Server power consumption management method and device, electronic equipment and storage medium

Abstract

The invention discloses a server power consumption management method and device, electronic equipment and a storage medium, and relates to the technical field of computers, the future load change trend of a server can be predicted according to the load data of the server at the current moment, and a power consumption management strategy of the server is generated according to the load data and the load change trend at the current moment. The power consumption management strategy is utilized to adjust the operation parameters of at least one component in the multiple components, and the power consumption management strategy is predicted in advance and adjusted in real time, so that the load fluctuation of the server can be dynamically adapted, and insufficient resource allocation of the server under different load conditions due to a fixed static power consumption management strategy is avoided. Therefore, the technical problems that the resource allocation of the server is unbalanced and the energy efficiency is low due to the fact that the power consumption management strategy cannot be dynamically adjusted in related technologies can be solved, and the purposes of optimizing the resource allocation of the server and improving the energy efficiency of the server are achieved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a server power consumption management method, device, electronic device, and storage medium. Background Art

[0002] As data centers continue to expand and demand for high-performance computing increases, server power management becomes crucial for improving energy efficiency and reducing operating costs. AMD (Advanced Micro Devices) architecture servers are widely used in data centers and high-performance computing scenarios due to their high performance and multi-core processing capabilities.

[0003] However, the BMC (Baseboard Management Controller) method for server power consumption management in related technologies mainly manages the power consumption of the entire machine and a fixed strategy. However, the BMC cannot obtain the load data of multiple components of the AMD server, and is limited by the fixed strategy, making it difficult to cope with the high-frequency load fluctuations of the AMD server, which in turn leads to unbalanced resource allocation and low energy efficiency of the server. Summary of the Invention

[0004] The present application provides a server power consumption management method, device, electronic device and storage medium to at least solve the problems in the related art such as the inability to dynamically adjust the power consumption management strategy according to real-time load data, which in turn leads to unbalanced resource allocation of the server and low energy efficiency.

[0005] The present application provides a server power consumption management method, wherein the server includes multiple components, and the multiple components include multiple processors that perform heterogeneous computing tasks. The method includes the following steps: obtaining load data of the server at the current moment; predicting the future load change trend of the server based on the load data of the server at the current moment; generating a power consumption management strategy for the server based on the load data and load change trend of the server at the current moment, and using the power consumption management strategy to adjust the operating parameters of at least one component among the multiple components.

[0006] The present application also provides a server power consumption management device, wherein the server includes multiple components, and the multiple components include multiple processors that perform heterogeneous computing tasks, wherein the device includes: an acquisition module for acquiring the load data of the server at the current moment; a prediction module for predicting the future load change trend of the server based on the load data of the server at the current moment; and an adjustment module for adjusting the operating parameters of at least one component among the multiple components using a power consumption management strategy.

[0007] The present application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for implementing the steps of any of the above-mentioned server power consumption management methods when executing the computer program.

[0008] The present application also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned server power consumption management methods are implemented.

[0009] Through the present application, since the power consumption management strategy in the related art is managed through the power consumption of the entire machine and a fixed strategy, it cannot adapt to the high-frequency load fluctuations of the server, which in turn leads to problems such as imbalanced resource allocation of the server and low energy efficiency, the present application can predict the future load change trend of the server based on the load data of the server at the current moment, generate the power consumption management strategy of the server based on the load data and load change trend at the current moment, and use the power consumption management strategy to adjust the operating parameters of at least one component among multiple components. By predicting in advance and adjusting the power consumption management strategy in real time, it is possible to dynamically adapt to the load fluctuations of the server, avoid insufficient performance of the server at high load or waste of power consumption at low load due to a fixed static power consumption management strategy, and increase the energy efficiency of the server by predicting in advance, thereby optimizing the resource allocation of the server and improving the energy efficiency of the server. Therefore, it can solve the technical problem that the related art cannot always dynamically adjust the power consumption management strategy, resulting in imbalanced resource allocation of the server and low energy efficiency, and achieve the technical effect of optimizing server resource allocation and improving server energy efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0011] Figure 1 A flowchart of a server power consumption management method provided according to an embodiment of the present application;

[0012] Figure 2 This is a flowchart of an execution method for managing server power consumption according to a specific embodiment of the present application;

[0013] Figure 3 A schematic diagram of a server power consumption management device provided according to an embodiment of the present application;

[0014] Figure 4 A schematic diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0015] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0016] It should be noted that, in the description of this application, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. The terms "first," "second," etc., in this application are used to distinguish similar objects, and are not used to describe a particular order or sequence.

