WO2021128084A1 - Data processing, acquisition, model training and power consumption control methods, system and device - Google Patents

Abstract

Data processing, acquisition, model training and power consumption control methods, a system and a device. The data processing method comprises: acquiring measurement information related to a first device (201) (101); and according to the measurement information, determining a processing unit frequency of the first device (201) (102). Compared to the existing processing unit frequency read from a BMC, the processing unit frequency obtained by means of said solution has higher accuracy, thereby helping to improve the power consumption management capability of devices.

Description

Data processing, acquisition, model training and power consumption control method, system and equipment Technical field

This application relates to the field of computer technology, and in particular to a method, system, and equipment for data processing, acquisition, model training, and power consumption control.

Background technique

Accurate power management capabilities are the key to improving system reliability and reducing costs. When the server is under low load, there should be a mechanism to reduce unnecessary power consumption. When power consumption exceeds a safe level, there should be ways to limit it. To manage and control power consumption, it is vital to accurately perceive the server's processing unit, such as the CPU (Central Processing Unit, central processing unit) frequency status.

However, in practical applications, the processing unit frequency cannot be accurately detected.

Summary of the invention

The embodiments of the present application provide a data processing, acquisition, model training, and power consumption control method, system, and equipment to solve or improve the problems in the prior art.

In an embodiment of the present application, a data processing method is provided. The method includes:

Acquiring measurement information related to the first device;

According to the measurement information, the frequency of the processing unit of the first device is determined.

In another embodiment of the present application, a data processing system is provided. The data processing system includes:

The first device is used to generate measurement information during work;

The first management device is configured to obtain measurement information related to the first device, and determine the frequency of the processing unit of the first device according to the measurement information.

In yet another embodiment of the present application, a data processing method is provided. The method includes:

Acquiring measurement information related to the first device;

Acquiring an inference model that uses training samples to complete training, where the training samples include multiple sample pairs, and the sample pairs include measurement information and processing unit frequencies;

The measurement information is used as an input of an inference model, and the inference model is executed to obtain the processing unit frequency of the first device.

In yet another embodiment of the present application, a data acquisition method is provided. The method includes:

Increase the test load for the second device for testing;

Acquiring a processing unit frequency of the second device under a test load and measurement information related to the second device;

The processing unit frequency and the measurement information are used as a sample pair in the training samples used to train the inference model to be trained.

In yet another embodiment of the present application, a model training method is provided. The model training methods include:

Acquiring training samples, the training samples including multiple sample pairs, the sample pairs including test information and processing unit frequencies corresponding to the test information;

Training the inference model to be trained based on the multiple samples;

Wherein, the trained inference model is used to determine the frequency of the processing unit of the first device according to the measurement information with the first device.

In yet another embodiment of the present application, a test system is provided. The test system includes:

The second device for testing, including the same hardware structure and performance as the first device, is used to run the test program to load the corresponding test load;

The second management device for testing is connected to the second device, and is used to obtain the processing unit frequency of the second device under the test load and the measurement information related to the second device; and the frequency of the processing unit And the measurement information is used as a sample pair in the training samples used to train the inference model to be trained.

In yet another embodiment of the present application, a device power consumption control method is provided. The method includes:

Acquiring measurement information related to the first device;

Determine the frequency of the processing unit of the first device according to the measurement information;

Resetting the first power consumption cap value based on the frequency of the processing unit;

The power consumption capping control for the first device is executed by using the first power consumption capping value after resetting.

In yet another embodiment of the present application, an electronic device is provided. The electronic device includes: a memory and a processor, among which,

The memory is used to store programs;

The processor is coupled with the memory, and is configured to execute the program stored in the memory for:

Acquiring measurement information related to the first device;

According to the measurement information, the frequency of the processing unit of the first device is determined.

In yet another embodiment of the present application, an electronic device is provided. The electronic device includes: a memory and a processor, among which,

The memory is used to store programs;

The processor is coupled with the memory, and is configured to execute the program stored in the memory for:

Acquiring measurement information related to the first device;

Acquiring an inference model that uses training samples to complete training, where the training samples include multiple sample pairs, and the sample pairs include measurement information and processing unit frequencies;

The measurement information is used as an input of an inference model, and the inference model is executed to obtain the processing unit frequency of the first device.

In yet another embodiment of the present application, an electronic device is provided. The electronic device includes: a memory and a processor, among which,

The memory is used to store programs;

The processor is coupled with the memory, and is configured to execute the program stored in the memory for:

Increase the test load for the second device for testing;

Acquiring a processing unit frequency of the second device under a test load and measurement information related to the second device;

The processing unit frequency and the measurement information are used as a sample pair in the training samples used to train the inference model to be trained.

In yet another embodiment of the present application, an electronic device is provided. The electronic device includes: a memory and a processor, among which,

The memory is used to store programs;

The processor is coupled with the memory, and is configured to execute the program stored in the memory for:

Acquiring training samples, the training samples including multiple sample pairs, the sample pairs including test information and processing unit frequencies corresponding to the test information;

Training the inference model to be trained based on the multiple samples;

Wherein, the trained inference model is used to determine the frequency of the processing unit of the first device according to the measurement information with the first device.

In yet another embodiment of the present application, an electronic device is provided. The electronic device includes: a memory and a processor, among which,

The memory is used to store programs;

The processor is coupled with the memory, and is configured to execute the program stored in the memory for:

Acquiring measurement information related to the first device;

Determine the frequency of the processing unit of the first device according to the measurement information;

Resetting the first power consumption cap value based on the frequency of the processing unit;

The power consumption capping control for the first device is executed by using the first power consumption capping value after resetting.

The inventor who implements the technical solutions provided by the embodiments of the present application discovered through a lot of creative work that the power of the processing unit of the device is related to measurement information, where the measurement information is measurable information related to the device. What needs to be added here is: the processing unit power mentioned in this article, that is, the clock frequency of the processing unit, is simply the operating frequency of the processing unit during operation. Therefore, in the technical solution provided by an embodiment of the present application, measurement information related to the first device is acquired; and the processing unit frequency of the first device is determined based on the measurement information. Practice has proved that the processing unit frequency obtained by using the technical solution provided in this embodiment is more accurate than the existing processing unit frequency read from the BMC (baseboard management controller); Helps improve the power management capabilities of the device.

In the technical solution provided by another embodiment of the present application, the self-learning ability of the training model is used to self-learn the correlation between the measurement information and the frequency of the processing unit; and then the inference model completed by the training is used to combine the measurement information related to the first device As an input of the inference model, the processing unit frequency of the first device is obtained by executing the inference model. Compared with the existing processing unit frequency read from the BMC, the processing unit frequency obtained by the technical solution provided in this embodiment has a higher accuracy rate.

