WO2020114313A1 - 预测硬盘故障发生时间的方法、装置及存储介质 - Google Patents

预测硬盘故障发生时间的方法、装置及存储介质 Download PDF

Info

Publication number
WO2020114313A1
WO2020114313A1 PCT/CN2019/121647 CN2019121647W WO2020114313A1 WO 2020114313 A1 WO2020114313 A1 WO 2020114313A1 CN 2019121647 W CN2019121647 W CN 2019121647W WO 2020114313 A1 WO2020114313 A1 WO 2020114313A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
hard disk
failure
training
data set
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/121647
Other languages
English (en)
French (fr)
Inventor
杨洪章
屠要峰
杨雅辉
李忠良
郭斌
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
ZTE Corp
Original Assignee
ZTE Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by ZTE Corp filed Critical ZTE Corp
Priority to US17/600,551 priority Critical patent/US11656943B2/en
Priority to EP19892295.7A priority patent/EP3879405A4/en
Priority to JP2021531832A priority patent/JP7158586B2/ja
Publication of WO2020114313A1 publication Critical patent/WO2020114313A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G11—INFORMATION STORAGE
    • G11B—INFORMATION STORAGE BASED ON RELATIVE MOVEMENT BETWEEN RECORD CARRIER AND TRANSDUCER
    • G11B27/00—Editing; Indexing; Addressing; Timing or synchronising; Monitoring; Measuring tape travel
    • G11B27/36—Monitoring, i.e. supervising the progress of recording or reproducing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
    • G06F11/08—Error detection or correction by redundancy in data representation, e.g. by using checking codes
    • G06F11/10—Adding special bits or symbols to the coded information, e.g. parity check, casting out 9's or 11's
    • G06F11/1076—Parity data used in redundant arrays of independent storages, e.g. in RAID systems
    • G06F11/1092—Rebuilding, e.g. when physically replacing a failing disk
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
    • G06F11/0703—Error or fault processing not based on redundancy, i.e. by taking additional measures to deal with the error or fault not making use of redundancy in operation, in hardware, or in data representation
    • G06F11/0751—Error or fault detection not based on redundancy
    • G06F11/0754—Error or fault detection not based on redundancy by exceeding limits
    • G06F11/076—Error or fault detection not based on redundancy by exceeding limits by exceeding a count or rate limit, e.g. word- or bit count limit
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
    • G06F11/0703—Error or fault processing not based on redundancy, i.e. by taking additional measures to deal with the error or fault not making use of redundancy in operation, in hardware, or in data representation
    • G06F11/0766—Error or fault reporting or storing
    • G06F11/0772—Means for error signaling, e.g. using interrupts, exception flags, dedicated error registers
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
    • G06F11/08—Error detection or correction by redundancy in data representation, e.g. by using checking codes
    • G06F11/10—Adding special bits or symbols to the coded information, e.g. parity check, casting out 9's or 11's
    • G06F11/1008—Adding special bits or symbols to the coded information, e.g. parity check, casting out 9's or 11's in individual solid state devices
    • G06F11/1048—Adding special bits or symbols to the coded information, e.g. parity check, casting out 9's or 11's in individual solid state devices using arrangements adapted for a specific error detection or correction feature
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/3003—Monitoring arrangements specially adapted to the computing system or computing system component being monitored
    • G06F11/3034—Monitoring arrangements specially adapted to the computing system or computing system component being monitored where the computing system component is a storage system, e.g. DASD based or network based
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/3055—Monitoring arrangements for monitoring the status of the computing system or of the computing system component, e.g. monitoring if the computing system is on, off, available, not available
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3452—Performance evaluation by statistical analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3466—Performance evaluation by tracing or monitoring
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting

Definitions

  • the present disclosure relates to the field of computer technology, and in particular, to a method, device, and storage medium for predicting the time when a hard disk failure occurs.
  • the hard disk is the main source of failure in the current data center.
  • the hard disk failure prediction technology in the related art blindly pursues the improvement of the detection rate, resulting in a high false detection rate, so many healthy hard disks will be mistaken for faulty hard disks.
  • the related hard disk failure prediction technology can only predict whether the hard disk will fail, but cannot predict the time when the failure may occur, resulting in the predicted failure distance being longer than the real failure time (short one to two weeks, long one to two months), so The life cycle of the hard disk is wasted, and the disk replacement cannot be effectively guided.
  • the efficiency of hard disk failure prediction is low.
  • the present disclosure provides a method, device and storage medium for predicting the occurrence time of a hard disk failure to solve the problem in the related art that the time when a hard disk failure occurs cannot be predicted.
  • a method for predicting the occurrence time of a hard disk failure includes: filtering out hard disks nearing failure from a plurality of hard disks based on the collected state data of the hard disks; calculating at the first preset time The variation and dispersion of each piece of data in the status data of the hard disks nearing failure collected in the segment to obtain the first prediction data set; input the first prediction data set into the first training model to obtain each hard disk The probability of a failure occurring in a second preset time period in the future, wherein the first training model is obtained by training based on a first training data set through a first artificial intelligence algorithm, and the first training data set includes positive sample data And negative sample data, wherein the change and dispersion corresponding to the failed hard disk are positive sample data, and the change and dispersion corresponding to the non-failed hard disk are negative sample data.
  • an apparatus for predicting the occurrence time of a hard disk failure including: a screening module for screening out hard disks nearing failure from a plurality of hard disks based on the collected state data of the hard disks;
  • the calculation module is used to calculate the variation and discrete amount of each piece of data in each state data of the hard disk nearing failure collected during the first preset time period to obtain the first prediction data set;
  • the input module is used to Input the first prediction data set into a first training model to obtain the probability of failure of each hard disk in a second preset time period in the future, wherein the first training model is based on the first training through a first artificial intelligence algorithm
  • the data set is obtained by training.
  • the first training data set includes positive sample data and negative sample data, wherein the variation and dispersion corresponding to the failed hard disk are positive sample data, and the variation and dispersion corresponding to the non-faulty hard disk are negative. sample.
  • an apparatus for predicting the occurrence time of a hard disk failure including: a processor; a memory for storing processor executable instructions; when the instructions are executed by the processor, the following operations are performed : Filter out the hard disks that are nearing failure from multiple hard disks based on the collected state data of the hard disks; calculate the data of each piece of data in the state data of hard disks that are nearing failure collected within the first preset time period The amount of change and the discrete amount to obtain the first prediction data set; input the first prediction data set into the first training model to obtain the probability of failure of each hard disk in a second preset time period in the future, wherein the A training model is obtained by training with a first artificial intelligence algorithm based on a first training data set, where the first training data set includes positive sample data and negative sample data, wherein the amount of change and the dispersion corresponding to the failed hard disk are positive sample data , The amount of change and discrete amount corresponding to the non-faulty hard disk
  • a non-transitory computer-readable storage medium which, when instructions in the storage medium are executed by a processor, enables the processor to execute the description according to the first aspect of the present disclosure Methods.
  • Fig. 1 is a flow chart of a method for predicting the occurrence time of a hard disk failure according to an exemplary embodiment
  • Fig. 2 is a flow chart showing a method for predicting the occurrence time of a hard disk failure according to an exemplary embodiment
  • Fig. 3 is a flow chart showing a method for predicting the occurrence time of a hard disk failure according to an exemplary embodiment
  • Fig. 4 is a block diagram of a device for predicting the occurrence time of a hard disk failure according to an exemplary embodiment
  • Fig. 5 is a block diagram of a device for predicting the time when a hard disk failure occurs according to an exemplary embodiment.
  • Fig. 1 is a flowchart of a method for predicting the occurrence time of a hard disk failure according to an exemplary embodiment. As shown in Fig. 1, the method includes the following steps:
  • Step 101 Screen out the hard disks nearing failure from multiple hard disks based on the collected status data of the hard disks;
  • the method can be applied to the failure prediction of a large number of hard disks in a data center.
  • the hard disk may be, for example, SATA (Serial Advanced Technology Attachment, Serial ATA interface specification ) Interface hard disk.
  • SATA Serial Advanced Technology Attachment, Serial ATA interface specification
  • the status data of all the hard disks in the data center can be collected at a fixed time interval N, and the value of N can be 3 hours.
  • the collection tool used to collect the status data of the hard disk can be the open source toolkit smartmontools, the instruction iostat One of them.
  • the state data of the hard disk collected in a continuous period of time M can be used as the original data set, and the value of M is, for example, 60 days. Assuming that there are g hard disks in the data center, the original data set has a total of g*M/N sample data (hereinafter also referred to as samples). In the method of predicting the occurrence time of a hard disk failure of the present disclosure, the operation of collecting status data of each hard disk can be continuously performed to provide the latest samples.
  • the 20,000 hard drives are selected from the 20,000 hard drives.
  • 20 hard drives you can only predict the failure time of the 20 hard drives, that is, after performing the above step 101, you can only perform the subsequent steps 102 and 103 for the hard drives that are near the fault selected in step 101, reducing The number of hard drives that need to be predicted for failure time reduces the amount of data that needs to be processed, and significantly improves the efficiency of predicting the failure time of hard drives.
  • the status data of the hard disk may include: SMART (Self-Monitoring Analysis and Reporting Technology) information of the hard disk and/or I/O (Input/Output of the hard disk) /Output) information.
  • SMART Self-Monitoring Analysis and Reporting Technology
  • I/O Input/Output of the hard disk
  • the SMART information includes at least one of the following: the number of sectors remapped on the hard disk, the count of unrepairable hardware errors, the error count of hardware ECC (Error Correcting Code, error checking and correction), the original read error rate, The error rate of the seek of the magnetic head and the number of write failures caused by the head being too high from the disc;
  • I/O information includes at least: the number of written blocks (blk_written).
  • the SMART information of a hard disk may include at least the NORMAL values of the following 3 entries: #5 (Reallocated Sector), the number of sectors remapped by the hard disk, #187 (Reported Uncorrectable Errors, hardware unrepairable error count), and # 195 (Hardware ECC Recovered, error count repaired by hardware ECC).
  • #5 Reallocated Sector
  • #187 Reported Uncorrectable Errors, hardware unrepairable error count
  • # 195 Hardware ECC Recovered, error count repaired by hardware ECC
  • the SMART information of the hard disk can be the NORMAL value of the following 6 entries: #1 (Raw Read Error Rate, original read error rate), #5 (Reallocated Sector Count, the number of sectors remapped by the hard disk), #7 ( Seek Error Rate, head seek error rate), #187 (Reported Uncorrectable Errors, hardware unrepairable error count), #189 (High Fly Write, the number of write failures caused by the head being too high) and #195 (Hardware ECC Recovered, error count repaired by hardware ECC), plus one item of I/O information, the state data of the hard disk can have 7 items in total.
  • Step 102 Calculate the change amount and discrete amount of each piece of data in each state data of the hard disk nearing failure collected during the first preset time period to obtain a first prediction data set;
  • step 102 for example, only the status data of each hard disk of the data center collected in the last 7 days may be acquired.
  • the first prediction data set may include various status data of the hard disk (for example, the 7 status data corresponding to the above 7 entries), and each status data includes data collected at different times within the first preset time period. Multiple data; the first prediction data set may also include only one kind of state data of the hard disk. In this case, in step 102, the data of the hard disk that is nearing failure and collected during the first preset time period may be directly calculated. The amount of variation and discreteness of each piece of data in this state data.
  • the above-mentioned discrete quantity can reflect the difference between one piece of status data of the hard disk and the overall mean, the discrete quantity can be variance or standard deviation, and the variation can reflect the degree of change of one piece of status data of the hard disk within the first preset time period.
  • Step 103 Input the first prediction data set into the first training model to obtain the probability of failure of each hard disk in a second preset time period in the future, wherein the first training model is based on the first artificial intelligence algorithm
  • the first training data set is obtained by training.
  • the first training data set includes positive sample data and negative sample data, wherein the variation and dispersion corresponding to the failed hard disk are positive sample data, and the variation and dispersion corresponding to the non-faulty hard disk are The amount is negative sample data.
  • the first preset time period may be greater than the second preset time period, or the two may be equal.
  • the state data of the hard disk in a period of time before the failure can be determined as the data corresponding to the failed hard disk; the state data of the hard disk in a period of time before the failure is determined as the data of the non-faulted hard disk, Or directly delete the part of the data; determine the status data of the non-faulty hard disk as the data corresponding to the non-faulty hard disk.
  • the first artificial intelligence algorithm may include: any one of the logistic regression algorithm, artificial neural network algorithm, random forest algorithm, for example, the logistic regression algorithm can be selected to train the first training data set to obtain the first training model .
  • the method for predicting the occurrence time of a hard disk failure calculates the amount of change and the discrete amount of each piece of state data based on the state data of the hard disk nearing failure collected during the first preset time period, thereby obtaining the first A prediction data set, inputting the first prediction data set into the first training module may also measure the probability of failure of each hard disk in a second preset time period in the future, thereby achieving the purpose of predicting the time when the hard disk fails.
  • the above steps 102 and 103 may be triggered to execute, assuming that the imminently faulty hard disk is h blocks, h ⁇ g.
  • Corresponding variance and variation are input into the first training model obtained in advance, and the probability of failure of each hard disk in the next 7 days can be obtained one by one.
  • the process of predicting the failure time of the h hard disks that are nearing failure may include:
  • the variance and change corresponding to the hard disk are input into the first training model, and the probability of failure in the next 7 days is p_1.
  • the variance and change corresponding to the hard disk are input into the first training model, and the probability of failure in the next 7 days is p_h.
  • the value of the first threshold may range from 50% to 90%, for example If the first threshold is 80%, it is considered that the hard disk will fail in the next 7 days, and the hard disk can be added to the disk replacement alarm list. O&M personnel can perform disk replacement processing operations based on the disk replacement alarm list. The disk replacement processing operation can be performed one by one, for example, the disk replacement can be performed in the order of the hard disk failure probability from high to low. During the disk replacement process, all the data in the failed hard disk can be copied to the newly added hard disk, so that the number of system hard disks remains unchanged.
  • ⁇ is the amount of change
  • a n is the first predetermined period of time to collect status data of n items
  • k is the total number of said first predetermined period of time to collect status data .
  • the method of the embodiment of the present disclosure may predict the time when the hard disk fails based on multiple samples collected within a period of time (that is, the first preset time period).
  • FIG. 2 is a flowchart of a method for predicting a time when a hard disk fails according to an exemplary embodiment. As shown in FIG. 2, the method may further include: on the basis of the method shown in FIG. 1:
  • Step 201 Before inputting the first prediction data set into the first training model, merge the state data of the hard disks in the first preset time period into one item of data, and calculate each item in the item of data Discrete amount and change amount of data; the combined data can include data collected at different times within the first preset time period;
  • step 201 the state data of the same state data among the state data of each hard disk collected in the first preset time period can be combined as An item of status data.
  • Step 202 Mark the dispersion and change of non-faulty hard disks in each hard disk as negative sample data, and mark the dispersion and change of the failed hard disks in each hard disk as positive sample data to obtain the first training data set.
  • step 201 and step 202 may be executed before step 101 and step 102, and may also be executed after step 101 and step 102.
  • FIG. 2 only shows an example case where step 201 is executed before step 101.
  • the data set After obtaining the first training data set, the data set can be trained based on the first artificial intelligence algorithm, and the first training model can be obtained.
  • filtering out hard disks that are nearing failure from multiple hard disks based on the collected state data of the hard disks may include: collecting the collected state data of each hard disk (hereinafter also referred to as the Two test data sets) input a second training model to obtain a classification result corresponding to each state data, and the classification result category includes the imminent failure and health, wherein the second training model uses a second artificial intelligence algorithm It is obtained by training based on a second training data set, which includes positive sample data and negative sample data, wherein the state data corresponding to the non-faulty hard disks in each hard disk is negative sample data, and the state corresponding to the failed hard disk The data is positive sample data; the hard disk whose proportion of the classification result in the category of endangered failure to the total result exceeds the second threshold is determined as the hard disk near the failure, and the second threshold is, for example, 80%.
  • the following describes the process of selecting hard drives that are on the verge of failure from multiple hard drives.
  • multiple samples generated by each hard disk of the above g hard disks in 3 days can be used as the second prediction data set. If the number of samples is too small, it is difficult to effectively collect the deterioration state of the hard disk; if the number of samples is too large, the processing volume is increased. It is appropriate to select 20-40 samples, for example, select 24 samples.
  • the hard disk has more than t classification results, the type is endangered, then it is determined that the predicted result is that the hard disk is endangered.
  • the value range of t may be [12, 22], for example, the value of t may be 18. If the classification result of the hard disk's imminent failure does not exceed t, the hard disk is determined to be a healthy hard disk.
  • the above example judges the hard disks that are on the verge of failure by voting, and can select the hard disks that are on the verge of failure among the hard disks in the data center.
  • the second artificial intelligence algorithm used for training the second training data set may be, for example, any one of a support vector machine algorithm, a Bayes algorithm, and a gradient thruster algorithm.
  • obtaining the second training data set may include: collecting the state data of the hard disks in a first preset period, the first preset period is, for example, 3 hours, to obtain the original data set; Mark each state data in the original data set, mark the state data of the hard disk in the third preset time period before failure as positive sample data, mark the hard disk in the third preset time period before failure
  • the external state data is marked as negative sample data, or the state data outside the third preset time period before the failure of the hard disk is deleted, and the state data of the hard disk that has not failed is marked as negative sample data to obtain the above second Training data set.
  • mark all samples of the original data set mark each sample of the healthy hard disk as a negative sample; mark the sample of the failed disk within K time before failure as a positive sample; mark the failed hard disk before K time of failure
  • the outer sample is marked as a negative sample or discarded.
  • the value of K is, for example, 7 days, and finally a second training data set is formed.
  • the process of obtaining the first training data set may include: reducing the above-mentioned original data set, and only retaining the samples of each hard disk in the last 7 days.
  • the samples of each hard disk in the 7 days are combined to obtain g samples, each sample has 7 kinds of state data, and each piece of data in each state data is used to find the discrete amount and the amount of change.
  • the discretization and change of the healthy hard disk are marked as negative samples, and the discretization and change of the failed hard disk are marked as positive samples to form a second training data set.
  • the method for predicting the occurrence time of a hard disk failure of the present disclosure may further include: updating the data to be updated in the original data set using the newly collected status data of each hard disk, wherein the data to be updated It is the state data collected in the earliest third preset time period in the original data set.
  • the state data collected in the most recent second preset period is periodically replaced with the state data collected in the latest second preset period in a second preset period.
  • State data, where the second preset period is, for example, 7 days, so in this embodiment, the state data collected in the last 7 days can be used to replace the state data collected in the earliest 7 days in the original data set, so that the original data Set to update.
  • the method for predicting the occurrence time of a hard disk failure of the present disclosure will be described below through an example.
  • a total of 2270400 samples of 4730 healthy hard disks are marked as negative samples, and a total of 3920 samples of 70 failed disks 7 days before the failure are marked as positive samples, and the remaining samples are discarded to obtain the second training data set.
  • the model of the support vector machine algorithm is used to obtain the second training model.
  • each hard disk Enter the second training model for each hard disk in a total of 24 samples in the last 3 days, and each hard disk will get 24 classification results. If a hard disk has more than 18 classification results that are deemed to be nearing failure, it is considered that the hard disk is nearing failure, and it is added to the list of hard disks nearing failure, for a total of 23 hard disks.
  • the samples of the 4800 hard disks from the last 7 days are combined, and each entry of each sample finds the variance and change within 7 days.
  • the state data of 4730 healthy hard disk targets are recorded as negative samples, and the state data of 70 failed hard drives are marked as positive samples to obtain the first training data set.
  • the first training model is obtained by training the model using a logistic regression algorithm for the first training data set.
  • the 23 hard drives in the list of hard disks on the verge of failure are combined for the last 7 days, and each entry is calculated for the variance and change within 7 days to obtain 23 samples.
  • 23 samples are input into the first training model one by one, and the probability of failure of each hard disk in the next 7 days is obtained one by one.
  • the hard disk is added to the disk replacement alarm list. It is assumed that there are four hard disks added to the disk replacement alarm list. Then, the operation and maintenance personnel can perform a disk replacement operation on the four hard disks. Add the latest 7-day collected status data of each hard disk to the original data set to replace the status data detected in the original data set at the earliest 7 days.
  • Fig. 3 is a block diagram of an apparatus for predicting the occurrence time of a hard disk failure according to an exemplary embodiment.
  • the apparatus 30 includes the following components: a filtering module 31, configured to Item status data filters out hard drives that are nearing failure from multiple hard drives; the calculation module 32 is used to calculate the amount of change in each piece of data in each state data of hard drives that are nearing failure collected within the first preset time period And a discrete quantity to obtain a first prediction data set; an input module 33 is used to input the first prediction data set into a first training model to obtain the probability of failure of each hard disk in a second preset time period in the future, where ,
  • the first training model is obtained by training based on a first training data set through a first artificial intelligence algorithm, the first training data set includes positive sample data and negative sample data, wherein the change amount and discrete amount corresponding to the failed hard disk It is positive sample data, and the amount of change and dispersion corresponding to non-faulty hard disks are negative sample data.
  • the screening module may include: an input unit configured to input the collected state data of each hard disk into the second training model to obtain a classification corresponding to the state data
  • the category of the classification result includes the imminent failure and health
  • the second training model is obtained by training based on a second training data set through a second artificial intelligence algorithm, and the second training data set includes positive sample data and Negative sample data, wherein the state data corresponding to the non-faulty hard drives in each hard disk is negative sample data, and the state data corresponding to the failed hard disks is positive sample data
  • a determining unit is used to classify the obtained classification results as being near failure
  • the hard disk whose classification result accounts for the total result exceeds the second threshold is determined as a hard disk nearing failure.
  • Fig. 4 is a block diagram of an apparatus for predicting the occurrence time of a hard disk failure according to an exemplary embodiment.
  • the apparatus may further include: a merge module 41 based on the apparatus shown in Fig. 3, with Before inputting the first prediction data set into the first training model, the state data of the hard disks in the first preset time period is combined into one piece of data, and the data of each piece of data in the piece of data is calculated. Discrete and change; the first marking module 42 is used to mark the discrete and change of non-faulty hard drives in each hard drive as negative sample data, and mark the discrete and change of the failed hard drives in each hard drive For positive sample data, the first training data set is obtained.
  • the status data may include SMART information of the hard disk and/or I/O information of the hard disk.
  • the SMART information may include at least one of the following: the number of sectors remapped by the hard disk, the count of unrepairable hardware errors, the count of error repaired by hardware ECC, the original read error rate, and head seek The error rate and the number of write failures caused by the head being too high off the disk; the I/O information includes at least: the number of write blocks.
  • the device may further include: an update module for updating the data to be updated in the original data set with the newly collected status data of each hard disk, wherein the data to be updated is The original data is the state data collected in the earliest third preset time period.
  • the amount of change is calculated by any of the following formulas: as well as Wherein, ⁇ is the amount of change, a n is the first predetermined period of time to collect status data of n items, k is the total number of said first predetermined period of time to collect status data .
  • Fig. 5 is a block diagram of a device 600 for predicting the occurrence time of a hard disk failure according to an exemplary embodiment.
  • the device 600 may be provided as a server. 5
  • the device 600 includes a processor 622, the number of which may be one or more, and a memory 632, which is used to store a computer program executable by the processor 622.
  • the computer program stored in the memory 632 may include one or more modules each corresponding to a set of instructions.
  • the processor 622 may be configured to execute the computer program to perform the above-described method of predicting the time when a hard disk failure occurs.
  • the device 600 may further include a power component 626 and a communication component 650, which may be configured to perform power management of the device 600, and the communication component 650 may be configured to implement communication of the device 600, for example, wired or wireless communication .
  • the device 600 may also include an input/output (I/O) interface 658.
  • the device 600 can operate an operating system based on the memory 632, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, and so on.
  • a non-transitory computer-readable storage medium including program instructions for example, a memory 632 including program instructions
  • the program instructions may be executed by the processor 622 of the device 600 to complete the above A method to predict the time when a hard drive fails.
  • the present disclosure can predict the time when the hard disk fails.
  • the beneficial effects of the present disclosure are as follows:
  • the method for predicting the occurrence time of a hard disk failure calculates the amount of change in each piece of state data based on each state data of a hard disk nearing failure collected during a first preset time period and Discrete quantity, to get the first prediction data set, input the first prediction data set to the first training module can also measure the probability of failure of each hard disk in the second preset time period in the future, so as to realize the prediction of hard disk occurrence The purpose of the time of failure.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Quality & Reliability (AREA)
  • Computing Systems (AREA)
  • Computer Hardware Design (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Debugging And Monitoring (AREA)
  • Test And Diagnosis Of Digital Computers (AREA)

