JP3832223B2 - Disk array disk failure recovery method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention mainly relates to a disk failure recovery method in an external storage system of a computer.
[0002]
[Prior art]
(1) Disk failure recovery method for disk array
The disk array system, also called RAID (Redundant Arrays of Inexpensive Disks), has a configuration in which a plurality of disk devices are arranged in an array, and a read request (data read request) from a host device (hereinafter abbreviated as a host). And a write device (data write request) that is processed at a high speed by the parallel operation of the disk device, and the reliability is improved by adding redundant data to the data. Disk array systems are classified into five levels according to the type and configuration of redundant data (Paper: "A Case for Redundant Arrays of Inexpensive Disks (RAID)", David A. Patterson, Garth Gibson, and Randy H. Katz. , Computer Science Division Department of Electrical Engineering and Computer Sciences, University of California Berkeley).
[0003]
In order to realize the disk array as described above, a read / write request from the host is converted into a read / write request to each disk device, data is distributed to each disk device at the time of writing, and each disk device at the time of reading. It is necessary to perform data distribution / set control that collects data from Such control is called disk array control.
[0004]
In the RAID 5 level, for example, where parity is added in the disk array, the contents of the disk can be recovered by guaranteeing parity with other disks even if one disk fails. When recovering a disk, the failed disk is replaced, data is read from the other disks constituting the parity group, subjected to XOR operation (exclusive OR), and the operation result is written to the replaced disk.
[0005]
(2) Duplication for snapshots
In general, data recorded on an external storage device of a computer such as a hard disk is regularly copied to tape so that the lost data can be recovered if the data is lost due to a device failure, software defect or incorrect operation. Backup is required. At this time, if the data is updated during the copying operation and the data is inconsistent, it does not make sense as a backup. Therefore, it is necessary to guarantee the data consistency during the copying operation.
[0006]
In order to guarantee the consistency of the data to be backed up, a program other than the backup program that accesses the data may be stopped. However, in a system that requires high availability, the program cannot be stopped for a long time. Therefore, it is necessary to provide a mechanism for creating a storage image of data at the start of backup without preventing the program from updating data during backup. Here, a storage image of data at a certain time point is called a snapshot, and a mechanism that provides a state in which data can be updated while creating a snapshot at a specified time point is called a snapshot management method. In addition, creating a snapshot by the snapshot management method is called snapshot acquisition, and the data for which the snapshot is acquired is called original data. Also, quitting the state in which the snapshot has been created is called snapshot deletion.
[0007]
One conventional snapshot management method is a data duplication method.
[0008]
In this method, in a normal state where a snapshot is not acquired, a program on a computer duplexes (mirrors) all data in two storage areas. When a snapshot is acquired, duplication is stopped, the two storage areas are separated into independent areas, one area is provided as original data, and the other area is provided as a snapshot.
[0009]
While the snapshot is acquired and duplication is stopped, updating of the data in the storage area of the original data is permitted, and when the data update occurs, the updated position is recorded. When the snapshot was deleted, data duplication was resumed, and update data whose contents did not match between the two storage areas were provided as a snapshot from the storage area of the original data based on the record of the update position Copy to storage area (mirror resynchronization). A method of duplicating data by a program on a computer is shown in US Pat. No. 5,051,887, for example.
[0010]
[Problems to be solved by the invention]
At the time of parity reconstruction after disk replacement, there is a problem in that normal access performance deteriorates because it is necessary to read data from all disks in the same parity group other than the replaced disk. Further, in nD + 1P representing a configuration such as RAID 5 forming a parity group, there is a problem that the performance is further deteriorated when the number n of data drives is increased.
[0011]
In the snapshot management method based on data duplication, during mirror resynchronization, normal access for updating / referencing and copy access for updated data are concentrated in the storage area of the original data, and the performance of normal access deteriorates. Since the time required for mirror resynchronization is proportional to the amount of data updated while taking a snapshot and stopping duplexing, assuming that update access occurs the same number of times per unit time, mirror resynchronization The time required for conversion increases in proportion to the stoppage time of duplexing. When backup is performed in parallel with parity reconstruction after disk replacement, it takes a long time to stop the duplication after taking a snapshot, resulting in longer mirror resynchronization time, which reduces normal access performance. is there.
[0012]
A first object of the present invention is to provide a disk failure recovery method that suppresses a decrease in normal access performance at the time of parity reconstruction after disk replacement in a disk array that is duplicated for snapshot acquisition. .
[0013]
The second object of the present invention is to perform mirror resynchronization by shortening the parity rebuilding time after disk replacement in a period when the duplexing is stopped in a disk array that is duplexed for snapshot acquisition. It is to provide a disk failure recovery method that reduces the amount of copy of update data in the disk and shortens the mirror resynchronization time when the performance deteriorates.