In the technical solution provided by another embodiment of the present application, the frequency of the processing unit of the second device under different load conditions and related measurement information is simulated by adding a load to the second device for testing, and the frequency of the processing unit and the measurement information are taken as The training samples of the inferred model can learn more accurately the relationship between the measurement information and the frequency of the processing unit, thereby improving the calculation accuracy of the frequency of the processing unit.

In the technical solution provided by another embodiment of the present application, the measurement information related to the first device is used to determine the processing unit frequency of the first device, and then based on the determined processing unit frequency, the power consumption capping control is reset. The reference first power consumption cap value; due to the high accuracy of the processing unit frequency, the accurate processing unit frequency is used to reset the first power consumption cap value, which is closer to the actual situation of the first device; it can be seen that the embodiment of the application is adopted The technical solutions provided will help improve the power consumption management capabilities of the equipment.

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 flowchart of a data processing method provided by an embodiment of this application;

FIG. 2 is a schematic structural diagram of a data processing system provided by an embodiment of this application;

FIG. 3 is a schematic flowchart of a data processing method provided by another embodiment of this application;

FIG. 4 is a schematic flowchart of a data acquisition method provided by an embodiment of the application;

FIG. 5 is a schematic flowchart of a model training method provided by an embodiment of this application;

FIG. 6 is a schematic structural diagram of a data processing system provided by another embodiment of this application;

FIG. 7 is a schematic diagram of a training sample generation process provided by an embodiment of the application;

FIG. 8 is a schematic diagram of an inference model training process provided by an embodiment of the application;

9 is a schematic diagram of a real-time processing unit frequency inference process when using an inference model to perform power capping control on a server according to an embodiment of the application;

FIG. 10 is a schematic flowchart of a method for controlling power consumption of a device according to an embodiment of the application;

FIG. 11 is a schematic structural diagram of a data processing device provided by an embodiment of the application;

FIG. 12 is a schematic structural diagram of a data processing device provided by another embodiment of this application;

FIG. 13 is a schematic structural diagram of a data acquisition device provided by an embodiment of this application;

FIG. 14 is a schematic structural diagram of a model training device provided by an embodiment of the application;

15 is a schematic structural diagram of a device power consumption control apparatus provided by an embodiment of the application;

FIG. 16 is a schematic structural diagram of an electronic device provided by an embodiment of the application.

Detailed ways

The power consumption management in the prior art includes three main parts: setting a capping value, monitoring operating power consumption, and performing power capping actions. That is, first set the power consumption cap value of each rack server according to the power distribution requirements of the cabinet, the actual power consumption of the normal operation of the server, business pressure requirements, etc., and then write the cap value into the out-of-band management device as the upper limit power consumption of the server operation . During the operation of the server, the out-of-band management device monitors the power consumption of the whole machine, and if it finds that the power consumption exceeds the capping value, it performs the capping action. Among them, the out-of-band management device has limited access to the server status. For example, the out-of-band management device cannot accurately and directly obtain the frequency of the processing unit. In power management, it is crucial to accurately obtain the processing unit frequency.

Currently, there is a method for obtaining the frequency of the processing unit: reading the frequency of the processing unit from a BMC (baseboard management controller). The BMC estimates the frequency through the instruction count during the sampling time. However, a lot of practice has proved that this estimate is quite rough. It has a reliable estimate only when all the processing unit cores are fully used. When the processing unit is not running at full capacity, the error can be ridiculously large.

To this end, each embodiment of the present application provides a solution to obtain a processing unit frequency with a higher accuracy rate, so as to perform power consumption management more accurately.

In order to enable those skilled in the art to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the accompanying drawings in the embodiments of the present application.

Some processes described in the specification, claims, and the above-mentioned drawings of the present application include multiple operations appearing in a specific order, and 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 101, 102, 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, first devices, modules, etc., and do not represent a sequence, nor do they limit "first" and "second". Two" are different types. In addition, the following 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 skilled in the art without creative work shall fall within the protection scope of this application.

Fig. 1 shows a schematic flowchart of a data processing method provided by an embodiment of the present application. As shown in Figure 1, the data processing method includes:

101. Acquire measurement information related to the first device.

102. Determine the frequency of the processing unit of the first device according to the measurement information.

In the above 101, the measurement information related to the first device can be understood as: all measurable information in the working process of the first device, including but not limited to: processing unit power consumption, processing unit temperature value, processing unit utilization Speed, motherboard temperature information, fan speed, etc.

In a specific implementation solution, the foregoing step 101 "obtain measurement information related to the first device" may include:

1011. In a case where the power consumption of the first device exceeds the first power capping threshold value (power capping threshold value), start power capping control.

Among them, the power consumption capping control can be simply understood as: monitoring the power consumption of the first device, and controlling the power consumption of the first device to not exceed the first power consumption capping value.

1012. Acquire measurement information related to the first device when the power consumption capping control is started by the first device.

In the above 102, the processing unit frequency, that is, the clock frequency of the processing unit, is simply the abbreviation of the operating frequency (the number of synchronization pulses generated in 1 second) of the processing unit during operation; it determines the operating speed of the computer.

In an achievable technical solution, the above step 102 "determine the frequency of the processing unit of the first device according to the measurement information" may include:

1021. Obtain an inference model that uses training samples to complete training, where the training samples include multiple sample pairs, and the sample pairs include measurement information and processing unit frequencies.

1022. Use the measurement information as an input of the inference model, and execute the inference model to obtain the processing unit frequency.

Among them, the "obtaining inference model" in the above step 1021 may specifically include:

10211. Obtain training samples, where the training samples include: processing unit frequencies and measurement information corresponding to the processing unit frequency samples;

10212. Use the measurement information as an input of an inference model to be trained, and execute the inference model to obtain an output result;

10213. In a case where it is determined that a convergence condition is satisfied based on the output result and the frequency of the processing unit, the inference model completes training;

10214. If it is determined based on the output result and the frequency of the processing unit that the convergence condition is not met, optimize the parameters in the inference model.

The inventor who implements the technical solutions provided by the embodiments of the present application discovered through a lot of creative work that the power of the processing unit of the device is related to measurement information, where the measurement information is measurable information related to the device. Therefore, in the technical solution provided by an embodiment of the present application, measurement information related to the first device is acquired; and the processing unit frequency of the first device is determined based on the measurement information. Practice has proved that the processing unit frequency obtained by using the technical solution provided in this embodiment is more accurate than the existing processing unit frequency read from the BMC (baseboard management controller); Helps improve the power management capabilities of the device.

Further, the method provided in this embodiment may further include the following steps:

103. In a case where the frequency of the processing unit does not meet the first preset condition, reset the first power consumption cap value.

