JP2011008583A - Disk array controller and disk array device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the deterioration in I/O performance in reconfiguration in an RAID6.SOLUTION: Two or more first storage devices stores data. A second storage device stores a first parity calculated by a first calculation method. A third storage device stores a second parity calculated by a second calculation method. When the failure of the third storage device is detected, a parity replacement functioning part 212 calculates the second parity by the second calculation method from the data stored in the first storage device. The parity replacement functioning part 212 overwrites the calculated second parity in the second storage device. When the third storage device is replaced with a fourth storage device, the reconfiguration functioning part 213 writes the second parity read from the second storage device in the fourth storage device. The reconfiguration functioning part 213 calculates the first parity by the first calculation method from the data read from the first storage device, and writes it in the second storage device.

Description

  The present invention relates to a disk array control device and a disk array device.

  In general, RAID (Redundant Arrays of Inexpensive Disks) is widely used as a method for improving the reliability of a faulty storage device associated with a computer or the like. Several levels of this RAID are standardized, and names of RAID0 to RAID6 are given respectively.

  Of these, RAID 1 and RAID 5 have been frequently used to improve reliability.

  RAID 1 is a method of completely duplicating data using two HDDs (Hard Disk Drives) as one set, and is called mirroring. RAID 1 is highly reliable and can be processed at high speed because it does not require calculation of parity (data) used in RAID 5 or the like. On the other hand, in RAID1, the available effective capacity is half of the physical capacity.

  RAID 5 stores (saves) data (data blocks) in (n−1) of the n HDDs, where n (n is an integer of 3 or more) HDDs as a set, and the remaining In this method, the parity of the data block is stored in one unit. This parity is calculated from data stored in HDDs other than the HDD in which the parity is stored. By using this parity, for example, even if one of the n HDDs fails, it is possible to restore (recalculate) the data block stored in the HDD.

  In RAID5, the effective capacity that can be used is (n-1) / n with respect to the physical capacity, so that more areas can be used compared to RAID1 described above. On the other hand, RAID5 needs to read data blocks from all HDDs and calculate parity when writing data, and thus write processing (writing) may be slow. In practice, the calculation of parity itself is a simple XOR operation (exclusive OR), so the influence is small, but it is a write process that data blocks must be read from all HDDs in order to calculate the parity. It is often the main cause of slowing down.

  Furthermore, in recent years, unlike RAID 5 (single parity) using one parity, RAID 6 (double parity) using two parities has gradually become widespread.

  RAID 6 stores data (data blocks) in (m-2) of m HDDs, with m (m is an integer of 4 or more) HDDs as a set, and the remaining two HDDs. One of these is a method of storing parity (XOR parity) similar to RAID5 and the other is storing a parity independent of the XOR parity. According to this method, even if any two HDDs out of the m HDDs fail, data can be completely restored.

  Hereinafter, of the two parities used in RAID 6, the XOR parity similar to RAID 5 is referred to as P parity, and the parity specific to RAID 6 (that is, parity independent of XOR parity) is referred to as Q parity. Here, 2D-XOR RAID 6 using two XOR parities is not assumed.

  By the way, the above-described Q parity needs to have a mathematically independent meaning from the P parity. That is, a binary simultaneous equation consisting of data blocks and two parities stored in (m−2) HDDs must have a solution.

  As a practical requirement, the range of the calculated parity (P and Q) values needs to be the same as the value range of each data block. That is, a 1-byte parity needs to be generated (calculated) from, for example, a 1-byte data block stored in (m−2) HDDs.

  For this reason, Q parity cannot be realized (calculated) by normal arithmetic four arithmetic operations.

  Therefore, in the actual RAID 6, the calculation of Q parity satisfying all such requirements is realized by defining an operation space called a Galois field. The four arithmetic operations in the Galois field cannot be calculated by the arithmetic four arithmetic operation circuit of a general processor. For this reason, it is common to store the four arithmetic operation tables in a memory or the like and calculate the Q parity by referring to the four arithmetic operation tables.

  As described above, since the calculation of Q parity is complicated, it takes much processing time compared with the calculation of P parity. However, during normal write processing (operation), the write time to the HDD is sufficiently longer than the Q parity calculation time, so that, for example, when there is a memory cache, the response is somewhat slow. , It will not be a big problem.

  As a technology related to RAID6, if an inconsistency is detected in the consistency check even if one of the multiple disk devices that make up RAID6 is in a failed state, the data illegal location is identified. Techniques that can be performed (hereinafter referred to as prior art) are disclosed (see, for example, Patent Document 1). According to this prior art, data fraud due to missing data write is detected using the time stamp and bitmap data recorded in the sector redundancy area.

JP 2008-71297 A

  By the way, when an HDD constituting a RAID fails, the RAID can be configured again by replacing the failed HDD (hereinafter referred to as a failed HDD) with a new HDD (hereinafter referred to as a new HDD). Is done. In this case, work (hereinafter referred to as reconfiguration) for restoring data or parity stored in the failed HDD to the new HDD is required.

  This reconfiguration is an operation of recalculating data or parity to be written to the new HDD from the data blocks and parity stored in the remaining HDD (HDD other than the failed HDD). For this reason, reconfiguration takes a very long time, and many computer resources (for example, CPU power and memory area) are used.

  In particular, in the above-described reconstruction in RAID 6, when the Q parity is stored in the failed HDD, it is necessary to recalculate the Q parity. As described above, since the calculation of Q parity requires a lot of processing time, more computer resources are required as compared with the above-described reconstruction in RAID 1 and 5, and the I / O performance is further adversely affected. Will be affected. That is, the reconstruction in RAID 6 has a large decrease in I / O performance compared to RAID 1 and 5.

  Therefore, in a system that needs to provide a certain level of performance to the user, such as a server for real-time data distribution, it is possible to reduce a decrease in I / O performance during reconfiguration in RAID 6. Necessary.