[0014]
In addition to the first and second objects, the third object of the present invention is a disk failure recovery which suppresses performance degradation due to an increase in the number of data drives n in nD + 1P representing a disk configuration such as RAID 5 forming a parity group Is to provide a method.
[0015]
[Means for Solving the Problems]
In order to achieve the first object, the present invention replaces parity rebuilding performed at the time of recovery of a failed disk with parity generation using parity redundancy in a disk array that is duplicated for snapshot acquisition. Thus, a disk failure recovery subprogram that is implemented by copying data from a disk at the same position in the snapshot mirror configuration is provided. By reducing the number of disk accesses related to parity reconstruction after disk replacement, it is possible to suppress a decrease in normal access performance.
[0016]
In order to achieve the second object, according to the present invention, a disk failure recovery subprogram similar to the above is provided in a disk array that is duplicated for snapshot acquisition. By shortening the parity rebuild time after disk replacement in the period when duplexing is stopped, the copy amount of update data during mirror resynchronization is reduced, and the mirror resynchronization time when performance is degraded is shortened. Can do.
[0017]
In order to achieve the second object, according to the present invention, a disk failure recovery subprogram similar to the above is provided in a disk array that is duplicated for snapshot acquisition. In nD + 1P representing a disk configuration such as RAID 5 that forms a parity group, copying is performed from one disk regardless of the number of n, so that a decrease in performance due to an increase in the number of data drives n can be suppressed.
[0018]
DETAILED DESCRIPTION OF THE INVENTION
The first embodiment of the present invention suppresses a decrease in normal access performance by reducing the number of disk accesses related to parity reconstruction after disk replacement in a disk array that is duplicated for snapshot acquisition. Is to do.
[0019]
In addition, by shortening the parity rebuild time after disk replacement during the period when duplexing is stopped, the amount of update data copied during mirror resynchronization is reduced, and the mirror resynchronization time when performance is degraded is shortened. Is to do.
[0020]
Also, by reducing the number of disk accesses related to parity reconstruction after disk replacement, it is possible to suppress performance degradation due to an increase in the number n of data drives in nD + 1P representing a disk configuration such as RAID 5 forming a parity group It is.
[0021]
In the present invention, a backup is taken as an example of using a snapshot, but it can also be used for other purposes such as OLAP (OnLine Analytical Processing) and system testing.
[0022]
(1) Description of configuration
The system configuration of the first embodiment of the present invention will be described with reference to FIG. In FIG. 1, a computer 100 and a disk array 200 are connected by a SCSI bus 300 via SCSI interfaces 140 and 240. The memory 120 of the computer 100 includes a database program 126 and a backup program 127 that are executed by the CPU 110 that controls the computer 100. The disk array 200 includes disk groups 251 to 252 controlled by the disk controller 250, and a snapshot management program 221 is stored in the memory 220 and is executed by the CPU 210. The disk group 251 has disks 271 to 275, the disk group 252 has disks 281 to 282, and each disk group has a striped array configuration with parity. In the present embodiment, the number of disks in the disk groups 251 to 252 is five, but each disk group 251 to 252 only needs to be composed of three or more disks.
[0023]
Storage areas in the disk groups 251 to 252 are accessed from the computer 100 as LUs (Logical Units) which are SCSI logical units. The storage areas corresponding to the disk groups 251 to 252 are referred to as LUs 261 to 262, respectively. There may be a plurality of LUs in each of the disk groups 251 to 252. In this embodiment, the snapshot management program 221 in the disk array 220 manages the LU 261 and LU 262 by duplicating the LU 261 as a mirror source LU having original data, and the LU 262 is a mirror destination LU which is a mirror of the original data. And The LU 262 is an LU used as a snapshot.
[0024]
Next, a program in the computer 100 will be described.
[0025]
The database program 126 has a function of accessing the LU 261 that is the mirror source LU during execution and switching to a backup mode that controls data update and ensures data consistency in the LU 261. Transition to the backup mode is made in accordance with an instruction from the user or the backup program 127. The backup program 127 has a function of reading data to be backed up to a tape or the like from the LU 262 storing a snapshot according to an instruction from the user, a function of issuing a SCSI ModeSelect command to the disk array 200, and a backup mode to the database program 126. It has a function to instruct to enable / disable.
[0026]
Next, programs and management tables in the disk array 200 will be described.
[0027]
The snapshot management program 221 of the disk array 200 has a disk access subprogram 230 for instructing disk access to the disk controller 250 in response to a request from the computer 100 and another update designated in advance by duplicating updates to one LU. This also applies to LUs, and has an LU mirror subprogram 231 that writes the same contents to two LUs. The disk access subprogram 230 performs disk array control for converting read / write requests from the computer 100 into read / write requests to the disks 271 to 275 and 281 to 285. The LU mirror subprogram 231 duplicates access to the LU 261 to the LU 262.
