WO2024099474A1 - 能效评估方法、装置、系统及相关设备 - Google Patents

能效评估方法、装置、系统及相关设备 Download PDF

Info

Publication number
WO2024099474A1
WO2024099474A1 PCT/CN2023/143424 CN2023143424W WO2024099474A1 WO 2024099474 A1 WO2024099474 A1 WO 2024099474A1 CN 2023143424 W CN2023143424 W CN 2023143424W WO 2024099474 A1 WO2024099474 A1 WO 2024099474A1
Authority
WO
WIPO (PCT)
Prior art keywords
energy
power consumption
hardware object
energy efficiency
saving
Prior art date
Application number
PCT/CN2023/143424
Other languages
English (en)
French (fr)
Inventor
吴俊杰
王江涛
谢海军
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Publication of WO2024099474A1 publication Critical patent/WO2024099474A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3409—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment for performance assessment
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/3055—Monitoring arrangements for monitoring the status of the computing system or of the computing system component, e.g. monitoring if the computing system is on, off, available, not available
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/3058—Monitoring arrangements for monitoring environmental properties or parameters of the computing system or of the computing system component, e.g. monitoring of power, currents, temperature, humidity, position, vibrations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment

Definitions

  • the present application relates to the field of computer technology, and in particular to an energy efficiency evaluation method, device, system and related equipment.
  • Energy efficiency refers to the ratio between the amount of energy used to provide services and the total amount of energy consumed during energy use.
  • the energy efficiency level of hardware objects is mainly evaluated through energy efficiency evaluation tools such as SPEC SERT and BenchSEE when the hardware objects are in an offline state.
  • the evaluation process can use standard loads to stress test the hardware objects and evaluate the energy efficiency level of the hardware objects by measuring the business throughput and energy consumption of the hardware objects under the load.
  • these energy efficiency level evaluation tools cannot evaluate the energy efficiency level of hardware objects in real time, and the cost of evaluating the energy efficiency level of hardware objects is usually high.
  • the present application provides an energy efficiency evaluation method to achieve real-time evaluation of the energy efficiency level of a hardware object and reduce the cost required to evaluate the energy efficiency level of the hardware object.
  • the present application also provides a corresponding device, an energy efficiency evaluation system, a computer-readable storage medium, and a computer program product.
  • the present application provides an energy efficiency evaluation method. Specifically, when evaluating the energy efficiency level of a hardware object, the operating state and actual power consumption of the hardware object are obtained.
  • the operating state of the hardware object is used to indicate the operating state of the hardware object at the current moment, such as indicating the utilization rate of the hardware object at the current moment, and then the ideal power consumption corresponding to the operating state is determined.
  • the ideal power consumption is used to indicate the minimum power consumption of the hardware object that can be operated in the operating state.
  • the minimum power consumption can be the lowest power consumption achieved by the hardware object in a historical time period, or it can be the minimum power consumption set based on human experience. Therefore, the energy-saving health of the hardware object is calculated based on the actual power consumption and ideal power consumption of the hardware object, and the energy-saving health is used to measure the energy efficiency level of the hardware object.
  • the energy-saving health degree used to measure the energy efficiency level of the hardware object can be calculated, thereby realizing real-time online evaluation of the energy efficiency level of the hardware object.
  • using the data of the hardware object in the operating state to evaluate its energy efficiency level can effectively reduce the time consumption of evaluating the energy efficiency level and the resource occupation of the hardware object, thereby effectively reducing the time cost and resource consumption required for energy efficiency evaluation.
  • the operating status of the hardware object can also reflect the characteristics of multiple dimensions such as the resource utilization rate of the hardware object, the operating conditions of the internal devices, and the sensor temperature required by the business load on the hardware object, thereby evaluating the energy efficiency level of the hardware object based on the operating status of the hardware object, which can make the accuracy, reliability and fairness of the energy efficiency level evaluation reach a high level.
  • the ideal power consumption corresponding to the operating state when determining the ideal power consumption corresponding to the operating state of the hardware object, can be determined based on the power consumption data of the hardware object in a historical time period. In this way, the power consumption of the hardware object during past operation can be used to guide the minimum power consumption that the hardware object can achieve in the current operating state, thereby improving the reliability and accuracy of determining the ideal power consumption.
  • the ideal power consumption based on the power consumption data of the hardware object in a historical time period when determining the ideal power consumption based on the power consumption data of the hardware object in a historical time period, it can be specifically based on the operating state of the hardware object, using the AI model for reasoning to obtain the ideal power consumption corresponding to the operating state output by the AI model.
  • the AI model uses the power consumption data of the hardware object in the historical time period to complete the training. For example, the various historical operating states of the hardware object in the historical time period can be used as input, and the lowest power consumption achieved by the hardware object in each historical operating state can be used as a label to train the AI model. In this way, the ideal power consumption of the hardware object can be determined by model reasoning.
  • the AI model for determining the ideal power consumption can be trained based on at least one set of training samples, the training samples including multiple historical operating states of the hardware object in a historical time period, and the minimum power consumption corresponding to each historical operating state, the minimum power consumption being the minimum value of multiple power consumptions corresponding to multiple sets of energy-saving control parameters under the same historical operating state, the energy-saving control parameters Used to control the energy consumption of hardware objects.
  • the AI model can be trained based on historical data so that the AI model can be used to achieve real-time reasoning on the ideal power consumption of the hardware object.
  • the training sample may also include energy-saving control parameters corresponding to the minimum power consumption corresponding to each historical operating state.
  • the energy-saving control parameters as inputs to the AI model, the data dimension of the AI model input can be increased, thereby performing reasoning based on multiple dimensional data, which can improve the reasoning accuracy of the AI model.
  • the ideal power consumption corresponding to the operating state of the hardware object can be corrected according to the energy-saving health.
  • the determined ideal power consumption can be made more reliable, and the interference of some factors in determining the ideal power consumption can be reduced, so that the accuracy of the calculated energy-saving health can be improved, thereby improving the accuracy of measuring the energy efficiency level of the hardware object.
  • the correction amount for the ideal power consumption when the energy-saving health of the hardware object is greater than the first threshold for a duration greater than the first duration, the correction amount for the ideal power consumption is reduced, and when the energy-saving health of the hardware object is less than the second threshold for a duration greater than the second duration, the correction amount for the ideal power consumption is increased, wherein the first threshold is greater than the second threshold.
  • the value of the ideal power consumption used to calculate the energy-saving health can be made closer to the minimum power consumption that the hardware object can actually achieve, thereby improving the accuracy of the calculated energy-saving health, and further improving the accuracy of measuring the energy efficiency level of the hardware object.
  • the energy-saving health of the hardware object can also be presented so that the user can perceive the energy efficiency level of the hardware object based on the energy-saving health; or, when the energy-saving health is lower than a threshold, it indicates that the energy efficiency level of the hardware object is low, that is, more energy is wasted during the operation of the hardware object. Therefore, energy-saving operations can be performed on the hardware object based on the energy-saving health, such as reducing the frequency of the hardware object, entering energy-saving mode, etc., so as to improve the energy efficiency level of the hardware object.
  • the present application provides an energy efficiency evaluation device, which includes various modules for executing the energy efficiency evaluation method in the first aspect or any possible implementation manner of the first aspect.
  • the present application provides an energy efficiency evaluation system, which includes a processor, a memory and a display.
  • the processor and the memory communicate with each other.
  • the processor is used to execute instructions stored in the memory so that the energy efficiency evaluation system performs the energy efficiency evaluation method in the first aspect or any one of the implementations of the first aspect.
  • the memory can be integrated into the processor or can be independent of the processor.
  • the energy efficiency evaluation system can also include a bus.
  • the processor is connected to the memory via a bus.
  • the memory can include a readable memory and a random access memory.
  • the present application provides a computer-readable storage medium, which stores instructions.
  • the computer-readable storage medium When the computer-readable storage medium is executed on a computing device, the computing device executes the operating steps of the energy efficiency evaluation method described in the first aspect or any one of the implementations of the first aspect.
  • the present application provides a computer program product comprising instructions, which, when executed on a computing device, enables the computing device to execute the operating steps of the energy efficiency evaluation method described in the first aspect or any one of the implementations of the first aspect.
  • FIG1 is a schematic diagram of an exemplary application scenario provided by the present application.
  • FIG2 is a flow chart of an energy efficiency evaluation method provided by the present application.
  • FIG3 is a schematic diagram of calculating the energy-saving health of a device according to the energy-saving health of an electronic device provided by the present application;
  • FIG4 is a flow chart of a method for training an AI model provided in the present application.
  • FIG5 is a schematic diagram of determining the minimum power consumption in each historical operating state through state projection provided by the present application.
  • FIG6 is a schematic diagram of a curve showing a change in energy-saving health of a hardware object provided by the present application.
  • FIG7 is a schematic diagram of the structure of an energy efficiency evaluation device provided by the present application.
  • FIG8 is a schematic diagram of the structure of an energy efficiency evaluation system provided in the present application.
  • this application provides an energy efficiency level evaluation method, which calculates the energy efficiency level of the hardware object according to the state of the hardware object during operation and the power consumption generated.
  • the energy-saving health level of the hardware object can be evaluated in real time, thereby reducing the time cost and resource consumption required for evaluating the energy efficiency level.
  • FIG 1 is a schematic diagram of an exemplary application scenario provided by the present application.
  • the levels of the hardware objects can be divided according to the granularity of the product form to which the hardware objects belong in the actual application scenario, or can be divided according to the size of the service scope of the object, etc., and this is not limited.
  • Figure 1 is an example of hardware objects including three levels, where the hardware objects of the first level are electronic devices, such as the central processing unit (CPU) 101, hard disk 102, fan 103, etc. shown in Figure 1.
  • CPU central processing unit
  • the hardware objects of the second level are devices, such as device 201, device 202 and device 203 shown in FIG1.
  • the devices may be computing servers, storage servers or terminals, etc.
  • the hardware objects of the second level may include multiple hardware objects of the first level, such as device 201 may include CPU 101, hard disk 102, fan 103, etc.
  • Different devices may communicate through a wired network or a wireless network.
  • the hardware objects of the third level are clusters, such as cluster 200 shown in FIG1, which may include multiple devices.
  • FIG1 takes cluster 200 including device 201, device 202 and device 203 as an example for illustrative description.
  • cluster 200 may be a data center including multiple computing devices, or may be an availability zone (AZ) including multiple computing devices, or may be a region including multiple computing devices, etc., which is not limited in this embodiment.
  • AZ availability zone
  • each AZ includes a data center or multiple data centers with similar geographical locations, and generally a region may include multiple AZs.
  • different clusters may communicate through a wired network or a wireless network.
  • the energy consumption generated by hardware objects at all levels during operation includes the amount of energy used to provide business services (such as data storage services, data computing services, etc.), as well as the amount of energy lost during energy conversion and heat dissipation. Therefore, the energy usage of hardware objects can be evaluated by calculating their energy efficiency.
  • tools such as SPEC SERT and BenchSEE are usually used to apply standard loads to offline hardware objects, and the energy efficiency level of the hardware objects is evaluated by measuring the business throughput and energy consumption of the hardware objects under the load. It is impossible to perform online real-time evaluation of the energy efficiency level of hardware objects.
  • the time required to measure the energy efficiency level of hardware objects using standard loads is usually high, such as 1 to 5 hours, etc., and it requires more resources of the hardware objects. Therefore, this method of measuring the energy efficiency level of hardware objects also requires high time and resource costs.
  • the present application adds an energy efficiency evaluation device 300 in the application scenario shown in FIG1.
  • the energy efficiency evaluation device 300 calculates the energy-saving health degree according to the actual power consumption of the hardware object in the current operating state and the ideal power consumption (that is, the lowest power consumption that the hardware object can achieve under ideal conditions).
  • the energy-saving health degree can be used to measure the energy efficiency level of the hardware object, so as to achieve real-time online evaluation of the energy efficiency level of the hardware object.
  • the hardware object can be an object of any level in FIG1.
  • the energy efficiency evaluation device 300 uses the data of the hardware object in the operating state to evaluate its energy efficiency level without applying additional load to the hardware object.
  • the energy efficiency evaluation device 300 evaluates the energy efficiency level of the hardware object based on the operating status of the hardware object, which can achieve a high level of accuracy, reliability and fairness in evaluating the energy efficiency level.
  • the energy efficiency evaluation device 300 can be implemented by software, for example, it can be implemented by at least one of a virtual machine, a container, and a computing engine.
  • the energy efficiency evaluation device 300 can be implemented by a physical device including a processor, wherein the processor can be a CPU, and an application-specific integrated circuit (ASIC), a programmable logic device (PLD), a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), a system on chip (SoC), a software-defined infrastructure (SDI) chip, an artificial intelligence (AI) chip, a data processing unit (DPU), and any other processor or any combination thereof.
  • ASIC application-specific integrated circuit
  • PLD programmable logic device
  • CPLD complex programmable logical device
  • FPGA field-programmable gate array
  • GAL generic array logic
  • SoC system on chip
  • SDI software-defined infrastructure
  • AI artificial intelligence
  • DPU data processing unit
  • the number of processors included in the energy efficiency evaluation device 300 can be one or more, and the type of processors included can be one or more. Specifically, the number and type of processors can be set according to the business requirements of the actual application, and this embodiment does not limit this.
  • the device 101 may include a larger number of electronic devices, or the cluster 200 may include a larger number of devices, or the device 101 may be a terminal device such as a smart phone or a smart terminal (such as an iPad), and this embodiment does not limit this.
  • FIG 2 is a flow chart of an energy efficiency evaluation method provided in an embodiment of the present application.
  • the method can be applied to the application scenario shown in Figure 1, or can be applied to other applicable application scenarios.
  • the hardware object whose energy efficiency level is evaluated can be an electronic device, equipment or cluster in Figure 1.
  • an exemplary explanation is given by taking the hardware object as the first-level electronic device in Figure 1 as an example.
  • the energy efficiency evaluation method shown in FIG. 2 may be performed by the energy efficiency evaluation device 300 in FIG. 1 , and the method may specifically include:
  • the energy efficiency evaluation device 300 obtains the operating state and actual power consumption of a hardware object, where the operating state is used to indicate the operating state of the hardware object at a current moment.
  • the hardware object may be a first-level electronic device.
  • the energy efficiency evaluation device 300 may collect the operation status of the hardware object and the actual power consumption generated by the hardware object in the current operation status in real time, such as using a sensor to collect data such as the operation status and the actual power consumption.
  • the operating status of a hardware object may be one or more of parameters such as CPU utilization, memory utilization, memory read/write rate, hard disk read/write rate, device sensor temperature, etc., or other data used to characterize the operating status of a hardware object.
  • the actual power consumption of the hardware object may be, for example, the power when the hardware object is running, or the voltage and current when the hardware object is running, or other data used to characterize the power consumption of the hardware object.
  • the energy efficiency evaluation device 300 can also collect energy-saving control parameters of the hardware object in the operating state, and the energy-saving control parameters are used to control the operation of the hardware object, such as CPU frequency, memory frequency, disk speed, fan speed, power supply mode, etc.
  • the energy-saving control parameters are used to control the operation of the hardware object, such as CPU frequency, memory frequency, disk speed, fan speed, power supply mode, etc.
  • CPU frequency as an example, by controlling the CPU to run based on different frequencies, the power consumption generated by the CPU per unit time will usually be different.
  • the energy efficiency evaluation device 300 determines an ideal power consumption corresponding to the operating state of the hardware object, where the ideal power consumption is used to indicate the lowest power consumption of the hardware object that can be operated in the operating state.
  • the energy efficiency evaluation device 300 can further obtain the minimum power consumption of the hardware object in the operating state, that is, the minimum power consumption that can be achieved, that is, the ideal power consumption described in step S202.
