JPH05289818A - Disk array control system - Google Patents

Abstract

(57) [Summary] [Objective] When a plurality of sequential commands from the host are access requests for continuous areas of the disk array device, the execution of the commands is accelerated. [Structure] When a plurality of commands are sequentially sent from a host computer, these commands are executed by the host command interpreting means 1
3, the disk commands of each disk device are generated, and when these commands are access requests for one continuous area, they are combined into one command by the command clustering means 14 and further by the host command interpreting means 13. From the compiled commands, the respective disk devices are executed at substantially the same time. [Effect] Since each disk device executes each disk command substantially simultaneously, the disk rotation waiting time can be reduced.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a disk array control system for high speed data transfer in a disk array controller.

[0002]

2. Description of the Related Art In a disk array device, a plurality of disk devices are connected, and data can be read and written in parallel, so that the processing speed is higher than that of a single disk device. ..

A disk array device is composed of a plurality of disk devices and a disk array control device for controlling them, but it looks like one disk device to a computer (upper computer) connected to its upper level. Therefore, it is necessary to convert the command issued from the host computer into a command to each disk device inside the disk array control device. Such a data transfer control method is disclosed in, for example, Japanese Patent Laid-Open No. 62-13132.
It is disclosed in Japanese Patent Publication No. 1 and the like. Hereinafter, this will be described with reference to FIG.

In FIG. 16, a means 13 for executing the above command conversion and distributing the command to each disk device.
And means for issuing a command to each disk device and detecting the end of read and write data transfer.
And means for receiving a command from the host computer and notifying the end of the command, means 17 for controlling the data transfer between the host computer and each disk device, and means 16 for controlling each of these means. A single command from the host computer is made to act as a single command to each of the multiple disk devices, allowing multiple disk devices to operate in parallel in the same way, achieving high-speed data transfer. To do.

Further, the disk array device is connected with a large number of disk devices, and the failure rate of the disks increases in proportion to the number of the disk devices, which poses a problem in terms of high reliability. Therefore, a group is made up of several data disk devices and one parity disk device. Further, data is divided into data units called stripes, and all stripes starting from the same logical address of all disk devices in the group form a group called a parity group. Parity information is generated by calculating the exclusive OR of all the data stripes in this parity group, and this is stored in the parity disk device as the parity stripe. As described above, by providing the data with redundancy rather than simply distributing the data to a plurality of disk devices, even if one disk device should fail, the parity disk device's parity information can be used to determine the failed disk device. Conventionally, a high reliability method configured so that the data of the above can be restored has been discussed.

Further, in the disk device, a method for controlling reading of file data over a plurality of continuous fixed length blocks at high speed is disclosed in Japanese Patent Laid-Open No. 64-53224. According to this method, when reading file data, it is checked whether there is a continuous block in the block in which this file data is stored,
When there are consecutive blocks, these blocks are read in a batch to enable high-speed data transfer with reduced rotation waiting time of the disk device.

[0007]

The host computer divides the command into a plurality of commands for accessing the continuous area on the logical address to the disk array device.
There are cases where these are continuously transferred and a request is issued. That is, the OS (operating system) of the host computer may handle the data by dividing it into units called pages of a certain fixed size. In this case, the long data access request is issued to the disk array device as a plurality of commands. Also, if the program running on the host computer is
When processing such as generating a number of data files in the same procedure is performed, similar operations occur because these data files are often arranged in continuous areas.

However, in the above-mentioned conventional technique, no consideration is given to the processing at the time of accessing the disk array device to such a continuous area a plurality of times (hereinafter referred to as a plurality of sequential accesses). According to such a conventional technique, even if the above-mentioned sequential continuous access request is issued to the disk array device, since this disk array device processes them as a single command, it is an access to a continuous area. There is a problem that the disk rotation wait occurs for each command and the processing speed is significantly reduced.

Further, in the above conventional technique, it is necessary to generate the parity information as described above, and prior to the data write, the old data and the old parity information already written in the area to be written are read, and New parity information is generated by calculating the exclusive OR with the new data to be written. According to such a conventional technique, the read processing of the old data for generating the parity information is performed at the time of the sequential write access.
Each command is executed one by one, and two disk rotation waits are required for each write command.
There is a problem that the reduction of the processing speed is promoted in addition to the waiting for the rotation for each command.

Further, in the above-mentioned prior art, since continuous data is distributed from a host computer to a plurality of disk devices on a logical address, the data is not arranged continuously on the disk device. Therefore, as a function that a disk device usually has, when issuing a read command, not only the data concerned but also the continuous data following this are prefetched, and when the next command accesses this area, the operation of the mechanical part such as seek and rotation wait There is a function (prefetch function) of immediately transferring this prefetch data without waiting time. However, according to the above-mentioned conventional technique, since the data is not continuously arranged on the disk device, the prefetch function of this disk device is provided. Does not work, and the operation of the mechanism unit waits for each access, which causes a problem that the processing speed at the time of a plurality of sequential accesses becomes lower than that of a single disk device.

