Power tuning method and related device

Abstract

The invention discloses a power adjusting and optimizing method and a related device, processor fault data of to-be-optimized equipment is obtained, the to-be-optimized equipment comprises a processor used for executing data processing business, and the processor fault data of the to-be-optimized equipment is analyzed and counted to obtain the fault probability of the to-be-optimized equipment. The power overload proportion represents the percentage that the transient power of the processor can exceed the rated power, and the transient power overshoot or too low of the processor can be suppressed by adjusting the power overload proportion, so that the current power overload proportion of the to-be-optimized equipment is adjusted based on the fault probability to obtain the target power overload proportion, and then the target power overload proportion of the to-be-optimized equipment is obtained. And controlling and adjusting the transient power of the processor according to the target power overload proportion so as to protect the processor in an expected power range. Transient power tuning of the processor is realized by adjusting the power overload proportion, and the average power does not need to be forcibly limited, so that the transient power tuning is realized on the premise that the performance and the computing power of the processor for executing data processing services are not influenced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computers, in particular to a power tuning method and related device. BACKGROUND

[0002] With the rapid development of information technology, the demand for data processing is growing exponentially. This trend of data growth puts higher requirements on the performance of processors, especially when dealing with large-scale data sets for complex model training, real-time analysis and complex computing tasks.

[0003] The power level has a direct impact on the performance of the processor. Within a certain range, increasing the power of the processor can improve its performance, but if the power is too high, not only does it increase energy costs, but it can also cause the processor to overheat, reduce performance, and even cause the processor to fail. Therefore, how to reduce power consumption while ensuring the performance of the processor has become a problem that needs to be solved.

[0004] Currently, the power of the processor is mainly optimized by forcibly limiting the average power. Taking a graphics processing unit (GPU) as an example, the power of the GPU is judged in real time to determine whether it is too high, and a power brake signal is triggered to reduce the frequency of the GPU when the power is too high. However, this method of forcibly limiting the average power may affect the average performance and computing power of the processor as a whole. SUMMARY

[0005] To solve the above technical problems, the present application provides a power tuning method and related device, which adjusts the power overload ratio to realize transient power tuning of the processor without forcibly limiting the average power, thereby realizing transient power tuning without affecting the performance and computing power of the processor when executing data processing services.

[0006] The present application embodiment discloses the following technical solutions:

[0007] In one aspect, the present application provides a power tuning method, which comprises:

[0008] Obtaining processor failure data of a device to be optimized, the device to be optimized including a processor for executing data processing services;

[0009] Analyzing the processor failure data of the device to be optimized to obtain a failure probability of the device to be optimized;

[0010] Adjusting the current power overload ratio of the device to be optimized based on the failure probability to obtain a target power overload ratio, the target power overload ratio being used to control and adjust the transient power of the processor.

[0011] In one aspect, the embodiments of the present application provide a power tuning method, which comprises:

[0012] obtaining a target power overload ratio of a device to be optimized, the device to be optimized comprising a processor for performing data processing services, the target power overload ratio being obtained by adjusting a current power overload ratio of the device to be optimized based on a failure probability, the failure probability being obtained by analyzing processor failure data of the device to be optimized;

[0013] calculating a transient power peak of the processor by using the target power overload ratio;

[0014] controlling adjustment of transient power of the processor based on the transient power peak.

[0015] In one aspect, the embodiments of the present application provide a power tuning device, which comprises an obtaining unit, an analyzing unit and an adjusting unit:

[0016] The obtaining unit is configured to obtain processor failure data of a device to be optimized, the device to be optimized comprising a processor for performing data processing services.

[0017] The analyzing unit is configured to analyze the processor failure data of the device to be optimized, and obtain a failure probability of the device to be optimized by statistics.

[0018] The adjusting unit is configured to adjust a current power overload ratio of the device to be optimized based on the failure probability, and obtain a target power overload ratio, the target power overload ratio being used for controlling adjustment of transient power of the processor.

[0019] In one aspect, the embodiments of the present application provide a power tuning device, which comprises an obtaining unit, a calculating unit and an adjusting unit:

[0020] The obtaining unit is configured to obtain a target power overload ratio of a device to be optimized, the device to be optimized comprising a processor for performing data processing services, the target power overload ratio being obtained by adjusting a current power overload ratio of the device to be optimized based on a failure probability, the failure probability being obtained by analyzing processor failure data of the device to be optimized.

[0021] The calculating unit is configured to calculate a transient power peak of the processor by using the target power overload ratio.

[0022] The adjusting unit is configured to control adjustment of transient power of the processor based on the transient power peak.

[0023] In an aspect, an embodiment of the present application provides a computer device, comprising a processor and a memory:

[0024] The memory is configured to store a computer program and transmit the computer program to the processor.

[0025] The processor is configured to execute the method according to the instructions in the computer program.

[0026] In an aspect, an embodiment of the present application provides a computer readable storage medium configured to store a computer program, the computer program, when executed by a processor, causing the processor to execute the method according to any one of the preceding aspects.

[0027] In an aspect, an embodiment of the present application provides a computer program product comprising a computer program, the computer program, when executed by a processor, implementing the method according to any one of the preceding aspects.

[0028] From the above technical solution, it can be seen that the present application adjusts the power overload ratio to optimize the transient power of the processor on the device to be optimized. Specifically, when power optimization is needed, the processor fault data of the device to be optimized can be obtained. The device to be optimized includes a processor for executing data processing services. The processor fault data can reflect the fault occurrence of the processor on the device to be optimized. Then, the processor fault data of the device to be optimized can be analyzed to obtain the fault probability. The fault probability can reflect the probability of the processor on the device to be optimized failing due to the transient power being too high up to the current time, and then reflect how to adjust the power overload ratio. The power overload ratio represents the percentage of the transient power of the processor that can exceed the rated power. By adjusting the power overload ratio, the transient power of the processor can be suppressed from overshooting or being too low. Thus, after adjusting the current power overload ratio of the device to be optimized based on the fault probability to obtain a target power overload ratio, since the target power overload ratio is the optimized power overload ratio, it can be reflected that the peak value that the transient power of the processor can reach is allowed under the condition of ensuring the performance of the processor. Therefore, the transient power of the processor can be controlled and adjusted according to the target power overload ratio to protect the processor within the expected power range. The present application adjusts the power overload ratio to optimize the transient power of the processor, without forcibly limiting the average power, so as to realize the transient power optimization without affecting the performance and computing power of the processor in executing data processing services. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced. Obviously, the accompanying drawings in the following description only only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0030] Figure 1 The application scenario architecture diagram of the power tuning method provided by the embodiment of the present application;

[0031] Figure 2 The flow chart of the power tuning method provided by the embodiment of the present application;

[0032] Figure 3 The overall flowchart of the power tuning method provided by the embodiment of the present application;

[0033] Figure 4 The flow chart of another power tuning method provided by the embodiment of the present application;

[0034] Figure 5 The flow chart of the setting of the target power overload ratio provided by the embodiment of the present application;

[0035] Figure 6 The flow chart of another power tuning method provided by the embodiment of the present application;

[0036] Figure 7 The example diagram of the adjustment of the transient power of the GPU provided by the embodiment of the present application;

[0037] Figure 8 The example diagram of the adjustment of the transient power of the GPU provided by the related art;

[0038] Figure 9 The structural diagram of the power tuning device provided by the embodiment of the present application;

[0039] Figure 10 The structural diagram of another power tuning device provided by the embodiment of the present application;

[0040] Figure 11 The structural diagram of the terminal provided by the embodiment of the present application;

[0041] Figure 12 The structural diagram of the server provided by the embodiment of the present application. DETAILED DESCRIPTION

[0042] The embodiments of the present application will be described below with reference to the accompanying drawings.

