Hard disk temperature prediction method and device, electronic equipment and storage medium

Abstract

The invention discloses a hard disk temperature prediction method and device, electronic equipment and a storage medium, and relates to the technical field of computer hardware management, and the method comprises the steps: obtaining a PCIe real-time bandwidth utilization rate, a hard disk IOPS and an environment parameter, inputting the PCIe real-time bandwidth utilization rate, the hard disk IOPS and the environment parameter into a pre-trained deep learning network prediction model, the initial temperature prediction result of the hard disk is obtained and then corrected, a target temperature prediction result of the hard disk is obtained, a temperature heat dissipation strategy of the hard disk is generated based on the target temperature prediction result, and the temperature of the hard disk is dynamically regulated and controlled according to the temperature heat dissipation strategy. The problem that the heat dissipation risk of the hard disk is increased due to the fact that the obtained hard disk temperature information is lagged and the read-write performance is poor is solved, the hard disk temperature is predicted through the PCIe bandwidth actual utilization rate and the hard disk IOPS, and therefore the hard disk temperature change in the future time period can be predicted in advance, and the heat dissipation safety of the hard disk is guaranteed.

Description

Technical Field

[0001] The present application relates to the technical field of computer hardware management, and in particular to a hard disk temperature prediction method, device, electronic device, and storage medium. Background Art

[0002] Server hardware health management technology is a key research direction in the data center and cloud computing fields. The BMC (Baseboard Management Controller) is responsible for server hardware status monitoring, fault warning, and thermal management. As data centers continue to increase their requirements for server hard disk storage performance, challenges such as high-density storage environments, dynamic load fluctuations, and operation and maintenance cost pressures have arisen. Therefore, more flexible and agile monitoring and prediction of hard disk temperature has become a key research direction for improving server reliability and energy efficiency.

[0003] In related technologies, hard drive temperature monitoring methods mainly rely on a fixed-frequency polling mechanism to obtain hard drive SMART (Self-Monitoring, Analysis, and Reporting Technology) information from the RAID (Redundant Array of Independent Disks Controller) card, or directly request SMART information from the hard drive through the transparent transmission capability provided by the RAID card.

[0004] However, obtaining hard disk SMART information through a fixed-frequency polling mechanism has the disadvantages of low update frequency and delayed acquisition of hard disk temperature information, which increases the risk of hard disk heat dissipation. At the same time, this method relies on the RAID card and its firmware, and has poor reliability. Directly requesting SMART information from the hard disk through the RAID card solves the lag problem, but may affect the hard disk read and write performance under high load conditions, which urgently needs to be solved. Summary of the Invention

[0005] The present application provides a hard disk temperature prediction method, device, electronic device and storage medium to at least solve the problem of increased hard disk heat dissipation risk caused by delayed acquisition of hard disk temperature information and poor read and write performance.

[0006] This application provides a hard disk temperature prediction method, including:

[0007] Obtaining the real-time bandwidth utilization of the disk array card's PCIe (Peripheral Component Interconnect Express, a high-speed serial computer expansion bus protocol), the hard disk's IOPS (Input / Output Operations Per Second), and environmental parameters of the service device where the hard disk is located;

[0008] Inputting the PCIe real-time bandwidth utilization, the IOPS of the hard disk, and the environmental parameters into a first target layer of a pre-trained deep learning network prediction model in a preset time series as input data to obtain an initial temperature prediction result of the hard disk, and inputting the initial temperature prediction result into a second target layer of the pre-trained deep learning network prediction model to obtain a target temperature prediction result of the hard disk;

[0009] A temperature heat dissipation strategy for the hard disk is generated based on the target temperature prediction result, and the temperature of the hard disk is dynamically regulated according to the temperature heat dissipation strategy.

[0010] The present application also provides a hard disk temperature prediction device, comprising:

[0011] An acquisition module is used to obtain the PCIe real-time bandwidth utilization of the disk array card, the IOPS of the hard disk, and the environmental parameters of the service device where the hard disk is located;

[0012] a temperature prediction module, configured to input the PCIe real-time bandwidth utilization, the IOPS of the hard disk, and the environmental parameters in a preset time series as input data into a first target layer of a pre-trained deep learning network prediction model to obtain an initial temperature prediction result of the hard disk, and input the initial temperature prediction result into a second target layer of the pre-trained deep learning network prediction model to obtain a target temperature prediction result of the hard disk;

[0013] The control module is configured to generate a temperature heat dissipation strategy for the hard disk based on the target temperature prediction result, and dynamically control the temperature of the hard disk according to the temperature heat dissipation strategy.

[0014] The present application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for implementing the steps of any of the above-mentioned hard disk temperature prediction methods when executing the computer program.

[0015] The present application also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned hard disk temperature prediction methods are implemented.

[0016] The present application also provides a computer program product, including a computer program, which implements the steps of any of the above-mentioned hard disk temperature prediction methods when executed by a processor.

[0017] Through this application, by obtaining the real-time PCIe bandwidth utilization, the IOPS of the hard disk and the environmental parameters, the PCIe real-time bandwidth utilization, the IOPS of the hard disk and the environmental parameters are input into a pre-trained deep learning network prediction model, and after obtaining the initial temperature prediction result of the hard disk, correction is made to obtain the target temperature prediction result of the hard disk, and a temperature heat dissipation strategy of the hard disk is generated based on the target temperature prediction result. The temperature of the hard disk is dynamically adjusted according to the temperature heat dissipation strategy, which solves the problems of increased hard disk heat dissipation risk due to delayed acquisition of hard disk temperature information and poor read and write performance. The hard disk temperature is predicted by the actual PCIe bandwidth utilization and hard disk IOPS, so that the hard disk temperature changes in the future time period can be predicted in advance to ensure the safety of hard disk heat dissipation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0019] Figure 1 A schematic diagram of a temperature prediction method for a hard disk in the related art;

[0020] Figure 2 A flowchart of a hard disk temperature prediction method provided in an embodiment of the present application;

[0021] Figure 3 A schematic diagram of a PCIe host bridge controller accessing a configuration space according to one embodiment of the present application;

[0022] Figure 4 A schematic diagram of a network structure design provided according to an embodiment of the present application;

[0023] Figure 5 A schematic diagram of temperature prediction implemented by a PCIe retimer chip according to one embodiment of the present application;

[0024] Figure 6 A schematic diagram of temperature prediction based on a PCIe switch chip according to one embodiment of the present application;

[0025] Figure 7 A block diagram of a temperature prediction device for a hard disk according to an embodiment of the present application;

[0026] Figure 8 FIG. 1 is a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0028] It should be noted that, in the description of this application, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. The terms "first," "second," etc., in this application are used to distinguish similar objects, and are not used to describe a particular order or sequence.