[0017] Before describing the solution of this application, we first introduce the relevant technologies of this application and the existing technical problems.

[0018] The problems that arise when implementing power management on AMD servers using BMC power management strategies in related technologies are essentially due to a mismatch between their architecture design and the hardware characteristics and load patterns of AMD servers. Specifically, they include:

[0019] 1. Insufficient power consumption management and the contradiction between hardware characteristics and monitoring capabilities.

[0020] 1. The multi-core and heterogeneous architecture complexity of AMD server processors.

[0021] (1) AMD server processors use a core complex or core chiplet die architecture. For example, the 7003 series processors contain multiple independent core chiplet dies, each with 8-16 cores. The BMC of related technologies only supports whole-machine power consumption monitoring (e.g., obtaining total current through motherboard power sensors) and cannot distinguish between different core load differences.

[0022] (2) AMD servers often integrate Radeon GPUs (AMD Radeon Graphics Processing Units), whose computing unit power consumption distribution is highly dynamic. Related technologies lack a component-level monitoring interface for the GPU (Graphics Processing Unit) in the BMC.

[0023] 2. The dynamic load adaptability is limited.

[0024] 1. High-frequency fluctuations in AMD server load.

[0025] Because AMD servers are often used in scenarios like AI (Artificial Intelligence) training and video transcoding, loads can jump from 10% to 90% in seconds. The fixed threshold strategy of the BMC in related technologies cannot respond to such sudden changes in a timely manner. Furthermore, the control logic of the BMC in related technologies is based on historical experience and cannot adapt to diverse real-time load patterns.

[0026] 3. Lack of heterogeneous computing optimization.

[0027] AMD servers' CPUs (Central Processing Units) and GPUs are connected via a high-speed bus, requiring dynamic allocation of bandwidth and power resources (for example, prioritizing video memory bandwidth for GPU-intensive tasks). Related technologies' BMCs lack hardware-level collaborative interfaces and are unable to participate in this cross-component scheduling. Furthermore, these BMCs cannot dynamically adjust power consumption based on task types, potentially limiting the performance of a single component in heterogeneous tasks.

[0028] In order to enable those skilled in the art to better understand the present application, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0029] An embodiment of the present application provides a server power consumption management method, and the method is described in detail in conjunction with the execution process of the server power consumption management method.

[0030] like Figure 1 As shown, the server power consumption management method includes the following steps:

[0031] The server of the present application includes multiple components, and the multiple components include multiple processors that perform heterogeneous computing tasks, wherein heterogeneous computing tasks refer to computing tasks completed by different types of processors in collaboration, such as CPU processing logic control, CPU processing large-scale parallel computing, etc. The server of the present application can be an AMD architecture server, and the following embodiments are all explained using the AMD architecture server as an example.

[0032] In step S101, the load data of the server at the current moment is obtained.

[0033] Load data includes the utilization of components such as CPU multi-core, GPU, and memory, as well as server bandwidth and traffic. Load data can be collected in real time through BMC sensors.

[0034] In step S102, the future load change trend of the server is predicted based on the load data of the server at the current moment.

[0035] It is understandable that the embodiments of the present application can predict the future load change trend of the server based on the load data of the server at the current moment, so as to dynamically adjust the power consumption management strategy of the server subsequently.

[0036] In an embodiment of the present application, the future load change trend of the server is predicted based on the load data of the server at the current moment, including: inputting the load data of the server at the current moment into a power consumption prediction model, and the power consumption prediction model outputs the load change trend of the server, wherein the power consumption prediction model is obtained based on training of a machine learning model.