[0028]
The snapshot management program 221 also includes a non-mirror time update monitoring subprogram 234 that detects an update to the mirror source LU when duplexing is stopped (non-mirror time), and an update / inconsistency whose update position will be described later. Non-mirrored update location management subprogram 235 recorded in the location management table 222, mirror resynchronization subprogram 232 for copying the update part of the mirror source LU to the mirror destination LU when mirror resynchronization is performed, and the failed disk And a disk failure recovery subprogram 233 for recovering to a striped array configuration with parity after replacement. The update / inconsistency position management table 222 is used to manage the data contents of the mirror source LU and the mirror destination LU. The update position of the mirror source LU updated at the time of non-mirror and the same LU in the replaced disk after the disk failure The position at which the parity is inconsistent is recorded.
[0029]
The update / inconsistency position management table 222 is a bitmap as shown in FIG. 2 and includes all LBA set numbers in the mirror source LU, update bits associated with the LBA set numbers, mirror source inconsistency bits, and mirror destinations. Consists of inconsistent bits. An LBA set is an individual set when the entire area of an LU is divided from the top in units of one or more equal numbers of LBAs (Logical Block Address), and the LBA set number is the top of the LBA. A serial number is assigned to each LBA set from the side.
[0030]
It is assumed that one LBA set is distributed to each disk from the head side by data distribution of the disk array, and a parity group number is added to a set of LBA sets at the same position in each disk. In the example of FIG. 2, the parity group number 0 is composed of LBA set numbers 0 to 4. In this embodiment, data distribution is performed for five disks, and a disk set number representing a set of LBA sets for each disk is added. In the example of FIG. 2, the disk set 0 is configured with LBA set numbers 0, 5, 10, and 15.
[0031]
The update bit indicates whether or not the LBA set of the mirror source LU has been updated at the time of non-mirroring, and designates 1 and 0 according to “update” and “non-update”, respectively. The initial value of the update bit is 0. The example of FIG. 2 shows a state in which only the area of LBA set number 1 is updated at the time of non-mirroring.
[0032]
The mirror source inconsistency bit indicates whether or not the parity is consistent with other disks in the same LU when the mirror source LU is exchanged, and is 1 and 0 in accordance with “inconsistency” and “match”, respectively. Is specified. The formula setting value of the mirror source mismatch bit is 0.
[0033]
The mirror destination inconsistency bit indicates whether the parity is consistent with other disks in the same LU when the mirror destination LU is exchanged, and is 1 or 0 in accordance with “inconsistency” or “match”, respectively. Is specified. The setting value of the mirror destination mismatch bit is 0. The example of FIG. 2 shows a state in which the area of the disk set 0 with the parity group numbers 1 to 3 is in a parity mismatch with other disks in the mirror destination LU. The area of disk set 0 with parity group number 0 indicates that parity consistency with other disks is achieved by a disk failure recovery method described later.
[0034]
(2) Snapshot acquisition / deletion
Taking the case of providing a snapshot of LU 261 as a mirror source LU as LU 262 as a mirror destination LU as an example, the operations of backup program 127 and snapshot management program 221 at the time of snapshot acquisition / deletion will be described with reference to the flowchart of FIG. explain.
[0035]
First, the backup program 127 of the computer 100 gives an instruction to the database program 126, and validates the backup mode to guarantee the consistency of the data for acquiring the snapshot (step 2000). Next, the backup program 127 issues a ModeSelect command for acquiring a snapshot to the disk array 200 (step 2001). Upon receipt of the ModeSelect command (step 3000), the snapshot management program 221 of the disk array 200 enables the non-mirror update monitoring subprogram 234 and the non-mirror update location management subprogram 235, and starts recording the update location of the LU 261. (Step 3001). Thereafter, when the LU 261 is updated, 1 is set in the update bit of the LBA set including the updated LBA in the update / inconsistent position management table 222, and the fact that there has been an update is recorded. Next, the snapshot management program 221 invalidates the LU mirror subprogram 231 and stops duplication of the LU 261 and LU 262 (step 3002). As a result, the update to the LU 261 that is the mirror source LU is not reflected in the LU 262 that is the mirror destination LU. Next, the snapshot management program 221 transmits the completion status of the ModeSelect command to the backup program 127 of the computer 100 (step 3003). When the backup program 127 receives the completion status of the ModeSelect command (step 2002), it gives an instruction to the database program 126 and invalidates the backup mode (step 2003).
