JP2006120042A - Disk array device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve the efficiency of automatic restoration for a read error of a disk array of an RAID 1. <P>SOLUTION: A block being, for example, a read and write unit of hard disk itself is used as a unit of automatic restoration in the RAID 1. When an error occurs in reading from one hard disk A of the disk array (S14), an RAID 1 control part reads data of a block of a read target from the other hard disk B (S16) and writes the read data to the hard disk A again (S20). When a read error is detected here, a block reallocation mechanism of the hard disk A itself reallocates a block in a spare area instead of the block where the read error has occurs, the rewrite of S20 is consequently performed for the reallocated block. Using the block as the unit of automatic restoration can greatly reduce the rate of error occurrences at the error restoration in comparison with in the conventional practice in which the entire disk or partition is used as the unit. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a disk array, and more particularly to recovery control of a RAID 1 disk array.

  A disk array device realizes high-speed data read / write and / or improved fault tolerance by connecting a plurality of hard disks in parallel and operating them as one disk device. As control levels of such disk array devices, RAID (redundant array of independent disks) 0 to 5 and combinations thereof are known.

Among these, RAID 1 uses two hard disks and records the same data on both disks, thereby recording data much higher than the reliability of the hard disk for the host system using the data recording mechanism. A mechanism is provided. Assuming that the failure rate of one hard disk is λ, the failure rate of RAID 1 is 2λ ^{2 because} both hard disks are likely to fail simultaneously. This is because the probability that one of the hard disks will fail is 2λ, the probability that the other will fail if one fails, and the probability that both will fail simultaneously is the product of them. For example, when λ is 1% (0.01), the failure rate of RAID1 is 0.02%.

  However, in order for RAID 1 to obtain this reliability, it is necessary to always maintain a redundant state in which the two units normally operate. On the other hand, when one of the hard disks has failed and stopped (this state is referred to as a degenerate state), the failure rate of RAID 1 is reduced to λ, which is one hard disk in terms of reliability. There is no change. Therefore, in order to maintain the high reliability of RAID1, even when an error occurs in one of the hard disks, it is necessary to restore the redundancy state as soon as possible.

  The hard disk failure is detected both at the time of writing and at the time of reading. There are two types of errors: (a) those caused by failures related to the entire hard disk, and (b) those caused by failures in the recording medium. If an error belonging to (a) occurs, this failure cannot be recovered both during writing and reading, so the hard disk in which the error was detected is disconnected. In the case of RAID 1, this results in a degenerate state, but this is inevitable because it is a failure of the entire disk. In this case, there is no other way than restoring the redundancy by replacing the hard disk in which the error has occurred with a new hard disk and copying the data of the normal disk.

  On the other hand, if the error belonging to (b) is an error at the time of writing, in the case of a hard disk to be verified, it is automatically repaired by the reassignment mechanism of the hard disk itself. However, when the verify process is not performed as in an IDE (Integrated Drive Electronics) hard disk drive, the error is detected at the time of reading because the repair is not performed at the time of writing.

  In the case of an error during reading belonging to (b) above, even if an error occurs in reading from one hard disk, reading from the other hard disk can satisfy the read request from the host system, but the hard disk itself Something needs to be done to get an error. Therefore, in the conventional RAID 1 disk array control, when such a partial failure occurs when one of the hard disks is read, the entire data of the other hard disk is copied to the hard disk that detected the error, Redundancy was restored. Some RAID 1 disk array systems based on software RAID are designed to restore redundancy by copying data from a normal hard disk in units of partitions in which an error has been detected, instead of the entire hard disk. Yes.

  Patent Document 1 discloses a recovery copy between disks in a redundant hard disk system. In the method shown in this document, when data is copied from a normal disk to a disk that has been repaired at the time of recovery, even if an error occurs in reading from the copy source, it is forced to continue at least, It is possible to attempt to continue the operation of the entire system by performing recovery of data in which no error occurs.

  Further, Patent Document 2 discloses a control method at the time of data reconstruction (rebuild) after replacing a failed disk in a disk array device. In this method, when a read error occurs during data reconstruction, if the occurrence location is an unused area, the error is ignored, thereby reducing the substantial error occurrence rate in the recovery process.

  Japanese Patent Application Laid-Open No. 2004-228561 discloses a method for supporting recovery from secondary failure in data recovery processing after disk replacement of a disk array. This method provides assistance to a failure recovery engineer when recovering from a situation where a read error has occurred in recovery processing.