[0037] Among them, the machine learning model can be LSTM (Long Short-Term Memory network).

[0038] It is understandable that the embodiment of the present application can input the current server load data into the power consumption prediction model, and the power consumption prediction model outputs the future load change trend of the server, so as to subsequently implement early adjustment of the server power consumption management strategy.

[0039] In an embodiment of the present application, in addition to inputting the load data of the server at the current moment into the power consumption prediction model, it also includes: obtaining the historical load data of the server; generating a training data set based on the historical load data, using the training data set to train the machine learning model, and obtaining the power consumption prediction model after the training is completed.

[0040] It is understandable that the embodiments of the present application can use the historical load data of the server to generate a training data set, and use the training data set to train a machine learning model, and obtain a power consumption prediction model after the training is completed.

[0041] For example, a training data set can be generated based on the server's historical load data for the past 30 days, and the historical load data for the past 30 days can be used to train a machine learning model.

[0042] Compared with the passive response of traditional BMC, this application can adjust the server's power consumption management strategy in advance through the power consumption prediction model. For example, when it is predicted that the load will drop at night, it will enter low power consumption mode in advance to reduce energy consumption by 30%.

[0043] In step S103, a power consumption management policy of the server is generated according to the load data and load change trend of the server at the current moment, and the operating parameters of at least one component among the multiple components are adjusted using the power consumption management policy.

[0044] Among them, operating parameters include adjustable parameters such as the frequency, number of cores, voltage, memory bandwidth, storage I / O performance, etc. of multiple processors.

[0045] It can be understood that the embodiments of the present application can generate a power consumption management strategy for the server based on the load data and load change trends at the current moment, and use the power consumption management strategy to adjust the operating parameters of at least one component among multiple components. By predicting in advance and adjusting the power consumption management strategy in real time, dynamic adaptation to load fluctuations can be achieved, avoiding insufficient performance of the server at high load or waste of power consumption at low load due to fixed static power consumption management strategies, and by predicting in advance, the operating efficiency of the server can be increased, thereby optimizing the resource allocation of the server and improving the energy efficiency of the server.

[0046] For example, when the server is currently executing a video transcoding task (high GPU load), based on the current load data, it is predicted that the future load change trend will continue to increase. In this case, the GPU operating frequency will be increased, idle CPU cores will be shut down, and the memory voltage will be reduced.

[0047] In an embodiment of the present application, a power consumption management strategy of a server is generated based on the load data of the server at the current moment and the load change trend, including: generating a first management strategy of at least one component based on the load data of the server at the current moment; generating a second management strategy of at least one component based on the load change trend.

[0048] It can be understood that the embodiment of the present application can generate a first management strategy for at least one component based on the load data of the server at the current moment, and generate a second management strategy for at least one component based on the load change trend, so as to take into account both real-time and advance prediction.

[0049] In addition, it should be noted that when the first management policy and the second management policy are different, the first management policy has the highest priority.

[0050] For example, if the load data at the current moment is high, indicating that the server is in a high-load state, the first management strategy is to enable more CPU cores or increase the CPU operating frequency, etc. However, the load data predicted in advance at the current moment is low, and the second management strategy is to shut down some CPU cores or reduce the CPU frequency. If the two management strategies are different, the first management strategy at the current moment shall prevail.

[0051] In an embodiment of the present application, a first management strategy for at least one component is generated based on the load data of the server at the current moment, including: identifying the actual load in the load data of the server at the current moment; if the actual load is greater than the load threshold, the first management strategy includes at least one of the following: shutting down at least one core of the central processing unit, or reducing the frequency of the central processing unit; reducing the operating frequency of the graphics processor of the server, or switching the graphics processor to an idle state; reducing the memory voltage of the server and the power consumption of the storage device; if the actual load is less than or equal to the load threshold, the first management strategy includes at least one of the following: turning on multiple cores of the central processing unit; increasing the operating frequency of the central processing unit; increasing the operating frequency of the graphics processor; and improving the memory bandwidth and storage input and output performance of the server.