Further, in the method of batch reading continuous blocks in the above-mentioned disk device, it is premised that the file is stored in a single disk device, and like the disk array device, the file is divided into a plurality of stripes. No consideration is given to the case of dividing and recording by distributing to multiple disk devices. Also, only the case where the file is divided into fixed lengths is shown, and no consideration is given to the case where the OS of the host computer divides the file into variable length block sizes. Further, only the read operation is shown, and as described above, in the disk array device, no particular consideration is given to speeding up the write operation, which reduces the processing speed.

An object of the present invention is to solve such a problem and to provide a disk array control system which does not generate a disk rotation wait at the time of a plurality of sequential accesses divided into blocks of arbitrary length.

Another object of the present invention is to provide a disk array control system which eliminates the need for a read operation for parity generation during a plurality of sequential accesses for performing the write operation.

Still another object of the present invention is to replace the prefetch function of the above-mentioned disk device with a function to replace the pre-fetch function, so that the data transfer can be performed not only in the read operation but also in the write operation without the operation waiting time of the mechanical portion of the disk device. It is to provide a disk array control method for performing the above.

[0015]

In order to achieve the above-mentioned object, the present invention receives a command from a host computer, notifies the end of the command, and interprets the command issued from the host computer. Means for converting into a command to each disk device, means for activating a disk command to each disk device, and detecting completion of read or write data transfer by the disk command,
In a disk array device control method comprising means for controlling data transfer between the host computer and each of the disk devices, and means for controlling each of these means, a command issued continuously by the host computer is issued. A means for temporarily storing in the queue and managing these, and when a plurality of commands stored in the queue are commands requesting sequential access, these are grouped and regarded as one command, and this grouped one A means for converting a command into a disk command for each disk device, and information about the grouped command and conversion information when converting the grouped command into a command for each disk device are stored and managed. And means are provided.

In order to achieve the above-mentioned other object, the present invention further provides a first control means for controlling such that continuous data following the data handled by the command issued by the host computer is prefetched from the disk device in advance. A data storage means for storing the preread data in the disk array control device or each disk device, and a second control means for controlling input / output of the preread data to the data storage means. Set up.

[0017]

The command queue management means sequentially stores a plurality of sequential access commands issued by the host computer for sequential access. The command interpreting means interprets these, recognizes that they are sequential access requests, and notifies these commands to the command grouping means. This command grouping means regards these command groups as one command and groups them, and stores this correspondence relationship in the conversion information management means. Further, the command grouping means converts this one grouped command into a disk command for each disk device, and stores the conversion correspondence relationship in the conversion information management means.

By the operation of each unit described above, a plurality of sequential access commands are converted into one sequential access command inside the disk array control device, and each data can be continuously and collectively handled. Therefore, there is no need to wait for disk rotation between commands, and the data transfer rate can be increased.

Further, at the time of writing, the command grouping means operates as follows. First, similarly to the above, a plurality of sequential write access commands are grouped. Then, the data area for writing all the data stripes in the parity stripe is cut out. In the areas other than the above data area, as in the conventional case, data read for parity generation is performed, and after parity is generated, write operation is performed.In the above data area, parity information is directly generated from the data stripe and only write operation is performed. Do. Since the grouping is performed as described above and the data is handled as a batch in a continuous manner, the data area is increased, and the read operation for parity generation is unnecessary for writing most of the data. Achieve higher speed.

If the command interpreting means interprets the command issued by the host computer and finds that the disk device storing the continuous data area is not accessed in the data area handled by this command, the data prefetching is carried out. The control means receives this and issues a read request to the disk device. This disk device does not have the data requested by the host computer, but responds to the read request issued by this data prefetch control means, and stores the next data following the data requested by the host computer in the prefetch data storage means. Send to At this time, the means for controlling the data transfer inputs this read data to the prefetch data storage means. Then, when the command issued by the host computer ends, and then the host computer requests continuous sequential data read to the data of the previous access, the data prefetch control means determines that the data previously input to the prefetch data storage means Judge that it is requested and immediately send this data to the host computer. In this way, since the disk device is not actually accessed, the speed of data transfer at the time of reading is realized.

In addition, since the read processing is performed for the parity generation as described above even at the time of writing, the data prefetch control means also operates to perform the prefetch operation in the same manner as described above. When the current write command ends and the next sequential write command is issued from the host computer, the old data required for parity generation is stored in the read-ahead data storage means, so read processing is not performed. Parity data can be generated and data write processing can be performed immediately.
Achieves high-speed data transfer when writing.

[0022]

DESCRIPTION OF THE PREFERRED EMBODIMENTS First, a disk array device using the present invention will be described with reference to the drawings. FIG. 2 is a block diagram showing the configuration of the disk array device. 1 is a host computer to which the disk array device 2 is connected, 3 is a disk array control device for controlling the disk devices 9a1-9en arranged on the array, Reference numeral 4 denotes a host interface (hereinafter, host I / F) that connects the disk array device 2 and the host computer 1.
5) data control means for controlling the flow of data in the disk array device 2, and 6 command control for interpreting a command sent from the host computer 1 and converting it into a command to each disk device 9a1-9en. Command control means 7 for executing the data buffer, 7 a data buffer, and 8a to 8e a disk interface for connecting the disk array control device 3 and the disk devices arranged on the array (hereinafter referred to as disk I
/ F), 10 is one disk device 9a1-9e
It is a parity generation unit that generates parity information provided so as not to lose the data even if any one of the n fails.

