JP2001100944A - RAID device - Google Patents

Abstract

(57) [Problem] To provide a RAID device in which even if it takes time to read a certain parity sequence, the influence of the reading does not reach other reading. SOLUTION: Disks 5A to 5D and 5P store data blocks and redundant data in a distributed manner.
The control unit 7 issues a second read request in order to read data blocks and redundant data from the disks 5A to 5D and 5P in response to the first read request transmitted from the host device. Furthermore, disks 5A to
From among 5D and 5P, a disk which no longer needs to read data blocks or redundant data is detected, and a read stop command is issued to stop reading from the detected disk.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a RAID (Red)
ununt Array of Inexpensi
More specifically, the storage capacity is increased and the transfer rate is improved by configuring a disk array using several disks (typically, magnetic disks or optical disks). In addition, the present invention relates to a RAID device that can provide higher system reliability.

[0002]

2. Description of the Related Art As is well known, RAID defines six basic architectures (levels 0 to 5). Hereinafter, the RAI adopting the level 3 architecture
The D device will be described with reference to FIG. In FIG. 39, the RAID device includes a controller 391 and disks 392A, 392B, 392C, 392D, and 392.
P. When data to be written arrives, the controller 391 divides the arrived data into blocks. The controller 391 generates redundant data based on the divided blocks. The controller 391 transfers the divided blocks to the disks 392A, 392B, 3
Write to 92C and 392D, and write redundant data to disk 392P. Hereinafter, a method of generating redundant data will be described with reference to FIG.

Data to be written arrives at the controller 391 in units of 2048 bytes. Now, FIG.
As shown in (a), the data that has arrived this time is called D-1. The controller 391 converts the arrived data D-1 into a data block DA as shown in FIG.
1, divided into D-B1, D-C1 and D-D1. Each data block DA1, DB1, DC1 and D
-D1 has a data length of 512 bytes.

Next, the controller 391 performs the operation shown in the following equation (1) using the data blocks DA1, DB1, DC1 and DD1. D-P1i = D-A1i xor D-B1i xor D-C1i xor D-D1i (1) However, in the above equation (1), the data blocks D-A1, D-
Since i-B1, D-C1, and D-D1 consist of 512 bytes, i takes a natural number of 1 to 512. That is, for example, when i = 1, the controller 391 sets the first byte of the data blocks D-A1, D-B1, D-C1, and D-D1 (that is, B-A11, B-B11, B-C1).
11 and BD11) are used to calculate DP11. When i = 2, DP12 is calculated using the second byte from the beginning of each data block (that is, BA12, BB12, BC12 and BD12). After that, the controller 391 repeatedly executes the above equation (1) until the last 512th byte to obtain DP11 and D-P11.
P12,... DP1512 are calculated. Controller 3
91 generates redundant data DP1 by arranging DP11, DP12,... DP1512 in order. That is, the redundant data DP1 is
Parities of A1, D-B1, D-C1, and D-D1.

[0005] The controller 391 stores the data blocks DA1, DB1, DC1 and DD1 in a distributed manner on the disks 392A, 392B, 392C and 392D. The redundant data D-P1 is stored on the disk 392P. The controller 391 controls data writing as described above.

The controller 391 controls not only writing but also data reading. Assume now that controller 391 has been requested to read data D-1. Under this assumption,
If no failure has occurred in any of the disks 392A to 392D, the controller 391 sends the data blocks D-A1 to
Read D-D1 from the disks 392A to 392D. The controller 391 forms the data D-1 of 2048 bytes by collecting the read data blocks D-A1 to D-D1. The controller 391 is
The data D-1 is transmitted to the request source of the data D-1.

[0007] Further, a disk may fail for various reasons. Assume now that disk 392C has failed. Under this assumption, even if the same read request as described above arrives at the controller 391, the controller 391 will continue to operate the data block D-
A1 to D-D1 are to be read from disks 392A to 392D. However, now disk 392C
, The data block D-C1 is not read. However, other data blocks D-A1,
Reading of DB1 and DD1 has been successful. If the controller 391 finds that the data block DC1 cannot be read, it reads the redundant data DP1 from the disk 392P.

Next, the controller 391 performs an operation shown in the following equation (2) using the data blocks DA1, DB1 and DD1 and the redundant data DP1.
Reproduce the data block D-C1. D−C1i = D−A1i xor D−B1i xor D−D1i xorD−P1i (2) However, in the above equation (2), as in the case of the above equation (1),
Data blocks D-A1, D-B1, D-C1, and D
Since i is 512 bytes, i takes a natural number from 1 to 512. The controller 391 repeatedly executes the above equation (2) from the first 1st byte to the last 512th byte to obtain DC11, DC12,... DC151.
2 is calculated. The controller 391 restores the data block D-C1 based on these calculation results. As a result, the data block D is stored in the controller 391.
-A1 to D-D1 are complete. The controller 391 collectively arranges the data blocks D-A1 to D-D1 to form 2048-byte data D-1. The controller 391 transmits the data D-1 to the request source of the data D-1.

As described above, the RAID device shown in FIG.
Discs 392A, 392B, 392C and 392D
In some cases, a necessary data block cannot be read from any one of the failed disks. But R
The AID device can restore the data blocks stored on the failed disk from the data blocks read from the other four normal disks except the failed disk and the redundant data.

Incidentally, data stored in a RAID device is roughly classified into two types. One is computer data and the other is video data. Computer data and video data have different properties. Therefore, the conditions required for the RAID device are different between the case where computer data is read and the case where video data is read.
More specifically, it is required that the computer data read from the disk be reliably transmitted to the requesting host terminal. Therefore, a failed disk occurs and RA
When the ID device cannot read the computer data of a certain parity series, it is required to reliably restore the unreadable computer data based on other data of the same parity series without losing it. For that purpose, it may take some time to read the parity sequence. On the other hand, the video data is reproduced as a video by the requesting host device. If the video data arrives at the host device with a delay, the video reproduced on the host device will be interrupted. Therefore, the RAID device must read the requested video data from the disk within a certain time and continuously transmit the video data to the host device.

Japanese Unexamined Patent Publication No. 2-81123 discloses a conventional RAID device suitable for video data (hereinafter referred to as a first R device).
(Referred to as an AID device) is disclosed. This R
The AID device has a storage device group including a plurality of storage devices for storing data and a redundant storage device for storing redundant data generated from the data. RAID
When reading data from a storage device group, if the reading of data from a certain storage device is delayed by a predetermined time or more compared to the start of reading data from another storage device, the device reads the certain storage device. It is determined as a failure storage device. R
In the AID device, data to be read from the storage device determined to have a failure is generated from data of other storage devices and redundant data.

Japanese Patent Laid-Open No. 9-69027 discloses another example of a conventional RAID device suitable for video data (hereinafter referred to as a second RAID device).
In the second RAID device, the redundant data is not normally read from the redundant storage device. The second RAID device reads the data from the redundant storage device when the read operation is not completed even after the time when the storage device storing the data has completed the read operation, and Restore data that has not been read.

[0013]

However, the first and second RAID devices have the following problems. First, in the first RAID device, as shown in FIG. 41A, even if a predetermined time has elapsed from the start time of the fourth reading (reading from the disk B), the disk D from which reading has not started is started. It is determined that a failure has occurred. Then, a parity operation is performed to restore the data on the disk D. However, in general, the time from the start to the completion of reading is not constant on a disk. That is, a certain disk may complete reading in a short period of time, and another disk may fail to read several times, and may take a long time to read. Such prolonged reading becomes more remarkable as the data size read in one access is larger, as in the case of reading video data. Therefore, in the first RAID device, FIG.
As shown in (b), even though the disk P starts reading earlier than the disk B (the fourth disk), reading of the disk B may be completed earlier than the disk P. In this case, even if the predetermined time has elapsed since the start of reading the disk B,
Since there is no data on the disk P, the RAID device cannot restore the data on the disk D. Since the data cannot be restored, there is a problem that the video data being read may be interrupted on the host device side.

Next, in the second RAID device, the reading of the redundant data is started after a lapse of a fixed time from the data block reading request (after a timeout). For this reason, as shown in FIG. 42A, there is a problem that the time until the data restoration is completed becomes long. Furthermore, in the second RAID device, FIG.
As shown in (1), data (data on the disk C in the figure) may be read immediately after the timeout. Immediately after the timeout, the second RAID device may be able to obtain data earlier than the end of the parity operation. In this case, the second RAID device can transmit the video data to the host device faster by using the data read immediately after the timeout. However, the second
Once the RAID device starts reading the redundant data, it does not use the data read immediately after the timeout. As a result, transmission of the video data to the host device may be delayed. Due to this delay, there is a problem that the video may be interrupted on the host device side.

In general, a read request that arrives later is often accumulated on a disk whose read operation is delayed. Therefore, if the disk fails to read data and retries reading the data, processing the next read request is delayed. As is apparent from the above, there is a problem that one reading failure has a bad influence on the next and subsequent readings.

Therefore, an object of the present invention is to provide a RAID apparatus in which even if it takes time to read a certain parity sequence, the influence of the reading does not affect other reading. Another object of the present invention is to provide a RAID device capable of efficiently executing a parity operation and reading a large amount of data per unit time.

[0017]

According to a first aspect of the present invention, a data block generated by dividing data and redundant data generated from the data block are recorded and transmitted from outside. R that executes data read processing in response to the first read request
An AID device, comprising: m disks in which data blocks and redundant data are stored in a distributed manner; and a control unit for controlling a read process, wherein the control unit responds to an input first read request. Then, in order to read the data blocks and the redundant data from the m disks, a second read request is issued, and from the m disks, the disk for which the reading of the data blocks or the redundant data is no longer necessary is read out. Then, a read stop command is issued to stop reading from the disk that has been detected and read is determined to be unnecessary.

According to the first aspect, when it is determined that reading of a certain data block or redundant data is unnecessary, the reading of the data block or redundant data is stopped. Therefore, the disk that has stopped reading can shift to the next reading at that time. This makes it possible to provide a RAID device in which even if reading of a certain disk is delayed, the adverse effect of the delay does not affect other reading.

A second invention is dependent on the first invention,
When the (m-1) disks have completed reading, the control unit determines that the reading performed on the remaining one disk is unnecessary, and sends a read stop command to the disks determined to be unnecessary. Issue According to the second aspect, even when it takes too much time to read one disk, the reading is stopped. This makes it possible to provide a RAID device in which even if reading of a certain disk is delayed, the adverse effect of the delay does not affect other reading.

A third invention is dependent on the first invention,
When detecting that two or more disks cannot be read, the control unit determines that reading of the disk being executed at the time of the detection is unnecessary, and issues a read stop command to the disk determined to be unnecessary. I do.
According to the third aspect, when it is determined that the parity operation cannot be performed, reading of another disk can be stopped. As a result, unnecessary reading is not continued, so that it is possible to provide a RAID device in which the adverse effect of the unnecessary reading does not affect other reading.

A fourth invention is dependent on the first invention,
When the (m-1) disks have completed reading, the control unit determines that reading that has not yet been performed on the remaining one disk is unnecessary, and stops reading on disks that are determined to be unnecessary. Issue a command. 4th
According to the invention, since unnecessary reading is not executed, it is possible to provide a RAID device in which the adverse effect of the unnecessary reading does not affect other reading.

According to a fifth aspect, a data block generated by dividing data and redundant data generated from the data block are recorded, and respond to a first read request from an external host device. A RAID device that executes data read processing, based on m disks on which data blocks and redundant data are stored in a distributed manner, and (m-2) data blocks and redundant data. And a control unit for controlling the read process by executing a parity operation to restore the remaining one data block, wherein the control unit responds to the input first read request. , A second read request is issued to read data blocks and redundant data from the m disks, and when (m-1) disks have completed reading, the (m-1) 1) It is detected whether or not the data read from the discs is a combination of a data block and redundant data. After waiting for the time, in order to restore the data blocks not read from the remaining one disk, a restoration instruction is issued to the parity operation unit, and m data blocks are prepared by the parity operation of the parity operation unit. Then, a process for transmitting data to the host device is performed, and the predetermined time is selected as a value that guarantees that data is transmitted to the host device without delay.

In the fifth invention, after the combination of the data block and the redundant data is read from the (m-1) disks, the control unit reads the remaining one data block for a predetermined time. Wait to be done. If the remaining one data block is read by the predetermined time, it is not necessary to execute the parity operation. As a result, the number of executions of the parity operation can be reduced.

A sixth invention is according to the fifth invention,
When the (m-1) disks have completed reading, the control unit further detects whether the data read from the (m-1) disks is a combination of a data block and redundant data. Then, when it is detected that the combination is not a combination of the data block and the redundant data, a process for transmitting data to the host device is executed without waiting for a predetermined time from the detection.

According to the sixth aspect, when the data read from the (m-1) disks are only data blocks, the control unit does not wait for a predetermined time, and
Since data is transmitted to the host device, it is possible to realize a RAID device that can read a large amount of data per unit time.

A seventh invention is dependent on the fifth invention,
The predetermined time is selected based on the probability that the reading is expected to be completed from the start of reading of each disk. According to the seventh aspect, in most cases, the remaining one data block is read within a predetermined time, so that the number of times of executing the parity operation can be reduced.

According to an eighth aspect of the present invention, a data block generated by dividing data and redundant data generated from the data block are recorded and respond to a first read request from an external host device. A RAID device that executes data read processing, based on m disks on which data blocks and redundant data are stored in a distributed manner, and (m-2) data blocks and redundant data. And a control unit for controlling the read process by executing a parity operation to restore the remaining one data block, wherein the control unit responds to the input first read request. , A second read request is issued to read data blocks and redundant data from the m disks, and when (m-1) disks have completed reading, the (m-1) 1) It is detected whether or not the data read from the discs is a combination of a data block and redundant data. After waiting for the time, a restoration instruction is issued to the parity operation unit to restore the data block that has not been read from the remaining one disk, and when m data blocks are aligned by the parity operation of the parity operation unit, A process for transmitting data to the host device is executed, and the restoration instruction is issued when the parity operation unit is not executing the parity operation.

According to the eighth aspect, the control unit reliably issues the restoration instruction when the parity operation is not being performed, so that no extra load is applied to the parity operation unit.
Thereby, the parity operation unit is used efficiently.

A ninth invention is according to the eighth invention,
The RAID device includes a table describing a time period during which the parity operation unit performs the parity operation, and the control unit further refers to the time period described in the table, and the parity operation unit does not execute the parity operation. At times, a restoration instruction is issued. In the ninth aspect, the control unit can easily know the issuance timing of the restoration instruction only by referring to the time zone described in the table.

According to a tenth aspect, a data block generated by dividing data and redundant data generated from the data block are recorded, and respond to a first read request from an external host device. A RAID device that executes data read processing, based on m disks on which data blocks and redundant data are stored in a distributed manner, and (m-2) data blocks and redundant data. And a control unit for controlling the read process by executing a parity operation to restore the remaining one data block, wherein the control unit responds to the input first read request. , (M-1) disks have failed in reading data blocks in the past, and determine that reading of each data block has not failed in the past. If you, for reading only the respective data block, the second read request (m-1)
When the data blocks are issued to the (m-1) disks and the data blocks are read from the (m-1) disks, a process for transmitting data to the host device is executed.

In the tenth aspect, the second read request may not be issued for redundant data. That is,
When redundant data is not required, unnecessary reading is not performed. As a result, the amount of data that can be read per unit time can be increased.

An eleventh invention is according to the tenth invention, and the control unit further comprises: (m-1) when it is determined that reading of any of the data blocks has failed in the past.
Issue a second read request to m disks to read the data blocks and redundant data,
When the (m-1) disks have completed reading, it is detected whether or not the data read from the (m-1) disks is a combination of a data block and redundant data. If a combination of redundant data is detected, in order to restore data blocks that have not been read from the remaining one disk,
A restoration instruction is issued to the parity operation unit, and when m data blocks are prepared by the parity operation of the parity operation unit, a process for transmitting data to the host device is executed.
In the eleventh invention, if necessary, a second read request is issued to read redundant data. As a result, the parity operation can be executed as soon as possible.

A twelfth invention is according to the eleventh invention, wherein the RAID device includes a table describing a storage location of a data block from which each disk failed to read in the past, and the control unit refers to the table. Then, it is determined whether to issue the second read request to the (m-1) disks or to issue the second read request to the m disks.
According to the twelfth aspect, the control unit can easily determine whether to issue the second read request for reading the redundant data simply by referring to the table.

A thirteenth invention is according to the twelfth invention, and when a defect occurs in a storage location of a data block or redundant data stored in each of the m disks, an alternative storage is provided instead of the defect storage location. The apparatus further includes a reassignment processing unit that executes a reassignment process for allocating a location, wherein the control unit, when an alternative recording location is reassigned to a storage location of the data block described in the table, the storage location of the data block. To delete the table. In the thirteenth aspect, an alternative recording area is allocated to a defective recording area, and a data block or redundant data is rewritten in the alternative recording area. This reduces the number of data blocks that take a long time to read in the table. This makes it possible to provide a RAID device that can read more data per unit time.

A fourteenth invention is according to the thirteenth invention, wherein an alternative recording area management unit for managing, as alternative recording area information, an address of an alternative recording area previously reserved in each of the m disks, An address management unit that manages address information of an alternative recording area allocated in place of the area, wherein the reassignment processing unit receives the second read request from the control unit and transmits the second read request to the m disks. The delay time in each disk is measured, and based on the measured delay time, it is determined whether or not the data block to be read by each of the second read requests or the recording area of the redundant data is defective. When the recording area of the redundant data is determined to be defective, based on the alternative recording area information managed by the alternative recording area management unit, Recessed assigns an alternate recording area Alternative recording area, to manage address information of the allocated replacement recording area in the address management unit, the control unit on the basis of the address information managed by the address management unit, the second
Is issued, and the delay time is an elapsed time calculated from a predetermined processing start time.

According to the fourteenth aspect, the reassignment processing section determines whether or not the recording area is defective based on the elapsed time calculated from the predetermined processing start time. If the delay of the response returned from the disk side is large, the reassignment processing unit determines that the recording area currently being accessed for reading is defective, and allocates an alternative recording area. As a result, the present RAID device can read and transmit data to the host device while suppressing occurrence of a large response delay.

A fifteenth invention is according to the first invention, and when a defect occurs in a storage location of a data block or redundant data stored in each of m disks, an alternative storage is provided instead of the defect storage location. The apparatus further includes a reassignment processing unit that executes a reassignment process for allocating a place.

