Server fault prediction method and device, electronic equipment and storage medium

Abstract

The invention provides a server fault prediction method and device, electronic equipment and a storage medium. The method comprises the following steps: obtaining time sequence data corresponding to an operation index when a server operates, dividing the time sequence data corresponding to the operation index into a plurality of time windows, and obtaining a target dynamic feature corresponding to each time window; further predicting the fault probability of the server based on the target dynamic feature corresponding to each time window under the operation index, and outputting a server fault alarm indication when the fault probability is greater than a preset threshold value; according to the method, the fault probability of the server is predicted according to the current time, namely the server operation condition before the current time, so that the possible fault of the server is predicted before the fault occurs.

Description

Technical Field

[0001] The present application relates to the technical field of fault prediction, and in particular to a server fault prediction method, device, electronic device and storage medium. Background Art

[0002] Currently, the baseboard management controller (BMC) included in the server can provide basic fault detection and alarm functions. For example, if the server has already failed, it can detect the fault and issue an alarm to remind relevant personnel to maintain the server. However, the BMC can only detect the fault when the server has already failed, and cannot predict the possible fault before the fault occurs. Summary of the invention

[0003] In view of this, the present application provides a server failure prediction method, device, electronic device and storage medium to predict server failures.

[0004] The technical solutions provided by this application are as follows:

[0005] According to an embodiment of the first aspect of the present application, a server failure prediction method is provided, the method comprising:

[0006] Obtaining time series data corresponding to the operating indicators of the server during operation; the operating indicators are used to describe the operating status of the server;

[0007] For the time series data corresponding to the operation indicator, the time series data is divided into multiple time windows to obtain the target dynamic features for describing the operation indicator in each time window; the target dynamic features are used to identify the local fluctuations and short-term trends of the operation indicator in each time window;

[0008] The failure probability of the server is predicted based on the target dynamic characteristics corresponding to each time window under the operation index, and a server failure alarm indication is output when the failure probability is greater than a preset threshold.

[0009] Optionally, obtaining time series data corresponding to the operating indicators of the server during operation includes:

[0010] Acquiring, through a baseboard management controller BMC deployed in the server, operating indicators collected within a set time period before a current moment;

[0011] The operating indicators include one or more of CPU temperature, fan speed, memory usage and power supply power.

[0012] Optionally, the target dynamic feature includes a CNN feature, or at least one of a statistical feature and a frequency domain feature and a CNN feature; the statistical feature is used to identify the overall trend and distribution of the operating indicator, and the frequency domain feature is used to identify the potential periodic fluctuation of the operating indicator; the CNN feature is determined by:

[0013] For each time window under the operation indicator, the time series data included in the time window is standardized to obtain preprocessed time series data;

[0014] The preprocessed time series data is input into a trained convolutional neural network (CNN) to obtain the CNN features corresponding to the time window.

[0015] Optionally, predicting the failure probability of the server based on the target dynamic feature corresponding to each time window under the operation indicator includes:

[0016] Inputting the target dynamic features corresponding to each time window under the operation index into the trained long short-term memory network LSTM to obtain the LSTM network output; the LSTM network output is used to characterize the long-term dependency relationship between the time series data corresponding to the operation index;

[0017] The LSTM network output is input into the deployed fully connected layer to obtain the failure probability of the server.

[0018] Optionally, the LSTM network is trained based on time series data corresponding to the operating indicators of the server during operation, system logs, and historical server fault information obtained during a training phase;

[0019] The system log includes one or more of an operation log, an event log and an SDS log, and the server historical fault information includes historical fault records of the server or other servers and the server operation status in a specified time period before and after the fault occurs.

[0020] Optionally, the method further comprises:

[0021] When a failure of the server is detected, the operating indicators collected within a set time period before the failure is obtained, and the operating indicators collected within the set time period are used as training data to train the LSTM network.

[0022] According to an embodiment of the second aspect of the present application, a server failure prediction device is provided, the device comprising:

[0023] A data acquisition unit, used to obtain time series data corresponding to the operation index of the server during operation; the operation index is used to describe the operation status of the server;

[0024] A feature acquisition unit is used to divide the time series data corresponding to the operation index into multiple time windows, and obtain the target dynamic features for describing the operation index in each time window; the target dynamic features are used to identify the local fluctuations and short-term trends of the operation index in each time window; the target dynamic features include CNN features, or at least one of statistical features and frequency domain features, and CNN features; the statistical features are used to identify the overall trend and distribution of the operation index, and the frequency domain features are used to identify the potential periodic fluctuations of the operation index;

[0025] A fault prediction unit is used to predict the failure probability of the server based on the target dynamic characteristics corresponding to each time window under the operation index, and output a server fault alarm indication when the failure probability is greater than a preset threshold.