[0029] In order to enable those skilled in the art to better understand the present application, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0030] An embodiment of the present application provides a hard disk temperature prediction method, and the method is described in detail in conjunction with the execution flow of the hard disk temperature prediction method.

[0031] Before introducing the embodiments of the present invention, we first introduce the specific implementation methods of the relevant technologies. With the continuous breakthroughs in the fields of artificial intelligence and high-performance computing technology, the storage performance of server hard disks has also been continuously improved to meet the high performance and reliability requirements of some scenarios such as AIGC (Artificial Intelligence Generated Content) and large models. Currently, server hard disk temperature management faces the following challenges: (1) High-density storage environment: The hard disks in a single server are generally managed by RAID cards, such as Figure 1As shown in Figure 1, multiple hard disks are densely deployed, and the heat dissipation space is limited. Local high temperature can easily lead to hard disk performance degradation or physical damage (such as head failure and media aging); (2) Dynamic load fluctuations: In scenarios such as artificial intelligence large model training and big data analysis, the hard disk I / O (Input / Output) pressure presents sudden and periodic characteristics, and traditional temperature monitoring solutions are difficult to predict transient temperature rise; (3) Operation and maintenance cost pressure: Business interruptions caused by hard disk failures will cause huge economic losses to users, and data recovery costs are high. Active preventive maintenance is needed to reduce the failure rate.

[0032] In the related art, the RAID card mainly requests SMART information from the hard disk it manages according to a fixed polling frequency. The SMART information of the hard disk contains some basic status information of the hard disk, such as the temperature of the hard disk. After the hard disk receives the request command from the RAID card, it transmits the SMART information stored in its own register to the RAID card, and then the RAID card stores the temperature information in a specific address of the RAID card register. However, the above temperature prediction method has the following defects: (1) The RAID card updates the hard disk temperature information at a low frequency, and the hard disk temperature information obtained by the BMC from the RAID card is relatively lagging. The heat dissipation cannot actively respond to the transient changes in the hard disk temperature, which increases the risk of failure of the heat dissipation strategy; (2) The solution is heavily dependent on the RAID card and its firmware. If there is a problem with the RAID card firmware, or the RAID card fails to obtain the hard disk temperature, the BMC loses the monitoring of the hard disk temperature, and the reliability is poor.

[0033] To address the technical deficiencies of the aforementioned temperature prediction method, a more mainstream technical solution is currently available: the BMC directly requests SMART information from the hard drive through the transparent transmission capability provided by the RAID card. The RAID card then passes the hard drive's response directly to the BMC without any processing. This method solves the problem of delayed temperature information, but it also introduces new risks. For example, in some high-load business scenarios, most of the processing power of the hard drive's main control chip is used to ensure business operations. Sending commands to the hard drive at this time may cause a brief drop in read and write performance when the hard drive responds to the command. This is unacceptable for some user business scenarios.

[0034] Therefore, based on the problems existing in the above-mentioned method, how BMC can monitor and predict the temperature of the hard disk more flexibly and agilely, and feed back the changes in the hard disk temperature into the cooling strategy, has become a key research direction for improving server reliability and energy efficiency. To overcome the above-mentioned defects, the embodiment of the present application predicts the hard disk temperature through the actual utilization of PCIe bandwidth and the hard disk IOPS. At the same time, thermodynamic equation constraints are introduced in the output layer of the neural network to force the prediction results to conform to the law of heat diffusion, so as to improve the temperature prediction accuracy.

[0035] Specifically, Figure 2 A flowchart of a hard disk temperature prediction method provided by an embodiment of the present invention is provided.

[0036] like Figure 2 As shown, the hard disk temperature prediction method includes the following steps:

[0037] In step S201, the PCIe real-time bandwidth utilization of the disk array card, the IOPS of the hard disk, and the environmental parameters of the service device where the hard disk is located are obtained.

[0038] According to one embodiment of the present application, obtaining the PCIe real-time bandwidth utilization of a disk array card includes: reading the number of transaction layer data packets of the disk array card based on a target chip, where the target chip includes a PCIe retimer chip and a PCIe switch chip; and calculating the PCIe real-time bandwidth utilization of the disk array card based on the number of transaction layer data packets and the average size of the transaction layer data packets.

