JP6610825B2 - Failure prediction apparatus, failure prediction method, and failure prediction program - Google Patents

Description

  The present invention relates to a failure prediction technique for a disk drive.

  A hard disk may have a bad sector due to minute defects on the surface of the disk, making it impossible to read and write, and repeating the retry operation may significantly reduce the data transfer speed. In addition, a failure may occur and the hard disk itself may fail to start up.

  Patent Document 1 discloses a technique for measuring a hard disk transfer time and detecting a hard disk failure sign from the transfer time. When the transfer time is measured every predetermined time (for example, one week) after product shipment and compared with the transfer time at the time of shipment from the factory, if the difference between the two transfer times satisfies the predetermined condition, Warning notification processing is performed to indicate that there may be a problem with the hard disk in the future.

JP 2011-68109 A

  In the prior art disclosed in Patent Document 1, a value obtained by multiplying a measurement value of transfer time by a predetermined coefficient is compared with a measurement time at the time of factory shipment, and when the former is long, it is determined as a delay region. A warning is issued when the delay area is greater than or equal to a predetermined threshold. However, no consideration was given to variations in access time depending on the head position of a hard disk drive (HDD). For example, even when accessing the same area (sector) of the HDD, if the rotation waiting time is the maximum, the waiting time for one rotation of the disk is required, and the access time is increased by that time, and the rotation waiting time is reduced. In the case of the minimum, the access time is shortened because the access can be made in the shortest time without waiting for the disk rotation. As described above, even when the same area of the HDD is accessed, the access time until the data read is completed varies depending on the head position before access and the access timing. Since no consideration was given, the access time could not be measured with sufficient accuracy. Therefore, even if failure prediction is performed based on such access time, failure prediction of the HDD cannot be performed with high accuracy.

  The present invention has been made in view of such a situation, and an object of the present invention is to provide a failure prediction technique capable of predicting the failure occurrence of a disk drive using device and preventing loss of data in the disk. is there.

  In order to solve the above problem, a failure prediction apparatus according to an aspect of the present invention measures an access time of an access target location using a predetermined access pattern for a disk drive, and calculates a threshold related to the access time of the access target location. A threshold calculating unit; and a control unit that detects an abnormality based on the threshold calculated by the threshold calculating unit, wherein the predetermined access pattern is an access target after the head of the disk drive is returned to the initial position. The pattern is to access the location.

  Another aspect of the present invention is a failure prediction method. In this method, a disk drive using device measures an access time of an access target location using a predetermined access pattern for the disk drive, calculates a threshold value related to the access time of the access target location, and the calculated threshold value And a step of detecting an abnormality based on the method, wherein the predetermined access pattern is a pattern for accessing the access target location after returning the head of the disk drive to the initial position.

  It should be noted that any combination of the above-described constituent elements and a conversion of the expression of the present invention between a method, an apparatus, a system, a recording medium, a computer program, etc. are also effective as an aspect of the present invention.

  According to the present invention, it is possible to predict the failure of a device using a disk drive and prevent loss of data in the disk.

It is a block diagram of the HDD failure prediction apparatus which concerns on embodiment. It is a flowchart which shows the failure prediction procedure by the HDD failure prediction apparatus of FIG. It is a flowchart which shows the detailed procedure of the failure prediction process of FIG. It is a flowchart which shows the detailed procedure of the warning determination process of FIG. It is a flowchart which shows the detailed procedure of the error determination process of FIG. It is a figure explaining the access pattern for measuring command execution time. It is a figure explaining the command execution time measured by the access pattern of FIG. It is the graph which showed the schematic diagram of FIG. 7A by the measured value. It is a figure explaining the evaluation data and threshold value data in the segment which grouped the command issue number. FIG. 9A shows the access time when the HDD 300 is normal, and FIG. 9B shows the access time when an abnormality occurs in the HDD 300. FIG. 10A shows an abnormal value database recorded in association with a command issue number and a threshold, and FIG. 10B shows an abnormal value database recorded in association with a segment number and a threshold. is there. FIG. 11A shows a warning counter counted for each command issue number, and FIG. 11B shows an error counter counted for each command issue number.