A sixteenth aspect is according to the fifteenth aspect, wherein an alternative recording area management unit for managing, as alternative recording area information, an address of an alternative recording area secured in advance in each of the m disks, An address management unit that manages address information of an alternative recording area allocated in place of the area, wherein the reassignment processing unit receives the second read request from the control unit and transmits the second read request to the m disks. The delay time in each disk is measured, and based on the measured delay time, it is determined whether or not the data block to be read by each of the second read requests or the recording area of the redundant data is defective. When the recording area of the redundant data is determined to be defective, based on the alternative recording area information managed by the alternative recording area management unit, Recessed assigns an alternate recording area Alternative recording area, to manage address information of the allocated replacement recording area in the address management unit, the control unit on the basis of the address information managed by the address management unit, the second
Is issued, and the delay time is an elapsed time calculated from a predetermined processing start time.

A seventeenth invention is according to the sixteenth invention, and the reassignment processing unit replaces the defective recording area with the defective recording area only when it determines that the same recording area is a defective recording area continuously for a specified number of times. Allocate a recording area. According to the seventeenth aspect, in the recording area, when the reassignment processing unit determines that there is a possibility of a defect continuously for the prescribed number of times, the substitute recording area is assigned. Therefore, even if the reassignment processing unit suddenly determines that the non-defective storage area is defective, no alternative recording area is allocated. As a result, it is possible to provide a RAID device that assigns an alternative recording area only to a true defect recording area.

An eighteenth invention is according to the sixteenth invention, and the predetermined processing start time is a time at which the second read request is transmitted to each of the m disks. A nineteenth invention is according to the sixteenth invention, and the predetermined processing start time is:
This is the time at which reading is started by the m disks based on the second reading request. According to the eighteenth or nineteenth aspect, the reassignment processing section can accurately know the delay time.

[0041]

DESCRIPTION OF THE PREFERRED EMBODIMENTS (First Embodiment) FIG. 1 is a block diagram showing a configuration of a RAID device according to a first embodiment of the present invention. In FIG. 1, the RAID device includes a host interface 1, a selector 2, six buffer units 3A to 3D, 3P and 3R, five SCSI interfaces 4A to 4D and 4P, and five disks 5
A to 5D and 5P, a parity operation unit 6, and a control unit 7
And Note that the issue time table 71 shown in the control unit 7 is not used in the first embodiment. Since the issue table 71 is required in the second embodiment, the description thereof is omitted here.

FIG. 2 shows the buffer units 3A-3 in FIG.
3 shows a detailed configuration of D, 3P and 3R. 2, the storage area of the buffer portion 3A is divided into a plurality of buffer areas 3A _1, 3A _2, 3A ₃ .... Each of the buffer areas 3A ₁ , 3A ₂ , 3A ₃ ... Has a capacity (that is, 512 bytes in the present embodiment) capable of storing a data block and redundant data described later. further,
Each of the buffer areas 3A ₁ , 3A ₂ , 3A ₃ ... Is assigned identification information (usually the head address of each buffer area) for specifying each buffer area. The other buffer units 3B to 3D, 3P, and 3R are also divided into a plurality of buffer areas similarly to the buffer unit 3A. In addition, each buffer region, as in the buffer area 3A _1, etc., the identification information is assigned.

Referring back to FIG. A host device (not shown) is provided outside the RAID device. The host device is connected to the RAID device so that bidirectional communication is possible. The host device can write data to the RAID device. At this time, the host device transmits a write request and data to the RAID device. In this embodiment, the host device transmits data having a size of 2048 bytes to the RAID device. The transmission data of the host device is
Typically, the video data is generated by dividing the video data every 2048 bytes. The RAID device starts the write process in response to the write request and the reception of the data. Less than,
This writing process will be briefly described with reference to FIG. R
This time, the transmission data D-1 (FIG. 3) of the host device is supplied to the selector 2 of the AID device through the host interface 1.
(See (a)) is input. The selector 2 divides the data D-1 into four to generate 512-byte data blocks D-A1, D-B1, D-C1, and D-D1. The selector 2 transfers the data block DA1 to the buffer unit 3A. Further, the selector 2 transfers the data blocks DB1 to DD1 to the buffer units 3B to 3D. The buffer units 3A to 3D hold the data blocks D-A1 to DD1 transferred thereto. Also,
The four data blocks D-A1 to D-D1 are also input to the parity operation unit 6. The parity operation unit 6 performs the parity operation described in the section of the related art to obtain the data block D.
-A1 to D-D1 to 512-byte redundant data D-
Generate P1. The redundant data DP1 is stored in the buffer unit 3
P and temporarily held by the buffer 3P.

At present, data blocks D-A1 to D-D1 are held in the buffer units 3A to 3D. Further, the buffer unit 3P holds redundant data D-P1. The four data blocks D-A1 to D-D1 and the redundant data D-P1 are created based on the same 2048-byte data D-1, and belong to the same parity sequence. For example, it is assumed that the data blocks D-A1 to D-D1 and the redundant data D-P1 belong to the parity sequence n. As described above, in this embodiment, the parity sequence means a combination of a data block and redundant data generated based on the same 2048-byte data.

A write request is input to the control unit 7 through the host interface 1. Control unit 7
Assigns the storage location of the parity sequence n generated this time in response to the input of the write request. The storage location of the data block is selected from the storage areas of the disks 5A to 5D. The storage location of the redundant data is selected from the disk 5P. The control unit 7 instructs the SCSI interface 4A of the storage location selected from the storage area of the disk 5A. Similarly, the control unit 7 instructs the SCSI interfaces 4B to 4D and 4P of the storage location selected from the storage areas of the disks 5B to 5D and 5P. SCS
The I interface 4A responds to the instruction of the control unit 7
From the buffer unit 3A connected to itself, the data block D
-Take out A1. The SCSI interface 4A stores the data block D-A1 in a storage area (storage location) of the disk 5A. Other SCSI interface 4B
To 4D are data blocks D-B1 to DD in the buffer units 3A to 3D, similarly to the SCSI interface 4A.
1 is stored in the storage area (storage location) of the disks 5B to 5D. The SCSI interface 4P stores the redundant data DP1 of the buffer unit 3P in a storage area (storage location) of the disk 5P.

In the RAID device, each time the transmission data of the host device arrives, the above-described write processing is performed. As a result, as shown in FIG. 3B, the data blocks and redundant data of the same parity sequence are stored in the disks 5A to 5A.
D and 5P are stored in a distributed manner. For example, as described above, as the parity sequence n (see the dot portion),
Data blocks D-A1, D-B1, D-C1, and D
-D1 and redundant data D-P1 are generated. The data blocks D-A1 to D-D1 are stored on the disks 5A to 5D. The redundant data D-P1 is stored in the disk 5P.
Is stored in Other parity sequences, like parity sequence n, are also stored on disks 5A, 5B, 5C, 5D, and 5D.
P is distributed and stored.

In the above write processing, the redundant data is stored only on the disk 5P, and the disk for storing the redundant data is fixed. That is, this writing process is described based on the RAID level 3 architecture. However, the RAID device according to the present embodiment is not limited to RAID level 3 but may be constructed based on level 5, for example. Level 5
Is different from Level 3 in that redundant data is not stored on a fixed disk but is stored in a distributed manner on each disk included in the RAID device.

In some cases, the host device wants to read data from the RAID device. In such a case, the host device transmits a first read request to the RAID device.
The first read request includes information indicating the data storage location, and thereby specifies the data storage location. The RAID device starts the read process in response to the arrival of the first read request. This reading process
Since this is a characteristic process of the present application, it will be described in detail with reference to the flowchart shown in FIG. The first read request is input to the control unit 7 through the host interface 1 (Step S1). The control unit 7 obtains a data storage location from the first read request. The control unit 7 specifies a storage location of a parity sequence (four data blocks and redundant data) generated based on the data based on the storage location of the data. Here, it should be noted that the process of obtaining the storage location of the parity sequence from the storage location of the data is well known and is defined by the RAID architecture.

Next, the control unit 7 issues a second read request set to read the parity sequence (step S2). In the present embodiment, five second read requests are issued because the parity sequence is distributed and stored in the disks 5A to 5D and 5P. The second read request is sent to the SCSI interfaces 4A to 4D and 4P by 1
Sent individually.

The second read request to the SCSI interface 4A indicates the storage location of the data block on the disk 5A. The disk 5A receives the second read request through the SCSI interface 4A. The disk 5A reads the data block from the storage location specified by the second read request. The read data block is transmitted to the SCSI interface 4A. By the way, the second read request to the SCSI interface 4A indicates not only the storage location of the disk 5A but also the storage location of the buffer unit 3A. More specifically, it is specified which buffer area (see FIG. 2) of the buffer section 3A stores the read data block. SCSI interface 4A
Are the designated buffer areas 3A ₁ , 3A ₂ , 3A ₃ ,
The data blocks read from the disk 5A are stored in any one of. The buffer unit 3A stores a 512-byte data block in one of the buffer areas 3A _i.
(I is a natural number of 1, 2, 3,...), The first read completion is transmitted to the control unit 7 to notify that the reading of the disk 5A has been completed.

Similarly, disks 5B to 5D are
Second input through the SI interfaces 4B to 4D
In response to the read request, the respective data blocks are started to be read. The data blocks read from each of the disks 5B to 5D are stored in the SCSI interfaces 4B to 4B.
Via D, they are stored in the buffer areas 3B _{i to} 3D _i . Here, when the data blocks have been stored in any of the buffer areas 3B _{i to} 3D _i , the buffer units 3B to 3D indicate that the reading of the disks 5B to 5D has been completed.
Is transmitted to the control unit 7. Similarly, when the second read request is input from the SCSI interface 4P, the disk 5P starts reading redundant data. Redundant data read via the SCSI interface 4P, it is stored in the buffer region 3-Way _i. Buffer 3P, when the redundant data has finished being stored in one of the buffer areas 3P _i, and sends the first read completion is notified to the control unit 7 that the read disk 5P is complete.

It should be noted that, in most cases, the completion of the first reading of the buffers 3A to 3D and 3P arrives at the controller 7 with a time lag.
For example, when it takes a long time for the disk 5A to read a data block, the first read completion of the disk 5A arrives at the control unit 7 later than the completion of the first read from another. As is clear from the above, each first read completion arrives at the control unit 7 in the order in which the reading on the disks 5A to 5D and 5P is completed. Control unit 7
When four first read completions arrive (FIG. 4)
In step S11 of (b), the process proceeds to step S12 without waiting for the completion of the remaining first reading. That is, the control unit 7 determines that reading has been completed on any of the four disks, and further determines that reading of the remaining one disk is delayed.

Next, the control unit 7 notifies the completion of the first read-out to the buffer units that have not been transmitted (the buffer units 3A to 3D and 3).
P), and the disk for which reading has not been completed (one of the disks 5A to 5D and 5P).
Number). The control unit 7 controls the disk whose reading has not been completed to forcibly stop the reading currently being performed.
A read stop command is issued (step S12).
The read abort command is issued through a SCSI interface connected in front of the disk that has not been read.
Input to the disc. As a result, the disk for which the reading has not been completed stops the reading currently being performed.

After step S12, the control section 7 determines whether or not a parity operation is necessary (step S13). At the present time, the control unit 7 controls the buffers 3A to 3D and 3P
, The first read completion is received from any four of them. Now, consider a case where the control unit 7 has received the first read completion from the four buffer units 3A to 3D.
In this case, there are four data blocks. Therefore, the control unit 7 determines that the data requested by the host device can be transmitted. Therefore, the control unit 7 determines that the parity operation is unnecessary, and determines in step S1
The process proceeds directly from step 3 to step S16. On the other hand, now, the control unit 7
Consider that the first read completion has been received from the buffer 3P. In this case, redundant data and three data blocks are present, and any one of the data blocks has not been read. Therefore, it is necessary to restore one data block that has not been read. Therefore, the control unit 7 determines that the parity operation is necessary. Therefore, the control unit 7 performs steps S13 to S13.
The process proceeds to 14, and a restoration instruction is issued to request the parity operation unit 6 to execute the parity operation (step S14).

In response to the restoration instruction, the parity calculation unit 6 converts the redundant data and the three data blocks into a buffer area 3P _i and three buffer areas (any one of the buffer areas 3A _{i to} 3D _i ). ). The parity calculation unit 6 performs the parity calculation described in the section of the related art, and calculates the parity data from the redundant data and the three data blocks.
One data block that has not been read is restored.
Reconstructed data block is temporarily held in one of the buffer areas 3R _i with buffer 3R it is. When the parity calculation is completed, the parity calculation unit 6 issues a restoration completion indicating that, and transmits the restoration completion to the control unit 7. Control unit 7
Receives the completion of restoration (step 15), determines that the four data blocks have been collected and the data requested by the host device is ready to be transmitted. Therefore, the control unit 7 proceeds to step S16.

In step S 16, the control section 7 issues a second read completion signal and sends it to the selector 2.
The completion of the second read specifies the four buffer areas where the data block is currently held. Selector 2
In accordance with the completion of the second reading, four buffer areas are sequentially selected, and four data blocks are sequentially read.
Further, the selector 2 assembles the four read data blocks to form 2048-byte data.
The configured data is transmitted to the host device through the host interface 1.

The general read processing of the present RAID device has been described above. A specific example of the reading process of the present RAID device will be described with reference to FIG. Hereinafter, the case where the host device requests data reading in the order of the parity sequences n and (n + 1) shown in FIG. 3B will be described as an example. Here, FIG. 5A shows the parity sequence n and (n +
It is a schematic diagram which shows the timing which 1) is read on a time axis.

First, the controller 7 issues a second read request to read the parity sequence n (FIG. 4A).
Steps S1 and S2). Thereafter, a second read request for reading the parity sequence (n + 1) is issued (steps S1 and S2). In the example of FIG. 5A, as indicated by the dot portion, the disk 5D starts reading data blocks first. Or later,
Reading of data blocks or redundant data starts in the order of the disks 5C → 5A → 5P → 5B. Over time, before the time t _1, the disk 5C, 5A and 5P complete the read. The disk 5B is at time t ₁
, The reading is completed fourth. However, the reading of the disk. 5D, will be delayed compared to the other reading, has continued even when the time t _1.

Therefore, immediately after time t ₁ , the controller 7
Are the first from the buffer units 3A, 3B, 3C and 3P.
4 have arrived (step 11 in FIG. 4B). Therefore, the control unit 7 issues a read stop command to the disk 5D as a disk for which reading has not been completed.
(Step S12). In response to the read stop command, the disk 5D stops the ongoing read, as indicated by the solid crosses in FIG. 5A. Hereafter, as is clear from the above, the control unit 7 executes step S
13 to S16 are executed. However, since the description of the processing after step S13 is not important here, the description of the processing is omitted.

When the time t ₁ has passed and the time t ₂ has elapsed, the disk 5D starts reading the data block of the parity sequence (n + 1) (see the hatched portion of the vertical line). At the time of time t _2, the disc 5A, 5C and 5
The reading process has started at P. Also, disk 5B
In from after a little time t _2, the reading process has begun. Time has elapsed, at the point of time t _3, disk 5
C, 5D, 5A and 5P have completed reading.
Therefore, as shown by the dashed crosses, the disk 5
The reading of B is forcibly stopped by the reading stop command of the control unit 7.

In the foregoing, a specific example of the reading process of the present RAID device has been described. As described above, in the present RAID device, when four data blocks are provided, redundant data is unnecessary, and when three data blocks and redundant data are provided, the remaining one data is used. Blocks are not required. Therefore, the RAID device issues a read stop command to the disk that is performing unnecessary read (step S12 in FIG. 4B), and forcibly stops the read. This is a characteristic process of the present RAID device.

Next, in order to highlight the features of the present RAID device, a reading process by a RAID device that does not execute step S12 will be described with reference to FIG. Here, FIG. 5B shows the RA without executing step S12.
FIG. 4 is a schematic diagram showing, on a time axis, timings at which parity sequences n and (n + 1) are read in the ID device. The condition in FIG. 5B is the same as the condition in FIG. 5A except that the RAID device does not execute step S12. That is, the host device requests data reading in the order of the parity sequences n and (n + 1) under the same conditions as described above.

Control unit 7 controls parity sequences n and (n +
To read 1), a second read request is issued in the order of arrival of the first read request. Then, in FIG. 5B, as in the case of FIG. 5A, the disk 5D →
Reading of data blocks or redundant data starts in the order of 5C → 5A → 5P → 5B. Further, At time t _1, as in the case of FIG. 5 (a), the disk 5C, 5
A, 5P and 5B have completed reading, while disk 5D is continuing reading. Reading disk 5D, since the read stop command is not issued, without being forced discontinued immediately after time t _1, finally ends at much time t ₄ after time t _1. At this time t ₁ , the data of the parity sequence n can be transmitted to the host device as in the case of FIG. 5A.

At time t ₄ , the disk 5
A, 5B, 5C and 5P are starting to read the data blocks and redundant data of the parity sequence (n + 1). However, disk 5D is finally at time t ₅ after the time t _4, it starts reading the data blocks of the parity series (n + 1). Further, immediately after time t _6, so the disk 5C, 5A, reading of 5P and 5B is completed, immediately thereafter, the data of the parity sequence (n + 1) is transmitted to the host device.

As can be seen by comparing FIG. 5A and FIG. 5B, the present RAID device has the following technical effects. That is, in both FIGS. 5A and 5B, the time t
_At the time point ₁ , since the three data blocks and the redundant data are prepared, the data blocks stored in the disk 5D can be restored. Therefore, the data of the parity sequence n can be transmitted to the host device. At this point, reading from the disk 5D becomes unnecessary. Therefore, as shown in FIG. 5 (a), the RAID device, immediately after time t _1, forcibly stop the reading of the disk 5D. Therefore, the disk 5D can start reading the data block of the parity sequence (n + 1) as soon as possible. On the other hand, as shown in FIG. 5 (b), RAID devices that do not perform step S12, subsequent time t _1, until the time t _4, not stop the unnecessary reading of disk 5D. Due to the useless processing time, as shown in FIG. 5B, the start of reading of the parity sequence (n + 1) of the disk 5D is delayed. As described above, the present RAID device stops reading from a disk whose reading has not been completed when a certain data block can be restored. As a result, a disk for which reading has not been completed can immediately start the next reading without continuing useless reading. That is, according to the present RAID device, the delay of one read does not adversely affect the subsequent read.

