Server fan speed regulation method and device based on BMC

Abstract

The invention provides a server fan speed regulation method and device based on a BMC (Baseboard Management Controller). The method provided by the invention comprises the following steps: determining the type of a server based on the condition that the server is equipped with a noise sensor; matching a corresponding noise determination method based on the server type, and determining a real-time noise value based on the noise determination method; comparing the real-time noise value with a noise threshold value, and determining a corresponding noise interval; acquiring a CPU utilization rate and a GPU utilization rate of the server; calculating a feedback value based on the CPU usage rate and the GPU usage rate; and in the noise interval, determining a corresponding speed regulation method based on a matching relationship between the feedback value and a plurality of feedback intervals, and regulating the speed of a fan on the server based on the speed regulation method. According to the server fan speed regulation method and device based on the BMC provided by the invention, dual optimization of heat dissipation efficiency and low noise is realized by intelligently regulating and controlling the rotating speed of the server fan.

Description

Technical Field

[0001] The present application relates to the field of server technology, and in particular to a server fan speed regulation method and device based on BMC. Background Art

[0002] With the deep penetration of information technology in fields like cloud computing, big data, and artificial intelligence, the scale of data centers and server clusters is expanding exponentially. During high-frequency computing, high-power components like CPUs and GPUs continuously generate significant heat. Failure to dissipate heat in a timely manner can lead to hardware overheating and damage, or even system downtime. Therefore, dynamically adjusting heat dissipation through fan speed regulation has become a core requirement for ensuring server hardware life and operational stability.

[0003] The current mainstream fan speed control strategy relies heavily on feedback from ambient temperature sensors. When the temperature sensor detects that the chassis temperature exceeds a threshold, the BMC (Baseboard Management Controller) drives the fan to increase speed to enhance heat dissipation. However, this single temperature-based control approach has significant technical flaws. First, temperature changes are not completely linearly correlated with the actual hardware load. Under low-load conditions, local temperature fluctuations can cause the fan to misjudge and run at high speed. Field measurements show that traditional solutions still produce high peak fan noise levels even under moderate loads. Second, they ignore real-time noise data feedback, making them unsuitable for noise-sensitive environments such as medical imaging centers and financial trading rooms. Field measurements in financial data centers have shown that traditional speed control solutions still operate at high noise levels for a significant portion of the nighttime low-load period, far exceeding the specified ambient noise standards. Furthermore, existing solutions often employ a crude control logic based on preset temperature ranges, lacking intelligent analysis of the noise-temperature coupling relationship, resulting in a persistent conflict between cooling efficiency and noise control.

[0004] Therefore, there is an urgent need for a method to achieve dual optimization of heat dissipation efficiency and low noise by intelligently regulating the speed of server fans. Summary of the Invention

[0005] In view of this, the present application provides a server fan speed regulation method and device based on BMC, which is used to achieve dual optimization of heat dissipation efficiency and low noise by intelligently regulating the server fan speed.

[0006] Specifically, this application is implemented through the following technical solutions:

[0007] In a first aspect, the present application provides a server fan speed adjustment method based on a BMC, the method comprising:

[0008] Determining a server type based on whether a noise sensor is provided on the server; the server type includes a first server and a second server, the first server is not provided with a noise sensor, the second server is provided with a noise sensor, and the noise sensor is installed on a BMC of the server;

[0009] Based on the noise determination method corresponding to the server type, a real-time noise value is determined based on the noise determination method; the real-time noise value includes the noise values ​​of all fans on the server;

[0010] Comparing the real-time noise value with a noise threshold to determine a corresponding noise interval;

[0011] Obtain the CPU usage and GPU usage of the server;

[0012] Calculating a feedback value based on the CPU usage rate and the GPU usage rate;

[0013] In the noise interval, based on the matching relationship between the feedback value and multiple feedback intervals, a corresponding speed regulation method is determined, and the speed of the fan on the server is regulated based on the speed regulation method; wherein, in the first noise interval, the temperature PID setting value of the target sensor is adjusted based on the feedback interval where the feedback value is located, and the fan speed is adjusted based on the adjusted temperature PID setting value; in the second noise interval, the noise value is adjusted based on the feedback interval where the feedback value is located, and the fan speed is adjusted based on the adjusted noise value.

[0014] A second aspect of the present application provides a server fan speed control device based on a BMC, the device comprising a determination module, an acquisition module, a calculation module, and a speed control module;

[0015] The determining module is configured to determine a server type based on whether a noise sensor is provided on the server; the server type includes a first server and a second server, the first server is not provided with a noise sensor, the second server is provided with a noise sensor, and the noise sensor is installed on a BMC of the server;

[0016] The determination module is further configured to match a corresponding noise determination method based on the server type and determine a real-time noise value based on the noise determination method; the real-time noise value includes noise values ​​of all fans on the server;

[0017] The determination module is further configured to compare the real-time noise value with a noise threshold to determine a corresponding noise interval;

[0018] The acquisition module is used to obtain the CPU usage and GPU usage of the server;

[0019] The calculation module is used to calculate a feedback value based on the CPU usage rate and the GPU usage rate;

[0020] The speed regulation module is used to determine a corresponding speed regulation method based on a matching relationship between the feedback value and multiple feedback intervals in the noise interval, and regulate the speed of the fan on the server based on the speed regulation method; wherein, in a first noise interval, the temperature PID set value of the target sensor is adjusted based on the feedback interval where the feedback value is located, and the fan speed is adjusted based on the adjusted temperature PID set value; in a second noise interval, the noise value is adjusted based on the feedback interval where the feedback value is located, and the fan speed is adjusted based on the adjusted noise value.