  Accordingly, an object of the present invention is to provide a disk array control device and a disk array device that can alleviate a decrease in I / O performance during reconfiguration in RAID6.

  According to one aspect of the present invention, two or more first storage devices that store data, and first parity data that is used to restore the data stored in the first storage device. A second storage device for storing first parity data calculated by the first calculation method from data stored in the first storage device, and the second storage device stored in the first storage device. Second parity data that is different from the first parity data used to restore the data stored in the first storage device, and is more complicated than the first calculation method from the data stored in the first storage device. A disk array controller connected to a third storage device that stores second parity data calculated by the second calculation method is provided. When the third storage device fails, the disk array control device detects a failure of the third storage device, and when a failure of the third storage device is detected, the disk array control device First calculation means for calculating second parity data from the data stored in one storage device by the second calculation method, and second parity data calculated by the first calculation means, When the overwriting means for overwriting the first parity data stored in the second storage device and the third storage device in which the failure is detected are replaced with a fourth storage device, the first storage device Read means for reading data stored in the storage device and second parity data overwritten in the second storage device; and second parity data read from the second storage device in the fourth First writing means for writing to the storage device; second calculation means for calculating first parity data from the data read from the first storage device by the first calculation method; and Second writing means for writing the first parity data calculated by the calculating means to the second storage device.

  The present invention makes it possible to reduce a decrease in I / O performance during reconfiguration in RAID6.

1 is a block diagram showing a configuration of a disk array device according to an embodiment of the present invention. FIG. 2 is a block diagram mainly showing a functional configuration of a RAID controller 20 shown in FIG. 1. The figure for demonstrating the area | region 301 which can store the replacement completion information on HDD30-1, 30-2, ..., 30-6 shown in FIG. FIG. 6 is a diagram showing an example of a data structure of HDDs 30-1, 30-2,..., 30-6 included in the disk array device 10 shown in FIG. 10 is a flowchart showing a processing procedure of the RAID controller 20 when a failure of one of the HDDs 30-1, 30-2,..., 30-6 is detected. The flowchart which shows the process sequence of a parity replacement process. The flowchart which shows the process sequence of I / O process at the time of 1 unit call. The flowchart which shows the process sequence of a reconstruction process. The flowchart which shows the process sequence of a stripe reconstruction process. The figure for demonstrating concretely operation | movement of the RAID controller 20 when failure of one of HDD30-1, 30-2, ..., 30-6 is detected.

  Embodiments of the present invention will be described below with reference to the drawings.

  FIG. 1 is a block diagram showing the configuration of the disk array device according to the present embodiment. As shown in FIG. 1, the disk array device 10 includes a RAID controller (disk array control device) 20 and four or more hard disk drives (HDDs) 30-1, 30-2,. k is an integer of 4 or more. In the following description, the disk array device 10 will be described as having six HDDs (that is, HDDs 30-1, 30-2,..., 30-6).

  The RAID controller 20 acquires an I / O request (write request or read request) transmitted from the outside, and writes data to the HDDs 30-1, 30-2,..., 30-6 according to the I / O request. / Execute read processing.

  In the present embodiment, it is assumed that the RAID controller 20 exists as hardware. Specifically, the format is a RAID card on PCI-Express. The RAID controller 20 includes an independent RAID controller processor 21 and a RAID controller memory 22. The RAID controller memory 22 stores a program (RAID program) executed by the RAID controller processor 21. The RAID controller processor 21 implements a RAID function by executing a RAID program stored in the RAID controller memory 22.

  Note that a data cache may be mounted in the RAID controller memory 22 in order to improve responsiveness.

  The HDDs 30-1, 30-2,..., 30-6 included in the disk array device 10 constitute, for example, a RAID (Redundant Array of Inexpensive Disks) 6.

  In this RAID 6, data (data block) is stored in four HDDs (first physical storage devices) out of the HDDs 30-1, 30-2,. One HDD (second physical storage device) includes first parity data (hereinafter referred to as P parity) calculated from the data block by an XOR operation (first calculation method), and the other HDD (third The second physical storage device) stores second parity data independent of the P parity (hereinafter referred to as Q parity). The Q parity has a mathematically independent meaning from the P parity, and is calculated by a Galois operation (second calculation method). The Galois operation is realized by defining an operation space called a Galois field, and is calculated by referring to, for example, a four arithmetic operation table prepared in advance. The Galois operation is a more complicated calculation method than the above XOR operation, and requires more processing time (processing amount) than the XOR operation.

  Here, since the disk array device 10 has six HDDs, data (data other than P parity and Q parity) is stored in four of the HDDs, but in RAID 6, two or more HDDs are stored. The data needs to be stored.

  Further, data blocks stored in four HDDs among the HDDs 30-1, 30-2,..., 30-6, and P parity and Q parity (a set) calculated from the data blocks are called stripes. . The HDDs 30-1, 30-2,..., 30-6 have a plurality of stripes.

  In the present embodiment, 2D-XOR RAID 6 using two XOR parities is not assumed.

  FIG. 2 is a block diagram mainly showing a functional configuration of the RAID controller 20 shown in FIG. As shown in FIG. 2, the RAID controller 20 includes a RAID function unit 211, a parity replacement function unit 212, and a reconfiguration function unit 212. In the present embodiment, these units 211 to 213 are realized by the RAID controller processor 21 illustrated in FIG. 1 executing a RAID program stored in the RAID controller memory 22. This RAID program can be stored in advance in a computer-readable storage medium and distributed. The RAID program may be downloaded to the computer 10 via a network, for example.

  The RAID controller 20 also includes a replacement completion information storage unit 221. In this embodiment, the replacement completion information storage unit 221 is stored in the RAID controller memory 22.

  The RAID function unit 211 detects the failure when one of the HDDs 30-1, 30-2,. When a failure of one of the HDDs 30-1, 30-2,..., 30-6 is detected, the RAID function unit 211 displays an I / O device such as a monitor or buzzer connected to the RAID controller 21. The failure is notified to the user via (not shown). As a result, the user can recognize that one of the HDDs 30-1, 30-2,..., 30-6 has failed, and can replace the failed HDD at an appropriate time.