[0036]
Next, the backup program 127 issues a ModeSelect command that instructs the LU 263 to delete the snapshot to the disk array 200 (step 2004). When receiving the ModeSelect command (step 3005), the snapshot management program 221 of the disk array 200 validates the LU mirror subprogram 231 and resumes duplication of the LU 261 and LU 262 (step 3006). Thereby, the update to the LU 261 is reflected in the LU 262. Next, the snapshot management program 221 invalidates the non-mirror time update monitoring subprogram 234 and the non-mirror time update position management subprogram 235, and stops the LU 261 update position recording (step 3007). Thereafter, the update bit in the update / inconsistent position management table 222 is not changed by the non-mirrored update position management subprogram 235. Next, the snapshot management program 221 activates the mirror resynchronization subprogram 232, refers to the update / inconsistency position management table 222, and copies the portion where the contents do not match between the LU 261 and LU 262 from the LU 261 to the LU 262 ( Step 3008). Next, the snapshot management program 221 invalidates the mirror resynchronization subprogram 232 (3009), and transmits the completion status of the ModeSelect command to the backup program 127 of the computer 100 (step 3010). The backup program 127 receives the end status of the ModeSelect command and ends the operation (step 2005).
[0037]
Here, the operation of the mirror resynchronization subprogram 232 for copying data from the LU 261 to the LU 262 in step 3008 will be described with reference to the flowchart of FIG. First, the mirror resynchronization subprogram 232 checks whether there is 1 as an update record in the update bit of the update / mismatch position management table 222 (step 1001). If there is no update record 1, mirror resynchronization is complete and the process ends (step 1002). If there is an update record, the update of the corresponding recording position is suppressed (step 1003), and the data of the corresponding recording position is stored.
Copying from the LU 261 that is the mirror source LU to the LU 262 that is the mirror destination LU (step 1004). Data is copied by issuing a READ command to one of the disks 271 to 275 of the disk group 251 including the LU 261 to read the specified LBA data, and issuing a READ command for the disk group 251 of the disk group 252 including the LU 262. The WRITE command is issued to any of the disks 281 to 285 corresponding to the selected disk, and the read data is written in the same LBA as the LBA designated in the LU 261. The COPY command may be used to copy the data. Next, the mirror resynchronization subprogram 232 cancels the update suppression of the corresponding recording position (step 1005), sets the corresponding update bit of the update / inconsistent position management table 222 to 0 and deletes the update record ( Step 1006) and return to Step 1001.
[0038]
The above is the operation of the backup program 127 and the snapshot management program 221 when acquiring / deleting a snapshot. The backup program 127 of the computer 100 can read the LU 262 that acquired the snapshot between step 2003 and step 2004.
[0039]
(3) Disk failure recovery method
In the present invention, parity reconstruction for a newly set disk by replacing a failed disk is performed by copying from a mirrored paired disk instead of using parity redundancy. The pair of disks corresponds to the relationship between the disk 271 and the disk 281 assuming that the disks 271 to 275 and the disks 281 to 285 in FIG. 1 form a disk group in this order, for example.
[0040]
In the present embodiment, parity reconstruction after replacing a failed disk will be described in three phases: a mirror phase, a non-mirror phase, and a resynchronization phase, depending on which stage of the snapshot acquisition / deletion operation is performed. . The non-mirror phase is from the start of step 3002 to the end of step 3007 in FIG. The resynchronization phase is from the start of step 3008 to the end of step 3009 in FIG. 3 in which the updated data is copied from the mirror source LU to the mirror destination LU. The mirror phase is the stage where it is mirrored, and the entire range other than the non-mirror phase and the resynchronization phase.
[0041]
Also, it is classified according to whether the disk of the mirror source LU or the mirror destination LU was replaced during each phase.
[0042]
The parity reconstruction after the replacement of the failed disk is performed by the snapshot management program 221 of the disk array 200 enabling the disk failure recovery subprogram 233, and is performed independently of the instruction from the computer 100. be able to.
[0043]
(3-1) Mirror phase
The operation of the disk failure recovery subprogram 233 in the mirror phase will be described with reference to the flowchart of FIG. In the mirror phase, the contents of the mirror source LU and the mirror destination LU are the same, and the operation of the disk failure recovery subprogram 233 is the same regardless of whether the mirror source LU or the mirror destination LU is replaced. Take the case where the disk of the destination LU is replaced as an example. Here, it is assumed that the disk replaced due to the failure is the disk 281 of the disk group 252 which is the mirror destination LU in FIG. When mirrored, the same data as the disk 271 of the disk group 251 that is the mirror source LU is stored in the disk 281. First, when a disk is replaced, the disk failure recovery subprogram 233 sets all the mirror destination inconsistency bits in the update / inconsistency position management table 222 corresponding to the replaced disk 281 to 1 (step 4001). Next, the disk failure recovery subprogram 233 checks whether there is 1 as the inconsistent record in the mirror destination inconsistent bit of the update / inconsistent position management table 222 (step 4002). If there is no inconsistent record 1, the failure recovery of the disk 281 is completed, and the parity of the disk group 252 is consistent, so the process ends (step 4003). If there is inconsistent recording, update of the corresponding recording position is suppressed (step 4004), and it is checked whether or not there is 1 as inconsistent recording in the mirror source inconsistent bit of the update / inconsistent position management table 222 (step 4005). . If there is no inconsistent record 1, the mirror source LU has parity consistency, and the data at the corresponding position is copied from the disk 271 to the disk 281 (step 4007). If there is 1 that is inconsistent recording, the mirror source LU is also recovering from the failure and the consistency of the parity is not taken. Is written in the corresponding position of the disk 281 (step 4006). After step 4007 or step 4006, the disk failure recovery subprogram 233 sets the inconsistency bit corresponding to the disk 281 in the update / inconsistency position management table 222 to 0 and deletes the inconsistency record (step 4008). ) Releases the update inhibition of the corresponding recording position (step 4009), and returns to step 4002.
