Fault prediction method and device of server, server and storage medium

Abstract

The invention discloses a server fault prediction method and device, a server and a storage medium, and relates to the technical field of servers, and the server fault prediction method comprises the steps: obtaining monitoring data from a target sensor, and enabling the monitoring data to represent a monitoring result of the target sensor for target hardware in the server, the substrate management controller is in communication connection with the target sensor; feature extraction processing is conducted on the monitoring data to obtain dynamic change features, and the dynamic change features represent the trend that the working state of the target hardware changes along with time; and predicting a fault of the target hardware based on the dynamic change characteristics to obtain a fault prediction result of the target hardware. The problem that hardware fault detection of the server lags behind in the prior art is solved, and the safety of the server is improved.

Description

Technical Field

[0001] This application relates to the field of server technology, and in particular to a method, apparatus, server, and storage medium for predicting server failures. Background Technology

[0002] As the fundamental unit of modern data centers, server hardware failures can lead to business interruptions, data loss, and increased operational costs. Traditional hardware failure prediction methods primarily rely on passive monitoring, such as threshold-based alarms. Consequently, these methods typically only detect and identify faults after they occur, failing to anticipate potential failures. Therefore, related technologies suffer from a lag in server hardware failure detection. Summary of the Invention

[0003] This application provides a method, apparatus, server, and storage medium for predicting server failures, in order to at least solve the problem of delayed hardware failure detection in servers that exists in related technologies.

[0004] This application provides a server fault prediction method, which is applied to a baseboard management controller in a server, including:

[0005] The monitoring data is obtained from the target sensor, which represents the monitoring results of the target sensor on the target hardware in the server. The baseboard management controller is in communication connection with the target sensor.

[0006] Feature extraction processing is performed on the monitoring data to obtain dynamic change features, which represent the trend of the target hardware's working state changing over time.

[0007] Fault prediction of the target hardware is performed based on dynamic change characteristics to obtain the fault prediction results of the target hardware.

[0008] This application also provides a server fault prediction device, comprising:

[0009] The data acquisition module is used to acquire monitoring data from the target sensor. The monitoring data represents the monitoring results of the target sensor on the target hardware in the server. The baseboard management controller is communicatively connected to the target sensor.

[0010] The feature extraction module is used to extract features from the monitoring data to obtain dynamic features, which represent the trend of the target hardware's working state changing over time.

[0011] The fault prediction module is used to perform feature extraction processing on monitoring data to obtain dynamic change features, which represent the trend of the target hardware's working state changing over time.

[0012] This application also provides a server, including: a memory for storing a computer program; and a processor for executing the computer program to implement the fault prediction method of any of the above-described servers.

[0013] This application also provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of any of the above-described server fault prediction methods.

[0014] In this embodiment, when the server fault prediction method is applied in a baseboard management controller, the baseboard management controller acquires monitoring data of the server target hardware from the target sensor; then, it extracts features from the monitoring data to obtain dynamic change features reflecting the trend of the target hardware's working state over time; finally, it predicts the target hardware fault based on these dynamic change features, thus obtaining the fault prediction result. Therefore, this embodiment acquires monitoring data from the target sensor through the baseboard management controller and performs feature extraction processing on the monitoring data to obtain dynamic change features. These dynamic change features reflect the trend of fault occurrence and / or development. Therefore, this application uses the obtained dynamic change features to predict faults, which can obtain fault prediction results before the fault occurs. This solves the problem of delayed hardware fault detection and inability to predict potential accidents in related technologies, thereby providing early warning, reducing operation and maintenance costs, reducing sudden accidents caused by hardware faults, and improving server security. Attached Figure Description

[0015] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A schematic diagram illustrating the process of a server fault prediction method provided in an embodiment of this application;

[0017] Figure 2 A schematic diagram illustrating the process of another server fault prediction method provided in this application embodiment;

[0018] Figure 3 A schematic diagram of a system for implementing a server fault prediction method provided in an embodiment of this application;

[0019] Figure 4 A schematic diagram illustrating the working principle of the server fault prediction method provided in the embodiments of this application;

[0020] Figure 5 A schematic diagram illustrating the visualization effect of the server's health score provided in an embodiment of this application;

[0021] Figure 6 A schematic diagram of a server fault prediction device provided in an embodiment of this application;

[0022] Figure 7 This is a schematic diagram of a server provided in an embodiment of this application. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0024] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., in this application are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0025] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] As the fundamental unit of modern data centers, server hardware failures can lead to business interruptions, data loss, and increased maintenance costs. Traditional hardware failure prediction methods mainly rely on passive monitoring, such as threshold-based alarms, which cannot predict potential failures in advance. Although the Baseboard Management Controller (BMC), as an independent management subsystem, can collect data from various sensors, related technologies lack efficient intelligent prediction mechanisms.