[0043] In optimizing the power of a processor, the power of the processor is currently mainly optimized by forcibly limiting the average power. Taking a processor as an example, the power of a GPU is judged in real time to determine whether the power of the GPU is too high, and the GPU is triggered to reduce the frequency when the power of the GPU is too high. However, this method of forcibly limiting the average power may affect the average performance and computing power of the processor as a whole.

[0044] For example, the processor is a GPU on a server in artificial intelligence (AI), and the GPU performs a data processing service. If the average power of the GPU is limited during the execution of the data processing service by the GPU, the power of the GPU during the entire process will be relatively low, thereby affecting the average performance and computing power of the GPU as a whole.

[0045] To solve the above technical problems, the embodiment of the present application provides a power optimization method, which adjusts the current power overload ratio of the to-be-optimized device based on the failure probability of the to-be-optimized device, and obtains a target power overload ratio. Since the target power overload ratio is the optimized power overload ratio, it can reflect the peak value of the transient power of the processor allowed under the condition of ensuring the performance of the processor, so the transient power of the processor can be controlled and adjusted according to the target power overload ratio to protect the processor within the expected power range. Thus, without forcibly limiting the average power, the transient power optimization can be achieved without affecting the performance and computing power of the processor during the execution of the data processing service.

[0046] It should be noted that the power optimization method provided by the embodiment of the present application can be applied to various products using processors, such as cloud computing products, mixed product large model training and inference, which are not limited by the embodiment of the present application. In cloud computing products, the AI computing power cluster optimizes the power of the GPU by the method provided by the embodiment of the present application, which can eliminate GPU high-load concurrent failures and support cloud computing to provide stable cloud products for multiple customers and multiple clusters. In mixed product large model training and inference, the AI computing power cluster optimizes the power of the GPU by the method provided by the embodiment of the present application, and the GPU cluster optimized by the method has zero new high-load GPU card drop failure cases. The optimized cluster is expected to reduce the failure rate by 0.36%, which significantly improves the continuous training and efficiency of the AI cluster.

[0047] The power optimization method provided by the embodiment of the present application can be executed by a computer device, which can be a server or a terminal. The server can be a standalone physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal can be a smartphone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, a vehicle terminal, a smart television, etc., but is not limited thereto.

[0048] As Figure 1 shown, Figure 1 An application scenario architecture diagram of a power tuning method is shown, which is introduced as a server.

[0049] The application scenario can include a server 100 and a device to be optimized 200. The device to be optimized 200 is a device in which a processor that can be power tuned is located, and the device to be optimized 200 can perform data processing services through the processor included therein. In the process of the processor performing data processing services, the power level directly affects the performance of the processor. Within a certain range, increasing the power of the processor can improve its performance, but if the power is too high, not only does it increase the energy cost, but it can also cause the processor to overheat, performance to decline, and even processor failure. Therefore, the power of the processor needs to be tuned by the power tuning method provided in the embodiments of the present application. The server 100 can be used to execute the power tuning method provided in the embodiments of the present application.

[0050] The embodiments of the present application mainly tune the transient power of the processor on the device to be optimized by adjusting the power overload ratio. Specifically, when power tuning is needed, the server 100 can obtain processor failure data of the device to be optimized 200. The processor failure data can reflect the occurrence of processor failure on the device to be optimized 200, and then the processor failure data of the device to be optimized 200 can be analyzed to obtain the failure probability of the device to be optimized 200.

[0051] The failure probability can reflect the probability of processor failure on the device to be optimized 200 due to transient power overload up to the current time, and then reflect how to adjust the power overload ratio. The power overload ratio represents the percentage of the transient power of the processor that can exceed the rated power. By adjusting the power overload ratio, the transient power of the processor can be suppressed from overshooting or being too low. In this way, after the server 100 adjusts the current power overload ratio of the device to be optimized 200 based on the failure probability to obtain a target power overload ratio, since the target power overload ratio is the tuned power overload ratio, it can be reflected in the peak value that the transient power of the processor can reach while ensuring the performance of the processor. Therefore, the transient power of the processor can be controlled and adjusted according to the target power overload ratio to protect the processor within the expected power range. The present application realizes the transient power tuning of the processor by adjusting the power overload ratio, without forcibly limiting the average power, so as to realize the transient power tuning without affecting the performance and computing power of the processor in executing data processing services.

[0052] It should be noted that the method provided by the embodiments of the present application can involve artificial intelligence technology. Artificial intelligence is the use of digital computers or computer-controlled machines to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making.

[0053] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software technologies. Artificial intelligence basic technologies generally include, such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, pre-training model technology, operation / interaction system, mechatronics, etc. Among them, the pre-training model is also called large model, basic model, which can be widely applied to downstream tasks in various directions of artificial intelligence after fine-tuning. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology and machine learning / deep learning, etc. For example, the method provided by the embodiments of the present application can be applied to power optimization of a processor using artificial intelligence technology.

[0054] The method provided by the embodiments of the present application can also involve cloud technology. Cloud technology refers to the integration of a series of resources such as hardware, software and network in a wide area network or local area network to realize a kind of hosting technology for data calculation, storage, processing and sharing. Specifically, it can involve cloud computing in cloud technology. The method provided by the embodiments of the present application can be applied to cloud computing products. Cloud computing is a computing mode that distributes computing tasks on a resource pool composed of a large number of computers, so that various application systems can obtain computing power, storage space and information services according to needs. The network providing resources is called "cloud". The resources in the "cloud" can be infinitely expanded for users, and can be obtained at any time, used on demand, expanded at any time, and paid according to use. Cloud computing is the product of the development and integration of traditional computer and network technologies such as grid computing, distributed computing, parallel computing, utility computing, network storage, virtualization, load balancing, etc.

[0055] With the development of the Internet, real-time data flow, and the diversification of connected devices, and the demand for search services, social networks, mobile commerce, and open collaboration, cloud computing has rapidly developed. Unlike previous parallel distributed computing, the emergence of cloud computing will revolutionize the entire Internet model and enterprise management model from a conceptual point of view.

[0056] It should be noted that in the specific embodiments of the present application, user information and other related data may be involved throughout the process. When the above embodiments of the present application are applied to specific products or technologies, the user's individual consent or individual permission is required, and the collection, use, and processing of related data need to comply with relevant laws, regulations, and standards of the relevant countries and regions.

[0057] Next, taking a server as an example, the power tuning method provided by the embodiments of the present application will be introduced in combination with the accompanying drawings. Referring to Figure 2 , Figure 2 A flowchart of a power tuning method is shown, which can include S201-S203, as shown below:

[0058] S201, obtain processor fault data of a to-be-optimized device, the to-be-optimized device including a processor for executing a data processing service.

[0059] The to-be-optimized device is a device in which a processor that can be power tuned is located. The to-be-optimized device can execute a data processing service through the processor included therein. The embodiments of the present application do not limit the type of to-be-optimized device, which can be, for example, a server, a terminal, a server cluster, an AI cluster (AI cluster refers to an artificial intelligence server cluster), etc. The processor can be various performance processors susceptible to power, such as GPU, Central Processing Units (CPU), etc.