[0044]
The above is the operation of the disk failure recovery subprogram 233 in the mirror phase.
[0045]
In step 4007, the data is copied by reading the LBA data designated by issuing a READ command to the disk 271 and reading the data read to the same LBA as the LBA designated by the disk 271 by issuing the WRITE command to the disk 281. Implement by writing. The COPY command may be used to copy the data.
[0046]
When a data update request is received from the computer 100 during the operation of the disk failure recovery subprogram 233, the snapshot management program 221 is to reflect both in the mirror source LU and the mirror destination LU. Set to zero. When a data read request is received from the computer 100, data is read from the LU whose inconsistency bit is 0. When the inconsistency bit of both LUs is 1, data is read from all the disks other than the replaced disk, and the requested data is restored by the parity calculation and transmitted to the computer 100.
[0047]
(3-2) Non-mirror phase
The operation of the disk failure recovery subprogram 233 in the non-mirror phase will be described with reference to the flowchart of FIG. In the non-mirror phase, the contents of the mirror source LU and the mirror destination LU do not match, but the operation of the disk failure recovery subprogram 233 is the same when either the mirror source LU or the mirror destination LU is replaced. Therefore, the case where the mirror source LU disk is replaced is taken as an example. Here, it is assumed that the disk replaced due to the failure is the disk 271 of the disk group 251 that is the mirror source LU in FIG.
[0048]
First, when a disk is replaced, the disk failure recovery subprogram 233 sets all the mirror source inconsistency bits in the update / inconsistency position management table 222 corresponding to the replaced disk 271 to 1 (step 5001). Next, the disk failure recovery subprogram 233 checks whether there is 1 as a mismatch record in the mirror source mismatch bit of the update / mismatch position management table 222 (step 5002). If there is no inconsistent record 1, the failure recovery of the disk 271 is completed, and the parity of the disk group 251 is consistent, so the process ends (step 5003). If there is inconsistent recording, update of the corresponding recording position is suppressed (step 5004), and it is checked whether or not there is 1 as inconsistent recording in the mirror destination inconsistent bit of the update / inconsistent position management table 222 (step 5005). . If there is 1 that is inconsistent recording, the mirror destination LU is also recovering from the failure and the parity is not consistent, so data is read from other disks 272 to 275 in the same disk group 251 as the disk 271 and parity calculation is performed. Thus, the data is restored and written in the corresponding position on the disk 271 (step 5006). If there is no inconsistent record 1, it is checked whether or not there is 1 as an update record in the corresponding update bit of the update / inconsistent position management table 222 (step 5007). If there is an update record 1, the data on the disk 271 and the disk 281 at the corresponding positions are different, so the disk failure recovery subprogram 233 reads the data from the other disks 272 to 275 in the same disk group 251 as the disk 271. Data is restored by parity operation and written to the corresponding position on the disk 271 (step 5006). If there is no update record 1, the data on the disk 271 and the disk 281 at the corresponding position should be the same, so the data at the corresponding position is copied from the disk 281 to the disk 271 (step 5008).
[0049]
After step 5006 or step 5008, the disk failure recovery subprogram 233 sets the inconsistency bit corresponding to the disk 271 in the update / inconsistency position management table 222 to 0 and deletes the inconsistency record (step 5009). ) Releases the update inhibition of the corresponding recording position (step 5010), and returns to step 5002. The above is the operation of the disk failure recovery subprogram 233 in the non-mirror phase.
[0050]
In step 5008, the data is copied by reading the LBA data designated by issuing a READ command to the disk 281 and reading the data read to the same LBA as the LBA designated by the disk 281 by issuing the WRITE command to the disk 271. Implement by writing. The COPY command may be used to copy the data.
[0051]
When a data read request is received from the computer 100 during the operation of the disk failure recovery subprogram 233, the snapshot management program 221 normally reads data from the mirror source LU, but the update bit is 0 and the mirror destination mismatch bit is 0. May be read from the mirror destination LU.
[0052]
(3-3) Resynchronization phase (mirror destination LU recovery)
The operation of the disk failure recovery subprogram 233 related to the mirror destination LU in the resynchronization phase will be described with reference to the flowchart of FIG. Here, it is assumed that the disk replaced due to the failure is the disk 281 of the disk group 252 which is the mirror destination LU in FIG.