[0021] This application provides a BMC-based server fan speed control method and device. First, in the context of diverse server hardware configurations, this application achieves full coverage through a "hardware perception-policy adaptation" approach. Specifically, servers are first classified based on whether they are equipped with noise sensors. For servers equipped with sensors, the real-time noise values ​​of all fans are directly collected using sensors on the BMC, ensuring that the speed control policy is based on actual noise data. For servers without sensors, the noise collection process is bypassed and the noise value is calculated based on fan speed, indirectly assessing server load and cooling requirements. This differentiated processing method avoids policy failures caused by missing hardware and enables servers with different configurations to achieve intelligent fan speed adjustment through a logical closed loop of "load-feedback value-speed control." For example, low-cost servers for small businesses may not be equipped with noise sensors, but can still dynamically adjust speeds based on CPU and GPU load. High-end servers in data centers can use sensor data to achieve a more precise balance between cooling and noise, significantly improving universality and practicality. Secondly, this application achieves dynamic optimization of the speed regulation strategy by constructing a dual control system: "initial noise screening - feedback fine-tuning." The first screening process focuses on real-time noise values, comparing them with preset thresholds and categorizing them into different noise ranges. This allows for rapid identification of current noise levels, providing a foundation for speed regulation. For example, in low-noise ranges, energy conservation is prioritized by reducing fan speed; in high-noise ranges, heat dissipation is prioritized. The second screening process focuses on feedback values, matching them with the feedback ranges to further refine the server's load scenarios (e.g., light load vs. heavy load). The speed regulation strategy is then dynamically adjusted based on the characteristics of different noise ranges. In the first noise range, the speed is indirectly controlled by adjusting the temperature PID setpoint to ensure heat dissipation safety in high-temperature scenarios. In the second noise range, speed is directly adjusted based on the noise value, taking into account quietness requirements under low loads. This hierarchical and progressive screening logic, like "first coarsely screening environmental conditions, then fine-tuning control parameters," avoids regulation errors caused by relying solely on a single metric (e.g., noise or load) while achieving a balance between heat dissipation efficiency and user experience. For example, when the server is in a high-load and high-noise range, the system will prioritize increasing the speed to ensure hardware safety; in a low-load and low-noise range, it will automatically reduce the speed to reduce energy consumption, ultimately achieving an efficient and intelligent speed regulation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Flowchart of the server fan speed adjustment method based on BMC provided in Example 1 of the present application;

[0023] Figure 2 This is a structural diagram of the BMC-based server fan speed regulation device provided in Example 2 of the present application. DETAILED DESCRIPTION

[0024] Exemplary embodiments are described in detail herein, with examples illustrated in the accompanying drawings. When the following description refers to the drawings, identical numerals in different drawings represent identical or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with this application.

[0025] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms "a," "the," and "the" used in this application are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0026] It should be understood that although the terms first, second, third, etc. may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0027] Specific embodiments are given below to introduce the technical solutions of the present application in detail.

[0028] Figure 1 This is a flow chart of the server fan speed adjustment method based on BMC provided in Example 1 of this application. Figure 1 The method provided in this embodiment may include:

[0029] S101: Determine the server type based on whether the server is equipped with a noise sensor.

[0030] The server types include a first server and a second server, the first server is not equipped with a noise sensor, the second server is equipped with a noise sensor, and the noise sensor is installed on a BMC of the server.

[0031] Specifically, server types are divided according to whether the server is equipped with a noise sensor. Servers are divided into two categories based on whether the noise sensor is integrated in the server hardware configuration as the core distinguishing criterion. Among them, the first server is a server that is not equipped with a noise sensor, that is, there is no sensor component for real-time monitoring of noise levels deployed inside the chassis, and it is impossible to directly obtain noise data during server operation. The first server is suitable for scenarios with low noise control requirements (such as data centers in non-office areas), or early server models that did not integrate noise monitoring functions. The second server is a server equipped with a noise sensor, and the noise sensor is installed on the BMC. The noise sensor is directly installed on the BMC (baseboard management controller), so that the noise sensor and the BMC form a hardware linkage to collect and transmit noise data in real time.

[0032] In specific implementations, the server hardware configuration is scanned through the BMC, or the server motherboard's hardware information list is directly read to confirm whether the server has a noise sensor component. If a noise sensor is detected, the system further queries the sensor's physical installation location parameters to confirm whether it is installed on the BMC. If a noise sensor is detected and installed on the BMC, the server is identified as the second server. If no noise sensor is detected, the server is identified as the first server.

[0033] Optionally, foam is provided on the noise sensor, and the foam compensates for the degree of noise attenuation at different distances by adjusting the thickness; the foam thickness is nonlinearly correlated with the distance from the noise sensor to the fan wall, and the foam thickness changes in a negative correlation with the distance.

[0034] Specifically, a foam pad is placed on the noise sensor, and its thickness is adjusted to compensate for noise attenuation at different distances. Because noise intensity is related to the detection location, the degree of noise attenuation during propagation varies when the distance from the noise sensor to the fan wall changes. Therefore, the noise data needs to be corrected by adjusting the foam pad thickness. The foam pad thickness and the distance from the noise sensor to the fan wall are nonlinearly related, and the two are negatively correlated. Specifically, as the distance increases, the foam pad thickness decreases accordingly; as the distance decreases, the foam pad thickness increases accordingly. This nonlinear relationship can be expressed as:

[0035] T(d')=15-14[(d-1) / 999]²;

[0036] Wherein, T(d') is the thickness of the foam; d is the distance from the noise sensor to the fan wall.

[0037] The method provided in this embodiment can accurately match the noise attenuation characteristics at different distances by adjusting the foam thickness, thereby correcting the data detected by the noise sensor to make it closer to the actual noise level.

[0038] S102: Match the server type to a corresponding noise determination method, and determine a real-time noise value based on the noise determination method.

[0039] The real-time noise value includes noise values ​​of all fans on the server.

[0040] Specifically, different server types require different noise determination methods. For server type 1, the real-time noise value is determined based on the fan speed. For server type 2, the real-time noise value is determined based on data collected by the noise sensor. The real-time noise value refers to the combined noise level within the chassis generated by all fans during server operation, expressed in decibels.

[0041] In a specific implementation, when the server type is the second server, the noise in the chassis is detected in real time by a noise sensor installed on the BMC, and the noise attenuation at different distances is corrected by foam, and the corrected noise data is transmitted to the BMC control module to obtain a real-time noise value.

[0042] Optionally, when the server type is a first server, determining the real-time noise value based on the noise determination method includes:

[0043] (1) For each fan on the server, calculate the current noise value of the fan based on the current rotation speed of the fan, the maximum rotation speed of the fan, and the maximum noise value of the fan at the maximum rotation speed.

[0044] Specifically, the fan's current speed refers to the fan's rotational speed in real-time operation. The fan's current speed is a real-time dynamic parameter that must be collected in real time by a sensor (such as a fan speed sensor) or the BMC. The fan's maximum speed refers to the maximum rotational speed that the fan can achieve within the design specifications (i.e., the rated maximum speed). The fan's maximum speed is a preset static parameter, set by the fan hardware manufacturer at the factory and is a fixed value. The fan's maximum noise value at maximum speed refers to the noise intensity generated when the fan is running at maximum speed. The fan's maximum noise value at maximum speed is a preset static parameter, determined by the fan manufacturer through testing. It is written into the hardware documentation as a fan specification parameter along with the maximum speed and stored in the server. It is a fixed value.

[0045] In a specific implementation, the current noise value of the fan is calculated based on the current speed of the fan, the maximum speed of the fan, and the maximum noise value of the fan at the maximum speed, including: calculating the quotient between the current speed of the fan and the maximum speed; performing a logarithmic operation on the quotient to obtain a logarithmic value; and determining the sum of the maximum noise value of the fan at the highest speed and the logarithmic value as the current noise value of the fan.