Further, in the example of FIG.
As a result of D starting reading of the data block at time t ₂ , the RAID device can transmit the data of the parity sequence (n + 1) to the host device immediately after time t ₃ . Therefore, this RAID device, until the time t _3,
Two data (parity sequence n
And (n + 1)). On the other hand, in the example of FIG. 5 (b), the disk 5D is finally to start reading data blocks at time t _5. Adversely affect the reading after a delay of this reading, the RAID device does not execute the step S12, at time t _3, no data parity sequence (n + 1) to a state that can be transmitted to the host device. Therefore, RAID devices that do not perform step S12, even if the time t _3, 2 pieces of data requested to the host device (parity sequence n and (n +
1)) cannot be sent. As is clear from the above, this book R
According to the AID device, the amount of data that can be read per unit time from a combination of disks 5A to 5P (a so-called disk array) increases. By this, this RAI
The D device can continuously send data to the host device. As a result, when the present RAID device handles video data, the video reproduced on the host device side is less likely to be interrupted.

By the way, the disk 5A of the first embodiment
6 to 5D and 5P may be of the type shown in FIG. Here, FIG. 6A shows parity sequences n to (n + 4) in one of the disks.
Indicates the physical recording position of the data block (or redundant data). In FIG. 6A, a data block (or redundant data) of a parity sequence n is recorded on the innermost side of the disk. Further, in the order along the outer circumferential direction, the parity sequence (n + 2) → (n + 4) → (n + 1)
Of data blocks (or redundant data). The data block (or redundant data) of the data block (or redundant data) of the parity sequence (n + 3) is recorded on the outermost peripheral side.

As shown in the upper section of FIG. 6B, the control unit 7 changes the parity sequence from n → (n + 1) → (n + 2) → (n
+3) → (n + 4), and a second read request for reading a data block (or redundant data)
Consider the case of issuing to the disk shown in FIG. The disk shown in FIG. 6A does not execute reading in the order of arrival of the second read request, but executes reading so that the seek distance of the read head becomes short. For example, the reading order of the disk is changed so that the read head moves linearly from the inner circumference to the outer circumference.
As a result, n → (n + 2) → (n + 4) → (n + 1) →
Data blocks (or redundant data) are read in the order of (n + 3). This allows the disk to store more data blocks (redundant data) per unit time.
Can be read efficiently.

The disks 5A to 5D and 5P shown in FIG.
The read processing of the present RAID device when a disk whose read order is changed is used for all or a part of the above will be specifically described. In the following, the host device determines that the parity sequence shown in FIG. 3B is n → (n +
1) → (n + 2) → (n + 3) → (n + 4). Here, FIG. 7A is a schematic diagram showing, on the time axis, the timing at which the parity sequences n to (n + 4) are read in the present RAID device.

First, the control unit 7 issues a second read request in the requested order. Therefore, each of the disks 5A to 5D and 5P has n → (n + 1) → (n +
The second read request arrives in the order of 2) → (n + 3) → (n + 4). However, the disks 5A to 5D and 5P determine the reading order independently of each other. Therefore, the actual reading order of each of the disks 5A to 5D and 5P is not the same as the requested order, and may be different from each other. In FIG. 7A, at time t ₇ , the disks 5A, 5B, 5D, and 5P have completed reading of the data blocks and redundant data of the parity sequence (n + 2) (see the hatched portions). There is. It should be noted that, in the time t _7, disk 5C is,
The reading of the data block of the parity sequence (n + 4) (see the hatched portion of the horizontal line) has been completed. In this case, the control unit 7, immediately after time t _7, the parity sequence (n + 2) will receive the first read a complete fourth one (step S11 in Figure 4 (b)). Therefore, a read stop command is issued to the disk 5C (step S12). Therefore, disk 5
C does not read the data block of the parity sequence (n + 2). Similarly, at time t _8, disk 5A, 5
B, 5C, and 5P have completed reading of the data block and redundant data (see the shaded portion of the vertical line) of the parity sequence (n + 4). In this case, the control unit 7, immediately after time t _8, issues a read stop command regarding parity sequence (n + 4) to the disk 5D. for that reason,
The disk 5D does not read the data block of the parity sequence (n + 4).

Next, in order to highlight the features of the present RAID device, a reading process by a RAID device that does not execute step S12 will be described with reference to FIG. FIG. 7B is a schematic diagram showing, on the time axis, the timing at which the parity sequences n to (n + 4) are read out in the RAID device that does not execute step S12. FIG.
The condition of FIG. 7B is the same as that of FIG. 7A except that the RAID device does not execute step S12. That is, the host device requests data reading in the order of the parity sequences n to (n + 4) under the same conditions as described above.

The discs 5A to 5D and 5P determine the reading order independently of each other. In FIG. 7B,
Figure 7 similarly to the In the case of (a), at time t _7, the disk 5A, 5B, 5D and 5P parity series (n + 2)
Has been read out of the data block and redundant data (see hatched portions). However, disk 5
C at time t _7, it does not start reading the data blocks of the parity series (n + 2). However, in the example of FIG. 7B, since the read stop command is not issued to the disk 5C, the data block of the parity sequence (n + 2) starts to be read from the disk 5C soon. However, it is unnecessary to read the parity sequence (n + 2) on the disk 5C, which is a waste of time. This is because, at time t _7, and is recorded on the disc 5C, data blocks of the parity series (n + 2) is because it is possible restoration. Similarly, at time t ₈ ,
The disks 5A, 5B, 5C and 5P have completed reading the parity sequence (n + 4). However, at time t
_{In 8} , the disk 5D has a parity sequence (n + 4)
Is not started, and the reading is started in due course. The parity sequence (n
The readout of +4) is unnecessary, and time is wasted.

As can be seen by comparing FIG. 7A and FIG. 7B, the present RAID device has the following technical effects. In other words, the present RAID device issues a read stop command to the remaining one disk when a certain data block can be restored. Thus, the disk that has received the read stop command does not start useless reading. Thus, the disk performs only necessary reading. Therefore, the present RAID device can transmit the requested data to the host device as soon as possible. That is, in the example of FIG. 7 (a), at time t _9, the parity sequence n, (n + 2), (n +
4) and (n + 1) four data can be transmitted to the host device. On the other hand, in the example of FIG.
Since C and 5D perform useless reading,
At time _{t 9, n, (n +} 2) and (n + 4) 3 pieces of data can only transmit to the host device. As is clear from the above, according to the present RAID device, the amount of data that can be read per unit time increases, and data can be continuously transmitted to the host device. Therefore, when the RAID device handles video data, the video reproduced on the host device side is less likely to be interrupted.

The disk shown in FIG. 6 does not process the second read request in the order of arrival, but switches the read order. Therefore, a plurality of second read requests may be waiting for read processing in the disk. Further, as is apparent from the above description, the control unit 7 may cancel the second read request waiting for processing in the disk. However, the disk may not be able to stop processing of only a specific one from a plurality of second read requests waiting to be processed. In this case, the control unit 7 temporarily suspends all processing of the second read request in the disk. Then, the control unit 7 issues a second read request again. However, at this time, the control unit 7 does not issue the second read request for which the processing must be stopped. Thus, the control unit 7 can cancel only the specific second read request.

(Second Embodiment) Next, a RAID device according to a second embodiment of the present invention will be described. Since the configuration of the present RAID device is the same as that shown in FIG. 1, FIG. 1 will be referred to below. Further, in order to make the technical effect of the present embodiment easier to understand, as shown in FIG. 6, any one of the disks 5A to 5D and 5P does not execute reading in the order of arrival of the second reading request. The read order is changed so that the seek distance of the read head becomes shorter.

The RAID device performs the write processing described in the first embodiment every time transmission data from the host device arrives. In some cases, the host device wants to read data from the RAID device. In such a case, the host device transmits a first read request specifying a data storage location to the RAID device. The RAID device starts the read process in response to the arrival of the first read request. Since this reading process is a characteristic process of the second embodiment, referring to the flowchart shown in FIG.
This will be described in detail. Note that the flowchart of FIG. 8A partially includes the same steps as those of FIG. 4A. Therefore, in FIG.
Steps similar to those shown in FIG. 4A are denoted by the same step numbers, and description thereof will be simplified.

The controller 7 issues a second read request set in response to the first read request (steps S1 and S2). When the control unit 7 issues the second read request, the control unit 7 creates one issue time table 71 as shown in FIG. 9 in the internal storage area (step S21). By the way, as described in the first embodiment, the buffer area 3A _i (see FIG. 2) is included in the second read request transmitted to the SCSI interface 4A.
Specifies whether to store the data block from the disk 5A. Other SCSI interfaces 4B-4
Similarly, the second read request sent to D and 4P includes in any of the buffer areas 3B _{i to} 3D _i and 3P _i
It is instructed whether to store the data blocks read from the disks 5B to 5D and 5P. The issue time table 71 describes in which of the buffer areas 3A _{i to} 3D _i and 3P _{i the} parity sequence to be read this time is stored. Further, the time t _ISSUE at which the control unit 7 issues the second read request is also described in the issue time table 71.

The control unit 7 executes the processing described in the first embodiment (see FIG. 4B) and the like, and transmits data requested by the host device. Here, the processing when the four first read completions have arrived is not directly related to the present embodiment, and thus the description thereof is omitted.

By the way, the control unit 7 sets the issue time t _ISSUE
Starting from the time when four first read completions must arrive (hereinafter, this time is referred to as a time limit T
Referred to as _{LIMI T)} stored in advance. That is, the time limit T
_LIMIT is the time during which at least four disks must have completed reading since the second read request was issued. If the completion of reading from any two disks exceeds the time limit T _LIMIT , the transmission of data requested by the host device will be delayed. As a result, when the RAID device handles video data, the video reproduced on the host device side is interrupted. Here, as described in the first embodiment, the present R
The AID device attempts to read data blocks and redundant data from the five disks 5A-5D and 5P. However, the present RAID device can transmit read-requested data to the host device as long as four data blocks are provided or three data blocks and one redundant data are provided. . Therefore, the second
If at least four disks have completed reading before the time limit T _LIMIT has elapsed since the issuance of the read request, data transmission to the host device is not delayed. vice versa,
If the reading has not been completed on the two disks when the time limit T _LIMIT has elapsed since the issuance of the second read request, data transmission to the host device is completely delayed. Therefore, reading of the other three disk devices is useless. In order to eliminate this useless reading, the control unit 7 executes a process according to a flowchart shown in FIG.

First, the control section 7 determines whether or not four first read completions have arrived before the time limit T _LIMIT (step S31). The determination in step S31 is
It is performed as follows. First, the control unit 7 controls the present time t _PRE internally measured at a predetermined timing.
And the issue time t _ISSUE described in the issue time table 71 of FIG. 9 is selected. The control unit 7 stores the above-described time limit T _LIMIT in advance. The control unit 7 determines (t
_PRE− t _ISSUE )> T _LIMIT , buffer area information 3A _{i to} 3D _i and 3P _i (see FIG. 9) described together with the currently selected issue time t _ISSUE are obtained. Incidentally, each first read completion includes information indicating the buffer areas 3A _{i to} 3D _i and 3P _i as described above. When the first read completion arrives, the control unit 7 sets the buffer areas 3A _i to 3A
The information indicating D _i and 3P _i is extracted and held. Next, the control unit 7 compares the buffer area information 3A _{i to} 3D _i and 3P _i obtained from the issue time table 71 with the buffer area information obtained from the first arrival completion of the currently arrived first read. Based on the comparison result, the control unit 7 can determine whether four first read completions have arrived before the time limit T _LIMIT .

If it is determined in step S 31 that four first read completions have arrived before the time limit T _LIMIT , the control unit 7 deletes the currently selected issue time table 71 (step S 31). S33), the processing of FIG. 8B ends. On the other hand, if the four first read completions have not arrived before the time limit T _LIMIT , the control unit 7
From the comparison result of the information of the buffer area, a disk (one of 5A to 5D and 5P) for which reading has not been completed is specified. The control unit 7 issues a read stop command in order to stop reading of the disk for which reading has not been completed (step S32). The disk that has not yet been read responds to the read abort command,
Abort the read that is currently continuing or has not yet started. Subsequent to step S33, the control unit 7 deletes the currently selected issue time table 71 (step S33), and ends the processing in FIG. 8B.

The general read processing of the present RAID device has been described above. A specific example of the reading process of the present RAID device will be described with reference to FIG. Hereinafter, the case where the host device requests data reading in the order of the parity sequences n and (n + 1) shown in FIG. 2B will be described as an example. Here, FIG.
In this RAID device, parity sequence n to (n + 2)
FIG. 5 is a schematic diagram showing the timing at which is read on the time axis.

[0083] The control unit 7 in response to a request from the host device, a second read request to read the parity sequence n, and issues a time t ₁₀ (see FIG. 10 (a)). After that, the control unit 7 creates one issue time table 71 shown in FIG. 9 for the process of reading the parity sequence n (step S21 in FIG. 8A). For convenience, the issue time table 71 which is created this time, referred to as the issue time table 71 _n. The issue time table 71 _n includes “3A ₁ ”, “3B ₁ ”, and “3
C ₁ ", and" 3D ₁ "and" 3-Way ₁ "is described. Further, the issue time table 71 _n stores the issue time t
“T ₁₀ ” is described as _ISSUE . Similarly, the second read request to read the parity sequence (n + 1), and a second read request to read the parity sequence (n + 2) is issued after time t _10. Further, parity sequences (n + 1) and (n +
The issue time tables 71 are created one by one for the read process of 2).

To the disks 5A to 5D and 5P, the second read requests of the parity sequences n, (n + 1) and (n + 2) are input, respectively. Each disk 5A-5D
And 5P independently determine the reading order. Therefore, for example, in the disk 5A, the parity sequence is n → (n
+2) → (n + 1) in an order of reading. The reading order on the disk 5B is (n + 2) → n →
(N + 1). The reading order on the disk 5C is as follows.
(N + 2) → (n + 1) → n. The reading order on the disk 5D is n → (n + 2) → (n + 1). The reading order on the disk 5P is n → (n + 1) → (n +
2). According to the above reading order, FIG.
As shown in (1), first, the disks 5A, 5D and 5P read the data block and the redundant data whose parity sequence is n (see the dot shading). Further, the disks 5B and 5C read the (n + 2) data block (see the hatched portion with hatching).

What should be noted here is the process of reading the disks 5A and 5P. Now, with the passage of time from the time t _10, and as a result, it becomes time t _11. This time t ₁₁ is,
Suppose that t ₁₁ = t ₁₀ + T _LIMIT . And now,
The current time t _PRE has passed the time t ₁₁ . Therefore,
(T _PRE −t _ISSUE )> T _LIMIT is satisfied. Therefore, the control unit 7, issue time t _ISSUE (= t ₁₀₎ information in the buffer area described with "3A _1" ~ "3
D ₁ "and obtain from the issue time table 71 _n" 3-Way ₁ "(see FIG. 9). Further, in the time t _11, the disk 5D is already finished reading the data blocks of the parity sequence n. Therefore, the control unit 7, from the buffer unit 3D, has received only the first read completion specifying a buffer area 3D _1. Thus, the control unit 7
Indicates that two or more first read completions have not arrived before the time limit T _LIMIT and the disks 5A to 5C and 5P
In step S32, it is determined that the reading of the parity sequence n is not completed (step S32). Thereby, the control unit 7 specifies the disks 5A to 5C and 5P for which reading of the parity sequence n takes too long. The control unit 7 issues a read stop command to the disks 5A to 5C and 5P specified as described above (step S32), and stops reading the parity sequence n from the disks 5A to 5C and 5P.

[0086] Therefore, the disc 5A and 5P, immediately after time t _11, as indicated by the × mark, stops the reading of the parity sequence n. Therefore, the disk 5A starts reading the parity sequence (n + 2) (see the hatched portion). The disk 5P starts reading the parity sequence (n + 1) (see the hatched portion of the vertical line). Disk 5B initially has a parity sequence of (n
+2) → n → (n + 1), but it is determined that the read operation of the parity sequence n is not started in response to the read stop command. Therefore, when the reading of the parity sequence (n + 2) is completed, the disk 5B starts reading (n + 1).
The disk 5C also skips the initially determined order and does not read the data block of the parity sequence n.

The above has described a specific example of the reading process of the present RAID device. As described above, the control unit 7 of the present RAID device determines that two or more data blocks belonging to the same parity sequence, or at least one data block and one redundant data belonging to the same parity sequence, It may detect that it is not read within _LIMIT . In this case, the control unit 7 specifies a disk for which the reading of the parity sequence has not been completed. Control unit 7
Issues a read stop command to all the specified disks to stop reading from the disks. This is a characteristic process of the present RAID device.

Next, in order to highlight the features of the present RAID device, the RA that does not execute the flowchart of FIG.
The reading process by the ID device will be described with reference to FIG. Here, FIG. 10B shows a parity sequence n in a RAID device that does not execute the processing of FIG.
The timing at which .about. (N + 2) are read is shown on the time axis. The conditions in FIG. 10B are the same as those in FIG. 10 except that the RAID device does not execute step S12.
The same conditions as (a) are set. That is, the host device requests data reading in the order of the parity sequences n to (n + 2) under the same conditions as described above.

The control section 7 reads out the parity sequence n.
In response to the second read request at time t _Ten(Figure 1
0 (b)). Similarly, parity sequence (n + 1)
A second read request to read data, and parity
Second read request for reading sequence (n + 2)
At time t_TenIssued at a later time.

The reading order of each of the disks 5A to 5D and 5P is determined independently. Therefore, for example, in the disk 5A, the parity sequence is n → (n + 2) → (n + 1)
Attempts to read in the order The reading order on the disk 5B is (n + 2) → n → (n + 1).
The reading order on the disk 5C is (n + 2) → (n +
1) → n. The reading order on the disk 5D is n →
(N + 2) → (n + 1). The reading order on the disk 5P is n → (n + 1) → (n + 2). According to the above reading order, as shown in FIG. 10B, each of the disks 5A to 5D and 5P reads the parity sequences n, (n + 1) and (n + 2), respectively.

As described above, the RAID device shown in FIG.
The processing of (b) is not executed. Therefore, disk 5A
In 5P and 5P, even if the reading process of the parity sequence n takes a long time (time longer than the time limit T _LIMIT ), the respective reading processes are not stopped. Furthermore, in this case, there is a very high possibility that the data blocks of the parity sequence n stored in the disks 5A and 5P are broken. Therefore, the RAID device uses the parity sequence n
Cannot be assembled and transmitted to the host device. It should be noted here that the disks 5B and 5C start reading data blocks of the parity sequence n when the turn comes, even though the data of the parity sequence n cannot be transmitted.