[0053]
First, when a disk is replaced, the disk failure recovery subprogram 233 sets all the mirror destination inconsistency bits in the update / inconsistency position management table 222 corresponding to the replaced disk 281 to 1 (step 6001). Next, the disk failure recovery subprogram 233 checks whether there is 1 as the inconsistent record in the mirror destination inconsistent bit of the update / inconsistent position management table 222 (step 6002). If there is no inconsistent recording 1, the failure recovery of the disk 281 is completed, and the parity of the disk group 252 is consistent, and the process is terminated (step 6003). If there is inconsistent recording, update of the corresponding recording position is suppressed (step 6004), and it is checked whether or not there is 1 as inconsistent recording in the mirror source inconsistent bit of the update / inconsistent position management table 222 (step 6005). . If there is no inconsistent record 1, the mirror source LU has parity consistency, so the data at the corresponding position is copied from the disk 271 to the disk 281 (step 6007), and the update / inconsistent position management table 222. The corresponding update bit is set to 0 and the update record is deleted regardless of whether or not there is an update (step 6008). If there is 1 inconsistent recording, the mirror source LU is also recovering from the failure and the parity is not consistent, so data is read from other disks 282 to 285 in the same disk group 252 as the disk 281 and parity calculation is performed. Thus, the data is restored and written in the corresponding position on the disk 281 (step 6006). After step 6008 or step 6006, the disk failure recovery subprogram 233 sets the inconsistency bit corresponding to the disk 281 in the update / inconsistency position management table 222 to 0 and deletes the inconsistency record (step 6006). ) Releases the update inhibition of the corresponding recording position (step 6010), and returns to step 6002.
[0054]
The above is the operation of the disk failure recovery subprogram 233 related to the mirror destination LU in the resynchronization phase.
[0055]
In step 6007, the data is copied by reading the LBA data designated by issuing a READ command to the disk 271 and reading the data read to the same LBA as the LBA designated by the disk 271 by issuing the WRITE command to the disk 281. Implement by writing. The COPY command may be used to copy the data.
[0056]
(3-4) Resynchronization phase (mirror source LU recovery)
The operation of the disk failure recovery subprogram 233 related to the mirror source LU in the resynchronization phase will be described with reference to the flowchart of FIG. Here, it is assumed that the disk replaced due to the failure is the disk 271 of the disk group 251 that is the mirror source LU in FIG.
[0057]
First, when a disk is replaced, the disk failure recovery subprogram 233 sets all the mirror source inconsistency bits of the update / inconsistency position management table 222 corresponding to the replaced disk 271 to 1 (step 7001). Next, the disk failure recovery subprogram 233 checks whether there is 1 as the inconsistent record in the mirror source inconsistent bit of the update / inconsistent position management table 222 (step 7002). If there is no inconsistent record 1, the failure recovery of the disk 271 is completed, and the parity of the disk group 251 is consistent, so the process ends (step 7003). If there is inconsistent recording, update of the corresponding recording position is inhibited (step 7004), and it is checked whether or not there is 1 as inconsistent recording in the mirror destination inconsistent bit of the update / inconsistent position management table 222 (step 7005). . If there is 1 that is inconsistent recording, the mirror destination LU is also recovering from the failure and the parity is not consistent, so data is read from other disks 272 to 275 in the same disk group 251 as the disk 271 and parity calculation is performed. Thus, the data is restored and written in the corresponding position on the disk 271 (step 7006). If there is no inconsistent record 1, it is checked whether or not there is 1 as an update record in the corresponding update bit of the update / inconsistent position management table 222 (step 7007). If there is an update record 1, the data on the disk 271 and the disk 281 at the corresponding positions are different, so the disk failure recovery subprogram 233 reads the data from the other disks 272 to 275 in the same disk group 251 as the disk 271. Data is restored by parity calculation and written to the corresponding position on the disk 271 (step 7006). If there is no update record 1, the data on the disk 271 and the disk 281 at the corresponding position may be the same, so the data at the corresponding position is copied from the disk 281 to the disk 271 (step 7008).
[0058]
After step 7006 or step 7008, the disk failure recovery subprogram 233 sets the inconsistency bit corresponding to the disk 271 in the update / inconsistency position management table 222 to 0 and deletes the inconsistency record (step 7009). ) Releases the update inhibition of the corresponding recording position (step 7010), and returns to step 7002. The above is the operation of the disk failure recovery subprogram 233 related to the mirror source LU in the resynchronization phase.
[0059]
In step 7008, the data is copied by reading the LBA data designated by issuing a READ command to the disk 281 and reading the data read to the same LBA as the LBA designated by the disk 281 by issuing the WRITE command to the disk 271. Implement by writing. The COPY command may be used to copy the data.
[0060]
When a data read request is received from the computer 100 during the operation of the disk failure recovery subprogram 233, the snapshot management program 221 normally reads data from the mirror source LU, but the update bit is 0 and the mirror destination mismatch bit is 0. May be read from the mirror destination LU.