[0046] Specifically, the fan's current speed is obtained by reading the feedback data from the fan speed sensor in real time through the server's BMC. The fan's specification parameters are retrieved from the server's hardware configuration database, and the fan's maximum speed and the fan's maximum noise value at the maximum speed are determined by searching the fan's specification parameters. Furthermore, the fan's current speed is divided by the fan's maximum speed to obtain the quotient of the two. A logarithmic operation is performed on the calculated quotient to obtain the corresponding logarithmic value; the fan's maximum noise value at the maximum speed is added to the above logarithmic value, and the sum of the two is ultimately determined as the fan's current noise value.

[0047] The calculation formula for the current fan noise value can be expressed as:

[0048] Lpt1= Lptmax + 50*(log(RPMcurrent / RPMmax));

[0049] Wherein, Lpt1 is the current noise value of the fan; Lptmax is the maximum speed of the fan; RPMcurrent is the current speed of the fan; and RPMmax is the maximum noise value of the fan at the highest speed.

[0050] It should be noted that since fan noise is primarily generated by aerodynamic effects and mechanical friction, its intensity is positively correlated with the power of the speed. Calculating the quotient essentially converts the actual speed to a ratio relative to rated operating conditions, eliminating variations across fan specifications and establishing a unified noise assessment benchmark. In acoustics, sound intensity (decibels) is defined using a logarithmic scale (due to the nonlinearity of human auditory perception of sound). The power-order relationship between fan noise and speed can be converted to a linear relationship through logarithmic operations. This maps the nonlinear noise characteristics of speed to a linear decibel scale, making noise calculations consistent with physical laws. For example, if noise is proportional to the cube of speed, after logarithmic operations, the noise increment will be linearly related to the logarithm of speed, facilitating subsequent quantification. The maximum noise value at the fan's highest speed is used as the baseline, representing the fan's noise baseline under full load. Logarithmic mapping converts the speed ratio into a decibel-scaled noise change (positive values ​​indicate an increase above the baseline, negative values ​​indicate a decrease). The sum of the two is essentially the linear superposition of the baseline noise and the speed-varying noise, consistent with the physical property of "noise dynamically adjusting with speed." For example, when the speed is reduced to 50% of the rated value, the logarithmic value is negative, and the current noise will be lower than the maximum noise value, which is consistent with the phenomenon that low-speed fans have lower noise during actual operation.

[0051] The method provided in this embodiment converts the absolute speeds of different fans into a uniform relative ratio by dividing the current speed by the maximum speed to obtain a quotient. This eliminates the computational barriers caused by differences in fan specifications and allows fans of different models to use the same logic to evaluate noise. A logarithmic operation is performed on the quotient. Considering the nonlinear logarithmic nature of the human ear's perception of sound, this allows the calculated noise value to better reflect actual human auditory perception. It also converts the complex nonlinear relationship between speed and noise into a more manageable linear relationship, simplifying the complexity of the computational model. The current noise value is determined by using the measured maximum noise value at the highest speed as a benchmark and adding the offset obtained by the logarithmic operation. This not only relies on measured data to ensure the accuracy of the calculation, but also dynamically adjusts the noise result based on the current speed. Furthermore, the entire process does not require additional sensors; it only requires basic parameters and can be calculated through simple division, logarithmic, and addition operations. This offers the advantages of low cost and strong adaptability in engineering. This setup not only conforms to the physical laws of noise generation, but also takes into account practicality and efficiency in engineering applications. By combining mathematical transformations with measured data, it achieves universal, real-time, and low-cost calculation of fan noise while ensuring the accuracy of acoustic principles.

[0052] (2) Based on the current noise value of the fan and the number of fans on the server, a real-time noise value is calculated by logarithmic addition and inverse transformation.

[0053] In a specific implementation, the real-time noise value is obtained by logarithmic addition and inverse transformation based on the current noise value of the fan and the number of fans on the server, including: converting the current noise value of each fan into logarithmic form, and calculating the sum of the noise values ​​of all fans based on the logarithmic addition formula and the number of fans on the server; and inversely transforming the sum of the noise values ​​to obtain the real-time noise value.

[0054] Specifically, the current noise value of each fan is converted from decibel units to a corresponding linear energy value. Based on acoustic principles, the decibel value on the logarithmic scale is restored to the sound power or sound pressure energy on the linear scale. Furthermore, the energy values ​​corresponding to all fans are added together to obtain the total energy value of all fan noise in the server. All fans on the server are traversed and the energy value of each fan is accumulated in sequence. The total energy value is again converted back to decibel units through logarithmic operations, and the linearly accumulated total energy is remapped to a decibel value on the logarithmic scale to obtain the final real-time noise value.

[0055] The calculation formula of real-time noise value can be expressed as:

[0056] Lpt total = 10log(10^(Lpt1 / 10)+10^(Lpt2 / 10)+......+10^(LptS / 10));

[0057] Wherein, Lpt is the real-time noise value; LptS / 10 is the current noise value of each fan; and S is the number of fans on the server.

[0058] The method provided in this embodiment matches noise determination methods to server types, achieving a balance between cost, accuracy, and applicability through differentiated technical approaches. For the first type of servers, hardware noise sensors are used to directly collect sound pressure signals and convert them into electrical signals to output noise values. This allows for high-precision real-time monitoring by directly capturing physical quantities, avoiding computational model errors and eliminating the need for complex modeling. This makes it suitable for scenarios with stringent noise control requirements or complex fan types and changing environments. For the second type of servers, noise values ​​are derived based on parameters such as fan speed through logarithmic operations and energy superposition models. This eliminates sensor costs and is suitable for low- and mid-range servers or mass-produced servers. This eliminates the need for specific sensors and facilitates parameter updates when fan models change. It also avoids the impact of sensor physical failures and electromagnetic interference. Furthermore, this approach enables scenario-specific matching. For example, critical business servers use sensors to ensure standards, general-purpose servers use computing to control costs, and some high-end servers can improve reliability through cross-validation of data from both. This creates a dynamic balance between noise monitoring accuracy, cost control, and engineering adaptability to meet the core requirements of different scenarios and optimize resources throughout the entire lifecycle.

[0059] S103: Compare the real-time noise value with a noise threshold to determine a corresponding noise interval.