[0026] Optionally, the data obtaining unit is specifically used for:

[0027] Acquiring, through a baseboard management controller BMC deployed in the server, operating indicators collected within a set time period before a current moment;

[0028] The operating index includes one or more of: CPU temperature, fan speed, memory usage, and power supply power;

[0029] And / or, the target dynamic feature includes a CNN feature, or at least one of a statistical feature and a frequency domain feature and a CNN feature; the statistical feature is used to identify the overall trend and distribution of the operating indicator, and the frequency domain feature is used to identify the potential periodic fluctuation of the operating indicator; the feature acquisition unit is specifically used to:

[0030] For each time window under the operation indicator, the time series data included in the time window is standardized to obtain preprocessed time series data;

[0031] Inputting the preprocessed time series data into a trained convolutional neural network (CNN) to obtain CNN features corresponding to the time window;

[0032] And / or, the fault prediction unit is specifically used for:

[0033] Inputting the target dynamic features corresponding to each time window under the operation index into the trained long short-term memory network LSTM to obtain the LSTM network output; the LSTM network output is used to characterize the long-term dependency relationship between the time series data corresponding to the operation index;

[0034] Input the LSTM network output into the deployed fully connected layer to obtain the failure probability of the server;

[0035] And / or, the LSTM network is trained based on time series data corresponding to the operating indicators of the server during operation, system logs, and historical server fault information obtained during a training phase;

[0036] The system log includes one or more of an operation log, an event log, and an SDS log, and the server historical fault information includes historical fault records of the server or other servers and the server operation status in a specified time period before and after the fault occurs;

[0037] And / or, the fault prediction unit is further used for:

[0038] When a failure of the server is detected, the operating indicators collected within a set time period before the failure is obtained, and the operating indicators collected within the set time period are used as training data to train the LSTM network.

[0039] According to an embodiment of the third aspect of the present application, there is provided an electronic device, including:

[0040] processor; and,

[0041] A computer-readable storage medium, wherein computer program instructions are stored in the computer-readable storage medium, and when the computer program instructions are executed by the processor, the processor is caused to execute the method according to the first aspect.

[0042] According to an embodiment of the fourth aspect of the present application, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor executes the method according to the first aspect.

[0043] It can be seen from the above technical scheme that the present application obtains the time series data corresponding to the operating indicators of the server during operation, divides the time series data corresponding to the operating indicators into multiple time windows, and obtains the target dynamic features corresponding to each time window, and further predicts the failure probability of the server based on the target dynamic features corresponding to each time window under the operating indicators, and outputs a server failure alarm indication when the failure probability is greater than a preset threshold. According to the time series data corresponding to the operating indicators of the server during operation, that is, the server operation status before the current moment, the failure probability of the server is predicted, thereby realizing the prediction of possible server failures before the failure occurs. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0045] Figure 1 A flowchart of a server failure prediction method provided in an embodiment of the present application;

[0046] Figure 2 A schematic diagram of the overall structure of a server fault prediction network provided in an embodiment of the present application;

[0047] Figure 3 A schematic diagram of the overall process of the server failure prediction method provided in the embodiment of the present application;

[0048] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application;

[0049] Figure 5 A structural diagram of a server fault prediction device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to enable those skilled in the art to better understand the technical solutions provided by the embodiments of the present application and to make the above-mentioned purposes, features and advantages of the embodiments of the present application more obvious and understandable, the technical solutions in the embodiments of the present application are further described in detail below in conjunction with the accompanying drawings.

[0051] Please refer to Figure 1 , Figure 1 A flowchart of a server failure prediction method provided in an embodiment of the present application.

[0052] like Figure 1 As shown, the method may include the following steps:

[0053] Step 101, obtaining time series data corresponding to the operating indicators of the server during operation.

[0054] In this embodiment, it is necessary to predict possible server failures based on server data before the current moment. Therefore, it is first necessary to obtain relevant parameters of the server within a period of time up to the current moment, such as time series data corresponding to the operating indicators of the server during operation.