[0052] The load threshold can be set according to specific circumstances and is used to distinguish between critical values ​​of high and low loads.

[0053] It is understandable that, in the embodiment of the present application, when the actual load is greater than the load threshold, the first management strategy includes at least one of the following:

[0054] Shut down at least one CPU core or reduce the CPU frequency; reduce the GPU operating frequency or switch the CPU to idle state; reduce the server's memory voltage and storage device power consumption;

[0055] When the actual load is less than or equal to the load threshold, the first management strategy includes at least one of the following:

[0056] Enable multiple CPU cores; increase the CPU operating frequency; increase the GPU operating frequency; and improve the server's memory bandwidth and storage input and output performance.

[0057] Through the above approach, fine-grained control of components such as the CPU and GPU can be achieved, avoiding the waste of resources caused by traditional fixed policy adjustments.

[0058] In an embodiment of the present application, after generating the power consumption management strategy of the server based on the load data and load change trend of the server at the current moment, it also includes: identifying the task type of the currently executed task of the server; and correcting the power consumption management strategy of the server based on the task type.

[0059] Among them, the task types include CPU intensive and CPU intensive.

[0060] It is understandable that the present application can identify the task type of the currently executed task of the server and modify the power consumption management strategy of the server based on the task type to achieve collaborative optimization of heterogeneous computing.

[0061] Specifically: if the task type of the currently executed task is a CPU-intensive task, the GPU power consumption will be reduced and the CPU power consumption will be increased. Conversely, if the task type of the currently executed task is a GPU-intensive task, the CPU power consumption will be reduced and the GPU power consumption will be increased, such as shutting down some CPU cores.

[0062] For example, the task currently being performed by the AMD server is image recognition, which is GPU-intensive. In this case, the GPU frequency is increased, and at the same time, the CPU memory bandwidth usage is reduced through Infinity Fabric, prioritizing GPU data throughput, thereby increasing the image recognition speed by 35% and reducing CPU power consumption by 25%.

[0063] In an embodiment of the present application, obtaining the load data of the server at the current moment includes: obtaining power consumption data of multiple components of the server and the operating status of the server, and determining the load data of the server at the current moment based on the power consumption data and the operating status.

[0064] Among them, power consumption data includes the power consumption of CPU, GPU, memory, storage devices, etc.; operating status includes temperature, fan speed, number of processes, etc.

[0065] It is understandable that the embodiments of the present application can comprehensively consider the operating status of the server and the power consumption data of multiple components to achieve accurate perception of the server load data and improve the accuracy of load data judgment.

[0066] In the embodiment of the present application, the running status of the server can be obtained through the BMC of the server, such as obtaining the temperature through the IPMI interface, and data such as the number of processes collected by the server system.

[0067] In an embodiment of the present application, a second management strategy for at least one component is generated based on a load change trend, including: identifying an inflection point in the load change trend; determining an adjustment moment based on the inflection point, and adjusting the first management strategy to the second management strategy at the adjustment moment.

[0068] The inflection point is a point where the load change trend suddenly changes, such as a peak point or a valley point.

[0069] It can be understood that the embodiments of the present application can identify the inflection point in the load change trend, and determine the adjustment time based on the inflection point, and adjust the first management strategy to the second management strategy at the adjustment time. The adjustment time can be a period of time before the inflection point, such as 5 seconds before the power consumption management strategy is adjusted to avoid server system oscillations caused by sudden changes in the strategy.

[0070] For example, when the server is currently performing a real-time analysis task of financial transaction data, the future load change trend is predicted based on the current load data. The inflection point of the change trend is obtained as time point A. The load will decrease at time point A, so the power consumption management strategy can be adjusted 5 seconds before time point A.

[0071] In an embodiment of the present application, it also includes: obtaining power consumption data of multiple components of the server after the power consumption management policy is adjusted, and updating the power consumption management policy and the power consumption prediction model based on the adjusted power consumption data of the multiple components.