  The RAID function unit 211 has a function of a normal RAID controller as will be described later.

  When a failure is detected by the RAID function unit 211, the parity replacement function unit 212 has stripes (HDDs 30-1, 30-2,..., 30-6 included in the HDDs 30-1, 30-2,. Parity replacement processing is executed every 30-6 stripes). In the parity replacement process, Q parity is recalculated by Galois operation from each data block of the stripe for the stripe in which Q parity was stored in the HDD where the failure was detected (that is, the stripe in which Q parity was lost) Is done. The Q parity is overwritten at the position where P is stored in the stripe where the Q parity is lost. When parity replacement processing is executed for each stripe, information (replacement completion information) indicating that parity replacement processing has been completed for that stripe (that is, the stripe for which parity replacement processing has been completed) is stored in the replacement completion information. Stored in the unit 221.

  The reconfiguration function unit 213 replaces the HDD (one of the HDDs 30-1, 30-2,..., 30-6) in which a failure is detected by the RAID function unit 211 by the user, and instructs to reconfigure RAID 6. If so, a reconstruction process (stripe reconstruction process) is executed for each stripe.

  The reconfiguration function unit 213 reads the data block and Q parity from the HDD other than the replaced HDD (new HDD) for the stripe on which the parity replacement process has been executed. The reconfiguration function unit 213 writes the read Q parity in the replaced new HDD.

  Further, the reconstruction function unit 213 calculates a P parity from the read data block by an XOR operation, and the calculated P parity is stored in the position where the original P parity is stored (that is, the Q parity is overwritten). To the position).

  The replacement completion information storage unit 221 stores information indicating that the parity replacement processing has been completed for each stripe as described above (replacement completion information). In the replacement completion information storage unit 221, for example, a 1-bit determination flag is stored as replacement completion information for each stripe of the HDDs 30-1, 30-2,. As a result, it is possible to recognize a stripe for which parity replacement processing has been completed.

  When the parity replacement process is performed sequentially, only the address of the stripe (replacement completion address) for which the parity replacement process has been completed may be stored. This indicates that the parity replacement processing has been completed up to the stripe corresponding to the replacement completion address.

  In addition, since the system including the disk array device 10 may be temporarily shut down, it is possible to prepare an area for storing replacement completion information in addition to the replacement completion information storage unit 221 on the RAID controller memory 22. preferable. Specifically, as shown in FIG. 3, for example, an area 301 that can store replacement completion information may be prepared on the HDDs 30-1, 30-2,..., 30-6.

  In the example shown in FIG. 3, the area 301 in which the replacement completion information can be stored is provided in the HDDs 30-1, 30-2, and 30-3 among the HDDs 30-1, 30-2,. That is, the replacement completion information includes three HDDs (here, HDDs 30-1, 30-2 and 30) so as to withstand, for example, up to two of the HDDs 30-1, 30-2,. -3). Further, RAID 6 may be applied to the replacement completion information itself so that it can withstand failure of up to two of the HDDs 30-1, 30-2,.

  Note that the area 301 in which the replacement completion information shown in FIG. 3 can be stored is an area of a plurality of stripes (that is, data blocks, P parity and Q parity) included in the HDDs 30-1, 30-2,. This is a different area.

  Here, with reference to FIG. 4, an example of the data structure of the HDDs 30-1, 30-2,..., 30-6 included in the disk array device 10 shown in FIG. Data blocks, P parity, and Q parity are stored in the HDDs 30-1, 30-2,..., 30-6. The HDDs 30-1, 30-2,..., 30-6 have a plurality of stripes (a set of four data blocks, P parity and Q parity). In the following description, the first stripe in the HDDs 30-1, 30-2,. The same applies to the second and subsequent stripes.

  In FIG. 4, Dij indicates the jth data block in the stripe with the stripe number i. Pi indicates the P parity in the stripe with the stripe number i. Qi indicates the Q parity in the stripe with the stripe number i.

  In the example shown in FIG. 4, in the stripe of stripe number 1, D11 for HDD 30-1, D12 for HDD 30-2, D13 for HDD 30-3, D14 for HDD 30-4, P1 for HDD 30-5, Q1 for HDD 30-6. Is stored.

  In the stripe with the stripe number 2, Q2 is stored in the HDD 30-1, D21 is stored in the HDD 30-2, D22 is stored in the HDD 30-3, D23 is stored in the HDD 30-4, D24 is stored in the HDD 30-5, and P1 is stored in the HDD 30-6. .

  In the stripe of stripe number 3, P3 is stored in HDD 30-1, Q3 is stored in HDD 30-2, D31 is stored in HDD 30-3, D32 is stored in HDD 30-4, D33 is stored in HDD 30-5, and D34 is stored in HDD 30-6. .

  Although the stripe numbers 1 to 3 have been described here, the same applies to the stripes after the stripe number 4.

  In the example shown in FIG. 4, HDDs that store P parity and Q parity are arranged so as to be shifted by one for each stripe. In this embodiment, how to arrange the P parity and the Q parity is not essential, and if the position of the P parity and the Q parity (HDD number) can be uniquely determined from the stripe number, the arrangement shown in FIG. It doesn't matter.

  Although not shown in FIG. 4, the HDD 30-1, 30-2,..., 30-6 have an area for storing the above replacement completion information (area 301 shown in FIG. 3). Yes.

  Next, referring to the flowchart of FIG. 5, the RAID controller 20 when a failure of one of the HDDs 30-1, 30-2,..., 30-6 configuring RAID 6 is detected by the RAID function unit 211. The processing procedure will be described. Here, a description will be given assuming that a failure of the HDD 30-2 among the HDDs 30-1, 30-2,.