Wherein, the first preset condition may be: whether it is within a set value range. It is assumed that the value range is between a first preset value and a second preset value, where the first preset value is smaller than the second preset value. Correspondingly, the above step 103 "resetting the first power consumption cap value when the frequency of the processing unit does not meet the first preset condition" may specifically include:

When the frequency of the processing unit is lower than the first preset value, increasing the first power consumption cap value to a second power consumption cap value;

When the frequency of the processing unit is higher than the second preset value, reducing the first power consumption cap value to a third power consumption cap value.

During specific implementation, the first preset value and the second preset value may be empirical values or values obtained through multiple experiments, etc., which are not specifically limited in this embodiment. The increase and decrease of the first power consumption cap value may be determined based on a preset reset rule. For example, the reset rule is: each adjustment is increased or decreased by a fixed value or a fixed ratio.

What needs to be explained here is: the processing unit mentioned in this article can be a general-purpose processor, such as a CPU; it can also be a dedicated processor or a heterogeneous computing unit, such as DSP (Digital Signal Processing, digital signal processor), ASIC (dedicated Integrated circuit), GPU (Graphics Processing Unit, graphics processor), FPGA (Field-Programmable Gate Array, field programmable gate array), network card acceleration chip, etc.

The method provided in the foregoing embodiment can be implemented based on the data processing system shown in FIG. 2. Specifically, as shown in Figure 2, the data processing system includes:

The first device 201 is used to generate measurement information during work;

The first management device 202 is configured to obtain measurement information related to the first device, and determine the frequency of the processing unit of the first device according to the measurement information.

Further, the first management device 202 is further configured to send a power consumption capping control instruction to the first device when the power consumption of the first device exceeds the first power consumption cap value. The first device 201 is also configured to perform a power capping operation according to the control instruction. The first management device 202 is further configured to obtain measurement information related to the first device when the power consumption capping control is activated by the first device, and determine the first device according to the measurement information The processing unit frequency.

In a specific application scenario, the first device is a server in a server cluster; the first management device is an out-of-band management device. The server includes multiple pieces of hardware, including but not limited to the following: motherboard, processing unit, power consumption, temperature sensor, and fan. The measurement information related to the first device acquired by the out-of-band management apparatus may include: power consumption of the processing unit, temperature value of the processing unit, utilization of the processing unit, temperature information of the main board, fan speed, and so on. By using the server out-of-band management device to collect and monitor the status information of the first device, the normal operation of the first device can be ensured.

In order to reduce the production cost of the out-of-band management device, a first interface may be provided in the first device, and the first interface is used to connect to the BMC. Among them, BMC can realize the collection function of part or all of the measurement information. The first interface may include, but is not limited to: a USB interface and a PCI (Peripheral Component Interconnect) slot.

The out-of-band management implemented by the out-of-band management device may include: out-of-band management of the power consumption of the device according to the out-of-band information sent by the server (that is, all measurement information related to the first device that can be measured), and/or Hardware shared by at least one server, such as a fan, is managed out-of-band.

It should be noted here that the out-of-band management device provided in this embodiment can implement other functions in addition to the above functions. For details, please refer to the description in the above method embodiment.

FIG. 3 shows a schematic flowchart of a data processing method provided by another embodiment of the present application. As shown in the figure, the data processing method includes:

301. Acquire measurement information related to a first device.

302. Obtain an inference model that uses training samples to complete training, where the training samples include multiple sample pairs, and the sample pairs include measurement information and processing unit frequencies.

303. Use the measurement information as an input of an inference model, and execute the inference model to obtain a processing unit frequency of the first device.

In a specific implementation solution, the above step 301 "obtain measurement information related to the first device" may include:

301. In the case where the power consumption of the first device exceeds the first power consumption cap value, start power consumption capping control;

302. Acquire measurement information related to the first device when the power consumption capping control is started by the first device.

The measurement information related to the first device can be understood as: all the information that can be measured during the working process of the first device, including but not limited to: processing unit power consumption, processing unit temperature value, processing unit utilization rate, main board Temperature information, fan speed, etc.

In an achievable technical solution, the above step 302 "obtain an inference model that uses training samples to complete training" may include:

3021. Use the measurement information in the sample pair as the input of the inference model to be trained, and execute the inference model to obtain an output result;

3022. In a case where it is determined that a preset end condition is met based on the output result and the frequency of the processing unit in the first sample pair, the inference model completes training;

3023. If it is determined based on the output result and the frequency of the processing unit in the sample pair that the preset end condition is not met, optimize the parameters in the inference model; and enter the next training process.

It should be noted here that the inference model to be trained in this embodiment can be a neural network model in the prior art, such as a convolutional neural network CNN, a long short-term memory network LSTM, etc., which is not specifically limited in this embodiment . In addition, the model training process can also refer to related content in the prior art.

FIG. 4 shows a schematic flowchart of a data acquisition method provided by an embodiment of the present application. The second device in this embodiment is a test device, which has the same type of processing unit and other hardware as the actual server (that is, a server that needs to be used on site and requires power consumption management), and it has a processing unit that can read accurately Frequency in-band sensor. Specifically, as shown in FIG. 4, the method includes:

401. Increase the test load for the second device for testing.

402. Obtain a processing unit frequency of the second device under a test load and measurement information related to the second device.

403. Use the processing unit frequency and the measurement information as a sample pair in the training samples used to train the inference model to be trained.

In the above 401, a test program, such as a Benchmark tool, can be loaded and run on the second device. For the content of the Benchmark tool, please refer to the prior art, which will not be repeated in this article.

In the above 402, the frequency of the processing unit can be obtained by an in-band sensor. The measurement information related to the first device can be understood as: all the information that can be measured during the working process of the first device, including but not limited to: processing unit power consumption, processing unit temperature value, processing unit utilization rate, main board Temperature information, fan speed, etc. The measurement information related to the first device is information that can be obtained by the out-of-band management device.

Further, the method provided in this embodiment may further include the following steps:

404. Set a first power consumption cap value for the second device.

Correspondingly, the foregoing step 402 "obtaining the frequency of the processing unit of the second device under test load and measurement information related to the second device" includes:

4021. When the test load increases until the power consumption of the second device reaches the first power consumption cap value, start power consumption capping control;

4022. When the second device starts power consumption capping control, acquire the processing unit frequency of the second device and measurement information related to the second device.

The "acquiring the frequency of the processing unit of the second device and the measurement information related to the second device" in the foregoing 4022 may specifically be:

After starting the power consumption capping control, continue to increase the test load for the second device until the preset maximum load is reached;

Record the processing frequency and measurement information of the second device during the test;

Wherein, the test period is a period of time from when the power consumption capping control is activated to when the test load of the second device is loaded to a preset maximum load; or from when the power consumption capping control is activated to the first The test load of the second device is loaded to the preset maximum load and continues for a period of time after the set duration.