[0055] The length of the time period corresponding to the acquired time series data can be set according to actual conditions, and this application does not impose any restrictions on this.

[0056] As an embodiment, a method for obtaining time series data corresponding to an operating indicator of a server during operation may include:

[0057] The baseboard management controller BMC deployed in the server is used to obtain the operation indicators collected within a set time period before the current moment;

[0058] The operation index may include one or more of: CPU temperature, fan speed, memory usage, and power supply power.

[0059] In this embodiment, various operating indicators of the server can be collected through the baseboard management controller BMC deployed in the server, such as CPU temperature, fan speed, memory usage, power supply power, etc., which are indicators used to describe the operating status of the server; various operating indicators of the server can also be collected through server management tools or remote monitoring platforms. This application does not limit the method of collecting server indicators and the specific parameters included in the server indicators.

[0060] At this point, the description of step 101 ends, and step 102 is executed below.

[0061] Step 102 , for the time series data corresponding to the operation indicator, the time series data is divided into a plurality of time windows, and a target dynamic feature for describing the operation indicator in each time window is obtained.

[0062] After the time series data corresponding to the operating indicators are collected, sliding window processing may be performed on the time series data corresponding to each indicator, that is, the time series data corresponding to each indicator may be divided into multiple time windows.

[0063] As an embodiment, the length of each time window can be set to the same fixed length, and the time series data included in each time window can overlap to improve the utilization of the time series data. In this case, the overlapping data helps to capture more local changes and time series information.

[0064] Furthermore, after the time series data is divided into a plurality of time windows, the target dynamic features corresponding to each time window may be obtained.

[0065] Since the time series data corresponding to the operating indicators are segmented to obtain multiple time windows, and the target dynamic features are determined based on the time series data corresponding to each time window, the target dynamic features can identify the local fluctuations and short-term trends of the operating indicators in each time window.

[0066] As an embodiment, the target dynamic features may include only CNN features, or may include at least one of statistical features and frequency domain features as well as CNN features; wherein the statistical features are used to identify the overall trend and distribution of the indicator, and the frequency domain features are used to identify potential periodic fluctuations of the indicator.

[0067] As an embodiment, before obtaining the target dynamic features corresponding to the time window, the time series data included in each time window may be standardized so that different types of data corresponding to various types of operating indicators can eliminate dimensional effects.

[0068] Specifically, the time series data included in each time window can be standardized by the following formula:

[0069]

[0070] Among them, x' is the data obtained after standardization processing, x is the original data, μ is the mean of the time series data included in the time window, and σ is the standard deviation of the time series data included in the time window.

[0071] For each value in the time series data included in each time window, the above-mentioned standardization processing is performed to obtain the time series data after the standardization processing. In this embodiment, the process of standardizing the time series data is actually the preprocessing of the data before feature extraction, and the time series data after the standardization processing is subsequently recorded as the preprocessed time series data.

[0072] As an example, CNN features may be determined in the following manner:

[0073] For each time window under the operation indicator, the time series data included in the time window is standardized to obtain preprocessed time series data;

[0074] The preprocessed time series data is input into the trained convolutional neural network (CNN) to obtain the CNN features corresponding to the time window.

[0075] In this embodiment, the time series data preprocessed as above can be used as the input of the trained CNN network, so as to perform feature extraction on the preprocessed time series data through the CNN network to obtain the CNN features corresponding to the time window.

[0076] As an embodiment, CNN features can be extracted according to the following formula:

[0077]

[0078] Among them, y i,j represents the obtained CNN features, x i+m,j+n Represents the value of the input data (in matrix form) at position (i+m, j+n), w m,n Represents the value of the convolution kernel w (in matrix form) at position (m, n).

[0079] As an embodiment, the statistical features may include features such as mean, standard deviation, maximum value, minimum value, etc., wherein the mean and standard deviation may be determined by the following formula:

[0080]

[0081] Among them, Mean represents the mean, n represents the total number of data included in the time window, and x i Represents the value of the i-th data in the time window.

[0082]

[0083] Among them, StdDev represents the standard deviation, n represents the total number of data included in the time window, and x i represents the value of the i-th data in the time window, and μ is the mean of the time series data included in the time window (i.e. the above-mentioned Mean).

[0084] As an example, the frequency domain features may be determined by Fourier transform:

[0085]

[0086] Among them, X(f) is the frequency domain feature of the time window, x(t) is the time series data included in the time window, and e -j2πft / N The frequency domain features can be used to identify potential periodic fluctuations in the time series data included in the time window.