In the figure, the disk array controller 3
Then, the read command or the write command from the host computer 1 received by the host I / F 4 is command control means 6
Transferred to. The command control means 6 interprets the received command, selects a disk device to be accessed, and generates a command for the disk device.

Here, in order to avoid confusion in the following description,
A command sent by the host computer 1 is called a host command, and a command unique to the disk device sent to each of the disk devices 9a1-9en is called a disk command.

The disk command generated by the command control means 6 is transferred to the target disk device via the disk I / F to which the target disk device is connected, and the read or write processing is executed. ..
At this time, it is also possible to select a plurality of disk devices and operate them simultaneously. For this reason, it is possible to expect high-performance operation as compared with one disk device.

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a specific configuration of the command control means 6 in FIG. 2 which uses an embodiment of the disk array control system according to the present invention. Reference numeral 11 denotes a host command via a host I / F 4 to a host computer 1. Host command input / output means for receiving and reporting the end of command, 12 for host command queuing means for temporarily storing and managing the received host command, 13 for interpreting the received host command and disk command Command clustering means for generating a command cluster by integrating a plurality of host commands, 15 issuing the generated disk command to each disk device, and command from each disk device. Start disk command to receive the end
Termination means, 16 is main control means for managing and controlling the entire command control means 6, 17 is data transfer timing control means for instructing the data control means 5 (FIG. 2) to an appropriate transfer timing, and 18 is command clustering means 1
4 is a disk command management table for managing the correspondence between the command cluster generated in 4 and the original disk command.

Next, the operation of the disk array device 2 will be described. In FIG. 2, the disk array device 2 appears as one large-capacity disk device to the host computer 1. Now, it is assumed that a command for continuous data access, which is divided into a plurality of times, is sent from the host computer 1 to the disk array device 2. Such an access request occurs in the following cases.

The first case is due to the restrictions of the OS. If you are dealing with long sequential files,
If S does not collectively handle the data length of this file,
Data will be divided and commands will be issued multiple times.

Secondly, it may occur due to the characteristics of the application. When an application regularly creates or reads multiple files, it is likely that they will be placed on consecutive logical addresses.
Multiple sequential accesses will be issued.

The operation when a plurality of such sequential accesses are issued from the host computer 1 to the disk array device 2 will be described with reference to FIG.

FIG. 3 shows a logical address map of the disk array system 2 as seen from the host computer 1 which is a host computer. In FIG. 3, the host computer 1 has a command A (a logical address AhA and a data length LA). Hereafter, simply A
, Logical address AhB, command B of data length LB (hereinafter simply referred to as B), logical address AhC, command C of data length LC (hereinafter simply referred to as C), and logical address AhD, command D of data length LD. It is assumed that four host commands (hereinafter, simply referred to as “D”) are continuously issued to the disk array device 2. The data handled by these host commands A, B, C and D are arranged consecutively on the address map as shown in FIG.

First, the host command A is the upper I / F 4
2 is supplied to the command input / output means 11 inside the command control means 6 shown in FIG. 1, and the command input / output means 11 transfers this to the host command queuing means 12. If the internal command queue is not full, the host command queuing means 12 receives this host command A and stores it at the end of the command queue. If the host command interpreting means 13 is not executing the command, the command One host command is taken out from the head of the queue and sent to the host command interpreting means 13.

Now, the host command queuing means 1
If the host command A is stored at the head of the command queue in 2 and the host command interpreting means 13 is not in execution, this host command A is read from this command queue and transferred to the host command interpreting means 13, The command is interpreted and converted into a disk command. At this time, the host command queuing means 12 also notifies the host computer 1 via the host command input / output means 11 that the host computer 1 can receive the next host command. Along with this, the top I /
F4 becomes ready to receive the next host command,
If the host computer 1 can send the next host command, the host computer 1 sends the host command B, and then the host commands C and D each time the host command queuing means 12 sends a host command. To do. The host command queuing means 12 sequentially stores these host commands in the command queue, and the host command interpreting means 13 stores these host commands B, C, D.
In the same manner as the host command A, each command is interpreted and converted into a disk command.

When a plurality of host commands are successively transmitted from the host computer 1 and stored in the command queue in this way, the command clustering means 14 immediately targets the disk command converted by the host command interpreting means 13. Disk command management table 18
To store. Then, if these continuously received host commands A to D are not continuous data access to continuous logical addresses, the command clustering means 14 executes the disk command in order to execute them sequentially in an appropriate order. The disk commands are fetched from the management table 18 and sent to the disk command start / end means 15 to issue these disk commands to each disk device.

On the other hand, as shown in FIG. 3, if these host commands A to D are accesses to the continuous area, the command clustering means 14 groups these host commands A to D. That is, as shown in FIG. 4, the host command A sent from the host computer 1,
B, C and D are logical address AhA and data length (LA +
This means equivalent conversion into one host command (LB + LC + LD) (this is called command clustering).

Next, the clustered host command is sent to the command interpreting means 13, converted into a disk command, and issued by the disk command starting / terminating means 15 to the target disk device.

