voice interaction and recognition technology, and users can submit tasks to be scheduled and their performance requirements through voice instructions. The performance requirements of different tasks to be scheduled will be different, which is not limited.

After that, according to the performance requirements of the task to be scheduled, from at least one resource device, a resource device that meets the performance requirement and the core state power consumption parameter value meets the set power consumption requirement can be selected from at least one resource device; the task to be scheduled is scheduled to meet the performance requirement And the resource device whose core state power consumption parameter value meets the set power consumption requirement.

Optionally, in the process of selecting a resource device that meets the performance requirements and that the core-mode power consumption parameter value meets the set power consumption requirement, the performance data and performance of at least one resource device under the respective core-mode power consumption parameter value may be selected. Consumption data. That is, according to the performance data and power consumption data of at least one resource device under the respective core-state power consumption parameter values, a resource device that meets the performance requirement and the core-mode power consumption parameter value meets the set power consumption requirement is selected.

In an optional implementation manner, according to the performance data and power consumption data of at least one resource device under the respective core-state power consumption parameter values, the one that meets the performance requirements and the core-state power consumption parameter values meets the set power consumption requirements is selected. The resource device includes: selecting a candidate resource device that meets the above performance requirements according to the performance data of at least one resource device under the respective core state power consumption parameter value; according to the power consumption data of the candidate resource device under the respective core state power consumption parameter value , Select a resource device whose power consumption data meets the set power consumption requirement from the candidate resource devices. Wherein, the number of candidate resource devices meeting the foregoing performance requirements may be one or more. The power consumption data of the candidate resource device under the kernel state power consumption parameter value meets the set power consumption requirement, which means that the kernel state power consumption parameter value of the candidate resource device meets the set power consumption requirement.

In another optional implementation manner, according to the performance data and power consumption data of at least one resource device under the respective core-state power consumption parameter values, it is selected to meet the performance requirements and the core-mode power consumption parameter values meet the set power consumption requirements The resource device includes: selecting a candidate resource device whose core state power consumption parameter value meets the set power consumption requirement according to the power consumption data of at least one resource device under the respective core state power consumption parameter value; according to the candidate resource device in the respective core According to the performance data under the state power consumption parameter value, a resource device whose power consumption data meets the set power consumption requirement is selected from the candidate resource devices.

In the foregoing embodiments, the “set power consumption requirement” is not limited, and can be flexibly set according to application requirements. For example, in an optional embodiment, setting the power consumption requirement may require that the core state power consumption parameter value brings the lowest power consumption to the resource device. Based on this, the power consumption can be selected from candidate resource devices that meet the above performance requirements. The resource device with the lowest data consumption; then the task to be scheduled is scheduled to the resource device with the lowest data consumption. In this alternative embodiment, the resource device with the best core-mode power consumption parameter value can be selected under the condition of guaranteed performance.

Before using the performance data and power consumption data of at least one resource device under the respective core-state power consumption parameter values, it is necessary to first obtain the performance data and power consumption data of the at least one resource device under the respective core-state power consumption parameter values.

In some optional embodiments, a load test tool may be used to collect performance data and power consumption data of each resource device under its own core state power consumption parameter value.

In other optional embodiments, an artificial intelligence model can be pre-trained, such as a performance-power consumption prediction model. When the kernel state power consumption parameter value of each resource device is known, the performance-power consumption prediction model can be used. The estimation model predicts the performance data and power consumption data of each resource device under its own core state power consumption parameter value. Regarding the training process and the use process of the performance-power consumption estimation model, refer to the foregoing embodiment, which will not be repeated here.

In still other embodiments, the core-state power consumption parameter value of each resource device is set using the method provided in the foregoing embodiment of this application, and in the process of setting the core-mode power consumption parameter value for each resource device, it has been learned The performance data and power consumption data of each resource device under its own core state power consumption parameter value, and the performance data and power consumption data of each resource device learned in advance under its own core state power consumption parameter value can be saved, so that, In the task scheduling process, the performance data and power consumption data of each resource device under the respective core state power consumption parameter values can be directly read. It should be noted that the core state power consumption parameter value of each resource device refers to the target value combination finally set by the method provided in the foregoing embodiment.

In the task scheduling method provided by the embodiment of the present application, in combination with the core state power consumption parameter value of the resource device, the resource device with the better core state power consumption parameter value can be preferentially selected to execute the task when the performance requirements are met. While taking into account performance requirements, reduce power consumption.

It should be noted that the execution subject of each step of the method provided in the foregoing embodiment may be the same device, or different devices may also be the execution subject of the method. For example, the execution subject of steps 501 to 504 may be device A; for another example, the execution subject of

steps

501, 503, and 504 may be device A, and the execution subject of step 502 may be device B; and so on.

