Server energy consumption control method, system, device, storage medium and program product

Abstract

The application discloses a kind of server energy consumption control method, system, equipment, storage medium and program product, it is related to server management technical field, method includes: the system index data of server is input into prediction model, first threshold and second threshold are predicted using prediction model;If the total load data of server is less than first threshold, control first target equipment to enter energy-saving state;If total load data is greater than second threshold, control second target equipment to enter working state.Can solve the problem that server energy consumption management can cause power supply frequently switch, increase fault risk, and energy-saving efficiency is lower.The method determines whether to carry out energy-saving operation according to two threshold values to determine the load state of server, realizes the intelligent optimization of server energy consumption, improves energy-saving efficiency.In addition, the weight coefficient of prediction model makes two threshold values can be automatically adjusted according to the actual operation of server, adapt to different business load mode and system state.

Description

Technical Field

[0001] This invention relates to the field of server management technology, specifically to a server energy consumption control method, system, device, storage medium, and program product. Background Technology

[0002] With the development of new technologies such as cloud computing and big data, the reliability requirements for servers are becoming increasingly stringent. Therefore, servers typically employ dual power supplies with redundancy during operation. This is so that if one power supply fails during operation, the other can continue to supply power to the server, preventing unexpected shutdowns due to insufficient power.

[0003] Currently, server scale is growing exponentially, resulting in enormous energy consumption. By implementing energy management for servers, enabling them to operate at low power consumption under non-full load conditions, this refined energy management strategy can bring significant energy savings to data centers with massive server clusters. Current energy management strategies set a switching threshold; when server energy consumption falls below this threshold, backup power is shut down. However, this leads to frequent power switching when server load fluctuates around the threshold, potentially causing insufficient system power and increasing the risk of failure. Furthermore, energy management strategies only schedule the server's power modules or virtual machine operations, without coordinating with other components such as the server's CPU, memory, and hard drive, resulting in relatively low energy efficiency. Summary of the Invention

[0004] In view of this, the present invention provides a server energy consumption control method, system, device, storage medium and program product to solve the problems that server energy consumption management leads to frequent power switching, increases the risk of failure and has low energy efficiency.

[0005] Firstly, this application provides a server energy consumption control method, the method comprising:

[0006] Obtain the server's total load data and system metrics data;

[0007] The system indicator data is input into the prediction model to obtain the first threshold and the second threshold. The first threshold is less than the second threshold. The first threshold and the second threshold are determined by the prediction model based on the weight coefficients and the system indicator data. The weight coefficients are determined based on the weight coefficient adjustment strategy, which is determined by a preset algorithm based on the server's status information.

[0008] When the total load data is less than the first threshold, the first target device in the control server enters the energy-saving state.

[0009] If the total load data exceeds the second threshold, the second target device in the control server enters the working state.

[0010] Secondly, this application provides a server energy consumption control system, which includes: a control device, a first target device, and a second target device;

[0011] The control device is used to acquire the server's total load data and system indicator data; input the system indicator data into the prediction model to obtain a first threshold and a second threshold, wherein the first threshold is less than the second threshold. The first threshold and the second threshold are determined by the prediction model based on the weight coefficients and the system indicator data. The weight coefficients are determined based on the weight coefficient adjustment strategy, which is determined by a preset algorithm based on the server's status information.

[0012] The control device is connected to the first target device and is also used to control the first target device in the server to enter an energy-saving state when the total load data is less than a first threshold.

[0013] The control device is connected to the second target device and is also used to control the second target device in the server to enter the working state when the total load data is greater than the second threshold.

[0014] Thirdly, this application provides a server power consumption control device, which includes:

[0015] The data acquisition module is used to acquire the server's total load data and system metric data;

[0016] The threshold determination module is used to input system indicator data into the prediction model to obtain a first threshold and a second threshold. The first threshold is less than the second threshold. The first threshold and the second threshold are determined by the prediction model based on the weight coefficients and system indicator data. The weight coefficients are determined based on the weight coefficient adjustment strategy, which is determined by a preset algorithm based on the server's status information.

[0017] The first control module is used to control the first target device in the server to enter an energy-saving state when the total load data is less than a first threshold.

[0018] The second control module is used to control the second target device in the server to enter the working state when the total load data is greater than the second threshold.

[0019] Fourthly, this application provides a computer device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the server energy consumption control method of the first aspect or any corresponding embodiment described above.

[0020] Fifthly, this application provides a computer-readable storage medium storing computer instructions for causing a computer to execute the server power consumption control method of the first aspect or any corresponding embodiment described above.

[0021] Sixthly, this application provides a computer program product, including computer instructions for causing a computer to execute the server energy consumption control method described in the first aspect or any corresponding embodiment.

[0022] This application inputs server system indicator data into a predictive model, which then predicts dynamic first and second thresholds. If the server's total load data is less than the first threshold, the first target device is controlled to enter an energy-saving state; if the total load data is greater than the second threshold, the second target device is controlled to enter an operating state. This solves the problem that server energy management leads to frequent power switching, increasing the risk of failure, and resulting in low energy efficiency. This method accurately determines the server load status based on the first and second thresholds, deciding whether to perform energy-saving operations, thus achieving intelligent optimization of server energy consumption, improving energy efficiency, and reducing server operating costs. Furthermore, by using a preset algorithm and server status information to adjust the weight coefficients of the predictive model, the first and second thresholds can be automatically adjusted according to the actual server operation, adapting to different business load patterns and system states, reducing manual intervention, and improving the intelligence level of energy management. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the specific embodiments or related technologies of this application, the drawings used in the description of the specific embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating a server power consumption control method according to an embodiment of this application;

[0025] Figure 2 This is a flowchart of server redundant power supply intelligent switching and energy consumption optimization according to an embodiment of this application;

[0026] Figure 3 This is a timing diagram of the zero-interruption switching operation according to an embodiment of this application;

[0027] Figure 4 This is a schematic diagram illustrating the generation of the first threshold and the second threshold according to an embodiment of this application;

[0028] Figure 5This is a schematic diagram of the server energy consumption control system according to an embodiment of this application;

[0029] Figure 6 This is a schematic diagram of a dual-channel communication mechanism according to an embodiment of this application;

[0030] Figure 7 This is a schematic diagram of the device functions in the server energy consumption control system according to an embodiment of this application;

[0031] Figure 8 This is a schematic diagram of a three-level optimization strategy according to an embodiment of this application;

[0032] Figure 9 This is a structural block diagram of a server power consumption control device according to an embodiment of this application;

[0033] Figure 10 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of this application. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0035] With the development of emerging technologies such as cloud computing and big data, the reliability requirements for servers are becoming increasingly stringent. Therefore, servers typically employ dual-power redundant supply systems. If one power supply fails during operation, the other can continue to power the system, preventing unexpected shutdowns due to insufficient power. As internet technology continues to evolve, the scale of servers in data center server rooms is growing exponentially, resulting in enormous annual power consumption. Through targeted technical optimization, servers can maintain low power consumption even under non-full load conditions. This refined energy control strategy allows data centers with massive server clusters to save significant amounts of electricity.