As is clear from FIGS. 10A and 10B, the RAID apparatus according to the present embodiment executes the processing of FIG. 8B, and transmits data to the host apparatus. When it is determined that the parity sequence cannot be read, all reading of the parity sequence is stopped. Therefore, FIG.
In the example of (a), disks 5A, 5B, 5C and 5P
Can start reading the next parity sequence earlier than in the case of FIG. by this,
The disks 5A and 5P can abort the useless reading without continuing, and can immediately start the next reading. In addition, the disks 5B and 5C skip reading of a parity sequence that cannot transmit data to the host device, and start reading of the next parity sequence. As a result, the amount of data that the RAID device can read per unit time increases. By this, this RAID
The device can continuously send data to the host device. As a result, when the present RAID device handles video data, the video reproduced on the host device side is less likely to be interrupted.

(Third Embodiment) In the previous embodiment, the control unit 7 immediately issues a restoration instruction to the parity operation unit 6 when redundant data is read. However, since the parity operation unit 6 executes a large amount of operations, the more the number of executions of the parity operation, the more the load is placed on the RAID device. Therefore, the third
In the embodiment, the control unit 7 controls the issuance timing of the restoration instruction to reduce the number of executions of the parity operation.
Implement an AID device. FIG.
It is a block diagram showing an AID device. RAID of FIG.
The device differs from that of FIG. 1 in that the control unit 7 includes a first timer 72. Since other configurations are the same, in FIG. 11, the same components as those shown in FIG. 1 are denoted by the same reference numerals, and description thereof will be simplified.

The present RAID device performs the write processing described in the first embodiment every time transmission data from the host device arrives. In some cases, the host device wants to read data from the RAID device. In such a case, the host device transmits a first read request specifying a data storage location to the RAID device. The RAID device starts the read process in response to the arrival of the first read request. This readout process is a characteristic process of the third embodiment, and will be described in detail with reference to the flowcharts shown in FIGS. FIG.
The flowchart of (a) is the same as that of FIG. Therefore, in FIG. 12A, the same steps as those shown in FIG. 2A are denoted by the same step numbers, and description thereof will be omitted. By executing the flowchart of FIG. 12A, a second read request (a read request for a parity sequence) is issued due to the first read request (steps S1 and S2). Further, the issuance time table 71 of FIG. 9 is created for the issued second read request (step S21).

The second command issued by the process shown in FIG.
Is transmitted to the disks 5A to 5D and 5P in the same manner as described in the first embodiment. Each of the disks 5A to 5D and 5P reads data blocks and redundant data in response to the second read request. The read data block and redundant data are
The data is stored in the buffer units 3A to 3D and 3P through the SCSI interfaces 4A to 4D and 4P. After storing the data blocks and the redundant data, the buffer units 3A to 3D and 3P transmit the first read completion to the control unit 7 to notify that the reading of the disks 5A to 5D and 5P is completed.

When the four first read completions arrive (FIG. 12B; step S11), the control unit 7 detects and holds the time t _{4th at} which the fourth first read completion arrives. (Step S41). Next, the control unit 7 determines whether the reading of the redundant data has been completed (Step S42). Now, consider a case where the redundant data has not been read (that is, a case where the first read completion has been received from the buffer units 3A to 3D). In this case, reading the redundant data being executed by the disk 5P is useless. Therefore, the control unit 7 issues a read stop command to stop the useless reading (step S12). Further, the control unit 7 issues a second read completion (step S16). The selector 2 extracts the data block from the buffers 3A to 3D in response to the completion of the second read, and assembles data to be transmitted to the host device. The selector 2 transmits the assembled data to the host device via the host interface 1.

It is also assumed that, in step S42, the redundant data has already been read (that is, the first read completion has been received from the buffer 3P). In this case, the control unit 7 proceeds to step S43, and firstly, the timeout value V to be set in the first timer 72.
Calculate _TO1 . Here, the timeout value V _TO1 will be described in detail.

Now, a simulation test is performed on the manufactured RAID device. In this simulation test, a second read request is repeatedly issued from the control unit 7 to one disk (one of 5A to 5D and 5P). Then
The control unit 7 receives the first read request for each issued second read request.
Read completion has arrived. In this simulation test, a time t from issuance of the second read request to arrival of completion of the first read is measured. This time t can be equated with the time required for reading from one disk. However, since each measured time t varies with a certain deviation, a probability distribution curve f (t) as shown in FIG. 13A can be obtained. In FIG. 13A, the horizontal axis indicates time t. The vertical axis represents the probability f (t) that the reading of the disk is completed at time t. Therefore, the probability P (t) that the first read completion has arrived by the time t after the issuance of the second read request is expressed by the following equation (3).

[0099]

(Equation 1)

Since the present RAID device includes five disks, all of the five first read completions have arrived by time t after the second read request of a certain parity sequence is issued. The probability P _all (t) is expressed by the following equation (4). P _all (t) = {P (t)} ⁵ … (4)

Now, the probability P_all(T) is a predetermined probability P₀When
Time t₀And In this case, P_all(T₀)
= P₀Becomes Where t₀And P₀Is the RAID device
An appropriate value is selected according to the design requirements.
Guarantees that data will be sent continuously to the host device.
Value must be chosen. In other words, t ₀
And P₀Guarantees that the video is not interrupted on the host device side
It is a possible value. As can be seen from the above, this RAID device
In one embodiment, the reading of a certain parity sequence is performed by the second reading.
Time t from the issue time of the request₀By the probability P₀Complete with
It is expected to be. In this sense, the present embodiment
Then, this time t₀With the expected completion value t₀Call
You. The control unit 7 calculates the expected completion value t₀The timeout
Value V _TO1Is stored in advance in order to calculate.

Now, since four first read completions have arrived at the control unit 7, each of the disks 5A to 5A
The progress of reading in D and 5P is, for example, as shown in FIG.
3 (b). In FIG. 13B, due to the second read request issued at time t _ISSUE ,
Reading is started on each of the disks 5A to 5D and 5P. Then, at the time t _4th , the disks 5A, 5A
The reading has been completed in B, 5D and 5P. Here, the reading of a certain parity sequence, based on the time t _ISSUE, to complete the expected value t _0, the probability P
₀ , it is expected that the reading will be completed at ₀ , so that reading of the disk 5C is performed as shown in FIGS.
_With a probability of ₀ , it is expected to be completed by the time (t _ISSUE + t ₀ ).

Therefore, in step S43, the control unit 7 first obtains the time t _4th obtained in step S41, the time t _ISSUE in the issue time table 71, and the expected completion value t ₀ stored in advance. Next, Δt ₀ − (t _4th −t
_ISSUE ) is calculated, whereby the value of FIG.
In (b), the time margin t _MARGIN indicated by a white arrow
Can be calculated. The control unit 7 sets the value of the calculated time margin t _MARGIN as the timeout value V _TO1 to the first
(Step S43). Thus, the first timer 72 is started, and starts counting down.

Next, the control section 7 determines whether or not the remaining first read completion has arrived (step S44).
In other words, the control unit 7 determines whether reading of the remaining data blocks is completed and four data blocks are completed. Now, a case where four data blocks are completed will be described with reference to FIG. in this case,
Before time margin T _MARGIN runs out from time t _4th (that is, before time (t _ISSUE + t ₀ )), disk 5A
５5D data blocks are all prepared. Further, since the reading of the redundant data has been completed, the control unit 7 does not need to issue the read stop command. Therefore, the control unit 7
Proceeds directly from step S44 to step S16. Then, the control unit 7 issues a second read completion (step S16). The selector 2 extracts the data block from the buffers 3A to 3D in response to the completion of the second read, and assembles data to be transmitted to the host device. The selector 2 transmits the assembled data to the host device via the host interface 1. The first timer 72 stops counting down as necessary.

On the other hand, a case where the remaining first read completion has not arrived in step S44 will be described. In this case, the control unit 7 determines whether the first timer 72 has timed out (step S45). That is, the control unit 7 determines whether or not the time margin T _MARGIN has elapsed from the time t _4th . If the first timer 72 has not timed out, the control unit 7 returns to step S44 and determines again whether or not the remaining first read completion has arrived. On the other hand, the case where the first timer 72 times out will be described with reference to FIG. in this case,
The control unit 7 knows that reading of the remaining one data block has not been completed before the time margin T _MARGIN elapses from the time t _4th . In the example of FIG. 14B, the disk 5C is still being read. Time margin T
_With the progress of _MARGIN , the control unit 7 determines that the RAID device cannot continuously transmit data when waiting for the remaining first read request any more. Therefore, the control unit 7
Proceeds from step S45 in FIG. 12 to step S14, issues a restoration instruction immediately after the time (t _{IS SUE} + t ₀ ), and requests the parity operation unit 6 to execute the parity operation (step S14). When the parity calculation is completed, the parity calculation unit 6 issues a restoration completion indicating that, and
It is transmitted to the control unit 7. When receiving the completion of the restoration (step 15), the control unit 7 determines that the four data blocks are completed and the data requested by the host device can be transmitted. Next, the control unit 7 issues a read stop command to stop useless reading in the remaining disk devices (step S12). Further, the control unit 7 issues a second read completion (step S1).
6). The selector 2 responds to the completion of the second read,
Retrieve data blocks from buffers 3A-3D,
The data to be transmitted to the host device is assembled. The selector 2 transfers the assembled data to the host interface 1
To the host device.

As described above, the RAI according to the present embodiment is
The D device is different from that of the first embodiment in that the fourth device
Immediately after the first read completion arrives,
Do not restore missing data blocks. In other words, this RAID
The device may return to the above after four first read completions have arrived.
Time margin T_MARGINJust read the rest of the data block
Wait for delivery to complete. This time margin T_MARGIN
For the first time when there is no more
An indication is issued. This time margin T_MARGINWithin the remaining
Data blocks are read, four data blocks are read.
The locks will be aligned. In this case, the RAID device
Data can be stored in the host device without performing parity operation.
Can be sent. It is important to note that
Room T _MARGINHas been described with reference to FIG.
In this way, it is possible to guarantee that the video is not interrupted on the host device side.
Value t₀ Is calculated based on More time
T _MARGINMeans that reading of the remaining data blocks is
Indicates the waiting time. Therefore, in most cases, four
The data block has time margin T_MARGINInside buffer part 3
A to 3D. This RAID device requires a large amount of computation
Hardly needs to perform parity operations that require
The number of times of execution can be minimized. Ma
When a fourth first read completion arrives, a redundant
Since the probability that data has not been read is 1/5,
With a 1/5 probability (that is, reading 5 sets of parity sequences)
Without performing a parity operation)
Data can be sent to the host device
You.

(Fourth Embodiment) By the way, in the previous embodiment, the control unit 7 issues a restoration instruction without being aware of the current state of the parity operation unit 6. Therefore, while the parity operation unit 6 is executing the parity operation,
The control unit 7 may issue the next restoration instruction to the parity operation unit 6 in some cases. However, the parity calculation unit 6 can process only one restoration instruction in a certain time zone,
Other restoration instructions cannot be accepted. Therefore, in the fourth embodiment, the control unit 7 controls the issuance timing of the restoration instruction, and during the execution of the parity operation,
A RAID device in which a new restoration instruction is not issued is realized.

FIG. 15 is a block diagram showing a RAID device according to the fourth embodiment of the present invention. RAI of FIG.
The D apparatus is different from the apparatus in FIG. 1 in that the control unit 7 further includes a reservation table 73 and a second timer.
Since other configurations are the same, in FIG. 15, the same components as those shown in FIG. 1 are denoted by the same reference numerals, and description thereof will be simplified.

The present RAID device performs the write processing described in the first embodiment every time transmission data from the host device arrives. In some cases, the host device wants to read data from the RAID device. In such a case, the host device transmits a first read request specifying a data storage location to the RAID device. The RAID device starts the read process in response to the arrival of the first read request. This reading process is a characteristic process of the fourth embodiment, and will be described in detail with reference to the drawings.

As shown in FIG. 12A, the second read request (parity sequence read request) is issued in response to the first read request (steps S1 and S2). Further, the issuance time table 71 of FIG. 9 is created for the issued second read request (step S21).

The second command issued by the process shown in FIG.
Is transmitted to the disks 5A to 5D and 5P as described in the first embodiment. Each of the disks 5A to 5D and 5P reads data blocks and redundant data in response to the second read request. The read data block and redundant data are stored in the SCSI
The data is stored in the buffer units 3A to 3D and 3P through the interfaces 4A to 4D and 4P. Buffer unit 3
When the data blocks and the redundant data have been stored, the A to 3D and 3P transmit the first read completion to the control unit 7 to notify that the reading of the disks 5A to 5D and 5P has been completed.

The control unit 7 periodically performs the processing procedure shown in the flowchart of FIG. The flowchart in FIG. 16 includes steps that are partially similar to those in FIG. Therefore, in FIG. 16, FIG.
Are given the same step numbers, and description thereof is omitted. The control unit 7 has four first
When the read completion of the data has arrived (step S1 in FIG.
1) The arrival time t _4th is held (step S41).
Next, the controller 7 determines whether or not redundant data has been read (step S42). If the redundant data has not been read out yet, as described in the fourth embodiment, the control unit 7 stops useless reading on the disk 5P (step S12) and issues a second reading completion (step S12). Step S16). As a result, the data assembled by the selector 2 is transmitted to the host device via the host interface 1.

The case where the redundant data has already been read in step S42 will be described. In this case, there is a possibility that the parity operation unit 6 executes the parity operation. Therefore, the control unit 7 writes necessary information into the reservation table 73 (step S51). Reservation table 73
As shown in FIG. 17, the usage time zone and the buffer area are written as necessary information. The use time zone is
The time zone in which the control unit 7 uses the parity operation unit 6 is shown.
The buffer area indicates where data blocks and redundant data used by the parity operation unit 6 are stored. The control unit 7 stores the information of the buffer area included in the first read completion obtained in step S11 in the reservation table 7
Write to 3.

In step S51, the reservation table 73
Describes when the parity operation starts and ends. Therefore, the control unit 7 calculates a timeout value V _TO2 (= t _4th −t _S ) from the parity calculation start time t _S and the fourth arrival time (current time) t _4th . The control unit 7 sets the calculated timeout value V _TO2 in the timer 74 (step S52). As a result, the timer 74 is activated and starts counting down. When the timer 74 times out, the parity operation unit 6 completes the previous parity operation and is ready to accept the next parity operation. That is, when the timer 74 times out, the control unit 7 can issue a restoration instruction.

Next, the control section 7 determines whether or not the remaining first read completion has arrived (step S44).
First, a case where the remaining first read completion has arrived will be described. In this case, the four data blocks are
Before the timeout occurs, everything will be ready. Therefore, there is no need to perform a parity operation. However, the use time of the parity operation unit 6 is written in the reservation table 73. Therefore, the control unit 7 deletes the information of the use time zone and the buffer area described in step S51 (step S53). Further, since the reading of the redundant data has been completed, the control unit 7 does not need to issue the read stop command. Therefore, the control unit 7 issues a second read completion (step S16). As a result, the data assembled by the selector 2 is transmitted to the host device via the host interface 1.
Note that the timer 74 stops counting down as necessary.

Next, a case where the remaining first read completion has not arrived in step S44 will be described. In this case, the controller 7 determines whether the timer 74 has timed out (step S54). That is, the control unit 7
Determines whether the timeout value V _TO2 has elapsed from the time t _4th . When the timer 74 has not timed out, the control unit 7 returns to step S44 and determines again whether or not the remaining first read completion has arrived.

[0117] On the other hand, if the timer 74 has timed out, the control unit 7, from the time t _4th, time-out value V
It can be seen that the reading of the remaining one data block has not been completed by the time _TO2 has elapsed, and that the time zone in which the parity operation unit 6 can be used has been reached. Therefore, the control unit 7 proceeds from step S54 to step S12, and stops useless reading on the remaining disks (step S12). Further, the control unit 7 issues a restoration instruction and requests the parity operation unit 6 to execute a parity operation (step S14). When the parity calculation is completed, the parity calculation unit 6 issues a restoration completion indicating that, and transmits the restoration completion to the control unit 7. When receiving the completion of restoration (step 15), the control unit 7 knows that the information on the use time zone and the buffer area described in step S51 is no longer necessary. Therefore, the control unit 7 deletes unnecessary information from the reservation table 73 (step S53).

Further, the control unit 7 determines that the four data blocks have been collected upon arrival of the completion of restoration, and that the data requested by the host device can be transmitted. Next, the control unit 7 issues a second read completion (step S16). As a result, the data assembled by the selector 2 is transmitted to the host device via the host interface 1.

The general read processing of the present RAID device has been described above. A specific example of the reading process of the present RAID device will be described with reference to FIG. Note that, in the following, the host device determines whether the parity sequence n shown in FIG.
A case where data reading is requested in the order of (n + 2) and (n + 4) will be described as an example. Here, FIG.
In this RAID device, parity sequence n, (n + 2)
FIG. 9 is a schematic diagram showing the timing at which (n + 4) and (n + 4) are read and the reservation status of the parity operation unit 6 on the time axis.

The disks 5A to 5D and 5P receive the second read requests for the parity sequences n, (n + 2) and (n + 4), respectively. Now, to simplify the description, it is assumed that each of the disks 5A to 5D and 5P reads the parity sequence in the order of arrival of the second read request. Further, as a precondition, the parity operation currently being executed is completed at time t _12, (see the hatched portion of the left edge) listed in the reservation table 73. Under the above conditions, each of the disks 5A to 5D and 5P first reads the parity sequence n. In the example of FIG. 18, since the disk 5B has finished reading the time t _12, the control unit 7, first read a complete fourth one arrives at time t ₁₂ (step S11 in FIG. 16). Control unit 7
Stores the time t ₁₂ as the arrival time t _{4t h} (step S41). Since the disk 5P has already read the redundant data, the control unit 7 proceeds to step S5.
Run 1, the reservation table 73 in FIG. 17, and writes the time t ₁₃ ~t ₁₄ as use time zone. Further, the control unit 7
Writes 3A _i , 3B _i , 3C _i and 3P _i as buffer areas (step S51). Further, the control unit 7 calculates a timeout value V _TO2 (T ₁ = t ₁₃ −t ₁₂ ) and sets the timeout value V _TO2 to the second timer 7.
4 (step S52).

[0121] at time t _12, the disk 5D is to continue with the reading of the data block. However, it is assumed that this reading is not completed until the time t _13. This assumption under the control unit 7, the second timer 74 times out at time t _13, stops the reading of the disk 5D, issues a restore instruction to the parity operation unit 6 (steps S12 and S14). The parity calculator 6 calculates the time t ₁₃
During ~t _14, restores the data blocks recorded on the disc 5D. Therefore, the control unit 7 stores the time t
_{At 14} , the completion of restoration from the parity calculation unit 6 arrives (step S 15), and therefore, from the reservation table 73, the use time zones t _{13 to} t ₁₄ and the buffer areas 3A _i , 3B _i ,
3C _i and 3P _i are deleted (step S53). Thereafter, the control unit 7 issues a second read completion (step S16).