FIG. 5 is a diagram showing mapping of each host command to the logical address of each disk device. In addition,
Here, for simplification of explanation, five disk devices a to e are connected in parallel (that is, in FIG. 2, five disk I / Fs 8a to 8e are provided in the disk array control device 3). That is, one disk device is connected to each of them), but of course the number of disk devices connected in parallel is arbitrary, and
The number of disk devices connected to the same disk I / F is also arbitrary.

Of the five disk devices a to e, the disk devices a to d are for reading and writing data, and the disk device e is for reading and writing parity data. ..

Data handled by one host command is divided into unit blocks called stripes shown in FIG. 5 and recorded on the disk. Here, the host command A
Is divided into three stripes A1, A2 and A3, and these stripes A1 to A3 are recorded in the disk devices a, b and c. Similarly, the data handled by the host commands B, C, and D are divided into two, five, and one stripes, respectively, and are recorded in the disk devices a to e as shown in FIG.

Next, the disk array control method according to the present invention used here will be divided into a read operation and a write operation, and the host command shown in FIG. 3 will be described as an example in detail.
First, the read operation will be described. For comparison, first, a case where command clustering is not performed as in the conventional method will be described.

As described above, the read data of the host command A is divided into stripes A1, A2, and A3 and recorded in the disk devices a, b, and c, so that the disk devices a, b, and c respectively record the data. Command Aa, Ab, A
Read by c is performed. Similarly, for host command B, stripes B1 and B2 are from disk devices d and a from disk commands Bd and Ba, and for host command C, stripes C1 and C5 are from disk device b to disk command Cb, and stripe C2, C3 and C4 are C from disk devices c, d and a respectively
For c, Cd, Ca, the host command D
The stripe D1 is read from the disk device C by the disk command Dc.

To explain the above with reference to FIG. 6, when the host commands A, B, C and D are successively supplied from the host computer 1 to the command converting means 6 via the host I / F 4,
In the command converting means 6 shown in FIG. 1, these are temporarily stored in the host command queuing means 12. The host command interpreting means 13 starts with the disk devices a, b,
host command A to disk command A for c
issue a, Ab, Ac, and issue the respective disk devices a, b,
These disk commands are executed on c at approximately the same time.
Upon receipt of the disk command, each of the disk devices a, b, and c interprets the disk command, waits for the seek operation and disk rotation, and then detects the target sector on the disk and reads the target data. In this way, the data A1, A2, A3 read from the disk devices a, b, c are recombined as the read data of the host command A, sent to the host computer 1, and the execution of the host command A ends.

At this time, the command conversion of the host command B has already been executed, and for the empty disk device d, at substantially the same time as the host command A, the disk command B is executed.
However, since the disk device a is executing the host command A, the disk command Ba is issued.
Cannot be issued and it is necessary to wait for the end of this execution. After confirming the end, send a disk command B to the disk device a.
a is issued, but as shown in FIG.
In spite of the fact that the data areas of stripes A1 and B2 are continuous, the start address of stripe B2 has passed during the overhead of command interpretation time, etc. Therefore, there is no need to wait for rotation around disk 1. must not. This also applies to the host commands C and D. Generally, when executing a host command for performing consecutive n sequential reads, a maximum of n disk rotation waits occur, which reduces the processing speed.

Next, the read operation by the disk array control system of the present invention used in FIG. 1 will be described. In FIG. 1, the host commands A, B, C, D are once grouped (command clustering) by the command clustering means 14 in the command control means 6 as described above, and one host is created as shown in FIG. After pretending to be equivalent to a command, the command is converted into a disk command and the disk command is issued to each disk device. This state is shown in FIG.

In the figure, the host commands A, B, C and D are received from the host computer 1, the command control means 6 interprets them and judges that they are sequential accesses to the continuous area, and as described above. The command clustering means 14 implements command clustering. Then, the host command interpreting means 13 interprets the stripe area for each disk device a to d to be read by this one command clustered host command, and from the result, the disk command CCa for each disk device a to d. , CCb, CCc, CCd are generated, and these are issued to the respective disk devices a to d. here,
The disk command CCa is for stripes A1, B2, C4 on the disk device 9a, the disk command CCb is for stripes A2, C1, C5 on the disk device 9b, and the disk command CCc is for stripes A3, C2, D1 on the disk device 9c. The disk command CCd is a disk command for successively reading the stripes B1 and C3 on the disk device 9d, respectively.

These disk commands CCa to CCd are simultaneously issued to the disk devices a to d, and after the seek and the rotation waiting, the respective target data are read. After these data are once transferred to the data buffer 7 (FIG. 2),
Based on the management information stored in the disk command management table 18 (FIG. 1), the data transfer timing means 17 transfers data from the data buffer 7 to stripes A1, A2, ...
..., data is read in the order of C5, D1 and the host computer 1
, And the host commands A, B, C, D are sequentially completed.

By carrying out the command clustering in this way, when n sequential access commands from the host computer 1 are issued to the disk array device 2, the number of rotation waits becomes only one, which is the same as the conventional method. In comparison, the number of times of waiting for rotation of the disk can be reduced by (n-1) times. Usually, the rotation waiting time is much longer than the data transfer time, so the performance improving effect by reducing the number of rotation waiting times is very large.