[0036] Currently, there are two methods for controlling server energy consumption. One method sets a fixed power consumption threshold. When the sum of the output power of the main power supply and the backup power supply is less than the threshold, the backup power supply is turned off, putting it into standby mode. When the sum of the output power of the main power supply and the backup power supply is greater than the threshold, the backup power supply is restarted, with the main power supply and backup power supply sharing the load equally. The other method first monitors and collects real-time energy consumption data for the server. Then, based on the energy consumption data, it performs server energy consumption analysis and modeling to gain a deeper understanding of energy consumption, reduce energy costs, and achieve sustainable energy management. Finally, it uses energy consumption optimization algorithms on the server energy consumption model to obtain energy consumption optimization strategies, automatically adjusting server power consumption and configuration. However, the above methods suffer from threshold rigidity and insufficient energy-saving efficiency. Using 2 / 3 of the rated power as a fixed power consumption threshold does not consider load fluctuation characteristics, leading to frequent power switching when the load fluctuates around the threshold, increasing the risk of failure. Furthermore, the backup power supply requires milliseconds to respond from standby to startup, which may cause insufficient system power supply during this period. Optimizing only the power supply module in the server or scheduling virtual machines without coordinating with components such as the central processing unit, memory, and hard drive results in an overall energy saving efficiency of less than 30% under light load. It also lacks predictive capabilities, relying on real-time monitoring rather than load forecasting, and cannot anticipate the risk of overload due to sudden load surges.

[0037] Based on the above, this application provides a server energy consumption control method. It dynamically generates power switching threshold ranges using an LSTM (Long Short-Term Memory) neural network model to accurately determine the system load status and then implement coordinated energy-saving operations or zero-interruption switching operations. Under low load, coordinated energy-saving measures such as shutting down redundant devices and reducing hard drive speed can effectively reduce energy waste. Under high load, zero-interruption switching ensures timely activation of the backup power supply, avoiding overload of the main power supply. Furthermore, during the switching process between the main and backup power supplies, a PID (Proportion-Integration-Differential) controller smoothly adjusts the output ratio of the main and backup power supplies, ensuring seamless operation of the entire switching process and uninterrupted server service. Based on a reinforcement learning framework, the dynamic threshold weight coefficients in the neural network model are optimized online, enabling the power switching threshold to automatically adjust according to the actual server operating conditions, adapting to different business load modes and system states. This method can accurately determine the system load status and then implement coordinated energy-saving operations or zero-interruption switching operations, achieving intelligent optimization of server energy consumption, significantly reducing server operating costs, and improving energy utilization efficiency.

[0038] According to an embodiment of this application, a server energy consumption control method embodiment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, for example, a computer, a server, etc., and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0039] This embodiment provides a server power consumption control method. Figure 1 This is a flowchart of a server energy consumption control method according to an embodiment of this application, such as... Figure 1 As shown, the process includes the following steps:

[0040] Step S101: Obtain the server's total load data and system metric data.

[0041] Specifically, during server operation, the server's total load data and system metrics are monitored in real time. Total load data includes, for example, the server's total output power ∑. P out System metrics data include: historical server load time-series data, business cycle characteristic identifiers, and real-time system metrics. Business cycle characteristic identifiers include: weekday pattern identifiers and holiday pattern identifiers. Real-time system metrics include: CPU utilization, network throughput, and memory usage. When collecting historical server load time-series data, the sampling interval should be ensured to be within 1 second to accurately grasp subtle trends in load changes.

[0042] Step S102: Input the system indicator data into the prediction model to obtain a first threshold and a second threshold. The first threshold is less than the second threshold. The first threshold and the second threshold are determined by the prediction model based on the weight coefficients and the system indicator data. The weight coefficients are determined based on the weight coefficient adjustment strategy, which is determined by a preset algorithm based on the server's status information.

[0043] Specifically, the prediction model is, for example, a pre-built and trained three-layer LSTM neural network model. The input layer of this model has 64 neurons to receive a load window of data with a time sequence length of 120 seconds; the hidden layer contains 128 LSTM units with a dropout rate set to 0.2 to prevent overfitting; and the output layer has 3 neurons, each used to determine a first threshold. Second threshold and the real-time output power of the main power supply The value. Real-time output power is typically read from the main power module via the PMBus (Power Management Bus).

[0044] In addition to using the prediction model to predict the first and second thresholds, the weight coefficients in the prediction model are also optimized online using a preset algorithm. A preset algorithm, such as Q-Learning, is a reinforcement learning algorithm. The preset algorithm can determine the operation on the weight coefficients based on the server's state information, such as increasing the weight coefficient by 10% or decreasing it by 5%. This operation on the weight coefficients is called the weight coefficient adjustment strategy. The initial weight coefficients in the prediction model are adjusted according to the weight coefficient adjustment strategy to obtain the final weight coefficients. Furthermore, the preset algorithm can update the weight coefficients of the prediction model at regular intervals, such as every 5 minutes or 10 minutes.

[0045] The system indicator data is input into the prediction model, which dynamically generates a power switching threshold range based on the weighting coefficients and the system indicator data. , The two boundary values ​​of the power switching threshold range are the first threshold. Second threshold The first threshold is less than the second threshold.

[0046] The above process is as follows Figure 2 As shown, dynamic threshold update; model prediction; reinforcement learning optimizes weight coefficients, and updating weight coefficients is shown for example: , , , .

[0047] Step S103: When the total load data is less than the first threshold, control the first target device in the server to enter the energy-saving state.

[0048] Specifically, the first target device includes, for example, a backup power supply, hard drive, redundant fans, and a central processing unit. When the total load data is less than a first threshold, for example: ∑ P out < It performs coordinated energy-saving operations, controlling the primary target device in the server to enter an energy-saving state. For example: sending a standby command to the backup power supply to put it into a cold redundancy state; synchronously triggering the hard drive to slow down to a preset safe speed, which is less than or equal to 5400 RPM; turning off redundant fans and adjusting the CPU to C1E energy-saving state. C1E energy-saving state is an automatic energy-saving mode that puts the CPU into a specific low-power sleep state, used to save energy when the CPU is idle.

[0049] The above process is as follows Figure 2As shown, the server's total output power is monitored in real time to determine whether the total output power is less than the first threshold. If it is less, collaborative energy-saving operations are performed, including cold redundancy, hard disk speed reduction, and fan control. A deep energy-saving check is also performed, and if the conditions are met, memory de-voltage and virtual machine migration are implemented.

[0050] Step S104: If the total load data is greater than the second threshold, control the second target device in the server to enter the working state.

[0051] Specifically, the second target equipment includes, for example, the main power supply, backup power supply, and supercapacitor buffer module. This occurs when the total load data exceeds a second threshold, for example: ∑ P out > The system performs a zero-interruption switchover operation, controlling the second target device in the server to enter the working state. For example: activating the supercapacitor buffer module for pre-charging; sending a start command to the backup power supply after the supercapacitor output voltage reaches its nominal value; and smoothly adjusting the output ratio of the main and backup power supplies through a PID controller. The above process is as follows: Figure 2 As shown, the process determines whether the total output power exceeds the second threshold. If it does, it performs zero-interrupt switching; supercapacitor pre-charging; backup power supply startup; and PID smooth switching of the main and backup power supply output ratio. The above process is as follows: Figure 3 As shown, after the detection module detects that the total load data is greater than the second threshold, it starts the supercapacitor for pre-charging within 5ms; after the supercapacitor is charged to 95% of the nominal voltage, it sends a start command to the backup power supply; the backup power supply starts the voltage rise control, and the PID controller adjusts the output ratio of the main power supply and the backup power supply; the backup power supply smoothly intervenes in the system load.