Each of the disks 5A to 5D and 5P
Is the parity sequence after the completion of reading the parity sequence n.
The reading of (n + 2) is started. Example of FIG.
Then, the completion of the first read from the disk 5D is determined at time t
_FifteenArrives at the control unit 7. The control unit 7 determines the time t_FifteenArrive
Time t_4th(Steps S11 and S4
1). Time t_FifteenIn, redundant data has been read
Therefore, the control unit 7 stores in the reservation table 73
Interzone t_Fifteen~ T₁₈And buffer area 3A_i, 3C _i,
3D_iAnd 3P_iIs written (step S
51). Here, time t_FifteenIs the time t₁₄Later than
Need attention. That is, the time t_FifteenNow, the parity operation
The unit 6 does not perform the parity operation. Therefore, time
Out value V_{TO Two}Is "0" (step S52),
The control unit 7 immediately reads the disk 5B that is currently continuing.
And a restoration instruction is issued to the parity operation unit 6.
(Steps S12 and S14). Explanation of subsequent processing
Is omitted from the above because it is clear from the above.

Further, each of the disks 5A to 5D and 5P
After the completion of the reading of the parity sequence (n + 2),
The reading of the key sequence (n + 4) is started. Day
Completion of the first read from the disk 5D occurs at time t₁₆To (time
t₁₈Before) the control unit 7. Time t
₁₆Since the redundant data has already been read,
7 is the time t in the reservation table 73 as the use time zone.₁₈
~ T₁₉Write. Further, the control unit 7 controls the buffer area
3A as identification information of_i, 3C_i, 3D_iAnd 3P
_iWrite. Further, the control unit 7 sets the timeout value V
_TO2(T_Two= T ₁₈-T₁₆) To calculate the timeout
G value V_TO2Is set in the second timer 74 (step S
52).

However, what should be noted here is that
Completion of the first read from the clock 5B is performed at time t₁₇(Time t
₁₈Before reaching the control unit 7). That is,
Completion of the first reading is performed by the second timer 74.
Arrives at the control unit 7 before going out. Therefore, the control unit
7 does not issue the restoration instruction, the parity operation unit 7
Time t₁₈~ T₁₆The parity operation that was to be executed
Do not execute (refer to the dotted line x). Therefore, the control unit 7
From the reservation table 73, use time zone t₁₈~ T₁₉Line
3A as identification information of the buffer area_i, 3B_i,
3C_iAnd 3P _iIs deleted (step S53), and the
Then, a read completion of No. 2 is issued (step S16).

As described above, the RAI according to the present embodiment is
Unlike the first embodiment, the D device writes the use time zone of the parity operation unit 6 to the reservation table 73 when four first read completions arrive. As the information on the use time zone, the time zone after the end of the currently executed parity calculation is written. Further, when the use time zone comes, the control unit 7 issues a restoration instruction. Therefore, the control unit 7 does not issue a restoration instruction to the parity operation unit 6 during execution of the parity operation. This can prevent an excessive load from being applied to the RAID device. When the remaining one data block arrives before the second timer 74 times out, the control unit 7 assembles and transmits the data from the four data blocks without issuing a restoration instruction. Issue a second read completion. As a result, the present RAID device can minimize the number of executions of the parity operation requiring a large amount of operation to a necessary minimum.

(Fifth Embodiment) FIG. 19 is a block diagram showing a RAID device according to a fifth embodiment of the present invention. The RAID device of FIG. 19 differs from that of FIG. 1 in that the control unit 7 further includes a failed block table 75. Since other configurations are the same, FIG.
In the figure, the same components as those shown in FIG. 1 are denoted by the same reference numerals, and the description thereof will be simplified. However, it should be noted that the present RAID device does not make the issue time table 71 an essential component.

It should be noted that data blocks and redundant data are not recorded on the disks 5A to 5D and 5P of this embodiment as shown in FIG. That is, the present RAID device is typically constructed based on level 5. At the level 5, the redundant data is not stored on a fixed disk (see FIG. 3), but is distributed and stored on the disks 5A to 5D and 5P included in the RAID device as shown in FIG.

In some cases, the host device wants to read data from the RAID device. In such a case, the host device transmits a first read request to the RAID device.
Further, the storage location of the data is specified by the first read request. The RAID device starts the read process in response to the arrival of the first read request. Since this reading process is a characteristic process of the present application, it will be described in detail with reference to the flowchart shown in FIG. However, FIG. 21 partially includes the same steps as those in FIG. Therefore, in FIG. 21, FIG.
Steps corresponding to the above steps are denoted by the same step numbers, and the description is simplified. The first read request is input to the control unit 7 through the host interface 1 (Step S1). The control unit 7 obtains the data storage location from the first read request. The control unit 7 controls a parity sequence (four data blocks and redundant data) generated based on the data storage location based on the data.
Identify the storage location of Here, it should be noted that the process of obtaining the storage location of the parity sequence from the storage location of the data is well known and is defined by the RAID architecture.

Next, the control unit 7 controls the four disks (5A
To 5D and 5P) determines whether or not reading of the four data blocks to be read this time has failed in the past (step S61). For this determination in step S61, the failed block table 75 is referred to. As shown in FIG. 22, the failed block table 75 lists storage locations of data blocks that have previously failed to be read. Note that the failed block table 75 may list storage locations of data blocks from which the disks 5A to 5D and 5P have retried reading. The failed block table 75 includes the disks 5A to 5D and 5P.
The storage locations of the data blocks that have taken a certain time or longer to read may be listed.

First, a case where four disks have never failed to read four data blocks will be described. In this case, the control unit 7 determines that the reading of the four data blocks is unlikely to fail this time, and issues a second read request set to read the parity sequence (step S62). . However, in step S62, the second read request is issued only to four disks on which data blocks are recorded, and is not issued to the remaining one disk on which redundant data is recorded.

Next, a description will be given of a case where some disks have failed in reading data blocks in the past in step S61. In this case, the control unit 7 determines that reading of at least one of the four data blocks is likely to fail this time, and sets the second read request to read the parity sequence. Issue (step S63). However, step S
At 63, a second read request is issued to the four disks on which data blocks are recorded and the remaining one disk on which redundant data is recorded.

The control unit 7 controls each of the disks 5A to 5D
When the completion of the first read from the P and 5P arrives, the processing shown in FIG. 2B is executed. However, if reading of the data block fails during this process,
The storage location of the data block is added to the failed block table 75.

As can be seen from the above, in this embodiment,
The number of issued second read requests differs depending on the determination result of step S61. By issuing such a second read request, in the present embodiment, FIG.
Technical effects as shown in FIG. Here, FIG.
(A) shows a case where a set of five second read requests is always issued as described in the previous embodiment. FIG. 23B shows a case where the number of second read requests included in one set changes according to the determination result of step S61. However, FIG.
Shows a case where a set of four second read requests is issued to clarify the technical effects of the present embodiment.

First, in FIG. 23A, since the redundant data is read every time, if the time required to read one data block (or redundant data) is T, the parity series n to (n + 4) are read. If at least 5
* T time is required. However, in FIG.
Since no redundant data is read, four disks
While one parity sequence is being read, the remaining one disk can read another parity sequence. Therefore, the RAID device may be able to read the parity sequences n to (n + 4) in a time shorter than the time 5 * T. FIG. 23B shows the shortest case, in which the RAID device has parity sequences n to (n +
4) shows a case of reading at time 4 * T.

As described above, in the present RAID device, when a data block whose reading has failed in the past is read from the disk, redundant data is read. On the other hand, when reading a data block from the disk for which reading has not failed in the past, redundant data is not read. As a result, as described with reference to FIGS. 23A and 23B, the present RAID device increases the amount of data that can be read per unit time. Further, when the reading of the data block is likely to fail, the redundant data is read. When the reading of the data block fails, the parity operation can be executed immediately. And data can be sent as soon as possible.

(Sixth Embodiment) By the way, the disk 5
One of the causes of the delay of reading in A to 5D and 5P is that defects occur in the recording areas of the disks 5A to 5D and 5P. If a data block or redundant data continues to be stored in such a defective recording area, reading of the data block or the redundant data will be delayed every time. Thus, in the sixth embodiment, a RAID device that executes a so-called reassignment process is realized.
Here, the reassignment process is a process of allocating an alternative recording area to a defective recording area and re-storing a data block or redundant data stored in the defective recording area in a newly allocated alternative recording area. Means

FIG. 24 is a block diagram showing a configuration of a RAID device according to the sixth embodiment of the present invention. The RAID device of FIG. 24 differs from that of FIG. 1 in that it further includes a reassignment processing unit 8, an alternative recording area information management unit 9, an address information management unit 10, and an address conversion processing unit 11. Since the present RAID device is the same in other configurations, in FIG. 24, the same components as those shown in FIG. 1 are denoted by the same reference numerals, and description thereof will be simplified. Here, it should be noted that although not shown in FIG. 24 for convenience, the control unit 7
The third embodiment includes the first timer 72 described in the third embodiment.

As is well known, each of the disks 5A to 5D and 5P manages its own recording area in units of 512-byte sectors. Each sector is assigned a number called an LBA in advance. Here, LBA is Logic
This is an acronym for al Block Address (logical block address). At the time of the initial startup of the present RAID device, some sectors of the recording area of the disks 5A to 5D and 5P are reserved for the alternative recording area. The alternative recording area information management unit 9 manages a first table 91 shown in FIG. 25 to manage such an alternative recording area.
In FIG. 25, the first table 91 describes LBAs that specify the reserved alternative recording areas.

A host device (not shown) is provided outside the RAID device. The host device is connected to the host interface 1 and requests the RAID device to write data or requests the RAID device to read data. This RAID device performs the same write processing as in the first embodiment. here,
When the present RAID device is constructed based on a RAID level 3 architecture, as shown in FIG.
The redundant data is fixedly recorded only on the disk 5P. When the present RAID device is constructed based on RAID level 5, as shown in FIG. 20, redundant data is recorded in a distributed manner on disks 5A to 5D and 5P. However, it should be noted that the data block and the redundant data are written in a recording area other than the alternative recording area when the data is written by the RAID device.

[0140] The host device also operates in the same manner as described above.
Of the RAI
Request to D device. At this time, the host device requests reading in units of a parity sequence. For example, to read five parity sequences n to (n + 4) (see FIG. 3),
The host device needs to send five first read requests to the RAID device. Here, the parity sequence is
As described above, this means a combination of a data block and redundant data generated based on the same 2048-byte data. As described above, each first read request includes information for specifying the storage location of the parity sequence to be read. In the sixth embodiment, LBA is used as information for specifying a storage location.

This RAID device starts read processing in response to the arrival of the first read request. Since this reading process is a characteristic process of the sixth embodiment,
This will be described with reference to FIGS. 26 and 28. Here, FIG.
6 is a flowchart showing a processing procedure of the control unit 7 after the arrival of the first read request. The flowchart of FIG. 26 partially includes the same steps as those of FIG. Therefore, in FIG. 26, the same steps as those shown in FIG. 12 are denoted by the same step numbers, and description thereof will be simplified. FIG. 28 is a flowchart showing a processing procedure of the control unit 7 after one read response arrives. The first read request is input to the control unit 7 through the host interface 1 (FIG. 26;
Step S1). The control unit 7 extracts the LBA as the storage location of the parity sequence to be read this time from the input first read request. The control unit 7 outputs the extracted L
BA is notified to the address conversion processing unit 11 (step S
71). The address translation processing unit 11 has a RAID level 3
Alternatively, an operation based on the policy of 5 is executed to derive an original LBA (described later) of a data block and redundant data from the notified storage location (LBA) of the parity sequence. As described above, the data block and the redundant data are stored in the RAI at the time of a write request from the host device.
The D device writes the data on the disks 5A to 5D and 5P. The original LBA means an LBA indicating a location where a data block or redundant data is stored immediately after a write request is made.

The operation of the address conversion processing section 11 will be described below. Since the present RAID device executes reassignment processing, the storage locations of data blocks and redundant data may change. In the following description, the current LB
A means an LBA indicating a current storage location of a data block or redundant data. Since the storage location may be changed as described above, the address conversion processing unit 11
Is notified of the storage location of the parity sequence, accesses the address information management unit 10 and specifies the original LBA of the data block or the redundant data. The address information management unit 10 stores the second table 1 as shown in FIG.
01 is managed. In FIG. 27, in the second table 101, for each original LBA of a data block or redundant data, the current LBA of the data block or the redundant data is registered. The current LBA registration process will be described later.

Here, when the current LBA is registered for the original LBA derived this time, the address conversion processing section 11 takes out the current LBA from the second table 101. Address translation processing unit 11
Determines that the data area to be read this time or redundant data is stored in the recording area indicated by the extracted current LBA. On the other hand, the original L derived this time
If the current LBA is not registered for the BA, the address conversion processing unit 11
It is determined that the data area or redundant data to be read this time is stored in the recording area indicated by. As described above, the address conversion processing unit 11 specifies the data block to be read this time and the LBA indicating the correct recording area of the redundant data. The address conversion processing unit 11 notifies the control unit 7 of the LBA specified as described above.

The control unit 7 issues a second read request set to read a parity sequence (four data blocks and redundant data) using the LBA notified by the address translation processing unit 11 ( Step S
2). In the present embodiment, as shown in FIG. 3 or FIG. 20, the parity sequence has five disks 5A to 5D and 5 disks.
Since the data is distributed and stored in P, five second read requests are issued. As described in the first embodiment, the second read request includes an LBA as a storage location for a data block or redundant data, and buffer areas 3A _i to 3A for storing the read data block or redundant data.
D _i and information indicating 3P _i . The second read request includes the SCSI interfaces 4A to 4D and 4P
Are transmitted one by one. When transmitting the second read requests to the SCSI interfaces 4A to 4D and 4P, the control unit 7 creates the issue time table 71 shown in FIG. 9 (Step S21). Since the creation processing of the issue time table 71 has already been described, the description in this embodiment will be omitted.

Each of the SCSI interfaces 4A to 4D and 4P transmits the input second read request to each of the disks 5A to 5D and 5P. Each disk 5A-5
D and 5P begin reading data blocks and redundant data in response to receiving the second read request. However, reading from each of the disks 5A to 5D and 5P is divided into a case where reading is successful and a case where reading is failed. If the reading is successful, each of the disks 5A to 5D and 5P stores the read data block and the redundant data in the SCSI interfaces 4A to 4D and 4P.
Send to Further, each of the disks 5A to 5D and 5P
Transmits a read response (success), which is a signal indicating that reading has been successful, to the SCSI interfaces 4A to 4D and 4P. SCSI interface 4A-4D
And 4P identify which second read request the read response (success) of each of the disks 5A to 5D and 5P is to the buffer area 3A specified by the control unit 7.
Read data blocks or redundant data are stored in _{i to} 3D _i and 3P _i . On the other hand, when the reading has failed, each of the disks 5A to 5D and 5P has a reading response (failure) indicating that the reading has failed.
To the SCSI interfaces 4A to 4D and 4P.

As is clear from the above, the SCSI interface 4A receives the combination of the data block and the read response (success) transmitted by the disk 5A, and the read response (failure) transmitted by the disk 5A. May be received. Other SCS
Similarly to the case of the SCSI interface 4A, the I interfaces 4B to 4D may also receive a combination of a data block and a read response (success) or may receive a read response (fail). Also, SC
The SI interface 4P receives a combination of redundant data and a read response (success) or a read response (fail). Hereinafter, first, an operation when each of the SCSI interfaces 4A to 4D and 4P receives a read response (success) will be described.
Thereafter, an operation when a read response (failure) is received will be described.

When receiving the read response (success), the SCSI interfaces 4A to 4D and 4P send the data blocks or redundant data transmitted by the disks 5A to 5D and 5P to the buffer units 3A to 3D and 3P.
To be stored. At this time, the data block or the redundant data is stored in the buffer areas 3A _{i to} 3D _i and 3P _i (see FIG. 2) specified by the read response (success). Further, the SCSI interfaces 4A to 4D and 4P transmit the read response (success) transmitted by the disks 5A to 5D and 5P to the control unit 7. on the other hand,
When the SCSI interfaces 4A to 4D and 4P receive the read response (failure) transmitted by the disks 5A to 5D and 5P, the read response (failure) is received.
Is transmitted to the control unit 7.

As is clear from the above, the control unit 7 includes:
One of the two types of read responses (that is, read response (success) or read response (fail))
It is transmitted from each of the SCSI interfaces 4A to 4D and 4P. It should be noted here that, in most cases, the read response from each of the SCSI interfaces 4A to 4D and 4P is accompanied by a time lag, and
It is a point that arrives at. For example, if the disk 5A takes a long time to read a data block, a SCSI
The read response from the interface 4A arrives at the control unit 7 later than the read response from the other.

The control section 7 executes the processing procedure shown in the flowchart of FIG. 28 every time one read response arrives. Upon receiving one read response (step S81), the control unit 7 determines whether the type of the received read response indicates success (step S82). When the control unit 7 determines that the read response (failure) has been received, the process proceeds to step S88 described below. Meanwhile, the control unit 7
Determines in step S82 that a read response (success) has been received, determines whether or not four data blocks constituting the same parity sequence have been collected (step S8).
3). More specifically, in step S83, it is determined whether or not reading of four data blocks from disks 5A to 5D has succeeded. That is, the control unit 7
It is determined whether or not all the read responses (success) have been received from the CSI interfaces 4A to 4D.

When the control section 7 determines that the four data blocks have been completed, it proceeds to step S84 described later. On the other hand, when determining in step S83 that the four data blocks are not aligned, the control unit 7 determines whether or not the data blocks can be restored by parity calculation (step S814). More specifically, in step S814, it is determined whether or not reading of three data blocks and redundant data forming the same parity sequence has been successful. That is, the control unit 7 determines whether a read response (success) from any three of the SCSI interfaces 4A to 4D and a read response (success) from the SCSI interface 4P are received. .

If the control unit 7 determines in step S812 that the data block cannot be restored, that is, if four read responses (success) have not been received at present, the processing procedure of the flowchart in FIG. Is temporarily terminated. Then, the control unit 7 determines that the new read response
It waits for arrival from any of the SI interfaces 4A to 4D and 4P.