Next, this effect will be described in more detail with a specific example. Now, it is assumed that the average seek time of each disk device is 15 milliseconds, the average rotation waiting time is 8 milliseconds, the rotation waiting time for one disk rotation is 16 milliseconds, and the data transfer rate of the disk device is 3 Mbytes / second. .. Also,
There are four disk devices used for the read operation. At this time, access requests are issued from the host computer 1 to the continuous area of 64 Kbytes divided into four 16 Kbyte host commands, and each 16 Kbyte is accessed by four units. It is assumed that the data that is evenly distributed and stored in the disk device is read. Furthermore, the overhead time for command interpretation is very small and can be ignored.

Under the above conditions, according to the above conventional method, the processing time T'is T '= (average seek time + average rotation waiting time) + (16 Kbyte data transfer time / 4) × 4
+ (1 rotation rotation waiting time) x 3 = 76.3 (milliseconds). On the other hand, according to the disk array control method of this embodiment, the processing time T is T 2 = (average seek time + average rotation waiting time) + (64 Kbyte data transfer time / 4) =
28.3 (milliseconds), which is about 3 times faster than the conventional method. The larger the number of sequential access commands from the host computer 1 to the continuous area, the greater the effect of this embodiment.

Further, according to this embodiment, FIGS.
As is clear from the comparison between and, the occupied time of the disk device is shorter than that of the conventional method. If the next host command is received while executing these host commands, the conventional method will wait a lot of time until the received command can be executed.
According to this embodiment, since the disk device occupies a short time, the next command can be executed with a short waiting time, the throughput of the process is increased as a whole, and the average command seen from the host computer is increased. This has the effect of shortening the response time.

Next, the operation at the time of writing will be described.
Similar to the case of reading, the host command shown in FIG. 3 is mapped to each disk device and written as shown in FIG. 5, but at the time of writing, it is necessary to generate the above-mentioned parity data, and read and write are performed. The control method is different. A method of creating this parity data will be described with reference to FIG.

In the disk array device of this embodiment, 5
One of the two disk devices is assigned as the parity disk device. As described above, the disk device e is the parity disk device in FIG. Parity is created in a group called a parity group, which has stripe units. One parity group is formed by stripes starting at the same logical address of each disk device a to e, and the exclusive OR of these data is used as parity data. In FIG. 5, for example, stripe A1,
A2, A3, B1 and P1 form one parity group, and a parity data stripe P1 is generated from the exclusive OR of the stripes A1, A2, A3 and B1.

Here again, for comparison, the conventional method will be described first with reference to FIG. The host computer 1, which is the host, sends a write request for continuous data to four host commands A, B,
The commands are issued as C and D, and the command control means 6 of the disk array control device 3 receives and interprets them as in the read operation, converts them into disk commands, and sequentially activates the disk devices. Here, the difference from when reading is
The point is that the read operation is performed to read the old data written in the disk before writing, and the exclusive OR of these and the new data to be written is taken to generate the new parity data.

Specifically, the old data A1 ', A of the stripes A1, A2, A3 corresponding to the host command A
2 ', A3' and old parity stripe P1 'are read, and then these old data and new data stripe A are read.
A new parity data stripe P1 is created by taking the exclusive OR of 1, A2 and A3. Then, after waiting for one rotation of the disk, the new data stripes A1, A2, A3 and the new parity data stripe P1 are written to complete a series of write operations. This process is similarly performed for B, C, and D.

According to this, similarly to the read operation, a rotation wait occurs between the read operation and the write operation for parity generation and between the host command and the execution of the host command, and the processing speed decreases. To do.

Next, the first write method in this embodiment will be described. When the host commands A, B, C, and D are received from the host computer 1, the command control means 6 interprets them as in the case of reading, and judges that these are sequential accesses to the continuous area. , B, C, D are grouped (command clustering) by the command clustering means 14 in the command control means 6, and are made to look like one host command equivalently, as shown in FIG. , Read disk commands CCa1, CCb1, CCc1, CCd1, CCe to the disk devices a to d
1 is generated and issued at the same time. Disk command CCa
1 is the old data stripes A1 ', B of the disk device a
2 ', C4', and the disk command CCb1 is the old data stripes A2 ', C1', C5 'of the disk device b.
The disk command CCc1 is the old data stripes A3 ', C2', D1 'of the disk device c, the disk command CCd1 is the old data stripes B1', C3 'of the disk device d, and the disk command CCe1 is the old data stripe of the disk device e. Parity stripes P1 ', P2', P
Lead 3'and P4 'respectively. Each disk device ae
Then, after the seek operation and the waiting for rotation, the respective target data are read out. And these old data and new data A1, A2, A3, B1, B2, C1, C2, C3, C
4, C5, D1 and new parity data P1, P2, P
3 and P4 are generated. Then, the new data A1 to D1 and the new parity data P1 to P4 are transferred to the read disk commands CCa1, CCb1, CCc1, CCd1, CCe1.
Using the same disk commands CCa2 to CCe2 as the write disk command, writing is performed in the same manner as in reading. This completes the write operation.