  When a failure of the HDD 30-2 is detected by the RAID function unit 211, the fact is notified to the user via the I / O device.

  When the failure of the HDD 30-2 is detected by the RAID function unit 211, the parity replacement function unit 212 included in the RAID controller 20 resets the replacement completion information stored in the replacement completion information storage unit 221 (step S1). . This replacement completion information is information indicating a stripe on which a parity replacement process described later has been executed.

  Next, the parity replacement function unit 212 determines whether or not there is an external I / O request, that is, whether or not an externally transmitted I / O request is acquired in the RAID controller 20 (step S2).

  The external I / O request includes a read request that is a request for reading data from the HDDs 30-1, 30-2,..., 30-6 constituting the RAID 6, and the HDDs 30-1, 30-2,. , 30-6, a write request which is a request for writing data is included. The I / O request includes the address of the data block to be accessed (read or write).

  When it is determined that there is no I / O request (NO in step S2), the parity replacement function unit 212 determines the HDDs 30-1 and 30-2 based on the replacement completion information stored in the replacement completion information storage unit 221. ,..., 30-6 determines whether parity replacement processing has been executed for all of the stripes (step S3). In other words, the parity replacement function unit 212 determines whether or not the parity replacement processing has been executed up to the final stripe in the HDDs 30-1, 30-2,.

  When it is determined that the parity replacement process has not been executed for all the stripes in the HDDs 30-1, 30-2,..., 30-6 (NO in step S3), the parity replacement function unit 212 performs the parity replacement process. The following steps S4 and S5 are executed for one of the stripes that has not been executed. Hereinafter, a stripe to be processed is referred to as a replacement target stripe.

  The parity replacement function unit 212 executes parity replacement processing for the replacement target stripe (step S4). In the parity replacement process, recalculation of the Q parity lost due to the failure of the HDD 30-2 detected by the RAID function unit 211 is executed. The recalculated Q parity is overwritten at the position of the P parity in the replacement target stripe. Details of the parity replacement processing will be described later.

  Next, the parity replacement function unit 212 stores replacement completion information indicating that the parity replacement processing has been completed for the replacement target stripe in the replacement completion information storage unit 221. That is, the parity replacement function unit 212 updates the replacement completion information stored in the replacement completion information storage unit 221 (step S5).

  On the other hand, when it is determined that there is an I / O request (YES in step S2), the RAID function unit 211 and the parity replacement function unit 212 execute the I / O process at the time of failure of one unit based on the I / O request. (Step S6). This one-unit failure I / O process is executed for a stripe to be accessed in an I / O request (a stripe corresponding to an address included in the I / O request). Details of the I / O process when one unit fails will be described later.

  When the processing in step S5 or step S6 described above is executed, it is determined whether or not a reconfiguration request has been detected (step S7). This reconfiguration request is sent to the HDD 30-2 (third physical storage device) by the user when a failure of the HDD 30-2 detected by the RAID function unit 211 is notified to the user. This is detected when a reconfiguration of RAID 6 is instructed by the user.

  When it is determined that a reconfiguration request has been detected (YES in step S7), the reconfiguration function unit 213 detects an HDD other than the HDD 30-2 in which a failure has been detected by the RAID function unit 211 (that is, HDD 30-1, 30-). 3,..., 30-6) and a new HDD (new HDD) replaced with the HDD 30-2, a process of reconfiguring RAID 6 (hereinafter referred to as a reconfiguration process) is executed (step S8). Details of the reconstruction process will be described later.

  Next, the procedure of the parity replacement process (the process of step S4 shown in FIG. 5) described above will be described with reference to the flowchart of FIG.

  First, the parity replacement function unit 212 loses Q parity in the stripe (the replacement target stripe described in FIG. 5) that is the target of the parity replacement process due to, for example, a failure of the HDD 30-2 detected by the RAID function unit 211. It is determined whether or not there is (step S11). That is, the parity replacement function unit 212 stores Q parity in the HDD 30-2 in which a failure is detected among a plurality of data blocks (here, four data blocks), P parity, and Q parity in the replacement target stripe. It is determined whether or not.

  4, the positions of the P parity and Q parity (HDD number) can be uniquely determined from the stripe number of the replacement target stripe. Therefore, if the Q parity position (HDD number) that can be determined from the stripe number of the replacement target stripe is the HDD 30-2 in which a failure is detected by the RAID function unit 211, it is determined that the Q parity is missing.

  Specifically, in the HDD 30-1, 30-3,..., 30-6 shown in FIG. 4 described above, when the replacement target stripe is, for example, the stripe with stripe number 3, the Q parity is set due to the failure of the HDD 30-2. Judged as missing.

  When it is determined that the Q parity is missing (YES in step S11), the parity replacement function unit 212 stores the data block in the replacement target stripe among the HDDs 30-1, 30-3,. The data block (that is, four data blocks) is read from the HDD (first physical storage device).

  The parity replacement function unit 212 recalculates the Q parity from the read data block by Galois calculation (step S12).

  Next, the parity replacement function unit 212 calculates (specifies) the position (HDD number) where the P parity is stored in the replacement target stripe (step S13).

  The parity replacement function unit 212 overwrites the recalculated Q parity at the position where the P parity is stored in the replacement target stripe (step S14).

  If it is determined in step S11 that the Q parity is not lost, the parity replacement process is terminated because it is not necessary to recalculate the Q parity.

  Next, with reference to the flowchart of FIG. 7, the processing procedure of the above-mentioned I / O processing at the time of one unit failure (processing of step S6 shown in FIG. 5) will be described. This one-unit failure I / O process is performed after a failure of one of the HDDs 30-1, 30-2,..., 30-6 (for example, the HDD 30-2) is detected by the RAID function unit 211. This is executed when an externally transmitted I / O request is acquired before a reconfiguration request is detected.