Specifically, the aforementioned continuous setting duration may be an empirical value, which is not specifically limited in this embodiment.

Further, the method provided in this embodiment may further include the following steps:

405. After increasing the test load for the second device to a preset maximum load, reset the first power consumption cap value; and enter the next round of testing until the reset first power consumption cap value is greater than or Equal to the threshold; or

405'. After increasing the load for the second device to a preset maximum load and continuing for a set duration, reset the first power consumption cap value; and enter the next round of testing until the reset of the first power The consumption cap value is greater than or equal to the threshold.

Wherein, the next round of testing returns to the above step 401 to repeat the above process again on the basis of resetting the first power consumption cap value to obtain a new sample pair. The foregoing threshold may be equal to or greater than the rated power consumption of the second device. Each round of resetting can increase a certain step length, this step length can be a fixed value, or it can be changed appropriately. A typical way is, for example, the rated power consumption of the second device (such as a server) is 500W, in the first round of testing, the first power consumption cap value is set to 350W; in the second round of testing, set the first power consumption The cap value is reset to 360W;..., in the Nth round of testing, the first cap value of power consumption is reset to 520W. Wait, the last threshold can be a little higher than the device's rated power consumption of 500W, because the rated power consumption does not represent the maximum power consumption of the machine in actual operation.

That is, the "resetting the first power consumption cap value" in the above steps 405 and 405' may specifically be:

On the basis of the first power consumption cap value, increase a set value to obtain a fourth power consumption cap value;

The first power consumption cap value is updated to the fourth power consumption cap value.

FIG. 5 shows a schematic flowchart of a model training method provided by an embodiment of the present application. As shown in Figure 5, the model training method includes:

501. Obtain a training sample, where the training sample includes a plurality of sample pairs, and the sample pairs include test information and a processing unit frequency corresponding to the test information.

502. Based on the multiple samples, train the inference model to be trained.

Wherein, the trained inference model is used to determine the frequency of the processing unit of the first device according to the measurement information with the first device.

The data acquisition method provided by the foregoing embodiment can be implemented based on the test system shown in FIG. 6. Specifically, as shown in Figure 6, the test system includes:

The second test device 601 includes the same hardware structure and performance as the first device, and is used to run the test program to load the corresponding test load;

The second management device 602 for testing is connected to the second device, and is used to obtain the frequency of the processing unit of the second device under the test load and the measurement information related to the second device; The frequency and the measurement information are used as a sample pair in the training samples used to train the inference model to be trained.

Wherein, the second management device for testing has the same hardware structure and functions as the first management device. In addition to this, it also has a function that the first management device does not have, that is, the frequency of the processing unit of the second device is acquired. In specific implementation, the frequency of the processing unit of the second device can be obtained through an in-band sensor.

Further, the second management device 602 is further configured to set a first power consumption cap value for the second device; the test load is increased until the power consumption of the second device reaches the first power consumption cap value When, sending a power consumption capping control instruction to the second device;

The second device 601 is also configured to perform a power consumption capping operation according to the control instruction;

The second management device 602 is further configured to obtain the processing unit frequency of the second device and the measurement information related to the second device when the power consumption capping control is activated by the second device; The processing unit frequency and the measurement information are used as a sample pair in the training samples used to train the inference model to be trained.

Further, the test system may further include:

The model training device is used to obtain training samples, the training samples include multiple sample pairs, the sample pairs include test information and processing unit frequencies corresponding to the test information; based on the multiple samples, the inference model to be trained is trained for the first A management device provides an inference model for completing the training.

It should be noted here that the second management device provided in this embodiment may implement other functions in addition to the above functions. For details, please refer to the description in the foregoing method embodiment.

The three major contents involved in each embodiment of the present application will be described below in conjunction with FIG. 7, FIG. 8, and FIG. 9.

The first part is to generate training samples.

The second part is to train the inference model.

The third part is the real-time processing unit frequency inference when using the inference model to control the power capping of the server.

As shown in Figure 7, generate training samples

To build a test system, select a test server with the same hardware structure and performance as the server in the actual application scenario and a test out-of-band management device with the same hardware structure and function as the out-of-band management device in the actual application scenario. Use the Benchmark tool to increase the load on the test server. The processing unit is the CPU.

Specifically, as shown in Figure 7, it includes:

S11. Set a first power consumption cap value for the test server.

S12. Run the Benchmark test program to gradually load the test server.

S13. When the power consumption of the test server reaches the first power consumption cap value, the power consumption cap control is enabled.

S14. Through sampling, record all related measurement information of the test server after the power consumption capping control is started, and the CPU frequency of the test server.

S15. After the load of the test server is loaded to the maximum load of Benchmark, reset the first power consumption cap value, and return to S11, until the reset first power consumption cap value is greater than or equal to the set threshold.

Among them, the set threshold is equal to the rated power consumption of the test server or 120% of the rated power consumption. In addition, after the Benchmark is loaded to the maximum load, it can continue for a period of time to allow enough time to record multiple pairs of samples when the power consumption cap takes effect.

As shown in Figure 8, the training inference model

S21. Data processing.

Perform data cleaning on the data collected by the method shown in Figure 7 above, for example, remove some incomplete samples and remove abnormal samples; integrate some samples into a feature, or derive some features from the sample . For example, for a processing unit with multiple cores, the frequency of each core should be integrated into a value for each processing unit.

It should be noted here that: the data processing process can refer to the corresponding content in the prior art, which is not specifically limited in this embodiment.

S22. Model training.

A machine learning model or any regression equation (ie, the above-mentioned inference model to be trained) is used to establish the relationship between the frequency of the processing unit and the measurement information. The training samples processed by the above S21 are used to train the inference model to be trained. Among them, the training process can refer to the related content of the prior art, which is not specifically limited here.

As shown in Figure 9, the real-time processing unit frequency is inferred when the inference model is used to control the power capping of the server.

S31. When the power consumption of the server reaches the power consumption cap value or a power consumption cap control instruction is received, start the power consumption cap control.

S32. Obtain measurement information of the server.

Among them, the measurement information includes at least: processing unit power, processing unit temperature value, processing unit usage rate, temperature information of the main board, fan speed, and so on.

S33. Input the measurement information into the inference model, and execute the inference model to obtain the CPU frequency.

The inferred processing unit frequency can be used as a basis for adjusting the existing power consumption cap value. For example, if the inferred frequency of the processing unit is too high, the existing power consumption cap value can be lowered; if the inferred frequency of the processing unit is too low, the existing power consumption cap value can be adjusted higher.