By the way, the control unit 7 executes steps S83 to S8.
When the process proceeds to 4, it means that four data blocks constituting the same parity sequence are prepared as described above. Here, in the RAID device described in the third embodiment, reading of the remaining data blocks is performed only during the time margin T _MARGIN from the time when three data blocks and redundant data are prepared (that is, time T _4th ). Wait for it to complete. Similarly, the RAID device according to the present embodiment is 3
Even if the number of data blocks and redundant data are complete, the process waits until reading of the remaining data blocks is completed. Therefore, at the time when step S84 is executed, the same parity sequence is configured only in the buffer units 3A to 3D.
There are a case where four data blocks are stored in the buffer units 3A to 3D and 3P and a case where four data blocks constituting the same parity sequence and redundant data are stored in the buffer units 3A to 3D and 3P. Therefore, the control unit 7 determines whether the reading of the redundant data has been completed (Step S).
84). That is, the control unit 7 determines whether a read response (success) has been received from the SCSI interface 4P.

If the control section 7 determines in step S84 that the reading of the redundant data is not completed, it generates a read stop request signal and transmits it to the reassignment processing section 8 (step S85). Hereinafter, the read stop request will be described. At the time of step S84, since four data blocks are provided, data can be assembled without executing a parity operation. Therefore, the control unit 7 understands that the unread redundant data is unnecessary. The read stop request transmitted in step S85 is a signal for requesting the reassignment processing unit 8 to stop reading of unnecessary redundant data. The read stop request includes information indicating a storage location (LBA) of unnecessary redundant data. The reassignment processing unit 8 executes the processing shown in the flowchart of FIG. 34 in response to the read stop request from the control unit 7. For convenience, the processing procedure shown in FIG. 34 will be described later. After ending the processing in step S85, the control unit 7 proceeds to step S86. On the other hand, when the control unit 7 determines in step S84 that the redundant data has been read, the process proceeds to step S72. In order to proceed to step S72, the condition is that reading of four data blocks and redundant data has been completed. That is, the condition is that the reading of the last one data block is completed while the first timer 72 set in step S815 described later is running. Therefore, there is no need to activate the first timer 72. Therefore, the control unit 7
Stops the activated first timer 72 (step S
87) Then, the process proceeds to step S86.

In step S86, the control section 7 generates a signal indicating that reading has been completed, and transmits the signal to the selector 2. The completion of reading is a signal for notifying the selector 2 that four data blocks of the same parity sequence have been arranged in the buffer units 3A to 3D and data can be assembled. Therefore, when reading is completed,
Information for specifying four buffer areas 3A _{i to} 3D _i storing four data blocks of the same parity sequence is included. The selector 2 sequentially selects the four buffer areas 3A _{i to} 3D _i according to the received read completion, and reads four data blocks. Further, the selector 2 assembles the four read data blocks to form 2048-byte data. The configured data is transmitted to the host device through the host interface 1.

By the way, the control unit 7 determines in steps S814 to S
When the process proceeds to 815, as described above, three data blocks and redundant data constituting the same parity sequence are prepared. As described above, the RA according to the present embodiment
The ID device waits for reading of the remaining data blocks to be completed. To this end, the control unit 7 calculates the timeout value V _TO1 and calculates the calculated timeout value V _TO1.
TO1 is set in the first timer 72 (step S81)
5). As a result, the first timer 72 starts up and starts counting down. The processing in step S815 is the same as the processing in step S43 in FIG. 12B, and thus the description in this embodiment is omitted. When the setting of the first timer 72 is completed in step S815, the control unit 7 sends a new read response to the SCSI interface 4
Wait for arrival from any of A-4D and 4P.

Now, the control section 7 executes steps S82 to S8.
When the process proceeds to step 8, a read response (failure) has arrived at the control unit 7. At this time, the control unit 7 determines whether the first timer 72 is running (step S8).
8). When the control unit 7 determines that the first timer 72 is not running, the process proceeds to step S811 described below. On the other hand, when it is determined that the first timer 72 is running, step S89 is executed. Here, if the first timer 72 is running, the read response (failure) that has arrived this time indicates that the reading of the remaining data block that has not been read at the time of execution of step S814 has failed. As a result, the control unit 7 recognizes that the countdown of the first timer 72 is unnecessary, and stops the countdown (step S89). Further, since it is known that the reading of the remaining data blocks has failed, the control unit 7 issues a restoration instruction and requests the parity operation unit 6 to execute the parity operation (step S810). The parity operation unit 6 restores the remaining one data block that could not be read out by the parity operation and temporarily stores the data block in the buffer unit 3P. Thereafter, the parity calculation unit 6 issues a restoration completion signal, which is a signal indicating that the restoration of the data block is completed, and transmits the signal to the control unit 7. The control unit 7 issues a read completion in response to the reception of the restoration completion, and transmits the read completion to the selector 2 (step S8).
6). As a result, data is transmitted to the host device.

By the way, the control section 7 executes steps S88 to S8.
When the process proceeds to 11, the control unit 7 has received at most three read responses. By the way, R of this embodiment
The AID device stores the parity sequences in the five disks 5A to 5D and 5P in a distributed manner. Therefore, when two of the five disks fail to read the data block or the redundant data, restoration of the data block by the parity calculation cannot be expected. Therefore, the control unit 7
Determines whether restoration of a data block can be expected by parity calculation in step S811. More specifically, in step S811, it is determined whether two of the currently arriving read responses indicate failure.

If the control unit 7 determines in step S811, that the restoration of the data block can be expected by the parity calculation (that is, if it is determined that the read response that has arrived this time indicates a failure for the first time), the control unit 7 shown in FIG. The processing procedure ends once. Then, the control unit 7 sends the new read response to the SCSI interfaces 4A to 4D and 4P.
Wait to arrive from one of the.

On the other hand, the control unit 7 determines in step S811 that
If it is determined that restoration of the data block cannot be expected by the parity operation (that is, it is determined that the data that has arrived this time is the second read response (failure)), the process proceeds to step S812. Next, the control section 7 issues a read stop request and sends it to the reassignment processing section 8 (step S812). Hereinafter, the read stop request transmitted at this time will be described. At the time of step S812, the reading has not been completed for some of the disks 5A to 5D and 5P. For example, if both the first read response and the second read response indicate failure, the reading has not been completed on three of the disks 5A to 5D and 5P. In addition, since restoration of the data block cannot be expected, the control unit 7 determines that the data block or the redundant data that has not been read at the time of step S812 is unnecessary. Therefore, step S812
Is a signal for requesting the reassignment processing unit 8 to stop reading of such unnecessary data blocks or redundant data.
The read stop request includes information indicating an LBA in which unnecessary data blocks or redundant data are stored. The reassignment processing unit 8 executes the processing shown in the flowchart of FIG. 34 in response to the read stop request from the control unit 7. For convenience, the processing procedure shown in FIG. 34 will be described later. After finishing the processing in step S812, the control unit 7 proceeds to step S813.

When the data block cannot be restored, the control unit 7 cannot transmit data to the host device, and thus generates a read failure signal indicating that (step S813). The generated read failure is
It is transmitted to the host device through the host interface 1.

The control unit 7 is provided with the first timer 7
If the time-out occurs, the processing procedure shown in FIG. 12B is executed. It should be noted that since this processing procedure has already been described, it is not described in the present embodiment. When issuing the second read request, the control unit 7 refers to the issue time table 71 and subtracts the issue time t _ISSUE from the current time t _PRE . Control unit 7
Is calculated (t _PRE −t _ISSUE ) is the time limit T _LIMIT
It is determined whether or not exceeds. And the time limit T _LIMIT
If the reading has not been completed on any two of the disks 5A to 5D and 5P at the time when it is determined that the number has exceeded the limit, the control unit 7 specifies the disk on which the reading has not been completed at that time. Then, the control unit 7 issues a read stop command to some of the identified disks 5A to 5D and 5P. Since such processing has already been described with reference to FIG. 8B, it should be noted that the description in this embodiment is omitted.

Next, an operation related to the reassignment processing section 8 will be described. As described above, new configurations are added to the SCSI interfaces 4A to 4D and 4P in connection with the reassignment processing unit 8. This new configuration is illustrated in FIG.
As shown in FIG. 9, the notification units 41A to 41D and 41P. The notifying units 41A to 41D and 41P
When the interfaces 4A to 4D and 4P transmit the second read request to the disks 5A to 5D and 5P, they create transmission notices that are signals indicating that. The created transmission notification is transmitted to the reassignment processing unit 8. Each transmission notification includes an ID for uniquely identifying the second read request transmitted this time, and the LB specified by the read request.
A is included. Further, the notification units 41A to 41D and 41P are provided with the SCSI interfaces 4A to 4D and 4C.
When P receives the read response from the disks 5A to 5D and 5P, it creates a reception notification, which is a signal indicating that. The created reception notification is transmitted to the reassignment processing unit 8. Each reception notification includes an ID for uniquely identifying a second read request corresponding to the read response received this time,
LBA specified by the second read request. Note that the reassignment processing unit 8 can operate correctly even if the reception notification does not include the LBA.

As shown in FIG. 29, the reassignment processing section 8 includes a third timer 81 for measuring the current time, a first list 82, and a second list 83, and receives a transmission notification. Each time the process is performed, the procedure for the reassignment process shown in the flowchart of FIG. 30 is executed. Now, in order to make the following description concrete, the reassignment processing unit 8
It is assumed that a transmission notification has been received this time from the interface 4A. In the transmission notification received this time, the ID "b"
, And “a” is included as the LBA. Is assumed to have been received from the SCSI interface 4A.

First, the reassignment processing section 8 detects the reception time of the transmission notification based on the current time measured by the third timer 81. This reception time is used by the reassignment processing unit 8 as the time at which the SCSI interface 4A transmitted the second read request to the disk 5A. now,
The transmission time of the second read request is _tt1 . further,
The reassignment processing unit 8 extracts “b” as the ID and “a” as the LBA from the received transmission notification (step S91).

Here, the first list 82 described above and the second list 82
Will be described. The first list 82 is
As shown in FIG. 31A-1, there are fields for registering ID, LBA, and processing start time. One first list 82 is created each time a second read request is sent. The reassignment processing unit 8 classifies and manages the created first list 82 for each transmission destination of the second read request. That is, the first list 82 includes the disks 5A to 5A.
5D and 5P (ie, SCSI interface 4A)
４4D and 4P). further,
The first list 82 for each disk is arranged in the transmission order of the second read request. That is, now, FIG.
In 1), a plurality of first lists 82 are shown. It is assumed that the plurality of first lists 82 are created in response to a second read request to the disk 5A, for example. In FIG. 31 (a-1), as indicated by the arrow, the first list 82 indicated on the near side of the drawing includes a new second read request (that is, a second read request whose transmission time is later). Is registered, and information of an old second read request (that is, a second read request with a new transmission time) is registered in the first list 82 shown on the back of the paper.

Also, the second list 83 is shown in FIG.
As shown in 1), information indicating the LBA storing the data block or the redundant data and the value N of the number-of-times counter FC
And a field for registering.

After step S91, the reassignment processing section 8 currently stores a plurality of second read requests at the transmission destination of the current second read request (hereinafter, simply referred to as “current transmission destination”). It is determined whether or not there is (Step S9)
2). The determination in step S92 will be specifically described. Under the above assumption, the current transmission destination is the disk 5A. Further, as described above, the first list 82 is created each time the second read request is transmitted to the disks 5A to 5D and 5P, and is classified and managed for each of the disks 5A to 5D and 5P. Further, the first list 82 indicates that the corresponding second read request indicates that the disk 5A to 5D or 5P
When the processing is completed or the processing is forcibly stopped, it is deleted. Therefore, the reassignment processing unit 8 manages the first list 82 managed for the current transmission destination (disk 5A).
Is counted, the number of second read requests accumulated in the disk 5A can be known. Here, it should be noted that if at least one first list 82 is managed in step S92, the reassignment processing unit 8
Is that it is determined that a plurality of second read requests are stored at the current transmission destination (disk 5A). The reason is described below. Regarding the second read request transmitted this time, the first list 82 has not been created yet at the time of step S91. At the time of step S91, the reassignment processing unit 8 manages only the first list 81 of the second read request transmitted to the disk 5A before that. Therefore, at the time of step S92, the previously transmitted second read request and the currently transmitted second read request are held by the current transmission destination (disk 5A). Therefore, the reassignment processing unit 8 performs step S
In 92, if at least one first list 82 is managed, a plurality of second lists are assigned to the current transmission destination (disk 5A).
It is determined that the read requests of are accumulated.

When the reassignment processing section 8 determines in step S92 that a plurality of second read requests have not been accumulated, it creates a new first list 82. The reassignment processing unit 8 registers the LBA “a” and the ID “b” obtained in step S81 in the newly created first list 82. Further, the reassignment processing section 8 determines in step S91
The transmission time “t _t1 ” detected in the above is set as the processing start time,
The BA “a” and the ID “b” are registered in the first list 82 registered. In addition, since the reassignment processing unit 8 has received the transmission notification from the SCSI interface 4A in step S91, the reassignment processing unit 8 stores the newly created first list 82
The data is classified and stored as the disk 5A (step S93). As a result, the first list 8 created this time
2, information as shown in FIG. 31 (a-2) is registered.

On the other hand, the reassignment processing section 8 determines in step S
If it is determined in step 92 that a plurality of second read requests are accumulated, the process proceeds to step S94. When another second read request is accumulated in the current transmission destination (disk 5A), the current second read request is transmitted by the current transmission destination unless the processing of the previous second read request is completed. Not processed. That is, the current second read request is
You have to wait inside this destination to be processed. Therefore, it is inappropriate to set the transmission time “t _t1 ” detected in step S91 as the processing start time in the first list 82 as in step S93. Therefore, in step S94, the reassignment processing unit 8 registers and manages only the LBA “a” and the ID “b” obtained in step S91 in the first list 82 (step S94). Here, it should be noted that the processing start time not registered in step S94 is registered later (for details, see step S104 described later).

The reassignment processing section 8 executes the processing procedure shown in the flowchart of FIG. 32 separately from the processing procedure of FIG. FIG. 32 shows a process for the reassignment processing unit 8 to detect a defect recording area. First,
The reassignment processing unit 8 stores the first list 82 currently held.
See, for measuring the delay time T _D of the second read request sent to each disk 5A~5D and 5P.
In the present embodiment, the delay time T _D, the disk 5A~5D and 5P has started processing the second read request, means the time to the current time.

Hereinafter, the measurement processing of the delay time T _D will be described more specifically. As is clear from the above, one first list 82 is created each time the SCSI interface 4A sends a second read request to the disk 5A. Similarly, the first list 82 is created for the other disks 5B to 5D and 5P. Some of the first lists 82 have registered therein the processing start times of the second read requests. First, the reassignment processing unit 8 selects one of the first lists 82 to be processed from the first lists 82 in which the processing start times are registered. Next, the reassignment processing unit 8 calculates, from the first list 82 to be processed,
Extract the processing start time. Next, the reassignment processing unit 8
Obtains a current time T _P which is measured by the timer 81.
Next, reassignment processing unit 8, the current time T _P, subtracts the processing start times retrieved. This subtraction result is defined as the delay time T _D of the second read request corresponding to the first list 82 to be processed.

Further, the reassignment processing section 8 sets a time limit T
_L is stored in advance internally. The limit time T _L is the actual RA
Each disk 5A to 5D is predetermined by the ID device.
And 5P are indexes for determining whether or not a defect recording area is included. Further, the limit time T _L is more preferably a time that guarantees that data and data can be transmitted without interruption of video and audio on the host device side. By the way, the reassignment processing unit 8 determines that the calculated delay time T _D is
It is determined whether or not a time limit T _L stored in advance is exceeded (step S101). The delay time T _D is equal to the limit time T _L
Is exceeded, the reassignment processing unit 8 determines that the processing of the second read request specified by the first list 82 to be processed is delayed, and determines the LBA specified by the second read request. It is determined that there is a possibility that a defect has occurred.

A specific example of the processing in step S101 will be described. Now, the reassignment processing unit 8 is configured as shown in FIG.
Assume that the first list 82 shown in 2) is selected. The first list 82 of FIG.
"A" and a processing start time " _tt1 " are registered. Therefore, the delay time T _D of the second read request is identified by the ID "b" is determined by (T _P -t _t1). Further, the reassignment processing section 8 calculates (T _D > T
_L ) It is determined whether or not the condition is satisfied. If (T _D> T _L) is not satisfied, and reassignment processing unit 8 selects a new first list 82 as a processing target, executes step S101. If the new first list 82 cannot be selected, the reassignment processing unit 8 ends the processing procedure of FIG.

[0174] On the other hand, "T _D> T _L" in step S101
When the condition is satisfied, the reassignment processing unit 8
The SCSI interface 4 is instructed to stop the process of the second read request specified by the list 82 (step S102). In step S102, the reassignment processing unit 8 transmits one of the SCSI messages A to stop the second read request.
A BORT_TAG is generated and transmitted to the SCSI interface 4. The SCSI interface 4 transmits the received ABORT_TAG to the disk 5 connected thereto. The disk 5 receives the received ABORT_TAG
, The processing of the second read request specified by the ID “b” is stopped. Here, as assumed above, the second read request specified by ID “b” is S
Since the data is transmitted to the disk 5A through the CSI interface 4A, the reassignment processing unit 8
The TAG is transmitted to the disk 5A via the SCSI interface 4A. As a result, the disk 5A has the ID
The processing of the second read request specified by “b” is stopped. The SCSI interface 4 is an ABO
After transmitting the RT_TAG, the control section 7 transmits a read response (failure), which is a signal indicating that the processing of the second read request specified by the ID “b” has failed.

The reassignment processing section 8 determines in step S102
Next, it is determined whether or not another first list 82 is managed for the disk specified by the first list 82 to be processed (hereinafter, referred to as the current processing target) (step S102). ). More specifically, the reassignment processing unit 8 determines whether a plurality of second read requests are currently stored in the current processing target (step S10).
3).

The reassignment processing section 8 determines in step S103
If it is determined that the other first list 82 is managed, the process proceeds to step S104. Now, under the above assumption, a plurality of first lists 82 are managed for the current processing target (disk 5A). In addition, step S described later
In 108 or S1013, the first list 8 to be processed
2 is deleted. Therefore, as shown in FIG. 31 (a-3), the reassignment processing unit 8 currently stores the first list 82 of the processing target and the first list 82 created next to the processing target.
(Hereinafter referred to as a “next first list 82”). In FIG. 31A-3, one of the following first lists 82 is indicated by a dotted line. It should be noted that the following first list 8
2 is that the processing start time is not registered. This is because the first list 82 created later is created in step S94 in FIG. To register a new processing start time, reassignment processor 8, in step S104, first, to obtain the current time T _P from the third timer 81. The reassignment processing unit 8 registers the current time T _P in the next first list 82 (step S
104). Thereafter, the reassignment processing unit 8 proceeds to step S1
Go to 05. On the other hand, the reassignment processing unit 8
If another second read request is not accumulated in 103, the process proceeds to step S105 without executing step S104.