  • the minimum power consumption that can be achieved by the hardware object in different operating states can be different, such as the minimum power consumption that can be achieved by the hardware object in operating state 1 is 100w (watt), and the minimum power consumption that can be achieved in operating state 2 is 150w.
  • the minimum power consumption that can be achieved by the hardware object in some operating states can be the same.
  • the minimum power consumption that can be achieved by the hardware object in the first operating state is the same as the minimum power consumption that can be achieved by the hardware object in the tenth operating state, but it is different from the minimum power consumption that can be achieved by the hardware object in the remaining 8 operating states.
  • this embodiment provides the following implementation examples for determining ideal power consumption.
  • the energy efficiency evaluation device 300 can determine the ideal power consumption corresponding to the hardware object in the current operating state based on the power consumption data of the hardware object in the historical time period, that is, the power consumption of the hardware object in the past period of time can be used to guide the minimum power consumption that the hardware object can achieve.
  • the energy efficiency evaluation device 300 can obtain the ideal power consumption through AI model reasoning.
  • the energy efficiency evaluation device 300 can obtain a trained AI model, which can be, for example, a model constructed based on a neural network model, such as a model constructed based on a recurrent neural network (RNN), a deep neural network (DNN), etc.; or, the AI model can be a regression tree model, a support vector machine (SVM) model, etc., and the specific implementation method of the AI model is not limited in this embodiment.
  • the training sample of the AI model can be, for example, the minimum power consumption achieved by the hardware object in various operating states during a historical time period.
  • the energy efficiency evaluation device 300 can input the obtained operating state into the AI model, and use the AI model to reason according to the input operating state to obtain the ideal power consumption corresponding to the operating state output by the AI model.
  • the energy efficiency evaluation device 300 can also obtain the energy-saving control parameters of the hardware object in the operating state, and input the operating state and energy-saving control parameters of the hardware object into the AI model, and the AI model infers the ideal power consumption of the hardware object. In this way, reasoning based on multiple dimensional data such as operating state and energy-saving control parameters can further improve the accuracy of the determined ideal power consumption, thereby further improving the accuracy of the subsequent evaluation of the energy efficiency level of the hardware object.
  • the energy efficiency evaluation device 300 may be configured with a mapping relationship between the operating state of the hardware object and the ideal power consumption, such as the mapping relationship may be configured in advance in the energy efficiency evaluation device 300 by a technician. In this way, after obtaining the operating state of the hardware object, the energy efficiency evaluation device 300 may determine the ideal power consumption corresponding to the operating state by searching the mapping relationship.
  • the operating state of the hardware object may be, for example, the CPU utilization rate
  • the energy efficiency evaluation device 300 may be configured with a mapping relationship between the CPU utilization rate and the ideal power consumption, such as configuring the ideal power consumption to be 100w when the CPU utilization rate is 10%; the ideal power consumption to be 300w when the CPU utilization rate is 50%. In this way, after obtaining the current CPU utilization rate, the energy efficiency evaluation device 300 may determine the ideal power consumption corresponding to the CPU utilization rate by searching the configured mapping relationship.
  • the mapping relationship in the energy efficiency evaluation device 300 can be determined according to the minimum power consumption achieved by the hardware object in various operating states during the historical time period. Taking the hardware object as a CPU as an example, assuming that in the past 30 days, the CPU utilization rate at multiple times is 10%, but the power consumption generated by the CPU at different times in the multiple times is 100w, 150w, and 300w respectively, then the minimum power consumption (100w) can be determined from the multiple power consumptions, and a mapping relationship between the CPU utilization rate (10%) and the minimum power consumption (100w) can be established. In actual application, the mapping relationship in the energy efficiency evaluation device 300 can also be determined by other methods, and this embodiment does not limit this.
  • the energy efficiency evaluation device 300 may also use other methods to determine the ideal power consumption of the object in the running state.
  • the energy efficiency evaluation device 300 calculates the energy-saving health of the hardware object according to the actual power consumption and the ideal power consumption. The energy-saving health is used to measure the energy efficiency level of the hardware object.
  • the ideal power consumption indicates the lowest power consumption that the hardware object can achieve in the current running state.
  • the lowest power consumption is the actual power consumption in the historical time period, that is, the lowest power consumption actually achieved by the hardware object during the operation of the past time period is taken as the theoretical value; and the actual power consumption is the power consumption actually generated by the hardware object in the current running state. Therefore, according to the actual power consumption and ideal power consumption of the hardware object, the energy efficiency level of the hardware object in the current running state can be reflected. Specifically, when the deviation between the actual power consumption and the ideal power consumption is small, it indicates that the current power consumption of the hardware object is small and is in a good energy-saving state; accordingly, the energy efficiency level of the hardware object is currently in a high state.
  • the energy efficiency level of the hardware object is currently in a low state.
  • the energy-saving health degree can be used to measure the energy efficiency level of the hardware object.
  • the energy-saving health degree is used to indicate the energy-saving effect of the hardware object, that is, it can be used to indicate the energy efficiency level of the hardware object. The greater the energy-saving health degree, the higher the energy efficiency level of the hardware object; the smaller the energy-saving health degree, the lower the energy efficiency level of the hardware object.
  • the energy efficiency evaluation device 300 can calculate the energy-saving health of the hardware object based on the following formula (1).
  • h1 is the energy-saving health of the hardware object
  • p is the actual power consumption
  • p * is the ideal power consumption.
  • the energy efficiency evaluation device 300 may also calculate the energy-saving health of the hardware object based on the following formula (2).
  • the energy efficiency evaluation device 300 can calculate the energy-saving health degree and realize online real-time evaluation of the energy efficiency level of the hardware object.
  • the energy efficiency evaluation device 300 may also perform the following steps:
  • the energy efficiency evaluation device 300 presents the energy-saving health of the hardware object.
  • users can learn the energy efficiency level of the hardware object based on the energy-saving health of the presented hardware object, so that users can understand the energy consumption of the hardware object during operation.
  • the energy-saving health of a hardware object is greater than or equal to a threshold value (such as 85%), it indicates that the current energy-saving state of the hardware object is good, and no further energy-saving operations need to be performed on the hardware object.
  • the current energy-saving state of the hardware object is poor, that is, there is a lot of energy waste during the operation of the hardware object.
  • the energy efficiency evaluation device 300 can perform energy-saving operations on the hardware object according to the energy-saving health to reduce the energy consumption generated by the hardware object during operation and improve the energy efficiency level of the hardware object.
  • the energy efficiency evaluation device 300 can obtain the energy-saving control parameters of the hardware object, generate new energy-saving control parameters according to the energy-saving health and the energy-saving control parameters, and perform energy-saving operations on the hardware object based on the energy-saving control parameters.
  • the energy-saving health of the CPU is 60%
  • the energy-saving control parameter is the CPU frequency
  • the frequency of the CPU in the current operating state is 3GHz (gigahertz). Since the energy-saving health of the CPU is lower than 85% (threshold), the energy efficiency evaluation device 300 can calculate that the frequency to be reached by the CPU after the CPU is down-clocked is 2GHz according to the energy-saving health and the CPU frequency of 3GHz in the current operating state; finally, the energy efficiency evaluation device 300 reduces the frequency of the CPU to the calculated 2GHz.
  • the frequency of the CPU during operation indicates the number of synchronization pulses that occur in the CPU within 1 second, which can determine the computing speed of the CPU.
  • the energy efficiency evaluation device 300 can reduce the CPU frequency to 2.8GHz, etc. according to the energy-saving health and the CPU frequency of 3GHz in the current operating state.
  • the energy efficiency evaluation device 300 can avoid missing energy-saving points during the operation of the hardware object by performing real-time evaluation on the hardware object and automatically performing corresponding energy-saving operations, and can perform energy-saving operations on the hardware object in a timely manner, thereby improving the energy-saving effect on the hardware object and ensuring that the energy efficiency level of the hardware object is always maintained at a high level.
  • the timing relationship between the steps shown in FIG. 2 is not used for limitation.
  • the energy efficiency evaluation device 300 may execute step S204 and step S205 simultaneously, or the energy efficiency evaluation device 300 may execute step S205 first and then execute step S204, etc., and this is not limited.
  • the specific implementation of the energy efficiency evaluation device 300 evaluating the energy efficiency level of the hardware object is introduced by taking the hardware object as an example of a first-level electronic device.
  • the energy efficiency evaluation device 300 can also calculate the energy-saving health level corresponding to the device or the energy-saving health level corresponding to the cluster based on the above similar method, so as to realize real-time evaluation of the energy efficiency level of the device or the cluster.
  • the energy efficiency evaluation device 300 may further calculate the energy efficiency health degree corresponding to the second-level device based on the energy-saving health degree corresponding to the electronic device of the first level, so as to achieve real-time evaluation of the energy efficiency level of the device. Then, the energy efficiency evaluation device 300 may further calculate the energy-saving health degree corresponding to the third-level cluster based on the energy-saving health degree corresponding to the device of the second level, so as to achieve real-time evaluation of the energy efficiency level of the cluster.
  • the energy efficiency evaluation device 300 can calculate the energy-saving health corresponding to the multiple electronic devices included in the single device of the second level based on the process described in the embodiment shown in FIG2 above, assuming that the single device includes N electronic devices (N is a positive integer).
  • the device may include multiple electronic devices such as CPU, memory, hard disk, fan, power supply unit (PSU), etc., and during the operation of the device, the multiple electronic devices in the device are in operation and generate energy consumption.
  • PSU power supply unit
  • the energy efficiency evaluation device 300 can use the AI model corresponding to the electronic device to infer the ideal power consumption of the electronic device according to the operating state of each electronic device, thereby calculating the energy-saving health of each electronic device based on the actual power consumption and ideal power consumption of each electronic device.
  • the energy efficiency evaluation device 300 performs weighted summation on the energy-saving health degrees corresponding to the N electronic devices, and calculates the energy-saving health degree of the entire device, as shown in Figure 3.
  • the energy efficiency evaluation device 300 can calculate the energy-saving health degree of the entire device based on the following formula (3).
  • h2 is the energy-saving health of a single device
  • N is the number of electronic devices that generate energy consumption included in the device
  • hi is the energy-saving health of the ith electronic device
  • wi is the weight corresponding to the ith electronic device.
  • the weights corresponding to each electronic device can be set in advance by technical personnel according to the needs of actual applications, such as setting according to the importance of each electronic device or the proportion of energy consumption, and configured in the energy efficiency evaluation device 300, so that the energy efficiency evaluation device 300 can obtain the energy-saving health of the whole machine by weighted summation based on the weights and energy-saving health of each electronic device. In this way, according to the energy-saving health of the electronic devices in each device, the energy-saving health of each device at the second level can be calculated, so as to realize real-time evaluation of the energy efficiency level of each device.
  • the energy efficiency evaluation device 300 can also perform corresponding energy-saving operations on each device. To improve the energy efficiency level of each device.
  • the energy efficiency evaluation device 300 can compare the energy-saving health of the device with a preset threshold. And, when the energy-saving health is greater than or equal to the threshold, the energy efficiency evaluation device 300 may not perform energy-saving operations. When the energy-saving health is less than the threshold, the energy efficiency evaluation device 300 can perform energy-saving operations on each electronic device in the device according to the energy-saving health or the energy-saving health of each electronic device in the device.
  • the device may include electronic devices such as CPU, memory, hard disk, fan, PSU, etc.
  • the energy efficiency evaluation device 300 can perform energy-saving operations such as dynamic voltage and frequency scaling (DVFS), sleep or shut down the processor core on the CPU; reduce the refresh rate of the memory, such as reducing the refresh rate of the memory from 2000MHz (megahertz) to 1300MHz; reduce the head speed of the hard disk, or switch the working mode of the hard disk to the sleep mode; reduce the speed of the fan; switch the power supply mode of the PSU, such as switching the power supply mode of the PSU from the load balancing mode to the active-standby mode, etc.
  • performing energy-saving operations on devices based on their energy-saving health can increase the energy efficiency level of the devices by an average of more than 10%.
  • the energy consumption of the cluster is the sum of the energy consumption generated by one or more devices in the cluster. Therefore, the energy efficiency evaluation device 300 can further calculate the energy saving health of the cluster based on the calculated energy saving health of each device.
  • the energy efficiency evaluation device 300 may calculate the energy-saving health of the cluster based on the following formula (4).
  • h3 is the energy-saving health of the cluster
  • M is the number of energy-generating devices included in the cluster
  • hj is the energy-saving health of the jth device
  • wj is the weight corresponding to the jth device.
  • the weight corresponding to each device can be set in advance by the technician according to the needs of the actual application, such as setting it according to the importance of each device or the proportion of energy consumption, and it is configured in the energy efficiency evaluation device 300, so that the energy efficiency evaluation device 300 can obtain the energy-saving health of the entire cluster by weighted summation according to the weight and energy-saving health of each device. In this way, the energy efficiency evaluation device 300 can realize real-time evaluation of the energy efficiency level of the cluster.
  • the energy efficiency evaluation device 300 can also perform corresponding energy-saving operations on the cluster to improve the energy efficiency level of the cluster. For example, the energy efficiency evaluation device 300 can shut down some devices in the cluster, or adjust some devices in the cluster from the running state to the dormant state, etc., so as to reduce the overall energy consumption of the cluster.
  • the energy efficiency evaluation device 300 determines the ideal power consumption of the hardware object and determines the energy-saving health according to the ideal power consumption, so as to realize the real-time evaluation of the hardware object.
  • the energy efficiency evaluation device 300 can use the pre-trained AI model to infer the ideal power consumption of the hardware object in the current operating state.
  • the process of training the AI model is introduced in detail.
  • the AI model can be trained by the energy efficiency evaluation device 300, or the AI model can be obtained by training other devices and then the AI model is provided to the energy efficiency evaluation device 300.
  • the following is an exemplary explanation of the training process of the AI model executed by the energy efficiency evaluation device 300.
  • FIG4 a flow chart of a method for training an AI model by the energy efficiency evaluation device 300 is shown. As shown in FIG4 , the method includes:
  • S401 The energy efficiency evaluation device 300 constructs an AI model.
  • the AI model may be a model constructed based on a neural network model, such as a model constructed based on RNN, DNN, etc.; or, the AI model may be a regression tree model, a SVM model.
  • the AI model may also be constructed by the user and then input into the energy efficiency evaluation device 300, which is not limited here.
  • the energy efficiency evaluation device 300 collects historical operation data of the hardware object, where the historical operation data includes a historical operation state of the hardware object in a historical time period and power consumption generated by the hardware object in the historical operation state.
  • the historical operation data collected by the energy efficiency evaluation device 300 may also include energy-saving control parameters adopted by the hardware object in the historical operation state, such as CPU frequency, memory frequency, disk speed, fan speed, power supply mode, etc.
  • the historical time period refers to a period of time in the past, such as the past 15 days, 30 days, 180 days, etc.
  • a hardware object may generate a corresponding log during operation, and the log is used to record relevant parameters of the hardware object during operation, such as operating status, operating power (and energy-saving control parameters), etc. Then, the energy efficiency evaluation device 300 may obtain the log generated by the hardware object in the historical time period, and read the historical operation data of the hardware object from the log.
  • the energy efficiency evaluation device 300 may record the relevant parameters of the hardware object during the operation of the hardware object, and when the amount of recorded data reaches a preset threshold or when the recording duration reaches a preset duration, the energy efficiency evaluation device 300 Stop recording data and use the recorded data as the historical operation data of the hardware object.
  • the energy efficiency evaluation device 300 generates a training sample based on the collected historical operation data.
  • the training sample includes a plurality of historical operation states of the hardware object within a historical time period and a historical minimum power consumption corresponding to each of the plurality of historical operation states.
  • This embodiment provides the following implementation examples for generating training samples.
  • the energy efficiency evaluation device 300 can traverse the historical operation data, determine the multiple historical operation states in which the hardware object is in the historical time period, and further determine one or more power consumptions generated by the hardware object in each historical operation state, and different power consumptions correspond to multiple moments in the historical time period, such as the power consumption of the hardware object at time A is 100w, the power consumption at time B is 150w, and the power consumption at time C is 300w, and the multiple energy consumptions are generated by the hardware object under the control of multiple sets of energy-saving control parameters.