FIG. 10 shows a schematic flowchart of a method for controlling power consumption of a device according to an embodiment of the present application. As shown in FIG. 10, the device power consumption control method includes:

701. Acquire measurement information related to the first device.

702. Determine the frequency of the processing unit of the first device according to the measurement information.

703. Reset the first power consumption cap value based on the frequency of the processing unit.

704. Use the reset first power consumption cap value to perform power consumption cap control for the first device.

For the above steps 701 and 702, please refer to the corresponding content in the above embodiments, which will not be repeated here.

The above step 703 "resetting the first power consumption cap value based on the frequency of the processing unit" may specifically include:

When the frequency of the processing unit is lower than the first preset value, increasing the first power consumption cap value to a second power consumption cap value;

In a case where the frequency of the processing unit is higher than the second preset value, reducing the first power consumption cap value to a third power consumption cap value;

Wherein, the first preset value is less than the second preset value.

The above-mentioned first preset value and second preset value may be preset values, which may be empirical values, or calculated through corresponding algorithms, or obtained through multiple experiments, and so on. The increase and decrease of the first power consumption cap value can be determined based on a preset reset rule. For example, the reset rule is: each adjustment is increased or decreased by a fixed value or a fixed ratio.

In addition, the method provided in the embodiment of the present application, that is, the execution premise of each of the foregoing steps 701 may be: when the power consumption capping control for the first device is started, an action of triggering the measurement information related to the first device.

The above step 702 "determine the frequency of the processing unit of the first device according to the measurement information" may include:

Acquiring an inference model that uses training samples to complete training, where the training samples include multiple sample pairs, and the sample pairs include measurement information and processing unit frequencies;

The measurement information is used as the input of the inference model, and the inference model is executed to obtain the processing unit frequency.

Further, the measurement information may include but is not limited to at least one of the following: processing unit power consumption, processing unit temperature value, and processing unit utilization rate.

In the technical solution provided in this embodiment, the measurement information related to the first device is used to determine the frequency of the processing unit of the first device, and then based on the determined frequency of the processing unit, the first reference to be referenced during power capping control is reset. 1. Power consumption cap value; due to the high accuracy of the processing unit frequency, the first power consumption cap value is reset with an accurate processing unit frequency, which is closer to the actual situation of the first device; it can be seen that the technology provided by the embodiment of this application is used The solution helps to improve the power management capability of the device.

FIG. 11 shows a schematic structural diagram of a data processing device provided by an embodiment of the present application. As shown in FIG. 11, the data processing device includes: an acquisition module 11 and a determination module 12. Wherein, the obtaining module 11 is used to obtain measurement information related to the first device; the determining module 12 is used to determine the frequency of the processing unit of the first device according to the measurement information.

Practice has proved that the processing unit frequency obtained by using the technical solution provided in this embodiment is more accurate than the existing processing unit frequency read from the BMC (baseboard management controller); Helps improve the power management capabilities of the device.

Further, the acquisition module 11 is also used for:

In the case where the power consumption of the first device exceeds the first power consumption capping value, start the power consumption capping control;

When the first device starts power consumption capping control, obtain measurement information related to the first device.

Further, the data processing device provided in this embodiment should include a reset module. The reset module is configured to reset the first power consumption cap value when the frequency of the processing unit does not meet the first preset condition.

Further, the reset module is also used for:

When the frequency of the processing unit is lower than the first preset value, increasing the first power consumption cap value to a second power consumption cap value;

In a case where the frequency of the processing unit is higher than the second preset value, reducing the first power consumption cap value to a third power consumption cap value;

Wherein, the first preset value is smaller than the second preset value.

Further, the determining module 12 is also used for:

Acquiring an inference model that uses training samples to complete training, where the training samples include multiple sample pairs, and the sample pairs include measurement information and processing unit frequencies;

The measurement information is used as the input of the inference model, and the inference model is executed to obtain the processing unit frequency.

Further, the acquisition module 11 is also used for:

Acquiring training samples, where the training samples include: processing unit frequencies and measurement information corresponding to the processing unit frequency samples;

Taking the measurement information as the input of the inference model to be trained, and executing the inference model to obtain an output result;

If it is determined based on the output result and the frequency of the processing unit that the convergence condition is satisfied, the inference model completes the training;

If it is determined based on the output result and the frequency of the processing unit that the convergence condition is not satisfied, the parameters in the inference model are optimized.

Further, the measurement information includes at least one of the following: processing unit power consumption, processing unit temperature value, and processing unit utilization rate.

What needs to be explained here is that the data processing device provided in the foregoing embodiment can implement the technical solutions described in the foregoing method embodiments. For the specific implementation principles of the foregoing modules or units, please refer to the corresponding content in the foregoing method embodiments. Here No longer.

FIG. 12 shows a schematic structural diagram of a data processing device provided by another embodiment of the present application. As shown in FIG. 12, the data processing device includes: an acquisition module 21 and an inference module 22. Wherein, the acquisition module 21 is used to acquire measurement information related to the first device; and to acquire an inference model that uses training samples to complete training, wherein the training samples include multiple sample pairs, and the sample pairs include measurement information and a processing unit. frequency. The inference module 22 is configured to use the measurement information as an input of an inference model, and execute the inference model to obtain the processing unit frequency of the first device.

In the technical solution provided in this embodiment, the self-learning ability of the training model is used to self-learn the association relationship between the measurement information and the frequency of the processing unit; then the trained inference model is used to use the measurement information related to the first device as the inference The input of the model, the frequency of the processing unit of the first device is obtained by executing the inferred model. Compared with the existing processing unit frequency read from the BMC, the processing unit frequency obtained by the technical solution provided in this embodiment has a higher accuracy rate.

Further, the acquisition module 21 is also used for:

In the case where the power consumption of the first device exceeds the first power consumption capping value, start the power consumption capping control;

When the first device starts power consumption capping control, obtain measurement information related to the first device.

Further, the acquisition module 21 is also used for:

Use the measurement information in the sample pair as the input of the inference model to be trained, and execute the inference model to obtain an output result;

If it is determined based on the output result and the frequency of the processing unit in the first sample pair that the preset end condition is met, the inference model completes the training;

If it is determined based on the output result and the frequency of the processing unit in the sample pair that the preset end condition is not met, the parameters in the inference model are optimized; and the next training process is entered.

What needs to be explained here is that the data processing device provided in the foregoing embodiment can implement the technical solutions described in the foregoing method embodiments. For the specific implementation principles of the foregoing modules or units, please refer to the corresponding content in the foregoing method embodiments. Here No longer.

