JPH09223049A - Disk array system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable the execution of arrangement control for further improving the efficiency on the assumption that data can be read out at an extremely high speed by the arrangement control of striping. SOLUTION: Based on the correspondent relation between the chunk of a logical disk and the chunks of disk devices A-D, arrangement control is performed so that the continuous chunks on a file can be respectively allocated to different disk devices. Basically, the allocation is to be performed in the order of idle chunks, namely, in the order of chunks 3 → 4 → 5 → 9 of logical disk, but since the chunks 3, 4 and 5 can be respectively allocated to the respective disk devices D, A and B, these chunks can be allocated as they are. In the case of chunk 9, on the other hand, since it is continuously allocated to the same disk device B as the chunk 5, the chunk 9 is skipped and the next chunk 10 is allocated.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a disk array device in which a plurality of disk devices are operated in parallel so as to respond to a request from a main computer by masquerading as one disk device. It is a thing.

[0002]

2. Description of the Related Art Conventionally, there has been known a disk array device in which a plurality of disk devices are operated in parallel so as to respond to a request from a main computer by seemingly one disk device. Therefore, a file system using this disk array device is also considered. Using a disk array device as one logical disk, and for the logical blocks (management units, called chunks) that make up the storage area of the logical disk, consecutive logical blocks are physical blocks (chunks) of different disk devices. There is a disk array access method that achieves a high-speed response to an access from the main computer by setting the disk array to be assigned to. This is called striping, and RAID (Red
undant Arrays of Inexpensive Disks) Level 0,
It is applied to 3, 4, 5, etc.

An example of this striping is shown in FIG. This is equivalent to RAID level 4, A to D
The chunks 0, 1, 2, and 3 are sequentially written in the four disk devices of No. 3, and the parity data P (0-3) corresponding to the chunks 0 to 3 are written in the remaining one E disk device. Similarly, chunks 4 are sequentially added to the A to D disk devices.
5, 6, and 7 are written, and the parity data P (4-7) corresponding to the data 4 to 7 is written to the remaining E disk device. The same applies to chunk 8 and thereafter.

When this striping is executed, when writing a file, if the file blocks constituting the file are sequentially arranged in continuous chunks on the logical disk, the continuous chunks are arranged in chunks of different disk devices. You. Therefore, when a file is sequentially read, the data is not continuously read from the same disk device, and a chunk of another disk device can be sought during a data transfer from a chunk of one disk device, or a cache can be read. Can read ahead to Since a plurality of disk devices can be operated in parallel in this way, very high-speed data reading as a whole has been possible.

[0005]

However, usually,
When files are deleted or generated repeatedly, the files are not always stored continuously due to the difference in file size, and it is highly conceivable that the free areas of the files are divided and generated. Then, when a file is newly stored after that, division may occur in the storage area on the logical disk.

In the striping described above, if consecutive chunks are sequentially allocated on the logical disk, consecutive chunks are always arranged in another disk device. This is not the case when division occurs in the memory area of. That is, if the chunks that are divided and exist on the logical disk are sequentially allocated, there is a possibility that the same disk device will be continuously accessed even when the files are read sequentially.

This point will be described with reference to the example shown in FIG.
In this case, as a result of deletion / generation of this file, chunks 0, 1, 2, 6, 7 on the logical disk are used, and chunks 3, 4, 5, 9, 10, 10, 11 ... Are empty. Has become. When allocating a storage area to a file in this state, if normal striping is applied, allocation is performed in the order of free chunks on the logical disk, that is, chunk 3 → chunk 4 → chunk 5 → chunk 9. Then, chunk 3, chunk 4, and chunk 5 may be allocated as D disk device, A disk device, and B disk device, respectively, but in the case of chunk 9, they are continuously allocated to the same B disk device as chunk 5. It will be assigned.

In that case, when reading, chunk 5
Will be read from chunk 9 after
During the data transfer of the above, it is not possible to exert the merit of seeking the next chunk with another disk device or even prefetching to the cache. Therefore, when the device is used for a video server, for example, smooth data reading may be interrupted.

The present invention has been made in order to solve the above-mentioned problems, and is basically based on the premise that very high-speed data reading can be performed by the layout control by the striping. Another object of the present invention is to provide a disk array device capable of executing layout control that further improves its efficiency.