  • the energy efficiency evaluation device 300 determines the minimum power consumption that the hardware object can achieve in the historical operation state, such as the energy efficiency evaluation device 300 can compare the power consumption of the hardware object at time A, time B and time C, and determine that the minimum power consumption that the hardware object can achieve in the historical operation state is 100w. In this way, the energy efficiency evaluation device 300 can determine the minimum power consumption corresponding to the hardware object in each historical operation state.
  • the energy efficiency evaluation device 300 uses the multiple historical operating states as model input, uses the minimum power consumption corresponding to each historical operating state as a training label of the AI model, and generates training samples for the AI model.
  • the energy efficiency evaluation device 300 can also determine the energy-saving control parameters corresponding to the minimum power consumption after determining the minimum power consumption corresponding to each historical operating state.
  • the energy efficiency evaluation device 300 uses the multiple historical operating states and the energy-saving control parameters corresponding to each historical operating state as model inputs, uses the minimum power consumption corresponding to each historical operating state as a training label for the AI model, and generates training samples for the AI model.
  • each historical operation data acquired by the energy efficiency evaluation device 300 includes a historical operation state and an energy-saving control parameter, as shown in the following formula (5).
  • si is the i-th historical operation data; is the i-th historical running state; is the i-th energy-saving control parameter.
  • the energy efficiency evaluation device 300 may project the historical operation data having the same historical operation state to obtain a projection state, as shown in the following formula (6).
  • the energy efficiency evaluation device 300 can traverse and compare multiple power consumptions with the same projection state based on the following formula (7) to determine the minimum power consumption in each projection state, that is, to determine the minimum power consumption that the hardware object can achieve in each historical operating state, as shown in FIG5 .
  • the power consumption of the hardware object is characterized by the power of the hardware object. In other embodiments, it can also be characterized by parameters such as voltage and current, which are not limited here.
  • the energy efficiency evaluation device 300 can determine the minimum power consumption corresponding to each historical operating state of the hardware object, and use the multiple historical operating states as model inputs, and use the minimum power consumption corresponding to each historical operating state as the training label of the AI model to generate training samples for the AI model.
  • multiple historical operating states and the energy-saving control parameters corresponding to each historical operating state can be used as model inputs, and the minimum power consumption corresponding to each historical operating state can be used as the training label of the AI model to generate training samples for the AI model.
  • the energy efficiency evaluation device 300 generates training samples based on the historical operation data of the hardware object as an example. In other embodiments, the energy efficiency evaluation device 300 may also generate training samples based on other data or based on other methods. For example, training samples may be generated based on test data, or the ideal power consumption of the hardware object in each operating state may be set by a technician based on experience.
  • S404 The energy efficiency evaluation device 300 uses training samples to train the constructed AI model.
  • the energy efficiency evaluation device 300 can input the historical operating status (and energy-saving control parameters) in the training sample into the AI model, and the AI model will infer based on the sample input and output the ideal power consumption. Then, the energy efficiency evaluation device 300 can compare the ideal power consumption output by the AI model with the minimum power consumption in the training sample, and adjust the parameters in the AI model according to the deviation between the ideal power consumption and the minimum power consumption, so as to realize the training of the AI model. In this way, based on multiple sets of historical operating status and minimum power consumption data in the training samples, the AI model can be trained multiple times until the AI model completes the training termination conditions, such as the AI model converges or The number of training sessions reaches the preset number, etc.
  • the energy efficiency evaluation device 300 can implement the training of the AI model based on the above steps S401 to S404. In this way, the energy efficiency evaluation device 300 can use the AI model to implement real-time evaluation of the energy efficiency level of the hardware object.
  • the energy efficiency evaluation device 300 can train different AI models for different electronic devices, such as training AI model 1 for the CPU and training AI model 2 for the memory, etc., so as to use different AI models to infer the ideal power consumption of different electronic devices.
  • the energy efficiency evaluation device 300 uses the AI model to infer the ideal power consumption of the hardware object in the current operating state, it can also correct the ideal power consumption output by the AI model, as shown in the following formula (8).
  • p ** p ** + ⁇ Formula (8)
  • p ** is the ideal power consumption after correction
  • p * is the ideal power consumption output by the AI model, that is, the ideal power consumption before correction
  • ⁇ is the correction amount, and the value of the correction amount corresponding to different operating states may be different.
  • the energy-saving health of the hardware object can be calculated based on the following formula (9).
  • the correction amount ⁇ can be set by a technician.
  • the correction amount ⁇ may be dynamically set by the energy efficiency evaluation device 300 according to the energy-saving health of the hardware object within a period of time.
  • the energy efficiency evaluation device can continuously monitor the energy-saving health of the hardware object, and when the energy-saving health of the hardware object is greater than or equal to the first threshold, it indicates that the energy-saving state of the hardware object is good, and there is no need to perform energy-saving operations to further improve the energy efficiency level of the hardware object.
  • the energy-saving health of the hardware object is greater than or equal to the first threshold for a duration greater than the first duration, that is, when the energy-saving health of the hardware object continues to be in a high energy-saving health interval for a long time, it may be that the ideal power consumption output by the AI model is too high, that is, higher than the minimum power consumption that the hardware object can actually achieve.
  • the energy efficiency evaluation device 300 can reduce the correction amount to reduce the value of the ideal power consumption, so that the value of the ideal power consumption used to calculate the energy-saving health is closer to the minimum power consumption that the hardware object can actually achieve, thereby improving the accuracy of the calculated energy-saving health.
  • the energy efficiency evaluation device 300 can perform corresponding energy-saving operations on the hardware object according to the energy-saving health to improve the energy efficiency level of the hardware object and achieve further energy saving of the hardware object.
  • the energy-saving health of the hardware object is less than the second threshold for a duration greater than the second duration, that is, when the energy-saving health of the hardware object continues to be in the low energy-saving health interval for a long time, it may be that the ideal power consumption output by the AI model is low, that is, lower than the minimum power consumption that the hardware object can actually achieve.
  • the energy efficiency evaluation device 300 can increase the correction amount to increase the value of the ideal power consumption, so that the value of the ideal power consumption used to calculate the energy-saving health is closer to the minimum power consumption that the hardware object can actually achieve, thereby improving the accuracy of the calculated energy-saving health.
  • the energy efficiency evaluation device 300 continuously adjusts the ideal power consumption output by the AI model, and the energy-saving health change curve of the hardware object calculated based on the continuously adjusted ideal power consumption can be shown in Figure 6.
  • the energy efficiency evaluation device 300 can gradually reduce the adjustment amount of the correction amount as the monitoring time of the energy-saving health of the hardware object increases until it reaches 0, so that the value of the ideal power consumption determined by the energy efficiency evaluation device 300 converges. At this time, the accuracy of the energy-saving health calculated by the energy efficiency evaluation device 300 based on the converged ideal power consumption can be continuously maintained at a high level, that is, the accuracy of the evaluation of the energy efficiency level of the hardware object is stably maintained at a high level.
  • the energy efficiency evaluation device 300 can also use the corrected ideal power consumption and the current operating state of the hardware object (and the energy-saving control parameters adopted) to update the AI model, so as to improve the accuracy of the ideal power consumption output by the AI model according to the operating state.
  • the updated AI model to infer the ideal power consumption of the hardware object in each operating state, the accuracy of the energy-saving health calculated based on the ideal power consumption can be further improved, and the energy efficiency level of the hardware object can be effectively evaluated.
  • the energy efficiency evaluation device 300 uses the operating power consumption of the hardware object in the historical time period to not only train an AI model for determining the ideal power consumption of the hardware object, so as to use the AI model to achieve real-time evaluation of the energy efficiency level of the hardware object, but also improve the reliability and credibility of the ideal power consumption inferred by the AI model. Therefore, the energy-saving health calculated by the energy efficiency evaluation device 300 based on the ideal power consumption output by the AI model can more accurately reflect the energy efficiency level of the hardware object.
  • the energy efficiency evaluation device 300 can further dynamically adjust the correction amount of the ideal power consumption, so that the determined ideal power consumption In order to make power consumption more reliable, the interference of some factors (such as the operating power consumption of the hardware object in the historical time period not reaching the actual minimum power consumption that can be achieved, etc.) in determining the ideal power consumption can be reduced. This can further improve the accuracy of the energy-saving health degree finally calculated by the energy efficiency evaluation device 300, thereby improving the accuracy of measuring the energy efficiency level of the hardware object.
  • the energy efficiency evaluation device 700 includes:
  • An acquisition module 701 is used to acquire the operating state and actual power consumption of a hardware object, where the operating state is used to indicate the operating state of the hardware object at a current moment;
  • a determination module 702 is used to determine an ideal power consumption corresponding to the operating state, where the ideal power consumption is used to indicate the minimum power consumption of the hardware object that can be operated in the operating state;
  • the calculation module 703 is used to calculate the energy-saving health of the hardware object according to the actual power consumption and the ideal power consumption, and the energy-saving health is used to measure the energy efficiency level of the hardware object.
  • the determination module 702 is configured to determine the ideal power consumption corresponding to the operating state according to power consumption data of the hardware object in a historical time period.
  • the determination module 702 is used to perform reasoning using an artificial intelligence AI model based on the operating state to obtain an ideal power consumption corresponding to the operating state output by the AI model, and the AI model completes training using power consumption data of the hardware object in a historical time period.
  • the AI model completes training based on at least one group of training samples, the training samples including multiple historical operating states of the hardware object within the historical time period, and the minimum power consumption corresponding to each of the multiple historical operating states, the minimum power consumption being the minimum value of multiple power consumptions corresponding to multiple groups of energy-saving control parameters under the same historical operating state, and the energy-saving control parameters are used to control the energy consumption of the hardware object.
  • the training sample further includes an energy-saving control parameter corresponding to the minimum power consumption.
  • the device 700 further includes:
  • the correction module 704 is used to correct the ideal power consumption corresponding to the operating state according to the energy-saving health level.
  • the correction module 704 is further configured to:
  • the correction amount for the ideal power consumption is increased, and the first threshold is greater than the second threshold.
  • the hardware object is one of a plurality of electronic devices included in the computing device
  • the acquisition module is further used to acquire the energy-saving health of the multiple electronic devices, wherein the energy-saving health of the multiple electronic devices includes the energy-saving health of the hardware object;
  • the calculation module is further used to calculate the energy-saving health of the computing device according to the energy-saving health of the multiple electronic components.
  • the device 700 further includes:
  • a presentation module 705 is used to present the energy-saving health of the hardware object
  • an execution module 706 is used to perform an energy-saving operation on the hardware object according to the energy-saving health level when the energy-saving health level is lower than a threshold.
  • the energy efficiency evaluation device 700 shown in FIG. 7 corresponds to the method shown in FIG. 4 , the specific implementation method of the energy efficiency evaluation device 700 shown in FIG. 7 and the technical effects thereof can be found in the relevant descriptions in the aforementioned embodiments and will not be described in detail here.
  • FIG8 is a schematic diagram of an energy efficiency evaluation system 800 provided by the present application.
  • the energy efficiency evaluation system 800 shown in FIG8 can be used to implement the method steps performed by the energy efficiency evaluation device 300 in the embodiment shown in FIG2.
  • the energy efficiency evaluation system 800 can be, for example, an independently operable card, a server, a processor in a server, etc., which is not limited in this embodiment.
  • the hardware structure of the energy efficiency evaluation system 800 is introduced below by taking the energy efficiency evaluation system 800 as a server as an example.
  • the energy efficiency evaluation system 800 includes a processor 801, a memory 802, and a communication interface 803.
  • the processor 801, the memory 802, and the communication interface 803 communicate through a bus 804, and may also communicate through other means such as wireless transmission.
  • the memory 802 is used to store instructions, and the processor 801 is used to execute the instructions stored in the memory 802.
  • the energy efficiency evaluation system 800 may also include a memory unit 805, and the memory unit 805 may be connected to the processor 801, the storage medium 802, and the communication interface 803 through the bus 804.
  • the memory 802 stores program code, and the processor 801 may call the program code stored in the memory 802 to perform the following operations:
  • the energy-saving health of the hardware object is calculated according to the actual power consumption and the ideal power consumption, and the energy-saving health is used to measure the energy efficiency level of the hardware object.
  • the processor 801 may be a CPU, and the processor 801 may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete device components, etc.
  • DSPs digital signal processors
  • ASICs application-specific integrated circuits
  • FPGAs field programmable gate arrays
  • a general-purpose processor may be a microprocessor or any conventional processor, etc.
  • the memory 802 may include a read-only memory and a random access memory, and provide instructions and data to the processor 801.
  • the memory 802 may also include a non-volatile random access memory.
  • the memory 802 may also store information about the device type.
  • the memory 802 may be a volatile memory or a nonvolatile memory, or may include both volatile and nonvolatile memories.
  • the nonvolatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory.
  • the volatile memory may be a random access memory (RAM), which is used as an external cache.
  • RAM random access memory
  • DRAM dynamic random access memory
  • SDRAM synchronous DRAM
  • double data rate synchronous dynamic random access memory double data rate SDRAM, DDR SDRAM
  • enhanced SDRAM enhanced synchronous dynamic random access memory
  • SLDRAM synchronous link dynamic random access memory
  • direct rambus RAM direct rambus RAM, DR RAM
  • the communication interface 803 is used to communicate with other devices connected to the energy efficiency evaluation system 800.
  • the bus 804 may also include a power bus, a control bus, and a status signal bus.
  • various buses are labeled as bus 804 in the figure.
  • the energy efficiency evaluation system 800 may correspond to the energy efficiency evaluation device 700 in the embodiment of the present application, and may correspond to the energy efficiency evaluation device 300 executing the method shown in Figure 2 according to the embodiment of the present application, and the above-mentioned and other operations and/or functions implemented by the energy efficiency evaluation system 800 are respectively for implementing the corresponding processes of the method in Figure 2, and for the sake of brevity, they will not be repeated here.
  • the embodiment of the present application also provides a computer-readable storage medium.
  • the computer-readable storage medium can be any available medium that can be stored by a computing device or a data storage device such as a data center containing one or more available media.
  • the available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk).
  • the computer-readable storage medium includes instructions that instruct the computing device to execute the above-mentioned energy efficiency evaluation method.
  • the embodiment of the present application also provides a computer program product.
  • the computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computing device, the process or function described in the embodiment of the present application is generated in whole or in part.
  • the computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium.
  • the computer instructions may be transmitted from one website, computer or data center to another website, computer or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.
  • wired e.g., coaxial cable, optical fiber, digital subscriber line (DSL)
  • wireless e.g., infrared, wireless, microwave, etc.
  • the computer program product may be a software installation package.
  • the computer program product may be downloaded and executed on a computing device.
  • the above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof.
  • the above embodiments may be implemented in whole or in part in the form of a computer program product.
  • the computer program product may include one or more computer programs. Machine instructions. When the computer program instructions are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part.
  • the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.
  • the computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium.
  • the computer instructions may be transmitted from a website site, a computer, a server, or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means to another website site, computer, server, or data center.
  • the computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media sets.
  • the available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium.
  • the semiconductor medium may be a solid-state hard disk.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Quality & Reliability (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • Computer Hardware Design (AREA)
  • Power Sources (AREA)