[0072] It is understandable that the embodiments of the present application can also update the power consumption management strategy and the power consumption prediction model based on the adjusted power consumption data to improve the effectiveness and pertinence of subsequent power consumption management strategies and the prediction accuracy of the power consumption prediction model.

[0073] Specifically, the server power consumption management method of this application has the following specific process: Figure 2 As shown, the following steps are described using an AMD server as an example:

[0074] Step 1: Initialize the server's BMC system and configure the hardware.

[0075] Start the AMD architecture-based server BMC, load the firmware and initialize the hardware modules, and configure the communication channels (such as I2C, PCIe, SMBus, etc.) between the BMC and the AMD CPU, GPU, memory, storage, and network interfaces. Enable the BMC's built-in power consumption sensor, configure its sampling frequency and accuracy, and obtain fine-grained power consumption data for components such as the CPU multi-core and GPU through AMD-provided performance monitoring interfaces (such as the AMD Energy Profiler). Load the preset power management policies, including low-power mode, balanced mode, and high-performance mode, initialize the machine learning model, and load historical load data for intelligent prediction.

[0076] Step 2: Real-time power consumption monitoring.

[0077] The BMC's sensor interface collects real-time power consumption data for each server component, including AMD CPU multi-core power consumption, GPU power consumption, memory power consumption, storage device power consumption, and network interface power consumption. This collected power consumption data is transmitted to the BMC's ARM processor for analysis and processing. The real-time power consumption data is then stored in the BMC's local storage unit for subsequent analysis and historical data training.

[0078] Step 3: Load analysis and dynamic adjustment.

[0079] First, based on load assessment, the current load situation is evaluated based on real-time power consumption data and server operating status, and indicators such as CPU and GPU utilization, memory bandwidth, storage I / O, and network traffic are analyzed.

[0080] The second is a dynamic adjustment strategy. When the load is low, some CPU cores are shut down or the CPU frequency is reduced, the GPU frequency is reduced or switched to idle state, and the memory voltage and storage device power consumption are reduced. When the load is high, more CPU cores are enabled and the CPU frequency is increased, and the GPU frequency is increased to accelerate computing tasks and increase memory bandwidth and storage I / O performance.

[0081] The third is heterogeneous computing optimization, which dynamically adjusts the power consumption distribution of the CPU and GPU according to the task type (CPU-intensive or GPU-intensive), and achieves power consumption balance between the CPU and GPU through the AMD Infinity Fabric architecture.

[0082] Step 4: Intelligent prediction and optimization.

[0083] Machine learning algorithms based on time series analysis (such as LSTM neural networks) are used to predict load and power consumption. By training historical data, the model can accurately predict future load changes, providing decision support for power management.

[0084] The first is historical data training, which uses historical load data to train machine learning models (such as LSTM neural networks) to learn the patterns of load changes.

[0085] The second is load forecasting, which predicts the load change trend in the future based on current load data and trained models.

[0086] The third is power consumption optimization. Based on the prediction results, the server's power consumption strategy is adjusted in advance. For example, when it is predicted that the load will increase in the future, the CPU and GPU performance will be improved in advance; or when it is predicted that the load will decrease in the future, the server will enter low power consumption mode in advance.

[0087] Step 5: Function adjustment and feedback.

[0088] The first is dynamic adjustment, which dynamically adjusts the power consumption status of each server component based on the results of real-time monitoring, load analysis and intelligent prediction, and sends adjustment instructions to AMD CPU, GPU and other components through the BMC firmware interface.

[0089] The second is the feedback mechanism, which feeds back the adjusted power consumption data to the BMC to form a closed-loop control and record the adjustment effect for optimizing the power consumption management strategy and machine learning model.

[0090] Step 6: System maintenance and updates.

[0091] The first is log recording, which records key data in the power consumption management process, including real-time power consumption, load changes, adjustment instructions and prediction results.

[0092] The second is strategy update, which regularly updates power consumption management strategies and machine learning models based on actual operating conditions.