[0060] Specifically, the noise threshold is a decibel value pre-set by the customer or server designers to measure whether the server's operating noise is within an acceptable range. This setting is independent of the installation distance of the noise sensor and is based on the noise control requirements of the application scenario (for example, a medical setting requiring a quiet environment might have a noise threshold of 45 decibels, while a typical data center might have a threshold of 60 decibels). The determination of the noise threshold requires a comprehensive consideration of the noise sensitivity of the application scenario (such as the strict noise restrictions in financial trading rooms), industry regulations (such as data center noise standards), and user requirements for equipment comfort. It is a manually defined control baseline parameter.

[0061] Furthermore, the noise interval is a level area that divides the real-time noise value into ranges based on the noise threshold. For example, it can be divided into a "low noise interval" (the noise value is more than 10 decibels lower than the noise threshold), a "medium noise interval" (the noise value is close to the noise threshold), and a "high noise interval" (the noise value exceeds the noise threshold). Different noise intervals correspond to different fan speed control strategies: in the low noise interval, the BMC can reduce the fan speed to save energy; in the medium noise interval, the current speed is maintained or slightly adjusted; in the high noise interval, the fan speed is increased first to ensure heat dissipation, and the temperature data is combined to dynamically balance the noise control requirements. For example, in one embodiment, the noise threshold is A, and multiple noise intervals are real-time noise values ​​less than A-6, and real-time noise values ​​are between A-6 and A.

[0062] During implementation, the user or designer pre-sets a noise threshold. Based on the noise threshold, the noise range is divided into multiple intervals (e.g., low noise interval: real-time noise value < noise threshold - X; medium noise interval: noise threshold - X ≤ real-time noise value ≤ noise threshold + Y; high noise interval: real-time noise value > noise threshold + Y). X and Y are user-defined interval width parameters. Furthermore, the real-time noise value is compared with the boundaries of each interval to determine the noise interval it belongs to (e.g., if the real-time noise value is 50 decibels, the noise threshold is 60 decibels, X = 10, and Y = 5, it belongs to the medium noise interval).

[0063] S104: Obtain the CPU usage and GPU usage of the server.

[0064] Specifically, CPU utilization refers to the percentage of time the central processing unit (CPU) spends executing computing tasks per unit time, reflecting how busy the CPU is. GPU utilization refers to the percentage of time the graphics processor (GPU) spends processing graphics or general computing tasks per unit time, reflecting the GPU's load status.

[0065] It's important to note that because the CPU and GPU are the most computationally intensive and heat-generating components in a server, their utilization is directly linked to cooling requirements. For example, when the CPU / GPU is operating at high load, the heat generated surges, necessitating an increase in fan speed to maintain a safe temperature. Utilization data provides a real-time reflection of the server's current workload. Compared to the lag of temperature sensors (temperatures take time to rise), utilization can predict cooling requirements in advance and enable proactive adjustment of fan speed. Dynamically adjusting fan speed based on utilization avoids the frequent starts and stops or excessive cooling caused by traditional "temperature triggering" strategies, ensuring hardware safety while reducing energy consumption and noise (for example, maintaining low speeds under low loads).

[0066] In practice, the BMC can directly read the CPU's hardware monitoring registers to obtain CPU usage. For GPUs, the BMC can communicate with the GPU's management bus (such as I2C / SMBus) to read GPU usage.

[0067] S105: Calculate a feedback value based on the CPU usage rate and the GPU usage rate.

[0068] Specifically, the feedback value is a comprehensive indicator calculated using a specific algorithm based on CPU and GPU utilization, used to quantify the server's current computing load. The feedback value serves as the core input to the speed regulation strategy, directly determining the direction and magnitude of fan speed adjustments.

[0069] In a specific implementation, the feedback value is calculated based on the CPU usage rate and the GPU usage rate, including: calculating the square root of the product of the CPU usage rate and the GPU usage rate; and determining the sum of the square root of the product and the CPU usage rate and the GPU usage rate as the feedback value.

[0070] Specifically, the CPU usage and GPU usage are multiplied to obtain the product. The square root of this product is then taken to obtain a new value. The original CPU usage and GPU usage are added together, and the square root of the sum is added to the sum. Finally, the feedback value is normalized (e.g., converted to a percentage or a value between 0 and 1).

[0071] The calculation formula of the feedback value can be expressed as:

[0072] +X+Y;

[0073] Among them, the is the feedback value; X is the CPU usage; and Y is the GPU usage.

[0074] The method provided in this embodiment combines the square root of the product of CPU and GPU utilization rates with their sum. Essentially, this approach captures the effects of hardware collaborative load while preserving the overall load base. The square root can discern the combined heating effect of high CPU and GPU loads (e.g., when both are fully loaded, the square root approaches 1, amplifying cooling requirements). It also suppresses excessive amplification of the feedback value caused by high load on a single hardware component (e.g., when only a single component is highly loaded, the square root is less than that component's load, preventing a sudden increase in fan speed). The sum calculation ensures a linear positive correlation between the feedback value and the total hardware load, preventing the product calculation from losing information about the load of a single component (e.g., when a single component is highly loaded, the sum still reflects the basic cooling requirements). This combination of factors allows the feedback value to trigger a strong cooling response to high collaborative loads on both hardware and maintain smooth speed regulation under high loads on a single hardware. Ultimately, this achieves a dynamic balance between hardware cooling safety and fan noise control—prioritizing cooling during high collaborative loads, avoiding noise waste during high single loads, and reducing energy consumption during low loads.

[0075] S106 . In the noise range, determine a corresponding speed regulation method based on a matching relationship between the feedback value and multiple feedback ranges, and regulate the speed of the fan on the server based on the speed regulation method.

[0076] Among them, in the first noise interval, the temperature PID set value of the target sensor is adjusted based on the feedback interval where the feedback value is located, and the fan speed is adjusted based on the adjusted temperature PID set value; in the second noise interval, the noise value is adjusted based on the feedback interval where the feedback value is located, and the fan speed is adjusted based on the adjusted noise value.

[0077] Specifically, the feedback interval is a pre-defined number of feedback value intervals used to categorize the feedback values ​​calculated in real time. For example, in one embodiment, the feedback intervals are set as feedback value greater than or equal to 1.5, feedback value between 1 and 1.5, feedback value between 0.5 and 1, and feedback value less than 0.5.