Abstract

一种预测硬盘故障发生时间的方法、装置及存储介质，用以解决相关技术无法对硬盘发生故障的时间进行预测的问题。该方法包括：根据采集到的硬盘的各项状态数据从多个硬盘中筛选出濒临故障的硬盘（101）；计算在第一预设时间段内采集到的濒临故障的硬盘的各项状态数据中的各条数据的变化量以及离散量，得到第一预测数据集（102）；将第一预测数据集输入第一训练模型，得到各硬盘在未来的第二预设时间段内发生故障的概率（103）。

Description

预测硬盘故障发生时间的方法、装置及存储介质
本公开要求享有2018年12月05日提交的名称为“预测硬盘故障发生时间的方法、装置及存储介质”的中国专利申请CN201811484874.3的优先权，其全部内容通过引用并入本文中。
技术领域
本公开涉及计算机技术领域，特别是涉及一种预测硬盘故障发生时间的方法、装置及存储介质。
背景技术
据国内外统计资料显示，硬盘是当前数据中心最主要的故障源。相关报告中指出：“在一个拥有4000个节点的数据中心，平均每6小时就会有一个硬盘发生故障”。硬盘故障如此频繁的发生势必会导致业务中断以及数据丢失等严重后果。如果能提前预测到硬盘故障的发生，将对数据中心的运营带来极大地便利。目前，相关技术中的硬盘故障预测技术在一味追求提升检出率的情况下，导致误检率偏高，因此很多健康硬盘会被误判为故障硬盘。且相关硬盘故障预测技术仅能预测硬盘是否会发生故障，而无法预测故障可能发生的时间，导致预测的故障距离真正故障时间较长(短则一两周，长则一两个月)，因此浪费硬盘生命周期，无法有效指导换盘。且在对硬盘的故障进行预测时受限于处理数据体量大、处理器处理能力有限等原因，导致需要较长的时间才能得出预测结果(短则十几分钟，长则数小时)，因此硬盘故障预测效率较低。
发明内容
本公开提供一种硬盘故障发生时间的预测方法、装置及存储介质，用以解决相关技术中无法对硬盘发生故障的时间进行预测的问题。
根据本公开的一个方面，提供了一种硬盘故障发生时间的预测方法，包括：根据采集到的硬盘的各项状态数据从多个硬盘中筛选出濒临故障的硬盘；计算在第一预设时间段内采集到的濒临故障的硬盘的各项状态数据中的各条数据的变化量以及离散量，得到第一预测数据集；将所述第一预测数据集输入第一训练模型，得到各硬盘在未来的第二预设时间段内发生故障的概率，其中，所述第一训练模型通过第一人工智能算法基于第一训练数据集进行训练得到，所述第一训练数据集中包括正样本数据以及负样本数据，其中，故障硬 盘对应的变化量以及离散量为正样本数据，非故障硬盘对应的变化量以及离散量为负样本数据。
根据本公开的第二个方面，提供了一种预测硬盘故障发生时间的装置，包括：筛选模块，用于根据采集到的硬盘的各项状态数据从多个硬盘中筛选出濒临故障的硬盘；计算模块，用于计算在第一预设时间段内采集到的濒临故障的硬盘的各项状态数据中的各条数据的变化量以及离散量，得到第一预测数据集；输入模块，用于将所述第一预测数据集输入第一训练模型，得到各硬盘在未来的第二预设时间段内发生故障的概率，其中，所述第一训练模型通过第一人工智能算法基于第一训练数据集进行训练得到，所述第一训练数据集中包括正样本数据以及负样本数据，其中，故障硬盘对应的变化量以及离散量为正样本数据，非故障硬盘对应的变化量以及离散量为负样本数据。
根据本公开的第三个方面，提供了一种预测硬盘故障发生时间的装置，包括：处理器；用于存储处理器可执行指令的存储器；当所述指令被处理器执行时，执行如下操作：根据采集到的硬盘的各项状态数据从多个硬盘中筛选出濒临故障的硬盘；计算在第一预设时间段内采集到的濒临故障的硬盘的各项状态数据中的各条数据的变化量以及离散量，得到第一预测数据集；将所述第一预测数据集输入第一训练模型，得到各硬盘在未来的第二预设时间段内发生故障的概率，其中，所述第一训练模型通过第一人工智能算法基于第一训练数据集进行训练得到，所述第一训练数据集中包括正样本数据以及负样本数据，其中，故障硬盘对应的变化量以及离散量为正样本数据，非故障硬盘对应的变化量以及离散量为负样本数据。
根据本公开的第四个方面，提供了一种非临时性计算机可读存储介质，当所述存储介质中的指令由处理器执行时，使得处理器能够执行根据本公开第一个方面所述的方法。
附图说明
图1是根据一示例性实施例示出的一种预测硬盘故障发生时间的方法的流程图；
图2是根据一示例性实施例示出的一种预测硬盘故障发生时间的方法的流程图；
图3是根据一示例性实施例示出的一种预测硬盘故障发生时间的方法的流程图；
图4是根据一示例性实施例示出的一种预测硬盘故障发生时间的装置的框图；
图5是根据一示例性实施例示出的一种预测硬盘故障发生时间的装置的框图。
具体实施方式
下面将结合本公开实施例中的附图，对本公开的实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本公开一部分实施例，而不是全部的实施例。基于 本公开中的实施例，本领域普通技术人员在没有作出创造性劳动的前提下所获得的所有其他实施例，都属于本公开保护的范围。
图1是根据一示例性实施例示出的一种预测硬盘故障发生时间的方法的流程图，如图1所示，该方法包括如下步骤：
步骤101：根据采集到的硬盘的各项状态数据从多个硬盘中筛选出濒临故障的硬盘；
在本公开的预测硬盘故障发生时间的方法的一个示例性实施场景中，该方法可应用于对数据中心的大量硬盘进行故障预测，硬盘例如可以是SATA(Serial Advanced Technology Attachment，串行ATA接口规范)接口的硬盘。在上述步骤101中，可以以固定的时间间隔N采集数据中心的所有硬盘的状态数据，N的取值可以为3小时，用来采集硬盘状态数据的采集工具可以是开源工具包smartmontools、指令iostat中的一种。可将一段连续时间M内的采集的硬盘的状态数据作为原始数据集，M的取值例如是60天。假设数据中心共有g块硬盘，那么原始数据集共计g*M/N条样本数据(以下也简称样本)。在本公开的预测硬盘故障发生时间的方法中，采集各硬盘的状态数据的操作可持续地进行，以提供最新的样本。
举例说明，假设数据中心共有20000块硬盘，根据一段时间内采集到的该20000块硬盘的各项状态数据确定出濒临故障的硬盘有20块，故从20000块硬盘中筛选出了濒临故障的该20块硬盘，则可仅对该20块硬盘进行故障时间的预测，即在执行了上述步骤101之后，可仅针对步骤101中筛选出的濒临故障的硬盘执行后续步骤102以及步骤103，减少了需进行故障时间预测的硬盘的个数，降低了需处理的数据量，显著提高了对硬盘故障时间的预测的效率。
在一种可实现方式中，硬盘的状态数据可包括：硬盘的SMART(Self-Monitoring Analysis and Reporting Technology，自我监测、分析及报告技术)信息和/或硬盘的I/O(Input/Output，输入/输出)信息。其中，所述SMART信息包括以下至少一项：硬盘重映射的扇区个数、硬件不可修复的错误计数、硬件ECC(Error Correcting Code，错误检查和纠正)修复的错误计数、原始读出错率、磁头寻道出错率以及磁头离盘片过高导致写失败的次数；所述I/O信息至少包括：写入块数(blk_written)。例如，硬盘的SMART信息可至少包括以下3个条目的NORMAL值：#5(Reallocated Sector Count，硬盘重映射的扇区个数)、#187(Reported Uncorrectable Errors，硬件不可修复的错误计数)以及#195(Hardware ECC Recovered，硬件ECC修复的错误计数)。又例如，硬盘的SMART信息可以下6个条目的NORMAL值：#1(Raw Read Error Rate，原始读出错率)、#5(Reallocated Sector Count，硬盘重映射的扇区个数)、#7(Seek Error Rate，磁头寻道出错率)、#187(Reported Uncorrectable Errors，硬件不可修复的错误计数)、#189(High Fly Write，磁头离盘片过高导致写失败的次数)以及#195(Hardware ECC Recovered，硬件ECC修复的错误计数)，再加上I/O信息的一个条目， 硬盘的状态数据可共有7个条目。