At this time, if the parity data generation time is sufficiently smaller than the one-round rotation waiting time of the disk, the halfway rotation waiting time is only once between the read and the write, so that the maximum ( 2n-2) times can be reduced, which is very effective in terms of speeding up.

Next, the second write method of this embodiment will be described. For this reason, the device shown in FIG. 9 is devised. That is, when the host command logical address is mapped as shown in FIG. 5, the upper part (head part) 2 of FIG.
Since one column writes all the data stripes of the parity group, they can generate parity data without reading the old data. Therefore, these two
As for the column, as shown in FIG. 9, all the data disk devices a, b, c, d and the parity disk device e can write data only without reading (disk commands CCa1 to CCe1).

In the third column from the top in FIG. 5, the old data C
4 ', C5', D1 'are disc commands CCa2, C
Read with Cb2, CCc2, CCe2, generate parity with these and new data, and then write disk command CC
New data is written by a3, CCb3, CCc3, CCe3.

According to the second method, the waiting time for rotation is increased by 1 to 3 times as compared with the first method. However, when handling data that requires a data transfer time that is sufficiently longer than the rotation waiting time, the read operation time for parity generation can be short, and the speed is sufficiently higher than that of the first method. Also, waiting for rotation (2
n-5) times can be reduced.

Similar to the read operation, the effect of the write operation will be specifically described with a specific example. here,
A write access request is issued from the host computer 1 to a continuous area of 64 Kbytes divided into four 16 Kbyte commands, and the data of each 16 Kbyte command is evenly distributed and written to the four disk devices. Also,
The stripe size is 4 Kbytes, and the overhead time for command interpretation is very small and can be ignored.

Therefore, according to the conventional method, the processing time T '
At worst, when writing after reading, T '=
(Average seek time + average rotation wait time) + (16 Kbyte data transfer time ÷ 4) x 4 x 2 + (1 round rotation wait time) x (total access count -1) = 145.7 (milliseconds) .. On the other hand, according to the first method of this embodiment, the processing time T1 is T1 = (average seek time + average rotation waiting time) + (64 Kbyte data transfer time / 4) × 2 +
(1 rotation rotation waiting time) = 49.6 (milliseconds), which is about 3 times faster than the conventional method.

When the access request is issued from the high-level computer 1 to the continuous area of 4 Mbytes divided into four 1 Mbyte commands, the processing time T'in the conventional method is used.
At worst, when writing after reading, T '=
(Average seek time + Average rotation wait time) + (1 Mbyte data transfer time / 4) × 4 × 2 + (1 rotation rotation wait time) × (total access count-1) = 802 (milliseconds). On the other hand, according to the first method in this embodiment, the processing time T1 is T1 = (average seek time + average rotation waiting time) + (4 Mbyte data transfer time / 4) × 2 +
(1 rotation rotation waiting time) = 706 (milliseconds), which is only about 1.1 times faster than the conventional method.

On the other hand, according to the second method of this embodiment, assuming that the processing time T2 is the worst waiting time for 5 times of rotation, and the area for performing the write operation after read is 16 Kbytes, T2 = (Average seek time + Average rotation waiting time) + ((4
Data transfer time of M bytes-16 Kbytes / 4) +
(16K-byte data transfer time) x 2 + (1 rotation rotation waiting time) x 4 = 438 (milliseconds), which is about twice as fast as the conventional method, and the effect is very large. ..

When the data is written in or read from the disk devices 9a to 9e after the above read or write operation is completed, the disk devices 9a to 9e in FIG. 1 to the disk command start / end means 15 of the command control means 6 shown in FIG. 1, and the disk command start / end means 15 notifies the command clustering means 14 of this. Then, when the above operation is a read operation, the data transfer timing means 17 operates as described above.
The data corresponding to the previous host command stored in the data buffer 7 (FIG. 2) is recombined and the host computer 1
(Fig. 2). Then, the command clustering means 14 deletes the management information corresponding to these commands from the disk command management table 18, and also notifies the host command queuing means 12 of this so as to delete the corresponding host command from the command queue. ,
In addition, the host command input / output unit 11 is notified. Upon receiving this notification, the host command input / output unit 11 notifies the host computer 1 of the command end via the host I / F 4 and ends the series of processes.

As described above, the effects of this embodiment can be summarized as follows: (1) At the time of sequential read, it is possible to reduce the maximum waiting time (n-1) of rotations. As a result, in the case of data read of 64 Kbytes in which a 16 Kbit host command is sequentially issued four times, for example, the total processing time is about 3 times faster than that of the conventional method.

(2) At the time of sequential writing of small data, the maximum (2n-2) is obtained according to the first method.
It is possible to reduce the number of times to wait for rotation. With this, for example, 16K
In the case of 64K-byte data write in which byte host commands are sequentially issued four times, the total processing time is about three times faster than that of the conventional method.

(3) At the time of sequential writing of large data, according to the second method, the maximum (2n-5) is obtained.
It is possible to reduce the number of times to wait for rotation, and further, it is possible to delete most of the old data read operations for parity generation. As a result, in the case of a 4 Mbyte data write in which a 1 Mbyte host command is sequentially issued four times, for example, the total processing time is at least twice as fast as the conventional method.