[0078] Furthermore, in conjunction with the above description, different noise ranges require different speed control methods. When the noise range is the first range, the fan speed is indirectly adjusted by adjusting the temperature sensor's PID control target value. PID control calculates the fan speed adjustment based on the deviation between the current temperature and the target temperature. A higher target temperature (setpoint) allows more room for hardware temperature rise and may result in lower fan speeds. Conversely, a lower target temperature requires higher fan speeds for forced cooling. When the feedback value falls within a certain feedback range, the temperature PID setpoint is dynamically lowered or raised. For example, if the feedback value is high (e.g., due to high load on dual hardware), indicating a strong cooling demand, the temperature setpoint is lowered (e.g., from 70°C to 65°C). This increases the deviation and causes the PID control to accelerate the fan. If the feedback value is low, the temperature setpoint is raised (e.g., from 70°C to 75°C), allowing for a moderate increase in hardware temperature and reducing fan speed to reduce noise. The essence of the speed regulation method in the first noise range is to indirectly adjust the fan speed by changing the "temperature control target". The core is to dynamically adjust the heat dissipation priority based on the hardware load (feedback value) to prioritize the accuracy of temperature control.

[0079] When the noise range falls within the second noise range, the noise level is directly used as the control target. The permissible noise threshold is adjusted by matching the feedback range, thereby controlling the fan speed. Since fan speed and noise are positively correlated (higher speed, greater noise), the permissible noise level (noise threshold) is dynamically adjusted when the feedback value falls within a certain feedback range. For example, a high feedback value indicates high hardware load, allowing the noise threshold to be raised (for example, from 60dB to 65dB), allowing the fan to operate at a higher speed to enhance heat dissipation. A low feedback value lowers the noise threshold (for example, from 60dB to 55dB), forcing the fan to reduce speed to meet low-noise requirements in low-noise scenarios (such as office environments). The speed control method in the second noise range essentially converts the feedback value into a noise constraint, prioritizing user experience (noise) and balancing hardware load and noise. This method is suitable for noise-sensitive scenarios (such as desktop servers and quiet areas in data centers).

[0080] Optionally, the method for determining the feedback interval includes: traversing all components on the server, and determining components whose power exceeds a preset threshold as target components; in each calculation, selecting any two components from the target components, and obtaining the usage rates of the two components; calculating the feedback value based on the usage rates of the two selected components, and arranging the calculated feedback values ​​in descending order of numerical values; counting the occurrence probabilities of different feedback values, generating a probability distribution curve based on the occurrence probabilities, determining a target interval with an occurrence probability greater than a preset probability threshold and a mutation point with an occurrence probability difference between adjacent intervals greater than a preset difference threshold based on the probability distribution curve, determining the interval endpoints based on the target interval, the mutation point, and the feedback values ​​corresponding to the noise thresholds under different target component combinations, and dividing the feedback value into multiple feedback intervals.

[0081] Specifically, target components refer to hardware components in the server whose power consumption exceeds a preset threshold. During server operation, by monitoring the power consumption of each component in real time, components whose actual power consumption exceeds the preset threshold are dynamically marked as target components. Target components may generate significant heat due to high power operation and are the primary targets for heat dissipation control. For example, in one embodiment, target components may include a central processing unit (CPU), graphics processing unit (GPU), storage device (hard disk), memory module, network interface card, motherboard chipset, power conversion module, etc.

[0082] It should be noted that the target components are not fixed. For example, when the server switches from an idle state to a computationally intensive task, the CPU and GPU may change from non-target components to target components.

[0083] In specific implementations, the server management system systematically traverses all hardware components on the server. During this traversal, real-time power data for each component is continuously monitored. The monitored power data is compared with a pre-set power threshold. When a component's power exceeds the threshold, it is marked as a target component. Furthermore, target components are selected and their utilization rates are obtained. During each feedback value calculation, two different components are randomly selected from the set of target components for combination. For example, in one possible implementation, a CPU and GPU combination is selected because, in many compute-intensive tasks, these two components often simultaneously carry a large amount of computing work, generate significant heat, and are highly correlated. In another possible implementation, a CPU and hard drive combination is selected because, in scenarios with frequent data read and write operations, their load significantly impacts overall server performance and cooling requirements. Real-time utilization data for the two selected components is obtained using the performance monitoring interface provided by the server operating system. Based on the previously determined feedback value calculation method, the obtained utilization data for the two components is substituted into the formula for calculation. Repeat the above process of selecting component combinations, obtaining usage rates, and calculating feedback values ​​multiple times. Collect all feedback values ​​obtained from these calculations and arrange them in descending order. Statistically process the sorted feedback values ​​to count the number of occurrences of each feedback value. Divide the number of occurrences of each feedback value by the total number of calculations to obtain the probability of occurrence. Plot the resulting feedback values ​​and their corresponding probability data into a probability distribution curve, with feedback value on the horizontal axis and probability of occurrence on the vertical axis. By observing the probability distribution curve, identify consecutive intervals where the probability exceeds a preset probability threshold (e.g., 5%) and mark these intervals as target intervals. Also, identify breakpoints where the probability difference between adjacent intervals exceeds a preset difference threshold (e.g., 3%). For example, if the probability difference between the preceding and following intervals at a feedback value of 1.0 in the probability distribution curve reaches 4%, exceeding the preset 3%, then 1.0 is considered a possible interval split point. Combine the feedback value information corresponding to the noise threshold for different target component combinations. For example, after extensive testing and practical application verification, we found that when the feedback value corresponding to the noise threshold for the CPU + GPU combination is 1.5, the server achieves a good balance between heat dissipation and noise control. By comprehensively considering the target range, the mutation point, and the feedback value corresponding to the noise threshold, the endpoints of the feedback range are ultimately determined, such as 0.5, 1.0, and 1.5. Based on these endpoints, the feedback value range is divided into multiple different intervals. For example, with 0.5, 1.0, and 1.5 as the endpoints, the feedback value is divided into intervals such as [0, 0.5), [0.5, 1.0), and [1.0, 1.5].

[0084] The method provided in this embodiment can accurately locate high-power heat-generating components such as CPU and GPU in the server by screening target components through power thresholds, avoid interference from low-power components, and reduce the amount of invalid calculations while allowing the feedback value to focus more on the core hardware that has the greatest impact on heat dissipation and noise; selecting the utilization rate combination of any two target components to calculate the feedback value can effectively capture the linkage characteristics of component loads in actual server operation, such as the scenario where the CPU and GPU are simultaneously highly loaded in deep learning tasks, to avoid the disconnection between the feedback value and the actual heat dissipation demand from the perspective of a single component; generating a distribution curve by statistically analyzing the probability of occurrence of feedback values, and determining the interval endpoints based on the probability threshold and mutation point, can dynamically identify high-frequency load states (such as low load in conventional office scenarios) Load interval) and the critical state of load switching (such as the feedback value mutation point when the computing task is started), so that the feedback interval division is freed from the mechanical nature of fixed thresholds and better adapts to the dynamic changes of server load; combined with the endpoints of the noise threshold calibration interval under different component combinations, the feedback interval can be directly bound to the fan speed regulation strategy. For example, when the feedback value falls into a certain interval, the corresponding speed strategy can be immediately matched, which not only avoids insufficient heat dissipation due to excessive noise control at high load, but also prevents energy consumption and noise waste caused by fan idling at low load. Ultimately, by dynamically adapting the feedback interval of the load, the server can achieve energy efficiency optimization by strengthening heat dissipation to ensure hardware safety at high load and reducing speed to reduce noise at low load, thereby improving the stability of system operation and user experience.