  When an HDD malfunctions, there is no guarantee of written data. Once a malfunction of the HDD occurs, internal data cannot be read basically, resulting in a large loss. That is, at the stage where it is noticed that there is a failure, it is impossible to avoid losing some data in the HDD. Therefore, the HDD failure prediction apparatus 100 according to the embodiment of the present invention predicts the life of the HDD as a limit at which the operational data can be normally read, and avoids losing data due to the HDD failure. With the goal.

  With regard to HDD life prediction based on comparison of transfer times, Patent Document 1 describes a method for evaluating access time. However, with regard to HDD failures, failures that come from wear of internal parts due to deterioration over time, in particular, tend not to stop at the location where the failure has occurred, but to expand over time, so the signs cannot be captured with high accuracy. As the time elapses, the possibility of losing the internal data of the HDD increases.

  In particular, re-accessing an area that is progressing to a failure is likely to progress to a serious failure, so the ability to predict the progression of the failure will determine how to take subsequent HDD internal data remedies. This is a very important guideline.

  Therefore, in the HDD failure prediction apparatus 100 according to the present embodiment, the accuracy of failure prediction is determined by determining a threshold to be used when predicting failure based on access time based on a scatter diagram of access time according to a specific access pattern. Adopt a method to increase

  FIG. 1 is a configuration diagram of an HDD failure prediction apparatus 100 according to the embodiment. The HDD failure prediction apparatus 100 has a function of driving the HDD 300 used by the host 200 and a function of predicting a failure of the HDD 300. The HDD failure prediction apparatus 100 includes an HDD controller 10, a temporary storage unit 20, a control unit 30, and an abnormal value DB recording unit 40. These configurations can be realized by hardware, software, or a combination thereof. It should be noted that the HDD failure prediction apparatus 100 and the host 200 can be configured integrally. Further, the HDD failure prediction apparatus 100, the host 200, and the HDD 300 may be configured integrally.

  The HDD controller 10 receives a read / write command for the HDD 300 from the host 200 based on the ATA standard of the hard disk drive, writes data to the HDD 300, and reads data from the HDD 300. Temporary storage unit 20 temporarily stores the transfer data in a FIFO structure.

  The control unit 30 measures the command execution time (access time) at the time of writing by the HDD controller 10 and determines the lifetime of the HDD 300.

  The abnormal value DB recording unit 40 stores a threshold value for determining an abnormality in the access time, and registers the command issue number and the delayed access time at which the delayed access occurred when the delayed access occurs as a database. In this embodiment, as will be described later, the HDD 300 is accessed in a predetermined number of sectors (specific capacity units). Corresponding to the first LBA (Logical Block Address) used for each access, 1, 2, 3,. . . Those given serial numbers such as are called command issue numbers. In the following description, LBA can be used instead of the command issue number. However, using the command issue number can reduce the data capacity required for storage and computation.

  The host 200 includes a display unit and an input unit (not shown). The input to the input unit of the host 200 is transmitted to the control unit 30 of the HDD failure prediction apparatus 100 and processed. The control unit 30 of the HDD failure prediction apparatus 100 also functions as a display control unit that controls the display unit of the host 200.

  When the host 200 issues a data write command to the HDD 300, the written data is temporarily stored in the temporary storage unit 20 via the HDD controller 10. When the temporarily stored write data reaches a predetermined capacity, the HDD controller 10 writes the old data on the time axis in accordance with the access pattern shown in FIG. Details of this processing will be described later.

  At this time, the control unit 30 measures the command execution time (access time) when data is written from the HDD controller 10 to the HDD 300, reads the threshold value stored in advance in the abnormal value DB storage unit, and reads the current access time. It is determined whether or not exceeds this threshold.

  When it is confirmed that the access time exceeds the threshold, the control unit 30 stores the LBA and access time whose access time exceeds the threshold, and counts a warning (warning) counter and an error counter, which will be described later. Failure prediction of the HDD 300 is performed based on the transition of the counter and the error counter, and when it is predicted that the life has been reached, an HDD stop warning is issued.

  Here, an outline of the access time of the HDD 300 will be described. FIG. 9A shows the access time when the HDD 300 is normal, and FIG. 9B shows the access time when the HDD 300 is abnormal.

  As shown in FIG. 9A, the normal access time can be represented by the sum of the data transfer time and the time depending on the head operation such as seek time, rotation waiting time, and focusing time.