  Further, in Patent Document 4, when a failure occurs during data reading, data is read from the other normal hard disk and rewritten to the failed hard disk to restore redundancy, and at the time of rewriting A method of writing in a new area while avoiding an area where an error has occurred in the past is disclosed. In this document, for recovery of a disk in which an error has occurred, in the case of RAID 3 to 5, a block (sector) area corresponding to a block (sector) area in which an error has occurred among other disks or parity disks It is disclosed that the data in the disk is read and rewritten to the disk in which the error occurred. For RAID1, “data read at this time is the same data read from another disk device 14 in which exactly the same data as the disk device 14 in which an error has occurred is recorded”. .

JP 05-242593 A Japanese Patent Laid-Open No. 08-185274 JP 09-305326 A Japanese Patent Laid-Open No. 2001-1000094

  The method of Patent Document 1 simply ignores the error and tries to continue the operation of the entire system, and normal data is written to the recovery target disk in the portion where the read error occurs in the normal disk. Recovery ends without being completed. In this case, when the portion where the read error has occurred is read after recovery, if the portion is read from the recovered disk, the read is performed from the portion where there is no correct data, so that illegal data is read and returned to the host system. The result is quite unpredictable. This is a very dangerous method.

  The method of Patent Document 2 is to reduce the error rate depending on the low usage rate of the hard disk, but the rate of occurrence is reduced to a fraction of at most by simply ignoring the unused portion error. Just do.

  Since the method of Patent Document 3 does not attempt automatic recovery, it is impossible to avoid the degeneration operation, and how much the degeneration operation period can be reduced by this method is not shown and is unknown.

  The method of Patent Document 4 can be applied to a RAID method that performs recording unit control of a small size, such as RAIDs 3 to 5. However, in the method of managing data recovery only with a large size such as the entire disk or partition such as RAID 1, even if a new area avoiding the area where an error has occurred in the past is secured, the entire disk or Providing an area of the same size as the partition is not cost effective, so it is virtually impossible to employ.

  Further, in the case of processing for reading data of a partition corresponding to an entire disk or a partition having an error from a normal hard disk and copying the data to a disk having an error as in the conventional RAID 1 data recovery, a read error occurs. Both disks become errors and recovery by automatic processing becomes difficult. In reading in units of files, the possibility that the hard disk will cause a read error is low, but in the case of reading a large size such as the whole disk or a partition, the possibility that a read error will occur cannot be ignored.

This point will be described in detail. For example, if the unrecoverable error rate when reading one bit of the hard disk is μ and the unit of recovery is S [bit], the probability that an error will occur in this recovery process is S · μ. It becomes. For example, if μ is 10 ⁻¹³ and the recovery unit S is 100 Gbytes (= 100 × 10 ⁹ × 8 [bit]) for the entire hard disk, the error rate when reading all normal hard disks is 0.08, that is, 8%. Probability that can never be ignored.

  In recent years, the capacity of hard disks has been steadily increasing, and hundreds of GBytes have become commonplace. Along with this, the size of hard disk partitions has also increased. Therefore, in the method of recovering the entire disk or partition as a unit, it is generally considered that even if the size of the recovery unit is small, it is about 10 GByte, and if it is large, it exceeds 300 GByte. If such a large unit is continuously read for recovery, a read error occurs at an error rate that cannot be ignored as described above even when reading from a normal hard disk. When a read error occurs in this way, both hard disks cannot be used. For this reason, manual recovery is necessary while logically comparing the contents of the hard disk. In this case, there is no guarantee that complete recovery is possible, and it is difficult to estimate the time required for the recovery operation. .

  RAID 1 is the most promising method for providing a low-cost and highly reliable hard disk system because it has the simplest configuration among various RAID methods and can be realized by software. However, as described above, RAID 1 is a promising method. The RAID 1 system does not provide an effective recovery method when a partial read error occurs on the hard disk.

  The present invention provides an effective recovery method when a partial read error occurs in any hard disk in a RAID 1 type disk array device.

  The present invention is a RAID 1 type disk array device comprising a plurality of hard disks provided with a reallocation mechanism and a RAID 1 control unit for controlling read / write operations on these hard disks, wherein the RAID 1 control unit Each recovery unit set to a size smaller than the partition of the hard disk is divided and managed, and if a read error occurs when reading from one of the hard disks, the recovery unit in which the read error has occurred There is provided a disk array device characterized in that data in a region is read from another one of the plurality of hard disks, and the read data is written to a hard disk in which a read error has occurred to restore the redundant configuration.