[0060] During the execution of the data processing service by the processor, the power level has a direct impact on the performance of the processor. Therefore, the power of the processor needs to be tuned by the power tuning method provided by the embodiments of the present application. The data processing service can be a service required to be executed by the processor, and can be different according to the application scenario of the to-be-optimized device where the processor is located. For example, the to-be-optimized device is an AI cluster, and the application scenario of the to-be-optimized device is a large model training and inference scenario, and at this time the data processing service can be the service of large model training and inference. For another example, the to-be-optimized device is a server cluster, and the application scenario of the to-be-optimized device is a cloud computing scenario, and at this time the data processing service can be the service of cloud computing.

[0061] It can be understood that the processor power is too high can cause processor failure, the processor failure frequency, can be considered that the processor is too high, otherwise, it can be considered that the power of the processor is moderate or low. Therefore, in the embodiment of the application, in order to determine whether the power of the processor is too high, the processor failure data of the to-be-optimized device can be acquired first. It should be noted that the power of the processor involved in the embodiment of the application is the instantaneous power of the processor.

[0062] The processor failure data can reflect the failure occurrence of the processor on the to-be-optimized device, for example, the processor on the to-be-optimized device works normally, the processor on the to-be-optimized device is disconnected from power supply and appears the failure card drop phenomenon due to various reasons, whether power failure, etc. The reasons for the processor on the to-be-optimized device being disconnected from power supply can be various, for example, the use of over current protection (OCP) mechanism, the use of over temperature protection (OT) mechanism, etc. The OCP can refer to that the current exceeds the protection current point, and the processor (for example, GPU) triggers the OCP and disconnects the processor power supply; the OT can refer to that the current exceeds the protection temperature point, and the processor triggers the OT and disconnects the processor power supply. Whether power failure can be judged by power state (Power Good, PG), and the PG can refer to that after the GPU power failure, the field programmable logic gate array (FPGA) on the processor acquires the power state of the processor to identify whether the processor is powered off.

[0063] S202, the processor failure data of the to-be-optimized device is parsed, and the failure probability of the to-be-optimized device is counted.

[0064] The processor failure data can reflect the failure occurrence of the processor on the to-be-optimized device, and then the processor failure data of the to-be-optimized device can be analyzed to obtain the failure probability of the to-be-optimized device. The failure probability can reflect the probability that the processor on the to-be-optimized device fails due to the transient power being too high until the current time, and then reflect how to adjust the power overload ratio. The power overload ratio can be the ratio of the transient power of the processor to the average power of the processor, and can represent the percentage of the transient power of the processor that can exceed the rated power. By adjusting the power overload ratio, the transient power overshoot or undershoot of the processor can be inhibited. The power overload ratio can be represented by EDPp percent, and EDPp (Electrical Design Point Peak transient currunt) is an electrical parameter of the processor to ensure the performance and reliability of the product. EDPp refers to the peak transient current maintained by the processor. The transient power of the processor can be as high as 1.4-1.6 times the rated power. Percent represents a percentage, so EDPp percent can represent the power overload ratio.

[0065] The embodiments of the present application rely on the processor failure data of the to-be-optimized device, and provide an automatic failure analysis function for the to-be-optimized device, so that the failure probability of the to-be-optimized device can be obtained, basic technical support is provided for adjusting EDPp percent based on the failure probability, and the power tuning method provided by the embodiments of the present application becomes possible.

[0066] It should be noted that the obtained processor failure data can include various failure conditions of the processor on the to-be-optimized device, some of which are caused by high power, and some of which can be caused by other reasons and are unrelated to power. In order to achieve power tuning, the failure probability obtained by the embodiments of the present application should be the occurrence probability of the power-related failure. Therefore, in order to obtain an accurate failure probability, in one possible implementation, the processor failure data of the to-be-optimized device can be analyzed to obtain the failure probability of the to-be-optimized device. The way to obtain the failure probability of the to-be-optimized device can be to analyze the processor failure data of the to-be-optimized device, obtain target type processor failure data from the processor failure data of the to-be-optimized device, and then based on the target type processor failure data, obtain the failure probability of the to-be-optimized device.

[0067] The target type processor failure data can be related data of power-related failures, such as failures caused by OCP mechanisms, failures caused by OT mechanisms, and failures caused by PG. Therefore, the failure probability obtained by the statistics is the occurrence probability of the power-related failure.

[0068] The processor fault data of the target type is obtained from the processor fault data, and then the occurrence probability of the power-related fault is obtained based on the processor fault data of the target type, so that a more accurate fault probability is obtained, and power tuning is more accurately performed to ensure the performance of the to-be-optimized device.

[0069] In S203, the current power overload ratio of the to-be-optimized device is adjusted based on the fault probability to obtain a target power overload ratio, and the target power overload ratio is used to control adjustment of the transient power of the processor.

[0070] The fault probability can reflect the probability that the processor on the to-be-optimized device fails due to the transient power being too high by the current time, and then reflect how to adjust the power overload ratio. The power overload ratio represents the percentage of the transient power of the processor that can exceed the rated power, and the transient power of the processor can be suppressed from overshooting or being too low by adjusting the power overload ratio. The power overload ratio of the to-be-optimized device at the current time can be referred to as the current power overload ratio, so the server can adjust the current power overload ratio of the to-be-optimized device based on the fault probability to obtain an adjusted power overload ratio, thereby changing the power overload ratio of the to-be-optimized device. The adjusted power overload ratio can be referred to as the target power overload ratio.

[0071] When adjusting the current power overload ratio, the adjustment method for the current power overload ratio can be different according to the size of the fault probability. The adjustment method can include lowering processing and rising processing, and a corresponding adjustment method is selected according to the size of the fault probability. Based on this, the way to adjust the current power overload ratio of the to-be-optimized device based on the fault probability to obtain the target power overload ratio can be to judge the size of the fault probability. If it is determined that the fault probability is greater than a first probability threshold, it can be considered that the fault caused by the transient power being too high affects the performance of the to-be-optimized device, and the current power overload ratio needs to be lowered to obtain the target power overload ratio, so that the transient power of the processor can be lowered according to the lowered target power overload ratio in the future.

[0072] The first probability threshold can be a fault probability value that affects the performance of the to-be-optimized device, and can be set according to actual business needs, for example, the first probability threshold can be 0.3%. The adjustment ratio value to be lowered during the lowering processing can also be set according to actual business needs, for example, the adjustment ratio value to be lowered can be 5%. Therefore, when the fault probability is greater than 0.3%, the current power overload ratio can be lowered by 5% to obtain the target power overload ratio.

[0073] The embodiment of the present application reduces the current power overload ratio when the failure probability is relatively high, thereby changing the current power overload ratio, so as to reduce the transient power of the processor by using the reduced power overload ratio (i.e., the target power overload ratio), thereby reducing the failure probability of the device to be optimized, and improving the performance of the device to be optimized.

[0074] It should be noted that when the current power overload ratio is reduced, the value of the failure probability is in different levels, and the degree of reducing the current power overload ratio can also be different. Based on this, if it is determined that the failure probability is greater than the first probability threshold, the way of reducing the current power overload ratio to obtain the target power overload ratio can be that if it is determined that the failure probability is greater than the first probability threshold, the adjustment ratio value corresponding to the first probability threshold is determined, the adjustment ratio value is positively correlated with the first probability threshold, and then the adjustment ratio value is reduced based on the current power overload ratio to obtain the target power overload ratio.