[0052] This embodiment provides a server energy consumption control method. The method inputs server system indicator data into a predictive model, which then predicts dynamic first and second thresholds. If the server's total load data is less than the first threshold, a first target device is controlled to enter an energy-saving state; if the total load data is greater than the second threshold, a second target device is controlled to enter an operating state. By accurately determining the server load status based on the first and second thresholds, and deciding whether to perform energy-saving operations, intelligent optimization of server energy consumption is achieved, improving energy efficiency and reducing server operating costs. Furthermore, by using a preset algorithm and server status information to adjust the weight coefficients of the predictive model, the first and second thresholds can be automatically adjusted according to the actual server operation, adapting to different business load modes and system states. This reduces manual intervention and improves the intelligence level of energy consumption management. This solves the problems of frequent power switching, increased failure risk, and low energy efficiency caused by server energy consumption management.

[0053] As an optional embodiment, system indicator data is input into the prediction model to obtain a first threshold and a second threshold, including:

[0054] Based on system metrics data, determine the load mean, load standard deviation, and load threshold;

[0055] Obtain the weight coefficient adjustment strategy and determine the weight coefficients in the prediction model based on the weight coefficient adjustment strategy;

[0056] The first data item is determined based on the load threshold and the first preset parameter;

[0057] The second and third data items are determined based on the load mean, load standard deviation, and weighting coefficients.

[0058] The larger value between the first data item and the second data item is used as the first threshold.

[0059] The fourth data item is determined based on the load threshold and the second preset parameter;

[0060] The smaller value between the third and fourth data items is used as the second threshold.

[0061] Specifically, this embodiment optimizes the weight coefficients in the prediction model online using a preset algorithm. The preset algorithm, for example, is the Q-learning algorithm. The preset algorithm can determine the operation for the weight coefficients based on the server's status information, such as increasing the weight coefficients by 10% or decreasing them by 5%. This operation on the weight coefficients is the weight coefficient adjustment strategy. The initial weight coefficients in the prediction model are adjusted according to the weight coefficient adjustment strategy to obtain the final weight coefficients. Furthermore, the preset algorithm can update the weight coefficients of the prediction model at regular intervals, such as every 5 minutes, 10 minutes, etc.

[0062] Determine the average load over T seconds based on system metrics data. and load standard deviation Determine the rated output power of the main unit. This refers to the maximum power that the main power supply can continuously output under standard operating conditions. As a load threshold.

[0063] The first preset parameter is, for example, 0.3, 0.4, or other values, depending on the load threshold. And the first preset parameter, determine the first data item, for example: the first data item is .

[0064] Weighting coefficients, for example: , , , Based on the average load Load standard deviation And weighting coefficients, to determine the second and third data items, for example: the second data item is... The third data item is .

[0065] The larger value between the first data item and the second data item is used as the first threshold. For example, as shown in formula (1), and The larger value in the range is used as the first threshold. .

[0066] (1)

[0067] The second preset parameter is, for example, 0.9, 0.8, or other values, depending on the load threshold. The second preset parameter determines the fourth data item, for example: the fourth data item is .

[0068] The smaller value between the third and fourth data items is used as the second threshold. For example, as shown in formula (2), and The smaller value is used as the second threshold. .

[0069] (2)

[0070] For example, a pre-built and trained three-layer LSTM neural network model. The input layer has 64 neurons to receive a 120-second load window of data; the hidden layer contains 128 LSTM units with a dropout rate of 0.2 to prevent overfitting; and the output layer has three neurons, each used to determine a first threshold. Second threshold Additionally, it can output a backup output, which can be another set. and If an error exists in the first group, such as when the first threshold is greater than the second threshold, a backup output can be used. The above process is as follows: Figure 4 As shown, historical load data, business cycle identifiers, and real-time system indicators are input into the model input layer, and the model is calculated using 128 LSTM units contained in the model hidden layer. The model output layer outputs the first threshold, the second threshold, and the backup output.

[0071] In this embodiment, the weight coefficients of the prediction model are adjusted by a preset algorithm and server status information, so that the first threshold and the second threshold can be automatically adjusted according to the actual operation of the server. The system load status can be accurately determined by the first threshold and the second threshold, and then collaborative energy-saving operation or zero-interruption switching operation can be taken to improve the intelligence level of energy consumption management.

[0072] As an optional embodiment, obtaining a weight coefficient adjustment strategy and determining the weight coefficients in the prediction model based on the weight coefficient adjustment strategy includes:

[0073] Obtain the server's status information and determine the weight coefficient adjustment strategy based on the preset algorithm and the server's status information;

[0074] Candidate weight coefficients are determined based on the weight coefficient adjustment strategy and the initial weight coefficients;

[0075] Determine the change in candidate weight coefficients from the initial weight coefficients, and determine the ratio of the change to the initial weight coefficients;

[0076] If the ratio is greater than the ratio threshold, a weighting coefficient is generated based on the initial weighting coefficient and the ratio threshold.

[0077] If the ratio is less than or equal to the ratio threshold, the candidate weight coefficient will be used as the weight coefficient.

[0078] Specifically, this embodiment optimizes the weight coefficients in the prediction model online using a preset algorithm. An example of such a preset algorithm is the Q-learning algorithm. The preset algorithm can determine the operations for the weight coefficients based on the server's status information; these operations constitute the weight coefficient adjustment strategy. Examples of weight coefficients include: , , , Weighting coefficient adjustment strategies, for example: Increase by 0.1, and Increase by 0.2, and Decrease by 0.3, Reduce by 0.2, etc. Additionally, the preset algorithm can update the weight coefficients of the prediction model at regular intervals, such as 5 minutes, 10 minutes, etc.

[0079] Candidate weight coefficients are obtained by adjusting the initial weight coefficients in the prediction model according to the weight coefficient adjustment strategy. For example, the initial weight coefficients include... =0.8、 =0.4、 =0.6、 =0.5, the weighting coefficient adjustment strategy is to... Increase by 0.1, and Increase by 0.2, and Decrease by 0.3, Decrease by 0.2, then the candidate weight coefficients include =0.9、 =0.6、 =0.3、 =0.3.

[0080] To prevent oscillations in the weighting coefficients, this embodiment introduces a restriction: the single adjustment range of each weighting coefficient shall not exceed a preset proportion of the weighting coefficient. The preset proportion may be, for example, 10%, 15%, or other proportions. This preset proportion is the threshold for adjusting the weighting coefficient each time.

[0081] Determine the change in candidate weight coefficients compared to initial weight coefficients, and determine the ratio of this change to the initial weight coefficients. For example, the initial weight coefficients include... =0.8、 =0.6、 =0.5, candidate weight coefficients include =0.85、 =0.6、 =0.3、 =0.3, with corresponding changes of 0.05, 0.2, 0.3, and 0.2, respectively. The ratios of these changes to the initial weighting coefficients are 6.25%, 50%, 50%, and 40%, respectively.

[0082] If the proportion value is greater than the proportion threshold, a weighting coefficient is generated based on the initial weighting coefficient and the proportion threshold. For example, if the proportion threshold is 10%, the initial weighting coefficient... The proportion is 50%. The proportion is 40%. The corresponding weighting coefficient needs to be generated based on the initial weighting coefficient and the proportion threshold. The initial weighting coefficient... =0.5, then the weighting coefficients are... =0.5 =0.55.