Abstract

一种能效评估方法，包括：获取硬件对象的运行状态以及实际功耗，该硬件对象的运行状态用于指示该硬件对象在当前时刻的运行状态，并确定该运行状态所对应的理想功耗，该理想功耗用于指示该硬件对象在该运行状态下的可运行的最低功耗，从而根据硬件对象的实际功耗以及理想功耗，计算硬件对象的节能健康度，该节能健康度用于衡量硬件对象的能效水平。如此，不仅能够实现对硬件对象的能效水平进行实时在线评估，而且，利用硬件对象在运行状态下的数据来评估其能效水平，可以有效减小评估能效水平的耗时以及对于硬件对象的资源占用，从而能够有效降低能效评估所需的时间成本以及资源消耗。

Description

能效评估方法、装置、系统及相关设备
本申请要求于2022年11月11日提交国家知识产权局、申请号为202211412374.5、申请名称为“能效评估方法、装置、系统及相关设备”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及计算机技术领域，尤其涉及一种能效评估方法、装置、系统及相关设备。
背景技术
随着技术的发展，在处理器、计算机等硬件对象的性能快速提升的同时，硬件对象的能耗也在随机增加。其中，能效(energy efficiency)可以用来作为评估硬件对象的能源使用情况的指标。能效，是指能源使用过程中用于提供服务的能源量与消耗的总能源量之间的比值。
目前，主要是通过SPEC SERT、BenchSEE等能效评估工具，在硬件对象处于离线状态下评估硬件对象的能效水平，评估过程可利用标准的负载对硬件对象进行压力测试，并通过测量硬件对象在该负载下的业务吞吐量和能耗来评估该硬件对象的能效水平。但是，这些能效水平评估工具无法对硬件对象的能效水平进行实时评估，并且，评估硬件对象的能效水平的成本通常较高。
发明内容
本申请提供了一种能效评估方法，以实现对硬件对象的能效水平进行实时评估，并降低评估硬件对象的能效水平所需的成本。此外，本申请还提供了对应的装置、能效评估系统、计算机可读存储介质以及计算机程序产品。
第一方面，本申请提供一种能效评估方法，具体地，在评估硬件对象的能效水平时，获取该硬件对象的运行状态以及实际功耗，该硬件对象的运行状态用于指示该硬件对象在当前时刻的运行状态，如指示硬件对象在当前时刻的利用率等，然后确定该运行状态所对应的理想功耗，该理想功耗用于指示该硬件对象在该运行状态下的可运行的最低功耗，该最低功耗可以是硬件对象在历史时间段内所达到的最低的功耗，也可以是基于人为经验所设定的最低功耗，从而根据硬件对象的实际功耗以及理想功耗，计算硬件对象的节能健康度，该节能健康度用于衡量硬件对象的能效水平。
如此，根据硬件对象的运行状态以及实际功耗，能够计算出用于衡量硬件对象的能效水平的节能健康度，从而实现对硬件对象的能效水平进行实时在线评估。并且，利用硬件对象在运行状态下的数据来评估其能效水平，可以有效减小评估能效水平的耗时以及对于硬件对象的资源占用，从而能够有效降低能效评估所需的时间成本以及资源消耗。另外，当硬件对象为设备或者集群时，硬件对象的运行状态，还可以反映硬件对象上的业务负载所要求的硬件对象的资源使用率、内部器件的运行情况、传感温度等多个维度的特征，从而基于该硬件对象的运行状态来评估硬件对象的能效水平，可以使得评估能效水平的准确性、可靠性以及公平性能够达到较高水平。
在一种可能的实施方式中，在确定硬件对象的运行状态对应的理想功耗时，具体可以是根据硬件对象在历史时间段内的功耗数据，确定该运行状态对应的理想功耗。如此，能够利用硬件对象在过去运行时的功耗情况，指导硬件对象在当前运行状态下所能达到的最低功耗，从而可以提高确定理想功耗的可靠性以及准确性。
在一种可能的实时方式中，在根据硬件对象在历史时间段内的功耗数据确定理想功耗时，具体可以是根据硬件对象的运行状态，利用AI模型进行推理，得到该AI模型输出的运行状态对应的理想功耗，该AI模型利用硬件对象在历史时间段内的功耗数据完成训练，例如可以将硬件对象在历史时间段内的各个历史运行状态作为输入，将硬件对象在各个历史运行状态下所达到的最低功耗作为标签，训练AI模型。如此，可以通过模型推理的方式确定硬件对象的理想功耗。
在一种可能的实施方式中，用于确定理想功耗的AI模型，可以基于至少一组训练样本完成训练，该训练样本包括硬件对象在历史时间段内的多个历史运行状态、每个历史运行状态对应的最低功耗，该最低功耗为同一历史运行状态下基于多组节能控制参数所对应的多个功耗中的最小值，该节能控制参数 用于控制硬件对象的能耗。如此，可以基于历史数据实现对AI模型的训练，以便后续利用该AI模型实现对硬件对象的理想功耗的实时推理。
在一种可能的实施方式中，训练样本还可以包括每个历史运行状态对应的最低功耗所对应的节能控制参数。如此，通过将节能控制参数也作为AI模型的输入，可以增加AI模型输入的数据维度，从而基于多个维度数据进行推理，可以提高AI模型的推理准确性。
在一种可能的实施方式中，在计算出硬件对象的节能健康度后，还可以根据该节能健康度，对硬件对象的运行状态对应的理想功耗进行修正。如此，通过对理想功耗的修正，可以使得所确定出的理想功耗更加可靠，减少部分因素对于确定理想功耗的干扰，以此可以提高所计算出的节能健康度的准确率，从而可以提高衡量硬件对象的能效水平的准确性。
在一种可能的实施方式中，当硬件对象的节能健康度大于第一阈值的持续时长大于第一时长时，减小针对该理想功耗的修正量，并且，当硬件对象的节能健康度小于第二阈值的持续时长大于第二时长时，增大针对该理想功耗的修正量，其中，该第一阈值大于第二阈值。如此，通过对修正量的大小调整，可以使得用于计算节能健康度的理想功耗的值更加接近于该硬件对象实际所能达到的最低功耗，从而提高所计算出的节能健康度的准确率，进而可以提高衡量硬件对象的能效水平的准确性。
在一种可能的实施方式中，在计算出硬件对象的节能健康度后，还可以呈现该硬件对象的节能健康度，以便用户根据该节能健康度感知硬件对象的能效水平；或者，当该节能健康度低于阈值时，表征硬件对象的能效水平较低，即该硬件对象在运行过程中存在较多的能源被浪费，因此，可以根据该节能健康度执行针对该硬件对象的节能操作，如对硬件对象进行降频、进入节能模式等处理，以此提高硬件对象的能效水平。
第二方面，本申请提供一种能效评估装置，所述能效评估装置包括用于执行第一方面或第一方面任一种可能实现方式中的能效评估方法的各个模块。
第三方面，本申请提供一种能效评估系统，所述能效评估系统包括处理器、存储器和显示器。所述处理器、所述存储器进行相互的通信。所述处理器用于执行存储器中存储的指令，以使得能效评估系统执行如第一方面或第一方面的任一种实现方式中的能效评估方法。需要说明的是，该存储器可以集成于处理器中，也可以是独立于处理器之外。能效评估系统还可以包括总线。其中，处理器通过总线连接存储器。其中，存储器可以包括可读存储器以及随机存取存储器。
第四方面，本申请提供一种计算机可读存储介质，所述计算机可读存储介质中存储有指令，当其在计算设备上运行时，使得计算设备执行上述第一方面或第一方面的任一种实现方式所述能效评估方法的操作步骤。
第五方面，本申请提供了一种包含指令的计算机程序产品，当其在计算设备上运行时，使得计算设备执行上述第一方面或第一方面的任一种实现方式所述能效评估方法的操作步骤。
本申请在上述各方面提供的实现方式的基础上，还可以进行进一步组合以提供更多实现方式。
附图说明
图1为本申请提供的一示例性应用场景示意图；
图2为本申请提供的一种能效评估方法的流程示意图；
图3为本申请提供的根据电子器件的节能健康度计算出设备的节能健康度的示意图；
图4为本申请提供的训练AI模型的方法流程示意图；
图5为本申请提供的通过状态投影确定每个历史运行状态下的最低功耗示意图；
图6为本申请提供的硬件对象的节能健康度变化曲线示意图；
图7为本申请提供的一种能效评估装置的结构示意图；
图8为本申请提供的一种能效评估系统的结构示意图。
具体实施方式
为了实现对硬件对象的能效水平进行实时评估、解决评估成本高的问题，本申请提供了一种能效水平评估方法，根据硬件对象在运行过程中的状态以及所产生的功耗，计算出用于评估硬件对象的能效水 平的节能健康度，以此实现实时评估硬件对象的能效水平、降低评估能效水平所需的时间成本以及资源消耗。
下面将结合本申请的附图，对本申请中的技术方案进行描述。
参见图1，为本申请提供的一示例性应用场景示意图。在图1所示的应用场景中，存在多个层级的硬件对象，硬件对象的层级例如可以根据硬件对象在实际应用场景中所属的产品形态的粒度进行划分，或者可以是根据对象的服务范围大小进行划分等，对此并不进行限定。图1中是以包括3个层级的硬件对象为例进行说明，其中，第一层级的硬件对象为电子器件，如图1所示的中央处理器(central processing unit，CPU)101、硬盘102、风扇103等，不同电子器件之间可以通过总线进行通信，如通过计算快速链接(compute express link，CXL)总线、外围元件快速互连(peripheral component interconnect express，PCIE)总线、内部集成电路(inter-integrated circuit，I2C)总线、统一总线(unified bus，UB或Ubus)中的一种或者多种总线进行通信等。第二层级的硬件对象为设备，如图1所示的设备201、设备202以及设备203，该设备可以是计算服务器、存储服务器或者终端等，第二层级的硬件对象可以包括多个第一层级的硬件对象，如设备201中可以包括CPU101、硬盘102、风扇103等，不同设备之间可以通过有线网络进行通信，或者可以通过无线网络进行通信。第三层级的硬件对象为集群，如图1所示的集群200，其可以包括多个设备，图1中以集群200包括设备201、设备202以及设备203为例进行示例性说明。实际应用时，集群200可以是包括多个计算设备的数据中心，或者可以是包括多个计算设备的可用区(availability zones，AZ)、或者可以是包括多个计算设备的分区(region)等，本实施例对此并不进行限定。其中，每个AZ包括一个数据中心或多个地理位置相近的数据中心，并且，通常一个region可以包括多个AZ。当第三层级的硬件对象包括多个集群时，不同集群之间可以通过有线网络或者无线网络进行通信。
各个层级的硬件对象在运行过程中所产生的能耗，包括用于提供业务服务(如数据存储服务、数据计算服务等)的能源量，以及在能量转换、散热等过程中发生损耗的能源量，因此，可以通过计算硬件对象的能效来评估硬件对象的能源使用情况。目前，通常是利用SPEC SERT、BenchSEE等工具，对处于离线状态的硬件对象施加标准的负载，并通过测量硬件对象在该负载下的业务吞吐量和能耗来评估该硬件对象的能效水平，无法实现对硬件对象的能效水平进行在线实时评估。并且，利用标准负载测量硬件对象的能效水平，所需的耗时通常较高，如可能需要1～5个小时等，需要占用硬件对象较多的资源，因此，这种测量硬件对象的能效水平的方式，所需的时间成本以及资源成本也较高。
基于此，本申请在图1所示的应用场景中增设了能效评估装置300，该能效评估装置300根据硬件对象在当前运行状态下的实际功耗以及理想功耗(也即理想情况下该硬件对象所能达到的最低功耗)，计算出节能健康度，该节能健康度能够用于衡量硬件对象的能效水平，以此实现对硬件对象的能效水平进行实时在线评估。其中，硬件对象可以是图1中任一层级的对象。并且，能效评估装置300利用硬件对象在运行状态下的数据来评估其能效水平，无需对硬件对象施加额外的负载，因此，可以有效减小评估能效水平的耗时以及对于硬件对象的资源占用，从而能够有效降低能效评估所需的时间成本以及资源消耗。另外，当硬件对象为设备或者集群时，硬件对象的运行状态，还可以反映硬件对象上的业务负载所要求的硬件对象的资源使用率、内部器件的运行情况、传感温度等多个维度的特征，从而能效评估装置300基于该硬件对象的运行状态来评估硬件对象的能效水平，可以使得评估能效水平的准确性、可靠性以及公平性能够达到较高水平。