[0075] That is, the embodiment of the present application sets different probability thresholds to distinguish different levels of failure probability, sets corresponding adjustment ratio values for different probability thresholds, and the probability threshold is positively correlated with the corresponding adjustment ratio value, that is, the higher the probability threshold, the higher the corresponding adjustment ratio value. For example, the probability threshold is 0.3%, and the corresponding adjustment ratio value is 5%; the probability threshold is 2%, and the corresponding adjustment ratio value is 10%; the probability threshold is 3%, and the corresponding adjustment ratio value is 20%, and so on. Therefore, if it is determined that the failure probability is greater than 0.3% (i.e., the first probability threshold), the corresponding adjustment ratio value is 5%, then 5% is reduced based on the current power overload ratio to obtain the target power overload ratio. If it is determined that the failure probability is greater than 2% (i.e., the first probability threshold), the corresponding adjustment ratio value is 10%, then 10% is reduced based on the current power overload ratio to obtain the target power overload ratio.

[0076] The embodiment of the present application determines the corresponding adjustment ratio value based on different levels of failure probability when the failure probability is relatively high, thereby reducing the adjustment ratio value based on the current power overload ratio, thereby more reasonably changing the current power overload ratio, so as to reduce the transient power of the processor by using the more reasonable power overload ratio (i.e., the target power overload ratio), thereby better reducing the failure probability of the device to be optimized, and effectively improving the performance of the device to be optimized.

[0077] In another possible implementation, the manner of adjusting the current power overload ratio of the device to be optimized based on the failure probability to obtain the target power overload ratio can be judging the size of the failure probability. If it is determined that the failure probability is less than or equal to a second probability threshold, it can be considered that the failure caused by the transient power overload does not affect the performance of the device to be optimized, or even there is no failure. At this time, in order to improve the performance of the processor within a certain range of transient power, the current power overload ratio can be increased to obtain the target power overload ratio, so that the transient power can be fully released according to the increased power overload ratio (i.e., the target power overload ratio).

[0078] The second probability threshold can be a maximum failure probability value that does not affect the performance of the device to be optimized, and can be set according to actual business requirements. For example, the second probability threshold can be 0%. The adjustment ratio value required for the increase can also be set according to actual business requirements. For example, the required adjustment ratio value for the decrease can be 1%. Therefore, when the failure probability is greater than 0%, the current power overload ratio can be increased by 1% based on the current power overload ratio, thereby obtaining the target power overload ratio.

[0079] The embodiment of the present application increases the current power overload ratio when the failure probability is relatively low or even there is no failure, thereby changing the current power overload ratio, so as to improve the transient power of the processor by using the increased power overload ratio (i.e., the target power overload ratio), thereby improving the performance of the processor within a certain range.

[0080] In another possible implementation, in order to avoid accidental influence on the judgment of the failure condition, the manner of adjusting the current power overload ratio of the device to be optimized based on the failure probability to obtain the target power overload ratio can be judging the size of the failure probability. If it is determined that the duration when the failure probability is less than or equal to the second probability threshold reaches a duration threshold, it indicates that the failure probability is relatively low for a long time or even there is no failure for a long time. At this time, in order to avoid that the transient power overload limits the performance of the processor, the current power overload ratio can be increased to obtain the target power overload ratio, so that the transient power can be fully released according to the increased power overload ratio (i.e., the target power overload ratio).

[0081] The embodiment of the present application further judges the duration when the failure probability is relatively low or even there is no failure when the failure probability is relatively low or even there is no failure, thereby increasing the current power overload ratio when the failure probability is relatively low for a long time or even there is no failure for a long time, avoiding that the current power overload ratio is adjusted incorrectly due to accidental factors when the failure probability is temporarily low, and thereby improving the accuracy of power tuning.

[0082] It should be noted that when the current power overload ratio of the device to be optimized is adjusted based on the failure probability, a one-time adjustment method can be used, that is, the adjustment ratio value to be adjusted is determined according to the failure probability, and the adjustment ratio value needs to make the transient power controlled after adjustment optimal, that is, the target power overload ratio obtained after adjustment is the optimal power overload ratio, so that the availability of the processor in the device to be optimized meets the preset condition (that is, the preset condition is reached), thereby improving the availability of the processor.

[0083] In some cases, in order to ensure flexibility and accuracy of adjustment, a step-by-step adjustment method can also be used, that is, the current power overload ratio of the device to be optimized is adjusted based on the failure probability to obtain the target power overload ratio. The method can be based on the failure probability, and the current power overload ratio is adjusted by a preset step value to obtain the target power overload ratio. After the transient power of the processor is adjusted based on the target power overload ratio control, if it is determined that the availability of the processor on the device to be optimized meets the preset condition, the step of adjusting the current power overload ratio based on the failure probability to obtain the target power overload ratio is stopped, otherwise, the failure probability is obtained again, the target power overload ratio obtained by the previous adjustment is taken as the new current power overload ratio, and the current power overload ratio is adjusted based on the failure probability again to obtain the target power overload ratio. The preset step value is used to adjust the current power overload ratio, and the step is repeated. The step-by-step adjustment method is to adjust the power overload ratio step by step through step-by-step approximation according to the preset step value. The preset step value can be the adjustment ratio value of each adjustment, which can be set according to actual business requirements.

[0084] The embodiment of the present application adjusts the power overload ratio by the step-by-step adjustment method, which can gradually approach the optimal power overload ratio, and ensures flexibility and accuracy of power optimization.

[0085] After obtaining the target power overload ratio, the target power overload ratio can be used to control and adjust the transient power of the processor. By changing the EDPp Percent of the underlying processor (such as a GPU), the adjustment of the EDPp Percent can be completed within 400us-50ms. Thus, the transient power adjustment time is short, and during the data processing service (such as continuous training of an AI cluster), the us-level transient power peak suppression is achieved while the transient power of the processor is still maintained above the rated power, and the performance of the processor is not affected. For the internal devices such as DrMOS in the GPU, the transient current overshoot suppression effect is achieved. DrMOS is a Driver and Metal-Oxide-Semiconductor Field-Effect Transistor (MOSFET) packaged chip, and the power supply of DrMOS supports flexible overclocking of the GPU.

[0086] In the embodiments of the present application, the server performing S201-S203 can be an in-band server, and of course, to avoid affecting the data processing service, the server performing S201-S203 can also be an out-of-band management server (out-of-band Management Server), and the embodiments of the present application do not limit this. When the server performing S201-S203 is an out-of-band management server, to adjust the transient power of the processor of the to-be-optimized device, after obtaining the target power overload ratio, the out-of-band management server can access the parameter transfer interface of the to-be-optimized device, and if the access is successful, the target power overload ratio is transferred to the processor of the to-be-optimized device through the parameter transfer interface.

[0087] The parameter transfer interface can be a Redfish interface, and Redfish is a modern, open standard, Representational State Transfer (RESTful) Application Programming Interface (API) for managing and detecting hardware devices in a data center, such as servers, storage, and network devices. It is designed to replace the traditional Intelligent Platform Management Interface (IPMI) to provide better security, scalability, and ease of use.