[0083] When the proportion is less than or equal to the proportion threshold, the candidate weight coefficient is used as the weight coefficient. For example, if the proportion threshold is 10%, the initial weight coefficient is... The proportion is 6.25%, which is less than 10%, and the candidate weight coefficients include... =0.85, then the weighting coefficients are... =0.85.

[0084] After continuous operation, the initial weight coefficients... , , , It will gradually converge to the value range most suitable for the current server cluster business mode, for example: for high-frequency fluctuating businesses, =0.8、 =1.2, a relatively wide range, reducing handover; stable long-cycle services: =1.5、 =0.6, narrow range, fast response.

[0085] In this embodiment, the dynamic threshold weight coefficients in the prediction model are optimized online based on a preset algorithm, so that the first threshold and the second threshold can be automatically adjusted according to the actual operation of the server to adapt to different business load modes and system states in the server.

[0086] As an optional embodiment, a weight coefficient adjustment strategy is determined based on a preset algorithm and server status information, including:

[0087] Based on the server's status information, determine whether the number of times the second target device's status has switched exceeds a first preset threshold.

[0088] If the number of switching times exceeds the first preset threshold, a weight coefficient adjustment strategy is generated based on the first initial strategy and the preset algorithm.

[0089] Based on the server's status information, determine whether the server's energy saving is less than the second preset threshold;

[0090] If the energy saving is less than the second preset threshold, a weight coefficient adjustment strategy is generated according to the second initial strategy and the preset algorithm.

[0091] Specifically, in this embodiment, the weighting coefficients of the prediction model include , , , There are four parameters in total, and two initial strategies are set for adjusting the weight coefficients. The first initial strategy is: when the number of times the second target device state switches is greater than the first preset threshold, or when the number of switches is close to the set maximum number of switches, the weight coefficient is increased first. and This achieves the goal of widening the difference between the first and second thresholds, thereby reducing the switching frequency. The second initial strategy is: when the server's energy saving is less than the second preset threshold, or when the energy saving significantly decreases, the weighting coefficient is appropriately reduced. and This improves sensitivity and facilitates the triggering of energy-saving operations.

[0092] The first preset threshold is, for example, 500, 1000, or other values. Based on the server's status information, the number of times the second target device's status needs to be switched is determined. It is then determined whether the number of switches exceeds the first preset threshold. If the number of switches exceeds the first preset threshold, a weight coefficient adjustment strategy is generated based on the first initial strategy and a preset algorithm. The weight coefficient adjustment strategy, for example, adjusts the weight coefficients in the initial weight coefficients... Increase by 0.1 Increase by 0.2.

[0093] The second preset threshold is, for example, 30%, 35%, or other values. Based on the server's status information, the current energy saving of the server is determined. It is then determined whether the server's energy saving is less than the second preset threshold. If the energy saving is less than the second preset threshold, a weighting coefficient adjustment strategy is generated based on the second initial strategy and the preset algorithm. The weighting coefficient adjustment strategy, for example, adjusts the initial weighting coefficient... Increase by 0.5 Increase by 0.3.

[0094] As an optional embodiment, the method further includes, before inputting system indicator data into the prediction model:

[0095] Determine the initial weight coefficients and create an initial model based on them;

[0096] The fifth data item is generated based on the energy saving, the energy saving threshold, and the first preset coefficient; the sixth data item is generated based on the number of switching, the number of switching threshold, and the second preset coefficient.

[0097] Generate a reward function based on the fifth and sixth data items;

[0098] Obtain historical load time-series data, historical business cycle characteristic identifiers, and historical system metrics for the server;

[0099] The initial model is trained based on historical load time-series data, historical business cycle characteristics, historical system indicators, and reward functions to obtain a prediction model.

[0100] Specifically, the initial weighting coefficients include , , , There are four parameters in total, which are used when the prediction model is first deployed. , , , The specific values ​​are set based on historical load analysis results and empirical values, for example: , The initial value is between 0.5 and 1.5, which affects the sensitivity of the threshold range to volatility; , Initial values ​​are recommended to be between 0.1 and 0.5, as this affects the magnitude of periodic adjustments. (In the initial weighting coefficients...) , , , The initial values ​​are determined through offline simulation to ensure that frequent switching between primary and backup power supplies or response delays are not caused. Initial weighting coefficients are determined, for example: =0.8、 =0.6、 =0.5.

[0101] An initial model is created based on the initial weight coefficients. For example, the initial model is a three-layer LSTM neural network model. The input layer of this model has 64 neurons, which are used to receive load window data with a time length of 120 seconds. The hidden layer contains 128 LSTM units with a dropout rate of 0.2. The output layer has 3 neurons. The weight coefficients of the three-layer LSTM neural network model are set to the initial weight coefficients mentioned above.

[0102] First preset coefficient Second preset coefficient The specific values ​​are set according to the experiment or actual needs, for example: =0.6, =0.4; =0.8, =0.6 etc. The ratio of energy saved to the maximum possible energy saved (i.e., the energy saved threshold) multiplied by a first preset coefficient. Subtract the ratio of the number of handovers to the maximum allowed number of handovers (i.e., the handover threshold) multiplied by the second preset coefficient. The reward function is obtained, which can strike a balance between energy saving and switching frequency, ensuring that the system can effectively save energy while reducing the potential risks caused by frequent switching. Therefore, a fifth data item is generated based on the energy saved, the energy saved threshold, and a first preset coefficient; a sixth data item is generated based on the number of switching, the switching frequency threshold, and a second preset coefficient; and the reward function is generated based on the fifth and sixth data items. The reward function is, for example, formula (3).

[0103] (3)

[0104] in, To conserve energy, As the energy-saving threshold, For the number of switching times, This is the threshold for the number of switching operations. The fifth data item is... The sixth data item is .

[0105] Acquire historical load time-series data of the server over a period of time, with a sampling interval of 0.5 seconds. Record historical business cycle characteristics of the server, such as weekday mode, holiday mode, etc. Simultaneously record historical system metrics of the server, such as CPU utilization, network throughput, memory usage, etc. Using the historical load time-series data, historical business cycle characteristics, historical system metrics, and a reward function, train the initial model. The training process does not require manual labeling of standard values ​​and employs a self-supervised learning mechanism to obtain the prediction model.

[0106] In this embodiment, the ratio of energy saving to energy saving threshold multiplied by a first preset coefficient is used, and the ratio of switching count to switching count threshold multiplied by a second preset coefficient is subtracted. A reward function is generated. The initial model is trained using this reward function, historical load time-series data, historical business cycle characteristic identifiers, and historical system indicators. This allows the predictive model to output accurate first and second thresholds, achieving a balance between energy-saving effects and the number of switching operations. This ensures that the server effectively saves energy while reducing the potential risks associated with frequent switching.

[0107] As an optional embodiment, the second target device in the control server enters the working state, including:

[0108] Get the server's system time;

[0109] The target index is determined based on the system time and the preset time constant;

[0110] The sixth data item is generated based on preset parameters and the target index;

[0111] The product of the sixth data item and the preset ratio coefficient is used as the target ratio;

[0112] The output ratio of the first power supply and the second power supply is controlled to a target ratio, wherein the first power supply and the second power supply are included in the second target device.

[0113] Specifically, obtain the server's system time. The preset time constant is... The preset proportional coefficient is The preset time constant and preset parameters can be set according to experimental or actual needs, for example: =2, 3, 4 or other values, =0.1, 0.3, 1.5, or other values. Determine the target index based on the system time and the preset time constant. For example, the target index could be: .