  As described above, at the time of one unit failure I / O processing, the RAID function unit 211 detects a failure of one of the HDDs 30-1, 30-2,. Therefore, in each of the stripes of the HDDs 30-1, 30-2,..., 30-6, one of the data block, P parity, and Q parity is missing.

  The parity replacement function unit 212 determines whether the external I / O request is a read request (step S21).

  When it is determined that the external I / O request is not a read request, that is, a write request (NO in step S21), the parity replacement function unit 212 determines the stripe (to be accessed by the I / O request) Hereinafter, it is determined whether or not the Q parity is missing in the access target stripe) (step S22). In other words, the parity replacement function unit 212 includes a plurality of data blocks (here, four data blocks), P parity, and Q parity in the access target stripes of the HDDs 30-1, 30-2,. Then, it is determined whether or not the Q parity is stored in the HDD 30-2 where the failure is detected.

  The access target stripe can be specified from the I / O request. Further, the position (HDD number) of the parity (P parity and Q parity) in the access target stripe can be determined (calculated) from the specified access target stripe (its stripe number). Thereby, the parity replacement function unit 212 can determine whether or not the access target stripe is a stripe in which the Q parity is lost.

  When it is determined that Q parity is missing in the access target stripe (YES in step S22), the parity replacement function unit 212 performs the access based on the replacement completion information stored in the replacement completion information storage unit 221. It is determined whether or not parity replacement processing has been executed for the target stripe (step S23).

  If it is determined that the parity replacement process is executed for the access target stripe (YES in step S23), the Q parity is stored (overwritten) at the position where the P parity is originally stored in the access target stripe. It is in a state.

  In this case, the parity replacement function unit 212 writes data blocks (write data) to the HDDs 30-1, 30-2,..., 30-6 in response to an I / O request (here, a write request) (step S24). .

  Next, the parity replacement function unit 212 recalculates the Q parity from a plurality of data blocks in the access target stripe by Galois operation, and overwrites the recalculated Q parity (step S25). That is, the recalculated Q parity is overwritten at the position where the P parity should originally be stored in the access target stripe.

  On the other hand, if it is determined in step S21 that the I / O request is a read request, if it is determined in step S22 that Q parity is not lost in the access target stripe, or if it is determined in step S23 that parity replacement is performed for the access target stripe. If it is determined that the process is not being executed, the RAID function unit 211 executes the same I / O process (operation) as that of normal RAID 6.

  Specifically, a case will be described in which it is determined in step S21 that the I / O request is a read request. At this time, if the data block is missing due to, for example, a failure of the HDD 30-2 detected by the RAID function unit 211, the RAID function unit 211 detects the missing data block from the P parity in the access target stripe. And the data block is returned as a response to the I / O request. If P parity or Q parity is missing due to the failure of the HDD 30-2, the data block in the access target stripe is returned.

  A case will be described in which it is determined in step S22 that no Q parity is lost in the access target stripe. At this time, when the P parity is missing in the access target stripe, the RAID function unit 211 writes the data block in response to the I / O request (here, the write request) as in step S24, and The Q parity in the stripe is recalculated, and the recalculated Q parity is overwritten (updated to the recalculated Q parity). On the other hand, when the data block is missing in the access target stripe, the RAID function unit 211 re-creates the P parity and Q parity from the data block (data block written in response to the I / O request) in the access stripe. Calculate and overwrite the recalculated P parity and Q parity (update to the recalculated P parity and Q parity).

  If it is determined in step S23 that the parity replacement process has not been executed for the access target stripe, the RAID function unit 211 performs a data block in response to an I / O request (here, a write request) as in step S24. And recalculates the P parity in the access target stripe and overwrites the recalculated P parity (updates to the recalculated P parity).

  Next, the processing procedure of the above-described reconstruction process (the process of step S8 shown in FIG. 5) will be described with reference to the flowchart of FIG. This reconfiguration process is executed when the HDD in which the failure is detected by the RAID function unit 211 is replaced with a new HDD. Here, it is assumed that a failure has been detected in the HDD 30-2 among the HDDs 30-1, 30-2,..., 30-6, and the HDD 30-2 has been replaced with a new HDD (new HDD).

  First, the reconstruction function unit 213 resets a reconstruction completion position indicating a position (stripe) where the reconstruction process is completed (step S31). As the reconstruction completion position, for example, a stripe number in the HDDs 30-1, 30-2,..., 30-6 is used.

  Next, the reconfiguration function unit 213 determines whether or not there is an external I / O request, that is, whether or not an externally transmitted I / O request is acquired in the RAID controller 20 (step S32). This I / O request includes a read request or a write request.

  When it is determined that there is an I / O request (YES in step S32), the reconfiguration function unit 213 executes reconfiguration I / O processing based on the I / O request (step S33).

  Here, the I / O processing at the time of reconfiguration will be briefly described. In this reconfiguration I / O processing, the reconfiguration function unit 213 determines whether or not the stripe (access target stripe) to be accessed by the I / O request has been reconfigured based on the reconfiguration completion position. Determine whether.

  When it is determined that the access target stripe has been reconfigured, the reconfiguration function unit 213 executes the same processing as that of normal RAID6. Specifically, when the I / O request is a read request, the data block in the access target stripe is returned as a response to the request. When the I / O request is a write request, the data block is written in response to the request, and the P parity and Q parity in the access target stripe are updated.

  On the other hand, when it is determined that the access target stripe has not been reconfigured, the same processing as the one-device failure I / O processing shown in FIG. 7 described above is executed.

  When the reconstruction function unit 213 executes the reconstruction I / O process, the process returns to the above-described step S32 and the process is repeated.

  If it is determined in step S32 that there is no I / O request, the reconstruction function unit 213 performs the following steps S34 to S1 for one of the stripes for which reconstruction processing has not been performed based on the reconstruction completion position. The process of S36 is executed. In this case, the reconstruction function unit 213 performs processing for the stripe with the smallest stripe number among the stripes for which reconstruction processing has not been performed. Hereinafter, the stripe to be processed is referred to as a reconstruction target stripe.