[0088] Taking the to-be-optimized device as an AI cluster for example, referring to Figure 3As shown, the AI cluster can include multiple nodes, each of which can be a server in the AI cluster. Each server includes multiple processors, which can be multiple GPUs, such as GPU1, GPU2, GPU3, GPU4, GPU5, GPU6, GPU7, and GPU8, which can constitute a GPU module. The out-of-band management server and the AI cluster can be connected through Ethernet (Ethnet), and the out-of-band management server provides a Redfish API for adjusting the EDPp Percent. After obtaining the target power overload ratio by performing S201-S203 through the out-of-band management server, the hardware can be parameterized. Specifically, the out-of-band management server can access the REDFISH API of the out-of-band management controller (BMC) of a certain server in the AI cluster. On the hardware, the BMC is connected to the HMC of the GPU module through the Universal Serial Bus (USB), and then connected to the FPGA through the peripheral component interconnect express (PCIe) of the HMC. The FPGA transmits the GPU modified EDPp Percent parameter through the Inter-Integrated Circuit (IIC), and controls the clock boost and clock down inside the GPU to control the transient power, for example, to control the transient power rise or fall of the GPU. The HMC is the management controller of the hyper-scale GPU accelerator.

[0089] When the clock boost and clock down inside the GPU are used to control the transient power, based on the target power overload ratio, the core clock frequency of the GPU can be automatically increased through the GPU Boost, and the computing power of the GPU is increased through the increased clock frequency, thereby increasing the transient power. Alternatively, based on the target power overload ratio, the core clock frequency of the GPU can be automatically reduced through the clock down, and the transient power is reduced through the reduced clock frequency.

[0090] The embodiments of the present application remotely adjust the power overload ratio through the out-of-band management server, and then configure the target power overload ratio on the device to be optimized through the Redfish interface, without the need for customer or business downtime authorization, thereby solving the problem of customer authorization difficulty. When the power overload ratio is changed, the data processing business is not aware.

[0091] In the embodiments of the present application, the adjusted power overload ratio (target power overload ratio) can be set on the to-be-optimized device in a pre-embedded manner, so as to not affect the data processing service. When the target power overload ratio is set on the to-be-optimized device, it is necessary to ensure that the target power overload ratio setting takes effect, so that the to-be-optimized device can control and adjust the transient power of the processor by using the target power overload ratio. Therefore, in order to know whether the target power overload ratio setting takes effect, in a possible implementation, the target power overload ratio can be marked by using a tag, so that it can be determined whether the target power overload ratio setting takes effect according to whether there is a tag mark. If the target power overload ratio setting takes effect in the to-be-optimized device, the tag is released, and if the target power overload ratio setting does not take effect in the to-be-optimized device, the tag mark is kept.

[0092] It should be noted that if the to-be-optimized device is an AI cluster, if there are multiple AI clusters, the target power overload ratio corresponding to each AI cluster can be set by using the above method. In the case of multiple AI clusters, different EDPp Percent parameter optimizations are supported, and the failure probability of the processor is the lowest under the highest EDPp percent parameter adjustment.

[0093] The embodiments of the present application indicate whether the target power overload ratio setting takes effect by using a tag, so that it can be accurately known whether the target power overload ratio setting takes effect, and then it is ensured that the subsequent transient power is controlled and adjusted by using the target power overload ratio, and the accuracy of the transient power adjustment is ensured.

[0094] It can be seen from the technical solution that the transient power of the processor of the to-be-optimized device is optimized by adjusting the power overload ratio. Specifically, when power optimization is needed, processor failure data of the to-be-optimized device can be obtained. The to-be-optimized device includes a processor for executing data processing services. The processor failure data can reflect the failure occurrence of the processor of the to-be-optimized device. Then, the processor failure data of the to-be-optimized device can be analyzed to obtain the failure probability of the to-be-optimized device. The failure probability can reflect the probability that the processor of the to-be-optimized device fails due to excessive transient power up to the current time, and then reflect how to adjust the power overload ratio. The power overload ratio represents the percentage of the transient power of the processor that can exceed the rated power. By adjusting the power overload ratio, the transient power of the processor can be suppressed from overshooting or being too low. Thus, after adjusting the current power overload ratio of the to-be-optimized device based on the failure probability to obtain the target power overload ratio, since the target power overload ratio is the optimized power overload ratio, it can be reflected that the peak value of the transient power of the processor is allowed under the condition of ensuring the performance of the processor. Therefore, the transient power of the processor can be controlled and adjusted according to the target power overload ratio to protect the processor within the expected power range. The transient power of the processor is optimized by adjusting the power overload ratio, without forcibly limiting the average power, so that the performance and computing power of the processor executing data processing services are not affected, and the transient power optimization is realized.

[0095] The power optimization method provided by the embodiments of the present application will be introduced from the to-be-optimized device side.

[0096] Referring to Figure 4 As shown in the figure, the method includes S401-S403, and the details are as follows:

[0097] S401, obtaining a target power overload ratio of a to-be-optimized device, the to-be-optimized device including a processor for executing data processing services, the target power overload ratio being obtained by adjusting a current power overload ratio of the to-be-optimized device based on a failure probability, the failure probability being obtained by analyzing processor failure data of the to-be-optimized device.

[0098] The determination method of the target power overload ratio can refer to the foregoing Figure 2 Corresponding embodiments, after obtaining the target power overload ratio, the target power overload ratio can be set on the to-be-optimized device, so that the target power overload ratio can be obtained to control and adjust the transient power of the processor.

[0099] It can be understood that the foregoing Figure 2The method provided by the corresponding embodiment can be executed by an in-band server or an out-of-band management server. When executed by the out-of-band management server, the target power overload ratio is obtained by adjusting the current power overload ratio based on the failure probability by the out-of-band management server, and at this time, the manner of obtaining the target power overload ratio of the device to be optimized can be to receive the target power overload ratio sent by the out-of-band management server through a parameter transmission interface. The specific manner in which the out-of-band management server sends the target power overload ratio to the device to be optimized can be referred to in the foregoing description, and will not be described here again.

[0100] It should be noted that, in order to ensure that the target power overload ratio takes effect on the device to be optimized, in a possible implementation manner, the processor of the device to be optimized can be automatically reloaded or reset when it is determined that the state of the device to be optimized is the idle state, so that the received target power overload ratio takes effect on the device to be optimized.

[0101] In a possible implementation manner, the manner of determining whether the state of the device to be optimized is the idle state can be to detect whether the state of the device to be optimized is the idle state by using a container platform to perform task management. The container platform can be a K8S container platform, and K8S is an abbreviation of kubernetes, which is an open source container orchestration engine that can support automatic deployment, large-scale scalability and application container management.

[0102] Referring to FIG. 7, Figure 5 , Figure 5 For example, the device to be optimized is a server cluster, and the processor is a GPU. The target power overload ratio (EDPp Percent) is pre-embedded and set (see S501 in FIG. 8), and if the setting takes effect, the label is removed, and if the setting does not take effect, the label is marked (see S502 in FIG. 8). If it is detected by using a container platform to perform task management that the state of a server in the server cluster is the idle state, the GPU is reloaded or reset (see S503 in FIG. 8), so as to implement the target power overload ratio (see S504 in FIG. 8). If there are multiple server clusters, corresponding target power overload ratios are set for the multiple server clusters (see S505 in FIG. 8). Figure 5 Figure 5 Figure 5 Figure 5 Figure 5

[0103] S402, calculating a transient power peak value of the processor by using the target power overload ratio.