[0087] In this embodiment, the CNN feature can be used alone as the target dynamic feature, or at least one of the statistical features or frequency domain features can be combined with the CNN feature to form a higher-dimensional feature vector as the target dynamic feature, so that the target dynamic feature can improve the ability to identify the overall trend and distribution of the indicator and identify the potential periodic fluctuations of the indicator, thereby further improving the accuracy of server fault prediction.

[0088] As an embodiment, a method for combining statistical features and frequency domain features with CNN features to obtain target dynamic features may include:

[0089] f fusion =[f CNN ,f stats ,f freq ]

[0090] Among them, f fusion represents the target dynamic characteristics, f CNN represents the CCN feature, f stats represents the statistical characteristics, f freq Similarly, it is also possible to combine only CNN features with statistical features, or only CNN features with frequency domain features, which is not limited in this application.

[0091] At this point, the description of step 102 ends, and step 103 is executed next.

[0092] Step 103 , predicting the failure probability of the server based on the target dynamic features corresponding to each time window under the operation index, and outputting a server failure alarm indication when the failure probability is greater than a preset threshold.

[0093] After determining the target dynamic features corresponding to each time window under the operation index, the failure probability of the server can be predicted based on the target dynamic features corresponding to each time window under the operation index.

[0094] As an embodiment, a method for predicting the failure probability of a server based on the target dynamic feature corresponding to each time window under the operation index may include:

[0095] The target dynamic features corresponding to each time window under the operation index are input into the trained long short-term memory network LSTM to obtain the LSTM network output; the LSTM network output is used to characterize the long-term dependency relationship between the time series data corresponding to the operation index;

[0096] The LSTM network output is input into the deployed fully connected layer to obtain the failure probability of the server.

[0097] In this embodiment, a trained LSTM network can be pre-deployed in the server, and the long-term dependencies between the time series data corresponding to the operating indicators can be captured through the LSTM network. The long-term dependencies between the time series data corresponding to the operating indicators are further converted into the failure probability of the server through the fully connected layer, so as to realize the prediction of server failure.

[0098] In this embodiment, the LSTM network is trained based on the time series data corresponding to the operating indicators of the server during operation obtained in the training phase, the system log, and the historical fault information of the server;

[0099] Among them, the system log includes one or more of the operation log, event log and SDS log, and the server historical fault information includes the historical fault records of this server or other servers and the server operation status in the specified time period before and after the fault occurs.

[0100] It should be noted that the time series data corresponding to the operating indicators obtained in the training phase are all time series data corresponding to the time period in which the health status of the server has been clearly determined (that is, whether the server has failed during the period). Based on this, the time series data corresponding to the operating indicators can be used as training data, and the health status of the server in the period (normal or faulty) can be used as the training label to train the CNN network and the LSTM network.

[0101] During the training phase, in addition to using the time series data corresponding to the time period in which the health status of the server has been determined (i.e., whether the server has failed during the period) as sample data, system logs (operation logs, event logs, and SDS logs, etc.) and historical server failure information can also be used as sample data to train the CNN network and LSTM network.

[0102] It is easy to understand that using system logs and historical server failure information (including historical failure records of this server or other servers and the server operation status in a specified time period before and after the failure) as sample data to train the CNN network and LSTM network can provide more dimensional references for possible server failures and parameters before and after the failure, making the final trained network prediction more accurate.

[0103] In this embodiment, LSTM can effectively capture the long-term dependencies in the time series data corresponding to the operation indicators through its gating mechanism (forget gate, input gate and output gate). This ability makes LSTM suitable for predicting future server failures and providing early warnings to administrators so that protective measures can be taken in time.

[0104] The following is a brief description of the LSTM network:

[0105] The LSTM network includes a forget gate, an input gate, and an output gate, which are used to adjust the cell state in the LSTM network (historical information stored in the network) to output the current hidden state (server failure prediction result).

[0106] Specifically, the forget gate is used to decide whether to forget certain information from previous moments based on the current operation of the server. For example, when the CPU temperature suddenly rises, the forget gate may decide to retain more past temperature information in order to capture trends that may lead to overheating.