步骤102：计算在第一预设时间段内采集到的濒临故障的硬盘的各项状态数据中的各条数据的变化量以及离散量，得到第一预测数据集；
上述第一预设时间段的例如7天，在步骤102中，例如，可以仅获取最近7天内采集到的数据中心的各硬盘的各项状态数据。
其中，第一预测数据集中可包括硬盘的多种的状态数据(例如上述7个条目对应的7种状态数据)，每种状态数据分别包括在第一预设时间段内不同的时间采集到的多条数据；第一预测数据集中也可仅包括硬盘的一种状态数据，该种情况下，在步骤102中，可直接计算在第一预设时间段内采集到的濒临故障的硬盘的该种状态数据中各条数据的变化量以及离散量。
上述离散量可以体现硬盘的一条状态数据与总体均数之间的差异，该离散量可以为方差或标准差，变化量可体现硬盘的一条状态数据在第一预设时间段内的变化程度。
步骤103：将所述第一预测数据集输入第一训练模型，得到各硬盘在未来的第二预设时间段内发生故障的概率，其中，所述第一训练模型通过第一人工智能算法基于第一训练数据集进行训练得到，所述第一训练数据集中包括正样本数据以及负样本数据，其中，故障硬盘对应的变化量以及离散量为正样本数据，非故障硬盘对应的变化量以及离散量为负样本数据。
其中，第一预设时间段可大于第二预设时间段，或者二者也可相等。
其中，可将硬盘在发生故障之前的一个时间段之内的状态数据确定为故障硬盘对应的数据；将硬盘在发生故障之前的一个时间段之外的状态数据确定为非故障硬盘对应的数据，或者直接删除该部分数据；将未发生故障的硬盘的状态数据确定为非故障硬盘对应的数据。
其中，上述第一人工智能算法可包括：逻辑回归算法、人工神经网络算法、随机森林算法中的任意一种算法，例如可选择逻辑回归算法对第一训练数据集进行训练，得到第一训练模型。
本公开实施例的预测硬盘故障发生时间的方法，将基于第一预设时间段内采集到的濒临故障的硬盘的各项状态数据计算得到各条状态数据的变化量以及离散量，从而得到第一预测数据集，将该第一预测数据集输入第一训练模块可也测出各硬盘在未来的第二预设时间段内发生故障的概率，从而实现了预测硬盘发生故障的时间的目的。
在一个例子中，在存在濒临故障硬盘的情况时，可触发执行上述步骤102以及步骤103，假设存在濒临故障的硬盘为h块，h≤g。将该h个硬盘在最近7天(为上述第一预设时间段的一个示例)的样本进行合并，得到h个样本，每个样本有7个条目，将各硬盘的最近7天的样本中每个条目分别求方差(为上述离散量的一个示例)和变化量，则每个样本有14 个条目，将计算得到的该方差以及变化量作为第一预测数据集，逐一将该h个样本对应的方差以及变化量输入预先训练得到的第一训练模型，可逐一得到各硬盘在未来7天发生故障的概率。
其中，对h块濒临故障的硬盘，逐一进行故障时间预测的过程可包括：
{对于第1块濒临故障硬盘，将该硬盘对应的方差以及变化量输入第一训练模型，得出其在未来7天发生故障的概率为p_1。
……(以此类推，不再赘述)
对于第h块濒临故障硬盘，将该硬盘对应的方差以及变化量输入第一训练模型，得出其在未来7天发生故障的概率为p_h。
综合上述过程得到的硬盘在未来七天发生故障的概率结果，如果得出某硬盘在未来7天发生故障的概率大于第一阈值，该第一阈值的取值范围可以是50％-90％，例如该第一阈值为80％，则认为该硬盘将在未来7天发生故障，可将该硬盘加入换盘告警列表。运维人员可根据该换盘告警列表进行换盘处理操作。换盘处理操作可逐一进行，例如，按照硬盘故障概率由高至低的顺序进行换盘。在进行换盘处理操作时，可将故障硬盘中的数据全部拷贝至新加入的硬盘，使得系统硬盘的个数保持不变。
在一种可实现方式中，其中，可以使用以下任一公式求上述变化量：
Figure PCTCN2019121647-appb-000001
以及
Figure PCTCN2019121647-appb-000002
Figure PCTCN2019121647-appb-000003
其中，Δ为所述变化量，a n为在所述第一预设时间段内采集到的第n项状态数据，k为在所述第一预设时间段内采集到的状态数据的总数。
由于硬盘是在其使用过程中逐渐发生恶化的，因此，仅将一次采集的一个硬盘的状态数据作为一个样本来预测硬盘故障，会忽略硬盘使用过程中逐渐恶化的因素。故本公开实施例的方法可基于一段时间内(即上述第一预设时间段)采集到的多条样本来预测硬盘发生故障的时间。例如，在硬盘逐渐恶化的过程中，发现硬盘#187是剧烈震荡的，那么可将最近一段时间，例如最近14天(为上述第一预设时间段的一个示例)内硬盘#187的所有样本抽出，合并为一个样本，确定出该样本中数据的变化程度(包括上述变化量以及离散量)，基于该变化程度预测硬盘在未来一段时间，例如未来5天(为上述第二预设时间段的一个示例)发生故障的概率。下结合图2对基于一段时间段采集到的样本对硬盘发生故障的时间进行预测的方法进行说明，图2是根据一示例性实施例示出的一种预测硬盘故障发生时间的方法的流程图，如图2所示，该方法在图1所示的方法的基础上还可包括：
步骤201：在将所述第一预测数据集输入第一训练模型之前，将所述各硬盘在所述第一预设时间段内的状态数据合并为一项数据，计算该项数据内各条数据的离散量以及变化量；合并得到的该项数据中可包括在第一预设时段内的不同时间采集到的数据；
其中，在每次采集硬盘的多种不同状态数据的情况下，在步骤201中，可将在第一预设时间段内采集的各硬盘的状态数据中属于同一种状态数据的状态数据组合为一项状态数据。
步骤202：将所述各硬盘中非故障硬盘的离散量以及变化量标记为负样本数据，将所述各硬盘中故障硬盘的离散量以及变化量标记为正样本数据，得到所述第一训练数据集。
其中，步骤201以及步骤202可在步骤101以及步骤102之前执行，也可在步骤101以及步骤102之后执行，图2中仅示出了步骤201在步骤101之前执行的一种示例情况。
在获得该第一训练数据集后，可基于第一人工智能算法对该数据集进行训练，即可得到上述第一训练模型。
在一种可实现方式中，根据采集到的硬盘的各项状态数据从多个硬盘中筛选出濒临故障的硬盘可包括：将采集到的所述各硬盘的各项状态数据(以下也称第二测试数据集)输入第二训练模型，得到与所述各项状态数据对应的分类结果，所述分类结果的类别包括濒临故障以及健康，其中，所述第二训练模型通过第二人工智能算法基于第二训练数据集进行训练得到，所述第二训练数据集中包括正样本数据以及负样本数据，其中，所述各硬盘中非故障硬盘对应的状态数据为负样本数据，故障硬盘对应的状态数据为正样本数据；将获得的分类结果中类别为濒临故障的分类结果占总结果的比例超过第二阈值的硬盘确定为濒临故障的硬盘，该第二阈值例如为80％。
以下通过一个例子对从多个硬盘中筛选出濒临故障的硬盘的处理过程进行说明。在该例子中，可将上述g块硬盘在3天中每块硬盘产生的多个样本作为第二预测数据集。如果样本个数太少，则难以有效采集到硬盘的恶化状态；如果样本个数太多，则增大了处理量。样本数量可选择20-40条为宜，例如，选择样本数量为24个。
对全体g块硬盘，逐一进行预测：
{对于第一块硬盘，逐条样本进行预测：
{将第一条样本输入第二训练模型，得出分类结果。
将第一块硬盘的第二条样本输入第二训练模块，得出分类结果。
……(以此类推，不再赘述)
对第24条样本，输入第二训练模型，得出分类结果。
综合上述24个分类结果。如果该硬盘具有超过t个分类结果为的类型为濒临故障，则确定预测结果为该硬盘濒临故障。其中，t的取值范围可以为[12，22]，例如t的取值可以为18。如果该硬盘的濒临故障的分类结果不超过t个，则判定该硬盘为健康硬盘。
}
对于第二块硬盘，逐条样本进行预测
……(以此类推，不再赘述)
对于第g块硬盘，逐条样本进行预测。
}
上述例子以投票的方式对濒临故障的硬盘进行判断，可筛选出数据中心的硬盘中濒临故障的硬盘。
将上述过程中被判定为濒临故障的硬盘添加入濒临故障硬盘列表。从而成功预测出了即将出现故障的硬盘。
其中，上述用于对第二训练数据集进行训练的第二人工智能算法例如可以是支持向量机算法、贝叶斯算法、梯度推进机算法中的任意一种算法。