[0061]
(4) Computer operation when reading and writing data and reading snapshots
First, the operation of the database program 126 when the computer 100 accesses the data of the LU 261 in the disk array 200 will be described. The database program 126 performs the same operation regardless of whether or not a snapshot is acquired.
[0062]
When the database program 126 reads LU 261 data, the database program 126 issues a READ command for reading the LU 261 data to the disk array 200. Finally, the database program 126 receives data and status from the disk array 200 and ends the operation. When the database program 126 writes data to the LU 261, the database program 126 issues a WRITE command for writing data to the LU 261 to the disk array 200, and transmits the data. Finally, the database program 126 receives the status from the disk array 200 and ends the operation.
[0063]
Next, the operation of the backup program 127 when the computer 100 reads a snapshot of the LU 261 in the disk array 200 will be described.
[0064]
When the backup program 127 reads the snapshot of the LU 261, the backup program 127 issues a READ command for reading the data of the LU 262, which is the mirror destination LU of the LU 261, to the disk array 200. Finally, the backup program 127 receives data and status from the disk array 200 and ends the operation.
[0065]
(5) Disk array operation when reading and writing data and reading snapshots
First, the operation of the snapshot management program 221 when the computer 100 accesses the data of the LU 261 in the disk array 200 will be described.
[0066]
When the computer 100 reads data from the LU 261, the snapshot management program 221 receives a READ command for the LU 261. Next, if the LU mirror subprogram 231 is valid and the copy of the updated portion by the mirror resynchronization subprogram 232 has been completed, data is read from the LU 261 or the LU 262 that is the mirror destination LU. Otherwise, data is read from the LU 261. Finally, the read data and status are transmitted to the computer 100. If the contents of the LU 261 and the LU 262 that is the mirror destination LU match, the load can be distributed by reading data from either of them.
[0067]
When the computer 100 writes data to the LU 261 and updates the stored contents, the snapshot management program 221 receives a WRITE command and data for the LU 261. Next, if the LU mirror subprogram 231 is valid, data is written to the LU 261 and the mirror destination LU 262, and if invalid, data is written to the LU 261. Next, if the non-mirror time update monitoring subprogram 234 and the non-mirror time update position management subprogram 235 are valid, the update bit of the LBA set including the updated LBA for the update / mismatch position management table 222 of the LU 261 Set to 1 and do nothing if disabled. Finally, the status is transmitted to the computer 100.
[0068]
Next, the operation of the snapshot management program 221 when the computer 100 reads a snapshot of the LU 261 in the disk array 200 will be described.
[0069]
When the computer 100 reads the snapshot of the LU 261, the snapshot management program 221 receives a READ command for the LU 262 that is the mirror destination LU of the LU 261. Next, the snapshot management program 221 reads data from the LU 262 that is the mirror destination LU. Finally, the read data and status are transmitted to the computer 100.
[0070]
Note that during the copying of the updated portion by the mirror resynchronization subprogram 232, the copy processing and the data access processing to the LU 261 by the computer 100 are concentrated on the same LU 261, so that the data access performance is degraded.
[0071]
(6) Effect
According to this embodiment, in a disk array that is operated in duplicate for snapshot acquisition, it is possible to suppress a decrease in normal access performance by reducing the number of disk accesses related to parity reconstruction after disk replacement. There is an effect that can be done.
[0072]
In addition, according to the present embodiment, by reducing the parity reconstruction time after disk replacement during the period when duplexing is stopped, the copy amount of update data during mirror resynchronization is reduced, and the performance is degraded. The mirror resynchronization time can be shortened.
[0073]
For example, assuming a RAID 5 5D + 1P configuration, when recovering a disk failure in the mirror phase, or when recovering a disk with parity redundancy, four disk reads and one disk write occur. By applying the present invention, it is possible to perform one disk read and one disk write, and it is possible to suppress a decrease in normal access performance. Even in the non-mirror phase, the same effect of reducing the number of disk accesses can be expected, it is possible to suppress read performance degradation when taking snapshots and taking backups, etc., and it is possible to suppress increase in read time. Mirror resynchronization time that degrades performance can be shortened.
[0074]
In addition, by reducing the number of disk accesses related to parity reconstruction after disk replacement, it is possible to suppress performance degradation due to an increase in the number n of data drives in nD + 1P representing a disk configuration such as RAID 5 forming a parity group. is there. In the case of nD + 1P, when disk recovery is performed by parity redundancy, n disk reads and one disk write occur. By applying the present invention, it is possible to perform one disk read and one disk write.
[0075]
In the present invention, the SCSI bus 300 is used as an interface for connecting the computer 100 and the disk array 200, but another interface such as Fiber Channel may be used.