  Here, the recovery unit area is preferably a block which is a read / write unit of the hard disk.

  Preferably, the RAID1 control unit includes a correspondence management table in which a correspondence relationship between each area of the recovery unit and a block group which is a read / write unit of the hard disk is registered, and a block in which a read error has occurred belongs. The block group of the recovery unit area is obtained from the correspondence management table, the data of the block group is read as the recovery unit area data, and written to the hard disk in which the read error has occurred.

  The best mode for carrying out the present invention (hereinafter referred to as “embodiment”) will be described below with reference to the drawings.

  The present embodiment provides a mechanism for improving the efficiency of redundancy recovery when a read error (corresponding to the failure (b) described above) occurs in a RAID 1 disk array device.

In RAID1, the unit for assuming that all recorded data is normal is set to be smaller than the unit of the entire disk or partition, and the unit to be restored when an error is detected is reduced. For example, if the unrecoverable error rate when reading one bit of the hard disk which is a conventional example similar to 10 ^-13 above, if a unit of recovery one megabyte [MByte] and, when reading the recovery unit The error rate can be reduced to 0.0000008, that is, 0.00008%. The error rate is 4 to 6 orders of magnitude lower than when reading the entire partition or disk for recovery on the scale of 10 to several hundred gigabytes. By making the unit of recovery into blocks, if the size is 4 to 16 [KByte], the error rate can be further reduced by about two digits.

  Of course, the recovery unit can be arbitrarily set by the user (RAID 1 system operator). However, if the read / write unit of the hard disk itself, such as a block or sector, is used as the recovery unit, the control is further simplified. Can be

  In the present embodiment, some functions of the current hard disk device are used.

  For example, the current hard disk device has a function of confirming that the rotational speed of the disk, its differential value, and the head position are within a range in which the writing process can be appropriately performed on the disk at the time of writing. Yes. For example, in the IDE standard, it is determined that writing is normally performed when the rotational speed or the like is within a certain range during writing. With this function, data is correctly written in an area to be written, and is prevented from being written in an area that should not be written. The SCSI standard further defines a verify process in which written data is read immediately and a write result is confirmed without a read error. However, since the verification process requires processing costs, the verification process is not implemented on the IDE standard hard disk in consideration of the probability of such an error.

  Further, data written to the hard disk is added with an error correction code called CRC (Cyclic Redundancy Code), and even if it is not written properly, it can be corrected if it is a slight error. If it exceeds the correctable range, it can be determined that an error has occurred during reading.

  Also, in a current hard disk device, when a certain block has a read error, it is determined that the block is a media error (an error that cannot be repaired due to a specific area on the disk being abnormal), and the disk Other blocks on the spare area prepared above are reallocated. The reallocated block is mapped to the block number of the original block. Therefore, although writing to the block is permitted when viewed from the host system, writing to the block reassigned to the block is performed inside the hard disk. This function is called a reassign mechanism or a replacement mechanism.

  In the present embodiment, efficient data recovery processing at the time of a read error in RAID 1 is realized by using a function that these current hard disk devices generally have. This recovery process will be described below.

  First, the system configuration of RAID 1 will be described with reference to FIG. In FIG. 1, a RAID 1 disk array device (which corresponds to a storage subsystem in the case of the entire computer system including the host system 10) includes a RAID 1 control unit 20, a hard disk 30A, and a hard disk 30B (hereinafter, “ Hard disk A "and" hard disk B "). Of these, the RAID1 control unit 20 controls reading and writing of data with respect to the hard disks A and B. A host system 10 such as a program higher than a storage device driver that operates on a personal computer or an operating system is connected to a RAID 1 control unit 20, and the RAID 1 control unit 20 reads data in the same manner as for a single hard disk device. Send a write request. The RAID1 control unit 20 that receives the request from the host system 10 writes the requested data to both the hard disks A and B in the case of a write request, and writes the requested data in the hard disks A and B in the case of a read request. (This may be preset by the system administrator in the RAID1 control unit 20 or may be automatically determined by the RAID1 control unit 20).

  Hereinafter, the control operation of the RAID1 control unit 20 will be described. In the following example, the unit of recovery when a read error occurs is a block (also referred to as a sector), which is a read / write unit of the hard disk itself. In this way, since the block is a recovery unit, the hard disks A and B have the same capacity and the same block size.