[0027] Currently, BMC fault prediction in related technologies mainly relies on monitoring metrics and log analysis. Monitoring metrics can include current CPU (Central Processing Unit) utilization, memory usage, disk I / O (Input / Output), and network traffic. Log analysis specifically includes system log analysis, application log analysis, error log analysis, and anomaly pattern identification. However, these technologies heavily depend on fixed threshold alarms and cannot predict progressive faults. Furthermore, traditional prediction methods do not consider the correlation analysis of multi-dimensional sensor data; they can generally only detect and identify faults after they occur, failing to predict potential faults in advance. Therefore, these technologies suffer from a lag in server hardware fault detection.

[0028] While related technologies have proposed methods for predicting software failures before they occur, improving server software diagnostic efficiency and enabling software and server maintenance before failures happen, thus reducing the probability of software failures, these technologies still require manual data collection and input. This method suffers from low execution efficiency and a high risk of data errors. Therefore, these technologies also suffer from low data collection and input efficiency, inaccurate collected data, and a high error rate in input data.

[0029] It should be noted that the server fault prediction method provided in this application is applied to the BMC (Server Management Center). The BMC can serve as a monitoring unit for the server, capable of acquiring relevant server data in real time.

[0030] To address the aforementioned problems, embodiments of this application provide a server fault prediction method, such as... Figure 1 As shown, Figure 1 The flowchart of a server fault prediction method provided in this application embodiment is as follows:

[0031] Step S101: Obtain monitoring data from the target sensor. The monitoring data represents the monitoring results of the target sensor on the target hardware in the server. The baseboard management controller is communicatively connected to the target sensor.

[0032] In this application, the server includes target hardware such as a CPU and hard drive, and target sensors such as temperature sensors, light sensors, voltage sensors, current sensors, and fan speed sensors. These sensors acquire operational data from the target hardware, i.e., monitoring data. This monitoring data can be real-time; the BMC can acquire real-time data from the target sensors, thereby capturing key indicators of the server hardware such as temperature, load, energy consumption, and fault codes. It can also connect to various sensors on the server motherboard (such as temperature sensors, voltage sensors, and fan speed sensors), enabling immediate identification of anomalies (such as sudden high loads or rapid increases in hardware temperature), achieving early warning and response to faults, and reducing the risk of downtime due to data latency. Simultaneously, it provides the latest data support for hardware health trend analysis, making operational decisions more aligned with the current hardware operating conditions, ultimately improving hardware stability, resource utilization efficiency, and fault handling timeliness, solving problems such as delayed fault detection caused by traditional offline data analysis.

[0033] Specifically, the target hardware refers to the hardware on the server's host, including but not limited to one or more of the following: fans, CPU, memory, hard drives, etc. In this embodiment, the BMC (Browser Control Center) is the core of the server fault prediction method. It achieves comprehensive collection of server operating data through various hardware interfaces and a sensor network, thereby enabling the BMC to acquire monitoring data from the target sensors. Compared to traditional manual data collection and input, the BMC, as the server's monitoring unit, can acquire relevant server data in real time, thus achieving intelligent and automated data collection and input.

[0034] For example, the BMC, as an independent subsystem, establishes communication connections with target sensors (such as temperature sensors, voltage sensors, fan speed sensors, etc.) on the server motherboard through the IPMI (Intelligent Platform Management Interface) protocol or the Redfish (Redfish Specification) protocol. It periodically collects key parameters such as CPU temperature, memory ECC (Error-Correcting Code) error count, hard disk SMART (Self-Monitoring, Analysis and Reporting Technology) health status, and power supply voltage fluctuations. Then, it stores the collected real-time data in its own cache or hardware management chip for local monitoring (such as triggering hardware alarms), thereby solving the problems of inaccurate data collected manually and high error rates of input data.

[0035] BMC communicates via the System Management Bus (SMBus) and I... 2 C (Inter-Integrated Circuit) buses and other communication devices can communicate directly with sensors, supporting real-time monitoring of components such as CPU, memory, and hard disk to achieve hardware interface support.

[0036] The IPMI protocol is responsible for packaging and transmitting sensor data to the management console, while the Redfish protocol provides a standardized data interface in JSON (JavaScript Object Notation) format through RESTful API (Representational State Transfer Application Programming Interface) to facilitate cloud platform integration and achieve protocol compatibility. It should be noted that the cloud platform integration mentioned above is a cloud platform service provided by the server to achieve protocol compatibility.