  As shown in FIG. 9B, the access time at the time of occurrence of an abnormality is several tens of times that at the time of normal access because, in addition to the access time at the normal time, a retry time, an alternative sector processing time, etc. are added when an alternative sector occurs. Processing time is required.

  Since the disk rotates in the HDD 300 at a constant speed, when the head accesses the target sector, a rotation waiting time corresponding to one rotation of the disk occurs at maximum depending on the access timing. In addition, since the seek time from the next access position varies depending on the head position when the previous access is completed, the total access time also varies. As a result, since the convergence time until access is different, the target sector cannot be accessed in the same access time as at the time of factory shipment, and varies for each access.

  The failure determination method of Patent Document 1 has a problem that HDD failure cannot be predicted with high accuracy because both the data at the time of factory shipment and the data at the time of life evaluation vary. Therefore, in the present embodiment, variations in access time are dealt with by using the access pattern shown in FIG.

  FIG. 2 is a flowchart showing a failure prediction procedure performed by the HDD failure prediction apparatus 100.

  In step S201, the HDD controller 10 accesses the HDD 300 according to the access pattern of FIG. 6, and the control unit 30 measures the command execution time at that time.

  In the access pattern shown in FIG. 6, specific capacity data is written corresponding to one command issue number. In this embodiment, the specific capacity is set to 256 sectors. However, the present invention is not limited to this. In the case of the HDD 300, if the head position at the time when the previous access is ended is unspecified, the seek time varies particularly, and the relatively accurate access time cannot be measured. Therefore, as shown in FIG. 6, after the end of the access, the head position is always reset to the initial position (sector 0 here) and then the specific capacity (256 sectors) is sequentially written, thereby providing a more accurate access time. Measurement is possible.

  FIG. 7A is a schematic diagram of access time when the HDD 300 is accessed according to the access pattern of FIG. The horizontal axis represents the command issue number, and the vertical axis represents the access time for each command issue number. FIG. 7B is a graph showing the schematic diagram of FIG. 7A with actually measured values. When the measurement data is plotted, a scatter diagram as shown in FIG. 7B is obtained, and the access time is distributed in the shape of a band rising upward. Hereinafter, a disk area (sector) having a specific capacity accessed for each command issue number is referred to as a “command issue number area”.

  If the HDD 300 is in an ideal state without waiting for rotation and focusing time, the scatter diagram plotting the access time for the sector to be the access target should be almost one line. However, as described above, since there is always a variation in the waiting time for rotation and the convergence time for disk access, plotting the access time for each specific capacity shown in the access pattern of FIG. 6 actually specifies as shown in FIG. 7B. There is a strong correlation between the command issue number and the access time. The inclination of the entire band in FIG. 7B is mainly due to an increase in seek time due to the fact that the sector to be accessed becomes far from the reset position of the head as the command issue number increases as shown in FIG.

  Note that the command execution time (access time) may be measured by the HDD failure prediction apparatus 100 or by an external device that generates the same access pattern.

  In step S202, a value corresponding to the upper end of the band in the scatter diagram measured in step S201 is detected. This value is used as a threshold value (failure prediction threshold value) for each command issue number. This value is the access time when the waiting time depending on the head operation such as seek time, rotation waiting time, and focusing time is maximized. When the HDD 300 is normal, the access time falls below this value. There is a feature. Therefore, by using this value as a failure prediction threshold, it is possible to grasp that the access of the HDD 300 is not normal when the access time exceeds this threshold. A specific method for detecting this threshold will be described later.

  In step S203, the threshold value calculated in step S202 is registered in the abnormal value DB recording unit 40. Specifically, as shown in FIG. 10A, a command issue number and a threshold value are recorded in association with each other.

  Steps S201 to S203 are pre-processing or initial setting processing for performing failure prediction.

  In step S204, a failure prediction operation is performed when the HDD 300 is used. The control unit 30 monitors whether or not the access time of the command issue area exceeds the threshold value recorded in the abnormal value DB recording unit 40, and for the command issue number area exceeding the threshold value, the command issue number is stored in the abnormal value DB recording unit 40. And record the access time. The control unit 30 counts the warning counter and the error counter recorded in the abnormal value DB recording unit 40, and as a result, when it reaches a level determined to be a failure prediction, it issues an HDD 300 stop warning and ends the processing. To do.