[0039] Specifically, before performing hard disk temperature prediction, it is necessary to collect data on relevant parameters involved in hard disk temperature prediction. As a feasible method, the embodiment of the present application adopts a PCIe retimer chip (PCIe retimer chip, a hardware component for enhancing PCIe signal integrity) or a PCIe switch chip (PCIe switch chip, providing expansion or aggregation capabilities and allowing more devices to be connected to a PCIe port) to read the PCIe real-time bandwidth utilization of the RAID card to reflect the hard disk load status. At the same time, a physical link is designed between the BMC and the PCIe Retimer and PCIe Switch, and communication is performed through the I2C (Inter-Integrated Circuit, internal integrated circuit bus) or SMBus (System Management Bus, system management bus) protocol. The PCIe real-time bandwidth utilization is calculated by the number of transaction layer packets (i.e., the number of TLP (Transaction Layer Packet)) of the disk array card and the average size of the transaction layer packets. That is, the BMC sends a message to the PCIe Retimer and PCIe Switch. The switch requests the number of TLPs sent or received by the PCIe device (such as a RAID card) connected to the chip. The switch then calculates the real-time PCIe bandwidth utilization of the disk array card based on the number of TLPs and the average TLP size. The specific expression is as follows:

[0040] PCIe real-time bandwidth utilization = (number of TLPs × average TLP size) / (theoretical bandwidth × time window);

[0041] It should be noted that PCIe real-time bandwidth utilization largely reflects the current workload of the RAID card and the hard disks under the RAID card. Since workload and temperature show a highly linear positive correlation, and workload is leading and causal, it is often the workload that experiences significant changes before the temperature does. Therefore, it has great advantages in temperature prediction.

[0042] As another feasible method, the embodiment of the present application can also obtain the PCIe real-time bandwidth utilization from the PCIe configuration space register by the BMC. The PCIe configuration space register is a set of standardized registers in each PCIe device, which is used to store the key information and control parameters of the device. The control parameters mainly include: (1) device ID (Identifier), manufacturer ID: identifying the device model and manufacturer; (2) Class Code: defining the device type (such as storage controller, network adapter, etc.); (3) BAR (Base Address Register): allocating the memory or I / O space address of the device; (4) PCIe Capability structure: including extended functions such as link status, speed, and power management.

[0043] Furthermore, in the PCIe configuration space register, there is a PCIe Capability structure (PCIe function register group, a storage area in the register). This data structure stores information related to the PCIe bus, including the current link bandwidth and negotiated rate. Among them, the Link Status Register is part of the PCIeCapability structure and is located at a specific offset address in the configuration space, such as Figure 3 As shown in FIG, it is implemented in the BMC firmware, and the PCIe Host Bridge controller can access the configuration space by writing code.

[0044] As another feasible way, in addition to PCIe real-time bandwidth utilization, the IOPS of the hard disk and the environmental parameters of the service device where the hard disk is located are also important factors affecting the hard disk temperature. Therefore, it is necessary to obtain the IOPS of the hard disk and the environmental parameters of the service device where the hard disk is located. Mainstream RAID cards and RAID cards developed by technicians in related fields generally provide relevant query interfaces, which can query the IOPS information of the specified hard disk. This IOPS information is not read from the hard disk by the RAID card. As the management end of the hard disk, the RAID card has designed logic for monitoring and counting the amount of hard disk read and write data in hardware and firmware. Therefore, the IOPS of the hard disk can be obtained directly from the RAID card interface without sending SMART query commands to the hard disk. In addition, there is no risk of the hard disk read and write performance being affected by the hard disk responding to the command sent by the RAID card.

[0045] The IOPS of the hard disk can be obtained from the RAID card interface through the built-in counter of the RAID card, without directly accessing the hard disk. The environmental parameters of the service device where the hard disk is located can include the current hard disk temperature, the current fan speed, the chassis inlet air temperature (°C), etc.

[0046] Therefore, by obtaining the real-time PCIe bandwidth utilization of the disk array card, the data transmission load between the RAID card and the hard disk is reflected in real time, indirectly indicating the working intensity of the hard disk, and solving the data lag problem caused by the RAID card cache in traditional temperature monitoring. At the same time, by obtaining the IOPS of the hard disk, the I / O operation frequency of the hard disk can be directly quantified to reflect its read and write pressure (positively correlated with heat generation), thereby avoiding performance interference caused by frequent requests for SMART information. The acquisition of the environmental parameters of the service device where the hard disk is located can provide the real-time status of the cooling system, which serves as the boundary condition for temperature prediction and identifies changes in cooling capacity.

[0047] In step S202, the PCIe real-time bandwidth utilization, the IOPS of the hard disk and the environmental parameters are input as input data in a preset time series into the first target layer of the pre-trained deep learning network prediction model to obtain the initial temperature prediction result of the hard disk, and the initial temperature prediction result is input into the second target layer of the pre-trained deep learning network prediction model to obtain the target temperature prediction result of the hard disk.

[0048] According to one embodiment of the present application, before the PCIe real-time bandwidth utilization, the IOPS of the hard disk and the environmental parameters are input as input data to the first target layer of the pre-trained deep learning network prediction model in a preset time series, it also includes: determining the input data of the deep learning network prediction model and the preset time series of the input data; determining the deep learning network structure of the deep learning network prediction model, and performing hard disk temperature prediction training on the input data based on the deep learning network structure, so as to adjust the training strategy according to the training results until the training results meet the temperature prediction conditions, and after the training results meet the temperature prediction conditions, obtain the deep learning network prediction model.

[0049] According to one embodiment of the present application, after the initial temperature prediction result is input into the second target layer in the pre-trained deep learning network prediction model, it also includes: judging whether the initial temperature prediction result meets the output condition of the thermodynamic diffusion equation; if the initial temperature prediction result does not meet the output condition of the thermodynamic diffusion equation, the initial temperature prediction result is corrected based on the second target layer until the initial temperature prediction result meets the output condition of the thermodynamic diffusion equation; otherwise, the initial temperature prediction result is used as the target temperature prediction result and output.

[0050] The preset time series and temperature prediction conditions can be set by those skilled in the art according to actual test requirements and are not specifically limited here.

[0051] Specifically, in order to be able to predict the hard disk temperature change trend in advance and thus trigger intervention measures before the temperature reaches the dangerous threshold, the embodiment of the present application needs to first train the deep learning network prediction model, such as designing a deep learning network based on LSTM (Long Short-Term Memory) and a physical constraint layer, wherein the hard disk temperature prediction algorithm is implemented using a deep learning method. Deep learning has the advantages of good generalization performance and good robustness. Therefore, the complexity of temperature prediction is not high and the relevant environmental parameters involved are relatively few. Therefore, in terms of the selection of network type and network scale, the embodiment of the present application adopts a lightweight LSTM network (suitable for time series prediction) + physical conduction correction factor to ensure that the prediction results conform to the law of heat diffusion. This step needs to consider relevant input parameters, including but not limited to the current hard disk temperature, hard disk IOPS, PCIe real-time bandwidth utilization, fan current speed and chassis inlet temperature, etc., to train the model and predict the hard disk temperature change trend within a period of time in the future (for example, 30s).