In addition, in some of the processes described in the above embodiments and drawings, multiple operations appearing in a specific order are included, but it should be clearly understood that these operations may be performed out of the order in which they appear in this document or performed in parallel. The sequence numbers of operations, such as 501, 502, etc., are only used to distinguish different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions of "first" and "second" in this article are used to distinguish different messages, devices, modules, etc., and do not represent a sequence, nor do they limit the "first" and "second" Are different types.

Fig. 8a is a schematic structural diagram of a physical device provided by an exemplary embodiment of this application. As shown in FIG. 8a, the physical device includes: a memory 81a and a processor 82a.

The memory 81a is used to store computer programs, and can be configured to store various other data to support operations on the physical device. Examples of such data include instructions for any application or method operated on the physical device, contact data, phone book data, messages, pictures, videos, etc.

The processor 82a is coupled to the memory 81a, and is configured to execute the computer program in the memory 81a to determine from the kernel parameters of the physical device related to the power management mechanism of the kernel mode supported by the physical device At least one kernel parameter; acquiring performance data and power consumption data of the physical device under multiple value combinations corresponding to the at least one kernel parameter, where at least part of the performance data and power consumption data under the value combination are based on The performance-power consumption prediction model predicts; according to the performance data and power consumption data of the physical device under the multiple value combinations, the target value combination corresponding to the at least one core parameter is determined; The target value combination sets the at least one kernel parameter so that the power consumption management mechanism operates according to the value in the target value combination.

In an optional embodiment, when the processor 82a determines the target value combination corresponding to the at least one kernel parameter, it is specifically configured to: according to the performance data and function of the physical device under the multiple value combinations Select at least one candidate value combination from the multiple value combinations; use a load test tool to test the performance data and power consumption data of the physical device under the at least one candidate value combination; The performance data and power consumption data of the physical device under the at least one candidate value combination tested by the load testing tool are used to determine the target value combination from the at least one candidate value combination.

Further optionally, the processor 82a is further configured to, after determining the target value combination, use the load test tool to test the performance data and power consumption data of the physical device under the target value combination , Revise the performance-power consumption prediction model.

Further, when the processor 82a uses a load testing tool to test the performance data and power consumption data of the physical device under the at least one candidate value combination, it is specifically configured to: for the first candidate value combination, set the operating system The value of the at least one kernel parameter in the source code is modified to the value in the first candidate value combination, and the operating system is installed on the physical device according to the modified operating system source code; Run the load test tool on the device to test the performance data and power consumption data of the physical device under the first candidate value combination; wherein, the first candidate value combination is the at least one candidate Any combination of candidate values in the value combination.

Further, when the processor 82a selects at least one candidate value combination from the multiple value combinations, it is specifically configured to: obtain power consumption data that the physical device actually needs to meet; and compare the power consumption data that the physical device actually needs to meet , To match the performance data and power consumption data of the physical device under multiple value combinations; from the multiple value combinations, obtain at least one of the matching degrees with the power consumption data that the physical device actually needs to meet. A combination of candidate values. or,

In an optional embodiment, when the processor 82a selects at least one candidate value combination from the multiple value combinations, it is specifically configured to: obtain performance data that the physical device actually needs to meet, and compare the physical device's actual needs Satisfied performance data is matched with the performance data and power consumption data of the physical device under multiple value combinations corresponding to at least one core parameter; from multiple value combinations, the performance data that the physical device actually needs to meet is obtained The matching degree of at least one candidate value combination that meets the matching degree requirement. or,

In an optional embodiment, when the processor 82a selects at least one candidate value combination from the multiple value combinations, it is specifically configured to: obtain the performance data and power consumption data that the physical device actually needs to meet, and The performance data and power consumption data that the physical device actually needs to meet are matched with the performance data and power consumption data of the physical device under multiple value combinations corresponding to at least one core parameter; from multiple value combinations, The matching degree of the performance data and the power consumption data that the physical device actually needs to meet all meets at least one candidate value combination that meets the matching degree requirement.

Further optionally, when obtaining the power consumption data that the physical device actually needs to meet, the processor 82a is specifically configured to: analyze the power consumption data that the physical device actually needs to meet according to the QoS of the service that the physical device actually needs to carry; or, Obtain the power consumption data required by the service deployment user as the power consumption data that the physical device actually needs to meet.

Further optionally, when the processor 82a obtains the performance data that the physical device actually needs to meet, it is specifically used to: analyze the performance data that the physical device actually needs to meet according to the QoS of the service that the physical device actually needs to carry; or, obtain the service Deploy the performance data required by the user as the actual performance data that the physical device needs to meet.