[0093] The third is fault handling, monitoring the operating status of BMC and various server components, and promptly detecting and handling faults.

[0094] In general, the server power consumption management method of the embodiment of the present application can be divided into four modules.

[0095] 1. Power consumption monitoring module.

[0096] Real-time monitoring of the power consumption of various server components, including AMD CPU multi-core, GPU, memory, PCIE devices, storage and network interfaces.

[0097] 2. Load analysis module.

[0098] Dynamically adjust power consumption strategies based on the real-time load of the server.

[0099] 3. Heterogeneous computing optimization module.

[0100] Optimize power consumption allocation based on the heterogeneous computing characteristics of AMD CPUs and GPUs.

[0101] 4. Power consumption adjustment module.

[0102] Dynamically adjust the power consumption status of each server component based on monitoring and analysis results.

[0103] The embodiments of the present application achieve intelligent management of server power consumption through fine-grained power consumption monitoring, dynamic load adaptation and heterogeneous computing optimization. Specifically, by real-time monitoring of the power consumption of each component of the server, more refined power consumption control can be achieved to avoid energy waste; the power consumption strategy can be dynamically adjusted according to the real-time load of the server to ensure sufficient performance under high load and reduce power consumption under low load; the power consumption prediction model can be used to predict future load and power consumption requirements, and power consumption optimization can be performed in advance to improve the energy efficiency of the data center; the BMC system based on the ARM architecture itself has low power consumption characteristics, which further reduces the overall energy consumption of the server.

[0104] The server power consumption management method of the present application is described below through a specific embodiment, taking an AMD server as an example.

[0105] 1. The server's BMC collects load data on components like the CPU, GPU, memory, and storage every second through the IPMI interface. For the CPU, AMD Energy Profiler is used to obtain the temperature, power consumption, and core utilization of each CCX module. For the GPU, the ROCm monitoring API is used to collect memory bandwidth, CUDA core utilization, and temperature. Memory bandwidth and other information are also read via the SMBus.

[0106] We collected server load data from different time periods over the past 30 days and processed it into a historical load sequence dataset. We then built an LSTM neural network model using Python and the TensorFlow framework, taking the historical load sequence as input and future load data as output. After 100 rounds of training, we saved the model for subsequent load forecasting.

[0107] 2. While the server is running, the BMC continuously calls the API every second to obtain current load data, including key indicators such as total CPU utilization, number of active GPU cores, and memory bandwidth. After obtaining the data, it is preprocessed into a format that conforms to the model input format and input into the trained LSTM model to predict the load trend for the next 10 minutes and determine whether the load will increase, decrease, or remain stable.

[0108] 3. If the current GPU utilization exceeds 85%, the actual load is determined to be greater than the threshold. At this time, the first management strategy is activated: N non-critical cores in the CPU are shut down to reduce CPU power consumption through ACPI C-states; the CPU operating frequency is reduced from 3.5GHz to 2.8GHz to reduce CPU energy consumption; and the GPU frequency is increased from 1.8GHz to 2.0GHz to ensure the performance of GPU-intensive tasks.

[0109] 4. If it is predicted that the load inflection point will be reached in 10 minutes and the load change trend is increasing, for example, the CPU frequency can be increased to 3.2 GHz 5 minutes in advance to preheat the CPU resources; the GPU frequency can be reduced to 1.5 GHz to reduce unnecessary power consumption.

[0110] 5. The BMC analyzes process information and identifies the current task as image rendering. It then modifies the power management strategy based on the task type. This prioritizes GPU memory bandwidth while reducing the ECC (Error-Correcting Code) checksum strength to achieve higher performance within an acceptable reliability range.

[0111] 6. After the BMC generates the policy, it controls the hardware through specific instructions. For example, it can call system commands to shut down the CPU core or use tools to adjust the GPU frequency.

[0112] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method.