[0085] Further, in combination with the above description, when the noise interval is the second noise interval, the corresponding speed regulation method is determined based on the matching relationship between the feedback value and multiple feedback intervals, and the fan on the server is regulated based on the speed regulation method, including: determining the target feedback interval corresponding to the feedback value based on the feedback value; determining the noise baseline value based on the target feedback interval; the noise baseline value is the benchmark for regulating the fan noise; with the noise baseline value as the target, by comparing the difference in real-time noise values ​​at consecutive moments, the PID control method is used to calculate the noise change at consecutive moments; based on the noise change at consecutive moments and the real-time noise value at the previous moment, the real-time noise value at the current moment is updated; based on the updated real-time noise value at the current moment, the maximum speed of the fan, and the maximum noise value of the fan at the highest speed, the adjusted fan speed is obtained.

[0086] Specifically, the noise baseline is the target reference value for server fan speed control, forming a unique one-to-one correspondence with the feedback interval. During server operation, the feedback value is calculated based on component load (such as CPU and GPU utilization) and divided into multiple feedback intervals (such as the first feedback interval and the second feedback interval). Each feedback interval corresponds to a typical load state (such as low load, medium-high load), and the noise baseline is the pre-set "target noise level" for each feedback interval. For example, when the feedback value falls into the "second noise interval," the corresponding noise baseline value (such as 55 decibels) is used as the reference target for fan speed control. For example, in one embodiment, the first feedback interval is for feedback values ​​greater than or equal to 1.5, and the corresponding noise baseline value is A; the second feedback interval is for feedback values ​​between 1 and 1.5, and the corresponding noise baseline value is A-3; the third feedback interval is for feedback values ​​between 0.5 and 1, and the corresponding noise baseline value is A-6.

[0087] In specific implementations, the calculated feedback value is compared with multiple pre-defined feedback intervals to determine the interval to which the feedback value belongs (e.g., the "third feedback interval"). A noise baseline value (e.g., A-6) uniquely corresponding to the target feedback interval (e.g., the "third feedback interval") is read from the system configuration. The real-time noise value at the current moment is compared with the previous moment to obtain the difference between the two. This difference is input into the PID control algorithm, and the noise change that needs to be adjusted is calculated through a comprehensive calculation of the three steps: proportional, integral, and differential. The real-time noise value at the previous moment is added to the calculated noise change to obtain the updated noise value at the current moment. Based on the conversion relationship between noise value and fan speed, the updated noise value at the current moment is converted to fan speed, and the fan speed to be adjusted is calculated. The calculated adjusted fan speed is converted into a PWM (pulse width modulation) signal or other control instruction, which is sent to the fan controller to perform speed adjustment. It should be noted that the conversion relationship between noise value and fan speed can be referred to the description in the above embodiment and will not be repeated here.

[0088] Updating the real-time noise value at the current moment can be expressed as:

[0089] dB(t)=dB(t-1)+ΔdB(t);

[0090] Wherein, the dB(t) is the updated real-time noise value at the current moment; the dB(t-1) is the real-time noise value at the previous moment; and the ΔdB(t) is the noise variation at consecutive moments.

[0091] Optionally, the noise baseline value is taken as the target, and the noise change at consecutive moments is calculated by comparing the difference of real-time noise values ​​at consecutive moments using a PID control method, including: calculating a first difference between the real-time noise value at the current moment and the real-time noise value at the previous moment, and determining the product of the first difference and the proportional coefficient as a proportional adjustment item; calculating a second difference between the real-time noise value at the current moment and the noise baseline value, and determining the product of the second difference and the integral coefficient as an integral adjustment item; calculating a third difference between the real-time noise value at the previous moment and the real-time noise value at the target moment, calculating a fourth difference between the first difference and the third difference, and determining the product of the differential coefficient and the fourth difference as a differential adjustment item; the target moment is the moment before the previous moment; and adding the proportional adjustment item, the integral adjustment item and the differential adjustment item to obtain the noise change at consecutive moments.

[0092] Specifically, PID (Proportional-Integral-Derivative) control is a feedback control algorithm that dynamically calculates a control variable to achieve system stability by comparing the deviation between the target value and the actual value. In this embodiment, PID control accurately calculates the fan speed adjustment based on the deviation between the noise baseline and the real-time noise value, allowing the noise value to quickly approach and stabilize near the baseline. The proportional control term is used to quickly respond to current noise trends. If the current noise level is rising rapidly (the first difference is positive), the proportional control term generates a large negative control variable, accelerating fan cooling; otherwise, the fan speed is reduced. The integral control term is used to eliminate long-term steady-state errors. If the noise value continuously deviates from the baseline, the integral control term continuously accumulates the deviation and generates a larger control variable until the noise value approaches the baseline. The differential control term is used to predict future noise changes and preemptively suppress fluctuations. If the noise increase rate is slowing (the first difference is less than the third difference), the differential control term generates a positive control variable to prevent excessive fan acceleration and noise overshoot.

[0093] It should be noted that the target time is the moment before the previous time. For example, the current time is t, the previous time is t-1, and the target time is t-2.

[0094] In a specific implementation, the difference between the real-time noise values ​​at the current moment and the previous moment is calculated to obtain a first difference. The first difference is multiplied by a preset proportional coefficient to obtain a proportional adjustment term. The difference between the real-time noise value at the current moment and the noise baseline is calculated to obtain a second difference. The second difference is multiplied by a preset integral coefficient to obtain an integral adjustment term. The difference between the real-time noise values ​​at the previous moment and the moment before that is calculated to obtain a third difference. The difference between the first difference and the third difference is calculated to obtain a fourth difference. The fourth difference is multiplied by a preset differential coefficient to obtain a differential adjustment term. The proportional adjustment term, the integral adjustment term, and the differential adjustment term are added together to obtain a final noise change.

[0095] The calculation formula of the noise variation can be expressed as: ΔdB(t) =Kp*[dB(t)-dB(t-1)] + Ki*[dB(t)-A]*dt + Kd*{[dB(t)-dB(t-1)]-[dB(t-1)-dB(t-2)]} / dt; dt = 1;

[0096] Among them, ΔdB(t) is the noise change; Kp is the proportional coefficient; dB(t) is the real-time noise value at the current moment; dB(t-1) is the real-time noise value at the previous moment; dB(t-2) is the real-time noise value at the target moment; Ki is the integral coefficient; A is the noise baseline value; Kd is the differential adjustment item; and dt is the time interval.