在一种可实现方式中，获得第二训练数据集可包括：以第一预设周期采集所述各硬盘的状态数据，该第一预设周期例如3个小时，得到原始数据集；对所述原始数据集中的各项状态数据进行标记，将硬盘在发生故障前的第三预设时间段之内的状态数据标记为正样本数据，将硬盘在发生故障前的第三预设时间段之外的状态数据标记为负样本数据，或删除硬盘在发生故障前的第三预设时间段之外的状态数据，将属于未发生故障的硬盘的状态数据标记为负样本数据，得到上述第二训练数据集。例如，对原始数据集的所有样本进行标记：对健康硬盘的每条样本标记为负样本；对故障盘在故障前K时间之内的样本标记为正样本；对故障硬盘在故障前K时间之外的样本标记为负样本或丢弃该样本。K的取值例如是7天，最终形成第二训练数据集。
在一个例子中，获得第一训练数据集的过程可包括：对上述原始数据集进行缩减，可仅保留各硬盘最近7天的样本。分别将各硬盘在该7天内的样本进行合并，得到g个样本，每个样本具有7种状态数据，分别各种状态数据中的各条数据求离散量以及变化量。最后对健康硬盘的离散量以及变化量标记为负样本，将故障硬盘的离散量以及变化量标记为正样本，形成第二训练数据集。在一种可实现方式中，本公开的预测硬盘故障发生时间的方法还可包括：使用新采集到的各硬盘的状态数据更新所述原始数据集中的待更新数据，其中，所述待更新数据是所述原始数据集中在最早的第三预设时间段内采集到的所述状态数据。例如，以第二预设周期周期性地使用最近一个所述第二预设周期内采集到的所述状态数据替换所述原始数据集中最早的所述第二预设周期内采集到的所述状态数据，其中，第二预设周期例如7天，故在本实施例中，可使用最近7天采集到的状态数据替换原始数据集中最早的7天所采集到的状态数据，从而对原始数据集进行更新。
以下通过一个例子对本公开的预测硬盘故障发生时间的方法进行说明，在该例子中，数据中心有4800块硬盘。每隔3小时对这4800块硬盘进行一次SMART信息的采集和IO的采集。每块硬盘的每次采集形成1个样本。采集在第60天结束时，共产生2304000条样 本作为原始数据集。假设在数据采集期间，共出现故障盘70块。
将4730块健康硬盘的共计2270400条样本标记为负样本，将70块故障盘在故障前7天的共计3920条样本标记为正样本，其余样本丢弃，得到第二训练数据集。针对第二训练数据集采用支持向量机算法的模型训练，得到第二练模型。
逐一对每一块硬盘最近3天共计24条的样本输入第二训练模型，每个硬盘均会得出24条分类结果。如果一块硬盘具有超过18条分类结果认定濒临故障，则认为该硬盘濒临故障，加入濒临故障硬盘列表，共计23块硬盘。
将4800块硬盘的最近7天的样本合并，每个样本的每个条目求7天内的方差和变化量。共计4800条样本，每条样本有14个条目。将4730块健康硬盘标的状态数据记为负样本，将70块故障硬盘的状态数据标记为正样本，得到第一训练数据集。针对第一训练数据集采用逻辑回归算法的模型训练，得到第一训练模型。
将濒临故障硬盘列表中的23块硬盘的最近7天的样本合并，每个条目求7天内的方差和变化量，得到23条样本。逐一将23条样本输入第一训练模型，逐一得到各硬盘在未来7天发生故障的概率。其中，当硬盘的该概率大于80％者，将硬盘加入换盘告警列表，假设被加入换盘告警列表的硬盘共有4块。则可由运维人员对该4块硬盘执行换盘操作。将最新的7天的采集的各硬盘的状态数据加入原始数据集，替代原始数据集中最早7天检测到的状态数据。
图3是根据一示例性实施例示出的一种预测硬盘故障发生时间的装置的框图，如图3所示，该装置30包括如下组成部分：筛选模块31，用于根据采集到的硬盘的各项状态数据从多个硬盘中筛选出濒临故障的硬盘；计算模块32，用于计算在第一预设时间段内采集到的濒临故障的硬盘的各项状态数据中的各条数据的变化量以及离散量，得到第一预测数据集；输入模块33，用于将所述第一预测数据集输入第一训练模型，得到各硬盘在未来的第二预设时间段内发生故障的概率，其中，所述第一训练模型通过第一人工智能算法基于第一训练数据集进行训练得到，所述第一训练数据集中包括正样本数据以及负样本数据，其中，故障硬盘对应的变化量以及离散量为正样本数据，非故障硬盘对应的变化量以及离散量为负样本数据。
在一种可实现方式中，所述筛选模块可包括：输入单元，用于将采集到的所述各硬盘的各项状态数据输入第二训练模型，得到与所述各项状态数据对应的分类结果，所述分类结果的类别包括濒临故障以及健康，其中，所述第二训练模型通过第二人工智能算法基于第二训练数据集进行训练得到，所述第二训练数据集中包括正样本数据以及负样本数据，其中，所述各硬盘中非故障硬盘对应的状态数据为负样本数据，故障硬盘对应的状态数据为正样本数据；确定单元，用于将获得的分类结果中类别为濒临故障的分类结果占总结果 的比例超过第二阈值的硬盘确定为濒临故障的硬盘。
图4是根据一示例性实施例示出的一种预测硬盘故障发生时间的装置的框图，如图4所示，该装置在图3所示的装置的基础上还可包括：合并模块41，用于在将所述第一预测数据集输入第一训练模型之前，将所述各硬盘在所述第一预设时间段内的状态数据合并为一项数据，计算该项数据内各条数据的离散量以及变化量；第一标记模块42，用于将所述各硬盘中非故障硬盘的离散量以及变化量标记为负样本数据，将所述各硬盘中故障硬盘的离散量以及变化量标记为正样本数据，得到所述第一训练数据集。
在一种可实现方式中，所述状态数据可包括：硬盘的SMART信息和/或硬盘的I/O信息。
在一种可实现方式中，所述SMART信息可包括以下至少一项：硬盘重映射的扇区个数、硬件不可修复的错误计数、硬件ECC修复的错误计数、原始读出错率、磁头寻道出错率以及磁头离盘片过高导致写失败的次数；所述I/O信息至少包括：写入块数。
在一种可实现方式中，所述装置还可包括：更新模块，用于使用新采集到的各硬盘的状态数据更新所述原始数据集中的待更新数据，其中，所述待更新数据是所述原始数据集中在最早的第三预设时间段内采集到的所述状态数据。
在一种可实现方式中，所述变化量通过以下任一公式进行计算：
Figure PCTCN2019121647-appb-000004
Figure PCTCN2019121647-appb-000005
以及
Figure PCTCN2019121647-appb-000006
其中，Δ为所述变化量，a n为在所述第一预设时间段内采集到的第n项状态数据，k为在所述第一预设时间段内采集到的状态数据的总数。
图5是根据一示例性实施例示出的一种预测硬盘故障发生时间的装置600的框图。例如，装置600可以被提供为一服务器。参照图5，装置600包括处理器622，其数量可以为一个或多个，以及存储器632，用于存储可由处理器622执行的计算机程序。存储器632中存储的计算机程序可以包括一个或一个以上的每一个对应于一组指令的模块。此外，处理器622可以被配置为执行该计算机程序，以执行上述的预测硬盘故障发生时间的方法。
另外，装置600还可以包括电源组件626和通信组件650，该电源组件626可以被配置为执行装置600的电源管理，该通信组件650可以被配置为实现装置600的通信，例如，有线或无线通信。此外，该装置600还可以包括输入/输出(I/O)接口658。装置600可以操作基于存储在存储器632的操作系统，例如Windows ServerTM，Mac OS XTM，UnixTM,LinuxTM等等。
在另一示例性实施例中，还提供了一种包括程序指令的非临时性计算机可读存储介质，例如包括程序指令的存储器632，上述程序指令可由装置600的处理器622执行以完成上述的预测硬盘故障发生时间的方法。
本公开能够对硬盘发生故障的时间进行预测。本公开有益效果如下：本公开实施例的预测硬盘故障发生时间的方法，将基于第一预设时间段内采集到的濒临故障的硬盘的各项状态数据计算得到各条状态数据的变化量以及离散量，从而得到第一预测数据集，将该第一预测数据集输入第一训练模块可也测出各硬盘在未来的第二预设时间段内发生故障的概率，从而实现了预测硬盘发生故障的时间的目的。
尽管为示例目的，已经公开了本公开的优选实施例，本领域的技术人员将意识到各种改进、增加和取代也是可能的，因此，本公开的范围应当不限于上述实施例。