FIG. 13 shows a schematic structural diagram of a data acquisition device provided by an embodiment of the present application. As shown in FIG. 13, the data acquisition device includes: a loading module 31 and an acquisition module 32. Wherein, the loading module is used to increase the test load for the second device for testing. The acquisition module is used to acquire the processing unit frequency of the second device under a test load and measurement information related to the second device; and use the processing unit frequency and the measurement information as the processing unit frequency and the measurement information to be used for training. A sample pair in the training sample for training the inference model.

In the technical solution provided by this embodiment, the second device for testing is added with a load to simulate the processing unit frequency of the second device under different load conditions and related measurement information, and the processing unit frequency and measurement information are used as the inference model. The training samples can learn more accurately the relationship between the measurement information and the frequency of the processing unit, thereby improving the calculation accuracy of the frequency of the processing unit.

Further, the data acquisition device provided in this embodiment may further include a setting module. The setting module is used to set a first power consumption cap value for the second device. Correspondingly, the acquisition module is also used for:

When the test load increases until the power consumption of the second device reaches the first power consumption cap value, start the power consumption cap control;

When the second device starts power consumption capping control, the processing unit frequency of the second device and the measurement information related to the second device are acquired.

Further, the acquisition module 32 is also used for:

After starting the power consumption capping control, continue to increase the test load for the second device until the preset maximum load is reached;

Record the processing frequency and measurement information of the second device during the test;

Wherein, the test period is a period of time from when the power consumption capping control is started to when the test load of the second device is loaded to a preset maximum load; or from when the power consumption capping control is started to the first The test load of the second device is loaded to the preset maximum load and continues for a period of time after the set duration.

Further, the data acquisition module provided in this embodiment may further include a reset module. Wherein, the reset module is used for:

After increasing the test load for the second device to the preset maximum load, reset the first power consumption cap value; and enter the next round of testing until the reset first power consumption cap value is greater than or equal to the threshold value ;or

After increasing the load for the second device to the preset maximum load and continuing for a set time period, reset the first power consumption cap value; and enter the next round of testing until the reset first power consumption cap value Greater than or equal to the threshold.

Further, the reset module is also used for:

On the basis of the first power consumption cap value, increase a set value to obtain a fourth power consumption cap value;

The first power consumption cap value is updated to the fourth power consumption cap value.

What needs to be explained here is that the data acquisition device provided in the foregoing embodiment can implement the technical solutions described in the foregoing method embodiments. For the specific implementation principles of the foregoing modules or units, please refer to the corresponding content in the foregoing method embodiments. No longer.

FIG. 14 shows a schematic structural diagram of a model training device provided by an embodiment of the present application. As shown in FIG. 14, the model training device includes: an acquisition module 41 and a training module 42. Wherein, the acquisition module 41 is used to acquire training samples, the training samples include multiple sample pairs, and the sample pairs include test information and processing unit frequencies corresponding to the test information. The training module 42 is configured to train the inference model to be trained based on the multiple samples; wherein the trained inference model is used to determine the frequency of the processing unit of the first device according to the measurement information with the first device .

FIG. 15 shows a schematic structural diagram of an apparatus for controlling power consumption of a device according to an embodiment of the present application. As shown in FIG. 15, the device power consumption control device includes: an acquisition module 51, a determination module 52, a reset module 53 and an execution module 54. Wherein, the obtaining module 51 is used to obtain measurement information related to the first device; the determining module 52 is used to determine the frequency of the processing unit of the first device according to the measurement information; the reset module 53 is used to obtain measurement information based on the processing unit Frequency, reset the first power consumption cap value; the execution module 54 is configured to use the reset first power consumption cap value to execute power consumption cap control for the first device.

Further, the reset module 53 is also used for:

When the frequency of the processing unit is lower than the first preset value, increasing the first power consumption cap value to a second power consumption cap value;

In a case where the frequency of the processing unit is higher than the second preset value, reducing the first power consumption cap value to a third power consumption cap value;

Wherein, the first preset value is smaller than the second preset value.

Further, the determining module 52 is also used for:

Acquiring an inference model that uses training samples to complete training, where the training samples include multiple sample pairs, and the sample pairs include measurement information and processing unit frequencies;

The measurement information is used as the input of the inference model, and the inference model is executed to obtain the processing unit frequency.

Further, the measurement information includes at least one of the following: processing unit power consumption, processing unit temperature value, and processing unit utilization rate.

FIG. 16 shows a schematic structural diagram of an electronic device provided by an embodiment of the present application. As shown in FIG. 16, the electronic device includes: a memory 61 and a processor 62, where:

The memory 61 is used to store programs;

The processor 62 is coupled with the memory 61, and is configured to execute the program stored in the memory 61 for:

Acquiring measurement information related to the first device;

According to the measurement information, the frequency of the processing unit of the first device is determined.

The aforementioned memory 61 may be configured to store other various data to support operations on the electronic device. Examples of such data include instructions for any application or method operating on the electronic device. The memory 61 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable and Programmable read only memory (EPROM), programmable read only memory (PROM), read only memory (ROM), magnetic memory, flash memory, magnetic or optical disk.

When the above-mentioned processor 62 executes the program in the memory 61, in addition to the above functions, it may also implement other functions. For details, please refer to the description of the previous embodiments.

Further, as shown in FIG. 16, the electronic device further includes: a communication component 63, a display 64, a power supply component 65, an audio component 66 and other components. Only some components are schematically shown in FIG. 16, which does not mean that the electronic device only includes the components shown in FIG. 16.

Another embodiment of the present application provides an electronic device. The structure of the electronic device is similar to the above-mentioned electronic device embodiment, which can be referred to as shown in FIG. 16 above. The electronic device includes a memory and a processor, among which,

The memory is used to store programs;

The processor is coupled with the memory, and is configured to execute the program stored in the memory for:

Acquiring measurement information related to the first device;

Acquiring an inference model that uses training samples to complete training, where the training samples include multiple sample pairs, and the sample pairs include measurement information and processing unit frequencies;

The measurement information is used as an input of an inference model, and the inference model is executed to obtain the processing unit frequency of the first device.

Wherein, when the processor executes the program in the memory, in addition to the above functions, other functions may also be implemented. For details, please refer to the description of the foregoing embodiments.

Correspondingly, an embodiment of the present application also provides a computer-readable storage medium storing a computer program, which can implement the steps or functions of the data processing method provided by the foregoing embodiments when the computer program is executed by a computer.

Another embodiment of the present application provides an electronic device. The structure of the electronic device is similar to the above-mentioned electronic device embodiment, which can be referred to as shown in FIG. 16 above. The electronic device includes a memory and a processor, among which,

The memory is used to store programs;

The processor is coupled with the memory, and is configured to execute the program stored in the memory for:

Increase the test load for the second device for testing;

Acquiring a processing unit frequency of the second device under a test load and measurement information related to the second device;

The processing unit frequency and the measurement information are used as a sample pair in the training samples used to train the inference model to be trained.