示例性地，能效评估装置300可以通过软件实现，例如可以是通过虚拟机、容器、计算引擎中的至少一种实现等。或者，能效评估装置300可以通过包括处理器的物理设备实现，其中，处理器可以是CPU，以及专用集成电路(application-specific integrated circuit，ASIC)、可编程逻辑器件(programmable logic device，PLD)、复杂程序逻辑器件(complex programmable logical device，CPLD)、现场可编程门阵列(field-programmable gate array，FPGA)、通用阵列逻辑(generic array logic，GAL)、片上系统(system on chip，SoC)、软件定义架构(software-defined infrastructure，SDI)芯片、人工智能(artificial intelligence，AI)芯片、数据处理单元(Data processing unit，DPU)等任意一种处理器或其任意组合。并且，能效评估装置300中所包括的处理器的数量可以是一个或者多个，所包括的处理器的种类可以是一种或者多种，具体可以根据实际应用的业务需求设定处理器的数量和种类，本实施例对此并不进行限定。
值得注意的是，上述图1所示的应用场景仅作为一种示例性说明，实际应用时，上述能效评估方法也可以应用于其它应用场景。比如，在其它可能的应用场景中，设备101可以包括更多数量的电子器件，或者集群200可以包括更多数量的设备等，或者，设备101可以是智能手机、智能终端(如iPad)等终端设备，本实施例对此并不进行限定。
为便于理解，下面结合附图，对本申请提供的能效评估方法的实施例进行描述。
参见图2，图2为本申请实施例提供的一种能效评估方法的流程示意图，该方法可以应用于图1所示的应用场景，或者可以应用于其它可适用的应用场景。其中，被评估能效水平的硬件对象可以是图1中的电子器件、设备或者集群。为便于说明，本实施例中以硬件对象具体为图1中的第一层级的电子器件为例进行示例性说明。
其中，图2所示的能效评估方法可以由图1中的能效评估装置300执行，该方法具体可以包括：
S201：能效评估装置300获取硬件对象的运行状态以及实际功耗，该运行状态用于指示硬件对象在当前时刻的运行状态。
本实施例中，硬件对象，具体可以是第一层级的电子器件。在硬件对象运行过程中，能效评估装置300可以实时采集硬件对象的运行状态以及该硬件对象在当前运行状态下所产生的实际功耗，如利用传感器采集运行状态以及实际功耗等数据。
示例性地，硬件对象的运行状态，例如可以是CPU利用率、内存利用率、内存读写速率、硬盘读写速率、器件传感温度等参数中的一种或者多种，也可以是其它用于表征硬件对象运行状态的数据等。
硬件对象的实际功耗，例如可以是硬件对象运行时的功率，或者可以是硬件对象运行时的电压以及电流，或者可以是其它用于表征硬件对象功耗的数据。
进一步地，能效评估装置300还可以采集硬件对象在该运行状态下的节能控制参数，该节能控制参数用于控制硬件对象的运行，例如可以是CPU频率、内存频率、磁盘转速、风扇转速、电源供电模式等。以CPU频率为例，通过控制CPU基于不同的频率运行，该CPU在单位时间内所产生的功耗通常会存在差异。
S202：能效评估装置300确定硬件对象的运行状态对应的理想功耗，该理想功耗用于指示该硬件对象在该运行状态下的可运行的最低功耗。
能效评估装置300在获取硬件对象的运行状态后，可以进一步获取硬件对象在该运行状态下可运行的最低功耗，也即所能达到的最低功耗，也即步骤S202中所述的理想功耗。其中，硬件对象在不同运行状态下所能达到的最低功耗可以不同，如硬件对象在运行状态1下所能达到的最低功耗为100w(瓦特)，而在运行状态2下所能达到的最低功耗为150w。或者，硬件对象在部分运行状态下所能达到的最低功耗可以相同，比如，在硬件对象所可能处于的10种不同的运行状态中，硬件对象在第1种运行状态下所能达到的最低功耗，与硬件对象在第10种运行状态下所能达到的最低功耗相同，但是与硬件对象在其余8种运行状态下所能达到的最低功耗均不同。为便于理解，本实施例提供了以下几种确定理想功耗的实现示例。
在第一种可能的实施方式中，能效评估装置300可以根据硬件对象在历史时间段内的功耗数据，确定硬件对象在当前的运行状态下所对应的理想功耗，即可以利用硬件对象在过去一段时间内的功耗情况，指导该硬件对象所能达到的最低功耗。
比如，能效评估装置300可以通过AI模型推理得到理想功耗。具体实现时，能效评估装置300可以获取完成训练的AI模型，该AI模型例如可以是基于神经网络模型进行构建的模型，如基于循环神经网络(recurrent neural network,RNN)、深度神经网络(deep neural networks，DNN)构建的模型等；或者，AI模型可以是回归树(regression tree)模型、支持向量机(support vector machine，SVM)模型等，本实施例中对于AI模型的具体实现方式并不进行限定。其中，AI模型的训练样本，例如可以是硬件对象在历史时间段内处于各个运行状态下所达到的最低功耗。关于训练AI模型的具体实现过程，可以参见后文描述，在此不做赘述。能效评估装置300在获取到硬件对象的运行状态后，可以将获取的运行状态输入至AI模型，利用AI模型根据输入的运行状态进行推理，得到AI模型输出的该运行状态对应的理想功耗。
进一步地，能效评估装置300还可以获取硬件对象在该运行状态下的节能控制参数，并将硬件对象的运行状态以及节能控制参数一并输入至AI模型中，由该AI模型推理得到该硬件对象的理想功耗。如此，基于运行状态以及节能控制参数等多个维度数据进行推理，可以进一步提高所确定的理想功耗的准确性，从而可以进一步提高后续评估硬件对象的能效水平的准确性。
在第二种可能的实施方式中，能效评估装置300中可以配置有硬件对象的运行状态与理想功耗之间的映射关系，如可以预先由技术人员将该映射关系配置于能效评估装置300等。这样，能效评估装置300在获取到硬件对象的运行状态后，通过查找该映射关系，可以确定出该运行状态对应的理想功耗。以硬件对象具体为CPU为例，硬件对象的运行状态例如可以是CPU利用率，并且，能效评估装置300中可以配置有CPU利用率与理想功耗之间的映射关系，如配置CPU利用率为10％时，理想功耗为100w；CPU利用率为50％时，理想功耗为300w等。这样，能效评估装置300可以在获取到当前的CPU利用率后，通过查找已配置的映射关系，确定该CPU利用率所对应的理想功耗。
示例性地，能效评估装置300中的映射关系，例如可以是根据该硬件对象在历史时间段内处于各个运行状态下所达到的最低功耗进行确定。以硬件对象具体为CPU为例，假设在过去的30天内，CPU在多个时刻的利用率均为10％，但是该CPU在该多个时刻中的不同时刻所产生的功耗分别为100w、150w、300w，则可以从多个功耗中确定出最低功耗(100w)，并建立该CPU利用率(10％)与最低功耗(100w)之间的映射关系。实际应用时，能效评估装置300中的映射关系也可以通过其它方式进行确定，本实施例对此并不进行限定。
上述确定理想功耗的实现方式仅作为一些示例性说明，在其它实施例中，能效评估装置300也可以采用其它方式确定件对象在运行状态下的理想功耗。
S203：能效评估装置300根据实际功耗以及理想功耗，计算硬件对象的节能健康度，该节能健康度用于衡量该硬件对象的能效水平。
可以理解，理想功耗指示了硬件对象在当前运行状态下所能达到的最低功耗，最低功耗是历史时间段内的实际功耗，即是将硬件对象在过去时间段内的运行过程中所实际达到的最低功耗作为理论值；而实际功耗为硬件对象在当前运行状态下所实际产生的功耗。因此，根据硬件对象的实际功耗与理想功耗，可以反映出硬件对象在当前运行状态下的能效水平的高低。具体的，当实际功耗与理想功耗之间的偏差较小时，表征硬件对象当前的功耗较小，处于较好的节能状态；相应的，硬件对象的能效水平当前处于较高的状态。当实际功耗与理想功耗之间的偏差较大时，通常为实际功耗远大于理想功耗，此时，硬件对象当前的功耗过高，存在较多的能量浪费；相应的，硬件对象的能效水平当前处于较低的状态。
本实施例中，可以利用节能健康度来衡量硬件对象的能效水平的高低，该节能健康度用于指示硬件对象的节能效果，也即可以用于指示硬件对象的能效水平高低。其中，节能健康度越大，表征硬件对象的能效水平越高；节能健康度越小，表征硬件对象的能效水平越低。
作为一种实现示例，能效评估装置300可以基于下述公式(1)计算得到硬件对象的节能健康度。
其中，h1为硬件对象的节能健康度；p为实际功耗；p＊为理想功耗。
或者，能效评估装置300也可以基于下述公式(2)计算得到硬件对象的节能健康度。
如此，能效评估装置300能够计算得到节能健康度，实现对硬件对象的能效水平的在线实时评估。
进一步地，能效评估装置300还可以执行以下步骤：
S204：能效评估装置300呈现该硬件对象的节能健康度。
这样，用户(如硬件对象的运维人员等)可以根据所呈现的硬件对象的节能健康度，获知硬件对象的能效水平，以便用户了解硬件对象在运行时的耗能情况。
S205：当硬件对象的节能健康度低于阈值时，能效评估装置300根据该节能健康度执行针对该硬件对象的节能操作。
可以理解，当硬件对象的节能健康度大于或者等于阈值(如85％等)，表征硬件对象当前的节能状态良好，可以不用对硬件对象执行进一步的节能操作。而当硬件对象的节能健康度小于该阈值，表针硬 件对象当前的节能状态较差，即硬件对象在运行过程中存在较多的能源浪费，此时，能效评估装置300可以根据该节能健康度执行针对该硬件对象的节能操作，以降低该硬件对象在运行过程中所产生的能耗，提高硬件对象的能效水平。
具体实现时，能效评估装置300可以获取该硬件对象的节能控制参数，并根据该节能健康度以及节能控制参数，生成新的节能控制参数，并基于该节能控制参数对硬件对象执行节能操作。
举例来说，假设硬件对象具体为CPU，该CPU的节能健康度为60％、节能控制参数为CPU频率，并且，CPU在当前运行状态下的频率为3GHz(吉赫兹)。由于CPU的节能健康度低于85％(阈值)，因此，能效评估装置300可以根据该节能健康度以及当前运行状态下的CPU频率3GHz，计算出对CPU进行降频后该CPU所要达到的频率为2GHz；最后，能效评估装置300将CPU的频率降低至所计算出的2GHz。其中，CPU运行时的频率，指示了CPU在1秒内所发生的同步脉冲数，能够决定CPU的运算速度。通常情况下，CPU频率越大，CPU的运算速度越快，相应的，CPU的能耗也就越高；反之，CPU频率越小，CPU的运算速度越慢，CPU的耗能也越低。又比如，当CPU的节能健康度为80％时，能效评估装置300可以根据该节能健康度以及当前运行状态下的CPU频率3GHz，将CPU的频率降低至2.8GHz等。
如此，能效评估装置300通过对硬件对象进行实时评估并自动执行相应的节能操作，可以避免在硬件对象运行过程中错过节能点，及时对硬件对象执行节能操作，从而可以提高针对该硬件对象的节能效果，使得硬件对象的能效水平始终保持在较高的水平。
本实施例中，图2所示的步骤之间的时序关系并不用于进行限定，比如，在其它实施例中，能效评估装置300可以同时执行步骤S204以及步骤S205，或者，能效评估装置300也可以先执行步骤S205，再执行步骤S204等，对此并不进行限定。
值得注意的是，上述实施例中是以硬件对象具体为第一层级的电子器件作为示例，对能效评估装置300评估硬件对象的能效水平的具体实现进行介绍。当硬件对象具体为第二层级的设备或者第三层级的集群时，能效评估装置300也可以基于上述类似方式，计算出设备对应的节能健康度或者集群对应的节能健康度，实现对设备或者集群的能效水平的实时评估。