[0052] Specifically, first, the PCIe real-time bandwidth utilization, hard disk IOPS, and environmental parameters obtained above are integrated into an input matrix, and a preset time series (for example, a 30-second time sliding window) is used as the input data of the deep learning network prediction model. The preset time series is the input basis of the LSTM model, which is used to capture the temporal correlation between load and temperature. Secondly, as Figure 4 As shown, the deep learning network structure of the deep learning network prediction model is determined, and the deep learning network structure mainly includes an input layer, a long short-term memory network layer (i.e., an LSTM layer), a discard layer (i.e., a Dropout layer), a fully connected layer (i.e., a Dense layer), a physical constraint layer (i.e., a PINN (Physics-Informed Neural Networks, physical information neural network) layer) and an output layer, wherein the LSTM layer is configured with 32 neurons and uses tanh as the activation function, which is mainly used to capture the time series characteristics in the input data and is particularly suitable for processing long-term dependent problems; the Dropout layer is used to prevent the model from overfitting, by randomly discarding a part of the neurons and temporarily setting the output of these neurons to zero during the training process. The Dropout layer of the embodiment of the present application sets a discard rate of 0.2, which means that each time the parameters are updated, 20% of the neurons will be randomly ignored; the Dense layer is configured with 16 neurons and uses ReLU (Rectified Linear Unit (rectified linear unit) is used as the activation function, and the ReLU function can increase nonlinear characteristics to help the model learn more complex patterns; the physical constraint layer is used to force the output to conform to the thermodynamic diffusion equation; the output layer is used to output the temperature prediction value for the next 30 seconds; finally, the hard disk temperature prediction is trained on the input data based on the deep learning network structure, and the obtained temperature prediction value is adjusted according to the training result until the training result meets the temperature prediction condition. After the training result meets the temperature prediction condition, a trained deep learning network prediction model is obtained. This deep learning network prediction model achieves high-precision prediction of the hard disk temperature by integrating data-driven and physical laws, and can accurately reflect the cumulative effect of hard disk load changes on temperature (such as continuous high load causing accelerated temperature rise).

[0053] Based on the collection of the above data, a training data set is produced and the deep learning network prediction model is trained. The trained deep learning network prediction model can predict the temperature of the hard disk within the next 30 seconds. Therefore, in some special scenarios, if the temperature fails to be obtained through the method in the relevant technology, or the hard disk temperature cannot be obtained using the method in the relevant technology due to various factors, it can be obtained through the temperature prediction method of this application.

[0054] Therefore, in the embodiment of the present application, the acquired PCIe real-time bandwidth utilization, hard disk IOPS and environmental parameters can be input into the first target layer (LSTM layer) in the pre-trained deep learning network prediction model with a time sliding window of 30 seconds to obtain the initial temperature prediction result of the hard disk. Since the LSTM layer is responsible for time series modeling, it is only used to capture the dynamic change law of the input data and then output the initial temperature prediction result. However, it is not certain whether the initial temperature prediction result meets the output condition of the thermodynamic diffusion equation. Therefore, after obtaining the initial temperature prediction result, it is also necessary to input the initial temperature prediction result into the pre-trained The second target layer (PINN layer) in the deep learning network prediction model can add thermodynamic constraints (such as forcing the temperature change rate to conform to the heat diffusion equation) in the physical constraint layer to determine whether the initial temperature prediction result meets the output conditions of the thermodynamic diffusion equation. If not, the initial temperature prediction result is smoothed and corrected, and after correction, the target temperature prediction result that conforms to the thermodynamic diffusion equation is output to improve the physical rationality of the prediction. For example, if the LSTM layer predicts that the hard disk temperature is "50°C higher in 1 second" due to noisy data, the physical constraint layer will correct it to a reasonable value that conforms to the thermodynamic diffusion equation (such as 5°C).

[0055] According to one embodiment of the present application, determining whether an initial temperature prediction result satisfies an output condition of a thermodynamic diffusion equation includes: obtaining a temperature change rate at each time point in the initial temperature prediction result, and calculating a spatial second-order derivative corresponding to each time point based on the temperature values of different spatial positions of the hard disk at each time point; determining, based on a preset thermal diffusion coefficient, whether the temperature change rate is within a temperature change interval of the product of the spatial second-order derivative and the thermal diffusion coefficient; if the temperature change rate is within the temperature change interval of the product of the spatial second-order derivative and the thermal diffusion coefficient, determining that the initial temperature prediction result satisfies the output condition of the thermodynamic diffusion equation.

[0056] Specifically, when determining whether the initial temperature prediction result meets the output conditions of the thermodynamic diffusion equation, it is usually necessary to combine specific physical models and mathematical formulas for determination. Assuming that an initial temperature prediction result T(t) of a time series has been obtained through the above-mentioned deep learning network prediction model, for example, the temperature value per second in the next 30 seconds is obtained, then it is necessary to calculate the temperature change rate and spatial second-order derivative, that is, the temperature change rate at each time point, and the spatial second-order derivative corresponding to each time point based on the temperature values at different spatial positions of the hard disk at each time point; then, verification is performed based on the temperature change rate and spatial second-order derivative. For example, based on a preset thermal diffusion coefficient, it is determined whether the temperature change rate is within the temperature change range of the product of the spatial second-order derivative and the thermal diffusion coefficient, that is, whether the temperature change rate is approximately equal to the product of the spatial second-order derivative and the thermal diffusion coefficient. If the temperature change rate is within the temperature change range of the product of the spatial second-order derivative and the thermal diffusion coefficient, then it is determined that the initial temperature prediction result meets the output conditions of the thermodynamic diffusion equation. Otherwise, the initial temperature prediction result needs to be corrected until it meets the output conditions of the thermodynamic diffusion equation.