[0113] According to the server power consumption management method proposed in the embodiment of the present application, the future load change trend of the server can be predicted based on the load data of the server at the current moment, and the power consumption management strategy of the server can be generated based on the load data and load change trend at the current moment. The operating parameters of at least one component among multiple components can be adjusted using the power consumption management strategy. By predicting in advance and adjusting the power consumption management strategy in real time, dynamic adaptation to the load fluctuations of the server can be achieved, avoiding insufficient performance of the server at high load or waste of power consumption at low load due to a fixed static power consumption management strategy. In addition, the energy efficiency of the server can be increased by predicting in advance, thereby optimizing the resource allocation of the server and improving the energy efficiency of the server.

[0114] An embodiment of the present application also provides a server power consumption management device.

[0115] like Figure 3 As shown, the server power consumption management device 10 includes multiple components, and the multiple components include multiple processors that perform heterogeneous computing tasks. The server power consumption management device 10 includes: an acquisition module 100, a prediction module 200 and an adjustment module 300.

[0116] Among them, the acquisition module 100 is used to obtain the load data of the server at the current moment; the prediction module 200 is used to predict the future load change trend of the server based on the load data of the server at the current moment; the adjustment module 300 is used to use the power consumption management strategy to adjust the operating parameters of at least one component among multiple components.

[0117] In an embodiment of the present application, the prediction module 200 is further used to: input the load data of the server at the current moment into the power consumption prediction model, and the power consumption prediction model outputs the prediction of the future load change trend of the server based on the load data of the server at the current moment, wherein the power consumption prediction model is obtained based on the training of the machine learning model.

[0118] In the embodiment of the present application, the server power consumption management device 10 of the embodiment of the present application further includes: a training module.

[0119] Among them, the training module is used to obtain the historical load data of the server before inputting the current server load data into the power consumption prediction model; generate a training data set based on the historical load data, use the training data set to train the machine learning model, and obtain the power consumption prediction model after the training is completed.

[0120] In an embodiment of the present application, the adjustment module 300 is further used to: generate a first management strategy for at least one component based on the load data of the server at the current moment; and generate a second management strategy for at least one component based on the load change trend.

[0121] In an embodiment of the present application, the adjustment module 300 is further used to: identify the actual load in the load data of the server at the current moment; if the actual load is greater than the load threshold, the first management strategy includes at least one of the following: shutting down at least one core of the central processing unit, or reducing the operating frequency of the central processing unit; reducing the operating frequency of the server's graphics processor, or switching the graphics processor to an idle state; reducing the server's memory voltage and the power consumption of the storage device; if the actual load is less than or equal to the load threshold, the first management strategy includes at least one of the following: turning on multiple cores of the central processing unit; increasing the operating frequency of the central processing unit; increasing the operating frequency of the graphics processor; and improving the server's memory bandwidth and storage input and output performance.

[0122] In the embodiment of the present application, the adjustment module 300 is further used to: identify an inflection point in the load change trend; determine an adjustment time according to the inflection point, and adjust the first management strategy to the second management strategy at the adjustment time.

[0123] In the embodiment of the present application, the server power consumption management device 10 of the embodiment of the present application further includes: a correction module.

[0124] The correction module is used to identify the task type of the server's currently executed task after generating the server's power consumption management policy based on the server's current load data and load change trend; and correct the server's power consumption management policy based on the task type.

[0125] It should be noted that, for the description of the features in the embodiment corresponding to the server power consumption management device, reference can be made to the relevant description of the embodiment corresponding to the server power consumption management method, which will not be repeated here.

[0126] According to the server power consumption management device proposed in the embodiment of the present application, the future load change trend of the server can be predicted based on the load data of the server at the current moment, and the power consumption management strategy of the server can be generated based on the load data and load change trend at the current moment. The operating parameters of at least one component among multiple components are adjusted using the power consumption management strategy. By predicting in advance and adjusting the power consumption management strategy in real time, dynamic adaptation to the load fluctuations of the server can be achieved, avoiding insufficient performance of the server at high load or waste of power consumption at low load due to a fixed static power consumption management strategy. In addition, the energy efficiency of the server can be increased by predicting in advance, thereby optimizing the resource allocation of the server and improving the energy efficiency of the server.