Further optionally, when the processor 82a obtains the power consumption data and performance data that the physical device actually needs to meet, it is specifically used to: analyze the power consumption data that the physical device actually needs to meet according to the QoS of the service that the physical device actually needs to carry And performance data; or, obtain the performance data and performance data required by the service deployment user as the power consumption data and performance data that the physical device actually needs to meet.

In an optional embodiment, when the processor 82a obtains the performance data and power consumption data of the physical device under multiple value combinations corresponding to at least one kernel parameter, it is specifically configured to: use a load test tool to test that the physical device is in the above-mentioned Performance data and power consumption data under partial value combinations among multiple value combinations; perform model training based on the performance data and power consumption data of the physical device under the above partial value combinations to obtain a performance-power consumption prediction model; Use the performance-power consumption prediction model to estimate the performance data and power consumption data of the physical device under other value combinations among multiple value combinations.

Further optionally, when the processor 82a obtains the performance-power consumption prediction model, it is specifically used to: perform regression analysis on the performance data and power consumption data of the physical device under partial value combinations to obtain the performance-power consumption prediction model. Estimate the model.

Furthermore, the processor 82a is specifically configured to: use partial value combinations as independent variables, and use the performance data and power consumption data of the physical device under the partial value combinations as dependent variables to perform linear regression analysis to obtain performance-power consumption estimates model.

In an optional embodiment, when the processor 82a obtains the performance data and power consumption data of the physical device under multiple value combinations corresponding to at least one kernel parameter, it is specifically configured to: receive the physical device sent by the model computing device Performance data and power consumption data under multiple value combinations corresponding to at least one core parameter; among them, the performance data and power consumption data under at least some value combinations are estimated by the model computing device based on the performance-power consumption prediction model of. For the detailed implementation of the performance-power consumption prediction model obtained by the model computing device and the performance-power consumption prediction model estimated based on the performance-power consumption prediction model, the detailed implementation of the performance data and power consumption data of the physical device under at least partial value combinations can be found in the foregoing embodiments. This will not be repeated here.

Optionally, in addition to the above-mentioned functions of the physical device of this embodiment, the processor 82 may also implement the following functions: determine at least one kernel parameter related to the performance of the device from the kernel parameters of the physical device; The performance data under multiple value combinations corresponding to the at least one kernel parameter, wherein the performance data under at least part of the value combinations is estimated based on the performance prediction model; according to the physical device in the multiple value combinations The performance data under the value combination, determine the target value combination corresponding to the at least one kernel parameter; set the at least one kernel parameter according to the target value combination, so that the physical device takes the value according to the target The value in the combination runs.

Optionally, in addition to the above-mentioned functions of the physical device of this embodiment, the processor 82 may also implement the following functions: determine at least one kernel parameter related to the power consumption of the device from the kernel parameters of the physical device; and obtain the physical device. The power data of the device under multiple value combinations corresponding to the at least one kernel parameter, wherein the power data under at least some of the value combinations is estimated based on the power estimation model; according to the physical device in the Determine the target value combination corresponding to the at least one kernel parameter based on the power data under multiple value combinations; set the at least one kernel parameter according to the target value combination, so that the physical device follows the The value in the target value combination runs.

Further, as shown in FIG. 8a, the physical device further includes: a communication component 83a, a display 84a, a power supply component 85a, an audio component 86a and other components. Only part of the components are schematically shown in FIG. 8a, which does not mean that the physical device only includes the components shown in FIG. 8a. In addition, depending on the implementation form of the physical device, the components in the dashed box in FIG. 8a are optional components, not mandatory components. For example, when the physical device is implemented as a terminal device such as a smart phone, a tablet computer, or a desktop computer, it can include the components in the dashed box in Figure 8a; when the physical device is implemented as a server such as a conventional server, a cloud server, a data center, or a server array When the device is used, the components in the dashed box in Figure 8a may not be included.

Correspondingly, an embodiment of the present application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor will cause the processor to implement the method in the method embodiment shown in FIG. 5a, FIG. 6a, or FIG. 7a. The steps.

Fig. 8b is a schematic structural diagram of a model computing device provided by an exemplary embodiment of this application. As shown in FIG. 8b, the physical device includes: a memory 81b and a processor 82b.

The memory 81b is used to store computer programs, and can be configured to store various other data to support operations on the physical device. Examples of such data include instructions for any application or method operated on the physical device, contact data, phone book data, messages, pictures, videos, etc.