Next, the reassignment processing section 8 extracts the LBA registered therein from the first list 82 to be processed. Hereinafter, the extracted LBA is referred to as an LBA to be inspected. Under the above assumption, the LBA to be inspected is “a”. The LBA to be inspected may be a defect recording area as described above. The reassignment processing unit 8 searches the currently managed second list 83 (see FIG. 31 (b-1)) based on the LBA to be inspected, and retrieves the second list 83 in which the LBA to be inspected is registered. It is determined whether or not there is a list 83 (step S105).

As described above, the second list 83 contains L
There is a field for registering the BA and the value N of the number counter FC. The value N of the number-of-times counter FC is equal to (T
_D > T _L ). Therefore, if the second list 83 can be found in step S105, the LBA to be inspected is at least at the time of the previous inspection.
It is determined that there is a possibility of a defect recording area.
In other words, the LBA to be inspected twice
, A second read request has been transmitted to read a data block or redundant data from. Moreover, in the two steps S101 executed in response to the transmission of each second read request, the reassignment processing unit 8
Is determined to continuously satisfy (T _D > T _L ). On the other hand, if the second list 83 cannot be found in step S105, the LBA is
It is determined that there is a possibility of a defect recording area.

Now, in step S105, the second list 8
If No. 3 has been found, step S109 is executed. On the other hand, the reassignment processing unit 8 determines in step S1
If the second list 83 cannot be found in step 05, step S106 is executed. Now, step S1
Assuming that the second list 83 could not be found at 05, the description continues. Under this assumption, the reassignment processing section 8 proceeds to step S106 and creates a new second list 83. The reassignment processing unit 8 is configured as shown in FIG.
As shown in (b-2), the newly created second list 8
In the corresponding field in 3, the LBA to be inspected is
(“A” under the above assumption). Further, the reassignment processing unit 8 registers “1” as an initial value of the number-of-times counter FC in the corresponding field (step S10).
6).

The reassignment processing unit 8 determines in step S106
Next, the second list 8 in which the LBA to be inspected is registered
3 (hereinafter, referred to as a second list 83 to be processed).
When the value N of the recorded number counter FC is equal to the specified number N_L Reached
It is determined whether or not it has been performed (step S107). Regulation
Number N_L Is reserved in the RAID device according to the present embodiment.
LBA to be inspected is in the defect recording area
This is a threshold for determining whether or not there is. Also, the prescribed times
Number N_L As one or more according to the specifications of the RAID device.
An appropriate natural number is selected. In the present embodiment, for convenience of explanation,
Above, specified number N _L Is chosen to be "2".
Now, the second list 83 to be processed is stored in step S106.
Immediately after it is newly created. Therefore, processing
In the second list 83 of the objects, the value N of the number counter is set.
"1" is registered (see FIG. 31 (b-2)).
Therefore, the reassignment processing unit 8 proceeds to step S107.
Where the value N is equal to the specified number N_L Judge that has not reached
You. As a result, the reassignment processing unit 8 determines in step S10
Proceed to 8.

Next, the reassignment processing section 8 determines that the first list 82 to be processed is no longer needed and deletes the first list 82 (step S108). This prevents the first list 82 from being redundantly selected as a processing target. According to the above assumption, at this time,
The first list 82 in which “b” is registered as the ID, “a” is registered as the LBA, and “t _t1 ” is registered as the processing start time is deleted. It should be noted here that the second list 83 to be processed is not deleted in step S108. When step S108 ends, the reassignment processing unit 8 returns to step S101 and sets the first
Is selected, and the process is subsequently executed. If the value N has reached the prescribed number of times _NL in step S106, step S109 described later is executed.

Further, another first read request arrives at the control unit 7 from the host device. In response to the arrival of the first read request, the control unit 7 transmits the second set of read requests to the SCSI interfaces 4A to 4D and 4P.
Send to The SCSI interfaces 4A to 4D and 4P transmit the received second read request to the disks 5A to 5D.
Send to 5D and 5P. Now, it is assumed that the second read request transmitted to the disk 5A indicates reading of a data block from LBA “a”. In this case, the notification unit 41A of the SCSI interface 4A creates a transmission notification for the second read request transmitted to the disk 5A, and transmits the transmission notification to the reassignment processing unit 8. This transmission notification includes “c” and “a” as ID and LBA.

The reassignment processing section 8 executes the processing shown in FIG. Therefore, first, the reassignment processing unit 8
When the transmission notification is received, the current time _TP is obtained from the third timer 81. This current time T _P is, as described above, SC
It is used as the time when the SI interface 4A transmitted the second read request to the disk 5A. Now, _assume that the transmission time of the second read request is “ _tt2 ”. Further, the reassignment processing unit 8 extracts “c” as the ID and “a” as the LBA from the received transmission notification (step S91). Next, the reassignment processing unit 8 determines in step S
92 → S93 processing procedure, or step S92 → S9
By executing the processing procedure of No. 4, the first list 82 is newly created for the second read request transmitted this time, and the processing of FIG. 30 ends. Now, this destination (disk 5
Assuming that A) holds only one second read request, the first list 82 contains "a" as the LBA, I
“C” is registered as D and “ _tt2 ” is registered as the processing start time (see FIG. 31A-4).

Further, the reassignment processing section 8 executes the processing shown in FIG. First, the reassignment processing unit 8 selects the first list 82 to be processed from the currently stored first list 82. Thereafter, reassignment processing unit 8, the delay time T _D which is calculated with reference to the first list 82 to be processed is determined whether it exceeds the limit time T _L (step S101). Now, when the first list 82 to be processed is assumed to be that shown in FIG. 31 (a-4), the delay time T _D is determined by the "T _P -t _t2". "T _D
(= T _P -t _t2)> When satisfying T _L ", reassignment processing unit 8, after stopping the process of the second read request specified by the first list 82 to be processed (step S
102), it is determined whether or not another first list 82 is managed for the current processing target (disk 5A) (step S103). Under the above assumption, since the current transmission destination (disk 5A) holds one second read request, the reassignment processing unit 8 proceeds directly from step S103 to S105. Next, the reassignment processing unit 8 extracts the LBA registered in the first list 82 to be processed as the LBA to be inspected (currently “a”). Further, the reassignment processing unit 8 searches the currently managed second list 83 based on the LBA to be inspected, and registers the LBA to be inspected (see FIG. 31 (b-2)). It is determined whether or not there is (step S105). As described above, now, the reassignment processing unit 8
Since the second list 83 shown in -2) is managed, the process proceeds to step S108. Here, the second list 83 in which the inspection target LBA is registered is handled by the reassignment processing unit 8 as a processing target, as described above.

Next, the reassignment processing section 8 sets the value N of the number counter FC registered in the second list 83 to be processed.
Is incremented by "1" (step S10).
9). According to the above assumption, the value N in FIG. 31B-2 is incremented by “1”. As a result, FIG.
As shown in (b-3), the value N is “2”. After step S109, the reassignment processing unit 8 determines whether or not the value N registered in the second list 83 to be processed has reached a specified number of times N _L (same as above, “2”). A determination is made (step S107). According to the above assumption, for the second list 83 to be processed, the value N
Is currently “2”, the reassignment processing unit 8 checks the LBA to be inspected (currently, the LBA of the disk 5A).
It is determined that the recording area specified by “a”) is a defect recording area, and the process proceeds to step S1010.

Next, the reassignment processing section 8 executes the first table 91 (FIG. 25) managed by the alternative recording area information management section 9.
), And selects one LBA from the LBAs specifying the currently available alternative recording area. Thereby, the reassignment processing section 8 selects an alternative recording area and assigns it to the defect recording area (step S101).
0). The size of the selected alternative recording area is the size of the data block or the redundant data (512 in this embodiment).
Bytes).

Next, the reassignment processing section 8 determines the LBA of the defect recording area (currently, the LBA of the disk 5A).
“A”) and the LBA of the selected alternative recording area are notified to the address conversion processing unit 11 (step S101).
1). The address conversion processing unit 11 includes a reassignment processing unit 8
The LBA of the defect recording area and the LBA of the replacement recording area received from the server are registered in the second table 101 (see FIG. 27) managed by the address information management unit 10. Attention should be paid here to the fact that in FIG.
The BA specifies the storage location where the data block or redundant data was originally recorded, and thus is described as the original LBA. Also, the LBA of the alternative recording area
Specifies the current recording area of the data block or the redundant data recorded in the defect recording area, and is described as the current LBA. As described above, since the address information is updated, the control unit 7 performs the second operation for reading the data block or the redundant data that has been reassigned this time.
When the next read request is created, the current LBA will be used.

Subsequent to step S1011, the reassignment processing section 8 sets the alternative recording area information management section 9 so that the alternative recording area selected in step S1010 is not redundantly selected.
Is updated in the first table 91 managed by (step S1012). By this updating process, the reassignment processing unit 8 prevents the redundant selection of the alternate recording area selected this time. Thus, the reassignment process ends. When the reassignment process is completed, the first list 82 to be processed
And the second list 83 of the processing target becomes unnecessary, so the reassignment processing unit 8 sets the first list 82 of the processing target to be unnecessary.
And the second list 83 are deleted (step S101).
3). Further, the reassignment processing section 8 generates a reassignment end notification, which is a signal indicating that the reassignment processing has been completed, and transmits it to the control section 7 (step S1014). The reassignment end notification includes information indicating the LBAs of the defect recording area and the replacement recording area.

In response to the reception of the reassignment completion notification from the reassignment processing unit 8, the control unit 7 restores data blocks or redundant data lost by reassignment in accordance with the RAID level policy adopted in the present RAID device. Then, the restored data block or the redundant data is written in the alternative recording area of the disk (reassigned disk) that is the transmission destination this time. Since this processing is a well-known technique, a detailed description thereof will be omitted. By writing the data block or the redundant data, the parity sequences recorded on the disks 5A to 5D and 5P are accurately matched before and after the execution of the reassignment process.

As described above, the RAID according to the present embodiment
In the apparatus, when a defect recording area is detected from the disks 5A to 5D or 5P, a reassignment process is performed. As a result, an alternative recording area is assigned to the defect recording area. In this alternative recording area, lost data blocks or redundant data are stored. That is, the data block or the redundant data in the defect recording area is not left as it is. As a result, after detecting a defective recording area, the RAID apparatus attempts to read a data block or redundant data by accessing an alternative recording area without accessing the defective recording area. As a result, it is possible to prevent a problem as described at the beginning of the present embodiment, that is, a delay in reading due to continuous access to the defective recording area.

In the sixth embodiment described above, in order to clarify the timing at which the substitute recording area is allocated, a part of the operation when the read response is received by each of the SCSI interfaces 4A to 4D and 4P is omitted. Explained. That is, the read response is the SCSI interface 4
Returning to A-4D and 4P, the first list 82
Changes depending on the time at which the read response is returned or the content of the read response (success or failure of the read). Next, a process in which the reassignment processing unit 8 updates the first list 82 when a read response is returned will be described.

The notification units 41A to 41D and 41P generate a signal of reception notification each time each of the SCSI interfaces 4A to 4D and 4P receives the read response from the disks 5A to 5D and 5P, 8 This reception notification includes the ID of the second read request that is the basis of the read response received this time, and the LBA specified by the second read request. Here, in order to make the following description concrete,
Assume that interface 4A has received a read response containing ID "b" and LBA "a". Under this assumption, the SCSI interface 4A has the ID
A reception notification including “b” and LBA “a” is transmitted to the reassignment processing unit 8. Here, it is necessary to pay attention to the first
The update processing of the list 82 has no relation to the success or failure of the read indicated by the read response. Therefore, the description of success or failure indicated by the read response is omitted.

The reassignment processing section 8 executes the processing procedure shown in the flowchart of FIG. 33 in response to the reception of the reception notification. First, the reassignment processing unit 8 outputs the ID “b”,
“A” as the LBA is extracted from the received reception notification. Further, the reassignment processing unit 8 searches the first list 82 currently managed for the one in which the ID “b” is registered (hereinafter, referred to as the first list 82 to be deleted) (step S <b> 1). S111). Here, despite the fact that the second read request has been transmitted, ID
The fact that the first list 82 in which “b” is registered is not managed means that the first list 82 in FIG.
013 means deleted. If the reassignment processing unit 8 cannot find the first list 82 to be deleted in step S111 as described above,
Steps S112 to S115 do not need to be performed, so the process proceeds to step S116.

On the other hand, in step S111, the first
"T _D is that was retreive the list 82 of the reception immediately before the reception notification (ie, immediately before coming back is this read response) until, in step S101 (see FIG. 32)
> T _L ”was not satisfied. Therefore, reassignment processing unit 8 at the moment, based on information registered in the first list 82 to be deleted, it is determined whether or not satisfied "T _D> T _L" (step S112). Reassignment processing unit 8, when the delay time T _D is greater than the prescribed time T _L, it is necessary to determine whether it is necessary to assign a replacement recording area to the defect recording area, surrounded by ○ mark "B As shown in FIG. 32, step S103 in FIG.
Execute the following processing.

[0195] On the other hand, the fact that the delay time T _D in step S112 does not exceed the prescribed time T _L, the disk 5
This means that reading in A did not take much time. That is, the LBA specified by “a” is not a defect recording area. Therefore, the reassignment processing unit 8
Is the same LBA as in the first list 82 to be deleted.
It is determined whether or not manages the registered second list 83 (step S113). Rear sign processing unit 8
If the second list 83 can be found in step S113, the second list 83 is deleted (step S114), and the process proceeds to step S115. On the other hand, if the second list 83 cannot be found in step S113, the reassignment processing section 8 proceeds directly from step S113 to S115, and proceeds to step S115 to delete the first list 82 to be deleted.
Is deleted (step S115).

Next, the reassignment processing section 8 checks whether another second read request is currently stored in the disk 5 (hereinafter, simply referred to as “current source”) of the source of the current read response. It is determined based on the number of first lists 82 managed for the transmission source (step S11).
6). If it is determined in step S116 that other read requests are accumulated, the first list 82 created in response to the other second read request (the next first list 82) has the processing start time. Is not registered. Therefore, the reassignment processing unit 8 obtains the current time from the third timer 81 and defines the time at which another second read request starts to be processed by the current transmission source. Rear sign processing unit 8
Sets the obtained current time as the processing start time of the second read request to be processed next, and
(Step S117), and the processing in FIG. 33 ends. On the other hand, when other second read requests are not accumulated in step S116, the reassignment processing unit 8 ends the processing in FIG. 33 without executing step S117.

By the way, in step S85 of FIG. 28, the control unit 7 transmits a read stop request for stopping reading of redundant data to the reassignment processing unit 8.
In addition, the control unit 7 has transmitted to the reassignment processing unit 8 a read stop request for stopping reading of unnecessary data blocks and / or redundant data in step S812 of FIG. As described above, each read stop request includes the LBA that specifies the data block to stop reading and / or the storage location of the redundant data. Next, a processing procedure when the reassignment processing unit 8 receives a read stop request will be described with reference to FIG.

First, the reassignment processing section 8 extracts the LBA from the received read stop request, and determines whether the reading of the data block or the redundant data from the LBA has been started (step S121). The process of step S121 will be specifically described. First, the reassignment processing unit 8 searches the currently managed first list 82 for a registered LBA whose reading should be stopped. The reassignment processing unit 8 searches for the first list 8
It is determined whether the processing start time is registered in No. 2 or not.
As is apparent from the above, the processing start time is not always registered when each first list 82 is created. for that reason,
At the start of the processing procedure in FIG. 34, the first list 82 in which the processing start time is registered is stored in the reassignment processing unit 8.
And those that are not registered are mixed. Here, the first
If the processing start time is already registered in the list 82, it can be considered that the reading of the data block or the redundant data from the corresponding LBA has started. Therefore, the reassignment processing unit 8 starts processing the second read request corresponding to the first list 82 on the disk 5 depending on whether the processing start time is registered in the found first list 82. It is determined whether or not.

The reassignment processing unit 8 determines in step S121
If it is determined that the reading from the LBA extracted from the reading stop request has started, the process of FIG. 34 ends. On the other hand, the reassignment processing unit 8
If it is determined that reading from
ABORT_TAG, which is one of the SI messages, is transmitted to the disk 5 having the extracted LBA through the SCSI interface 4 to stop the execution of the second read request corresponding to the first list 82 found. (Step S122). Further, the SCSI interface 4 transmits to the control unit 7 a read response (failure) indicating that the reading of the corresponding second read request has failed. After step S112, the reassignment processing unit 8 deletes the first list 82 found in step S121 (step S123).

[0200] As described above, the reassignment processing section 8 responds to the read stop request from the control section 7 and returns to Step S.
Only when the condition of 111 is satisfied, the processing of the second read request is stopped. Thereby, the reassignment processing section 8 can correctly detect the defect recording area from the disks 5A to 5D and 5P. If the reassignment processing unit 8 stops the process unconditionally in response to a read stop request will not satisfy the delay time T _D> limit time T _L for most of the second read request. As a result, the reassignment processing unit 8 may not be able to accurately detect the defect recording area.

(Seventh Embodiment) In the RAID device according to the fifth embodiment described above, the failed block table 75
Describes a storage location of a data block that takes a long time to read. The control unit 7 refers to the failed block table 75 to determine whether to transmit five second read requests or four second read requests. As a result, a RAID device capable of reading a large amount of data per unit time has been realized. However, the more data blocks that take longer to read are listed in the failed block table 75, the more frequently the RAID device sends five second read requests. As a result, there is a problem that more data cannot be read per unit time. Therefore, in the seventh embodiment, such a problem is solved, and a RAID device capable of reading more data per unit time is realized.

FIG. 35 is a block diagram showing a configuration of a RAID device according to the seventh embodiment of the present invention. The RAID device of FIG. 35 is different from that of FIG.
Includes a failed block table 75. Since other configurations are the same, in FIG. 35, the same components as those shown in FIG. 24 are denoted by the same reference numerals, and description thereof will be omitted. Here, it should be noted that the failed block table 75 in FIG. 35 is the same as that shown in FIG. It should be further noted that in the present embodiment, the redundant data is stored in the disk 5A as shown in FIG.
5D and 5P.