[0010]

In order to achieve this object, the invention according to claim 1 controls a plurality of disk devices by a disk array controller, and responds to an access from a main computer. Are configured to respond as an apparently one disk device, and a plurality of logical blocks that form the storage area of the apparently one disk device and a plurality of logical blocks that form each storage area of the plurality of disk devices. A physical block group is a disk array device in which consecutive logical blocks are set in a correspondence relationship in which they are assigned to physical blocks of the different disk device, and the disk array controller responds to a request from the main computer. File management means for managing logical block groups in file units When arranging a plurality of file blocks constituting a file in a logical block group as a part of file management by the file management means, continuous file blocks on a file are created based on the correspondence between the logical blocks and physical blocks. A disk array device, comprising: a file allocation control means for controlling allocation so as to be allocated to different physical block groups.

In the disk array device of the present invention, a logical block group and a plurality of physical block groups of a plurality of disk devices are set in a correspondence relationship in which continuous logical blocks are assigned to physical blocks of different disk devices ( When writing a file, when the file blocks that make up the file are sequentially arranged in consecutive logical blocks, the consecutive logical blocks are arranged in different physical blocks. Therefore, when the file is sequentially read by the file management means, the file is not continuously read from the same physical block, and the physical block of another disk device is sought during data transfer from a certain physical block, Furthermore, since pre-reading to the cache can be performed, very high-speed data reading can be performed as a whole.

However, when the file management means repeatedly deletes or creates a file, it is conceivable that an empty area of the file may be divided. In this case, if the logical blocks that are divided and exist in sequence are sequentially allocated, the same disk device may be accessed consecutively when the file is read sequentially. The benefits are lost.

According to the present invention, it is possible to deal with the case where the free area of such a file is divided and generated. That is, when arranging a plurality of file blocks that make up a file in a logical block group as part of file management by the file management means, the file allocation control means determines on the file based on the correspondence between the logical blocks and the physical blocks. The placement is controlled so that consecutive file blocks are assigned to different physical block groups.

Therefore, in the example shown in FIG. 1B, basically, the logical blocks that are vacant (empty chunks) are allocated in the order of chunk 3 → chunk 4 → chunk 5 → chunk 9. This is done by allocating while determining whether or not they will be different physical block groups, that is, different disk devices. Since chunk 3, chunk 4, and chunk 5 are allocated as D disk device, A disk device, and B disk device, respectively, they are allocated as they are. On the other hand, in the case of chunk 9, since it is continuously allocated to the same B disk device as chunk 5, chunk 9 is skipped and an attempt is made to allocate to the next chunk 10. Since the chunk 10 is a disk device different from the chunk 5, it is allocated to the chunk 10.

In this case, when reading, the chunk 5
After that, the data is read from the chunk 10, and the advantages such as seeking the chunk 10 in another disk device during data transfer of the chunk 5 and further prefetching to the cache can be exhibited.

As described above, not only when a file is continuously allocated to a group of logical blocks, but also when an empty logical block (empty chunk) is generated in a divided manner, the continuous file The respective file blocks can be allocated to different physical block groups, and the advantage that a plurality of disk devices are operated in parallel and data can be read at extremely high speed as a whole is not impaired. Therefore, it is particularly effective in a system such as a video server that continuously takes out data at a constant speed.

As described above, the layout control may be performed so that the continuous file blocks on the file are allocated to different physical block groups, but the layout control described in claim 2 may be further performed. . That is,
Allocating physical block groups that can be allocated based on a predetermined allocation order, and allocating consecutive file blocks on a file to different physical block groups.

If the arrangement control is performed so that consecutive file blocks are allocated to different physical block groups, for example, in the example shown in FIG. 1B, for example, chunk 3 → chunk 4 → chunk 5 → chunk 9 in this order. Since the chunk 9 is the same B disk device as the chunk 5 when it is allocated, it may be allocated to the chunk 12 of the A disk device. Of course, the effect is still exhibited, but it can be said that allocation in the chunk 10 based on a predetermined allocation order is preferable in this case.