FIG. 10 is a block diagram showing the internal structure of the command control means 6 of the disk array device 2 of FIG. 2 which uses the disk array control system according to the present invention. The command control means 6 shown in FIG. Control means 19
Is provided. In the following description, it is assumed that the command clustering unit 14 and the host command queuing unit 12 described in FIG. 1 are not provided, but these may be used as shown in FIG.

In FIG. 10, prefetch control means 1
Reference numeral 9 is a means for determining whether or not a prefetch described later has hit, and if hit, controls to read data from a prefetch buffer described later.
The other components are the same as those of the command control means 6 shown in FIG.

FIG. 11 is a block diagram showing a main part of the disk array device 2 shown in FIG. 2 using the command control means 6 shown in FIG.
9 is prefetch control means inside the command control means 6 shown in FIG. 10, and 20a to 20e are disk devices 9a, 9
b ... A prefetch buffer for inputting and storing the data stored above in accordance with a command from the prefetch control means 19, and the portions corresponding to those in FIG.

Next, the operation of this embodiment will be described. It is assumed that the host computer 1 reads the data in the continuous area on the logical address map of the disk array device 2 by a plurality of host commands. Think The case where such sequential access occurs a plurality of times is the same as in the first embodiment described above. In such a case, assuming that one disk device is connected to the host computer 1, if the data is stored in a prefetch data storage area called a prefetch buffer of the disk device, the data is stored from this storage area. By reading the data, there is no delay depending on the mechanism of the disk device, such as seek and rotation wait, and since it is a read from an electrical storage area such as a memory, it is possible to access at very high speed. .. Therefore, in a disk device, when a read request is issued from a high-order computer, prefetch control is performed in which not only the data but also the data following this data is preread and stored in the prefetch buffer.

However, in the disk array device 2, since a plurality of disk devices are arranged in parallel and data is recorded in different disk devices for each stripe, even if the logical addresses seen from the host side are continuous, each disk device is Data is discontinuous on the logical address of the device. Therefore, the prefetch function of the disk device cannot be expected except in the case of large data access exceeding the data size of the parity group. The second embodiment enables a speed-up method that compensates for the drawbacks of this disk array device.

Next, the operation of this embodiment will be described. Here, as an example, it is assumed that two sequential access requests A and B are issued from the host computer 1. And
As shown in FIG. 12, A1 and A2 are stripes of disk devices a and b accessed by the host command A, B1 and
B2 is a stripe of the disk devices c and d accessed by the host command B, and P is a parity stripe in the disk device e of the parity group including A1, A2, B1 and B2.

First, the write operation will be described. For comparison, first, the conventional method will be described with reference to FIG. 13 as in the first embodiment.
Will be explained. First, the host command A is executed. That is, first, the old data A1 ', A2', P'is read from the disk devices a, b, e. Then, new parity data P is generated by calculation of an exclusive logical wheel of these and new data, new data A1 and A2 are written to the disk devices a and b, and new parity data P is written to the disk device e, respectively. The command end is reported to the computer 1, and the host command A ends. The host command B is similarly executed, the end of the command B is reported to the host computer 1, and these two sequential write commands are ended.

Next, the write operation according to the method of this embodiment will be described with reference to FIGS. 2, 10, 11 and 14. First, the host command A is the command control means 6
The host command interpreting means 13 determines that the access request is for the disk devices 9a, 9b, 9e, generates disk commands for these disk devices 9a, 9b, 9e, and starts / ends the disk command. To the means 15. The disk command activation / termination means 15 activates these disk commands.

At the same time, the host command interpreting means 1
3 judges that the disk devices 9c and 9d are unused, and notifies the prefetch control means 19 of this.
Upon receiving this notification, the prefetch control means 19 generates a disk command to read the data in the logical address area following the access area of the host command A of the disk devices 9c and 9d, and sends it to the disk command start / end means 15. To do. Also, this prefetch control means 1
A request for read to the disk devices 9c and 9d is stored in the storage unit 9 inside. Therefore, the disk command start / end means 15 issues a read disk command to the disk devices 9c and 9d.

As in the conventional method, the host command A is sent from the disk devices a, b, e to the old data A1 ', A2',
The new parity data P is generated by reading P ′ and taking the exclusive OR with the new data, and the new data A1,
The data is written to the disk devices 9a, 9b, 9e together with A2, and this is reported to the host computer 1. This completes all processing for the host command A.

At the same time, the prefetch control means 19
Disk device 9c generated from the host command B by
The read disk command to 9d is executed, and the old data B1 'and B2' are transferred from them, but the prefetch control means 19 uses the prefetch buffers 20c and 2c.
0d is set in a data input enabled state, and the read old data B1 ′ and B2 ′ are stored here. At this time, the old data B1 'and B2' are not transferred to the data transfer means 5'and the data buffer 7. When the transfer of the old data to the prefetch buffers 20c and 20d is completed, the prefetch control means 19 records this state in the internal storage area. Also, this prefetch control means 1
Reference numeral 9 is write data A1 and A for the disk devices 9a and 9b.
Also for 2 and P, when they are transferred from the data buffer 7 to these disk devices 9a, 9b and 9e, they are controlled to be simultaneously stored in the prefetch buffers 20a, 20b and 20e.

