Disk storage system and data update method used therefor

Details Disk storage system and data update method used therefor Disk storage system and data update method used therefor

1998-08-06

2001-05-15

Yoo, Do Hyun (Department: 2187)

Electrical computers and digital processing systems: memory

Addressing combined with specific memory configuration or...

Dynamic-type storage device

C711S103000, C711S118000, C711S170000, C711S205000

Reexamination Certificate

active

06233648

ABSTRACT:

BACKGROUND OF THE INVENTION
The entire contents of Japanese Patent Application No. H09-366782 filed on Dec. 26, 1997 are incorporated herein by reference.
The present invention relates to a disk storage system including a disk array unit called a RAID (Redundant Arrays of Inexpensive Disks) and, more particularly, to a disk storage system having an improved data update function.
A disk array unit (called a RAID) constituted by a plurality of disk drives is known. For example, a RAID is a disk storage system which is connected to a network server and can store a large volume of data. As a high-speed data write method for such a RAID, in particular, a method of writing new data in bulk in data update processing without rewriting the new data in the storage area of old data has been proposed. More specifically, the new data is stored in a prepared write buffer, and the stored new data is written in empty areas prepared in disk drives in bulk (see Jpn. Pat. Appln. KOKAI Publication Nos. 6-214720 and 6-266510).
A conventional write method (data update method) will be briefly described with reference to FIG.
1
. Referring to
FIG. 1
, reference symbols L
1
through L
99
denote logical blocks; and P
1
through P
56
, physical blocks in a disk unit (disk drives). Assume that logical blocks L
6
, L
4
, L
2
, L
12
, L
7
, and L
11
are to be updated. The old data of these logical blocks are present in physical blocks P
6
, P
4
, P
2
, P
12
, P
7
, and P
11
in the disk unit. According to a general write method, the contents of physical blocks P
6
, P
4
, P
2
, P
12
, P
7
, and P
11
are updated. According to a high-speed write method, however, while the data of physical blocks P
6
, P
4
, P
2
, P
12
, P
7
, and P
11
are maintained, the new data of logical blocks L
6
, L
4
, L
2
, L
12
, L
7
, and L
11
are written in physical blocks P
51
, P
52
, P
53
, P
54
, P
55
, and P
56
as other prepared empty areas in bulk. With this operation, the number of writes (six writes are required in the general write method) can be reduced to three in the high-speed write method when, for example, two physical blocks are written in bulk, thereby attaining an improvement in write performance.
For the sake of descriptive convenience, in the example shown in
FIG. 1
, two physical blocks are written in one disk in bulk. In practice, however, several ten physical blocks are written in bulk. According to a RAID architecture of level 4 (RAID4 architecture) and a RAID architecture of level 5 (RAID5 architecture), since one stripe (N*K logical blocks; N is the number of drives, and K is the number of blocks) can be rewritten in bulk, no disk read is required for parity maintenance, and the overhead in a write can be reduced. According to the RAID4 architecture, data are arranged in large units such as sectors and blocks instead of small units such as bits and bytes, and each disk is independently operated in accordance with a request to read small-volume data. According to the RAID5 architecture, redundant data (parity data) is cyclically set in the respective data disks instead of being stored in a dedicated parity disk.
Since the latest data of logical blocks L
6
, L
5
, L
2
, L
12
, L
7
, and L
11
are present in physical blocks P
51
through P
56
in the disk unit, the contents of an indirect map are rewritten to indicate correct disk positions. The data of logical blocks are read out after the latest disk positions are obtained by looking up this indirect map. No old data is therefore erroneously read out.
In the conventional high-speed write method described above, as shown in
FIG. 1
, a write buffer is prepared in a nonvolatile memory, the empty area in the disk unit is divided into blocks, and data are sequentially compacted/stored in the empty blocks through the write buffer. For this reason, even if contiguous data are to be read out, the data must be divided into blocks, and the addresses must be converted in units of blocks. When, therefore, a request to read data exceeding the data length of a logical block is sent from a host system, the requested data must be divided into a plurality of logical blocks.
In addition, every time data update processing is performed, in particular, a search for data to be updated in the entries corresponding to a buffer management table must be performed by using a conversion map. The processing time is therefore prolonged by the time required for this search. Furthermore, the conversion map is dependent on the storage capacity of the disk unit and must be created in the main memory of the system. This conversion map is therefore a factor that limits the capacity used by the main memory.
BRIEF SUMMARY OF THE INVENTION
It is an object of the present invention to improve performance associated with data update processing when a high-speed write method is applied to a disk storage system having particularly a RAID architecture.
The present invention relates to a scheme of executing read processing in accordance with a request to read contiguous data without theoretically dividing any logic block. The present invention also relates to a scheme of improving the efficiency of search processing of searching the entries of a conversion map corresponding to a buffer management table in data update processing.
More specifically, according to the present invention, there is provided a disk storage system constituted by N disk drives, comprising write buffer means having a storage capacity corresponding to N*K (K is an integer indicating the number of blocks) logical blocks, cache management means for managing the write buffer as a cache memory for storing read and write data to be transferred between the respective disk drives, and storing, in the write buffer, the logical blocks whose data lengths are changed, as needed, and data update means for delaying updating of logical blocks stored in the write buffer in data update processing until the number of stored logical blocks reaches N*K−1, creating a logical address tag block constituted by logical addresses of the respective logical blocks stored in the write buffer, selecting contiguous storage areas (in units of sectors) from empty areas different from areas of the respective disk drives in which old data to be updated are stored, and sequentially writing N*K logical blocks, obtained by adding the logical address tag block to the N*K−1 logical blocks, in the selected areas by a continuous write operation.
According to another aspect of the present invention, in the disk storage system, the data update means has a function of delaying updating of logical blocks stored in the write buffer in data update processing until the number of stored logical blocks reaches N*K−1, creating a logical address tag block constituted by logical addresses of the respective logical blocks stored in the write buffer, creating K parity blocks from N*K logical blocks obtained by adding the logical address tag block to the N*K−1 logical blocks, selecting contiguous storage areas (in units of sectors) from empty areas different from areas of the respective disk drives in which old data to be updated are stored, and sequentially writing N*K logical blocks having the parity blocks added thereto in the selected areas by a continuous write operation.
With this arrangement, when contiguous data are to be read out, an empty area of the write buffer in the nonvolatile memory need not be divided in units of logical blocks.
According to the present invention, since the data to be read out from the disk unit is divided into physically contiguous blocks to be compacted/stored in empty areas of the write buffer in the nonvolatile memory. With this scheme, addresses can be converted in bulk. An input/output request for a read need not be divided. In addition, since a conversion map can be created in the main memory instead of the nonvolatile memory independently of the capacity of the disk unit, the storage area of the main memory can be efficiently used by paging. If the present invention is applied to a disk storage system having

Affiliated with

Also associated with

Rating

Disk storage system and data update method used therefor does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Disk storage system and data update method used therefor, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Disk storage system and data update method used therefor will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-2528027