Wherein, when the processor executes the program in the memory, in addition to the above functions, other functions may also be implemented. For details, please refer to the description of the previous embodiments.

Correspondingly, an embodiment of the present application also provides a computer-readable storage medium storing a computer program, which when executed by a computer can implement the steps or functions of the data acquisition method provided in the foregoing embodiments.

Another embodiment of the present application provides an electronic device. The structure of the electronic device is similar to the above-mentioned electronic device embodiment, which can be referred to as shown in FIG. The electronic device includes a memory and a processor, among which,

The memory is used to store programs;

The processor is coupled with the memory, and is configured to execute the program stored in the memory for:

Acquiring training samples, the training samples including multiple sample pairs, the sample pairs including test information and processing unit frequencies corresponding to the test information;

Training the inference model to be trained based on the multiple samples;

Wherein, the trained inference model is used to determine the frequency of the processing unit of the first device according to the measurement information with the first device.

Wherein, when the processor executes the program in the memory, in addition to the above functions, other functions may also be implemented. For details, please refer to the description of the previous embodiments.

Correspondingly, an embodiment of the present application also provides a computer-readable storage medium storing a computer program, which can implement the steps or functions of the model training method provided in the foregoing embodiments when the computer program is executed by a computer.

Another embodiment of the present application provides an electronic device. The structure of the electronic device is similar to the above-mentioned electronic device embodiment, which can be referred to as shown in FIG. 16 above. The electronic device includes a memory and a processor, among which,

The memory is used to store programs;

The processor is coupled with the memory, and is configured to execute the program stored in the memory for:

Acquiring measurement information related to the first device;

Determine the frequency of the processing unit of the first device according to the measurement information;

Resetting the first power consumption cap value based on the frequency of the processing unit;

The power consumption capping control for the first device is executed by using the first power consumption capping value after resetting.

Wherein, when the processor executes the program in the memory, in addition to the above functions, other functions may also be implemented. For details, please refer to the description of the previous embodiments.

Correspondingly, an embodiment of the present application also provides a computer-readable storage medium storing a computer program, which when executed by a computer can implement the steps or functions of the device power consumption control method provided by the foregoing embodiments.

The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in One place, or it can be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the objectives of the solutions of the embodiments. Those of ordinary skill in the art can understand and implement it without creative work.

Through the description of the above implementation manners, those skilled in the art can clearly understand that each implementation manner can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM/RAM, magnetic A disc, an optical disc, etc., include several instructions to make a computer first device (which may be a personal computer, a server, or a network first device, etc.) execute the methods described in each embodiment or some parts of the embodiment.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the application, not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: The technical solutions recorded in the foregoing embodiments are modified, or some of the technical features are equivalently replaced; these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims (31)

1. A data processing method, characterized in that it comprises:

   Acquiring measurement information related to the first device;

   According to the measurement information, the frequency of the processing unit of the first device is determined.
2. The method according to claim 1, wherein acquiring measurement information related to the first device comprises:

   In the case where the power consumption of the first device exceeds the first power consumption capping value, start the power consumption capping control;

   When the first device starts power consumption capping control, obtain measurement information related to the first device.
3. The method according to claim 2, further comprising:

   When the frequency of the processing unit does not meet the first preset condition, reset the first power consumption cap value.
4. The method according to claim 3, wherein, when the frequency of the processing unit does not meet a first preset condition, resetting the first power consumption cap value includes:

   When the frequency of the processing unit is lower than the first preset value, increasing the first power consumption cap value to a second power consumption cap value;

   In a case where the frequency of the processing unit is higher than the second preset value, reducing the first power consumption cap value to a third power consumption cap value;

   Wherein, the first preset value is smaller than the second preset value.
5. The method according to any one of claims 1 to 4, wherein determining the frequency of the processing unit of the first device according to the measurement information comprises:

   Acquiring an inference model that uses training samples to complete training, where the training samples include multiple sample pairs, and the sample pairs include measurement information and processing unit frequencies;

   The measurement information is used as the input of the inference model, and the inference model is executed to obtain the processing unit frequency.
6. The method according to claim 5, wherein obtaining an inference model comprises:

   Acquiring training samples, where the training samples include: processing unit frequencies and measurement information corresponding to the processing unit frequency samples;

   Taking the measurement information as the input of the inference model to be trained, and executing the inference model to obtain an output result;

   If it is determined based on the output result and the frequency of the processing unit that the convergence condition is satisfied, the inference model completes the training;

   If it is determined based on the output result and the frequency of the processing unit that the convergence condition is not satisfied, the parameters in the inference model are optimized.
7. The method according to any one of claims 1 to 4, wherein the measurement information includes at least one of the following: processing unit power consumption, processing unit temperature value, and processing unit utilization rate.
8. A data processing system, characterized in that it comprises:

   The first device is used to generate measurement information during work;

   The first management device is configured to obtain measurement information related to the first device, and determine the frequency of the processing unit of the first device according to the measurement information.
9. The system according to claim 8, wherein:

   The first management device is further configured to send a power consumption capping control instruction to the first device when the power consumption of the first device exceeds a first power consumption cap value;

   The first device is further configured to perform a power consumption capping operation according to the control instruction;

   The first management apparatus is further configured to obtain measurement information related to the first device when the power consumption capping control is activated by the first device, and determine the measurement information of the first device according to the measurement information. Processing unit frequency.
10. The system according to claim 8 or 9, wherein the first device is a server in a server cluster; and the first management device is an out-of-band management device.
11. A data processing method, characterized in that it comprises:

    Acquiring measurement information related to the first device;

    Acquiring an inference model that uses training samples to complete training, where the training samples include multiple sample pairs, and the sample pairs include measurement information and processing unit frequencies;

    The measurement information is used as an input of an inference model, and the inference model is executed to obtain the processing unit frequency of the first device.
12. The method according to claim 11, wherein the measurement information related to the first device is acquired;

    In the case where the power consumption of the first device exceeds the first power consumption capping value, start the power consumption capping control;

    When the first device starts power consumption capping control, obtain measurement information related to the first device.
13. The method according to claim 11 or 12, wherein obtaining an inference model that uses training samples to complete training includes:

    Use the measurement information in the sample pair as the input of the inference model to be trained, and execute the inference model to obtain an output result;

    If it is determined based on the output result and the frequency of the processing unit in the first sample pair that the preset end condition is met, the inference model completes the training;

    If it is determined based on the output result and the frequency of the processing unit in the sample pair that the preset end condition is not met, the parameters in the inference model are optimized; and the next training process is entered.
14. A data acquisition method, characterized in that it comprises:

    Increase the test load for the second device for testing;

    Acquiring a processing unit frequency of the second device under a test load and measurement information related to the second device;

    The processing unit frequency and the measurement information are used as a sample pair in the training samples used to train the inference model to be trained.
15. The method according to claim 14, further comprising:

    Setting a first power consumption cap value for the second device;

    And, acquiring the processing unit frequency of the second device under the test load and the measurement information related to the second device includes:

    When the test load increases until the power consumption of the second device reaches the first power consumption cap value, start the power consumption cap control;

    When the second device starts power consumption capping control, the processing unit frequency of the second device and the measurement information related to the second device are acquired.
16. The method according to claim 15, wherein acquiring the processing unit frequency of the second device and measurement information related to the second device comprises:

    After starting the power consumption capping control, continue to increase the test load for the second device until the preset maximum load is reached;

    Record the processing frequency and measurement information of the second device during the test;

    Wherein, the test period is a period of time from when the power consumption capping control is activated to when the test load of the second device is loaded to a preset maximum load; or from when the power consumption capping control is activated to the first The test load of the second device is loaded to the preset maximum load and continues for a period of time after the set duration.
17. The method according to claim 15 or 16, further comprising:

    After increasing the test load for the second device to the preset maximum load, reset the first power consumption cap value; and enter the next round of testing until the reset first power consumption cap value is greater than or equal to the threshold ;or

    After increasing the load for the second device to the preset maximum load and continuing for a set time period, reset the first power consumption cap value; and enter the next round of testing until the reset first power consumption cap value Greater than or equal to the threshold.
18. The method of claim 17, wherein resetting the first power consumption cap value comprises:

    On the basis of the first power consumption cap value, increase a set value to obtain a fourth power consumption cap value;

    The first power consumption cap value is updated to the fourth power consumption cap value.
19. A model training method is characterized in that it includes:

    Acquiring training samples, the training samples including multiple sample pairs, the sample pairs including test information and processing unit frequencies corresponding to the test information;

    Training the inference model to be trained based on the multiple samples;

    Wherein, the trained inference model is used to determine the frequency of the processing unit of the first device according to the measurement information with the first device.
20. A test system including:

    The second device for testing, including the same hardware structure and performance as the first device, is used to run the test program to load the corresponding test load;

    The second management device for testing is connected to the second device, and is used to obtain the processing unit frequency of the second device under the test load and the measurement information related to the second device; and the frequency of the processing unit And the measurement information is used as a sample pair in the training samples used to train the inference model to be trained.
21. The system of claim 20, wherein:

    The second management device is further configured to set a first power consumption cap value for the second device; when the test load increases until the power consumption of the second device reaches the first power consumption cap value, the The second device sends a power consumption capping control instruction;

    The second device is further configured to perform a power consumption capping operation according to the control instruction;

    The second management device is further configured to obtain the processing unit frequency of the second device and the measurement information related to the second device when the power consumption capping control is activated by the second device; The processing unit frequency and the measurement information are used as a sample pair in the training samples used to train the inference model to be trained.
22. The system according to claim 20 or 21, further comprising:

    The model training device is used to obtain training samples, the training samples include multiple sample pairs, the sample pairs include test information and processing unit frequencies corresponding to the test information; based on the multiple samples, the inference model to be trained is trained for the first A management device provides an inference model for completing the training.
23. A device power consumption control method, characterized in that it includes:

    Acquiring measurement information related to the first device;

    Determine the frequency of the processing unit of the first device according to the measurement information;

    Resetting the first power consumption cap value based on the frequency of the processing unit;

    The power consumption capping control for the first device is executed by using the first power consumption capping value after resetting.
24. The method according to claim 23, wherein, based on the frequency of the processing unit, resetting the first power consumption cap value comprises:

    When the frequency of the processing unit is lower than the first preset value, increasing the first power consumption cap value to a second power consumption cap value;

    In a case where the frequency of the processing unit is higher than the second preset value, reducing the first power consumption cap value to a third power consumption cap value;

    Wherein, the first preset value is smaller than the second preset value.
25. The method according to claim 23, wherein determining the frequency of the processing unit of the first device according to the measurement information comprises:

    Acquiring an inference model that uses training samples to complete training, where the training samples include multiple sample pairs, and the sample pairs include measurement information and processing unit frequencies;

    The measurement information is used as the input of the inference model, and the inference model is executed to obtain the processing unit frequency.
26. The method according to any one of claims 23 to 25, wherein the measurement information includes at least one of the following: processing unit power consumption, processing unit temperature value, and processing unit utilization rate.
27. An electronic device, characterized by comprising: a memory and a processor, wherein:

    The memory is used to store programs;

    The processor is coupled with the memory, and is configured to execute the program stored in the memory for:

    Acquiring measurement information related to the first device;

    According to the measurement information, the frequency of the processing unit of the first device is determined.
28. An electronic device, characterized by comprising: a memory and a processor, wherein:

    The memory is used to store programs;

    The processor is coupled with the memory, and is configured to execute the program stored in the memory for:

    Acquiring measurement information related to the first device;

    Acquiring an inference model that uses training samples to complete training, where the training samples include multiple sample pairs, and the sample pairs include measurement information and processing unit frequencies;

    The measurement information is used as an input of an inference model, and the inference model is executed to obtain the processing unit frequency of the first device.
29. An electronic device, characterized by comprising: a memory and a processor, wherein:

    The memory is used to store programs;

    The processor is coupled with the memory, and is configured to execute the program stored in the memory for:

    Increase the test load for the second device for testing;

    Acquiring a processing unit frequency of the second device under a test load and measurement information related to the second device;

    The processing unit frequency and the measurement information are used as a sample pair in the training samples used to train the inference model to be trained.
30. An electronic device, characterized by comprising: a memory and a processor, wherein:

    The memory is used to store programs;

    The processor is coupled with the memory, and is configured to execute the program stored in the memory for:

    Acquiring training samples, the training samples including multiple sample pairs, the sample pairs including test information and processing unit frequencies corresponding to the test information;

    Training the inference model to be trained based on the multiple samples;

    Wherein, the trained inference model is used to determine the frequency of the processing unit of the first device according to the measurement information with the first device.
31. An electronic device, characterized by comprising: a memory and a processor, wherein:

    The memory is used to store programs;

    The processor is coupled with the memory, and is configured to execute the program stored in the memory for:

    Acquiring measurement information related to the first device;

    Determine the frequency of the processing unit of the first device according to the measurement information;

    Resetting the first power consumption cap value based on the frequency of the processing unit;

    The power consumption capping control for the first device is executed by using the first power consumption capping value after resetting.

Images