[0037] In addition, BMC can still run independently when the server's main operating system is not started, ensuring the continuity of data acquisition and enabling out-of-band management capabilities.

[0038] Some BMC systems integrate in-band monitoring data (such as operating system logs) with out-of-band hardware data through extended firmware, improving data coverage and enabling multimodal data fusion. It should be noted that the aforementioned systems can be systems on the BMC, such as RTOS (Real-Time Operating System). The extended firmware can enable the RTOS to recognize and control specific chip peripherals, such as GPIO (General Purpose Input / Output), UAPI (User API), SPI (Serial Peripheral Interface), or external modules.

[0039] Step S102: Perform feature extraction processing on the monitoring data to obtain dynamic change features, which represent the trend of the target hardware's working state changing over time.

[0040] Optionally, before performing feature extraction on the monitoring data, the server's fault prediction method further includes: the BMC preprocesses the raw data collected by the target sensor, and the preprocessed data is the monitoring data. The preprocessing of the raw data includes data cleaning and interpolation. For example, the data cleaning process uses a sliding window averaging method to eliminate instantaneous noise (such as voltage spikes), and the interpolation process uses an interpolation algorithm to complete missing data caused by target sensor failures. Taking the use of an interpolation algorithm to process monitoring data as an example: for missing data caused by sensor failures, the location of the missing data and adjacent valid data points (e.g., three valid data points before and after the missing point) are first located. Then, a suitable interpolation algorithm (e.g., linear interpolation, polynomial interpolation, etc.) is selected based on the data's variation pattern. For example, linear interpolation constructs a straight line equation using the two nearest valid data points on either side of the missing point, and substitutes the time coordinate of the missing point to calculate the completed value. Polynomial interpolation uses more adjacent valid data to fit a curve to derive the completed value, ultimately achieving reasonable completion of the missing data and ensuring the continuity and availability of the entire data sequence. By preprocessing the raw data, this embodiment can improve the quality of monitoring data, increase the efficiency of the feature extraction process, and ensure the accuracy and reliability of dynamically changing features, thereby providing data support for subsequent hardware fault identification and improving the accuracy of fault identification.

[0041] The aforementioned dynamic change characteristics are key indicator data features after calculation, and may include, but are not limited to, one or more of the following: dynamic change characteristics, discrete event density, and signal enhancement characteristics. For example, the aforementioned dynamic change characteristics may include time-series indicators such as temperature change rate (ΔT / Δt, where ΔT represents the amount of temperature change and Δt represents the amount of time change), standard deviation of power consumption fluctuation, and slope of fan speed trend; the aforementioned discrete event density may include discrete event densities such as the hourly frequency of memory ECC errors and hard disk read / write error rate; the aforementioned signal enhancement characteristics may include: eliminating high-frequency noise interference through low-pass filtering, or extracting temperature periodic fluctuation characteristics using Fourier transform.

[0042] Step S103: Based on the dynamic change characteristics, predict the faults of the target hardware to obtain the fault prediction results of the target hardware.

[0043] Taking the CPU, memory, hard drive, and fan on a server host as an example, the dynamic change characteristics reflect the key indicator information of the target hardware. From this, we can obtain one or more key indicator information such as temperature and power consumption. By analyzing and processing the indicator information of the target hardware, we can obtain the changing trend and current status of different key indicator information of the target hardware, i.e., dynamic change characteristics. For example, we can compare or judge the dynamic change characteristics with preset thresholds to determine the result of fault prediction.

[0044] Optionally, based on the dynamic change characteristics obtained above, a prediction model is constructed using LSTM (Long Short-Term Memory) and SVM (Support Vector Machine) to achieve fault risk classification and type discrimination.

[0045] In the embodiments of this application, the BMC acquires monitoring data from a target sensor, which represents the monitoring results of the target sensor on the target hardware in the server. The baseboard management controller is communicatively connected to the target sensor. Feature extraction processing is performed on the monitoring data to obtain dynamic change features, which represent the trend of the target hardware's operating state changing over time. Based on these dynamic change features, fault prediction of the target hardware is obtained to achieve a fault prediction result. Therefore, in this embodiment, the BMC acquires monitoring data, processes it to obtain dynamic change features, and then predicts faults in the target hardware. This solves the problem of delayed hardware fault detection in related technologies. Since this method is based on BMC applications, it also solves the problems of low data acquisition and input efficiency, inaccurate acquired data, and high error rates in input data in related technologies. Therefore, this embodiment provides early warning, reduces operating and maintenance costs, significantly reduces sudden accidents caused by hardware failures, and improves server security.