  Based on the replacement completion information stored in the replacement completion information storage unit 221, the reconstruction function unit 213 determines whether or not the parity replacement process has been completed for the reconstruction target stripe (step S34). As described in the parity replacement process, a stripe for which the parity replacement process has been completed always has a Q parity.

  When it is determined that the parity replacement process has been completed for the reconstruction target stripe (YES in step S34), the reconstruction function unit 213 reconfigures the reconstruction target stripe (hereinafter referred to as a stripe reconstruction process). ) Is executed (step S35). Details of the stripe reconstruction processing will be described later.

  On the other hand, when it is determined that the parity replacement process has not been performed for the reconstruction target stripe (NO in step S34), the RAID function unit 211 performs the reconstruction process similar to the normal RAID 6 for the reconstruction target stripe. (Step S36). In this case, the reconstruction target stripe is in a state in which any of the data block, P parity, and Q parity is missing.

  Specifically, when a data block is missing in the reconstruction target stripe, the RAID function unit 211 determines the data block (data block other than the missing data block) in the reconstruction target stripe and the P parity from the data block. The missing data block is restored, and the restored data block is written to the new HDD (location corresponding to the reconfigured stripe). In this case, although it is possible to restore the data block using Q parity instead of P parity, it is preferable to use P parity in order to reduce the amount of calculation processing.

  Further, when the P parity is missing in the reconstruction target stripe, the RAID function unit 211 recalculates the P parity from the data block in the reconstruction target stripe by the XOR operation, and renews the recalculated P parity. Write to HDD.

  Further, when the Q parity is missing in the reconstruction target stripe, the RAID function unit 211 recalculates the Q parity from the data block in the reconstruction target stripe by Galois operation, and renews the recalculated Q parity. Write to HDD.

  When the parity replacement process is not completed for the reconstruction target stripe, the reconstruction process similar to the normal RAID 6 as described above is executed.

  When the process of step S35 or step S36 described above is executed, the reconstruction function unit 213 determines whether the reconstruction process has been performed up to the final stripe (stripe number is the last stripe) based on the reconstruction completion position. Is determined (step S37).

  When it is determined that the reconstruction process has not been executed up to the final stripe (NO in step S37), the reconstruction function unit 213 updates the reconstruction completion position (step S38). That is, the reconfiguration function unit 213 updates the reconfiguration completion position with the stripe number of the reconfiguration target stripe for which the processing in steps S34 to S36 described above has been executed.

  If the process of step S38 is performed, it will return to above-mentioned step S32 and a process will be repeated. As described above, when the processing is repeated after returning to step S32, the processing of step S34 to step S36 is executed with the next stripe at the reconstruction completion position (the stripe) as the reconstruction target stripe.

  On the other hand, when it is determined that the reconstruction process has been executed up to the final stripe (YES in step S37), the reconstruction process is terminated. When the reconfiguration process ends, the disk array device 10 returns to the normal process.

  Next, the procedure of the stripe reconstruction process (the process of step S35 shown in FIG. 8) will be described with reference to the flowchart of FIG. This stripe reconstruction process is executed for the stripe (reconstruction target stripe) for which the parity replacement process has been completed as described above. Note that the reconstruction target stripe always has Q parity by the parity replacement process described above.

  The reconstruction function unit 213 determines whether or not the Q parity is lost in the reconstruction target stripe due to, for example, a failure of the HDD 30-2 detected by the RAID function unit 211 (step S41). In this case, the reconstruction function unit 213 calculates (determines) the position (HDD number) of the parity (P parity and Q parity) from the stripe number of the reconstruction target stripe. Thereby, it is determined whether or not the Q parity is stored in the HDD 30-2 in which the failure is detected by the RAID function unit 211 in the reconfiguration target stripe (that is, the Q parity is lost due to the failure).

  Assume that it is determined that Q parity is missing in the reconstruction target stripe (YES in step S41). In this case, in the stripe to be reconfigured, the parity replacement process is executed for the stripe with the missing Q parity, so the recalculated Q parity is overwritten at the position where the P parity should originally be stored. It is in a state.

  Therefore, the reconfiguration function unit 213 reads the Q parity overwritten at the position where the P parity is originally stored in the reconfiguration target stripe by the parity replacement process, and writes the Q parity to the new HDD (step S42).

  Next, the reconstruction function unit 213 reads all data blocks (here, four data blocks) in the reconstruction target stripe. The reconstruction function unit 213 recalculates the P parity by the XOR operation from the read data block (step S43).

  The reconstruction function unit 213 writes the recalculated P parity in the position where the P parity should originally be stored in the reconstruction target stripe (that is, the position where the Q parity is stored) (step S44).

  On the other hand, when it is determined that the Q parity is not lost in the reconfiguration target stripe (NO in step S41), the reconfiguration function unit 213 detects the reconfiguration target stripe due to the HDD failure detected by the RAID function unit 211. In step S45, it is determined whether or not the P parity is lost. In this case, the reconstruction function unit 213 can determine whether or not the P parity is lost, in the same manner as the determination process of whether or not the Q parity is lost.

  Assume that it is determined that the P parity is missing in the reconstruction target stripe (YES in step S45). In this case, the P parity is lost in the reconstruction target stripe.

  In this case, the reconstruction function unit 213 reads all data blocks in the reconstruction target stripe. The reconstruction function unit 213 recalculates the P parity by the XOR operation from the read data block (step S46).

  Next, the reconfiguration function unit 213 writes the recalculated P parity in the new HDD (step S47).

  On the other hand, when it is determined that the P parity is not missing in the reconstruction target stripe (NO in step S45), that is, when it is determined that one of the data blocks is missing in the reconstruction target stripe, the reconstruction is performed. The functional unit 213 recalculates the missing data block from the data block (data block other than the missing data block) and the P parity in the reconstruction target stripe (step S48). For recalculation of missing data blocks, Q parity in the reconstruction target stripe may be used instead of P parity.