[0114] Preset parameters, for example: 1, 2, or other values. Based on the preset parameters and the target index, a sixth data item is generated. The sixth data item is, for example: .

[0115] The product of the sixth data item and the preset proportional coefficient is used as the target proportion, as shown in formula (4).

[0116] (4)

[0117] The first and second power supplies are included in the second target device. The first power supply is the main power supply of the server, and the second power supply is the backup power supply of the server. During the zero-interruption switching operation, the output ratio of the first and second power supplies is smoothly adjusted within 10-100ms using a PID controller to control the output ratio of the first and second power supplies to the target ratio, avoiding system shock or instability caused by sudden changes in power output.

[0118] In this embodiment, the output ratio of the primary and backup power supplies is smoothly adjusted by a PID controller, and the entire switching process is seamlessly connected, ensuring uninterrupted operation of server services. This method achieves a smooth transition, ensures zero interruption in the switching process, and guarantees the continuity and reliability of server services.

[0119] As an optional embodiment, the above-described server energy consumption control method may further include steps A1 to A4, wherein steps A1 to A4 are used to perform conduction tests on the redundant power modules in the server using steps S101 to S104.

[0120] Step A1: Obtain the number of redundant power supply modules in the server.

[0121] Specifically, in this embodiment, both the main power supply and backup power supply of the server include one or more redundant power supply modules. Therefore, it is necessary to perform conduction tests on the redundant power supply modules in the server.

[0122] The method involves obtaining the power supply type and number of redundant power supply modules for the server. Common server power supply methods include three types of redundancy: redundant cold backup, parallel current sharing N+1 backup power supply, and redundant hot backup. Therefore, this embodiment needs to clearly identify the type and number of redundant power supply modules for the server to be tested. For example, in N+1 redundancy, the number of redundant power supply modules is 1.

[0123] Step A2: Calculate the conduction test drop voltage based on the number of redundant power modules in the server.

[0124] Specifically, the number of redundant power supply modules required to operate when one of the power supplies supplying the server fails is calculated based on the number of redundant power supply modules. The corresponding voltage drop value is then calculated based on the number of redundant power supply modules involved in power supply. For example, using a 1+1 power supply, if the redundant power supply module supplying the server fails, another redundant power supply module is needed to ensure the stability of the server's power output. If the 1+1 power supply uses a parallel current-sharing method to supply power to the server simultaneously, then when one of the redundant power supply modules fails, the output current of the redundant power supply module will double, assuming the server power remains constant. Therefore, to simulate a redundant power supply module failure, this embodiment uses a method of halving the voltage supplying the server's redundant power supply modules. This causes both power supplies to double their current while the voltage is halved and the power remains constant, simulating the electromagnetic effect (conduction test) caused by the current doubling when a redundant power supply module fails. Therefore, the corresponding voltage drop value needs to be calculated based on the number of redundant power supply modules.

[0125] Step A3: Using a voltage drop simulation device, the power supply voltage of the server is subjected to a voltage drop, and the current value of the server power supply is obtained and the corresponding conducted interference value is calculated.

[0126] Specifically, after obtaining the corresponding voltage drop value, a voltage drop simulation device is used to perform a voltage drop test on the server's power supply voltage. After the voltage drop test, and for a period of time following the voltage drop and the subsequent increase in power supply current, the conducted interference value generated by the current in the current server is calculated. The calculated conducted interference value is compared with the corresponding rules, and the comparison result is used as the conducted interference test result.

[0127] In addition, when simulating a voltage drop in the server's power supply voltage using a voltage drop simulation device, a drop period needs to be set. The specific period is adjusted based on the type and performance of the redundant power supply. Because it simulates the change in the output current of the redundant power supply when the server's power supply fails, the voltage drop simulation device drops the server's power supply voltage according to a predetermined period, rather than directly reducing the voltage of the simulation device to the corresponding value.

[0128] In this embodiment, the corresponding conduction detection can be achieved without increasing the server's power consumption. Furthermore, it can accurately determine whether the server's conducted readings meet the conduction rules when the redundant power supply fails.

[0129] According to an embodiment of this application, a server energy consumption control system is provided, such as... Figure 5 As shown, the system includes: a control device, a first target device, and a second target device;

[0130] The control device is used to acquire the server's total load data and system indicator data; input the system indicator data into the prediction model to obtain a first threshold and a second threshold, wherein the first threshold is less than the second threshold. The first threshold and the second threshold are determined by the prediction model based on the weight coefficients and the system indicator data. The weight coefficients are determined based on the weight coefficient adjustment strategy, which is determined by a preset algorithm based on the server's status information.

[0131] The control device is connected to the first target device and is also used to control the first target device in the server to enter an energy-saving state when the total load data is less than a first threshold.

[0132] The control device is connected to the second target device and is also used to control the second target device in the server to enter the working state when the total load data is greater than the second threshold.

[0133] Specifically, control devices include, for example, a baseboard management controller and a central processing unit (CPU). During server operation, these control devices monitor the server's total load data and system metrics in real time. Total load data includes, for example, the server's total output power ∑. P out System metrics data include: historical server load time-series data, business cycle characteristic identifiers, and real-time system metrics. Business cycle characteristic identifiers include: weekday mode identifiers and holiday mode identifiers. Real-time system metrics include: CPU utilization, network throughput, and memory usage. When collecting historical server load time-series data, the sampling interval must be less than 1 second.

[0134] Predictive models, for example, include pre-built and trained three-layer LSTM neural network models. The control device inputs system performance data into the predictive model, which dynamically generates power switching threshold ranges based on weight coefficients and the system performance data. , The two boundary values ​​of the power switching threshold range are the first threshold. Second threshold The first threshold is less than the second threshold. In addition to using a prediction model to predict the first and second thresholds, a preset algorithm is used to optimize the weight coefficients in the prediction model online. The preset algorithm can determine the weight coefficient adjustment strategy based on the server's status information. The initial weight coefficients in the prediction model are then adjusted according to the weight coefficient adjustment strategy to obtain the final weight coefficients.

[0135] like Figure 5As shown, the control device is connected to the first target device and the second target device via a communication channel. The communication channel can employ a dual-channel communication mechanism, including a primary communication channel and a backup communication channel. For example, the primary communication channel transmits control commands via the PMBus bus, while the backup communication channel transmits a heartbeat signal via an independent GPIO (General-purpose input / output) bus, with a heartbeat signal frequency ≥100Hz. When the PMBus bus times out without response, it automatically switches to the backup communication channel. The switching method between the primary and backup communication channels is as follows: Figure 6 As shown, during the PMBus main channel command transmission process, it checks whether communication is normal. If communication is normal, the PMBus main channel is maintained. If the PMBus main channel communication times out, the GPIO backup channel is activated to transmit a heartbeat signal. The frequency of the heartbeat signal is ≥100Hz, and the control GPIO backup channel is maintained. The above content is as follows... Figure 7 As shown, the controller is a baseboard management controller, which employs dual-channel communication: a PMBus main channel and a GPIO backup channel. The frequency of the GPIO backup channel heartbeat signal is ≥100Hz. This dual-channel communication mechanism ensures reliable transmission of control commands.

[0136] For example, the first target device and the second target device: Figure 7 As shown, the first target device includes a temperature-adaptive heatsink, a virtual machine manager, a power module, etc.; the second target device includes a supercapacitor array on the power backplane, a power module, etc. The power module can transmit data to the power backplane through N+1 power management buses, where N is any positive integer.