或者，能效评估装置300也可以根据第一层级的电子器件对应的节能健康度，进一步计算出第二层级的设备对应的节能健康度，实现对设备的能效水平的实时评估。然后，能效评估装置300还可以根据第二层级的设备对应的节能健康度，进一步计算出第三层级的集群对应的节能健康度，实现对集群的能效水平的实时评估。
具体实现时，能效评估装置300可以基于上述图2所示实施例描述的过程，计算出第二层级的单个设备所包括的多个电子器件分别对应的节能健康度，假设单个设备包括N个电子器件(N为正整数)。举例来说，如图3所示，设备可以包括CPU、内存、硬盘、风扇、电源供应器(power supply unit，PSU)等多个电子器件，并且，在该设备运行过程中，该设备中的多个电子器件均处于运行状态并产生能耗，因此，能效评估装置300可以分别根据各个电子器件的运行状态，采用该电子器件所对应的AI模型推理出该电子器件的理想功耗，从而基于各个电子器件的实际功耗与理想功耗计算出各个电子器件的节能健康度。
然后，能效评估装置300对N个电子器件分别对应的节能健康度进行加权求和，计算得到整个设备的节能健康度，如图3所示。示例性地，能效评估装置300可以基于下述公式(3)计算出整个设备的节能健康度。
其中，h2为单个设备的节能健康度；N为该设备所包括的产生能耗的电子器件的数量；hi为第i个电子器件的节能健康度；wi为第i个电子器件对应的权重。其中，各个电子器件对应的权重，可以预先由技术人员根据实际应用的需要进行设定，如根据各个电子器件的重要程度或者耗能占比进行设定等，并将其配置于能效评估装置300中，以便能效评估装置300根据各个电子器件的权重以及节能健康度，加权求和得到整机的节能健康度。如此，根据各个设备内的电子器件的节能健康度，可以计算得到第二层级的各个设备的节能健康度，实现对各个设备的能效水平的实时评估。
并且，能效评估装置300在得到各个设备的节能健康度后，还可以对各个设备执行相应的节能操作， 以提高各个设备的能效水平。具体实现时，针对每个设备，能效评估装置300可以将该设备的节能健康度与预先设定的阈值进行比较。并且，当节能健康度大于或者等于该阈值时，能效评估装置300可以不执行节能操作。当节能健康度小于该阈值时，能效评估装置300可以根据该节能健康度或者该设备内的各个电子器件的节能健康度，对该设备内的各个电子器件执行节能操作。比如，设备内可以包括CPU、内存、硬盘、风扇、PSU等电子器件，则能效评估装置300可以对CPU执行动态电压频率调整(dynamic voltage and frequency scaling，DVFS)、休眠或者关闭处理器核等节能操作；降低内存的刷新率，如将内存的刷新率由2000MHz(兆赫兹)降低至1300MHz；降低硬盘的磁头转速，或者将硬盘的工作模式切换至休眠模式；降低风扇的转速；切换PSU的供电模式，如将PSU的供电模式由负载均衡模式切换至主备模式等。实际测试场景中，基于节能健康度对设备执行节能操作，可以使得设备的能效水平平均提升10％以上。
由于第三层级的集群，通常包括一个或者多个设备，该集群的能耗即为该集群内的一个或者多个设备所产生的能耗之和。因此，能效评估装置300还可以根据计算出的各个设备的节能健康度，进一步计算出集群的节能健康度。
示例性地，能效评估装置300可以基于下述公式(4)计算出集群的节能健康度。
其中，h3为集群的节能健康度；M为该集群所包括的产生能耗的设备的数量；hj为第j个设备的节能健康度；wj为第j个设备对应的权重。其中，各个设备对应的权重，可以预先由技术人员根据实际应用的需要进行设定，如根据各个设备的重要程度或者耗能占比进行设定等，并将其配置于能效评估装置300中，以便能效评估装置300根据各个设备的权重以及节能健康度，加权求和得到整个集群的节能健康度。如此，能效评估装置300能够实现对集群的能效水平的实时评估。
能效评估装置300在得到集群的节能健康度后，还可以对集群执行相应的节能操作，以提高集群的能效水平。比如，能效评估装置300可以关闭集群中的部分设备，或者将集群中的部分设备由运行状态调整为休眠状态等，以此降低集群的整体能耗。
上述图2所示实施例中，主要介绍了能效评估装置300确定硬件对象的理想功耗以及根据该理想功耗确定节能健康度，以实现对硬件对象的实时评估。其中，能效评估装置300可以利用预先完成训练的AI模型推理得到硬件对象在当前运行状态下的理想功耗。下面，对训练AI模型的过程进行详细介绍。其中，AI模型可以由能效评估装置300完成训练，或者可以是由其它装置训练得到AI模型后再将该AI模型提供给能效评估装置300。为便于描述，以下以能效评估装置300执行AI模型的训练过程为例进行示例性说明。
参见图4，示出了能效评估装置300训练AI模型的方法流程示意图。如图4所示，该方法包括：
S401：能效评估装置300构建AI模型。
示例性地，AI模型例如可以是基于神经网络模型进行构建的模型，如基于RNN、DNN构建的模型等；或者，AI模型可以是回归树模型、SVM模型。
在其它实施例中，AI模型也可以是由用户完成构建后，将其输入至能效评估装置300中，在此不做限定。
S402：能效评估装置300采集硬件对象的历史运行数据，该历史运行数据包括硬件对象在历史时间段内的历史运行状态、硬件对象在该历史运行状态下所产生的功耗。
进一步地，能效评估装置300所采集的历史运行数据中还可以包括硬件对象在该历史运行状态下所采用的节能控制参数，如CPU频率、内存频率、磁盘转速、风扇转速、电源供电模式等。
其中，历史时间段，即为过去的一段时间，如过去的15天、30天、180天等。
在一种可能的实施方式中，硬件对象在运行过程中，可以生成相应的日志，该日志用于记录该硬件对象在运行时的相关参数，如运行状态、运行功率(以及节能控制参数)等。则，能效评估装置300可以获取硬件对象在历史时间段内所生成的日志，并从该日志中读取到硬件对象的历史运行数据。
在又一种可能的实施方式中，能效评估装置300可以在硬件对象运行过程中，记录硬件对象运行时的相关参数，并且，当记录的数据量达到预设阈值或者当记录的时长达到预设时长时，能效评估装置300 停止记录数据，并将已记录的数据作为硬件对象的历史运行数据。
S403：能效评估装置300基于采集的历史运行数据，生成训练样本，该训练样本包括硬件对象在历史时间段内的多个历史运行状态、该多个历史运行状态中每个历史运行状态对应的历史最低功耗。
本实施例提供了以下几种生成训练样本的实现示例。
在第一种实现示例中，能效评估装置300可以对历史运行数据进行遍历，确定硬件对象在历史时间段内所处于的多个历史运行状态，并进一步确定硬件对象在每个历史运行状态下所产生的一个或者多个功耗，不同功耗对应于历史时间段内的多个时刻，如硬件对象在同一历史运行状态下在A时刻的功耗为100w，在B时刻的功耗为150w，在C时刻的功耗为300w，并且该多个能耗是由硬件对象分别在多组节能控制参数的控制下产生。然后，针对每个历史运行状态，能效评估装置300确定硬件对象在历史运行状态下所能达到的最低功耗，如能效评估装置300可以比较硬件对象在A时刻、B时刻以及C时刻的功耗，并从中确定出硬件对象在该历史运行状态下所能达到的最低功耗为100w。如此，能效评估装置300可以确定硬件对象在各个历史运行状态下分别对应的最低功耗。接着，能效评估装置300将该多个历史运行状态作为模型输入，将每个历史运行状态对应的最低功耗作为AI模型的训练标签，生成针对该AI模型的训练样本。
当能效评估装置300所获取的历史运行数据中还包括节能控制参数时，能效评估装置300在确定出各个历史运行状态对应的最低功耗后，还可以确定该最低功耗对应的节能控制参数，从而能效评估装置300将该多个历史运行状态以及每个历史运行状态对应的节能控制参数作为模型输入，将每个历史运行状态对应的最低功耗作为AI模型的训练标签，生成针对该AI模型的训练样本。
在第二种实现示例中，能效评估装置300所获取的每个历史运行数据包括历史运行状态以及节能控制参数，如下述公式(5)所示。
其中，si为第i个历史运行数据；为第i个历史运行状态；为第i个节能控制参数。
然后，能效评估装置300可以对具有相同历史运行状态的历史运行数据进行投影，得到投影状态，如下述公式(6)所示。
其中，为投影状态的历史运行数据。
接着，能效评估装置300可以基于下述公式(7)对具有相同投影状态的多个功耗进行遍历和比较，确定每个投影状态下的最低功耗，也即确定硬件对象在每个历史运行状态下所能达到的最低功耗，如图5所示。
其中，为第i个历史运行状态下所能达到的最低功耗，本实施例中，硬件对象的功耗通过该硬件对象的功率进行表征，在其它实施例中，也可以是通过电压和电流等参数进行表征，在此不做限定。
最后，能效评估装置300可以确定硬件对象在各个历史运行状态下分别对应的最低功耗，并将该多个历史运行状态作为模型输入，将每个历史运行状态对应的最低功耗作为AI模型的训练标签，生成针对该AI模型的训练样本。或者，可以将多个历史运行状态以及每个历史运行状态对应的节能控制参数作为模型输入，将每个历史运行状态对应的最低功耗作为AI模型的训练标签，生成针对该AI模型的训练样本。
本实施例中，是以能效评估装置300基于硬件对象的历史运行数据生成训练样本为例，在其它实施例中，能效评估装置300也可以基于其它数据或者基于其它方式生成训练样本，如可以基于测试数据生成训练样本，或者可以由技术人员根据经验设定硬件对象在各个运行状态下的理想功耗等。
S404：能效评估装置300利用训练样本对构建的AI模型进行训练。
具体实现时，能效评估装置300可以将训练样本中的历史运行状态(以及节能控制参数)输入至AI模型中，由该AI模型根据样本输入进行推理并输出理想功耗。然后，能效评估装置300可以将AI模型输出的理想功耗与训练样本中的最低功耗进行比较，并根据理想功耗与最低功耗之间的偏差，对AI模型中的参数进行调整，以此实现对AI模型的训练。如此，基于训练样本中的多组历史运行状态以及最低功耗的数据，实现对AI模型的多次训练，直至AI模型完成训练终止条件，如AI模型收敛或者 训练次数达到预设次数等。
如此，能效评估装置300可以基于上述步骤S401至步骤S404实现对AI模型的训练。这样，能效评估装置300可以利用该AI模型实现对硬件对象的能效水平的实时评估。
值得注意的是，当硬件对象具体为电子器件时，针对不同的电子器件，能效评估装置300可以分别训练出不同的AI模型，如针对CPU训练出AI模型1、针对内存训练出AI模型2等，以便利用不同的AI模型分别推理不同电子器件的理想功耗。
进一步地，当AI模型的训练样本基于硬件对象在历史时间段内的历史运行数据生成时，硬件对象在部分历史运行状态下的最低功耗，可能并非为该硬件对象在该部分历史运行状态下所实际能达到的最低功耗，因此，能效评估装置300在利用AI模型推理硬件对象在当前运行状态下的理想功耗时，还可以对AI模型输出的理想功耗进行修正，如下述公式(8)所示。
p＊＊＝p＊+δ     公式(8)
其中，p＊＊为修正后的理想功耗；p＊为AI模型输出的理想功耗，也即修正前的理想功耗；δ为修正量，不同运行状态对应的修正量的取值可以存在差异。
相应地，该硬件对象的节能健康度可以基于下述公式(9)进行计算得到。
其中，修正量δ可以由技术人员进行设定。
或者，修正量δ可以由能效评估装置300根据硬件对象在一段时间段内的节能健康度进行动态设定。
具体实现时，能效评估装置可以持续监测硬件对象的节能健康度，并且，当硬件对象的节能健康度大于或者等于第一阈值时，表征硬件对象的节能状态良好，可以无需执行节能操作来进一步提升硬件对象的能效水平。但是，当硬件对象的节能健康度大于或者等于第一阈值的持续时长大于第一时长时，也即硬件对象的节能健康度持续处于高节能健康度区间的时长较长时，可能是AI模型所输出的理想功耗偏高，即高于硬件对象实际所能达到最低功耗。此时，能效评估装置300可以减小修正量，以降低理想功耗的值，使得用于计算节能健康度的理想功耗的值更加接近于该硬件对象实际所能达到的最低功耗，从而提高所计算出的节能健康度的准确率。