[0046] This embodiment provides a method for predicting server failures. Figure 2 This is a flowchart illustrating a server fault prediction method according to an embodiment of this application, as shown below. Figure 2 As shown, the process includes the following steps:

[0047] Step S201: Acquire monitoring data from the target sensor. The monitoring data represents the monitoring results of the target sensor on the target hardware in the server. The baseboard management controller is communicatively connected to the target sensor. For details, please refer to step S101 above, which will not be repeated here.

[0048] Step S202 involves performing feature extraction processing on the monitoring data to obtain dynamic change features, which represent the trend of the target hardware's operating status changing over time. For details, please refer to step S102 above, which will not be repeated here.

[0049] Step S203: Input the dynamic change features into the preset machine learning model so as to output the fault prediction result through the preset machine learning model.

[0050] The preset machine learning model is used to determine the fluctuation of the target hardware's working state by analyzing dynamic change characteristics, and to determine the fault prediction result by classifying the fluctuation.

[0051] Optionally, the model part used in this embodiment to determine the fluctuation of the working state of the target hardware by analyzing dynamic change characteristics can use LSTM (Long Short-Term Memory Network) to achieve time series prediction. In this embodiment, the model part used to determine the fault prediction result by classifying the fluctuation can use tools such as random forest and SVM to achieve the model classification operation. It can be seen that the preset machine learning model in this embodiment may include a combination of two sub-models, such as a combination of LSTM and SVM.

[0052] To achieve time-series prediction, embodiments of this application employ LSTM to capture the long-term dependencies of parameters such as temperature and power consumption. Optionally, an ARIMA (Autoregressive Integrated Moving Average) model can also be used to analyze periodic fluctuations. In this model, the temperature captured by LSTM is input into the machine learning model to predict future temperatures based on the current temperature. To achieve model classification, embodiments of this application may further employ random forest for fault type discrimination under multi-dimensional feature fusion, such as hard drive failure and fan failure. SVM is suitable for anomaly detection in small sample scenarios.

[0053] During the training process of the pre-set machine learning model, the input data with dynamically changing features is divided into training, validation, and test sets according to a preset ratio, such as a 7:2:1 ratio, to avoid overfitting and thus enhance the reliability and practicality of the machine learning model.

[0054] After the above process is completed, BMC interprets the model's decision-making logic by adjusting hyperparameters, thereby adjusting the model and optimizing dynamically changing features. For example, BMC adjusts the SHAP values ​​(SHapley Additive exPlanations) to optimize the handling of dynamically changing features. The machine learning model involved in this embodiment supports incremental training to continuously learn new data (new dynamically changing features) while retaining as much of the old data as possible.

[0055] Optionally, the machine learning model can be adjusted based on dynamic alarm thresholds. For example, to address server heat dissipation issues, due to the day-night temperature difference, the preset threshold for fan speed can be increased during the day when temperatures are high to mitigate the adverse effects of high temperatures on heat dissipation; conversely, at night when temperatures are low, the preset threshold for fan speed can be decreased to save power. As another example, when the server CPU is running continuously, the preset threshold for its remaining utilization rate can be appropriately increased to prevent the utilization rate from spiked beyond the previously set threshold, thereby reducing server malfunctions.

[0056] In the embodiments of this application, a machine learning model is used to classify and identify fault risks, and to dynamically update the model to adapt to environmental changes such as hardware aging, thereby reducing the operational and maintenance pressure on staff.

[0057] As an optional embodiment, a preset machine learning model is used to classify fluctuations based on dynamic alarm thresholds, and the server fault prediction method further includes:

[0058] Step S301: Obtain monitoring data within the first preset time period.

[0059] Optionally, the first preset time period is a user-defined time period, such as the last 24 hours; the above monitoring data is the monitoring data obtained by BMC from the target sensor, and the monitoring data represents the monitoring results of the target sensor on the target hardware in the server, such as periodically collecting key parameters such as CPU temperature, memory ECC error count, hard disk SMART health status, and power supply voltage fluctuation.

[0060] Step S302: Adjust the dynamic alarm threshold based on the monitoring data within the first preset time period.

[0061] Specifically, based on the data within the first preset time period, the threshold is dynamically adjusted. For example, the preset threshold for data within a user-defined 24-hour period is dynamically adjusted to balance the impact of the external environment on the server's internal hardware performance.

