JPH0736633A - Magnetic disk array - Google Patents

Abstract

PURPOSE:To reduce the overhead which occurs at the time of data write to a magnetic disk array which has redundancy and is capable of parallel access to plural drives. CONSTITUTION:When a data access from a host system 2 occurs, accessed data and peripheral data are held in a data buffer memory 103 together with parity data, thereby eliminating the data read from a drive 30N at the time of write to the area held in the data buffer memory 103.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a redundant disk array using a magnetic disk used in an information processing apparatus, and more particularly to a redundant disk array for improving data access performance.

[0002]

2. Description of the Related Art Conventionally, an external storage device used in an information processing apparatus has been moving toward the use of a larger-capacity storage device due to the need of the system side. Then, a large amount of data is lost at one time due to the failure of one storage device.

Further, since the data stored in different areas of the same external storage device cannot be accessed in parallel, the throughput of the entire system is lowered.

A method devised to eliminate the adverse effects of using a single large-capacity external storage device as described above was proposed by the Department of Computer Science, University of California, Berkeley, entitled "Redundant Array of Low Cost Disks ( R
edundant Arrays of inexpe
"native Disk; abbreviated RAID)".

In contrast to this RAID, in the Department of Computer Science, University of California, Berkeley (EESC) report number UCV / CSD / 87/391, RAID 1
Five types of realization methods up to RAID5 have been proposed. In the methods defined as RAID4 and RAID5, the data for the RAID array is divided into blocks, the divided data is distributed and stored in each drive constituting the array, and the parity data is stored in another drive. is doing.

When writing to these RAIDs 4 and 5, it is necessary to read the corresponding data and the parity data corresponding thereto once and then rewrite. For this reason, extra disk access is required as compared with normal data writing to a single drive. The overhead required for reading and rewriting is called a write penalty. (For example, JP-A-61-1
90644, JP-A-63-132354,
JP-A-64-76346 and JP-A-3-289
44).

[0007]

As described above, the RAI
In the D array, the throughput of the external storage device is improved by parallel access.
In No. 5, there is a problem that the overhead at the time of writing becomes large due to the existence of the write penalty.

Therefore, the object of the present invention is to
It is an object of the present invention to provide a magnetic disk array for reducing the write penalty in the RAID5 array to improve the throughput, which is one of the purposes of the RAID array.

[0009]

According to the present invention, when a data access to a drive under its control occurs, the data existing in the vicinity where the data access is stored and the parity data of the data are prefetched. However, there is obtained a magnetic disk array having redundancy capable of holding the data.

Further, according to the present invention, an address indicating an area where data is held is stored in the prefetch buffer,
When the address is accessed, it is possible to obtain a detection device characterized by detecting that the address has been accessed.

Further, according to the present invention, the detection device comprises:
When the access to the data address where the data is held in the prefetch buffer is detected, it is possible to use the data stored in the prefetch buffer without reading the data from the disk. A characteristic magnetic disk array is obtained.

Further, according to the present invention, when the data is rewritten to the data area stored in the prefetch buffer, the data stored in the prefetch buffer is accessed without accessing the drive. There is obtained a magnetic disk array characterized by having redundancy that can be used to reconstruct redundant data.

[0013]

In the magnetic disk array according to the present invention, the data storage area that is likely to be accessed in the future and the parity data corresponding to the data storage area are held in the memory on the control mechanism of the array, not on the drive. Thus, when the data area stored in the memory is accessed, data is read from the drive and the access to the drive is reduced to reduce the write penalty in the RAID4 and RAID5 arrays.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the magnetic disk array of the present invention will be described below with reference to the drawings. FIG. 1 shows the configuration of a RAID4 disk array that is connected to an information processing apparatus that serves as a host and performs data access.

Referring to FIG. 1, a disk array 1 which realizes RAID 4 includes a host interface 101 and a host bus 20 connected to the host interface 101.
1 to the host system 2.

The disk array 1 has an array control circuit 102, and the array control circuit 102 distributes a data access command from the host system 2 to each drive, generates parity data and checks it.

The disk array 1 includes a drive 30.
, 30N, and each drive is connected to drive interfaces 111, 112, ... 11N via drive buses 311, 312, ... 30N. In RAID 4, the drive that holds the parity data is specified, but in this embodiment, drive 30P (see FIG. 2) is assumed to be the parity disk.

The disk array 1 also includes a data buffer memory 103 for storing data in the read access area, data in the vicinity of the data area, and parity data corresponding to each data area, a memory controller 104 for controlling the data buffer memory 103, and data. Buffer memory 1
A data address table 105 for storing which data address corresponds to the data stored in 03 and its parity data, and a central processing unit 106 for controlling the overall operation of the disk array 1.

Next, a method of realizing RAID4 in the present invention will be described.

RAID 4 is a type of magnetic array disk that allows data to be accessed in parallel by dividing the data for the array into blocks and allocating the divided data to a plurality of drives in a distributed manner. . Since a plurality of drives are used, the failure rate when viewed as an array is lower than that of one drive for each drive. In order to prevent this, in RAID 4, a parity drive that holds the parity of data is prepared in addition to the drive that holds the divided data.