[0097] The method provided in this embodiment uses PID control logic to decompose the dynamic adjustment of noise at the physical level into multi-dimensional synergy in the noise change calculation process: first, the proportional adjustment term calculates the first difference of the noise values ​​at adjacent moments and multiplies it by the proportional coefficient. In essence, it quantifies the "instantaneous rate of change" of the noise, and can make rapid adjustments to the speed of sudden increase or decrease of the noise, ensuring immediate response to the current noise fluctuation; secondly, the integral adjustment term is based on the second difference between the current noise and the reference value and the integral coefficient. In essence, it continuously tracks the "accumulated noise deviation", which can eliminate the steady-state error caused by the long-term noise deviation from the reference value, so that the noise gradually returns to the target value over a long period of time; thirdly, the differential adjustment term calculates the difference between the previous moment and the target value. The third difference of the noise value at the moment (the previous moment) is compared with the first difference to obtain the fourth difference and multiplied by the differential coefficient. This corresponds to the capture of the "acceleration" of the noise change, which can perceive the acceleration or deceleration trend of the noise fluctuation in advance, and avoid noise oscillation caused by over-response during the adjustment process; finally, the noise change amount obtained by the superposition of the three not only realizes the "instant force control" of the current noise fluctuation, but also completes the "continuous cumulative correction" of historical deviations, and also adds "predictive adjustment" of future fluctuation trends, so that the fan speed adjustment forms a closed-loop control of "quick response-deviation elimination-trend prediction" at the physical level, ensuring that the noise stably approaches the baseline value in dynamic changes, and balancing the actual application scenarios of heat dissipation requirements and noise control.

[0098] The method provided in this embodiment first determines server type based on whether the server is equipped with a noise sensor. For servers equipped with noise sensors, the sensor installed on the baseboard management unit (BMC) directly collects real-time noise values, ensuring that the speed control strategy is based on actual noise data. For servers without sensors, the current noise value is calculated using the fan speed, maximum speed, and maximum noise value. Logarithmic addition and inverse transformation are then used to obtain the real-time noise value, indirectly assessing cooling requirements. This differentiated processing method avoids speed control strategy failures caused by hardware differences, enabling servers with different configurations to achieve intelligent fan speed control through a closed-loop "load-feedback-speed control" logic loop.

[0099] On the second aspect, this application has constructed a dual control system of "initial noise screening-feedback fine-tuning". The first screening compares the real-time noise value with the threshold to determine the low, medium and high noise ranges, and quickly identifies the noise level to provide a basic direction for speed regulation. For example, in the low noise range, energy saving is prioritized to reduce the speed, and in the high noise range, heat dissipation is prioritized to increase the speed; the second screening is based on the matching of feedback value and feedback range, and the load scenario is further refined. In the first noise range, the speed is indirectly controlled by adjusting the temperature PID set value to ensure heat dissipation safety in high-temperature scenes. In the second noise range, the noise value is directly used as the target for adjustment, taking into account the low-load and quiet requirements. This hierarchical and progressive screening method avoids the deviation of single indicator adjustment and achieves a balance between heat dissipation efficiency and user experience. For example, when the server is highly loaded and noisy, the speed is prioritized to ensure hardware safety, and when the load is low and the noise is low, the speed is automatically reduced to reduce energy consumption.

[0100] Third, in the speed regulation method for the second noise range, a noise baseline is obtained by determining the target feedback range corresponding to the feedback value. Using this baseline as the target, a PID control method is used to calculate the noise change. The proportional control term quickly responds to current noise trends, the integral control term eliminates long-term steady-state errors, and the differential control term predicts future noise changes and suppresses fluctuations in advance. The three terms are summed to obtain the noise change, which is then used to update the real-time noise value. The adjusted speed is then calculated by combining the fan's maximum speed and maximum noise value. This speed regulation method, through the synergistic effect of the three PID control terms, ensures that the noise change includes immediate response to current fluctuations, continuous correction of historical deviations, and predictive adjustment of future trends. This achieves a closed-loop control mechanism for fan speed regulation: "rapid response—deviation elimination—trend prediction." This ensures that noise remains stable near the baseline during dynamic changes, balancing heat dissipation requirements with noise control. For example, when noise suddenly increases, a counter-regulation is immediately generated to suppress the increase in speed. When noise deviates continuously, the regulation is gradually increased until it returns to normal. When noise slows, the regulation is reduced in advance to avoid overshoot, ultimately achieving precise noise control and dynamic balance.

[0101] Corresponding to the aforementioned embodiment of a server fan speed adjustment method based on BMC, the present application also provides an embodiment of a server fan speed adjustment device based on BMC.

[0102] Figure 2 This is a schematic diagram of the structure of the server fan speed regulating device based on BMC provided in Example 2 of this application. Figure 2 The device provided in this embodiment includes a determination module 210, an acquisition module 220, a calculation module 230 and a speed regulation module 240;

[0103] The determining module 210 is configured to determine a server type based on whether the server is equipped with a noise sensor; the server type includes a first server and a second server, the first server is not equipped with a noise sensor, the second server is equipped with a noise sensor, and the noise sensor is installed on a BMC of the server;

[0104] The determination module 210 is further configured to match a corresponding noise determination method based on the server type and determine a real-time noise value based on the noise determination method; the real-time noise value includes noise values ​​of all fans on the server;

[0105] The determination module 210 is further configured to compare the real-time noise value with a noise threshold to determine a corresponding noise interval;

[0106] The acquisition module 220 is used to obtain the CPU usage and GPU usage of the server;

[0107] The calculation module 230 is configured to calculate a feedback value based on the CPU usage rate and the GPU usage rate;

[0108] The speed regulation module 240 is used to determine the corresponding speed regulation method based on the matching relationship between the feedback value and multiple feedback intervals in the noise interval, and regulate the speed of the fan on the server based on the speed regulation method; wherein, in the first noise interval, the temperature PID setting value of the target sensor is adjusted based on the feedback interval where the feedback value is located, and the fan speed is adjusted based on the adjusted temperature PID setting value; in the second noise interval, the noise value is adjusted based on the feedback interval where the feedback value is located, and the fan speed is adjusted based on the adjusted noise value.

[0109] The device of this embodiment can be used to perform Figure 1 The steps, specific implementation principles and implementation processes of the method embodiment shown are similar and will not be repeated here.