This RAID device also starts read processing in response to the arrival of the first read request from the host device, as in the sixth embodiment. Since this reading process is a characteristic process of the seventh embodiment, FIG.
This will be described with reference to FIG. 36. First, FIG.
7 is a flowchart showing a processing procedure from the arrival of a read request to the transmission of a second set of read requests. The flowchart of FIG. 36 partially includes the same steps as those of FIG. Therefore, in FIG. 36, the same steps as those shown in FIG. 26 are denoted by the same step S numbers, and the description thereof will be simplified.

When the first read request is input (step S1), the control unit 7 converts the LBA indicating the storage location of the parity sequence to be read this time into the address conversion processing unit 11.
(Step S71). That is, the control unit 7 obtains an LBA indicating a storage location of a data block of the same parity sequence and redundant data.

Next, the control unit 7 determines whether the four disks (any four of the disks 5A to 5D and 5P) have failed in the past to read the four data blocks to be read this time. Is determined (step S13).
1). For the determination in step S131, the failed block table 75 is referred to. Failed block table 75
As shown in FIG. 22, storage locations of data blocks that have failed to be read in the past (note that the storage locations are indicated by LBAs in this embodiment need to be noted). Therefore, the control unit 7 sets the LB indicating the storage location of each data block obtained from the address conversion processing unit 11.
By comparing A with the LBAs listed in the failed block table 75, the determination in step S131 can be made easily.

First, a case will be described in which the control unit 7 determines that the reading of four data blocks has not failed in step S131. In this case, the control unit 7
Determines that reading of the four data blocks is unlikely to fail this time, and issues a second read request set to read the parity sequence (step S132). However, in step S132, the second read request is issued only to the four disks 5 storing data blocks, and is not issued to the remaining one disk storing redundant data.

Next, the case where the control unit 7 determines that reading of some data blocks has failed in step S131 will be described. In this case, the control unit 7 determines that reading of the four data blocks is likely to fail this time, and issues a second read request set to read the parity sequence (step S133). However, in step S133, the second read request is issued to the four disks storing data blocks and the remaining one disk storing redundant data.

[0208] The second read request issued in step S132 is read by four disks storing data blocks constituting the same parity sequence. The second read request issued in step S133 is read by five disks that store data blocks and redundant data that make up the same parity sequence. In either case, four or five disks generate a read response indicating success or failure of reading. Each of the four or five disks transmits the read response generated by the disk to the control unit 7 through a SCSI interface connected to the disk. Control unit 7
Executes the processing procedure shown in the flowchart of FIG. 37 every time a read response arrives. The flowchart in FIG. 37 includes the same steps as those in the flowchart in FIG. 28 except that the flowchart further includes step S141. Therefore, in FIG. 37, steps corresponding to those shown in FIG. 28 are denoted by the same step numbers, and description thereof will be omitted.

When the controller 7 determines that the read response (failure) has arrived (step S82), it extracts the LBA from the arrived read response (failure). The LBA included in the read response (failure) indicates the storage location of the data block or redundant data that failed to be read. The control unit 7 extracts the LBA extracted from the read response (failure),
Register in the failed block table 75 (step S14)
1). It should be noted that step S141 may be executed at any timing after it is determined in step S82 that the read response (failure) has arrived. That is, the execution timing of step S141 is not limited to immediately after it is determined that the read response (failure) has arrived in step S82.

Incidentally, the reassignment processing section 8 executes the processing procedure described in the sixth embodiment. This processing procedure has already been described in the sixth embodiment, and will not be described again in this embodiment. An important operation of the reassignment processing unit 8 in the present embodiment is to transmit a processing completion notification, which is a signal indicating that, to the control unit 7 when the reassignment processing ends. This processing completion notification includes the LBA indicating the storage location determined to be defective by the reassignment processing unit 8. As described above, it takes time to read data from the defect storage location. Therefore, the LBA indicating the defect storage location is also described in the failed block table 75.

Upon receiving the processing completion notification, the control unit 7
The procedure is executed according to the flowchart shown in FIG. Upon receiving the processing completion notification, the control unit 7 determines that the reassignment processing unit 8 has executed the reassignment process (step S151), and proceeds to step S152. In step S152, the control unit 7 extracts the LBA from the processing completion notification. Next, the controller 7 sets the failed block table 75
And deletes the LBA that matches the one extracted from the processing completion notification from the failed block table 75. Thus, the control unit 7 sets the failed block table 7
5 is updated (step S152).

As described above, the RA according to the seventh embodiment
The ID device is also considered to have a defect in a storage location that takes a long time to read, and is assigned an alternative storage location. That is, the storage location of the data block or redundant data is changed from the defective storage location to the alternative storage location. In response to the reassignment process, the control unit 7 updates the failed block table 75. Therefore, the data block or the redundant data is not stored in the defective storage location for a long time. Further, each time the reassignment process is executed, the number of LBAs described in the failed block table 75 decreases. As a result, the storage location (LBA) of the data block from the address conversion processing unit 11
Is less likely to be described in the failed block table 75. Therefore, the control unit 7 often transmits four second read requests. As a result, it is possible to realize a RAID device that can read more data per unit time.

In the above embodiment, the RAID device has been described as including five disks. However, this number may not be the same as the data length or data block length.
It is changed according to the design requirements of the ID device. Therefore, R
The number of AID devices should not be limited to five.
Note that “m” in the claims corresponds to the number of disks included in the RAID device.

In the above embodiment, the host device stores 2048 bytes of data. It has been described that the data is transmitted to the RAID device of each physical form, and the RAID device divides the data arriving from the host device into 512-byte data blocks. However, the size is selected to simplify the description of the embodiment, and the sizes of the data and the data block may be other than 2048 bytes and 512 bytes.

[Brief description of the drawings]

FIG. 1 shows a block configuration of a RAID device according to a first embodiment.

FIG. 2 shows buffer units 3A to 3D, 3P, and 3R of FIG.
3 shows a detailed configuration.

FIG. 3 shows a parity sequence stored in disks 5A to 5D and 5P of FIG.

FIG. 4 is a flowchart illustrating a procedure of a process executed by a control unit 7 of FIG. 1;

FIG. 5 is a diagram for explaining one of the technical effects of the RAID device of FIG. 1;

FIG. 6 shows the disks 5A to 5D and 5P of FIG. 1, and particularly shows a type in which the order of read processing is changed.

FIG. 7 is a diagram for explaining one of technical effects of the RAID device of FIG. 1;

FIG. 8 is a flowchart illustrating a procedure of a process executed by a control unit 7 according to the second embodiment.

FIG. 9 is an issue time table 7 created internally by the control unit 7;
1 is shown.

FIG. 10 is a diagram for explaining one of the technical effects of the RAID device according to the second embodiment.

FIG. 11 shows a block configuration of a RAID device according to a third embodiment.

FIG. 12 is a flowchart executed by the control unit 7 in FIG. 11 at the time of read processing.

FIG. 13 shows a probability distribution curve f (t) and each of disks 5A to 5A.
5D shows the progress of reading in 5D.

FIG. 14 is a diagram for explaining a case where four data blocks are prepared in step S44 of FIG. 12 and a case where the first timer 72 times out in step S45.

FIG. 15 is a block diagram illustrating a RAID device according to a fourth embodiment.

FIG. 16 is a flowchart executed by the control unit 7 of FIG. 15 during a read process.

FIG. 17 shows a reservation table 73 of FIG.

FIG. 18 illustrates a specific example of a read process of the RAID device in FIG. 15;

FIG. 19 is a block diagram showing a RAID device according to a fifth embodiment.

20 shows data blocks and redundant data stored in a distributed manner on each of the disks 5A to 5D and 5P in FIG. 19;

FIG. 21 is a flowchart executed by the control unit 7 in FIG. 19 at the time of read processing.

FIG. 22 shows a failed block table 75 of FIG.

FIG. 23 is a view for explaining one of the technical effects of the RAID device according to the fifth embodiment.

FIG. 24 is a block diagram illustrating a configuration of a RAID device according to a sixth embodiment of the present invention.

FIG. 25 is a diagram showing a first table 91 managed by the alternative recording area information management unit 9 of FIG. 24;

FIG. 26 is a flowchart showing a processing procedure of the control unit 7 after a first read request arrives.

FIG. 27 is a diagram showing a second table 101 managed by the address information management unit 10 of FIG.

FIG. 28 is a flowchart showing a processing procedure of the control unit 7 after one read response arrives.

FIG. 29 is a block diagram showing a detailed configuration of the reassignment processing unit 8 of FIG.

FIG. 30 is a flowchart showing a part of a procedure of a reassignment process executed by the reassignment processing unit 8;

FIG. 31 is a diagram for describing a first list 82 and a second list 83 shown in FIG. 29.

FIG. 32 is a flowchart illustrating another part of the procedure of the reassignment process executed by the reassignment processing unit 8 in FIG. 24;

FIG. 33 is a flowchart showing a procedure of processing executed by the reassignment processing unit 8 in FIG. 24 in response to reception of a reception notification.

FIG. 34 is a flowchart showing a processing procedure executed when the reassignment processing unit 8 of FIG. 24 receives a read stop request.

FIG. 35 is a block diagram illustrating a configuration of a RAID device according to a seventh embodiment of the present invention.

FIG. 36 is a flowchart showing a processing procedure from when a first read request arrives at the control unit 7 in FIG. 35 to when a second read request set is transmitted.

FIG. 37 is a flowchart showing a processing procedure executed by the control unit 7 of FIG. 35 each time a read response arrives.

FIG. 38 is a flowchart showing a processing procedure executed when the control unit 7 of FIG. 35 receives a processing completion notification.

FIG. 39 shows a conventional RAID device employing a level 3 architecture.

FIG. 40 is a diagram for explaining a method of generating redundant data in a conventional RAID device.

FIG. 41 is a diagram for explaining a problem of the first RAID device disclosed in Japanese Unexamined Patent Publication No. 2-81123.

FIG. 42 is a diagram for explaining a problem of the second RAID device disclosed in Japanese Unexamined Patent Application Publication No. 9-69027.

[Explanation of symbols]

 5A to 5D, 5P disk 6 parity operation unit 7 control unit 71 issue time table 72 first timer 73 reservation table 74 second timer 75 failed block table 8 reassign processing unit 9 substitute Recording area information management unit 10: Address information management unit 11: Address conversion processing unit

 ──────────────────────────────────────────────────の Continuing on the front page (72) Inventor Yukiko Ito 1006 Kazuma Kadoma, Osaka Prefecture Matsushita Electric Industrial Co., Ltd. Terms (reference) 5B065 BA01 CA07 CA12 CA15 CA30 CC08 CH13 EA02 EA04 EA12 EA15 EA18 EA19 EA24

Claims (19)

[Claims] 1. A data block generated by dividing data and redundant data generated from the data block are recorded. The data block responds to a first read request transmitted from the outside. A RAID device that executes a read process of m, wherein data blocks and redundant data are stored in a distributed manner.
And a control unit for controlling the read process, wherein the control unit responds to the input first read request by
Issue a second read request to read data blocks and redundant data from the number of disks, and detect a disk that no longer needs to read data blocks or redundant data from the m disks; A RAID device that issues a read stop command in order to stop reading from a disk for which reading has been detected as unnecessary. 2. The control section, when (m-1) the disks have completed reading,
2. The RAID device according to claim 1, wherein it is determined that the reading performed on the remaining one disk is unnecessary, and a read stop command is issued to the disk for which the reading is determined to be unnecessary. 3. When the control unit detects that two or more disks cannot be read, the control unit determines that reading of the disk executed at the time of the detection is unnecessary, and determines that the disk is not required to be read. The RAID device according to claim 1, wherein a read stop command is issued to the RAID device. 4. The control unit, when (m-1) the disks have completed reading,
The RAID device according to claim 1, wherein it is determined that a read that has not been executed on the remaining one disk is unnecessary, and a read stop command is issued to the disk for which the read is determined to be unnecessary. 5. A data block generated by dividing data and redundant data generated from the data block are recorded. In response to a first read request from an external host device, the data block A RAID device that performs a read process of (a), wherein m disks in which the data blocks and the redundant data are stored in a distributed manner, and (m−2) data blocks and the redundant data, A parity operation unit that executes an operation to restore the remaining one data block; and a control unit that controls the read process. The control unit responds to the input first read request. And the m
Issue a second read request to read data blocks and redundant data from said disks, and (m-1) said disks complete reading,
What is read from the (m-1) disks is
Detects whether the combination is a combination of data block and redundant data. If the combination is detected as a combination of data block and redundant data, after waiting for a predetermined time from the time of the detection, the remaining one disk In order to restore the data block that has not been read, a restoration instruction is issued to the parity operation unit. When m data blocks are prepared by the parity operation of the parity operation unit, data is transmitted to the host device. RAID device, wherein the predetermined time is selected as a value that guarantees that data is transmitted to the host device without delay. 6. The control unit further comprises: when (m-1) disks have been read,
What is read from the (m-1) disks is
It is detected whether or not the combination is a data block and redundant data, and when it is detected that the combination is not a combination of the data block and redundant data, the data is transmitted to the host device without waiting for a predetermined time from the detection. The RAID device according to claim 5, wherein the RAID device performs a process for transmitting. 7. The RAID according to claim 5, wherein the predetermined time is selected based on a probability that the reading is expected to be completed from the start of reading of each of the disks.
apparatus. 8. A data block generated by dividing data and redundant data generated from the data block are recorded, and the data block is stored in response to a first read request from an external host device. A RAID device that executes a read process of (a), wherein m disks in which the data blocks and the redundant data are stored in a distributed manner, and (m−2) data blocks and the redundant data, A parity operation unit that performs an operation to restore the remaining one data block; and a control unit that controls the read process. The control unit responds to the input first read request. And the m
Issue a second read request to read data blocks and redundant data from said disks, and when (m-1) said disks have completed reading,
What is read from the (m-1) disks is
Detects whether the combination is a combination of data block and redundant data. If the combination is detected as a combination of data block and redundant data, after waiting for a predetermined time from the time of the detection, the remaining one disk A restoration instruction is issued to the parity operation unit to restore the data block that has not been read, and when m data blocks are aligned by the parity operation of the parity operation unit, the data is transmitted to the host device. A RAID device that executes processing, and wherein the restoration instruction is issued when the parity operation unit is not executing a parity operation. 9. A parity operation unit includes a table describing a time period in which the parity operation is performed, and the control unit further refers to the time period described in the table, and the parity operation unit performs the parity operation. 9. The RAID device according to claim 8, wherein a restoration instruction is issued when not executing. 10. A data block generated by dividing data and redundant data generated from the data block are recorded. In response to a first read request from an external host device, the data block A RAID device that executes a read process of (a), wherein m disks in which the data blocks and the redundant data are stored in a distributed manner, and (m−2) data blocks and the redundant data, A parity operation unit that performs an operation to restore the remaining one data block; and a control unit that controls the read process. The control unit responds to the input first read request. And (m-
1) It is determined whether or not the disks have failed in reading data blocks in the past, and if it is determined that reading of each data block has not failed in the past, only the data blocks are read. Therefore, a second read request is issued to the (m-1) disks, and when a data block is read from the (m-1) disks, data is transmitted to the host device. A RAID device that performs processing. 11. The control unit further comprising: if it is determined that reading of any of the data blocks has failed in the past, reading out the (m−1) data blocks and redundant data. 2 is issued to the m disks, and when (m-1) disks have completed reading,
What is read from the (m-1) disks is
Detecting whether or not the combination is a data block and redundant data, and if detecting a combination of the data block and the redundant data, recovering the data block that has not been read from the remaining one disk 11. The apparatus according to claim 10, wherein a restoration instruction is issued to the parity operation unit, and when m data blocks are prepared by the parity operation of the parity operation unit, a process for transmitting data to the host device is executed. 12. RAID
apparatus. 12. The control unit according to claim 1, further comprising: a table describing a storage location of a data block from which each disk has failed to read in the past, wherein the control unit refers to the table,
1) issue a second read request to the disks,
The RAID device according to claim 10, wherein it is determined whether to issue a second read request to the m disks. 13. When a defect occurs in a storage location of a data block or redundant data stored in each of the m disks, a reassignment processing unit that executes a reassignment process of assigning an alternative storage location instead of the defective storage location is provided. Further, the control unit, when an alternative recording location is reassigned to a storage location of the data block described in the table by the reassignment processing unit, deletes the storage location of the data block from the table. R described in 12
AID device. 14. An alternative recording area management unit that manages, as alternative recording area information, an address of an alternative recording area previously reserved in each of the m disks, and an address of an alternative recording area assigned instead of a defective recording area. And an address management unit that manages information, wherein the reassignment processing unit, when a second read request from the control unit is transmitted to the m disks, measures a delay time in each of the disks, Based on the measured delay time, it is determined whether the data block or the redundant data recording area to be read by each of the second read requests is defective, and the data block or the redundant data recording area is defective. If it is determined that the defective recording area is replaced based on the alternative recording area information managed by the alternative recording area management unit. And assigning the address information of the assigned alternative recording area to the address management unit. The control unit issues a second read request based on the address information managed by the address management unit. The RAID device according to claim 13, wherein the delay time is an elapsed time calculated from a predetermined processing start time. 15. When a defect occurs in a storage location of a data block or redundant data stored in each of the m disks, a reassignment processing unit that executes a reassignment process of assigning an alternative storage location instead of the defective storage location is provided. The RAID device according to claim 1, further comprising: 16. An alternative recording area management unit that manages, as alternative recording area information, an address of an alternative recording area secured in advance in each of the m disks, and an address of an alternative recording area assigned instead of a defective recording area. And an address management unit that manages information, wherein the reassignment processing unit, when a second read request from the control unit is transmitted to the m disks, measures a delay time in each of the disks, Based on the measured delay time, it is determined whether the data block or the redundant data recording area to be read by each of the second read requests is defective, and the data block or the redundant data recording area is defective. If it is determined that the defective recording area is replaced based on the alternative recording area information managed by the alternative recording area management unit. And assigning an address management unit to manage the address information of the assigned alternative recording area. The control unit issues a second read request based on the address information managed by the address management unit. The RAID device according to claim 15, wherein the delay time is issued, and the delay time is an elapsed time calculated from a predetermined processing start time. 17. The RAID according to claim 16, wherein the reassignment processing unit assigns an alternative recording area to the defective recording area only when the same recording area is determined to be a defective recording area continuously for a specified number of times. apparatus. 18. The predetermined processing start time is set to
17. The RAID device according to claim 16, wherein the time is a time at which the second read request is transmitted to the plurality of disks. 19. The RAI according to claim 16, wherein the predetermined processing start time is a time at which reading by the m disks based on a second read request is started.
D device.