[0107] The output formula of the forget gate is as follows:

[0108] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0109] Among them, f t Represents the output of the forget gate, which determines how much information in the cell state at the previous moment needs to be "forgotten" or discarded. In hardware state prediction, this can be seen as determining how much influence the current hardware state has on the future state.

[0110] σ: Sigmoid activation function, the output value is between 0 and 1. 0 means complete forgetting, and 1 means complete retention.

[0111] W f : Weight matrix, which represents the parameters used to calculate the forget gate in this layer of neural network, and is used to adjust the degree of forgetting according to the input data and the hidden state at the previous moment. This matrix is ​​obtained by optimizing the parameters during the training process.

[0112] h t-1 : The hidden state at the previous moment contains the network's "memory" information about the hardware state at the previous moment, which is calculated from the predicted output or hidden state at the previous moment.

[0113] x t : Input data at the current moment, which is the hardware status data collected from the BMC platform, such as temperature, load, etc., which enters the LSTM after preprocessing and feature extraction.

[0114] b f : The bias term of the forget gate controls the bias of forgetting and helps the model adjust the degree of “forgetting”.

[0115] In this embodiment, the input gate is used to write important information at the current moment into the cell state. For example, if a sudden increase in CPU power consumption is detected, the input gate will calculate the cell state that needs to be updated to reflect the impact of this change on future fault predictions.

[0116] The output formula of the input gate is as follows:

[0117] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0118] Candidate cell states:

[0119]

[0120] Among them, i t represents the output of the input gate, which determines how much information of the input data at the current moment needs to be written into the cell state. For the hardware state prediction task, this means that we need to decide which hardware data (e.g., sensor readings, load changes) is helpful for the current prediction.

[0121] W i : The weight matrix of the input gate determines how the combination of the current input data and the hidden state of the previous moment affects the output of the input gate, thereby affecting the update of the cell state.

[0122] The candidate cell state is a "potential" update value that represents the new information of the candidate under the current input and hidden state. After being processed by the tanh activation function, the value is ensured to be between -1 and 1.

[0123] W C : The weight matrix of the candidate cell state, together with the input data and the hidden state at the previous moment, determines the new information that needs to be added to the cell state.

[0124] b i and b C : Bias term between input gate and candidate cell state, helping to adjust the impact of input on cell state update.

[0125] Furthermore, the cell state can be updated through the output results of the forget gate and the input gate, ensuring that the model can correctly accumulate and update the historical information of the server operation status, thereby improving the accuracy of fault prediction.

[0126] The cell state update formula is:

[0127]

[0128] Among them, C t Represents the cell state at the current moment, which contains all long-term memories from the previous moment to the current moment. It is the result of the combined action of the forget gate and the input gate. In hardware prediction, the cell state represents the historical memory of the hardware state and the update based on the current input. For example, it reflects the accumulated CPU operation history.

[0129] C t-1 : The cell state at the previous moment, indicating the hardware state memory at the previous moment.

[0130] f t ·C t-1 : The part controlled by the forget gate determines the information that needs to be retained in the cell state at the previous moment.

[0131] The part controlled by the input gate determines which part of the input data at the current moment needs to be added to the cell state.

[0132] In this embodiment, the output gate is used to control the output of the LSTM unit, that is, to generate the prediction result at the current moment, such as predicting the probability that the server is in an abnormal state.

[0133] Output gate control signal formula:

[0134] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0135] Hidden state update formula:

[0136] h t =o t tanh(C t )

[0137] Among them, t Represents the control signal of the output gate, which determines which information in the cell state will eventually be output to the next layer or as the final prediction result.

[0138] h t : Hidden state, which is also the output at the current moment, represents the prediction of the server state at the next moment W o : The weight matrix of the output gate, which controls the contribution of the hidden state to the current prediction.

[0139] b o : The bias term of the output gate, which adjusts the output deviation.

[0140] This completes the introduction to the LSTM network.

[0141] After obtaining the LSTM network features output by the LSTM network, it is also necessary to pass the features to the fully connected layer so that the long-term dependency features captured by the LSTM can be linearly transformed and mapped through the fully connected layer so that they can correspond to the server failure prediction.

[0142] As an example, the output of the fully connected layer can be calculated by the following formula:

[0143] y=Softmax(W d ·h t +b d )

[0144] Among them, y is the predicted value of server failure probability, Softmax() is the activation function, and h t is the hidden state output by the LSTM network, W d is the weight matrix of the fully connected layer, b d is the bias vector.