The processor 82b is coupled with the memory 81b, and is configured to execute the computer program in the memory 81b to determine at least one related to the power management mechanism of the kernel state supported by the physical device from the kernel parameters of the physical device Kernel parameters; use a load test tool to test the performance data and power consumption data of the physical device under the partial value combination corresponding to the at least one kernel parameter; according to the performance data of the physical device under the partial value combination Perform model training with power consumption data to obtain a performance-power consumption prediction model; use the performance-power consumption prediction model to predict the performance of the physical device under other value combinations corresponding to the at least one core parameter Data and power consumption data.

In an optional embodiment, when the processor 82b obtains the performance-power consumption prediction model, it is specifically configured to: perform regression analysis on the performance data and power consumption data of the physical device under the partial value combination, In order to get the performance-power consumption prediction model.

Further optionally, the processor 82b is specifically configured to: use the partial value combination as an independent variable, and use the performance data and power consumption data of the physical device under the partial value combination as the dependent variable to perform linear regression analysis , Get the performance-power consumption prediction model.

Optionally, in addition to the above-mentioned functions, the model computing device of this embodiment may also implement the following functions: determine at least one kernel parameter related to device performance from the kernel parameters of the physical device; use a load test tool Testing the performance data of the physical device under the partial value combination corresponding to the at least one kernel parameter; performing model training according to the performance data of the physical device under the partial value combination to obtain a performance prediction model; The performance prediction model is used to predict the performance data of the physical device under other value combinations corresponding to the at least one kernel parameter.

Optionally, in addition to the above-mentioned functions, the model computing device of this embodiment may also implement the following functions: determine at least one kernel parameter related to the power consumption of the device from the kernel parameters of the physical device; A tool to test the power data of the physical device under the partial value combination corresponding to the at least one kernel parameter; perform model training according to the power data of the physical device under the partial value combination to obtain a power estimation model ; Using the power estimation model to predict the power data of the physical device under other value combinations corresponding to the at least one kernel parameter.

Optionally, in addition to the above-mentioned functions, the model computing device of this embodiment may also implement the following functions: from the kernel parameters of the physical device, determine the power management mechanism of the kernel state supported by the physical device Relevant at least one kernel parameter; use a load test tool to test the performance data and power consumption data of the physical device under the partial value combination corresponding to the at least one kernel parameter; according to the physical device in the partial value combination Perform model training on the performance data and power consumption data under the following performance data to obtain a performance-power consumption prediction model; use the performance-power consumption prediction model to estimate that the physical device is in multiple selections corresponding to the at least one core parameter. Performance data and power consumption data under value combinations; wherein the multiple value combinations include the partial value combinations.

Further, as shown in FIG. 8b, the model computing device further includes: a communication component 83b, a power supply component 85b and other components. Only part of the components are schematically shown in FIG. 8b, which does not mean that the model computing device only includes the components shown in FIG. 8b. Optionally, the model computing device can be implemented as a server device such as a conventional server, a cloud server, a data center, or a server array.

Correspondingly, an embodiment of the present application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor causes the processor to implement the method shown in FIG. 5b, FIG. 6b, FIG. 7b, or FIG. 7c. The steps in the example.

Fig. 8c is a schematic structural diagram of a task scheduling device provided by an exemplary embodiment of this application. As shown in FIG. 8c, the task scheduling device includes: a memory 81c and a processor 82c.

The memory 81c is used to store computer programs, and can be configured to store other various data to support operations on the task scheduling device. Examples of such data include instructions for any application or method to operate on the task scheduling device, task lists, messages, pictures, videos, etc.

The processor 82c is coupled with the memory 81c, and is configured to execute the computer program in the memory 81c to obtain the tasks to be scheduled and the performance requirements of the tasks to be scheduled; from at least one resource device, select the core device that meets the performance requirements and The resource device whose value of the power consumption parameter meets the set power consumption requirement; schedule the task to be scheduled to the resource device that meets the performance requirement and the value of the kernel power consumption parameter meets the set power consumption requirement; wherein, the kernel power consumption parameter The value refers to a value combination of at least one kernel parameter related to the power management mechanism of the kernel mode supported by the resource device.

Optionally, the resource device may be a server or a server cluster.

In an optional embodiment, when the processor 82c selects a resource device that meets the performance requirements and the core state power consumption parameter value meets the set power consumption requirement, it is specifically configured to: The performance data and power consumption data under the consumption parameter value are selected to select a resource device that meets the performance requirement and the core state power consumption parameter value meets the set power consumption requirement.

Further optionally, the processor 82c is specifically configured to: select a candidate resource device that meets the performance requirements according to the performance data of at least one resource device under the respective core state power consumption parameter value; according to the candidate resource device's performance in the respective core state For the power consumption data under the consumption parameter value, a resource device whose power consumption data meets the set power consumption requirement is selected from the candidate resource devices.