The reason will be described. Operating the disk devices in parallel leads to the basic effects of the disk array device, and the speed of data reading is realized. Considering this point, in the case shown in FIG. 1, since the chunks 3, 4, 5, and 10 exist in different disk devices, respectively, these four chunks can be read simultaneously. On the other hand, chunks 3, 4, 5, 12
In this case, since the chunks 3, 4, and 5 exist in different disk devices, the simultaneous reading is possible, but the chunk 12 is not read after the reading of the chunk 4 is completed. And cannot read. Therefore, in order to further improve the overall read efficiency, the basic allocation control of allocating consecutive file blocks on the file to different physical block groups is maintained, while the physical block groups that can be allocated are further allocated. It is effective to perform allocation based on a predetermined allocation order.

[0020]

Next, an embodiment of the present invention will be described. The disk array device 30 according to the embodiment of the present invention shown in FIG. 2 includes five disk devices 31, 32,
33, 34, 35 and their respective disk devices 31 to 35
And a disk array controller 40 capable of individually controlling the above. In the following description, in order to distinguish between the five disk devices, they will be described as A disk device 31, B disk device 32, C disk device 33, D disk device 34, and E disk device 35. The disk devices 31 to 35 are so-called physical hard disk drives and a control board for controlling the same, which are integrated.

The disk array device 30 includes a main computer 10 and a SCSI (Small Computer System Interfac).
e) Connected via a standard bus 20, and the main computer 10 looks like a single disk device. The main computer 10 has a keyboard 11 as an input means and a CRT as a display means.
12 are connected.

The disk array controller 40 includes a CPU 41 as control means and array management software 42.
a and the like are provided with a ROM 42 as a storage means for storing the operation program of the CPU 41, a RAM 43 as a work area of the CPU 41, an interface device 44 for a disk device, etc., and the interface device 44 based on the array management software 42a etc. The five disk devices 31 to 35 can be simultaneously operated in parallel via the. Then, the disk array controller 40
Logical blocks (chunks) constituting a storage area of a logical disk (virtual disk) and A to D disk devices 31 to 3
4 stores the correspondence with the physical blocks (chunks) forming the storage area, and the data write control is performed based on the correspondence.

The disk array device 30 of this embodiment is
This is equivalent to RAID level 4, and chunks 0, 1, 2, and 3 are sequentially written in four disk devices 31 to 34 of A to D, and parity data P (0-3) corresponding to the chunks 0 to 3 is written. Striping can be executed to write the data to the remaining one E disk device 35. Similarly, A
~ Chunks 4, 5, to D disk devices 31-34 sequentially
6, 7 are written, and the parity data P (4-7) corresponding to the data 4-7 are written in the remaining E disk device 35. The same applies to chunk 8 and thereafter.

Since the parity data P is stored in the E disk unit 35, even if the data from any one of the four units A to D disk units 31 to 34 cannot be read out, the data is read from the other three units. By performing a predetermined error correction function on the basis of the data of No. 1 and the parity data P written in the disk device 35, it is possible to create (restore) data from the disk device that could not read the data. .

The execution of the above striping has the following advantages. That is, when writing a file, when the file blocks forming the file are sequentially arranged in consecutive chunks on the logical disk, the consecutive chunks are arranged in chunks of different disk devices 31 to 34, respectively. Therefore, when reading files sequentially, the same disk devices 31 to 31
It is possible to seek a chunk of another disk device 31 to 34 while data is being transferred from a chunk of one disk device 31 to 34 without being continuously read from 34, or to perform prefetch to a cache. As described above, since the plurality of disk devices 31 to 34 can be operated in parallel, it is possible to read data at extremely high speed as a whole.

However, when files are repeatedly deleted or created, the files are not always stored continuously because of the difference in file size, and the free areas of the files are often fragmented. Conceivable. Then, when a file is newly stored after that, division may occur in the storage area on the logical disk.

Also in the above striping, if consecutive chunks are sequentially allocated on the logical disk, consecutive chunks are always allocated to different disk devices 31 to 31.
However, if division occurs in the storage area on the logical disk, this is not the case. That is, if the chunks that are divided and exist on the logical disk are sequentially allocated, there is a possibility that the same disk device will be continuously accessed even in the sequential reading of files.