[0062] For example, server equipment generates heat during operation. Therefore, to address server heat dissipation, heat sinks are installed on critical hardware such as processors. Inside these heat sinks are built-in fans. The faster the built-in fan rotates, the more heat the hardware generates, and the stronger the corresponding heat dissipation. Therefore, due to the temperature difference between day and night, the built-in fan needs to rotate faster during the day (when temperatures are lower) than at night. A preset threshold can be adjusted to 20% of the built-in fan's full speed to mitigate the adverse effects of high temperatures on processor cooling. Similarly, at night, the preset threshold can be adjusted to 15% of the built-in fan's full speed to save power. It can be seen that although the built-in fan speed decreases at night, the lower temperature often results in a stronger cooling effect at the same fan speed. Even at lower speeds, the lower temperature can compensate for the difference in heat dissipation caused by the reduced fan speed.

[0063] For example, when a server's CPU is running continuously, the number of running processes often increases, and the remaining CPU utilization rate decreases. However, when the remaining CPU utilization rate is too low or even zero, system crashes, errors, and other malfunctions are very likely to occur, potentially leading to data loss, system corruption, and hardware failures. Therefore, when the server CPU is operating under high load, a preset threshold for the remaining CPU utilization rate can be appropriately increased to prevent sudden spikes in CPU utilization that exceed the preset threshold, thereby minimizing server failures.

[0064] In the embodiments of this application, by dynamically adjusting the threshold, the impact of external and internal factors on the hardware performance of the server is balanced, which helps the server, especially the server hardware, to adapt to different operating conditions, avoids misjudgment caused by fixed thresholds, and further improves the accuracy of judgment.

[0065] As an optional embodiment, the server fault prediction method further includes:

[0066] Step S401: Judge the fault prediction results to determine false alarm events.

[0067] Specifically, the fault prediction result is compared with the specific fault result. The specific fault result is the specific hardware fault information obtained by BMC through analysis of the collected real-time data, such as one or more fault information such as hard drive overheating or fan stoppage. The fault prediction result and the specific fault result include, but are not limited to, one or more faults such as hard drive failure or fan failure. If the fault prediction result contains a result that does not exist in the specific fault result, then a false alarm event is identified.

[0068] Step S402: Adjust the dynamic alarm threshold based on false alarm events.

[0069] Specifically, BMC adjusts the dynamic alarm threshold based on false alarm events, for example, by widening or tightening the dynamic alarm threshold range during consecutive false alarms.

[0070] In the embodiments of this application, a feedback mechanism is used to correct the dynamic alarm threshold in conjunction with false alarm events, thereby reducing the false alarm rate, avoiding misjudgments caused by problems in the setting of the dynamic alarm threshold, and improving the accuracy of the judgment.

[0071] As an optional embodiment, step S402 above, which adjusts the dynamic alarm threshold based on false alarm events, includes:

[0072] Step S4021: Obtain the number of false alarm events that occur consecutively within the second preset time period.

[0073] Specifically, the second preset time period is a user-defined time period, such as 48 hours of continuous use of the solid-state drive, where 48 hours is a user-defined time period, which is the aforementioned second preset time period.

[0074] Step S4022: If the number of consecutive false alarm events is greater than or equal to a preset value, then increase or decrease the dynamic alarm threshold.

[0075] Specifically, the dynamic alarm threshold is tightened based on the number of false alarms occurring consecutively within the second preset time period. For example, the fluctuation range of fan speed is increased in high-temperature environments, such as increasing the fan speed (dynamic alarm threshold) from 1500 RPM to 2500 RPM; after the CPU has been working continuously at high intensity for 48 hours, the remaining CPU utilization rate (dynamic alarm threshold) is reduced from 20% to 10% to adapt to the working environment that matches the hardware operation.

[0076] In the embodiments of this application, by adjusting the dynamic alarm threshold, the impact of the external environment on the internal hardware performance of the server is balanced, which helps the server, especially the server hardware, to adapt to different environments, avoids misjudgment caused by fixed thresholds, further improves the accuracy of judgment, and enhances the reliability of the server.

[0077] As an optional embodiment, the monitoring data includes monitoring results of multiple target hardware, and the server fault prediction method further includes:

[0078] Step a1: Determine the multiple health levels corresponding to the fault prediction results of multiple monitoring results.

[0079] Specifically, BMC will detect the monitoring results of specific hardware and match the detected monitoring results with the corresponding hardware fault detection results, such as matching the health data of CPU, memory, and hard drive.

[0080] Step a2 involves performing a weighted summation on multiple health scores to obtain the server's first score.

[0081] Specifically, a weighted scoring algorithm is used to aggregate the health of components such as CPU, memory, and hard drive according to weights, such as 40%, 30%, and 30%, into a total score, which is the first scoring result.

[0082] For example, the first rating result = 40% * CPU health + 30% * Memory health + 30% * Hard disk health.

[0083] Step a3: Attenuate the first score result according to the preset attenuation factor to obtain the second score result.