Claims (16)

  1. 一种预测硬盘故障发生时间的方法，其中，包括：
    根据采集到的硬盘的各项状态数据从多个硬盘中筛选出濒临故障的硬盘；
    计算在第一预设时间段内采集到的濒临故障的硬盘的各项状态数据中的各条数据的变化量以及离散量，得到第一预测数据集；
    将所述第一预测数据集输入第一训练模型，得到各硬盘在未来的第二预设时间段内发生故障的概率，其中，所述第一训练模型通过第一人工智能算法基于第一训练数据集进行训练得到，所述第一训练数据集中包括正样本数据以及负样本数据，其中，故障硬盘对应的变化量以及离散量为正样本数据，非故障硬盘对应的变化量以及离散量为负样本数据。
  2. 根据权利要求1所述的方法，其中，根据采集到的硬盘的状态数据从多个硬盘中筛选出濒临故障的硬盘，包括：
    将采集到的所述各硬盘的各项状态数据输入第二训练模型，得到与所述各项状态数据对应的分类结果，所述分类结果的类别包括濒临故障以及健康，其中，所述第二训练模型通过第二人工智能算法基于第二训练数据集进行训练得到，所述第二训练数据集中包括正样本数据以及负样本数据，其中，所述各硬盘中非故障硬盘对应的状态数据为负样本数据，故障硬盘对应的状态数据为正样本数据；
    将获得的分类结果中类别为濒临故障的分类结果占总结果的比例超过第二阈值的硬盘确定为濒临故障的硬盘。
  3. 根据权利要求1所述的方法，其中，所述状态数据包括：
    硬盘的SMART信息和/或硬盘的读/写I/O信息。
  4. 根据权利要求3所述的方法，其中，所述SMART信息包括以下至少一项：
    硬盘重映射的扇区个数、硬件不可修复的错误计数、硬件错误检查和纠正ECC修复的错误计数、原始读出错率、磁头寻道出错率以及磁头离盘片过高导致写失败的次数；
    所述I/O信息至少包括：写入块数。
  5. 根据权利要求1所述的方法，其中，所述方法还包括：
    在将所述第一预测数据集输入第一训练模型之前，将所述各硬盘在所述第一预设时间段内的状态数据合并为一项数据，计算该项数据内各条数据的离散量以及变化量；
    将所述各硬盘中非故障硬盘的离散量以及变化量标记为负样本数据，将所述各硬盘中故障硬盘的离散量以及变化量标记为正样本数据，得到所述第一训练数据集。
  6. 根据权利要求1所述的方法，其中，所述方法还包括：
    使用新采集到的各硬盘的状态数据更新所述原始数据集中的待更新数据，其中，所述待更新数据是所述原始数据集中在最早的第三预设时间段内采集到的所述状态数据。
  7. 根据权利要求1至6任一项所述的方法，其中，所述变化量通过以下任一公式进行计算：
    Figure PCTCN2019121647-appb-100001
    以及
    Figure PCTCN2019121647-appb-100002
    Figure PCTCN2019121647-appb-100003
    其中，Δ为所述变化量，a n为在所述第一预设时间段内采集到的第n项状态数据，k为在所述第一预设时间段内采集到的状态数据的总数。
  8. 一种预测硬盘故障发生时间的装置，其中，包括：
    筛选模块，用于根据采集到的硬盘的各项状态数据从多个硬盘中筛选出濒临故障的硬盘；
    计算模块，用于计算在第一预设时间段内采集到的濒临故障的硬盘的各项状态数据中的各条数据的变化量以及离散量，得到第一预测数据集；
    输入模块，用于将所述第一预测数据集输入第一训练模型，得到各硬盘在未来的第二预设时间段内发生故障的概率，其中，所述第一训练模型通过第一人工智能算法基于第一训练数据集进行训练得到，所述第一训练数据集中包括正样本数据以及负样本数据，其中，故障硬盘对应的变化量以及离散量为正样本数据，非故障硬盘对应的变化量以及离散量为负样本数据。
  9. 根据权利要求8所述的装置，其中，所述筛选模块包括：
    输入单元，用于将采集到的所述各硬盘的各项状态数据输入第二训练模型，得到与所述各项状态数据对应的分类结果，所述分类结果的类别包括濒临故障以及健康，其中，所述第二训练模型通过第二人工智能算法基于第二训练数据集进行训练得到，所述第二训练数据集中包括正样本数据以及负样本数据，其中，所述各硬盘中非故障硬盘对应的状态数据为负样本数据，故障硬盘对应的状态数据为正样本数据；
    确定单元，用于将获得的分类结果中类别为濒临故障的分类结果占总结果的比例超过第二阈值的硬盘确定为濒临故障的硬盘。
  10. 根据权利要求8所述的装置，其中，所述状态数据包括：
    硬盘的SMART信息和/或硬盘的读/写I/O信息。
  11. 根据权利要求10所述的装置，其中，所述SMART信息包括以下至少一项：
    硬盘重映射的扇区个数、硬件不可修复的错误计数、硬件错误检查和纠正ECC修复的错误计数、原始读出错率、磁头寻道出错率以及磁头离盘片过高导致写失败的次数；
    所述I/O信息至少包括：写入块数。
  12. 根据权利要求8所述的装置，其中，所述装置还包括：
    合并模块，用于在将所述第一预测数据集输入第一训练模型之前，将所述各硬盘在所述第一预设时间段内的状态数据合并为一项数据，计算该项数据内各条数据的离散量以及 变化量；
    第一标记模块，用于将所述各硬盘中非故障硬盘的离散量以及变化量标记为负样本数据，将所述各硬盘中故障硬盘的离散量以及变化量标记为正样本数据，得到所述第一训练数据集。
  13. 根据权利要求8所述的装置，其中，所述装置还包括：
    更新模块，用于使用新采集到的各硬盘的状态数据更新所述原始数据集中的待更新数据，其中，所述待更新数据是所述原始数据集中在最早的第三预设时间段内采集到的所述状态数据。
  14. 根据权利要求8至13任一项所述的装置，其中，所述变化量通过以下任一公式进行计算：
    Figure PCTCN2019121647-appb-100004
    以及
    Figure PCTCN2019121647-appb-100005
    Figure PCTCN2019121647-appb-100006
    其中，Δ为所述变化量，a n为在所述第一预设时间段内采集到的第n项状态数据，k为在所述第一预设时间段内采集到的状态数据的总数。
  15. 一种预测硬盘故障发生时间的装置，其中，包括：
    处理器；
    用于存储处理器可执行指令的存储器；
    当所述指令被处理器执行时，执行如下操作：
    根据采集到的硬盘的各项状态数据从多个硬盘中筛选出濒临故障的硬盘；
    计算在第一预设时间段内采集到的濒临故障的硬盘的各项状态数据中的各条数据的变化量以及离散量，得到第一预测数据集；
    将所述第一预测数据集输入第一训练模型，得到各硬盘在未来的第二预设时间段内发生故障的概率，其中，所述第一训练模型通过第一人工智能算法基于第一训练数据集进行训练得到，所述第一训练数据集中包括正样本数据以及负样本数据，其中，故障硬盘对应的变化量以及离散量为正样本数据，非故障硬盘对应的变化量以及离散量为负样本数据。
  16. 一种非临时性计算机可读存储介质，当所述存储介质中的指令由处理器执行时，使得处理器能够执行根据权利要求1至7任一项所述的方法。
PCT/CN2019/121647 2018-12-05 2019-11-28 预测硬盘故障发生时间的方法、装置及存储介质 Ceased WO2020114313A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
US17/600,551 US11656943B2 (en) 2018-12-05 2019-11-28 Method and apparatus for predicting hard disk fault occurrence time, and storage medium
EP19892295.7A EP3879405A4 (en) 2018-12-05 2019-11-28 Method and apparatus for predicting hard disk fault occurrence time, and storage medium
JP2021531832A JP7158586B2 (ja) 2018-12-05 2019-11-28 ハードディスク故障発生時期の予測方法、装置及び記憶媒体

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811484874.3A CN109828869B (zh) 2018-12-05 2018-12-05 预测硬盘故障发生时间的方法、装置及存储介质
CN201811484874.3 2018-12-05

Publications (1)

Publication Number Publication Date
WO2020114313A1 true WO2020114313A1 (zh) 2020-06-11

Family

ID=66858675

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/121647 Ceased WO2020114313A1 (zh) 2018-12-05 2019-11-28 预测硬盘故障发生时间的方法、装置及存储介质

Country Status (5)

Country Link
US (1) US11656943B2 (zh)
EP (1) EP3879405A4 (zh)
JP (1) JP7158586B2 (zh)
CN (1) CN109828869B (zh)
WO (1) WO2020114313A1 (zh)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111858283A (zh) * 2020-07-24 2020-10-30 山东海量信息技术研究院 一种边缘数据中心的硬盘故障预处理方法及相关组件
CN112308126A (zh) * 2020-10-27 2021-02-02 深圳前海微众银行股份有限公司 故障识别模型训练方法、故障识别方法、装置及电子设备
CN112737834A (zh) * 2020-12-25 2021-04-30 北京浪潮数据技术有限公司 一种云硬盘故障预测方法、装置、设备及存储介质
CN113688564A (zh) * 2021-07-30 2021-11-23 济南浪潮数据技术有限公司 一种预测ssd硬盘剩余寿命的方法、装置、终端及存储介质
CN113822336A (zh) * 2021-08-20 2021-12-21 济南浪潮数据技术有限公司 一种云硬盘故障预测方法、装置、系统及可读存储介质
CN114327962A (zh) * 2020-09-27 2022-04-12 中国移动通信集团浙江有限公司 一种硬盘故障分级预警的方法和装置