当硬件对象的节能健康度小于第二阈值时(该第二阈值小于第一阈值)，表征硬件对象的节能状态较差，此时，能效评估装置300可以根据该节能健康度对硬件对象执行相应的节能操作，以提高硬件对象的能效水平，实现对硬件对象的进一步节能。但是，当硬件对象的节能健康度小于第二阈值的持续时长大于第二时长时，也即硬件对象的节能健康度持续处于低节能健康度区间的时长较长时，可能是AI模型所输出的理想功耗偏低，即低于硬件对象实际所能达到的最低功耗。此时，能效评估装置300可以增大修正量，以增大理想功耗的值，使得用于计算节能健康度的理想功耗的值更加接近于该硬件对象实际所能达到的最低功耗，从而提高所计算出的节能健康度的准确率。
如此，在硬件对象运行过程中，能效评估装置300持续对AI模型输出的理想功耗进行调整后，基于持续调整后的理想功耗所计算出的硬件对象的节能健康度变化曲线可以如图6所示。
实际应用场景中，能效评估装置300对于修正量的调整量大小，可以随着对于硬件对象的节能健康度的监测时长的增加而逐渐减小，直至为0，以使得能效评估装置300所确定的理想功耗的值收敛。此时，能效评估装置300基于收敛的理想功耗所计算出的节能健康度的准确率可以持续保持在较高水平，也即将针对硬件对象的能效水平的评估准确性稳定保持在较高水平。
进一步地，能效评估装置300还可以利用修正后的理想功耗、硬件对象所处的当前运行状态(以及所采用的节能控制参数)对AI模型进行更新，以此可以提高AI模型根据运行状态输出的理想功耗的准确性。这样，利用更新后的AI模型推理硬件对象在各个运行状态下的理想功耗，可以进一步提高根据该理想功耗所计算出的节能健康度的准确性，实现对硬件对象的能效水平的有效评估。
本实施例中，能效评估装置300利用硬件对象在历史时间段内的运行功耗，不仅可以训练得到用于确定硬件对象的理想功耗的AI模型，以便利用该AI模型实现对硬件对象的能效水平的实时评估，而且，也能提高AI模型所推理得到的理想功耗的可靠性以及可信度，从而能效评估装置300基于该AI模型输出的理想功耗所计算出的节能健康度，能够更加准确性的反映硬件对象的能效水平。
并且，能效评估装置300还可以进一步对理想功耗的修正量进行动态调整，可以使得所确定出的理 想功耗更加可靠，减少部分因素(如硬件对象在历史时间段内运行功耗未达到实际所能达到的最低功耗等)对于确定理想功耗的干扰，以此可以进一步提高能效评估装置300最终所计算出的节能健康度的准确率，从而可以提高衡量硬件对象的能效水平的准确性。
值得注意的是，本领域的技术人员根据以上描述的内容，能够想到的其他合理的步骤组合，也属于本申请的保护范围内。其次，本领域技术人员也应该熟悉，说明书中所描述的实施例均属于优选实施例，所涉及的动作并不一定是本申请所必须的。
以上结合图1至图6对本申请实施例提供的能效评估方法进行介绍，接下来结合附图对本申请实施例提供的能效评估装置的功能以及实现该能效评估装置的能效评估系统进行介绍。
参见图7，示出了一种能效评估装置的结构示意图，该能效评估装置700包括：
获取模块701，用于获取硬件对象的运行状态以及实际功耗，所述运行状态用于指示所述硬件对象当前时刻的运行状态；
确定模块702，用于确定所述运行状态对应的理想功耗，所述理想功耗用于指示所述硬件对象在所述运行状态下的可运行的最低功耗；
计算模块703，用于根据所述实际功耗以及所述理想功耗，计算所述硬件对象的节能健康度，所述节能健康度用于衡量所述硬件对象的能效水平。
在一种可能的实施方式中，所述确定模块702，用于根据所述硬件对象在历史时间段内的功耗数据，确定所述运行状态对应的理想功耗。
在一种可能的实施方式中，所述确定模块702，用于根据所述运行状态，利用人工智能AI模型进行推理，得到所述AI模型输出的所述运行状态对应的理想功耗，所述AI模型利用所述硬件对象在历史时间段内的功耗数据完成训练。
在一种可能的实施方式中，所述AI模型基于至少一组训练样本完成训练，所述训练样本包括所述硬件对象在所述历史时间段内的多个历史运行状态、所述多个历史运行状态中每个历史运行状态对应的最低功耗，所述最低功耗为同一历史运行状态下基于多组节能控制参数所对应的多个功耗中的最小值，所述节能控制参数用于控制所述硬件对象的能耗。
在一种可能的实施方式中，所述训练样本还包括所述最低功耗对应的节能控制参数。
在一种可能的实施方式中，所述装置700还包括：
修正模块704，用于根据所述节能健康度，对所述运行状态对应的理想功耗进行修正。
在一种可能的实施方式中，所述修正模块704，还用于：
当所述硬件对象的节能健康度大于第一阈值的持续时长大于第一时长时，减小针对所述理想功耗的修正量；
当所述硬件对象的节能健康度小于第二阈值的持续时长大于第二时长时，增大针对所述理想功耗的修正量，所述第一阈值大于所述第二阈值。
在一种可能的实施方式中，所述硬件对象为计算设备包括的多个电子器件中的一个电子器件；
所述获取模块，还用于获取所述多个电子器件的节能健康度，所述多个电子器件的节能健康度包括所述硬件对象的节能健康度；
所述计算模块，还用于根据所述多个电子器件的节能健康度，计算得到所述计算设备的节能健康度。
在一种可能的实施方式中，所述装置700还包括：
呈现模块705，用于呈现所述硬件对象的节能健康度；
或，执行模块706，用于当所述节能健康度低于阈值时，根据所述节能健康度执行针对所述硬件对象的节能操作。
由于图7所示的能效评估装置700对应于图4所示的方法，故图7所示的能效评估装置700的具体实现方式及其所具有的技术效果，可以参见前述实施例中的相关之处描述，在此不做赘述。
图8为本申请提供的一种能效评估系统800的示意图。图8所示的能效评估系统800能够用于实现图2所示实施例中的能效评估装置300所执行的方法步骤。实际应用时，能效评估系统800例如可以是可独立运行的卡、服务器、服务器中处理器等，本实施例对此并不进行限定。为便于理解，下面以能效评估系统800为服务器为例介绍能效评估系统800的硬件结构。
如图8所示，所述能效评估系统800包括处理器801、存储器802、通信接口803。其中，处理器801、存储器802、通信接口803通过总线804进行通信，也可以通过无线传输等其他手段实现通信。该存储器802用于存储指令，该处理器801用于执行该存储器802存储的指令。进一步的，能效评估系统800还可以包括内存单元805，还内存单元805可以通过总线804与处理器801、存储介质802以及通信接口803连接。其中，该存储器802存储程序代码，且处理器801可以调用存储器802中存储的程序代码执行以下操作：
获取硬件对象的运行状态以及实际功耗，所述运行状态用于指示所述硬件对象当前时刻的运行状态；
确定所述运行状态对应的理想功耗，所述理想功耗用于指示所述硬件对象在所述运行状态下的可运行的最低功耗；
根据所述实际功耗以及所述理想功耗，计算所述硬件对象的节能健康度，所述节能健康度用于衡量所述硬件对象的能效水平。
应理解，在本申请实施例中，该处理器801可以是CPU，该处理器801还可以是其他通用处理器、数字信号处理器(DSP)、专用集成电路(ASIC)、现场可编程门阵列(FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立器件组件等。通用处理器可以是微处理器或者是任何常规的处理器等。
该存储器802可以包括只读存储器和随机存取存储器，并向处理器801提供指令和数据。存储器802还可以包括非易失性随机存取存储器。例如，存储器802还可以存储设备类型的信息。
该存储器802可以是易失性存储器或非易失性存储器，或可包括易失性和非易失性存储器两者。其中，非易失性存储器可以是只读存储器(read-only memory，ROM)、可编程只读存储器(programmable ROM，PROM)、可擦除可编程只读存储器(erasable PROM，EPROM)、电可擦除可编程只读存储器(electrically EPROM，EEPROM)或闪存。易失性存储器可以是随机存取存储器(random access memory，RAM)，其用作外部高速缓存。通过示例性但不是限制性说明，许多形式的RAM可用，例如静态随机存取存储器(static RAM，SRAM)、动态随机存取存储器(DRAM)、同步动态随机存取存储器(synchronous DRAM，SDRAM)、双倍数据速率同步动态随机存取存储器(double data date SDRAM,DDR SDRAM)、增强型同步动态随机存取存储器(enhanced SDRAM，ESDRAM)、同步连接动态随机存取存储器(synchlink DRAM，SLDRAM)和直接内存总线随机存取存储器(direct rambus RAM，DR RAM)。
该通信接口803用于与能效评估系统800连接的其它设备进行通信。该总线804除包括数据总线之外，还可以包括电源总线、控制总线和状态信号总线等。但是为了清楚说明起见，在图中将各种总线都标为总线804。
应理解，根据本申请实施例的能效评估系统800可对应于本申请实施例中的能效评估装置700，并可以对应于执行根据本申请实施例中图2所示方法中的能效评估装置300，并且能效评估系统800所实现的上述和其它操作和/或功能分别为了实现图2中方法的相应流程，为了简洁，在此不再赘述。
本申请实施例还提供了一种计算机可读存储介质。所述计算机可读存储介质可以是计算设备能够存储的任何可用介质或者是包含一个或多个可用介质的数据中心等数据存储设备。所述可用介质可以是磁性介质，(例如，软盘、硬盘、磁带)、光介质(例如，DVD)、或者半导体介质(例如固态硬盘)等。该计算机可读存储介质包括指令，所述指令指示计算设备执行上述能效评估方法。
本申请实施例还提供了一种计算机程序产品。所述计算机程序产品包括一个或多个计算机指令。在计算设备上加载和执行所述计算机指令时，全部或部分地产生按照本申请实施例所述的流程或功能。
所述计算机指令可以存储在计算机可读存储介质中，或者从一个计算机可读存储介质向另一计算机可读存储介质传输，例如，所述计算机指令可以从一个网站站点、计算机或数据中心通过有线(例如同轴电缆、光纤、数字用户线(DSL))或无线(例如红外、无线、微波等)方式向另一个网站站点、计算机或数据中心进行传输。
所述计算机程序产品可以为一个软件安装包，在需要使用前述能效评估方法的任一方法的情况下，可以下载该计算机程序产品并在计算设备上执行该计算机程序产品。
上述实施例，可以全部或部分地通过软件、硬件、固件或其他任意组合来实现。当使用软件实现时，上述实施例可以全部或部分地以计算机程序产品的形式实现。所述计算机程序产品包括一个或多个计算 机指令。在计算机上加载或执行所述计算机程序指令时，全部或部分地产生按照本申请实施例所述的流程或功能。所述计算机可以为通用计算机、专用计算机、计算机网络、或者其他可编程装置。所述计算机指令可以存储在计算机可读存储介质中，或者从一个计算机可读存储介质向另一个计算机可读存储介质传输，例如，所述计算机指令可以从一个网站站点、计算机、服务器或数据中心通过有线(例如同轴电缆、光纤、数字用户线(DSL))或无线(例如红外、无线、微波等)方式向另一个网站站点、计算机、服务器或数据中心进行传输。所述计算机可读存储介质可以是计算机能够存取的任何可用介质或者是包含一个或多个可用介质集合的服务器、数据中心等数据存储设备。所述可用介质可以是磁性介质(例如，软盘、硬盘、磁带)、光介质(例如，DVD)、或者半导体介质。半导体介质可以是固态硬盘。
以上所述，仅为本申请的具体实施方式，但本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，可轻易想到各种等效的修改或替换，这些修改或替换都应涵盖在本申请的保护范围之内。因此，本申请的保护范围应以权利要求的保护范围为准。