[0057] Therefore, based on the judgment of whether the initial temperature prediction result meets the output conditions of the thermodynamic diffusion equation, the prediction result can be made more reliable, reducing inaccurate predictions caused by data noise or model overfitting, and being able to more accurately predict future temperature change trends, so that measures can be taken in advance to prevent the temperature from being too high. Therefore, this step not only improves the accuracy and reliability of the model, but also enhances the stability and energy efficiency management capabilities of the system.

[0058] In step S203, a temperature heat dissipation strategy for the hard disk is generated based on the target temperature prediction result, and the temperature of the hard disk is dynamically regulated according to the temperature heat dissipation strategy.

[0059] According to one embodiment of the present application, a temperature heat dissipation strategy for the hard disk is generated based on the target temperature prediction result, and the temperature of the hard disk is dynamically regulated according to the temperature heat dissipation strategy, including: judging whether the target temperature prediction result meets the preset temperature safety threshold condition; if the target temperature prediction result does not meet the preset temperature safety threshold condition, then performing heat dissipation regulation on the environmental parameters of the service device where the hard disk is located; otherwise, judging whether the target temperature prediction result is within the temperature warning interval of the preset temperature safety threshold condition; if the target temperature prediction result is within the temperature warning interval, then issuing a temperature warning reminder for the hard disk, so that when the target temperature prediction result exceeds the temperature warning interval, the environmental parameters of the service device where the hard disk is located are heat-dissipated and regulated.

[0060] Among them, the preset temperature safety threshold conditions and temperature warning intervals can be set by technical personnel in this field according to actual testing requirements and are not specifically limited here.

[0061] Specifically, after outputting the target temperature prediction result that conforms to the thermodynamic diffusion equation, the BMC will automatically determine whether the target temperature prediction result meets the preset temperature safety threshold condition. In other words, the target temperature prediction result needs to ensure that the operating temperature of the hard disk is within a safe range to avoid hardware damage or performance degradation due to overheating, thereby ensuring the working performance of the hard disk. Therefore, after obtaining the target temperature prediction result of the hard disk, the target temperature prediction result is further judged. If the target temperature prediction result will exceed the safety range within a period of time in the future (for example, 30 seconds), heat dissipation control is required.

[0062] Specifically, first, the BMC determines whether the target temperature prediction result meets the preset temperature safety threshold condition (the temperature range in which the hard drive can safely operate, such as 40°C to 70°C, which can be determined based on different hard drive types), that is, the model predicts whether the hard drive temperature will exceed the preset temperature safety threshold condition within the next 30 seconds; secondly, if the target temperature prediction result does not meet the preset temperature safety threshold condition, that is, the target temperature prediction result exceeds the preset temperature safety threshold condition, that is, the hard drive temperature will exceed the preset temperature safety threshold condition within the next 30 seconds, then it means that the hard drive operating temperature is too high at this time, and immediate action is required to prevent hard drive damage. Therefore, it is necessary to regulate the environmental parameters of the service equipment where the hard drive is located to quickly reduce the temperature and protect the hard drive safety. For example, the hard drive temperature can be cooled and cooled by dynamically increasing the fan speed, activating the cooling device, optimizing the load distribution, etc., so that when the hard drive temperature is too high, effective heat dissipation measures can be used to intervene in advance to effectively prevent hard drive failure caused by overheating.

[0063] Furthermore, if the target temperature prediction result meets the preset temperature safety threshold condition, the hard disk temperature is lower than the temperature safety threshold condition, that is, the hard disk temperature will not exceed the preset temperature safety threshold condition in the next 30 seconds, then the current state of the hard disk is safe. However, to avoid abnormal hard disk temperature, it is possible to further determine whether the target temperature prediction result is within the temperature warning range of the preset temperature safety threshold condition. That is, even if the target temperature prediction result meets the preset temperature safety threshold condition, to further ensure the safety of the hard disk, it is possible to determine whether the target temperature prediction result is within the temperature warning range. If the target temperature prediction result is within the temperature warning range (for example, 30°C to 40°C), that is, close to the critical point of the hard disk temperature safety range, a hard disk temperature warning reminder needs to be issued. When the target temperature prediction result exceeds the temperature warning range, the environmental parameters of the service device where the hard disk is located are heat-dissipated and controlled. The specific control method can also be used to dissipate heat and cool the hard disk temperature by dynamically increasing the fan speed, activating the cooling device, optimizing the load distribution, etc., so that by real-time monitoring and prediction of the hard disk temperature, and timely adjustment of the heat dissipation strategy when necessary, unnecessary energy waste is avoided, data security and system stability are guaranteed, and hard disk failures caused by overheating are effectively prevented, thereby extending the service life of the hard disk.

[0064] According to one embodiment of the present application, after determining whether the target temperature prediction result meets the preset temperature safety threshold condition, it also includes: if the target temperature prediction result meets the preset temperature safety threshold condition, then the target temperature prediction result is given a positive reward based on a preset reward function, so that when the target temperature prediction result does not meet the preset temperature safety threshold condition, the environmental parameters are regulated for heat dissipation based on the minimization of energy consumption mode.

[0065] The preset reward function can be set by those skilled in the art according to actual testing requirements and is not specifically limited here.