Furthermore, the processor 82c is specifically configured to select the resource device with the lowest power consumption data from the candidate resource devices according to the power consumption data of the candidate resource devices under their respective core state power consumption parameter values.

Further, as shown in FIG. 8c, the task scheduling device further includes: a communication component 83c, a display 84c, a power supply component 85c, an audio component 86c and other components. Only some components are schematically shown in FIG. 8c, which does not mean that the task scheduling device only includes the components shown in FIG. 8c. In addition, according to the different implementation forms of the task scheduling device, the components in the dashed box in FIG. 8c are optional components, not mandatory components. For example, when the task scheduling device is implemented as a terminal device such as a smart phone, a tablet computer, or a desktop computer, it can include the components in the dashed box in Figure 8c; when the task scheduling device is implemented as a conventional server, cloud server, data center, or server array, etc. When the server device is used, the components in the dashed box in Figure 8c may not be included.

Correspondingly, an embodiment of the present application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor causes the processor to implement the steps in the method embodiment shown in FIG. 7d.

The communication components in Figs. 8a to 8c are configured to facilitate wired or wireless communication between the device where the communication component is located and other devices. The device where the communication component is located can access wireless networks based on communication standards, such as WiFi, 2G, 3G, 4G/LTE, 5G and other mobile communication networks, or a combination of them. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component may further include a near field communication (NFC) module, radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra wideband (UWB) technology, Bluetooth (BT) technology Wait.

The above-mentioned display in Figs. 8a-8c includes a screen, and the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, sliding, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure related to the touch or slide operation.

The power supply components in Figures 8a to 8c above provide power for various components of the equipment where the power supply components are located. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the device where the power supply component is located.

The audio components in Figs. 8a to 8c may be configured to output and/or input audio signals. For example, the audio component includes a microphone (MIC). When the device where the audio component is located is in an operating mode, such as call mode, recording mode, and voice recognition mode, the microphone is configured to receive external audio signals. The received audio signal can be further stored in a memory or sent via a communication component. In some embodiments, the audio component further includes a speaker for outputting audio signals.

Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

The present invention is described with reference to flowcharts and/or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and/or block in the flowchart and/or block diagram, and the combination of processes and/or blocks in the flowchart and/or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing equipment to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing equipment are generated It is a device that realizes the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing equipment to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction device. The device implements the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

These computer program instructions can also be loaded on a computer or other programmable data processing equipment, so that a series of operation steps are executed on the computer or other programmable equipment to produce computer-implemented processing, so as to execute on the computer or other programmable equipment. The instructions provide steps for implementing the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

In a typical configuration, the computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include non-permanent memory in a computer readable medium, random access memory (RAM) and/or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). Memory is an example of computer readable media.

Computer-readable media include permanent and non-permanent, removable and non-removable media, and information storage can be realized by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, CD-ROM, digital versatile disc (DVD) or other optical storage, Magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media can be used to store information that can be accessed by computing devices. According to the definition in this article, computer-readable media does not include transitory media, such as modulated data signals and carrier waves.

It should also be noted that the terms "include", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or equipment including a series of elements not only includes those elements, but also includes Other elements that are not explicitly listed, or they also include elements inherent to such processes, methods, commodities, or equipment. If there are no more restrictions, the element defined by the sentence "including a..." does not exclude the existence of other identical elements in the process, method, commodity, or equipment that includes the element.

The foregoing descriptions are only examples of the present application, and are not used to limit the present application. For those skilled in the art, this application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included in the scope of the claims of this application.

Claims

A data processing method, characterized in that it comprises:

From the kernel parameters of the physical device, determine at least one kernel parameter related to the power management mechanism of the kernel mode supported by the physical device;

Acquire performance data and power consumption data of the physical device under multiple value combinations corresponding to the at least one kernel parameter, where at least some of the performance data and power consumption data under the value combination are based on performance-power consumption prediction Estimated by the estimation model;

Determine the target value combination corresponding to the at least one kernel parameter according to the performance data and power consumption data of the physical device under the multiple value combinations;

The at least one kernel parameter is set according to the target value combination, so that the power consumption management mechanism operates according to the value in the target value combination.
The method according to claim 1, wherein determining the target value combination corresponding to the at least one kernel parameter according to the performance data and power consumption data of the physical device under the multiple value combinations, comprising :

Selecting at least one candidate value combination from the multiple value combinations according to the performance data and power consumption data of the physical device under the multiple value combinations;