[0137] When the total load data is less than a first threshold, the control device in the control server enters an energy-saving state. When the total load data is greater than a second threshold, the control device in the control server enters an operating state.

[0138] The server energy consumption control system provided in this embodiment inputs the server's system indicator data into a prediction model, which then predicts dynamic first and second thresholds. If the server's total load data is less than the first threshold, the control device controls a first target device to enter an energy-saving state; if the total load data is greater than the second threshold, the control device controls a second target device to enter an operating state. By accurately determining the server load status based on the first and second thresholds, and deciding whether to perform energy-saving operations, intelligent optimization of server energy consumption is achieved, improving energy efficiency and reducing server operating costs. Furthermore, by using a preset algorithm and server status information to adjust the weight coefficients of the prediction model, the first and second thresholds can be automatically adjusted according to the actual server operation, adapting to different business load modes and system states, reducing manual intervention, and improving the intelligence level of energy consumption management. This solves the problems of frequent power switching, increased failure risk, and low energy efficiency caused by server energy consumption management.

[0139] As an optional embodiment, the control device includes: an acceleration unit and a storage unit;

[0140] Storage unit, used to store prediction models;

[0141] The acceleration unit is used to calculate the first threshold and the second threshold based on the prediction model and system indicator data.

[0142] Specifically, the control device is illustrated using a baseboard management controller as an example. The baseboard management controller integrates an acceleration unit and a storage unit. The acceleration unit is a dedicated matrix multiplication unit, such as an NPU (Neural Processing Unit) accelerator. An NPU accelerator is a processor specifically designed for handling artificial intelligence and machine learning tasks, with a peak computing power ≥ 2 TOPS. The storage unit is, for example, ROM (Read-Only Memory). The controller's interrupt response latency is < 10 microseconds.

[0143] The prediction model, including its weight coefficients, is stored in a memory unit. The acceleration unit then calculates the first and second thresholds based on the prediction model and system indicator data.

[0144] The above content is as follows Figure 7 As shown, the substrate management controller includes an NPU accelerator, the peak computing power of the matrix multiplication unit of the NPU accelerator is ≥2 TOPS, and the above weighting coefficients are fixed in the read-only memory of the NPU accelerator.

[0145] As an optional embodiment, the first target device includes a first power supply, a first storage device, a redundant cooling device, and a computing device;

[0146] Control equipment is used to send a standby command to the first power supply and control the first power supply to enter a cold redundancy state.

[0147] The control device is also used to send energy consumption control instructions to the first storage device to control the energy consumption of the first storage device to be less than or equal to a third preset threshold.

[0148] The control equipment is also used to send shutdown commands to redundant cooling equipment to shut down the redundant cooling equipment;

[0149] The control device is also used to send status control commands to the computing device to control the computing device to enter the target energy-saving state.

[0150] Specifically, the first target device includes a first power supply, a first storage device, redundant cooling equipment, and a computing device. The first power supply is, for example, a backup power supply, which includes one or more redundant power modules supporting the PMBus protocol. The first storage device is, for example, a hard disk. The redundant cooling equipment is, for example, redundant fans. The computing device is, for example, a central processing unit (CPU).

[0151] The control device sends a Standby command (i.e., a standby command) to the primary power supply, controlling it to enter a cold redundancy state. The control device sends an energy consumption control command to the primary storage device, controlling its energy consumption to be less than or equal to a third preset threshold, such as 5W, 10W, or other values. When the primary storage device is a hard drive, the energy consumption control command can be a command to adjust the hard drive's rotational speed, reducing it to a preset safe rotational speed, which is less than or equal to 5400 RPM. The control device sends a shutdown command to the redundant cooling device, shutting it down. The control device sends a status control command to the computing device, controlling it to enter a target energy-saving state, such as C1E energy-saving state.

[0152] In this embodiment, when the server is under low load, coordinated energy-saving measures such as shutting down redundant devices and reducing hard drive speed can effectively reduce energy waste.

[0153] As an optional embodiment, the first target device includes a second storage device, a cooling device, and a virtual machine management device;

[0154] The control device is used to determine the duration for which the total load data is less than a fourth preset threshold. If the duration exceeds a fifth preset threshold, it sends a voltage regulation command to the second storage device to adjust the voltage of the second storage device to a preset voltage.

[0155] The control device is also used to send power regulation commands to the cooling device to control the power of the cooling device to be less than the sixth preset threshold.

[0156] The control device is also used to send virtual machine migration instructions to the virtual machine management device to migrate idle virtual machines to energy-saving nodes.

[0157] Specifically, the first target device includes a second storage device, a cooling device, and a virtual machine management device. The second storage device is, for example, memory. The cooling device is, for example, a heat dissipation system. The virtual machine management device is, for example, a virtual machine manager or a virtual machine cluster manager.

[0158] The fourth preset threshold is, for example, 0.2 × rated output power. 0.3× Or other values. The fifth preset threshold is, for example, 300 seconds, 350 seconds, or other durations.

[0159] Control equipment determines total load data ∑ P out If the duration is less than the fourth preset threshold, and if the duration exceeds the fifth preset threshold, a voltage regulation command is sent to the second storage device to adjust the second storage device to a preset voltage, such as a voltage less than or equal to 1.0V.

[0160] The control device sends a power regulation command to the cooling device, controlling the cooling device to enter a low-power mode, so that its power is less than a sixth preset threshold, such as 30%, 40% or other percentages of the rated power.

[0161] The control device sends virtual machine migration instructions to the virtual machine management device to migrate idle virtual machines to energy-saving nodes, concentrate resources, improve resource utilization efficiency, and further reduce overall energy consumption. Energy-saving nodes are, for example, servers with relatively low power consumption.

[0162] In this embodiment, based on factors such as the duration during which the total load data is less than a fourth preset threshold, the memory voltage and cooling system power are further optimized to achieve intelligent energy-saving linkage among multiple components. This not only improves the server's energy consumption performance under various operating conditions but also reduces manual intervention and enhances the system's intelligent management level.

[0163] As an optional embodiment, the second target device includes a first power supply, a second power supply, and a capacitor buffer device;

[0164] The control device is used to send a pre-charge command to the capacitor buffer device to activate the capacitor buffer device for pre-charging.

[0165] The control device is also used to send a start command to the first power supply to start the first power supply after the output voltage of the capacitor buffer device reaches the target value;

[0166] The control device is also used to send a first output control command to the first power supply and a second output control command to the second power supply, controlling the ratio between the output of the first power supply and the output of the second power supply to reach a target ratio, wherein the target ratio is determined according to a preset formula.

[0167] Specifically, the second target device includes a first power supply, a second power supply, and a capacitor buffer device. The first power supply is, for example, a backup power supply. The second power supply is, for example, a main power supply. The backup power supply includes one or more redundant power modules that support the PMBus protocol. The capacitor buffer device is, for example, a distributed supercapacitor array integrated on the power supply backplane. The capacitor array consists of multiple 2.7V / 100F supercapacitors connected in series to provide sufficient capacitance and voltage levels to meet the energy buffering requirements during power switching. Simultaneously, the capacitor array is equipped with a bidirectional DC / DC converter with an efficiency of over 95% to achieve efficient energy conversion between the supercapacitors and the power system. Furthermore, the capacitor array is equipped with an overvoltage protection circuit that triggers protection when the voltage reaches 110% of the nominal voltage (e.g., a trigger threshold of 12V) to prevent overvoltage damage to the supercapacitors and the entire power system.