[0104] S403, controlling and adjusting the transient power of the processor based on the transient power peak value.​​​​​

[0105] The power overload ratio represents the percentage of the transient power of the processor that exceeds the rated power. By adjusting the power overload ratio, the transient power overshoot or undershoot of the processor can be suppressed. Thus, after adjusting the current power overload ratio of the to-be-optimized device based on the failure probability to obtain the target power overload ratio, since the target power overload ratio is the adjusted power overload ratio, the peak value of the transient power of the processor that can be reached under the condition of ensuring the performance of the processor can be reflected. Therefore, when the transient power of the processor is controlled and adjusted by using the target power overload ratio, the transient power peak value of the processor can be calculated by using the target power overload ratio, and then the transient power of the processor is controlled and adjusted based on the transient power peak value.

[0106] When the processor is a GPU, the control calculation formula of the target power overload ratio on the transient power is as follows: transient power peak value = TGP * Boost Rate * EDPp percent. Wherein, TGP is the average power maximum limit (THE POWER LIMIT OF GPU), Boost Rate is the overshoot ratio, which can be the ratio of the maximum value (Max) of the transient power of the GPU to TGP, and EDPp percent is the target power overload ratio. Taking H800 EDPp percent 90% as an example, if Boost Rate is 1.6 and TGP is 700W, the transient power peak value = 700W * 1.6 * 90% = 1008W.

[0107] The transient power optimization of the processor is realized by adjusting the power overload ratio in the embodiments of the present application, without forcibly limiting the average power, so that the transient power optimization is realized without affecting the performance and computing power of the processor in executing data processing services.

[0108] The foregoing embodiments introduce the power optimization method in detail. Next, the power optimization method provided by the embodiments of the present application will be introduced in combination with an actual application scenario. In this application scenario, the to-be-optimized device is an AI cluster, the AI cluster is used to implement large model training and inference services, and the AI cluster includes multiple servers, and the processors on the servers are GPUs. In order to avoid the transient power overshoot of the GPU in the GPU high-load scene such as large model pre-training and inference, the transient power overshoot of the GPU can be suppressed by the power optimization method provided by the embodiments of the present application, and the transient power of the GPU is protected within the expected power range, avoiding the high failure probability when the 700W power specification GPU is loaded to 800-900W by the training service.

[0109] Specifically, referring to FIG. 6, the power overload ratio of the AI cluster is set to operate by default (see FIG. 6). Figure 6 Figure 6 ​In S601, the current power overload ratio of the AI cluster is changed (see Figure 6 In S602, the failure probability is obtained by analyzing the processor failure data (see Figure 6 In S603, if the failure probability is 0%, the duration of the failure probability being 0% is counted. The current power overload ratio is adjusted based on the failure probability (see Figure 6 In S604, the current power overload ratio of the AI cluster is changed using the adjusted power overload ratio (target power overload ratio). When adjusting, if the failure probability is greater than 0.3%, the current power overload ratio is reduced by 5%; if the failure probability is 0% for a long time, the current power overload ratio is increased by 1%. In the embodiments of the present application, the power overload ratio is adjusted in a step-by-step manner. Each time the adjustment is performed, S602-S604 are executed until the availability of the GPU in the AI cluster meets the preset condition (see Figure 6 In S605, the availability of the GPU in the AI cluster meets the preset condition. The preset condition can be that the availability reaches a preset threshold. The preset threshold can be set according to actual needs. For example, the preset threshold can be 95%. For example, the AI cluster includes 50 GPUs, and 48 GPUs are available. At this time, the availability of the GPU is 48 / 50=96%, which is greater than the preset threshold. Therefore, the availability of the GPU in the AI cluster meets the preset threshold, and the performance of the AI cluster in large model training and inference business meets the standard.

[0110] After obtaining the target power overload ratio, the transient power of the GPU can be controlled based on the target power overload ratio. An example of adjusting the transient power of the GPU can be seen in Figure 7 (a) and (b) of FIG. 1, Figure 7 (a) of FIG. 1 can be a change curve of the transient power of the GPU under the current power overload ratio, Figure 7 (b) of FIG. 1 can be a change curve of the transient power of the GPU under the target power overload ratio. By comparing (a) and (b), it can be seen that the method provided in the embodiments of the present application can reduce the proportion of GPU over-rating and reduce the power fluctuation range of the GPU while maintaining the high performance state of the GPU. For example, in (a) of FIG. 1, the transient power of the GPU reaches 1000W and then gradually decreases to 700W. After the current power overload ratio is reduced to obtain the target power overload ratio, in (b) of FIG. 1, the transient power of the GPU reaches 900W and then gradually decreases to 700W. Therefore, when the transient performance of the power supply device (Power Supply Unit, PSU) fluctuates, better load balancing of the PSU and protection of the dynamic load capacity of the PSU can be maintained.

[0111] Compared with reducing the average power of the GPU by the Power Brake in the related art, the related art can cause AI cluster performance jitter, server single machine direct performance is pulled down, and large model training business exception is triggered. After changing the version, cancel the power protection, the GPU load is fully released, the transient power of the GPU can reach more than 800W, and the GPU failure probability is 3-5%.

[0112] The related art can also use GPU power limit to realize power optimization, but this way can cause the average power of the GPU to also decrease, directly affecting the GPU cluster computing power, as shown in Figure 8 Figure 8 In the (a) graph, the transient power curve of the GPU when the transient power peak is 1000W can be, Figure 8 In the (b) graph, the transient power curve of the GPU when the transient power peak is 800W can be. By comparing the (a) graph and the (b) graph, it can be seen that the method provided by the related art can reduce the proportion of GPU over-rating power and narrow the power fluctuation range of the GPU, but the average power of the GPU decreases from 700W to 500W, causing the average power of the GPU to also decrease, directly affecting the GPU cluster computing power.

[0113] The method provided by the embodiments of the present application adopts the step-by-step modification of EDPp percent, so that the supplier server model reduces the GPU failure probability to 0% in multiple AI clusters under high load; in individual AI cluster large model training under high load, the failure probability is maintained at 3%, and further reducing EDPp percent can further improve the reliability of the AI cluster computing power GPU, and the failure rate is reduced to less than 3%.

[0114] For servers with good heat dissipation capacity, the failure probability is 1.8% on average when responding to AI cluster training, and the failure probability of GPU card failure due to high load is 0.36%. After adjusting EDPp percent to 90%, no new high-load GPU card is lost, and the continuity of AI cluster training is greatly improved.

[0115] It should be noted that the implementation manners provided by the present application in the above aspects can be further combined to provide more implementation manners.

[0116] Based on the power optimization method provided in the foregoing embodiments, the embodiments of the present application further provide a power optimization device 900. As shown in Figure 9 The power optimization device 900 includes an acquisition unit 901, an analysis unit 902, and an adjustment unit 903:

[0117] ​The acquisition unit 901 is configured to acquire processor failure data of a to-be-optimized device, wherein the to-be-optimized device comprises a processor configured to perform a data processing service;

[0118] The analysis unit 902 is configured to analyze the processor failure data of the to-be-optimized device to obtain a failure probability of the to-be-optimized device.

[0119] The adjustment unit 903 is configured to adjust a current power overload ratio of the to-be-optimized device based on the failure probability to obtain a target power overload ratio, wherein the target power overload ratio is used to control adjustment of transient power of the processor.

[0120] In a possible implementation, the adjustment unit 903 is configured to:

[0121] If it is determined that the failure probability is greater than a first probability threshold, the current power overload ratio is reduced to obtain the target power overload ratio.