[0066] Specifically, in the hard disk temperature prediction and heat dissipation strategy, in addition to ensuring that the operating temperature of the hard disk and other key components is within a safe range, it is also necessary to reduce energy consumption as much as possible. Therefore, in order to achieve the above goals, the embodiment of the present application can design a multi-dimensional reward function, for example, it can include (1) temperature safety: if the temperature of the hard disk or other components inside the server meets the preset temperature safety threshold condition, that is, the temperature remains within the safe range, then the target temperature prediction result is given a positive reward based on the preset reward function; on the contrary, if the temperature is too high, the target temperature prediction result is given a negative reward based on the preset reward function; (2) energy consumption minimization: the energy consumption can be calculated based on factors such as fan speed and frequency of use of the cooling device. Lower energy consumption corresponds to higher reward values; (3) system performance: ensure that the overall performance of the system is not affected, that is, avoid performance degradation caused by excessive cooling measures. The system performance can be evaluated by monitoring indicators such as IOPS and latency, and incorporated into the design of the reward function.

[0067] It should be noted that when the hard drive temperature is too high, the hard drive and other key components may face the risk of performance degradation, data loss, or even physical damage. Therefore, a reward mechanism is needed. The reinforcement learning algorithm will tend to select operations that can control the temperature within a safe range, thereby ensuring the stability and long-term reliability of the system.

[0068] Furthermore, by introducing temperature safety as a reward factor, the algorithm can learn to reduce the fan speed or the frequency of use of other cooling equipment as much as possible while meeting safety requirements, thereby reducing overall energy consumption. By monitoring the temperature in real time and giving corresponding rewards, the reinforcement learning model can automatically adjust the cooling strategy under various conditions, thereby ensuring that the optimal temperature state is maintained regardless of high load or low load.

[0069] Therefore, by incorporating temperature safety into the reward function, not only can the immediate safety of the system be directly improved, but it can also indirectly bring a series of long-term benefits, including but not limited to improving energy efficiency, extending hardware life, and enhancing the system's adaptability to complex working conditions. This is crucial for building an efficient and reliable data center infrastructure.

[0070] To facilitate those skilled in the art to better understand the present application, the following description will be based on specific embodiments:

[0071] Specifically, if Figure 5As shown, in order to solve the signal attenuation problem caused by long-distance transmission and improve the stability and performance of the entire system, this application generally embeds a PCIe Retimer chip on the PCIe Riser adapter card. This application adds a hardware link based on the I2C protocol or SMBus between the BMC and the PCIe retimer, so that the BMC can interact and communicate with the PCIe retimer. When the server is running, first, the BMC sends a hard disk temperature query command to the RAID card every 30 seconds. After receiving the BMC's request, the RAID card parses the command and returns the hard disk temperature cached in the RAID card register to the BMC.

[0072] Secondly, the BMC also sends a hard disk IOPS query command to the RAID card every 30 seconds to query the hard disk IOPS. The BMC also communicates with the PCIe retimer through I2C or SMBus. The PCIe retimer stores the number of TLPs sent or received by the PCIe device in a specific register, and then calculates the actual PCIe bandwidth utilization based on the average TLP size. For machines without PCIe retimers, the BMC attempts to read the PCIe connection rate of a specific storage area in the PCIe configuration space of the RAID card.

[0073] Again, using a 30-second sliding window, the BMC integrates the current temperature, IOPS, PCIe bandwidth utilization, fan speed, and air inlet temperature into an array and inputs it into a pre-trained deep learning network prediction model. The deep learning network prediction model then trains the hard drive temperature prediction based on the input data. After correcting the training results, it outputs a target temperature prediction result that conforms to the thermodynamic diffusion equation, thereby predicting the hard drive temperature change trend over the next 30 seconds.

[0074] Alternatively, as Figure 6 As shown in the figure, for AI (Artificial Intelligence) servers and some heterogeneous servers, the server topology is relatively complex. In order to expand more PCIe devices, AI servers generally have PCIe Switch chips. Due to the large number of PCIe slots and devices, during the hardware design process, the PCIe retimer chip may not have a hardware link for direct communication with the BMC. Therefore, a PCIe Switch chip can also be selected to communicate with the BMC. The content obtained by the BMC from the PCIe retimer can all be obtained through the PCIe switch. The specific data acquisition content has been detailed above and is not repeated here to avoid redundancy.

[0075] It should be noted that the embodiment of the present application is not only aimed at the temperature of the hard disk under the RAID card, but also the temperature of the hard disk that is not connected to the RAID card but directly connected to the backplane can also be predicted using the above-mentioned method.

[0076] In summary, the embodiments of the present application combine multiple parameters (such as the PCIe real-time bandwidth utilization of the disk array card, the IOPS of the hard disk, and the environmental parameters of the service device where the hard disk is located) and use a deep learning model to predict temperature, and then dynamically adjust the heat dissipation strategy based on the prediction results, thereby achieving the following beneficial effects:

[0077] (1) Improve hard drive reliability: By accurately predicting the hard drive's operating temperature and taking measures to control the temperature when necessary, hardware failures caused by overheating can be effectively reduced, the risk of data loss can be lowered, and the overall stability of the system can be enhanced;

[0078] (2) Optimize energy consumption management: Based on accurate temperature prediction results, intelligently adjust the cooling strategy (such as fan speed) to avoid unnecessary energy consumption, reduce costs, and reduce the burden on the data center cooling system, helping to achieve green computing goals;

[0079] (3) Improve system performance: Ensure that the hard disk always operates within its optimal temperature range to prevent performance degradation caused by overheating, so as to ensure that data reading and writing speeds are not affected;

[0080] (4) Enhanced preventive maintenance capabilities: Early warning of potential overheating problems allows administrators to take preventive measures or arrange maintenance, reducing the frequency of sudden failures.