[0084] Specifically, the attenuation factor is a coefficient that introduces a time attenuation coefficient to external environment (such as continuous high temperature) or frequent error events, emphasizing the influence of recent state. By introducing the attenuation factor into the first score result for attenuation processing, the second score result can be obtained.

[0085] In the embodiments of this application, the BMC health scoring system quantifies the health of the server, concretizes the abstract server hardware status parameters, and simply and clearly evaluates the current health status of the server. This helps operators to carry out targeted maintenance work with the help of the first and second scoring results, realizes predictive maintenance, and ensures the stability and security of the server.

[0086] As an optional embodiment, the server fault prediction method further includes:

[0087] Step b1: Generate a visual analysis result of the server's health status based on the second rating result, the first rating result, and at least one of multiple health levels.

[0088] Specifically, the results of this visualization analysis can be presented in various forms, such as a health trend chart combining health scores, or a component-level rating radar chart combining the first and second rating results.

[0089] Step b2: Send the visualization analysis results to the client so that the visualization analysis results can be displayed through the client.

[0090] Specifically, BMC sends the visualization analysis results to the client, which then outputs the analysis results in a visual manner, such as displaying health trend charts, component-level score radar charts, and the top 3 risk items through a Web (World Wide Web) interface.

[0091] In the embodiments of this application, BMC generates a visual analysis result of the server's health status through a first rating result, a second rating result, and at least one of multiple health levels, and outputs the result in a visual form, which concretizes and visualizes the abstract hardware data content, lowers the threshold for discovering faults, reduces operating costs, and further improves maintenance efficiency.

[0092] like Figure 3 As shown, Figure 3 This diagram illustrates a system schematic of the fault prediction method for the implementation server provided in an embodiment of this application. The entire system may consist of specific hardware and / or software modules, such as, but not limited to, a sensor data acquisition module, a feature extraction module, a machine learning prediction module, a dynamic threshold adjustment module, and a health scoring system connected in sequence; wherein, the dynamic threshold adjustment module can be used to adjust or modify the dynamic alarm threshold used by the machine learning prediction module.

[0093] like Figure 4 As shown, Figure 4 This illustration shows a complete flowchart of a server fault prediction method provided in an embodiment of this application. The specific process is as follows:

[0094] The BMC acquires various sensor data through the data acquisition module; then, the feature extraction module preprocesses the acquired data to provide structured data for subsequent analysis; subsequently, the aforementioned feature data is used to build a model to achieve fault assessment and classification; the dynamic threshold adjustment module adjusts the dynamic alarm threshold based on historical operation records, thereby reducing the false alarm rate; finally, a health assessment of the system is performed to alert users to risks.

[0095] The functions of each module are described in detail below:

[0096] 1) Sensor data acquisition module

[0097] This module, centered on the BMC (Browser Control Center), comprehensively collects server operating data through various hardware interfaces and sensor networks. As an independent subsystem, the BMC connects to sensors on the server motherboard (such as temperature sensors, voltage sensors, and fan speed sensors) via IPMI or Redfish protocols, periodically collecting key parameters such as CPU temperature, memory ECC error count, hard drive SMART health status, and power supply voltage fluctuations. This module includes the following features:

[0098] Hardware interface support: BMC supports interfaces via System Management Bus (SMBus), I / O, and I / O interfaces. 2 The C-bus and other communication devices communicate directly with sensors, supporting real-time monitoring of components such as CPU, memory, and hard drive.

[0099] Protocol compatibility: The IPMI protocol is responsible for packaging and transmitting sensor data to the management console, while the Redfish protocol provides a standardized data interface in JSON format via RESTful API, which facilitates integration with cloud platforms.

[0100] Out-of-band management capability: BMC can still run independently when the server's main operating system is not started, ensuring the continuity of data collection.

[0101] Multimodal data fusion: Some systems integrate in-band monitoring data (such as operating system logs) with out-of-band hardware data by extending firmware, thereby improving the data coverage dimensions.

[0102] 2) Feature extraction module

[0103] This module preprocesses and performs feature engineering on the raw data, providing structured features for subsequent analysis. Data cleaning: A sliding window averaging method is used to eliminate transient noise (such as voltage spikes), and interpolation algorithms are used to complete missing data caused by sensor malfunctions. This module can calculate key indicators, including the following features:

[0104] Dynamic change characteristics: Calculate time-series indicators such as temperature change rate (ΔT / Δt), power consumption fluctuation standard deviation, and fan speed trend slope.

[0105] Discrete event statistics: Statistics on the hourly frequency of memory ECC errors, hard disk read / write error rate, and other discrete event densities.