  Next, the reconstruction function unit 213 writes the recalculated data block in the new HDD (step S49).

  As described above, by executing the stripe reconstruction process on the reconstruction target stripe, the Q parity is already recalculated in the parity replacement process described above in the reconstruction target stripe in which the Q parity is lost. It is possible to omit Q parity calculation processing, which has a larger processing amount (processing time) than P parity calculation processing.

  Here, referring to FIG. 10, the operation of the RAID controller 20 when a failure of one of the HDDs 30-1, 30-2,..., 30-6 is detected by the RAID function unit 211 (mainly the packet The replacement process and stripe reconstruction process) will be specifically described.

  Here, a description will be given assuming that a failure of the HDD 30-2 among the HDDs 30-1, 30-2,..., 30-6 included in the disk array device 10 shown in FIG.

  In the example shown in FIG. 10, in the stripes with stripe numbers 1, 2, 4, 5 and 6 (the first, second, fourth, fifth and sixth stripes from the top), the data block or P parity is changed due to the failure of the HDD 30-2. It is missing. On the other hand, in the stripe with the stripe number 3 (the third stripe from the top), the Q parity is lost due to the failure of the HDD 30-2.

  Note that, according to the number of HDDs and the data arrangement rule shown in FIG. 10, for example, for stripes after stripe number 6 as well, a stripe in which Q parity is lost due to the failure of the HDD 30-2 can be calculated. In the example shown in FIG. 10, it can be calculated that the Q parity is lost due to the failure of the HDD 30-2 in every six stripes, that is, in stripes with stripe numbers 6N + 3 (N = 0, 1, 2,...). Similarly, in the stripe with the stripe number 6N + 4, it can be calculated that the P parity is missing due to the failure of the HDD 30-2.

  Here, it is assumed that the parity replacement process is executed with the stripe of stripe number 3 as the replacement target stripe before the failed HDD 30-2 is replaced with a new HDD (step S51). In this case, as shown in FIG. 10, the data block in the stripe of stripe number 3 (here, D31, D32, D33 and D34 stored in HDDs 30-3, 30-4, 30-5 and 30-6) Q parity (Q3) is recalculated by Galois calculation. This recalculated Q3 is overwritten at the position (here, HDD 30-1) where the P parity (P3) is to be stored in the stripe of stripe number 3.

  In the parity replacement process in which a stripe other than the stripe of stripe number 3 is used as a replacement target stripe, the Q parity is not lost in the replacement target stripe, and therefore processing such as recalculation and overwriting of Q parity is not executed.

  Here, it is assumed that the failed HDD 30-2 is replaced with a new HDD 100 and the reconfiguration process is executed (step S52). In this case, in the stripe of stripe number 1 in which the data block (D12) shown in FIG. 10 is lost, the data blocks in the stripe (here, D11, D11 stored in the HDDs 30-1, 30-3, and 30-4). D13 and D14) and P parity (here, P1 stored in the HDD 30-5) are restored (recalculated), and the restored D12 is written to the new HDD 100. Since the same applies to the stripes with stripe numbers 2, 5, and 6, detailed description thereof is omitted.

  Further, in the stripe of stripe number 3 in which the Q parity (Q3) is missing, the Q parity (here, Q3) of the position (here, HDD 30-1) where the P parity is originally stored in the stripe is the new HDD 100. Is written to. Also, P parity (here, D31, D32, D33, and D34 stored in HDDs 30-3, 30-4, 30-5, and 30-6) is obtained from the data block in the stripe of stripe number 3 by XOR operation. Then, P3) is recalculated, and the P3 is written in the HDD 30-1 (position where P3 should be stored) where Q3 was overwritten. That is, even if Q3 is lost due to a failure of the HDD 30-2, the stripe can be reconfigured without recalculating Q3 during the reconfiguration process.

  Further, in the stripe of stripe number 4 in which P parity (P4) is lost, D44, D41 stored in the data blocks (here, HDDs 30-1, 30-4, 30-5 and 30-6) in the stripe. , D42 and D43), the P parity (here, P4) is recalculated by the XOR operation, and the recalculated P4 is written in the new HDD 100.

  As described above, in the present embodiment, in a stripe in which Q parity is lost due to a failure of one of a plurality of HDDs (HDDs 30-1, 30-2,. The missing Q parity is recalculated from all the data blocks by the Galois operation, and the recalculated Q parity is overwritten on the P parity (position) in the stripe. In this embodiment, when the RIAD 6 is reconfigured by replacing the HDD in which the failure is detected with a new HDD, the Q parity overwritten at the P parity position in the stripe in which the Q parity is lost due to the failure. P parity is recalculated by XOR operation from all data blocks in the stripe, written to the new HDD, and the recalculated P is stored at the position where the P parity is originally stored (position where the Q parity is overwritten). Parity is written.

  For example, in the case of a configuration in which the redundancy is one HDD such as RAID 5 (that is, a configuration that can withstand a failure up to one HDD), it can withstand a situation in which another HDD failure occurs after one failure. Since there is no room, the priority of the RAID reconfiguration work is considerably high, and it is often necessary to accept a decrease in I / O performance.

  However, since the redundancy in RAID 6 is two HDDs, there is room to withstand one more failure when one unit fails. Compared to RAID 5 and the like, RAID reconfiguration work in this case The urgency of is not necessarily high. Therefore, in many cases, replacement of a physical HDD in which a failure is detected is postponed.

  In RAI6, when a RAID is reconfigured by replacing an HDD after a failure of one HDD, it can withstand a failure of another HDD as described above, so normal I / O processing is given priority. It is efficient that the reconfiguration operation is executed when the I / O processing is not performed (or few). However, in such an operation, it takes a long time until the reconfiguration is completed. If another HDD fails during this time, an urgent reconfiguration may be necessary.