Using a load test tool to test the performance data and power consumption data of the physical device under the at least one candidate value combination;

According to the performance data and power consumption data of the physical device under the at least one candidate value combination tested by the load test tool, the target value combination is determined from the at least one candidate value combination .
The method according to claim 2, wherein using a load test tool to test the performance data and power consumption data of the physical device under the at least one candidate value combination comprises:

For the first candidate value combination, modify the value of the at least one kernel parameter in the operating system source code to the value in the first candidate value combination, and use the modified operating system source code in the Install the operating system on the physical device;

Run the load test tool on the physical device to test the performance data and power consumption data of the physical device under the first candidate value combination; wherein, the first candidate value combination is the Any one of the at least one candidate value combination.
The method according to claim 2, characterized in that, according to the performance data and power consumption data of the physical device under the multiple value combinations, at least one candidate is selected from the multiple value combinations. Value combinations, including:

Acquiring power consumption data and/or performance data that the physical device actually needs to meet;

Matching power consumption data and/or performance data that the physical device actually needs to meet with the performance data and power consumption data of the physical device under the multiple value combinations;

From the multiple value combinations, obtain at least one candidate value combination whose matching degree with the power consumption data and/or performance data that the physical device actually needs to meet meets the matching degree requirement.
The method according to claim 4, wherein obtaining power consumption data and/or performance data that the physical device actually needs to meet includes:

According to the QoS of the service that the physical device actually needs to carry, analyze the power consumption data and/or performance data that the physical device actually needs to meet; or

Acquire the power consumption data and/or performance data required by the service deployment user as the power consumption data and/or performance data that the physical device actually needs to meet.
The method according to claim 2, wherein after determining the target value combination, the method further comprises:

The performance data and power consumption data of the physical device under the target value combination tested by the load test tool are used to modify the performance-power consumption prediction model.
The method according to claim 1, wherein the power consumption management mechanism comprises: DVFS, and the at least one kernel parameter comprises: the lowest operating frequency at which the processor can run, the highest operating frequency at which the processor can run, and At least one of the adjustment modes of the operating frequency of the processor.
7. The method according to claim 7, wherein the power consumption management mechanism further comprises: a C-state mechanism, and the at least one kernel parameter further comprises: entry time thresholds corresponding to the C mode of each level; wherein, The entry time threshold represents at least the time that the physical device needs to be maintained after entering the corresponding C mode.
The method according to claim 1, wherein the power management mechanism of the kernel state supported by the physical device comprises: a C-state mechanism, and the at least one kernel parameter comprises: entry corresponding to each level of C mode Time threshold; wherein, the entry time threshold indicates the time that the physical device needs to be maintained at least after entering the corresponding C mode.
The method according to any one of claims 1-9, wherein obtaining performance data and power consumption data of the physical device under multiple value combinations corresponding to the at least one kernel parameter comprises:

Using a load test tool to test the performance data and power consumption data of the physical device under partial value combinations among the multiple value combinations;

Performing model training according to the performance data and power consumption data of the physical device under the partial value combination to obtain a performance-power consumption prediction model;

The performance-power consumption prediction model is used to predict the performance data and power consumption data of the physical device under other value combinations among the multiple value combinations.
The method according to claim 10, wherein performing model training according to the performance data and power consumption data of the physical device under the partial value combination to obtain the performance-power consumption prediction model comprises:

Perform regression analysis on the performance data and power consumption data of the physical device under the partial value combination to obtain a performance-power consumption prediction model.
The method according to claim 11, wherein performing regression analysis on the performance data and power consumption data of the physical device under the partial value combination to obtain a performance-power consumption prediction model comprises:

The partial value combination is used as an independent variable, and the performance data and power consumption data of the physical device under the partial value combination are used as dependent variables to perform linear regression analysis to obtain a performance-power consumption prediction model.
The method according to any one of claims 1-9, wherein acquiring the performance data and power consumption data of the physical device under multiple value combinations corresponding to the at least one kernel parameter comprises:

Receive performance data and power consumption data of the physical device under multiple value combinations corresponding to the at least one kernel parameter sent by the model computing device; wherein, at least part of the performance data and power consumption data under the value combination are all values The model calculation equipment is estimated based on the performance-power consumption estimation model.
A data processing method, characterized in that it comprises:

From the kernel parameters of the physical device, determine at least one kernel parameter related to the power management mechanism of the kernel mode supported by the physical device;

Using a load test tool to test the performance data and power consumption data of the physical device under a partial value combination corresponding to the at least one kernel parameter;

Performing model training according to the performance data and power consumption data of the physical device under the partial value combination to obtain a performance-power consumption prediction model;