[0145] It is easy to understand that during the training process, for the LSTM model and the fully connected layer, the parameters such as the weight matrices and bias terms in the input gate, forget gate, and output gate formulas can be adjusted.

[0146] After predicting the server failure probability, if it is determined that the failure probability is greater than a preset threshold, a server failure alarm indication may be output to notify relevant personnel to take preventive measures.

[0147] This concludes the description of step 103.

[0148] In this embodiment, the cross entropy loss function can be used to measure the classification error for the training of CNN network and LSTM network:

[0149]

[0150] Among them, Loss is the loss value, y i is the predicted value, is the true value.

[0151] In this embodiment, an optimization algorithm such as an Adam optimizer or an SGD optimizer can be used to adjust the model parameters to minimize the loss function, and the accuracy, precision, recall and other indicators of the model can be evaluated through a validation set to adjust the hyperparameters of the model.

[0152] As an embodiment, when a server failure is detected, operating indicators collected within a set time period before the failure occurs are obtained, and the operating indicators collected within the set time period are used as training data to train the LSTM network.

[0153] When a server failure is detected, the actual server failure situation can be fed back to the network and network-related parameters can be updated to improve the accuracy of the prediction.

[0154] This concludes Figure 1 Description of the server failure prediction method in .

[0155] The present application obtains time series data corresponding to the operating indicators of the server during runtime, divides the time series data corresponding to each indicator into multiple time windows, and obtains the target dynamic features corresponding to each time window, further predicts the failure probability of the server based on the target dynamic features corresponding to each time window under the operating indicators, and outputs a server failure alarm indication when the failure probability is greater than a preset threshold. The application predicts the failure probability of the server based on the time series data corresponding to the operating indicators of the server during runtime, that is, the server operating status before the current moment, thereby realizing the prediction of possible server failures before the failure occurs.

[0156] Specifically, the present application can process the collected multi-dimensional time series data (such as temperature, power consumption, voltage, etc.) of key hardware in the server by deploying CNN network and LSTM network in the server to predict the failure probability of the server. Among them, the CNN network is used to extract features from multi-dimensional time series data, capture local patterns and spatial relationships in the data, and can be combined with frequency domain features and statistical features as target dynamic features; the LSTM network is used to further process the target dynamic features, perform time series modeling, and capture long-term dependencies in the data.

[0157] The server failure prediction method proposed in this application is described below through a specific embodiment.

[0158] Example 1

[0159] Please refer to Figure 2 as well as Figure 3 , Figure 2 A schematic diagram of the overall structure of a server fault prediction network provided in an embodiment of the present application. Figure 3 A schematic diagram of the overall process of the server failure prediction method provided in an embodiment of the present application.

[0160] The server failure prediction method proposed in this embodiment includes the following steps:

[0161] Step 301: data collection.

[0162] In this embodiment, the BMC may be used to collect time series data corresponding to multiple indicators in the operation of the server, such as CPU temperature and power consumption time series data.

[0163] Step 302: data preprocessing.

[0164] In this embodiment, after the collected time series data is divided into multiple time windows, the time series data included in each time window can be preprocessed, such as standardizing the time series data, and the standardized time series data is used as input data.

[0165] Step 303: feature extraction.

[0166] In this embodiment, the time series data after the normalization process can be used as the input in the input layer (corresponding to Figure 2 The input layer in the image is input into the CNN network, and the CNN network performs pooling, node expansion and other processing (corresponding to Figure 2 ), and obtain the target dynamic features.

[0167] Step 304: Fault prediction.

[0168] In this embodiment, the extracted multiple target dynamic features can be input into the LSTM network (corresponding to Figure 2 In the LSTM layer, the LSTM network features are obtained to characterize the long-term dependencies between the target dynamic features.

[0169] The LSTM network features are further input into the fully connected layer to convert the LSTM network features into server failure prediction probability (corresponding to Figure 2 The fully connected layer in (corresponding to Figure 2 ).

[0170] This concludes the description of Embodiment 1.