  In the processing of steps S201 to S203, the control unit 30 operates as a threshold value calculation unit. The threshold calculation unit may be a circuit different from the control unit 30.

  If the HDD model number and firmware are the same, the threshold value obtained from the band of the scatter diagram in FIG. 7 is the same, so there is no need to create a new threshold value, so steps S201 to S203 are omitted. The failure prediction in step S204 can be started using the threshold value measured by another HDD. When using a threshold value measured with another HDD, the HDD failure prediction apparatus 100 does not need to include a threshold value calculation unit.

  Here, the method for detecting the threshold value in step S202 will be described in detail. The first method in step S202 will be described.

  As shown in FIG. 8, the command issue numbers are grouped for every predetermined number N of command issue numbers. Hereinafter, this group is referred to as “command issue segment” or simply “segment”. The predetermined number N is referred to as “segment length”. Here, the predetermined number N (segment length) is set so that one segment includes approximately one or more points at which the waiting time depending on the head operation such as seek time, rotation waiting time, and focusing time is maximized. To do. Typically, N = 30-50 is good. For example, when N = 30, command issue number = 1-30 is segment 1, command issue number = 31-60 is segment 2, command issue number = 61-90 is segment 3, and so on. And the segment.

  Next, the maximum value of the access time is detected for each segment. The maximum value is set as a threshold value in the segment. For example, when N = 30 and the command issue number = 12, the access time is maximum and the maximum value is 30 msec in the segment 1, all the threshold values corresponding to the command issue numbers 1 to 30 are set to 30 msec. . Alternatively, a value obtained by multiplying the maximum value of the access time in each segment by a predetermined magnification may be used as the threshold of the segment. For example, the predetermined magnification = 1.2 and the maximum value 30 msec × 1.2 = 36 msec may be set as the threshold value of the segment. Or it is good also considering the value which added predetermined value to the maximum value of the access time in each segment as a threshold value. For example, the predetermined value = 5 msec and the maximum value 30 msec + 5 msec = 35 msec may be set as the threshold value.

  The threshold values calculated by the first method are all the same for the command issue numbers corresponding to one segment. Therefore, in step S203, the segment number may be recorded in correspondence with the threshold as shown in FIG. 10B without recording the threshold for each command issue number.

  The second method of step S202 will be described. First, as in the first method, a segment is formed for each predetermined number of command issue numbers. This segment is for use in subsequent processing steps, and in step S202, no segment is used.

  Next, for a certain command issue number (command issue number i), the maximum value of the access time is detected for command issue numbers (i-w to i + w) in a predetermined range before and after the command issue number. That is, the threshold value θ [i] corresponding to the command issue number i is calculated according to Equation (1). Here, a [i] is the command execution time (access time) corresponding to the command issue number i, w is a positive integer, and max is a function that returns the maximum value from the values specified in the arguments. is there. According to Equation (1), the maximum value is detected for (2w + 1) command issue numbers. The positive integer w may be set so that one or more maximum values of access time are included in (2w + 1) command issue numbers. Typically, w = 15-25 may be used. (2w + 1) may be the same as or different from the segment length N. Assuming that the maximum command issue number used for the measurement in step S201 is P, threshold values θ [w + 1] to θ [Pw] corresponding to i = (w + 1) to (P−w) are respectively calculated according to mathematical formulas. . For i = 1 to w, θ [w + 1] may be used, and for i = (P−w + 1) to P, θ [Pw] may be used.

  In addition, the threshold may be calculated by further performing a moving average process on the value calculated according to Equation (1). For example, the left side of Expression (1) is substituted into the temporary variable μ [i], and the moving average of μ [i] is calculated according to Expression (2) as the threshold θ [i]. Here, ε [j] is a weighting coefficient that satisfies Equation (3). L is a positive integer and is typically set to 5-10. By calculating the threshold value according to the mathematical formula (2), the change in the threshold value becomes smooth and failure prediction may be possible with high accuracy.

  FIG. 3 is a flowchart showing a detailed procedure of the failure prediction process shown in step S204.

  The failure of the HDD 300 indicates how long the access time of the area delimited by the specific capacity unit exceeds the threshold which is the upper end of the band of the scatter diagram shown in FIG. 7, and how the occurrence of the access delay determines the area. Predict based on whether it spreads over again.