[0122] In a possible implementation, the adjustment unit 903 is configured to:

[0123] If it is determined that the failure probability is greater than the first probability threshold, a first adjustment ratio value corresponding to the first probability threshold is determined, wherein the first adjustment ratio value is positively correlated with the first probability threshold.

[0124] The target power overload ratio is obtained by reducing the first adjustment ratio value on the basis of the current power overload ratio.

[0125] In a possible implementation, the adjustment unit 903 is configured to:

[0126] If it is determined that the failure probability is less than or equal to a second probability threshold, the current power overload ratio is increased to obtain the target power overload ratio.

[0127] Alternatively, if it is determined that a time length during which the failure probability is less than or equal to the second probability threshold reaches a time length threshold, the current power overload ratio is increased to obtain the target power overload ratio.

[0128] In a possible implementation, the adjustment unit 903 is configured to:

[0129] The target power overload ratio is obtained by adjusting the current power overload ratio by a preset step value based on the failure probability.

[0130] The apparatus further comprises a determination unit:

[0131] The determining unit is configured to, after adjusting the transient power of the processor based on the target power overload ratio, determine whether the availability of the processor on the device to be optimized reaches a preset condition, and stop the step of adjusting the current power overload ratio according to a preset step value to obtain the target power overload ratio based on the failure probability.

[0132] In a possible implementation, the parsing unit 902 is configured to:

[0133] parse the processor failure data of the device to be optimized, and obtain target type processor failure data from the processor failure data of the device to be optimized;

[0134] based on the target type processor failure data, statistically obtain the failure probability of the device to be optimized.

[0135] In a possible implementation, the apparatus further includes an accessing unit and a delivering unit:

[0136] The accessing unit is configured to access a parameter delivery interface of the device to be optimized.

[0137] The delivering unit is configured to, if the access is successful, deliver the target power overload ratio to the processor of the device to be optimized through the parameter delivery interface.

[0138] In a possible implementation, the apparatus further includes a marking unit and a canceling unit:

[0139] The marking unit is configured to mark the target power overload ratio with a label.

[0140] The canceling unit is configured to, if the target power overload ratio is set to take effect in the device to be optimized, cancel the label.

[0141] It can be seen from the technical solution that the transient power of the processor of the to-be-optimized device is optimized by adjusting the power overload ratio. Specifically, when power optimization is needed, processor failure data of the to-be-optimized device can be obtained. The to-be-optimized device includes a processor for executing data processing services. The processor failure data can reflect the failure occurrence of the processor of the to-be-optimized device. Then, the processor failure data of the to-be-optimized device can be analyzed to obtain a failure probability. The failure probability can reflect the probability that the processor of the to-be-optimized device fails due to excessive transient power up to the current time, and then reflect how to adjust the power overload ratio. The power overload ratio represents the percentage of the transient power of the processor that can exceed the rated power. By adjusting the power overload ratio, the transient power of the processor can be suppressed from overshooting or being too low. Therefore, after adjusting the current power overload ratio of the to-be-optimized device based on the failure probability to obtain a target power overload ratio, the target power overload ratio is the optimized power overload ratio, which can reflect the peak value of the transient power of the processor that can be reached while ensuring the performance of the processor. Therefore, the transient power of the processor can be controlled and adjusted according to the target power overload ratio to protect the processor within the expected power range. The transient power of the processor is optimized by adjusting the power overload ratio, without forcibly limiting the average power, so that the performance and computing power of the processor executing data processing services are not affected.

[0142] Based on the power optimization method provided in the foregoing embodiments, another power optimization apparatus 1000 is further provided in the embodiments of the present application. Referring to FIG. 10, Figure 10 As shown in FIG. 10, the power optimization apparatus 1000 includes an obtaining unit 1001, a calculating unit 1002, and an adjusting unit 1003.

[0143] The obtaining unit 1001 is configured to obtain a target power overload ratio of a to-be-optimized device. The to-be-optimized device includes a processor for executing data processing services. The target power overload ratio is obtained by adjusting a current power overload ratio of the to-be-optimized device based on a failure probability. The failure probability is obtained by analyzing processor failure data of the to-be-optimized device.

[0144] The calculating unit 1002 is configured to calculate a transient power peak value of the processor by using the target power overload ratio.

[0145] The adjusting unit 1003 is configured to control and adjust the transient power of the processor based on the transient power peak value.

[0146] In a possible implementation, the target power overload ratio is obtained by adjusting the current power overload ratio based on the failure probability by an out-of-band management server. The obtaining unit 1001 is configured to:

[0147] The target power overload ratio sent by the out-of-band management server is received through the parameter passing interface.

[0148] In a possible implementation, the apparatus further includes a setting unit.

[0149] The setting unit is configured to, if it is determined that the state of the device to be optimized is an idle state, reload or reset a processor of the device to be optimized, so that the received target power overload ratio is set to take effect on the device to be optimized.

[0150] Embodiments of the present application also provide a computer device, which can execute the power tuning method. The computer device can be a terminal, Figure 11 A structure diagram of a terminal provided by an embodiment of the present application is shown. In Figure 11 In the embodiment, a terminal is taken as an example of a smartphone:

[0151] Referring to Figure 11 , the smartphone includes a radio frequency (RF) circuit 1110, a memory 1120, an input unit 1130, a display unit 1140, a sensor 1150, an audio circuit 1160, a wireless fidelity (WiFi) module 1170, a processor 1180, and a power supply 1190, and the like. The input unit 1130 can include a touch panel 1131 and other input devices 1132, and the display unit 1140 can include a display panel 1141. The audio circuit 1160 can include a speaker 1161 and a microphone 1162. It can be understood that Figure 11 The structure of the smartphone shown in the embodiment is not a limitation on the smartphone, and can include more or fewer components than shown, or combine certain components, or different component arrangements.

[0152] The memory 1120 can be used to store software programs and modules, and the processor 1180 executes various function applications and data processing of the smartphone by running the software programs and modules stored in the memory 1120. The memory 1120 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), and the like; and the data storage area can store data created according to the use of the smartphone (such as audio data, a phone book, etc.), and the like. In addition, the memory 1120 can include a high-speed random access memory, and can also include a non-volatile memory, for example, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state memory device.

[0153] The processor 1180 is the control center of the smartphone, connecting various parts of the smartphone via various interfaces and lines. It performs various functions and processes data by running or executing software programs and / or modules stored in the memory 1120 and calling data stored in the memory 1120. Optionally, the processor 1180 may include one or more processing units; preferably, the processor 1180 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 1180.

[0154] In this embodiment, the processor 1180 in the smartphone can execute the power tuning methods provided in the various embodiments of this application.

[0155] The computer device provided in this application embodiment can also be a server. Please refer to [link / reference]. Figure 12 As shown, Figure 12 The diagram illustrates the structure of a server 1200 provided in this embodiment. The server 1200 can vary significantly due to different configurations or performance characteristics. It may include one or more processors, such as a central processing unit 1222, and a memory 1232, as well as one or more storage media 1230 (e.g., one or more mass storage devices) for storing application programs 1242 or data 1244. The memory 1232 and storage media 1230 may be temporary or persistent storage. The program stored in the storage media 1230 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the server. Furthermore, the central processing unit 1222 may be configured to communicate with the storage media 1230 and execute the series of instruction operations stored in the storage media 1230 on the server 1200.