  First, in the process at the time of writing, when the RAID 1 control unit 20 receives a write request from the host system 10, it issues a write request for the data to both hard disks A and B. In response to this write request, a response indicating a write success or an error is returned from each of the hard disks A and B. If there is an error response from both hard disks A and B, the RAID1 control unit 20 notifies the host system 10 of the stop of the storage subsystem and then stops the sub storage system. If there is an error response from only one of the hard disks A and B and a write success notification is received from the other, the hard disk that has received the error response is disconnected from the storage subsystem and shifted to a degeneration operation. And the notification which reports that it changed to degenerate operation is performed with respect to the high-order system 10. FIG.

  The above is the control at the time of writing, which is basically the same as the conventional RAID1 control. Next, the control operation of the RAID1 control unit 20 during reading will be described with reference to FIG.

  In this procedure, the RAID1 control unit 20 waits for a read request from the host system 10 (S10). The read request from the host system 10 includes designation of the block number to be read. Upon receiving such a read request, the RAID1 control unit 20 reads the data of the block specified in the request from the hard disk selected in the reading target (hard disk A in the illustrated example) (S12). If no error occurs in this reading (the determination result in S14 is negative (N)), the RAID1 control unit 20 returns the read data to the host system 10 (S24).

  On the other hand, if there is a read error notification from the hard disk A during the read operation in S12 (the determination result in S14 is affirmative (Y)), the RAID1 control unit 20 tried to read from the hard disk A to the hard disk B. Is requested to read the same block (S16). If data can be correctly read from the hard disk B in response to this request (determination result in S18 is negative (N)), the data is rewritten to the corresponding block of the hard disk A (S20). Note that the control unit inside the hard disk A performs reassignment processing of the block in which the read error has occurred when the read error is detected. Therefore, in the rewriting in S20, if a write request is made to the hard disk A for the block in which the read error is detected in S12, 14, the hard disk A automatically performs the reassignment block mapped to the block. Data will be written to. If there is no notification of a write error from the hard disk A at the time of this rewriting (the determination result in S22 is negative (N)), the RAID1 control unit 20 returns the data read in S16 to the host system 10 (S24). .

  If a read error occurs during reading from the hard disk B in S16 (the determination result in S18 is Y), the RAID1 control unit 20 notifies the host system 10 of the stop of the storage subsystem (S26). The storage system is stopped (S28).

  If a write error occurs during the recovery (rewriting) of the hard disk A in S20, the RAID1 control unit 20 disconnects the hard disk A from the storage subsystem and shifts to the degeneration operation of only the hard disk B. (S30) The host system 10 is notified of the transition to the degeneration operation (S32).

  The example of the control operation of the RAID1 control unit 20 at the time of reading has been described above. In the above example, the data read from the hard disk B in S16 is not immediately returned to the host system 10, but the data is rewritten to the hard disk A in S18 and the redundancy is restored. Is returned to the host system 10. However, this order is merely an example. On the contrary, the system can be established by returning the data to the host system 10 immediately after reading the data from the hard disk B in S16 and then rewriting the data to the hard disk A. However, in this case, there is a possibility that a read error will occur again if the next request from the host system 10 comes before the completion of the redundant recovery of the hard disk A due to timing problems. As a procedure for avoiding such a problem, the procedure shown in FIG. 2 is preferable.

  In the above description, the block, which is the read / write unit of the hard disk, is used as the recovery unit in the event of a read error, but it is of course possible to employ a larger size as the recovery unit. If the recovery unit is made smaller than the size of the hard disk or general partition, as described above, the read error occurrence rate during the recovery operation can be reduced. Further, if the recovery unit is made smaller, the time for reading and rewriting at the time of recovery is shortened, and the time required for the recovery work can be shortened as a whole. Since the storage subsystem does not have redundancy until the restoration process is completed, reducing this processing time can greatly improve the reliability of the data recording mechanism as seen from the host system. If the recovery unit is set to a size several orders of magnitude smaller than that of the hard disk or partition, such as 1 megabyte, the read error rate during recovery and the time required for recovery processing can be greatly reduced.

  In the above-described example in which a block, which is a read / write unit of the hard disk, is used as a recovery unit, the RAID1 control unit 20 may recover a block requested to be read from the host system 10 if there is a read error. Therefore, there was no need to perform special hard disk space management for recovery. On the other hand, in order to realize recovery in a unit larger than such a block, the RAID1 control unit 20 has a correspondence management table as shown in FIG. 3 for managing the association between the recovery unit and the block. Make it. As shown in the figure, for each recovery unit number, the number of the block corresponding to the recovery unit is registered in the correspondence management table. In the illustrated example, since a continuous block group is a recovery unit, the corresponding block number indicates the block number of the first and last blocks. Even if an error occurs in a block in the recovery unit, it is reassigned by the control in the hard disk, so there is no need to modify the correspondence management table itself.