Claims (12)

  1. 一种能效评估方法，其特征在于，所述方法包括：
    获取硬件对象的运行状态以及实际功耗，所述运行状态用于指示所述硬件对象当前时刻的运行状态；
    确定所述运行状态对应的理想功耗，所述理想功耗用于指示所述硬件对象在所述运行状态下的可运行的最低功耗；
    根据所述实际功耗以及所述理想功耗，计算所述硬件对象的节能健康度，所述节能健康度用于衡量所述硬件对象的能效水平。
  2. 根据权利要求1所述的方法，其特征在于，所述确定所述运行状态对应的理想功耗，包括：
    根据所述硬件对象在历史时间段内的功耗数据，确定所述运行状态对应的理想功耗。
  3. 根据权利要求2所述的方法，其特征在于，所述根据所述硬件对象在历史时间段内的功耗数据，确定所述运行状态对应的理想功耗，包括：
    根据所述运行状态，利用人工智能AI模型进行推理，得到所述AI模型输出的所述运行状态对应的理想功耗，所述AI模型利用所述硬件对象在历史时间段内的功耗数据完成训练。
  4. 根据权利要求3所述的方法，其特征在于，所述AI模型基于至少一组训练样本完成训练，所述训练样本包括所述硬件对象在所述历史时间段内的多个历史运行状态、所述多个历史运行状态中每个历史运行状态对应的最低功耗，所述最低功耗为同一历史运行状态下基于多组节能控制参数所对应的多个功耗中的最小值，所述节能控制参数用于控制所述硬件对象的能耗。
  5. 根据权利要求4所述的方法，其特征在于，所述训练样本还包括所述最低功耗对应的节能控制参数。
  6. 根据权利要求1至5任一项所述的方法，其特征在于，所述方法还包括：
    根据所述节能健康度，对所述运行状态对应的理想功耗进行修正。
  7. 根据权利要求6所述的方法，其特征在于，所述方法还包括：
    当所述硬件对象的节能健康度大于第一阈值的持续时长大于第一时长时，减小针对所述理想功耗的修正量；
    当所述硬件对象的节能健康度小于第二阈值的持续时长大于第二时长时，增大针对所述理想功耗的修正量，所述第一阈值大于所述第二阈值。
  8. 根据权利要求1至7任一项所述的方法，其特征在于，所述硬件对象为计算设备包括的多个电子器件中的一个电子器件，所述方法还包括：
    获取所述多个电子器件的节能健康度，所述多个电子器件的节能健康度包括所述硬件对象的节能健康度；
    根据所述多个电子器件的节能健康度，计算得到所述计算设备的节能健康度。
  9. 根据权利要求1至8任一项所述的方法，其特征在于，所述方法还包括：
    呈现所述硬件对象的节能健康度；
    或，当所述节能健康度低于阈值时，根据所述节能健康度执行针对所述硬件对象的节能操作。
  10. 一种能效评估装置，其特征在于，所述能效评估装置包括：
    获取模块，用于获取硬件对象的运行状态以及实际功耗，所述运行状态用于指示所述硬件对象当前时刻的运行状态；
    确定模块，用于确定所述运行状态对应的理想功耗，所述理想功耗用于指示所述硬件对象在所述运行状态下的可运行的最低功耗；
    计算模块，用于根据所述实际功耗以及所述理想功耗，计算所述硬件对象的节能健康度，所述节能健康度用于衡量所述硬件对象的能效水平。
  11. 一种能效评估系统，其特征在于，包括处理器、存储器；
    所述处理器用于执行所述存储器中存储的指令，以使所述能效评估系统执行如权利要求1至9任一项所述方法的步骤。
  12. 一种计算机可读存储介质，其特征在于，包括指令，当其在计算设备上运行时，使得所述计算设备执行如权利要求1至9中任一项所述方法的步骤。
PCT/CN2023/143424 2022-11-11 2023-12-29 能效评估方法、装置、系统及相关设备 WO2024099474A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211412374.5 2022-11-11
CN202211412374.5A CN118035051A (zh) 2022-11-11 2022-11-11 能效评估方法、装置、系统及相关设备

Publications (1)

Publication Number Publication Date
WO2024099474A1 true WO2024099474A1 (zh) 2024-05-16

Family

ID=90993926

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/143424 WO2024099474A1 (zh) 2022-11-11 2023-12-29 能效评估方法、装置、系统及相关设备

Country Status (2)

Country Link
CN (1) CN118035051A (zh)
WO (1) WO2024099474A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118467296A (zh) * 2024-05-21 2024-08-09 深圳市奥盛通科技有限公司 一种加固计算机的智能控制功耗方法及装置
CN118519784A (zh) * 2024-07-22 2024-08-20 超云数字技术集团有限公司 服务器功耗均衡处理方法、装置及电子设备
CN118959349A (zh) * 2024-10-12 2024-11-15 广东鑫风风机有限公司 一种多叶离心风机的能耗评估方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2007018322A (ja) * 2005-07-08 2007-01-25 Yamatake Corp 省エネルギー量推定装置、方法、およびプログラム
CN103092744A (zh) * 2012-05-09 2013-05-08 华南理工大学 一种动态电源管理策略性能评估方法
CN109840180A (zh) * 2018-12-18 2019-06-04 平安科技（深圳）有限公司 服务器运行功耗管理方法、装置及计算机可读存储介质
CN113448249A (zh) * 2021-06-28 2021-09-28 南京天溯自动化控制系统有限公司 一种能源系统节能诊断方法、装置、存储介质及电子设备
CN114237181A (zh) * 2021-12-17 2022-03-25 广东德尔智慧工厂科技有限公司 多车间健康监测系统及方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2007018322A (ja) * 2005-07-08 2007-01-25 Yamatake Corp 省エネルギー量推定装置、方法、およびプログラム
CN103092744A (zh) * 2012-05-09 2013-05-08 华南理工大学 一种动态电源管理策略性能评估方法
CN109840180A (zh) * 2018-12-18 2019-06-04 平安科技（深圳）有限公司 服务器运行功耗管理方法、装置及计算机可读存储介质
CN113448249A (zh) * 2021-06-28 2021-09-28 南京天溯自动化控制系统有限公司 一种能源系统节能诊断方法、装置、存储介质及电子设备
CN114237181A (zh) * 2021-12-17 2022-03-25 广东德尔智慧工厂科技有限公司 多车间健康监测系统及方法

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118467296A (zh) * 2024-05-21 2024-08-09 深圳市奥盛通科技有限公司 一种加固计算机的智能控制功耗方法及装置
CN118519784A (zh) * 2024-07-22 2024-08-20 超云数字技术集团有限公司 服务器功耗均衡处理方法、装置及电子设备
CN118519784B (zh) * 2024-07-22 2024-11-15 超云数字技术集团有限公司 服务器功耗均衡处理方法、装置及电子设备
CN118959349A (zh) * 2024-10-12 2024-11-15 广东鑫风风机有限公司 一种多叶离心风机的能耗评估方法及系统

Also Published As

Publication number Publication date
CN118035051A (zh) 2024-05-14

Similar Documents

Publication Publication Date Title
WO2024099474A1 (zh) 能效评估方法、装置、系统及相关设备
CN107678855B (zh) 处理器动态调节方法、装置及处理器芯片
CN103201702B (zh) 对计算工作负载进行管理的方法和系统
US9927857B2 (en) Profiling a job power and energy consumption for a data processing system
US9460483B2 (en) Methods, systems and apparatus to manage power consumption of a graphics engine
CN104798034B (zh) 用于协调频率的方法、处理器以及系统
WO2020172818A1 (zh) 一种动态电压频率调整系统和方法
WO2021073130A1 (zh) 处理器的调频方法及装置、计算设备
US20210234890A1 (en) Predictive rate limiting system for cloud computing services
Dey et al. User interaction aware reinforcement learning for power and thermal efficiency of CPU-GPU mobile MPSoCs
CN110221771B (zh) Ssd存储系统中自我调节功率使用和功耗的方法和装置
CN104049716A (zh) 一种结合温度感知的计算机节能方法及系统
WO2022052626A1 (zh) 功耗管理的方法和相关设备
US20160363976A1 (en) Method and Apparatus for Demand Estimation for Energy Management of Client Systems
CN113852135B (zh) 一种虚拟电厂能源调度方法、装置、存储介质及平台
US7925873B2 (en) Method and apparatus for controlling operating parameters in a computer system
CN117389344B (zh) 用电设备及其表面温度控制方法、装置、介质和芯片系统
CN115167981A (zh) 一种云网负载预测系统、方法、电子设备及存储介质
CN104956345B (zh) 基于响应定时优化在多处理器系统中执行频率协调
CN116680131A (zh) 服务器测试系统、服务器测试方法、电子设备及存储介质
WO2022236782A1 (zh) 一种功耗调节方法及装置
WO2024139467A1 (zh) 调频方法、装置、处理器、芯片和计算机设备
CN110928683A (zh) 基于两类密集型虚拟机的边缘计算资源分配方法
US11720464B1 (en) System and method for reduction of data transmission by threshold adaptation
CN116893865B (zh) 微服务实例调整方法、装置、电子设备及可读存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23888159

Country of ref document: EP

Kind code of ref document: A1