  In such an operation, according to the above-described embodiment, since the recalculation of the Q parity that requires a lot of processing time is performed before the reconfiguration, the Q parity is not calculated at the time of the reconfiguration. Instead, P parity calculation is performed.

  This P parity calculation can be executed at a higher speed and with fewer resources than the Q parity calculation. For this reason, in this embodiment, it is possible to reduce a decrease in I / O performance during reconfiguration and to shorten the time required for reconfiguration.

  In other words, in the present embodiment, not all of the reconfiguration work is performed after replacement of the HDD in which a failure is detected, but a part of the work (Q parity recalculation process requiring a lot of processing time) is performed on the HDD. Since it is executed before replacement (that is, before reconfiguration work), adverse effects on I / O performance can be reduced.

  In this embodiment, recalculation of the Q parity is executed in the parity replacement process before replacing the HDD. However, since this process may be stopped or interrupted at any time, the normal I / O process is affected. It can be distributed so that it is executed in a range not given.

  In the present embodiment, the RAID controller 20 included in the disk array device 10 has been described as being in the form of a RAID card on PCI-Express, for example, which exists as hardware on the I / O bus. The RAID controller 20 may have a different format. Specifically, the RAID controller 20 is, for example, a format in which the RAID controller 20 is mounted as a circuit in the CPU, a format in which the RAID controller 20 is built in the chipset and connected to the internal bus, or a RAID The present invention can be applied to any form in which the hardware of the controller 20 does not exist and the software is executed by the main processor and main memory of the computer.

  In the present embodiment, the disk array device 10 has been described as having an HDD as a physical storage device. However, for example, a case where the RAID 6 is configured using another physical storage device such as a semiconductor storage device can be easily performed. Applicable.

  Further, the present invention is not limited to the above-described embodiments as they are, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. Moreover, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment.

  DESCRIPTION OF SYMBOLS 10 ... Disk array apparatus, 20 ... RAID controller (disk array control apparatus), 21 ... RAID controller processor, 22 ... RAID controller memory, 30-1, 30-2, ..., 30-n ... HDD (physical storage device), 211: RAID function unit, 212: Parity replacement function unit, 213: Reconfiguration function unit, 221: Replacement completion position storage unit.

Claims (4)

Two or more first storage devices for storing data, and first parity data used for restoring the data stored in the first storage device, the first storage device A second storage device for storing first parity data calculated by the first calculation method from stored data, and the data used for restoring the data stored in the first storage device Second parity data that is different from the first parity data and is calculated from data stored in the first storage device by a second calculation method that is more complicated than the first calculation method. In the disk array controller connected to the third storage device for storing the parity data of
Detecting means for detecting a failure of the third storage device when the third storage device fails;
A first calculation means for calculating second parity data from the data stored in the first storage device by the second calculation method when a failure of the third storage device is detected;
Overwrite means for overwriting the first parity data stored in the second storage device with the second parity data calculated by the first calculation means;
When the third storage device in which the failure is detected is replaced with a fourth storage device, the data stored in the first storage device and the second parity overwritten in the second storage device Reading means for reading data;
First writing means for writing second parity data read from the second storage device to the fourth storage device;
Second calculation means for calculating first parity data from the data read from the first storage device by the first calculation method;
A disk array control device comprising: second write means for writing the first parity data calculated by the second calculation means to the second storage device. Obtaining means for obtaining a write request for requesting writing of data to the first storage device;
Third writing means for writing data to the first storage device in response to the acquired write request;
Determination means for determining whether the second parity data is overwritten on the first parity data stored in the second storage device; and
When it is determined that the second parity data has been overwritten, the first calculation means converts the second parity data from the data written by the third writing means by the second calculation method. Calculate
The overwriting means overwrites the second storage device with the second parity data calculated by the second calculation method from the data written by the third writing means. 2. The disk array control device according to 1. It further comprises replacement completion information storage processing means and replacement completion information storage means,
The first, second, and third storage devices include data stored in the first storage device, first parity data stored in the second storage device, and the third storage device. Having a plurality of stripes composed of second parity data stored in
The first calculating means stores, when a failure of the third storage device is detected, for each stripe of the first, second, and third storage devices stored in the first storage device. Second parity data is calculated from the existing data by the second calculation method,
The overwriting means overwrites the second parity data calculated for each stripe on the first parity data constituting the stripe,
The replacement completion information storage processing means replaces the replacement completion information indicating that the second parity data calculated for each stripe by the overwriting means is overwritten on the first parity data constituting the stripe. Stored in the completion information storage means,
The determination means includes second parity data calculated for each stripe composed of data written by the third writing means based on the replacement completion information stored in the replacement completion information storage means. The disk array control apparatus according to claim 2, wherein it is determined whether or not the first parity data constituting the stripe is overwritten. Two or more first storage devices for storing data, and first parity data used for restoring the data stored in the first storage device, the first storage device A second storage device for storing first parity data calculated by the first calculation method from stored data, and the data used for restoring the data stored in the first storage device Second parity data that is different from the first parity data and is calculated from data stored in the first storage device by a second calculation method that is more complicated than the first calculation method. A disk array device comprising: a third storage device for storing the parity data; and a disk array control device connected to the first storage device, the second storage device, and the third storage device ,
The disk array controller is
Detecting means for detecting a failure of the third storage device when the third storage device fails;
A first calculation means for calculating second parity data from the data stored in the first storage device by the second calculation method when a failure of the third storage device is detected;
Overwrite means for overwriting the first parity data stored in the second storage device with the second parity data calculated by the first calculation means;
When the third storage device in which the failure is detected is replaced with a fourth storage device, the data stored in the first storage device and the second parity overwritten in the second storage device Reading means for reading data;
First writing means for writing second parity data read from the second storage device to the fourth storage device;
Second calculation means for calculating first parity data from the data read from the first storage device by the first calculation method;
And a second writing means for writing the first parity data calculated by the second calculating means to the second storage device.

Images