  In the case of the configuration having the correspondence management table as described above, when the RAID1 control unit 20 receives a read error notification from the hard disk, the number of the recovery unit to which the number of the block to be read at that time belongs is shown in FIG. It is obtained from the correspondence management table as illustrated. Then, the block group belonging to the obtained recovery unit is obtained from the correspondence management table, the data of the block group is read from the other hard disk, and written to the hard disk having the read error. With the above control, it is possible to restore redundancy when a read error occurs.

  The size of the recovery unit can also be set by the system administrator via a parameter setting program of the RAID1 control unit 20 or the like. In this case, if the administrator specifies the size of the recovery unit, for example, in units of bytes or blocks, the program calculates the block numbers at the beginning and end of each recovery unit according to the size, and manages them accordingly. It should be registered in the table.

  The RAID1 control unit 20 described above can be configured as a hardware RAID controller, a RAID control program, or a hardware and software compromise system.

  As described above, according to this embodiment, when a read error occurs in a part of a hard disk in a RAID 1 disk array device, automatic recovery to a redundant state can be performed. In addition, since the unit of automatic recovery is significantly smaller than the conventional RAID 1 system, the probability of occurrence of a normal hard disk read error during automatic recovery can be greatly reduced, and the time required for automatic recovery is also greatly increased. Can be shortened. The method of this embodiment is most effective especially in a disk array device using an IDE disk drive. This is because the IDE disk drive does not perform the verification process of the written data, so that there is a high probability that an error will occur during reading.

It is a figure which shows the system configuration | structure of RAID1. It is a flowchart which shows the example of control operation of the RAID1 control part at the time of reading. It is a figure which shows the example of a management table of a recovery unit and the block of a hard disk.

Explanation of symbols

  10 host system, 20 RAID1 control unit, 30A, 30B hard disk.

Claims (6)

A RAID 1 type disk array device comprising a plurality of hard disks provided with a reallocation mechanism and a RAID 1 controller for controlling read / write operations on these hard disks,
The RAID1 control unit manages the hard disk by dividing the area into recovery units set to a size smaller than the partition of the hard disk, and a read error occurs when reading from one of the hard disks Recovering the redundant configuration by reading the data of the recovery unit area in which the read error has occurred from the other one of the plurality of hard disks and writing the read data to the hard disk in which the read error has occurred. A featured disk array device.   2. The disk array device according to claim 1, wherein the recovery unit area is a block which is a read / write unit of the hard disk.   The RAID1 control unit includes a correspondence management table in which a correspondence relationship between each area of the recovery unit and a block group that is a read / write unit of the hard disk is registered, and a block of a recovery unit area to which a block in which a read error occurs belongs 2. The disk array device according to claim 1, wherein a group is obtained from the correspondence management table, the data of the block group is read as data in the area of the recovery unit, and written to the hard disk in which a read error has occurred. A disk array control method for controlling read / write operations in a RAID 1 system for a plurality of hard disks having a reallocation mechanism,
The hard disk is divided into areas for each recovery unit set to a size smaller than the partition of the hard disk, and is managed.
If a read error occurs when reading from one of the plurality of hard disks, the data of the recovery unit area where the read error has occurred is read from the other one of the plurality of hard disks, and the read data is read. Restore the redundant configuration by writing to the hard disk where the error occurred.
And a disk array control method. A program for causing a computer system to function as a disk array control device that controls read / write operations in RAID 1 with respect to a plurality of hard disks having a reallocation mechanism,
Managing the hard disk by dividing an area for each recovery unit set to a size smaller than the partition of the hard disk;
If a read error occurs when reading from one of the plurality of hard disks, reading the data of the recovery unit area where the read error has occurred from the other one of the plurality of hard disks;
Restoring the redundant configuration by writing the read data to the hard disk in which the read error has occurred; and
A program that executes A program for causing a computer system to function as a disk array control device that controls read / write operations in RAID 1 with respect to a plurality of hard disks having a reallocation mechanism,
If a read error occurs when reading data from one of the plurality of hard disks, reading the block data in which the read error has occurred from the other one of the plurality of hard disks;
Restoring the redundant configuration by writing the read block data to the corresponding block of the hard disk in which the read error has occurred; and
A program that executes

Images