[0081] According to the hard disk temperature prediction method proposed in an embodiment of the present invention, the PCIe real-time bandwidth utilization, the hard disk IOPS and environmental parameters are obtained, and the PCIe real-time bandwidth utilization, the hard disk IOPS and environmental parameters are input into a pre-trained deep learning network prediction model. After obtaining the initial temperature prediction result of the hard disk, the model is corrected to obtain the target temperature prediction result of the hard disk. A temperature heat dissipation strategy for the hard disk is generated based on the target temperature prediction result, and the temperature of the hard disk is dynamically controlled according to the temperature heat dissipation strategy. This solves the problems of increased hard disk heat dissipation risks caused by delayed acquisition of hard disk temperature information and poor read and write performance. The hard disk temperature is predicted by the actual PCIe bandwidth utilization and hard disk IOPS, so that the hard disk temperature changes in the future time period can be predicted in advance to ensure the safety of hard disk heat dissipation.

[0082] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method.

[0083] An embodiment of the present application also provides a temperature prediction device for a hard disk.

[0084] like Figure 7 As shown, the hard disk temperature prediction device 10 includes: an acquisition module 100 , a temperature prediction module 200 and a control module 300 .

[0085] The acquisition module 100 is used to obtain the PCIe real-time bandwidth utilization of the disk array card, the IOPS of the hard disk, and the environmental parameters of the service device where the hard disk is located;

[0086] The temperature prediction module 200 is configured to input the PCIe real-time bandwidth utilization, the hard disk IOPS, and environmental parameters in a preset time series as input data into the first target layer of a pre-trained deep learning network prediction model to obtain an initial hard disk temperature prediction result, and then input the initial temperature prediction result into the second target layer of the pre-trained deep learning network prediction model to obtain a target hard disk temperature prediction result;

[0087] The control module 300 is used to generate a temperature heat dissipation strategy for the hard disk based on the target temperature prediction result, and dynamically control the temperature of the hard disk according to the temperature heat dissipation strategy.

[0088] According to one embodiment of the present application, the acquisition module 100 includes:

[0089] A reading unit is used to read the number of transaction layer data packets of the disk array card based on a target chip, wherein the target chip includes a PCIe retimer chip and a PCIe switch chip;

[0090] The calculation unit is used to calculate the PCIe real-time bandwidth utilization of the disk array card based on the number of transaction layer data packets and the average size of the transaction layer data packets.

[0091] According to one embodiment of the present application, before inputting the PCIe real-time bandwidth utilization, the hard disk IOPS, and environmental parameters in a preset time series as input data into the first target layer of the pre-trained deep learning network prediction model, the temperature prediction module 200 further includes:

[0092] a determination unit, configured to determine input data of a deep learning network prediction model and a preset time series of the input data;

[0093] The model training unit is used to determine the deep learning network structure of the deep learning network prediction model, and perform hard disk temperature prediction training on the input data based on the deep learning network structure, so as to adjust the training strategy according to the training results until the training results meet the temperature prediction conditions, and obtain the deep learning network prediction model after the training results meet the temperature prediction conditions.

[0094] According to one embodiment of the present application, after inputting the initial temperature prediction result into the second target layer of the pre-trained deep learning network prediction model, the temperature prediction module 200 further includes:

[0095] A first judging unit is used to judge whether the initial temperature prediction result meets the output condition of the thermodynamic diffusion equation;

[0096] A correction unit is used to correct the initial temperature prediction result based on the second target layer until the initial temperature prediction result meets the output condition of the thermodynamic diffusion equation if the initial temperature prediction result does not meet the output condition of the thermodynamic diffusion equation; otherwise, the initial temperature prediction result is used as the target temperature prediction result and outputted.

[0097] According to one embodiment of the present application, the judging unit includes:

[0098] The acquisition subunit is used to obtain the temperature change rate at each time point in the initial temperature prediction result, and calculate the spatial second-order derivative corresponding to each time point based on the temperature values of different spatial positions of the hard disk at each time point;

[0099] a judgment subunit, for judging whether the temperature change rate is within a temperature change range of the product of the spatial second-order derivative and the thermal diffusion coefficient according to a preset thermal diffusion coefficient;

[0100] The determination subunit is used to determine whether the initial temperature prediction result meets the output condition of the thermodynamic diffusion equation if the temperature change rate is in the temperature change range of the product of the spatial second-order derivative and the thermal diffusion coefficient.

[0101] According to one embodiment of the present application, the control module 300 includes:

[0102] The second judgment unit is used to judge whether the target temperature prediction result meets the preset temperature safety threshold condition;

[0103] A control unit is configured to control the heat dissipation of the environmental parameters of the service device where the hard disk is located if the target temperature prediction result does not meet the preset temperature safety threshold condition; otherwise, it is configured to determine whether the target temperature prediction result is within the temperature warning range of the preset temperature safety threshold condition;

[0104] The early warning unit is used to issue a temperature warning reminder for the hard disk if the target temperature prediction result is within the temperature warning range, so as to adjust the heat dissipation of the environmental parameters of the service device where the hard disk is located when the target temperature prediction result exceeds the temperature warning range.

[0105] According to one embodiment of the present application, after determining whether the target temperature prediction result meets a preset temperature safety threshold condition, the second determination unit further includes:

[0106] The control subunit is used to give positive rewards to the target temperature prediction result based on a preset reward function if the target temperature prediction result meets the preset temperature safety threshold condition, so as to control the heat dissipation of the environmental parameters based on the minimization of energy consumption mode when the target temperature prediction result does not meet the preset temperature safety threshold condition.

[0107] In summary, the description of the features in the embodiment corresponding to the hard disk temperature prediction device can refer to the relevant description of the embodiment corresponding to the hard disk temperature prediction method, and will not be repeated here.

[0108] An embodiment of the present application further provides an electronic device, which may include:

[0109] A memory 801 , a processor 802 , and a computer program stored in the memory 801 and executable on the processor 802 .

[0110] When the processor 802 executes the program, the hard disk temperature prediction method provided in the above embodiment is implemented.

[0111] Furthermore, the electronic device further includes:

[0112] The communication interface 803 is used for communication between the memory 801 and the processor 802 .

[0113] The memory 801 is used to store computer programs that can be run on the processor 802.