The data distribution state in the embodiment shown in FIG. 1 is shown in FIG. It is assumed that the drive 30N forming the array stores data 4N1, 4N2, ... Divided into block units. When the data stored in the drive 30P for parity data is 4P1, 4P2, 4P3 ..., The data relational expression shown in Expression 1 must be satisfied in order to configure RAID4.

[0022]

[Equation 1]

The following procedure is required to rewrite the data in the RAID4 array for which this relational expression holds. Drive 3 specified by an arbitrary data address
If the drive data 4NX existing on 0N is rewritten, it is also necessary to rewrite the parity data 4PX generated by this drive data 4NX. At this time, the data before rewriting of the drive data 4NX is 4NX _OLD , the data to be rewritten is 4NX _NEW , the data of the parity data 4PX before rewriting is 4PX _OLD , and the data after rewriting is 4PX.
_{When set to NEW} , the relationship of Equation 2 holds.

[0024]

[Equation 2]

Therefore, drive data 4NX _OLD , 4P
If X _OLD and 4NX _NEW are known, the rewritten parity data 4PX _NEW can be generated, so it is not necessary to read the data of all the drives connected to the disk array 1 for rewriting the parity, but at least the drive 30N in which the data exists With respect to the parity drive 30P, it is necessary to access the disk twice from reading to rewriting.

In this embodiment, the overhead due to the above-mentioned parity rewriting is reduced by the following method. In the circuit of FIG. 1, the central control unit 106 includes the drive 30
When reading arbitrary drive data 4NX and its parity data 4PX from N and 30P, it is instructed not to read only the drive data 4NX, 4PX which is a necessary block but to read a specific length L area including 30N, 30P Then, the read data of the specific length L and the parity data corresponding to the data are written in the data buffer memory 103. Also, at this time, the length L held in the data buffer memory 103 at the same time.
Minute data is data corresponding to which data address, and the data and parity are data buffer memory 1
Which area of 03 is recorded is recorded in the data address table 105.

When a write access request is issued from the host system 2 while the data is held in the data buffer memory 103, the central control unit 106
Is drive data 4NY for which write access is specified
Based on the data address table 105, it is checked whether the area is held in the data buffer memory 103.

When the data is written in the area in which the data is written in the data buffer memory 103, the drives 30P and 30N are not accessed, and the parity data 4PY corresponding to the drive data 4NY _OLD from the data buffer memory 103 is not accessed. _{Read OLD} . Here, the central control unit 106 uses the data read using the relational expression shown in Expression 2 and the drive data 4NY _NEW transferred from the host system 2 and to be newly written in 4NY.
Then, future parity data 4PY _NEW is generated and stored again in the data buffer memory 103.

For this reason, even in RAID4, which normally requires reading from the drive and then writing data back to the drive as described above, the write operation can be completed without access to the drive. The calculation result temporarily written in the data buffer memory 103 is written to the drives 30N and 30P by an instruction of the central processing unit 106 immediately or when the disk array 1 is not accessed.

Although RAID 4 is shown in this embodiment, the read-ahead data and the corresponding parity are similarly stored in the buffer memory even in the RAID 5 disk array in which the parity data is distributed not in the fixed drive but in all the drives. This makes it possible to reduce the overhead during writing.

[0031]

As described above, according to the present invention, in a magnetic disk array having redundancy and capable of parallel access to a plurality of drives, when data access is performed, the data and data in the vicinity thereof are buffered. By accumulating in the memory, redundant data generated at the time of writing data is read, and the overhead due to rewriting is reduced.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a magnetic disk array of the present invention.

FIG. 2 is an explanatory diagram showing a data distribution state in the RAID4 disk array of the embodiment of FIG.

[Explanation of symbols]

 1 Disk Array 2 Host System 101 Host Interface 102 Array Control Circuit 103 Data Buffer Memory 104 Memory Controller 105 Data Address Table 106 Central Processing Unit 111 Drive Interface 301 Drive 311 Drive Bus

Claims (4)

[Claims] 1. When a data access to a drive under its control occurs, the data existing in the vicinity where the data access is stored and the parity data of the data can be prefetched and the data can be held. Magnetic disk array characterized by high redundancy. 2. The magnetic disk array according to claim 1, wherein an address indicating an area in which data is held is stored in a read-ahead buffer, and when the address is accessed, the access is detected. Characteristic detection device. 3. The magnetic disk array according to claim 2, wherein when the detecting device detects an access to a data address where data is held in the prefetch buffer, the prefetch is not read from the disk. A magnetic disk array characterized by having redundancy capable of using the data stored in the buffer. 4. The redundant magnetic disk array according to claim 3, wherein when data rewriting is performed on a data area stored in the prefetch buffer,
A magnetic disk array characterized by having redundancy capable of reconstructing redundant data by using data stored in the prefetch buffer without accessing a drive.