[0076]
In this embodiment, the mirror source LU and the mirror destination LU are duplicated to obtain a snapshot. However, the present invention can also be applied to a multiple mirror provided with a plurality of mirror destination LUs. In this case, the mirror destination inconsistency bits of the update / inconsistency position management table 222 are provided for the mirror destination LU, and the disk failure recovery subprogram 233 performs the same operation as when the mirror destination LU and the mirror source LU are duplicated. LU mirror subprogram 231 is the mirror source
An operation of multiplexing access to the LU to a plurality of mirror destination LUs may be performed.
[0077]
【The invention's effect】
As described above, according to the present invention, in a disk array that is duplicated for snapshot acquisition, normal access performance is reduced by reducing the number of disk accesses related to parity reconstruction after disk replacement. There is an effect that can be suppressed.
[0078]
In addition, according to the present embodiment, by reducing the parity reconstruction time after disk replacement during the period when duplexing is stopped, the copy amount of update data during mirror resynchronization is reduced, and the performance is degraded. The mirror resynchronization time can be shortened.
[0079]
In addition, by reducing the number of disk accesses related to parity reconstruction after disk replacement, it is possible to suppress performance degradation due to an increase in the number n of data drives in nD + 1P representing a disk configuration such as RAID 5 forming a parity group. is there.
[Brief description of the drawings]
FIG. 1 is a system configuration diagram according to a first embodiment.
FIG. 2 is an explanatory diagram of an update / inconsistency position management table in the first embodiment.
FIG. 3 is a snapshot acquisition / deletion flow according to the first embodiment.
FIG. 4 is an operation flow of a mirror resynchronization subprogram in the first embodiment.
FIG. 5 is an operational flow of a disk failure recovery subprogram in the mirror phase in the first embodiment.
FIG. 6 is a non-mirror phase disk failure recovery subprogram operation flow in the first embodiment.
FIG. 7 is a disk failure recovery subprogram operation flow related to mirror destination LU recovery in the resynchronization phase in the first embodiment.
FIG. 8 is a disk failure recovery subprogram operation flow related to the mirror source LU recovery in the resynchronization phase in the first embodiment.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 100 ... Computer, 200 ... Disk array, 221 ... Snapshot management program, 222 ... Update / inconsistency position management table, 233 ... Disk failure recovery subprogram, 271-275, 281-285 ... Disk.

Claims (2)

A disk failure recovery method for an external storage device having a plurality of disks connected to a computer,
Forming a parity group using the first plurality of disks, and arranging a first logical unit on the first plurality of disks;
Using the same number of disks as the first plurality of disks, and forming a parity group using the second plurality of disks so as to form a pair with the first plurality of disks, the second plurality of disks Having a second step of placing a second logical unit on the disk;
A third step of writing the data write request data to both the first logical unit and the second logical unit when the computer makes a data write request;
When there is a data write request from the computer, the first logical unit and the second logical unit are configured so as to perform the data write to the first logical unit and not to the second logical unit. A fourth step for controlling the writing of data to the logical unit;
A fifth step of storing the position on the disk and the update record at which the data was written when the data was written to the first logical unit in the fourth step;
A sixth step of identifying a disk in the first plurality of disks that is paired with the replaced disk when one of the second plurality of disks is replaced with a new disk; ,
A seventh step of copying the data at the location to the new disc location corresponding to the location of the identified disc if there is no update record at the location of the data on the identified disc;
If the update record exists at the position of the data on the specified disc, the data corresponding to the position of the new disc is transferred to the position of the new disc corresponding to the position of the second plurality of discs. A disk failure recovery method for an external storage device, comprising an eighth step of obtaining and recording from other disk groups other than the replaced disk by parity calculation. A disk failure recovery method for an external storage device having a plurality of disks connected to a computer,
Forming a parity group using the first plurality of disks, and arranging a first logical unit on the first plurality of disks;
Using the same number of disks as the first plurality of disks, and forming a parity group using the second plurality of disks so as to form a pair with the first plurality of disks, the second plurality of disks Having a second step of placing a second logical unit on the disk;
A third step of writing the data write request data to both the first logical unit and the second logical unit when the computer makes a data write request;
When there is a data write request from the computer, the first logical unit and the second logical unit are configured so as to perform the data write to the first logical unit and not to the second logical unit. A fourth step for controlling the writing of data to the logical unit;
A fifth step of storing the position on the disk and the update record at which the data was written when the data was written to the first logical unit in the fourth step;
A sixth step of identifying a disk in the second plurality of disks that is paired with the replaced disk if one of the first plurality of disks is replaced with a new disk; ,
If there is no update record at the position of the data on the one exchanged disk , the data on the position of the specified disk paired with the position on the one exchanged disk is changed to the position. A seventh step of copying to the location of the new disk corresponding to
If there is the update record at the position of the data on the one exchanged disk , the data corresponding to the position of the new disk is transferred to the position of the new disk corresponding to the position. A disk failure recovery method for an external storage device, comprising: an eighth step of obtaining and recording by a parity calculation from a disk group other than the replaced one of the disks among the disks.

Images