[0168] In this embodiment, the control device performs a zero-interruption switching operation. Once the total load data is detected to be greater than the second threshold, the control device immediately sends a pre-charge command to the capacitor buffer device, activating the capacitor buffer device for pre-charging within 5ms. This ensures that the capacitor buffer device can quickly store enough energy to support subsequent power switching.

[0169] Once the output voltage of the capacitor buffer device reaches the target value, the control device sends a WorkOn command (start command) to the primary power supply, activating it to begin handling part of the load. This achieves seamless switching between primary and backup power supplies, ensuring uninterrupted server operation. The target value is, for example, 95% or 96% of the capacitor buffer device's nominal value, or other values.

[0170] During the zero-interruption switching operation, the control device sends a first output control command to the first power supply within 10-100ms. The output ratio of the first and second power supplies is smoothly adjusted to the target ratio by the PID controller to avoid impact or instability on the system due to sudden changes in power supply output. The target ratio is determined according to a preset formula, such as formula (4).

[0171] The above content is as follows Figure 7 As shown, the capacitor backplane includes a supercapacitor array consisting of eight 2.7V / 100F supercapacitors. The capacitor array is equipped with a bidirectional DC / DC converter with an efficiency of over 95% and an overvoltage protection circuit that triggers protection when the voltage reaches 110% of the nominal voltage.

[0172] In this embodiment, the output ratio of the first power supply and the second power supply is controlled to reach a target ratio, ensuring that the backup power supply is activated in a timely manner and avoiding overload of the main power supply. Overall, intelligent optimization of server energy consumption is achieved, improving energy utilization efficiency. Activating the capacitor buffer device for pre-charging further enhances the stability and reliability of the system, reducing the risk of server downtime due to power failure.

[0173] As an optional embodiment, the control device includes: a curve segmentation unit, a feature extraction unit, and a strategy generation unit;

[0174] The curve segmentation unit is used to obtain the server's load curve and segment the load curve into a preset number of load curve segments;

[0175] The feature extraction unit is used to extract preset feature information from the load curve segment;

[0176] The strategy generation unit is used to generate a first control strategy, a second control strategy, and a third control strategy according to a preset algorithm and preset feature information. The first control strategy is used to adjust the difference between the first threshold and the second threshold, the second control strategy is used to pre-activate the capacitor buffer device, and the third control strategy is used to start the first power supply in advance.

[0177] Specifically, a three-level optimization strategy was designed for the control equipment, executed by a short-term optimization layer, a medium-term optimization layer, and a long-term optimization layer. The short-term optimization layer performs PID control every 100ms, adjusting the output ratio of the first and second power supplies according to a preset formula. The medium-term optimization layer updates the weight coefficients in the prediction model every 5 minutes using a Q-learning algorithm. The long-term optimization layer performs load pattern clustering analysis every 24 hours to generate typical scenario templates.

[0178] This embodiment describes the long-term optimization layer of the control device, which includes a curve segmentation unit, a feature extraction unit, and a strategy generation unit.

[0179] The curve segmentation unit retrieves the server's load curve, such as the server's daily load curve. Preset quantities include, for example, 288, 300, 340, or other values. Load curve segments include, for example, 5-minute load curve segments, 10-minute load curve segments, etc. The curve segmentation unit divides the load curve into a preset number of load curve segments, for example, a preset number of 288, which consists of 288 5-minute load curve segments.

[0180] The feature extraction unit performs feature extraction on the load curve segments, extracting preset feature information from each load curve segment. The preset feature information includes one or more features such as the mean, variance, peak-to-peak ratio, skewness, kurtosis, and autocorrelation coefficient of the curve.

[0181] Preset algorithms include, for example, the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm. The strategy generation unit inputs preset feature information into the preset algorithm to generate a first control strategy, a second control strategy, and a third control strategy. The first control strategy includes a steady-state template, the second control strategy includes an impulse template, and the third control strategy includes a ramp template. The steady-state template, impulse template, and ramp template are generated by the preset algorithm based on the preset feature information. The steady-state template is used to increase the switching threshold range, i.e., adjust the difference between the first threshold and the second threshold; the impulse template is used to pre-activate the capacitor buffer device, activating and charging it before the control device performs a zero-interruption switching operation; the ramp template is used to pre-start the first power supply, which is the server's backup power supply, before the control device performs a zero-interruption switching operation.

[0182] In addition, as server energy consumption is controlled, the aforementioned steady-state template, pulse template, and ramp template can be continuously updated using new data.

[0183] The above content is as follows Figure 8 As shown, the long-term optimization layer performs cluster analysis every 24 hours to generate scene templates. Pattern recognition is then performed, and the weight coefficients are updated every 5 minutes using a preset algorithm, which can be a Q-learning algorithm. After parameter adjustments in the mid-term optimization layer, the short-term optimization layer uses PID control every 100ms to adjust the output ratio of the main and backup power supplies, thus controlling the power output in real time.

[0184] This embodiment also provides a server power consumption control device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0185] This embodiment provides a server power consumption control device, such as... Figure 9 As shown, it includes:

[0186] Data acquisition module 901 is used to acquire the server's total load data and system indicator data;

[0187] The threshold determination module 902 is used to input system indicator data into the prediction model to obtain a first threshold and a second threshold, wherein the first threshold is less than the second threshold. The first threshold and the second threshold are determined by the prediction model based on the weight coefficients and the system indicator data. The weight coefficients are determined based on the weight coefficient adjustment strategy, which is determined by a preset algorithm based on the server's status information.

[0188] The first control module 903 is used to control the first target device in the server to enter an energy-saving state when the total load data is less than a first threshold.

[0189] The second control module 904 is used to control the second target device in the server to enter the working state when the total load data is greater than the second threshold.

[0190] Further functional descriptions of the above modules are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0191] In this embodiment, the server power consumption control device is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0192] This application also provides a computer device having the above-described features. Figure 9 The server power consumption control device shown is shown.

[0193] Please see Figure 10 , Figure 10 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of this application, such as... Figure 10 As shown, the computer device includes one or more processors 1001, a memory 1002, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 10 Take processor 1001 as an example.

[0194] Processor 1001 may be a central processing unit, a network processor, or a combination thereof. Processor 1001 may further include an integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field-programmable gate array (FPGA), a general-purpose array logic (GPA), or any combination thereof.

[0195] The memory 1002 stores instructions executable by at least one processor 1001 to cause at least one processor 1001 to perform the method shown in the above embodiments.

[0196] The memory 1002 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 1002 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk drive, flash memory device, or other non-transitory solid-state memory. In some alternative embodiments, the memory 1002 may optionally include memory remotely located relative to the processor 1001, and these remote memories can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0197] The memory 1002 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 1002 may also include a combination of the above types of memory.

[0198] The computer device also includes a communication interface 1003 for communicating with other devices or communication networks.

[0199] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include memory capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0200] A portion of this application can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0201] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and such modifications and variations all fall within the scope defined by this application.