This point will be described with reference to FIG.
In this case, chunks 0, 1, 2, on the logical disk
6,7 are used, chunks 3,4,5,9,1
0 is 11, 11 ... is vacant. When allocating a storage area to a file in this state, if normal striping is performed, allocation will be performed in the order of empty chunks, that is, chunk 3 → chunk 4 → chunk 5 → chunk 9. Then, chunk 3, chunk 4, and chunk 5 may be allocated as D disk device 34, A disk device 31, and B disk device 32, respectively, but in the case of chunk 9, the same B disk device 32 as chunk 5 is allocated. Will be allocated continuously.

In that case, when reading, chunk 5
Will be read from chunk 9 after
While the data is being transferred, another disk device 31 to 34 cannot seek the next chunk, and the prefetch to the cache cannot be achieved.

The disk array device 30 is capable of coping with the case where such a free area of a file is divided and generated. That is, as shown in FIG. 1B, when allocating a plurality of file blocks (chunks 0 to 4 in this case) forming a file to chunks on the logical disk, chunks of the logical disk and A to
Based on the correspondence with the chunks forming the storage areas of the D disk devices 31 to 34, the allocation control is performed so that consecutive chunks on the file are allocated to different disk devices 31 to 34, respectively.

The allocation control process will be further described with reference to the flowchart of FIG. FIG. 3 shows a file area allocation control process executed by the disk array controller 40 when a file write request is issued from the main computer 10.

When there is a file write request from the main computer 10 and the designated write position is an area which has not been allocated yet, it is necessary to allocate the area to the file. In that case, the first step S
In 100, an unused logical block (empty chunk) on the logical disk is extracted from the area management information written on the disk. This is the area to be newly allocated.

At S110, in order to associate the unused logical blocks with the physical blocks (chunks) in the disk devices 31 to 34, the physical numbers in the disk devices 31 to 34 are calculated from the numbers of the unused logical blocks. Convert to block number. This S11
After the number conversion process of 0, the process proceeds to S120.

In S120, an unused logical block, which is a new area to be allocated, is a physical block of the immediately preceding file block, that is, a block in the file and which is the same as the last block already allocated to the same disk device. In order to confirm whether or not the above, the logical block number of the immediately preceding file block is extracted based on the file management information written on the disk together with the above-mentioned area management information.

In the following S130, the logical block corresponding to the file block immediately before that and the disk devices 31 to 31.
In order to make correspondence with the physical block in 4,
The logical block number is converted to the physical block number in the disk devices 31 to 34. This S1
After the number conversion process in 30, the process proceeds to S140.

In S140, it is determined whether or not the physical blocks corresponding to the two logical blocks obtained as described above belong to the same disk device 31-34. Then, if a negative determination is made in S140, that is, if they do not belong to the same disk devices 31 to 34, the process proceeds to S160 to allocate to the empty area, and then this area allocation control processing ends.

On the other hand, if the affirmative judgment is made in S140, that is, if the disks belong to the same disk device 31 to 34, the allocation to the empty area is inappropriate.
At 150, the next free area is newly extracted based on the area management information, the process returns to S110, and the processing from S110 onward is repeated. Then, when a negative determination is made in S140, the processing of S160 is executed and then the area allocation control processing is ended.

FIG. 1 shows this area allocation control processing.
The example shown in (b) will be further described. Basically, the logical blocks are vacant (empty chunks) in the order, that is, chunk 3 → chunk 4 → chunk 5 → chunk 9, but they are different disk devices 31 to 34, respectively. It will be allotted while judging whether or not. For chunk 3, chunk 4 and chunk 5, D disk device 34,
Since the A disk device 31 and the B disk device 32 are allocated, they are allocated as they are. On the other hand, in the case of chunk 9, the same B disk device 32 as chunk 5
The chunk 9 is skipped and the next chunk 10 is to be allocated. Since the chunk 10 is different from the chunk 5 in the disk devices 31 to 34, it is allocated to the chunk 10.

In this case, when reading, chunk 5
After that, the data is read from the chunk 10, and the advantages such as seeking the chunk 10 in another disk device 31 to 34 during data transfer of the chunk 5 and further prefetching to the cache can be exhibited.