[0110] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0111] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application scheme. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0112] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A server fan speed adjustment method based on BMC, characterized in that: The method comprises: Determining a server type based on whether a noise sensor is provided on the server; the server type includes a first server and a second server, the first server is not provided with a noise sensor, the second server is provided with a noise sensor, and the noise sensor is installed on a BMC of the server; Based on the noise determination method corresponding to the server type, a real-time noise value is determined based on the noise determination method; the real-time noise value includes the noise values ​​of all fans on the server; Comparing the real-time noise value with a noise threshold to determine a corresponding noise interval; Obtain the CPU usage and GPU usage of the server; Calculating a feedback value based on the CPU usage rate and the GPU usage rate; In the noise interval, based on the matching relationship between the feedback value and multiple feedback intervals, a corresponding speed regulation method is determined, and the speed of the fan on the server is regulated based on the speed regulation method; wherein, in the first noise interval, the temperature PID setting value of the target sensor is adjusted based on the feedback interval where the feedback value is located, and the fan speed is adjusted based on the adjusted temperature PID setting value; in the second noise interval, the noise value is adjusted based on the feedback interval where the feedback value is located, and the fan speed is adjusted based on the adjusted noise value.

2. The method according to claim 1, characterized in that In the noise range, determining a corresponding speed regulation method based on a matching relationship between the feedback value and a plurality of feedback ranges, and regulating the speed of the fan on the server based on the speed regulation method, including: Based on the feedback value, determining a target feedback interval corresponding to the feedback value; Determining a noise baseline value based on the target feedback interval; the noise baseline value is a benchmark for adjusting fan noise; Taking the noise baseline value as a target, by comparing the real-time noise value differences at consecutive moments, the noise variation at consecutive moments is calculated using the PID control method; Based on the noise variation at the consecutive moments and the real-time noise value at the previous moment, updating the real-time noise value at the current moment; An adjusted fan speed is obtained based on the updated real-time noise value at the current moment, the maximum speed of the fan, and the maximum noise value of the fan at the maximum speed.

3. The method according to claim 2, characterized in that The method of calculating the noise variation at consecutive moments by comparing the real-time noise value differences at consecutive moments with the noise baseline value as the target includes: Calculating a first difference between a real-time noise value at a current moment and a real-time noise value at a previous moment, and determining a product of the first difference and a proportional coefficient as a proportional adjustment item; Calculating a second difference between the current real-time noise value and the noise reference value, and determining a product of the second difference and an integral coefficient as an integral adjustment item; calculating a third difference between a real-time noise value at a previous moment and a real-time noise value at a target moment, calculating a fourth difference between the first difference and the third difference, and determining a product of a differential coefficient and the fourth difference as a differential adjustment item; the target moment being the moment before the previous moment; The proportional adjustment term, the integral adjustment term and the differential adjustment term are added together to obtain the noise variation at consecutive moments.

4. The method according to claim 1, wherein When the server type is a first server, determining the real-time noise value based on the noise determination method includes: For each fan on the server, calculate a current noise value of the fan based on a current rotation speed of the fan, a maximum rotation speed of the fan, and a maximum noise value of the fan at the maximum rotation speed; Based on the current noise value of the fan and the number of fans on the server, a real-time noise value is obtained by logarithmic addition and inverse transformation.

5. The method according to claim 4, characterized in that Calculating a current noise value of the fan based on a current rotation speed of the fan, a maximum rotation speed of the fan, and a maximum noise value of the fan at the maximum rotation speed includes: Calculating a quotient between the current rotation speed of the fan and the maximum rotation speed; performing a logarithmic operation on the quotient to obtain a logarithmic value; The sum of the maximum noise value of the fan at the highest rotation speed and the logarithmic value is determined as the current noise value of the fan.

6. The method according to claim 4, characterized in that The step of calculating the real-time noise value based on the current noise value of the fan and the number of fans on the server by logarithmic addition and inverse transformation includes: Converting the current noise value of each fan into a logarithmic form, and calculating the sum of the noise values ​​of all fans based on a logarithmic addition formula and the number of fans on the server; Perform an inverse transformation on the noise value sum to obtain a real-time noise value.

7. The method according to claim 1, characterized in that The method for determining the feedback interval includes: Traverse all components on the server and identify components whose power exceeds a preset threshold as target components; In each calculation, any two components are selected from the target components, and the usage rates of the two components are obtained; Calculate feedback values ​​based on the utilization rates of the two selected components, and arrange the calculated feedback values ​​in descending order; The occurrence probabilities of different feedback values ​​are counted, a probability distribution curve is generated based on the occurrence probabilities, a target interval with an occurrence probability greater than a preset probability threshold and a mutation point with a difference in occurrence probability between adjacent intervals greater than a preset difference threshold are determined based on the probability distribution curve, interval endpoints are determined based on the target interval and the mutation point, combined with feedback values ​​corresponding to noise thresholds under different target component combinations, and the feedback value is divided into multiple feedback intervals.

8. The method according to claim 1, characterized in that The noise sensor is provided with foam, and the foam compensates for the noise attenuation degree at different distances by adjusting the thickness; the foam thickness is nonlinearly correlated with the distance from the noise sensor to the fan wall, and the foam thickness changes in a negative correlation with the distance.

9. The method according to claim 1, characterized in that The calculating of the feedback value based on the CPU usage rate and the GPU usage rate includes: Calculating the square root of the product of the CPU usage rate and the GPU usage rate; A sum of the square root of the product, the CPU usage rate, and the GPU usage rate is determined as a feedback value.

10. A server fan speed control device based on BMC, characterized in that: The device includes a determination module, an acquisition module, a calculation module and a speed regulation module; The determining module is configured to determine a server type based on whether a noise sensor is provided on the server; the server type includes a first server and a second server, the first server is not provided with a noise sensor, the second server is provided with a noise sensor, and the noise sensor is installed on a BMC of the server; The determination module is further configured to match a corresponding noise determination method based on the server type and determine a real-time noise value based on the noise determination method; the real-time noise value includes noise values ​​of all fans on the server; The determination module is further configured to compare the real-time noise value with a noise threshold to determine a corresponding noise interval; The acquisition module is used to obtain the CPU usage and GPU usage of the server; The calculation module is used to calculate a feedback value based on the CPU usage rate and the GPU usage rate; The speed regulation module is used to determine a corresponding speed regulation method based on a matching relationship between the feedback value and multiple feedback intervals in the noise interval, and regulate the speed of the fan on the server based on the speed regulation method; wherein, in a first noise interval, the temperature PID set value of the target sensor is adjusted based on the feedback interval where the feedback value is located, and the fan speed is adjusted based on the adjusted temperature PID set value; in a second noise interval, the noise value is adjusted based on the feedback interval where the feedback value is located, and the fan speed is adjusted based on the adjusted noise value.

Images