[0127] The embodiment of the present application also provides an electronic device, such as Figure 4 As shown, it includes a memory 401 and a processor 402. The memory 401 stores a computer program, and the processor 402 is configured to run the computer program to execute the steps in any of the above server power consumption management method embodiments.

[0128] An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored. The computer program is configured to execute the steps of any of the above-mentioned server power consumption management method embodiments when running.

[0129] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.

[0130] An embodiment of the present application further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of any of the above-mentioned server power consumption management method embodiments are implemented.

[0131] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0132] The above is a detailed introduction to the server power consumption management provided by this application. This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of this application, several improvements and modifications can be made to this application, and these improvements and modifications also fall within the scope of protection of the claims of this application.

Claims

1. A server power consumption management method, characterized in that: The server includes multiple components, each of which includes multiple processors that perform heterogeneous computing tasks, wherein the method includes the following steps: Obtaining the load data of the server at the current moment; Predict the future load change trend of the server based on the current server load data; A power consumption management strategy for the server is generated according to the load data of the server at a current moment and the load change trend, and the operating parameters of at least one component among the multiple components are adjusted using the power consumption management strategy.

2. The server power consumption management method according to claim 1, wherein: The predicting of the future load change trend of the server based on the current server load data includes: The load data of the server at the current moment is input into a power consumption prediction model, and the power consumption prediction model outputs a future load change trend of the server, wherein the power consumption prediction model is obtained based on training of a machine learning model.

3. The server power consumption management method according to claim 2, characterized in that: Before inputting the load data of the server at the current moment into the power consumption prediction model, the method further includes: Obtaining historical load data of the server; A training data set is generated based on the historical load data, and a machine learning model is trained using the training data set, and the power consumption prediction model is obtained after the training is completed.

4. The server power consumption management method according to claim 1, wherein: Generating the power consumption management strategy of the server according to the load data of the server at the current moment and the load change trend includes: generating a first management policy for at least one component according to the load data of the server at the current moment; A second management strategy for at least one component is generated according to the load variation trend.

5. The server power consumption management method according to claim 4, characterized in that: The processor includes a central processing unit and a graphics processing unit, and generating a first management strategy of at least one component according to the load data of the server at the current moment includes: Identifying the actual load in the load data of the server at the current moment; If the actual load is greater than the load threshold, the first management strategy includes at least one of the following: shutting down at least one core of the central processing unit, reducing the operating frequency of the central processing unit, reducing the operating frequency of the graphics processor of the server, switching the graphics processor to an idle state, and reducing the memory voltage and storage device power consumption of the server; If the actual load is less than or equal to the load threshold, the first management strategy includes at least one of the following: turning on multiple cores of the central processing unit, increasing the operating frequency of the central processing unit, increasing the operating frequency of the graphics processing unit, and improving the memory bandwidth and storage input and output performance of the server.

6. The server power consumption management method according to claim 4, characterized in that: Generating a second management strategy for at least one component according to the load change trend includes: Identify inflection points in load change trends; An adjustment time is determined according to the inflection point, and the first management strategy is adjusted to the second management strategy at the adjustment time.

7. The server power consumption management method according to claim 1, characterized in that: After generating the power consumption management strategy of the server according to the load data of the server at the current moment and the load change trend, the method further includes: Identifying the task type of the currently executed task of the server; A power consumption management policy of the server is modified based on the task type.

8. A server power consumption management device, characterized in that: The server includes multiple components, and the multiple components include multiple processors that perform heterogeneous computing tasks, wherein the apparatus includes: An acquisition module, used to obtain the load data of the server at the current moment; The prediction module is used to predict the future load change trend of the server based on the current server load data; An adjustment module is configured to adjust an operating parameter of at least one component among the multiple components using the power consumption management strategy.

9. An electronic device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the steps of the server power consumption management method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of the server power consumption management method according to any one of claims 1 to 7 are implemented.

Images