[0171] Please refer to Figure 4 , Figure 4 It is a schematic structural diagram of an electronic device proposed in an embodiment of the present application. At the hardware level, the electronic device includes a processor, an internal bus, a network interface, and a computer-readable storage medium, and of course may also include hardware required for other services. The processor reads the corresponding computer program instructions from the computer-readable storage medium and runs them to form a terminal interaction device at the logical level. Of course, in addition to the software implementation method, the present application does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0172] Please refer to Figure 5 , Figure 5 : is a structural diagram of a server fault prediction device proposed in an embodiment of the present application. Figure 5As shown, the device may include a data acquisition unit 501, a feature acquisition unit 502, and a fault prediction unit 503. Specifically, the device includes:

[0173] The data acquisition unit 501 is used to obtain time series data corresponding to the operation index of the server during operation; the operation index is used to describe the operation status of the server;

[0174] The feature acquisition unit 502 is used to divide the time series data corresponding to the operation index into multiple time windows, and obtain the target dynamic feature for describing the operation index in each time window; the target dynamic feature is used to identify the local fluctuation and short-term trend of the operation index in each time window;

[0175] The fault prediction unit 503 is used to predict the failure probability of the server based on the target dynamic characteristics corresponding to each time window under the operation index, and output a server failure alarm indication when the failure probability is greater than a preset threshold.

[0176] Optionally, the data obtaining unit 501 is specifically used for:

[0177] The baseboard management controller BMC deployed in the server is used to obtain the operation indicators collected within a set time period before the current moment;

[0178] The operation index includes one or more of: CPU temperature, fan speed, memory usage, and power supply power;

[0179] And / or, the target dynamic feature includes a CNN feature, or at least one of a statistical feature and a frequency domain feature and a CNN feature; the statistical feature is used to identify the overall trend and distribution of the operating indicator, and the frequency domain feature is used to identify the potential periodic fluctuation of the operating indicator; the feature acquisition unit 502 is specifically used to:

[0180] For each time window under the operation indicator, the time series data included in the time window is standardized to obtain preprocessed time series data;

[0181] Input the preprocessed time series data into the trained convolutional neural network (CNN) to obtain the CNN features corresponding to the time window;

[0182] And / or, the fault prediction unit 503 is specifically used for:

[0183] The target dynamic features corresponding to each time window under the operation index are input into the trained long short-term memory network LSTM to obtain the LSTM network output; the LSTM network output is used to characterize the long-term dependency relationship between the time series data corresponding to the operation index;

[0184] Input the LSTM network output into the deployed fully connected layer to obtain the failure probability of the server;

[0185] And / or, the LSTM network is trained based on time series data corresponding to the operating indicators of the server during runtime obtained in the training phase, system logs, and historical server fault information;

[0186] The system log includes one or more of the operation log, event log and SDS log. The server historical fault information includes the historical fault records of the server or other servers and the server operation status in a specified time period before and after the fault occurs.

[0187] And / or, the fault prediction unit 503 is further configured to:

[0188] When a server failure is detected, the operating indicators collected within a set time period before the failure is obtained, and the operating indicators collected within the set time period are used as training data to train the LSTM network.

[0189] This concludes Figure 5 Description of the server failure prediction device.

[0190] Correspondingly, an embodiment of the present application further provides a computer-readable storage medium, on which a number of computer instructions are stored. When the computer instructions are executed, the method disclosed in the above example of the present application can be implemented.

[0191] Exemplarily, the computer-readable storage medium may be any electronic, magnetic, optical or other physical storage device that may contain or store information, such as executable instructions, data, etc. For example, the computer-readable storage medium may be: RAM (Radom Access Memory), volatile memory, non-volatile memory, flash memory, storage drive (such as hard disk drive), solid state drive, any type of storage disk (such as optical disk, DVD, etc.), or similar storage medium, or a combination thereof.

[0192] The above are only preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the scope of protection of the present application.

Claims

1. A server failure prediction method, characterized in that: The method includes: Obtaining time series data corresponding to the operating indicators of the server during operation; the operating indicators are used to describe the operating status of the server; For the time series data corresponding to the operation indicator, the time series data is divided into multiple time windows to obtain the target dynamic features for describing the operation indicator in each time window; the target dynamic features are used to identify the local fluctuations and short-term trends of the operation indicator in each time window; The failure probability of the server is predicted based on the target dynamic characteristics corresponding to each time window under the operation index, and a server failure alarm indication is output when the failure probability is greater than a preset threshold.

2. The method according to claim 1, characterized in that The obtaining of time series data corresponding to the operating indicators of the server during operation includes: Acquiring, through a baseboard management controller BMC deployed in the server, operating indicators collected within a set time period before a current moment; The operating indicators include one or more of CPU temperature, fan speed, memory usage and power supply power.