[0106] Signal enhancement: Eliminate high-frequency noise interference through low-pass filtering, or extract temperature periodic fluctuation characteristics using Fourier transform.

[0107] 3) Machine Learning Prediction Module

[0108] This module builds a prediction model based on feature data to achieve fault risk classification and type identification. This module can perform algorithm selection, training, and validation, and the algorithm selection includes the following features:

[0109] Time series prediction: Use LSTM (Long Short-Term Memory) network to capture the long-term dependence of parameters such as temperature and power consumption, or use ARIMA model to analyze periodic fluctuations.

[0110] Classification models: Random forests are used for fault type discrimination under multi-dimensional feature fusion (such as hard drive failure / fan failure), while SVMs are suitable for anomaly detection in small sample scenarios.

[0111] Training and validation include the following features:

[0112] Data partitioning: The training set, validation set, and test set are allocated in a 7:2:1 ratio to avoid overfitting.

[0113] Feature Importance Analysis: Explaining the model's decision-making logic through SHAP values ​​and optimizing feature engineering.

[0114] Online learning: Supports incremental training and dynamically updates the model to adapt to environmental changes such as hardware aging.

[0115] 4) Dynamic threshold adjustment module

[0116] This module optimizes dynamic alarm thresholds through historical data analysis to reduce false alarm rates. This module includes the following features:

[0117] Baseline modeling: Calculate the mean (μ) and standard deviation (σ) of each parameter and their distribution patterns based on historical normal data, and set the preset threshold to μ+3σ.

[0118] Adaptive strategy:

[0119] Sliding window statistics: Dynamic alarm thresholds are dynamically adjusted based on data from the most recent 24 hours to adapt to the impact of day-night temperature differences on fan speed.

[0120] Feedback mechanism: Adjust the dynamic alarm threshold based on false alarm events, for example, widen the dynamic alarm threshold range when there are consecutive false alarms.

[0121] Multi-parameter correlation: Automatically widens the fan speed fluctuation range under high temperature environment, and tightens the SMART error threshold after the hard drive has been used for more than 20,000 hours.

[0122] 5) Health scoring system

[0123] This system quantifies server health and generates executable operation and maintenance policies, for example... Figure 5 As shown, this module can perform scoring models, provide visualization output, and offer operation and maintenance suggestions. The scoring model includes the following features:

[0124] Weighted scoring method: The health of components such as CPU, memory, and hard drive is aggregated into a total score according to weights (e.g., 40%, 30%, 30%).

[0125] Decay factor: Introduces a time decay coefficient for sustained high temperature or frequent erroneous events, emphasizing the impact of recent conditions.

[0126] The visualization output includes the following features:

[0127] Visual output: Displays health trend charts, component-level score radar charts, and the top 3 risk items through a web interface.

[0128] The operation and maintenance recommendations provided include the following features:

[0129] Predictive maintenance: Recommending time windows for replacing hard drives or cleaning fans based on failure probability.

[0130] Energy efficiency optimization: Based on power consumption fluctuations, we recommend enabling energy-saving mode during low-load periods.

[0131] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0132] This embodiment also provides a server fault prediction device, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as described herein. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0133] Embodiments of this application also provide a server fault prediction device, such as... Figure 6 As shown, this device is used in a baseboard management controller in a server and includes:

[0134] The data acquisition module 601 is used to acquire monitoring data from the target sensor. The monitoring data represents the monitoring results of the target sensor on the target hardware in the server. The baseboard management controller is communicatively connected to the target sensor.

[0135] The feature extraction module 602 is used to perform feature extraction processing on the monitoring data to obtain dynamic change features, which represent the trend of the target hardware's working state changing over time.

[0136] The fault prediction module 603 is used to perform feature extraction processing on the monitoring data to obtain dynamic change features, which represent the trend of the target hardware's working state changing over time.

[0137] As an optional embodiment, the fault prediction module 603 is specifically used to input dynamically changing features into a preset machine learning model, so as to output fault prediction results through the preset machine learning model;

[0138] The preset machine learning model is used to determine the fluctuation of the target hardware's working state by analyzing dynamic change characteristics, and to determine the fault prediction result by classifying the fluctuation.

[0139] As an optional embodiment, a preset machine learning model is used to classify fluctuations based on dynamic alarm thresholds, and the fault prediction module 603 includes:

[0140] The acquisition submodule is used to acquire monitoring data within a first preset time period;

[0141] The adjustment submodule is used to adjust the dynamic alarm threshold based on the monitoring data within the first preset time period.