Families Citing this family (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109828869B (zh) * 2018-12-05 2020-12-04 南京中兴软件有限责任公司 预测硬盘故障发生时间的方法、装置及存储介质
CN110413227B (zh) * 2019-06-22 2021-06-11 华中科技大学 一种硬盘设备的剩余使用寿命在线预测方法和系统
CN111008119A (zh) * 2019-12-13 2020-04-14 浪潮电子信息产业股份有限公司 一种硬盘预测模型的更新方法、装置、设备及介质
US10733512B1 (en) * 2019-12-17 2020-08-04 SparkCognition, Inc. Cooperative use of a genetic algorithm and an optimization trainer for autoencoder generation
CN111414286B (zh) * 2020-03-06 2021-11-09 同济大学 一种基于深度学习的不平衡硬盘数据的故障诊断方法
CN111858108B (zh) * 2020-06-23 2024-05-10 新华三技术有限公司 一种硬盘故障预测方法、装置、电子设备和存储介质
CN114595085B (zh) * 2020-12-03 2025-12-30 中兴通讯股份有限公司 磁盘失效预测方法、预测模型训练方法、电子设备
CN114756420A (zh) * 2020-12-29 2022-07-15 超聚变数字技术有限公司 故障预测方法及相关装置
CN113295635A (zh) * 2021-05-27 2021-08-24 河北先河环保科技股份有限公司 一种基于动态更新数据集的水质污染报警方法
US20230008268A1 (en) * 2021-07-07 2023-01-12 Hewlett-Packard Development Company, L.P. Extrapolated usage data
CN115700549A (zh) * 2021-07-23 2023-02-07 伊姆西Ip控股有限责任公司 模型训练方法、故障确定方法、电子设备和程序产品
CN113590406B (zh) * 2021-08-16 2024-12-03 湖南博匠信息科技有限公司 一种基于电变量进行固态硬盘故障检测的方法及系统
CN116738301B (zh) * 2022-03-01 2026-04-17 英业达科技有限公司 硬盘效能问题分类模型的建立方法及系统、分析方法
CN115130517B (zh) * 2022-07-05 2025-08-26 重庆大学 一种基于LightGBM的汽车胎压故障精准定位方法
CN115687038B (zh) * 2022-08-23 2023-09-26 江苏臻云技术有限公司 一种基于大数据的硬盘故障预测系统及方法
CN116595452A (zh) * 2023-05-09 2023-08-15 苏州浪潮智能科技有限公司 服务器故障预测方法、装置、计算机设备及存储介质
CN116701150B (zh) * 2023-06-19 2024-01-16 深圳市银闪科技有限公司 一种基于物联网的存储数据安全监管系统及方法
CN116610484B (zh) * 2023-07-19 2023-11-03 苏州浪潮智能科技有限公司 一种模型训练方法、故障预测方法、系统、设备以及介质
CN117170998B (zh) * 2023-11-03 2024-03-01 凌雄技术(深圳)有限公司 一种智能化设备生命周期管理系统
KR20250089851A (ko) * 2023-12-12 2025-06-19 삼성전자주식회사 전자 장치 및 이의 제어 방법
CN118656273B (zh) * 2024-08-21 2025-02-11 天津理工大学 面向低质量数据集的硬盘故障预测及数据迁移方法
CN119597580B (zh) * 2024-11-25 2025-07-04 鸿秦（北京）科技有限公司 一种硬盘异常实时监测方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104503874A (zh) * 2014-12-29 2015-04-08 南京大学 一种云计算平台的硬盘故障预测方法
US20160232450A1 (en) * 2015-02-05 2016-08-11 Wistron Corporation Storage device lifetime monitoring system and storage device lifetime monitoring method thereof
CN108446734A (zh) * 2018-03-20 2018-08-24 中科边缘智慧信息科技（苏州）有限公司 基于人工智能的磁盘故障自动预测方法
CN108647136A (zh) * 2018-05-10 2018-10-12 南京道熵信息技术有限公司 基于smart信息和深度学习的硬盘损坏预测方法及装置
CN109828869A (zh) * 2018-12-05 2019-05-31 中兴通讯股份有限公司 预测硬盘故障发生时间的方法、装置及存储介质

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2015184823A (ja) * 2014-03-20 2015-10-22 株式会社東芝 モデルパラメータ算出装置、モデルパラメータ算出方法およびコンピュータプログラム
CN105260279B (zh) * 2015-11-04 2019-01-01 四川效率源信息安全技术股份有限公司 基于smart数据动态诊断硬盘故障的方法和装置
CN107025153B (zh) * 2016-01-29 2021-02-12 阿里巴巴集团控股有限公司 磁盘的故障预测方法和装置
US10191668B1 (en) * 2016-06-27 2019-01-29 EMC IP Holding Company LLC Method for dynamically modeling medium error evolution to predict disk failure
US10310749B2 (en) * 2016-09-16 2019-06-04 Netscout Systems Texas, Llc System and method for predicting disk failure
US10216558B1 (en) * 2016-09-30 2019-02-26 EMC IP Holding Company LLC Predicting drive failures
JP2019053474A (ja) * 2017-09-14 2019-04-04 先智雲端數據股▲分▼有限公司ＰｒｏｐｈｅｔＳｔｏｒ Ｄａｔａ Ｓｅｒｖｉｃｅｓ， Ｉｎｃ． クラウドベースサービスのデータ保護方法
CN108073486B (zh) * 2017-12-28 2022-05-10 新华三大数据技术有限公司 一种硬盘故障的预测方法和装置
CN108228377B (zh) * 2017-12-29 2020-07-07 华中科技大学 一种面向磁盘故障检测的smart阈值优化方法
CN109491850A (zh) * 2018-11-21 2019-03-19 北京北信源软件股份有限公司 一种磁盘故障预测方法及装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104503874A (zh) * 2014-12-29 2015-04-08 南京大学 一种云计算平台的硬盘故障预测方法
US20160232450A1 (en) * 2015-02-05 2016-08-11 Wistron Corporation Storage device lifetime monitoring system and storage device lifetime monitoring method thereof
CN108446734A (zh) * 2018-03-20 2018-08-24 中科边缘智慧信息科技（苏州）有限公司 基于人工智能的磁盘故障自动预测方法
CN108647136A (zh) * 2018-05-10 2018-10-12 南京道熵信息技术有限公司 基于smart信息和深度学习的硬盘损坏预测方法及装置
CN109828869A (zh) * 2018-12-05 2019-05-31 中兴通讯股份有限公司 预测硬盘故障发生时间的方法、装置及存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3879405A4 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111858283A (zh) * 2020-07-24 2020-10-30 山东海量信息技术研究院 一种边缘数据中心的硬盘故障预处理方法及相关组件
CN114327962A (zh) * 2020-09-27 2022-04-12 中国移动通信集团浙江有限公司 一种硬盘故障分级预警的方法和装置
CN112308126A (zh) * 2020-10-27 2021-02-02 深圳前海微众银行股份有限公司 故障识别模型训练方法、故障识别方法、装置及电子设备
CN112737834A (zh) * 2020-12-25 2021-04-30 北京浪潮数据技术有限公司 一种云硬盘故障预测方法、装置、设备及存储介质
CN113688564A (zh) * 2021-07-30 2021-11-23 济南浪潮数据技术有限公司 一种预测ssd硬盘剩余寿命的方法、装置、终端及存储介质
CN113688564B (zh) * 2021-07-30 2024-02-27 济南浪潮数据技术有限公司 一种预测ssd硬盘剩余寿命的方法、装置、终端及存储介质
CN113822336A (zh) * 2021-08-20 2021-12-21 济南浪潮数据技术有限公司 一种云硬盘故障预测方法、装置、系统及可读存储介质

Also Published As

Publication number Publication date
EP3879405A1 (en) 2021-09-15
JP2022508320A (ja) 2022-01-19
US20220206898A1 (en) 2022-06-30
CN109828869B (zh) 2020-12-04
JP7158586B2 (ja) 2022-10-21
CN109828869A (zh) 2019-05-31
EP3879405A4 (en) 2022-01-19
US11656943B2 (en) 2023-05-23

Similar Documents

Publication Publication Date Title
WO2020114313A1 (zh) 预测硬盘故障发生时间的方法、装置及存储介质
US8478569B2 (en) Auto adjustment of baseline on configuration change
Zhang et al. Multi-view feature-based {SSD} failure prediction: what, when, and why
US20150205657A1 (en) Predicting failure of a storage device
CN111858108B (zh) 一种硬盘故障预测方法、装置、电子设备和存储介质
CN113778766B (zh) 基于多维特征的硬盘故障预测模型建立方法及其应用
WO2021238258A1 (zh) 一种磁盘故障预测方法和系统
US11734103B2 (en) Behavior-driven die management on solid-state drives
KR20210108874A (ko) 기계 학습을 사용하여 스토리지 장치 장애를 예측하는 시스템 및 장치
JP7273669B2 (ja) ストレージシステム及びその制御方法
CN110119344B (zh) 基于s.m.a.r.t参数的硬盘健康状态分析方法
Ghiasvand et al. Anomaly detection in high performance computers: A vicinity perspective
JP2021022416A (ja) ハードディスクの耐用年数の最適化方法及びそのシステム
CN115794451A (zh) 基于存储设备健康状态的执行策略预测方法、装置及系统
CN110175100A (zh) 一种存储盘故障预测方法及预测系统
US10574552B2 (en) Operation of data network
Tsai et al. A study of soft error consequences in hard disk drives
Yang et al. Zte-predictor: Disk failure prediction system based on lstm
US10776240B2 (en) Non-intrusive performance monitor and service engine
CN114930299A (zh) 一种管理数据存储阵列的方法、装置及存储介质
CN114661566B (zh) 硬盘故障预测方法及装置
Zhou et al. Asldp: An active semi-supervised learning method for disk failure prediction
CN114816878A (zh) 一种硬盘故障预测方法、设备及介质
US8780471B2 (en) Linking errors to particular tapes or particular tape drives
CN118885349B (zh) 目标不可纠正故障预测模型训练方法以及相关设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19892295

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2021531832

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2019892295

Country of ref document: EP

Effective date: 20210610

WWW Wipo information: withdrawn in national office

Ref document number: 2019892295

Country of ref document: EP