3. The method according to claim 1, characterized in that The target dynamic feature includes a CNN feature, or at least one of a statistical feature and a frequency domain feature and a CNN feature; the statistical feature is used to identify the overall trend and distribution of the operating indicator, and the frequency domain feature is used to identify the potential periodic fluctuation of the operating indicator; the CNN feature is determined by: For each time window under the operation indicator, the time series data included in the time window is standardized to obtain preprocessed time series data; The preprocessed time series data is input into a trained convolutional neural network (CNN) to obtain the CNN features corresponding to the time window.

4. The method according to claim 1, characterized in that: The predicting the failure probability of the server based on the target dynamic feature corresponding to each time window under the operation indicator includes: Inputting the target dynamic features corresponding to each time window under the operation index into the trained long short-term memory network LSTM to obtain the LSTM network output; the LSTM network output is used to characterize the long-term dependency relationship between the time series data corresponding to the operation index; The LSTM network output is input into the deployed fully connected layer to obtain the failure probability of the server.

5. The method according to claim 4, characterized in that The LSTM network is trained based on the time series data corresponding to the operating indicators of the server during operation, the system log, and the historical fault information of the server obtained during the training phase; The system log includes one or more of an operation log, an event log and an SDS log, and the server historical fault information includes historical fault records of the server or other servers and the server operation status in a specified time period before and after the fault occurs.

6. The method according to claim 4, characterized in that The method further comprises: When a failure of the server is detected, the operating indicators collected within a set time period before the failure is obtained, and the operating indicators collected within the set time period are used as training data to train the LSTM network.

7. A server failure prediction device, characterized in that: The device includes: A data acquisition unit, used to obtain time series data corresponding to the operation index of the server during operation; the operation index is used to describe the operation status of the server; A feature acquisition unit is used to divide the time series data corresponding to the operation index into multiple time windows, and obtain the target dynamic features for describing the operation index in each time window; the target dynamic features are used to identify the local fluctuations and short-term trends of the operation index in each time window; the target dynamic features include CNN features, or at least one of statistical features and frequency domain features, and CNN features; the statistical features are used to identify the overall trend and distribution of the operation index, and the frequency domain features are used to identify the potential periodic fluctuations of the operation index; A fault prediction unit is used to predict the failure probability of the server based on the target dynamic characteristics corresponding to each time window under the operation index, and output a server fault alarm indication when the failure probability is greater than a preset threshold.

8. The device according to claim 7, characterized in that The data acquisition unit is specifically used for: Acquiring, through a baseboard management controller BMC deployed in the server, operating indicators collected within a set time period before a current moment; The operating index includes one or more of: CPU temperature, fan speed, memory usage, and power supply power; And / or, the target dynamic feature includes a CNN feature, or at least one of a statistical feature and a frequency domain feature and a CNN feature; the statistical feature is used to identify the overall trend and distribution of the operating indicator, and the frequency domain feature is used to identify the potential periodic fluctuation of the operating indicator; the feature acquisition unit is specifically used to: For each time window under the operation indicator, the time series data included in the time window is standardized to obtain preprocessed time series data; Inputting the preprocessed time series data into a trained convolutional neural network (CNN) to obtain CNN features corresponding to the time window; And / or, the fault prediction unit is specifically used for: Inputting the target dynamic features corresponding to each time window under the operation index into the trained long short-term memory network LSTM to obtain the LSTM network output; the LSTM network output is used to characterize the long-term dependency relationship between the time series data corresponding to the operation index; Input the LSTM network output into the deployed fully connected layer to obtain the failure probability of the server; And / or, the LSTM network is trained based on time series data corresponding to the operating indicators of the server during operation, system logs, and historical server fault information obtained during a training phase; The system log includes one or more of an operation log, an event log, and an SDS log, and the server historical fault information includes historical fault records of the server or other servers and the server operation status in a specified time period before and after the fault occurs; And / or, the fault prediction unit is further used for: When a failure of the server is detected, the operating indicators collected within a set time period before the failure is obtained, and the operating indicators collected within the set time period are used as training data to train the LSTM network.

9. An electronic device, characterized in that: include: processor; as well as, A computer-readable storage medium, wherein computer program instructions are stored in the computer-readable storage medium, and when the computer program instructions are executed by the processor, the processor is caused to perform the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, which, when executed by a processor, enable the processor to perform the method according to any one of claims 1 to 6.

Images