[0142] As an optional embodiment, the server fault prediction device further includes:

[0143] The false alarm detection module is used to judge the fault prediction results in order to identify false alarm events;

[0144] The threshold adjustment module is used to adjust the dynamic alarm threshold based on false alarm events.

[0145] As an optional embodiment, the threshold adjustment module includes:

[0146] The event statistics submodule is used to obtain the number of false alarm events that occur consecutively within a second preset time period;

[0147] The threshold update submodule is used to increase or decrease the dynamic alarm threshold if the number of consecutive false alarm events is greater than or equal to a preset value.

[0148] As an optional embodiment, the monitoring data includes monitoring results from multiple target hardware components; the server fault prediction device further includes:

[0149] The determination module is used to determine multiple health levels corresponding to the fault prediction results of multiple monitoring results;

[0150] The calculation module is used to perform weighted summation on multiple health scores to obtain the server's first score result;

[0151] The attenuation processing module is used to attenuate the first score result according to the preset attenuation factor to obtain the second score result.

[0152] As an optional embodiment, the server fault prediction device further includes:

[0153] The generation module is used to generate a visual analysis result of the server's health status based on the second rating result, the first rating result, and at least one of multiple health levels.

[0154] The sending module is used to send the visualization analysis results to the client so that the visualization analysis results can be displayed on the client.

[0155] Embodiments of this application also provide a server, such as Figure 7 As shown, it includes a memory 10 and a processor 20. The memory 10 stores a computer program, and the processor 20 is configured to run the computer program to perform the steps in any of the above-described embodiments of the server fault prediction method.

[0156] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above-described server fault prediction method embodiments when running.

[0157] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0158] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0159] The foregoing has provided a detailed description of a server fault prediction method, apparatus, server, and storage medium provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only intended to aid in understanding the method and core ideas of this application. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A method for predicting server failures, characterized in that, The method is applied to a baseboard management controller in a server, and the method includes: The monitoring data is obtained from the target sensor, which represents the monitoring result of the target sensor on the target hardware in the server. The baseboard management controller is communicatively connected to the target sensor. The monitoring data is processed by feature extraction to obtain dynamic change features, which represent the trend of the working state of the target hardware changing over time. Based on the dynamic change characteristics, the faults of the target hardware are predicted to obtain the fault prediction results of the target hardware.

2. The method according to claim 1, characterized in that, The method of predicting faults in the target hardware based on the dynamic change characteristics includes: The dynamic change features are input into a preset machine learning model, and the fault prediction result is output through the preset machine learning model; The preset machine learning model is used to determine the fluctuation of the working state of the target hardware by analyzing the dynamic change characteristics, and to determine the fault prediction result by classifying the fluctuation.

3. The method according to claim 2, characterized in that, The preset machine learning model is used to classify the fluctuation situation based on a dynamic alarm threshold; the method further includes: Obtain the monitoring data within the first preset time period; The dynamic alarm threshold is adjusted based on the monitoring data within the first preset time period.

4. The method according to claim 3, characterized in that, The method further includes: The fault prediction results are evaluated to determine false alarm events; The dynamic alarm threshold is adjusted based on the false alarm events.

5. The method according to claim 4, characterized in that, The step of adjusting the dynamic alarm threshold based on the false alarm event includes: Obtain the number of false alarm events that occur consecutively within a second preset time period; If the number of consecutive false alarm events is greater than or equal to a preset value, the dynamic alarm threshold is increased or decreased.

6. The method according to any one of claims 1 to 5, wherein the monitoring data includes monitoring results of a plurality of the target hardware; characterized in that, The method further includes: Determine multiple health levels corresponding to the fault prediction results of multiple monitoring results; The multiple health scores are weighted and summed to obtain the first score result of the server. The first scoring result is attenuated according to a preset attenuation factor to obtain the second scoring result.

7. The method according to claim 6, characterized in that, The method further includes: Based on the second rating result, the first rating result, and at least one of the multiple health scores, a visual analysis result of the server's health status is generated; The visualization analysis results are sent to the client so that the visualization analysis results can be displayed through the client.

8. A fault prediction device for a server, characterized in that, The device includes: A data acquisition module is used to acquire monitoring data from a target sensor, wherein the monitoring data represents the monitoring results of the target sensor on the target hardware in the server, and the baseboard management controller is communicatively connected to the target sensor. The feature extraction module is used to perform feature extraction processing on the monitoring data to obtain dynamic change features, which represent the trend of the working state of the target hardware changing over time. The fault prediction module is used to perform feature extraction processing on the monitoring data to obtain dynamic change features, which represent the trend of the working state of the target hardware changing over time.

9. A server, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the server fault prediction method as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of the server fault prediction method as described in any one of claims 1 to 7.

Images