  The cause of the access delay exceeding the threshold value of the HDD 300 is that an abnormality has occurred at the time of access, and in addition to the processing at the time of normal access, the processing time at the time of occurrence of an abnormality such as retry or occurrence of alternative sector is added.

  However, the occurrence of a replacement sector due to damage to a specific sector causes a large access delay only when the replacement sector occurs. However, even if the same sector is accessed thereafter, the HDD 300 is treated as having returned to normal operation. However, even if the same area is accessed again, the access delay does not occur. In this case, the occurrence of the abnormal sector is not confirmed, and the expansion of the abnormal sector cannot be confirmed, and the expansion beyond the alternative sector generated at that time is not seen. On the other hand, the access delay due to deterioration over time gradually increases as the deterioration progresses, and the number exceeding the threshold of the area delimited by the segment length increases for each access, but the pattern cannot be specified, so it is delimited by segment. A failure is predicted by registering in the abnormal value DB recording unit 40 the transition of how the threshold of the specified area is exceeded.

  In step S301 in FIG. 3, it is identified which command issue number corresponds to the top data transferred from the temporary storage unit 20 to the HDD 300. For example, in the temporary storage unit 20, the data to be transferred is managed using the same LBA that is normally written from the host 200 to the HDD 300, and a value obtained by dividing the head LBA by a specific capacity at the time of transfer is calculated and the command issue number and do it. This identified command issue number (first command issue number used for data transfer) is set to i.

  In step S305, it is calculated how many command issue numbers (specific capacities) the data capacity (write capacity) transferred from the temporary storage unit 20 to the HDD 300 corresponds to. Specifically, the data capacity is divided by the specific capacity, and the quotient and remainder are calculated. Then, the quotient is set as the write count M.

  Writing to the HDD 300 can be controlled in units of sectors, but the specific capacity is more than that (here, 256 sectors), so if the last write data is less than the specific capacity, the existing data is in the specific capacity to be written last. May exist.

Therefore, in step S310, it is determined whether or not the remainder calculated in step S305 is “0”. That is, it is determined whether the write capacity is divisible by the specific capacity. As a result, when it is not divisible (when there is a remainder) (NO in S310), the process proceeds to step S315. In step S315, it is confirmed whether or not there is other data in the M + 1-th write area corresponding to the fractional data. If other data exists (YES in S315), the existing data is temporarily stored in step S320. After reading into the storage unit 20 and combining with the (M + 1) th write data, the data is prepared as the (M + 1) th write data, and the process proceeds to step S325. If there is no other data in step S315 (NO in S315), it is not necessary to combine the write data, and the process directly proceeds to step S325.
As a result, the number of times of writing increases by one per fraction, so M is incremented by 1 in step S325, and the process proceeds to step S330.

  If the write capacity is divisible by the specific capacity in step S310 (YES in S310), the process proceeds to step S330.

  In step S330, the specific capacity corresponding to the command issue number i in the access pattern shown in FIG. 6 is written.

  In step S350, a warning determination process for detecting a non-fatal failure is executed. The warning determination process will be described in detail later with reference to the flowchart of FIG.

  In step S355, an error determination process for detecting a fatal failure is executed. In this error determination process, a fatal abnormal sector that is not expected to be recovered is monitored to detect the arrival of the end of life, and when the end of the life is detected, the HDD 300 is finally alerted to stop operation. The user is prompted to save data from the HDD 300. The error determination process will be described later in detail with reference to the flowchart of FIG.

  In step S360, the warning counter value and error counter value of the HDD 300 updated in the processes in steps S350 and S355 are displayed. The user (operator) can monitor the status of the HDD 300 by using these counter values, and the operator can judge from this numerical value and stop the HDD 300 independently when necessary.

  In step S365, when it is confirmed that the error counter processed in response to the result of step S360 has exceeded the operation stop parameter value (YES in S365), the failure prediction process ends. Note that immediately after YES in step S365, a warning message that further alerts the user may be displayed, or a message notifying that the failure prediction process is to be terminated may be displayed.

  While data is being written to the HDD 300, the processing of step S350 and step S355 is performed on the command issue number to predict failure of the HDD 300 being used. For this reason, 1 is subtracted from the write count M in step S370, and 1 is added to the command issue area i because the access moves to the next area.