[0114] The memory 801 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0115] If the memory 801, processor 802, and communication interface 803 are implemented independently, the communication interface 803, memory 801, and processor 802 can be interconnected via a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0116] Optionally, in a specific implementation, if the memory 801, the processor 802 and the communication interface 803 are integrated on a chip, the memory 801, the processor 802 and the communication interface 803 can communicate with each other through an internal interface.

[0117] The processor 802 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention.

[0118] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps of any of the above-mentioned hard disk temperature prediction method embodiments when running.

[0119] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.

[0120] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of any of the above hard disk temperature prediction method embodiments are implemented.

[0121] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0122] The above describes in detail the hard disk temperature prediction method provided by the present invention. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are intended only to facilitate understanding of the present invention's method and its core concepts. It should be noted that those skilled in the art may make various improvements and modifications to the present invention without departing from the principles of the present invention, and such improvements and modifications fall within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting hard disk temperature, characterized in that: include: Obtaining the PCIe real-time bandwidth utilization of the disk array card, the IOPS of the hard disk, and the environmental parameters of the service device where the hard disk is located; Inputting the PCIe real-time bandwidth utilization, the IOPS of the hard disk, and the environmental parameters into a first target layer of a pre-trained deep learning network prediction model in a preset time series as input data to obtain an initial temperature prediction result of the hard disk, and inputting the initial temperature prediction result into a second target layer of the pre-trained deep learning network prediction model to obtain a target temperature prediction result of the hard disk; A temperature heat dissipation strategy for the hard disk is generated based on the target temperature prediction result, and the temperature of the hard disk is dynamically regulated according to the temperature heat dissipation strategy.

2. The method according to claim 1, characterized in that The obtaining of the PCIe real-time bandwidth utilization of the disk array card includes: Based on a target chip, the number of transaction layer data packets of the disk array card is read, wherein the target chip includes a PCIe timer chip and a PCIe switch chip; The PCIe real-time bandwidth utilization of the disk array card is calculated based on the number of transaction layer data packets and the average size of the transaction layer data packets.

3. The method according to claim 1, characterized in that Before inputting the PCIe real-time bandwidth utilization, the IOPS of the hard disk, and the environmental parameters as input data in a preset time series into the first target layer of the pre-trained deep learning network prediction model, the method further includes: Determining input data of the deep learning network prediction model and a preset time series of the input data; Determine the deep learning network structure of the deep learning network prediction model, and perform temperature prediction training of the hard disk on the input data based on the deep learning network structure, so as to adjust the training strategy according to the training results until the training results meet the temperature prediction conditions, and obtain the deep learning network prediction model after the training results meet the temperature prediction conditions.

4. The method according to claim 1, wherein After inputting the initial temperature prediction result into the second target layer of the pre-trained deep learning network prediction model, the method further includes: Determining whether the initial temperature prediction result meets the output condition of the thermodynamic diffusion equation; If the initial temperature prediction result does not meet the output condition of the thermodynamic diffusion equation, the initial temperature prediction result is corrected based on the second target layer until the initial temperature prediction result meets the output condition of the thermodynamic diffusion equation; otherwise, the initial temperature prediction result is used as the target temperature prediction result and output.

5. The method according to claim 4, characterized in that The determining whether the initial temperature prediction result satisfies the output condition of the thermodynamic diffusion equation includes: Obtaining the temperature change rate at each time point in the initial temperature prediction result, and calculating the spatial second-order derivative corresponding to each time point based on the temperature values of different spatial positions of the hard disk at each time point; According to a preset thermal diffusion coefficient, determining whether the temperature change rate is within a temperature change range of the product of the spatial second-order derivative and the thermal diffusion coefficient; If the temperature change rate is within the temperature change range of the product of the spatial second-order derivative and the thermal diffusion coefficient, it is determined that the initial temperature prediction result meets the output condition of the thermodynamic diffusion equation.

6. The method according to claim 1, characterized in that Generating a temperature heat dissipation strategy for the hard disk based on the target temperature prediction result, and dynamically regulating the temperature of the hard disk according to the temperature heat dissipation strategy, includes: Determine whether the target temperature prediction result meets a preset temperature safety threshold condition; If the target temperature prediction result does not meet the preset temperature safety threshold condition, heat dissipation control is performed on the environmental parameters of the service device where the hard disk is located; otherwise, it is determined whether the target temperature prediction result is within the temperature warning range of the preset temperature safety threshold condition; If the target temperature prediction result is within the temperature warning range, a temperature warning reminder for the hard disk is issued, so that when the target temperature prediction result exceeds the temperature warning range, the environmental parameters of the service device where the hard disk is located are regulated for heat dissipation.

7. The method according to claim 6, characterized in that After determining whether the target temperature prediction result meets a preset temperature safety threshold condition, the method further includes: If the target temperature prediction result meets the preset temperature safety threshold condition, a positive reward is given to the target temperature prediction result based on a preset reward function, so that when the target temperature prediction result does not meet the preset temperature safety threshold condition, the environmental parameters are heat-dissipating regulated based on the minimization of energy consumption mode.

8. A temperature prediction device for a hard disk, characterized in that: include: An acquisition module is used to obtain the PCIe real-time bandwidth utilization of the disk array card, the IOPS of the hard disk, and the environmental parameters of the service device where the hard disk is located; a temperature prediction module, configured to input the PCIe real-time bandwidth utilization, the IOPS of the hard disk, and the environmental parameters in a preset time series as input data into a first target layer of a pre-trained deep learning network prediction model to obtain an initial temperature prediction result of the hard disk, and input the initial temperature prediction result into a second target layer of the pre-trained deep learning network prediction model to obtain a target temperature prediction result of the hard disk; The control module is configured to generate a temperature heat dissipation strategy for the hard disk based on the target temperature prediction result, and dynamically control the temperature of the hard disk according to the temperature heat dissipation strategy.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the hard disk temperature prediction method according to any one of claims 1 to 7.

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

Images