Claims

1. A server energy consumption control method, characterized in that, The method includes: Obtain the server's total load data and system metrics data; The system indicator data is input into the prediction model to obtain a first threshold and a second threshold, wherein the first threshold is less than the second threshold. The first threshold and the second threshold are determined by the prediction model based on the weight coefficients and the system indicator data. The weight coefficients are determined based on a weight coefficient adjustment strategy, which is determined by a preset algorithm based on the server's status information. The step of inputting the system indicator data into the prediction model to obtain a first threshold and a second threshold includes: determining the load mean, load standard deviation, and load threshold based on the system indicator data; obtaining the weight coefficient adjustment strategy and determining the weight coefficients in the prediction model based on the weight coefficient adjustment strategy; determining a first data item based on the load threshold and a first preset parameter; determining a second data item and a third data item based on the load mean, the load standard deviation, and the weight coefficients; using the larger value between the first data item and the second data item as the first threshold; determining a fourth data item based on the load threshold and the second preset parameter; and using the smaller value between the third data item and the fourth data item as the second threshold. If the total load data is less than the first threshold, control the first target device in the server to enter an energy-saving state; If the total load data exceeds the second threshold, the second target device in the server is controlled to enter the working state.

2. The method according to claim 1, characterized in that, The step of obtaining the weight coefficient adjustment strategy and determining the weight coefficients in the prediction model based on the weight coefficient adjustment strategy includes: Obtain the status information of the server, and determine the weight coefficient adjustment strategy based on the preset algorithm and the status information of the server; Based on the weight coefficient adjustment strategy and the initial weight coefficient, candidate weight coefficients are determined; Determine the change between the candidate weight coefficient and the initial weight coefficient, and determine the ratio of the change to the initial weight coefficient; If the ratio value is greater than the ratio threshold, the weight coefficient is generated based on the initial weight coefficient and the ratio threshold. If the ratio value is less than or equal to the ratio threshold, the candidate weight coefficient is used as the weight coefficient.

3. The method according to claim 2, characterized in that, The step of determining the weight coefficient adjustment strategy based on the preset algorithm and the server's status information includes: Based on the server's status information, determine whether the number of times the second target device's status has changed is greater than a first preset threshold. If the number of switching times exceeds a first preset threshold, the weight coefficient adjustment strategy is generated according to the first initial strategy and the preset algorithm. Based on the server's status information, determine whether the server's energy saving is less than a second preset threshold; If the energy saving is less than the second preset threshold, the weight coefficient adjustment strategy is generated according to the second initial strategy and the preset algorithm.

4. The method according to claim 1, characterized in that, Before inputting the system indicator data into the prediction model, the method further includes: Determine the initial weight coefficients and create an initial model based on the initial weight coefficients; The fifth data item is generated based on the energy saving, the energy saving threshold, and the first preset coefficient; the sixth data item is generated based on the number of switching, the number of switching threshold, and the second preset coefficient. Generate a reward function based on the fifth and sixth data items; Obtain historical load time-series data, historical business cycle characteristic identifiers, and historical system metrics of the server; The initial model is trained based on the historical load time-series data, the historical business cycle feature identifiers, the historical system indicators, and the reward function to obtain the prediction model.

5. The method according to claim 1, characterized in that, The process of controlling the second target device in the server to enter a working state includes: Obtain the system time of the server; The target index is determined based on the system time and the preset time constant; The sixth data item is generated based on the preset parameters and the target index; The product of the sixth data item and the preset ratio coefficient is taken as the target ratio; The output ratio of the first power supply and the second power supply is controlled to the target ratio, wherein the first power supply and the second power supply are included in the second target device.

6. A server energy consumption control system, characterized in that, The system includes: a control device, a first target device, and a second target device; The control device is used to acquire the server's total load data and system indicator data; input the system indicator data into a prediction model to obtain a first threshold and a second threshold, wherein the first threshold is less than the second threshold, the first threshold and the second threshold are determined by the prediction model based on weight coefficients and the system indicator data, the weight coefficients are determined based on a weight coefficient adjustment strategy, and the weight coefficient adjustment strategy is determined by a preset algorithm based on the server's status information; The control device inputs the system indicator data into a prediction model to obtain a first threshold and a second threshold, including: determining the load mean, load standard deviation, and load threshold based on the system indicator data; obtaining the weight coefficient adjustment strategy and determining the weight coefficients in the prediction model based on the weight coefficient adjustment strategy; determining a first data item based on the load threshold and a first preset parameter; determining a second data item and a third data item based on the load mean, the load standard deviation, and the weight coefficients; using the larger value between the first data item and the second data item as the first threshold; determining a fourth data item based on the load threshold and the second preset parameter; and using the smaller value between the third data item and the fourth data item as the second threshold. The control device is connected to the first target device and is also used to control the first target device in the server to enter an energy-saving state when the total load data is less than the first threshold. The control device is connected to the second target device and is also used to control the second target device in the server to enter the working state when the total load data is greater than the second threshold.

7. The system according to claim 6, characterized in that, The control device includes: an acceleration unit and a storage unit; The storage unit is used to store the prediction model; The acceleration unit is used to calculate the first threshold and the second threshold based on the prediction model and the system indicator data.

8. The system according to claim 6, characterized in that, The first target device includes a first power supply, a first storage device, a redundant cooling device, and a computing device; The control device is used to send a standby command to the first power supply and control the first power supply to enter a cold redundancy state. The control device is also used to send an energy consumption control command to the first storage device to control the energy consumption of the first storage device to be less than or equal to a third preset threshold. The control device is also used to send a shutdown command to the redundant cooling device to shut down the redundant cooling device; The control device is also used to send state control commands to the computing device to control the computing device to enter a target energy-saving state.

9. The system according to claim 6, characterized in that, The first target device includes a second storage device, a cooling device, and a virtual machine management device; The control device is used to determine the duration for which the total load data is less than a fourth preset threshold, and when the duration exceeds a fifth preset threshold, to send a voltage regulation command to the second storage device to adjust the voltage of the second storage device to a preset voltage. The control device is also used to send a power regulation command to the cooling device to control the power of the cooling device to be less than a sixth preset threshold. The control device is also used to send virtual machine migration instructions to the virtual machine management device to migrate idle virtual machines to energy-saving nodes.

10. The system according to claim 6, characterized in that, The second target device includes a first power supply, a second power supply, and a capacitor buffer device; The control device is used to send a pre-charge command to the capacitor buffer device to activate the capacitor buffer device for pre-charging. The control device is also used to send a start command to the first power supply to start the first power supply after the output voltage of the capacitor buffer device reaches the target value. The control device is further configured to send a first output control command to the first power supply and a second output control command to the second power supply, thereby controlling the ratio between the output of the first power supply and the output of the second power supply to reach a target ratio, wherein the target ratio is determined according to a preset formula.

11. The system according to claim 10, characterized in that, The control device includes: a curve segmentation unit, a feature extraction unit, and a strategy generation unit; The curve segmentation unit is used to obtain the load curve of the server and segment the load curve into a preset number of load curve segments; The feature extraction unit is used to extract preset feature information from the load curve segment; The strategy generation unit is used to generate a first control strategy, a second control strategy, and a third control strategy according to a preset algorithm and the preset feature information. The first control strategy is used to adjust the difference between the first threshold and the second threshold, the second control strategy is used to pre-activate the capacitor buffer device, and the third control strategy is used to start the first power supply in advance.

12. A computer device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the server energy consumption control method according to any one of claims 1 to 5.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the server energy consumption control method according to any one of claims 1 to 5.

14. A computer program product, characterized in that, Includes computer instructions for causing a computer to execute the server energy consumption control method according to any one of claims 1 to 5.

Images