As described above, not only when a file is continuously allocated to a group of logical blocks, but also when an empty logical block (empty chunk) is divided and generated, continuous files The respective file blocks can be allocated to different physical block groups, and the advantage that a plurality of disk devices are operated in parallel and data can be read at extremely high speed as a whole is not impaired. Therefore, it is particularly effective in a system such as a video server that continuously takes out data at a constant speed.

As described above, the present invention is not limited to such an embodiment, and can be implemented in various forms without departing from the gist of the present invention. For example, the basic idea of the allocation control of the present invention is that the allocation control is performed so that successive file blocks on the file are allocated to the physical blocks of different disk devices 31 to 34, but such allocation control is not necessary. For example, in the example shown in FIG. 1B, when the chunk 9 is allocated in the order of chunk 3 → chunk 4 → chunk 5 → chunk 9, the chunk 9 is the same B disk device 32 as the chunk 5, so the B disk device 32 Other than the above, for example, the D disk device 3
4 chunk 11 and A disk device 31 chunk 12
You can assign it to. Of course, the effect is still exhibited, but it can be said that allocation in the chunk 10 of the C disk device 33 based on a predetermined allocation order is preferable in this case.

The reason will be described. Operating the disk devices in parallel leads to the basic effects of the disk array device, and the speed of data reading is realized. Considering this point, in the case shown in FIG. 1, since the chunks 3, 4, 5, and 10 exist in different disk devices, respectively, these four chunks can be read simultaneously.

On the other hand, when it is allocated to the chunk 11 of the D disk device 34, it is stored in the chunks 3, 4, 5, and 11, and the three chunks 3, 4, and 5 are stored in different disk devices. It can be read simultaneously because it exists, but chunk 3 is chunk 3
Can only be read after the completion of reading. Further, when the data is allocated to the chunk 12 of the A disk device 31, it is stored in the chunks 3, 4, 5 and 12, and since the three chunks 3, 4 and 5 exist in different disk devices, simultaneous reading is possible. Possible, but chunk 12
Can be read only after the reading of chunk 4 is completed. Therefore, in order to further improve the overall read efficiency, the basic allocation control of allocating consecutive file blocks on the file to different physical block groups is maintained, while the physical block groups that can be allocated are further allocated. It is effective to perform allocation based on a predetermined allocation order. In this respect, in the case of this example, allocation to the chunk 10 of the C disk device 33 based on a predetermined allocation order is preferable.

[Brief description of drawings]

FIG. 1A is an explanatory diagram of striping in a disk array, and FIG. 1B is an explanatory diagram regarding allocation of file chunks when free areas are dispersedly present in a logical disk.

FIG. 2 is an explanatory diagram showing a schematic configuration of a disk array device and a main computer connected thereto as an embodiment.

FIG. 3 is a flowchart showing an area allocation control process executed in a disk array controller.

[Explanation of symbols]

10 ... Main computer 20 ... Bus 30 ... Disk array device 31 ... A disk device 32 ... B disk device 33 ... C disk device 34 ... D disk device 35 ... E disk device 40 ... Disk array controller 41 ... CPU 42 ... ROM 43 ... RAM 44 ... Interface device

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁶ Identification number Agency reference number FI Technical display location G11B 20/18 572 G11B 20/18 572F

Claims (2)

[Claims] 1. A disk array controller controls a plurality of disk devices, and is configured to respond to an access from a main computer as an apparently single disk device. A logical block group forming a storage area and a plurality of physical block groups forming each storage area of the plurality of disk devices are arranged such that the continuous logical blocks are respectively allocated to the physical blocks of the different disk device. In the disk array device set by the relationship, the disk array controller includes a file management unit for managing a logical block group in file units in response to a request from the main computer, and a file management unit for managing the file. Multiple files that make up a file as part File allocation control means for arranging each of the logical blocks into a logical block group so that successive file blocks on the file are allocated to different physical block groups based on the correspondence between the logical blocks and the physical blocks. A disk array device comprising: 2. The file allocation control means allocates to the allocable physical block group based on a predetermined allocation order so that successive file blocks on the file are allocated to different physical block groups. The disk array device according to claim 1, wherein the disk array device controls the arrangement.