The performance-power consumption prediction model is used to predict the performance data and power consumption data of the physical device under other value combinations corresponding to the at least one core parameter.
The method according to claim 14, wherein performing model training according to the performance data and power consumption data of the physical device under the partial value combination to obtain the performance-power consumption prediction model comprises:

Perform regression analysis on the performance data and power consumption data of the physical device under the partial value combination to obtain a performance-power consumption prediction model.
The method according to claim 15, wherein performing regression analysis on the performance data and power consumption data of the physical device under the partial value combination to obtain a performance-power consumption prediction model comprises:

The partial value combination is used as an independent variable, and the performance data and power consumption data of the physical device under the partial value combination are used as dependent variables to perform linear regression analysis to obtain a performance-power consumption prediction model.
A data processing method, characterized in that it comprises:

From the kernel parameters of the physical device, determine at least one kernel parameter related to the performance of the device;

Acquiring performance data of the physical device under multiple value combinations corresponding to the at least one kernel parameter, where at least part of the performance data under the value combination is estimated based on a performance prediction model;

Determine the target value combination corresponding to the at least one kernel parameter according to the performance data of the physical device under the multiple value combinations;

The at least one kernel parameter is set according to the target value combination, so that the physical device operates according to the value in the target value combination.
A data processing method, characterized in that it comprises:

From the kernel parameters of the physical device, determine at least one kernel parameter related to the performance of the device;

Using a load test tool to test the performance data of the physical device under the partial value combination corresponding to the at least one kernel parameter;

Performing model training according to the performance data of the physical device under the partial value combination to obtain a performance prediction model;

The performance prediction model is used to predict the performance data of the physical device under other value combinations corresponding to the at least one kernel parameter.
A data processing method, characterized in that it comprises:

From the kernel parameters of the physical device, determine at least one kernel parameter related to the power consumption of the device;

Acquiring power data of the physical device under multiple value combinations corresponding to the at least one kernel parameter, where at least part of the power data under the value combination is estimated based on a power estimation model;

Determine the target value combination corresponding to the at least one kernel parameter according to the power data of the physical device under the multiple value combinations;

The at least one kernel parameter is set according to the target value combination, so that the physical device operates according to the value in the target value combination.
A data processing method, characterized in that it comprises:

From the kernel parameters of the physical device, determine at least one kernel parameter related to the power consumption of the device;

Using a load test tool to test the power data of the physical device under the partial value combination corresponding to the at least one kernel parameter;

Performing model training according to the power data of the physical device under the partial value combination to obtain a power estimation model;

Using the power estimation model to predict the power data of the physical device under other value combinations corresponding to the at least one kernel parameter.
A data processing method, characterized in that it comprises:

From the kernel parameters of the physical device, determine at least one kernel parameter related to the power management mechanism of the kernel mode supported by the physical device;

Using a load test tool to test the performance data and power consumption data of the physical device under a partial value combination corresponding to the at least one kernel parameter;

Performing model training according to the performance data and power consumption data of the physical device under the partial value combination to obtain a performance-power consumption prediction model;

The performance-power consumption estimation model is used to estimate the performance data and power consumption data of the physical device under multiple value combinations corresponding to the at least one core parameter; wherein, the multiple value combinations include all Combinations of the values mentioned above.
A device management system, characterized by comprising: at least one physical device and at least one model computing device; wherein the at least one physical device respectively supports a kernel-mode power management mechanism;

The at least one model computing device is configured to determine from the kernel parameters of the target device at least one kernel parameter related to the power management mechanism of the kernel state supported by the target device, and obtain the target based on the artificial intelligence model Performance data and power consumption data of the device under multiple value combinations corresponding to the at least one kernel parameter; the target device is any one of the at least one physical device;

The target device is configured to determine that the at least one kernel parameter corresponds to the performance data and power consumption data of the target device under multiple value combinations corresponding to the at least one kernel parameter obtained by the model calculation device The target value combination of, and the at least one kernel parameter is set according to the target value combination.
The system according to claim 19, wherein the model calculation device is specifically used for:

Using a load test tool to test the performance data and power consumption data of the target device under a partial value combination corresponding to the at least one kernel parameter;

Perform model training according to the performance data and power consumption data of the physical device under the partial value combination to obtain a performance-power consumption prediction model; and

The performance-power consumption estimation model is used to estimate the performance data and power consumption data of the physical device under other combinations of values corresponding to the at least one core parameter, so as to obtain that the target device is in the physical device Performance data and power consumption data under the multiple value combinations.
A data center system is characterized by comprising: a model computing device and at least one computer room, the at least one computer room includes at least one physical device, and the at least one physical device respectively supports a core-mode power management mechanism;