[0156] Server 1200 may also include one or more power supplies 1226, one or more wired or wireless network interfaces 1250, one or more input / output interfaces 1258, and / or one or more operating systems 1241, such as Windows Server. TM Mac OS X TM Unix TM Linux TM FreeBSD TM etc.

[0157] In this embodiment, the central processing unit 1222 in the server 1200 can execute the power tuning methods provided in the various embodiments of this application.

[0158] According to an aspect of the present application, a computer readable storage medium is provided for storing a computer program for performing the power tuning method described in the various embodiments.

[0159] According to an aspect of the present application, a computer program product is provided, which includes a computer program stored in a computer readable storage medium. A processor of a computer device reads the computer program from the computer readable storage medium, and the processor executes the computer program, so that the computer device performs the method provided in the various optional implementation manners of the above embodiments.

[0160] The descriptions of the corresponding flow or structure of each of the above figures are each focused on, and the parts not described in detail in a certain flow or structure can be referred to the related descriptions of other flows or structures.

[0161] The terms "first", "second", "third", "fourth" and the like in the description of the present application and the above figures (if any) are used to distinguish like objects, and are not necessarily used to describe a particular sequential or chronological order. It should be understood that the data thus used can be interchanged, where appropriate, so that the embodiments of the present application described herein can be carried out in other than the order shown or described herein. Furthermore, the terms "comprising" and "including" and any of their derivatives, are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or apparatus that comprises a list of steps or units need not be limited to those steps or units that are clearly listed, but can include other steps or units that are not clearly listed or inherent to such processes, methods, products, or apparatuses.

[0162] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are merely illustrative, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0163] The units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0164] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0165] When the integrated unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application, essentially or the part that contributes to the prior art, or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a terminal, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various computer program storage media.

[0166] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program with a predetermined function, and works together with other related parts to achieve a predetermined target, and can be realized in whole or in part by using software, hardware (such as a processing circuit or a memory) or a combination thereof. Similarly, one processor (or multiple processors or memories) can be used to realize one or more modules or units. In addition, each module or unit can be a part of an integral module or unit that includes the functions of the module or unit.

[0167] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications or replacements do not cause the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A power tuning method, characterized in that, The method includes: Obtain processor fault data of the device to be optimized, wherein the device to be optimized includes a processor for performing data processing services; The processor fault data of the device to be optimized is analyzed, and the fault probability of the device to be optimized is statistically obtained. Based on the failure probability, the current power overload ratio of the device to be optimized is adjusted to obtain the target power overload ratio, which is used to control and adjust the transient power of the processor.

2. The method according to claim 1, characterized in that, The step of adjusting the current power overload ratio of the device to be optimized based on the fault probability to obtain the target power overload ratio includes: If the fault probability is determined to be greater than the first probability threshold, the current power overload ratio is reduced to obtain the target power overload ratio.

3. The method according to claim 2, characterized in that, If the fault probability is determined to be greater than a first probability threshold, the current power overload ratio is reduced to obtain the target power overload ratio, including: If it is determined that the failure probability is greater than the first probability threshold, an adjustment ratio value corresponding to the first probability threshold is determined, and the adjustment ratio value is positively correlated with the first probability threshold; Based on the current power overload ratio, the adjustment ratio value is reduced to obtain the target power overload ratio.

4. The method according to claim 1, characterized in that, The step of adjusting the current power overload ratio of the device to be optimized based on the fault probability to obtain the target power overload ratio includes: If the fault probability is determined to be less than or equal to the second probability threshold, the current power overload ratio is increased to obtain the target power overload ratio. Alternatively, if the duration for which the fault probability is less than or equal to the second probability threshold reaches a duration threshold, the current power overload ratio is increased to obtain the target power overload ratio.

5. The method according to claim 1, characterized in that, The step of adjusting the current power overload ratio of the device to be optimized based on the fault probability to obtain the target power overload ratio includes: Based on the fault probability, the current power overload ratio is adjusted according to a preset step value to obtain the target power overload ratio; The method further includes: If, after adjusting the transient power of the processor based on the target power overload ratio, it is determined that the availability of the processor on the device to be optimized reaches a preset condition, the step of adjusting the current power overload ratio based on the fault probability according to a preset step value to obtain the target power overload ratio is stopped.

6. The method according to claim 1, characterized in that, The step of parsing the processor fault data of the device to be optimized and statistically obtaining the fault probability of the device to be optimized includes: The processor fault data of the device to be optimized is parsed to obtain the processor fault data of the target type. Based on the processor fault data of the target type, the failure probability of the device to be optimized is statistically obtained.

7. The method according to any one of claims 1-6, characterized in that, The method further includes: Access the parameter transmission interface of the device to be optimized; If the access is successful, the target power overload ratio is transmitted to the processor of the device to be optimized through the parameter transmission interface.

8. The method according to claim 7, characterized in that, The method further includes: The target power overload ratio is marked using tags; If the target power overload ratio is set in effect in the device to be optimized, the label is removed.

9. A power optimization method, characterized in that, The method further includes: The target power overload ratio of the device to be optimized is obtained. The device to be optimized includes a processor for performing data processing services. The target power overload ratio is obtained by adjusting the current power overload ratio of the device to be optimized based on the failure probability. The failure probability is obtained by parsing the processor failure data of the device to be optimized. The transient power peak of the processor is calculated using the target power overload ratio; The transient power of the processor is adjusted based on the transient power peak control.

10. The method according to claim 9, characterized in that, The target power overload ratio is obtained by adjusting the current power overload ratio based on the failure probability by the out-of-band management server. Obtaining the target power overload ratio of the device to be optimized includes: The target power overload ratio is received from the out-of-band management server via the parameter transmission interface.

11. The method according to claim 10, characterized in that, The method further includes: If the state of the device to be optimized is determined to be idle, the processor of the device to be optimized is reloaded or reset so that the received target power overload ratio is set on the device to be optimized.

12. A power optimization device, characterized in that, The device includes an acquisition unit, an analysis unit, and an adjustment unit: The acquisition unit is used to acquire processor fault data of the device to be optimized, wherein the device to be optimized includes a processor for performing data processing services; The parsing unit is used to parse the processor fault data of the device to be optimized and to statistically obtain the fault probability of the device to be optimized. The adjustment unit is used to adjust the current power overload ratio of the device to be optimized based on the fault probability to obtain a target power overload ratio, which is used to control and adjust the transient power of the processor.

13. A power optimization device, characterized in that, The device includes an acquisition unit, a calculation unit, and an adjustment unit: The acquisition unit is used to acquire the target power overload ratio of the device to be optimized. The device to be optimized includes a processor for performing data processing services. The target power overload ratio is obtained by adjusting the current power overload ratio of the device to be optimized based on the failure probability. The failure probability is obtained by parsing the processor failure data of the device to be optimized. The computing unit is used to calculate the transient power peak of the processor using the target power overload ratio; The adjustment unit is used to control and adjust the transient power of the processor based on the transient power peak value.

14. A computer device, characterized in that, The computer device includes a processor and memory: The memory is used to store computer programs and to transfer the computer programs to the processor; The processor is configured to execute the method according to any one of claims 1-8 or 9-11 according to instructions in the computer program.

15. A computer-readable storage medium for storing a computer program that, when executed by a processor, causes the processor to perform the method of any one of claims 1-8 or 9-11.

16. A computer program product comprising a computer program that, when executed by a processor, implements the method of any one of claims 1-8 or 9-11.

Images