  Finally, in step S375, if the number of times of writing M is greater than 0 (YES in S375), a series of processes from step S330 to step S370 are repeated until the predetermined number of times of writing is reached. When the number M of times of writing becomes 0 (NO in S375), the failure prediction process is terminated.

  FIG. 4 is a flowchart showing a detailed procedure of the warning determination process in step S350. In the warning determination process, an area where an abnormality has occurred is identified, and it is determined how the abnormality occurrence area is distributed (spread).

  The warning determination is performed by adding the value of the warning counter for each command issue number shown in FIG.

  In step S401, it is determined whether or not the command execution time a [i] corresponding to the command issue number i has exceeded the threshold value θ [i].

  As a feature with respect to the access time of the HDD 300, the HDD 300 in which failure and deterioration have not progressed finishes within the normal processing time, so the segment access time falls within the threshold. However, the HDD 300 that has deteriorated over time and the HDD 300 in which a failure has occurred have increased internal failures due to head contamination, the effect of accumulated dust, scratches on the board surface, and the like. Since time requires processing corresponding to a failure and time is required for internal processing compared to normal processing, the access time exceeds a predetermined threshold value, and the delay of the access time for the write address increases. In step S401, it is confirmed whether or not such a symptom has occurred.

  When the command execution time exceeds the threshold (YES in S401), in step S402, the warning counter is incremented by 1 and the command issue number and the command execution time (access time) at that time are registered in the abnormal value DB recording unit 40. To do.

  If the command execution time does not exceed the threshold (NO in S401), the process proceeds to step S403. In step S403, when past warnings corresponding to the command issue number are counted (YES in S403), it is considered that there is a temporary factor at the time of access, and in step S407, the warning count in the command issue number area is set to 0. The warning determination process is terminated.

  Next, the process proceeds from step S402 to step S404. In step S404, in order to reliably detect the spread of the fault, in the narrow neighboring segment area (first neighboring segment area) in which five segments are taken before the segment to be accessed and five segments are taken after the segment, Check if the scatter plot band is spreading beyond the threshold. Specifically, it is determined whether 10 or more non-zero warning counters have occurred in the first neighboring segment area. When ten or more warning counters are confirmed (YES in step S404), it is determined that the delay is increased in this neighboring segment area, and a warning of stopping the HDD 300 can be issued in the next error determination process. In step S405, a failure prediction threshold (here, 10) is added to the error counter.

  Scratches caused by contact with the head that physically occur on the disk show sector continuity in the access delay, and the cause of the occurrence can be identified relatively easily. However, in the life prediction, the access delay occurs once. Since there is a characteristic that an access delay occurs in the neighboring sectors centering on some sectors where the access time does not recover thereafter, the distribution of the access delay for each sector is recorded, and how this is illustrated. It is necessary to determine whether or not the number of sectors in which the threshold is exceeded and access delay occurs in the scatter diagram band of 7 is increased. Regarding this determination, when attention is paid only to the access delay in the scatter diagram of FIG. 7, a certain problem clearly occurs in the sector that exceeds the alternative sector generation prediction threshold as in S501 of FIG. However, if an access delay occurs in an area that is less than or equal to the alternative sector occurrence prediction threshold and exceeds the threshold shown in the scatter diagram of FIG. It is difficult to determine whether or not an error has occurred in that sector.

  Therefore, in addition to the fact that the access time of the command issue number i exceeds the threshold, it is determined whether the same behavior is seen in the sector near the command issue number i, and the progress of the failure reaching the lifetime is predicted. This is because the access delay of each sector is not so great, but many access delays occur in the area near the command issue number i. It means that some kind of obstacle that extends the access time is in progress. In step S404, an area delimited by the same number of sectors is used as a neighboring segment area so that accurate determination can be made using the number of sectors (command issue numbers) in which the access delay occurs in the neighboring area. In Step S404, the predetermined number such as five segments before, five segments after, and 10 or more warning counters is an example, and a predetermined number other than the above-described values may be used.

  In step S406, when N command issue numbers in the segment to be accessed are viewed, if there is a predetermined number (for example, 2) of non-zero warning counters, a failure that cannot be recovered in the segment has occurred. In step S408, the error count is incremented by 1, and the error counter value is passed to the error determination process. At this time, since the area where the error has occurred can be specified, it is possible to extend the life of the entire HDD by taking measures such as avoiding the use of the problem area.