The model computing device is used to determine from the kernel parameters of the target device at least one kernel parameter related to the power management mechanism of the kernel state supported by the target device, and obtain the target device's location based on the artificial intelligence model. Performance data and power consumption data under multiple value combinations corresponding to the at least one kernel parameter; the target device is any one of the at least one physical device;

The target device is configured to determine that the at least one kernel parameter corresponds to the performance data and power consumption data of the target device under multiple value combinations corresponding to the at least one kernel parameter obtained by the model calculation device The target value combination of, and the at least one kernel parameter is set according to the target value combination.
A physical device, characterized by comprising: a memory and a processor;

The memory is used to store a computer program;

The processor, coupled with the memory, is configured to execute the computer program for:

From the kernel parameters of the physical device, determine at least one kernel parameter related to the power management mechanism of the kernel mode supported by the physical device;

Acquire performance data and power consumption data of the physical device under multiple value combinations corresponding to the at least one kernel parameter, where at least some of the performance data and power consumption data under the value combination are based on performance-power consumption prediction Estimated by the estimation model;

Determine the target value combination corresponding to the at least one kernel parameter according to the performance data and power consumption data of the physical device under the multiple value combinations;

The at least one kernel parameter is set according to the target value combination, so that the power consumption management mechanism operates according to the value in the target value combination.
A model computing device, which is characterized by comprising: a memory and a processor;

The memory is used to store a computer program;

The processor, coupled with the memory, is configured to execute the computer program for:

From the kernel parameters of the physical device, determine at least one kernel parameter related to the power management mechanism of the kernel mode supported by the physical device;

Using a load test tool to test the performance data and power consumption data of the physical device under a partial value combination corresponding to the at least one kernel parameter;

Performing model training according to the performance data and power consumption data of the physical device under the partial value combination to obtain a performance-power consumption prediction model;

The performance-power consumption prediction model is used to predict the performance data and power consumption data of the physical device under other value combinations corresponding to the at least one core parameter.
A computer storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor is caused to implement the steps in the method of any one of claims 1-21.
A task scheduling method, characterized in that it comprises:

Acquiring the task to be scheduled and the performance requirements of the task to be scheduled;

From at least one resource device, select a resource device that meets the performance requirements and the value of the core state power consumption parameter meets the set power consumption requirement;

Scheduling the to-be-scheduled task to a resource device that meets the performance requirement and the kernel-mode power consumption parameter value meets the set power consumption requirement;

Wherein, the value of the kernel mode power consumption parameter refers to a value combination of at least one kernel parameter related to the kernel mode power consumption management mechanism supported by the resource device.
The method according to claim 28, characterized in that, from at least one resource device, selecting a resource device that meets the performance requirement and that the value of the core state power consumption parameter meets the set power consumption requirement comprises:

According to the performance data and power consumption data of the at least one resource device under the respective core-state power consumption parameter values, a resource device that meets the performance requirements and the core-mode power consumption parameter values meets the set power consumption requirements is selected.
The method according to claim 29, characterized in that, according to the performance data and power consumption data of the at least one resource device under the respective core-state power consumption parameter values, select the core-mode power consumption parameter values that meet the performance requirements Resource equipment that meets the set power consumption requirements, including:

Selecting a candidate resource device that meets the performance requirement according to the performance data of the at least one resource device under the respective core state power consumption parameter value;

According to the power consumption data of the candidate resource devices under the respective core state power consumption parameter values, resource devices whose power consumption data meets the set power consumption requirements are selected from the candidate resource devices.
The method according to claim 30, characterized in that, according to the power consumption data of the candidate resource devices under the respective core state power consumption parameter values, power consumption data is selected from the candidate resource devices to meet the set power consumption requirements The resource equipment, including:

According to the power consumption data of the candidate resource devices under the respective core state power consumption parameter values, the resource device with the lowest power consumption data is selected from the candidate resource devices.
A task scheduling device, which is characterized by comprising: a memory and a processor;

The memory is used to store a computer program;

The processor, coupled with the memory, is configured to execute the computer program for:

Acquiring the task to be scheduled and the performance requirements of the task to be scheduled;

From at least one resource device, select a resource device that meets the performance requirements and the value of the core state power consumption parameter meets the set power consumption requirement;

Scheduling the to-be-scheduled task to a resource device that meets the performance requirement and the core state power consumption parameter value meets the set power consumption requirement;

Wherein, the value of the kernel mode power consumption parameter refers to a value combination of at least one kernel parameter related to the kernel mode power consumption management mechanism supported by the resource device.
A computer storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor is caused to implement the steps in the method of any one of claims 28-31.