  As described above, in the warning determination process, it is possible to determine that a failure that is not counted as an error is accumulated in the HDD 300 by adding the warning counter. Further, when a predetermined number or more of warning counters are generated in a certain segment area, it is possible to immediately connect to an error determination process by adding an error counter.

  FIG. 5 is a flowchart showing a detailed procedure of the error determination process in step S355. In the error determination process, it is determined how much access time is actually required for access to the command issue number area whose access time exceeds the threshold in the warning determination process of FIG. 4, and from the accumulated error count so far, Identify what abnormalities you have.

  The error determination process is performed by adding the error counter value for each command issue number shown in FIG.

  In step S501, it is determined whether the access time exceeds the alternative sector occurrence prediction threshold. If the command execution time exceeds the normal access time shown in FIG. 9A and is significantly delayed due to the abnormal processing time determined for each type of HDD 300 in advance, recovery processing is performed. There is a high possibility that it is in the alternative sector processing. The processing time when an abnormality occurs in the HDD 300 required for such alternative processing or the like is determined for each HDD 300. When it can be determined that such alternative sector processing has been entered (YES in S501), since it is difficult to read data, even if the warning counter is registered in the warning determination processing of FIG. In S502, an error counter is added.

  As described above, when the access time exceeds the alternative sector occurrence prediction threshold, when the number of occurrences is small and the single sector is generated, the access time at the subsequent re-access is restored if the access is shifted to the alternative sector. However, since it is clear that there is a problem that obstructs access clearly at the location where the error occurred, even if the warning counter is reset in step S407 in FIG. The counter is not initialized by accumulating as a counter.

  Next, in step S503, when five or more non-zero error counters are confirmed in the second neighboring segment area in which 10 segments are taken before the access target segment and 10 segments are taken later (YES in S503). Then, it is determined that the access delay is extended in the second neighboring segment area, and the process proceeds to step S505 to issue an HDD stop warning. This indicates that failures are concentrated in the second neighboring segment area. If errors occur continuously in adjacent segments where LBA continues, a problem of physical write failure occurs in this area due to scratches on the disk surface on the same circumference due to rubbing of the head or defective heads. This is because there is a high possibility.

    In the case of BR> A, it is highly possible that the physical failure stays in the command issue number area that is currently being evaluated, and has spread to the command issue number area that has not yet been evaluated. Since it may develop into an error and data cannot be read out, it is necessary to promptly issue an HDD stop warning and stop using the HDD.

  The number of adjacent second neighboring segment areas that are 10 before the access target segment and 10 after that, and the number of error counters within the second neighboring segment area are 5 or more are examples, and other than this. The value of may be used. It is desirable to set an optimum value for each HDD 300 so that the smallest scratch can be found early on the innermost circumference of the disk with the fewest consecutive sectors in order to predict the scratches of one maximum disk at an early stage. .

  In step S503, when the number of non-zero error counters in the second neighboring segment area is less than 5 (NO in S503), the process proceeds to step S504. If the number of error counters for all segments exceeds a predetermined value (for example, 10) in step S504 (YES in S504), an HDD stop warning is issued in step S505.

  As described above, according to the failure prediction procedure performed by the HDD failure prediction apparatus 100 according to the present embodiment, failure prediction of the HDD 300 can be performed with high accuracy as follows, and loss of data in the HDD 300 can be prevented.

  If abnormal writing due to particles being added under the head or accidental writing failure when the head could not completely write data to the sector, the access time temporarily exceeds the threshold, but again, By overwriting the same command issue number area, the access time is recovered, and thereafter the operation is performed with a normal access time. In such a case, it can be determined that the warning counter is reset and the HDD 300 has returned to normal operation.

  In addition, when the function is restored by the generation of a substitute sector, if it is a single shot, the accumulation of error counts will not proceed since the function of the HDD 300 is normalized, but a substitute sector is continuously generated in the neighboring segment. It is considered that a writing error due to a scratch or a defective head has occurred. This is because error counter generation progresses in the sector before and after the problem sector, or when the access time exceeds the alternative sector occurrence prediction threshold, the error count is added assuming that the alternative sector is generated It can be determined that the failure of the HDD 300 is near.

  Further, a failure due to the life of the HDD 300 can be determined by checking whether or not the range in which the access time in which the warning counter is registered exceeds the threshold value is generated before and after the specific segment. In this case, it can be judged from the fact that the scatter diagram of FIG.

  In particular, if a predetermined number (for example, two) or more of the warning counter occurs in the re-access of the command issue number area in which the warning counter is registered, recovery due to write failure is not expected, and there is a possibility of data writing abnormality in this command issue number area. Therefore, it is possible to more accurately determine that the occurrence of the failure is near by accelerating the progress from the warning counter to the error counter.

  In this way, by accumulating the warning counter and shifting to the error counter, in addition to the general failure determination of the HDD 300, it is possible to more accurately determine the failure that progresses over time.

  Finally, the error counter is used to determine from the accumulated warning counter value that a threshold value for failure prediction has been exceeded and to notify the HDD 300 of a stop warning by an indicator lamp or the like. This may be a warning by a buzzer or the like, and may be linked with a stop function of the present system. Also, in cooperation with external systems, display information related to the purchase of storage devices such as HDDs (advertising information, etc.) according to warning counter and error counter values, and use of backup services using the cloud etc. A preferential service (such as the presentation of a coupon) may be implemented to promote or promote the purchase of a storage device or the use of a backup service. In this way, by providing pinpoint appropriate information to highly necessary users, it is possible to improve user convenience and increase sales of related products and services.

  Thus, by deriving a threshold value from the maximum access time during normal operation of the HDD 300, the range of access time during normal operation of the HDD 300 can be known, so abnormal operation can be accurately performed in the command issue number area where the access time exceeds the threshold value. The failure prediction of the HDD 300 can be performed with high accuracy.

  The present invention has been described based on the embodiments. The embodiments are exemplifications, and it will be understood by those skilled in the art that various modifications can be made to combinations of the respective constituent elements and processing processes, and such modifications are within the scope of the present invention. .

  In the above description, the failure prediction is performed using the warning counter that may be reset and the error counter that is not reset. However, the failure depends on whether the warning counter exceeds a predetermined number using only the warning counter. May be determined.

  In the above description, the failure prediction technique has been described by taking a hard disk as an example of the disk. However, the failure prediction technique of the present embodiment can be applied to any magnetic disk or optical disk.

10 HDD controller, 20 temporary storage unit, 30 control unit, 40 abnormal value D
B recording unit, 100 HDD failure prediction device, 200 host, 300 HDD.

Claims (6)

A threshold calculation unit that measures an access time of an access target location using a predetermined access pattern to the disk drive, and calculates a threshold value related to the access time of the access target location;
A control unit that detects an abnormality based on the threshold value calculated by the threshold value calculation unit;
With
The failure prediction apparatus according to claim 1, wherein the predetermined access pattern is a pattern for accessing an access target location after returning the head of the disk drive to an initial position.   The threshold value calculation unit measures an access time when a plurality of access target locations are accessed, and uses the access times of other access target locations located in the vicinity of the first access target location. The failure prediction apparatus according to claim 1, wherein the threshold value corresponding to the access target location is calculated.   The threshold calculation unit groups a plurality of adjacent access target locations, and calculates the threshold corresponding to the first access target location based on a maximum value of access time for each group. Item 3. The failure prediction apparatus according to Item 2.   The threshold calculation unit calculates a moving average value related to the maximum access time for each group, and calculates the threshold corresponding to the first access target location based on the moving average value. The failure prediction apparatus according to claim 3. The device using the disk drive
Measuring an access time of an access target location using a predetermined access pattern for the disk drive, and calculating a threshold value related to the access time of the access target location;
Detecting an abnormality based on the calculated threshold;
A failure prediction method for executing
The failure prediction method according to claim 1, wherein the predetermined access pattern is a pattern for accessing an access target portion after returning the head of the disk drive to an initial position. Measuring the access time of the access target location using a predetermined access pattern for the disk drive, and calculating a threshold value for the access time of the access target location;
Detecting an abnormality based on the calculated threshold;
A failure prediction program for causing a computer to execute

The failure prediction program according to claim 1, wherein the predetermined access pattern is a pattern for accessing an access target location after returning the head of the disk drive to an initial position.

Images