The operation of reading the data area following the current access area and storing it in the prefetch buffer,
Saving the write data to the disk device in the prefetch buffer is generically called data prefetch.

When the host computer 1 receives the end report of the host command A, the next host command B is issued to the disk array device 2. In the command control means 6, when this is received, the host command interpreting means 13 interprets this host command B, and disk devices 9c, 9d, 9
The disk command of e is generated and sent to the disk command starting / terminating means 15.

Further, the host command interpreting means 13 indicates that the disk devices 9c, 9d and 9e are to be accessed.
Further, the disk devices 9a and 9b determine that they are not currently used and notify the prefetch control means 19 of them. The prefetch control means 19 uses the old data B1 ', B2', which is used for generating the parity at the time of this write.
It is determined whether P'is stored in the prefetch buffers 20c, 20d, 20e (this is referred to as prefetch hit determination). In this case, since the old data has been prefetched into the prefetch buffers 20c, 20d, 20e first, this is notified to the disk command activation / termination means 15 and the data transfer timing control means 17 as a prefetch hit.

Therefore, the disk command starting / terminating means 15 judges that reading to the disk devices 9c, 9d, 9e is unnecessary, and issues a write disk command to the disk devices 9c, 9d, 9e at an appropriate timing. The data transfer timing control means 17 transfers the old data B1 ', B2', P'from the prefetch buffers 20c, 20d, 20e to the data transfer means 5 and the new data B1, B2 in the data buffer 7 to the parity generation means 10 (FIG. 2). To the new parity P.

After the completion of the generation of this new parity P, the command control means 6 receives this end notification and issues a write disk command to the disk devices 9c, 9d, 9e to write the data B1, B2 and the parity P to them. .. When this write operation is completed and this is reported to the host computer 1, the write processing is completed. FIG. 14 shows the above situation.

Further, the prefetch control means 19 does not execute the disk devices 9a and 9b while the disk devices 9c, 9d and 9e are performing the write operation, so that the next continuous data is prefetched into the prefetch buffers 20a and 20b.
In order to prepare for the next sequential access, a read disk command is generated in order to read the data and is issued to the disk command start / end means 15. Furthermore, the data currently written is stored in the prefetch buffers 20c, 2c.
Save in 0d, 20e.

As described above, when the prefetch hits, the read operation for generating the parity at the time of writing becomes unnecessary, and the processing speed is doubled as compared with the conventional one.

Further, although the write operation has been described, it goes without saying that the same operation also applies to the sequential read operation. In this case, if all the data is stored in the prefetch buffer, the data read process can be executed without accessing the disk device at all, so that a very high speed access without seek time and rotation waiting time is realized.

Further, prefetch read is performed while predicting that the next continuous access will come during execution of a certain command, but if another access to this disk device comes during prefetch read, the prefetch operation is performed. Disk command activation / termination means 15 so as to immediately interrupt
Should be configured. At this time, the prefetched data becomes invalid, and the access requested by the host computer 1 is immediately executed (this is called prefetch abort).

Further, in this embodiment, the prefetch buffer 20 is provided between the data transfer means 5'and the disk I / F 8 as shown in FIG. 11, but it is provided logically on the data buffer 7. It doesn't matter.

Furthermore, in this embodiment, the prefetch buffer 20 is provided on the disk array control device 3, but it may be provided inside the disk device. FIG. 15 is a block diagram showing a specific configuration of the disk device in this case, in which 21 is an upper I / F, 22 is a data buffer, and 23
Is a data control means, 24 is a command control means, and 25 is a drive device.

Although the data buffer 22 is used here, a prefetch buffer may be provided inside the disk device 9 separately from the data buffer 22.

As described above, the prefetch control means 19
A read command for prefetch is generated by (FIG. 10), and the disk command activation / termination means 15 issues the read command to the disk device 9 via the disk I / F 8. This read command is received by the host I / F 21, interpreted by the command control means 24, and the data control means 23 reads the corresponding data from the drive device 25 and stores it in the data buffer 22. Then, the data control means 23 reads this data from the data buffer 22, and the upper I / F 21 and the disk I / F 8
The data is transferred to the disk array control device 3 (FIG. 2) via the disk array control device 3 and the data control means 5 of the disk array transfer device 3 may read and discard this data.

By this operation, the data is pre-fetched in the data buffer 22 inside the disk device 9, and the next time this data is accessed, this pre-fetch hits and seek operation, without waiting for rotation, the disk Data can be read from the device 9.

In this case, since the disk device 9 is accessed once, there is an overhead of the disk I / F 8. Therefore, the speed is slightly slower than that of the system in which the prefetch buffer is formed inside the disk array control device 3 described above. However, since there is no seek and rotation waiting, it is sufficiently faster than the conventional method, and further, the prefetch buffer is unnecessary compared to the method in which the prefetch buffer is configured inside the disk array control device 3 described above. Can be realized at a low price. Further, it is not necessary to perform complicated prefetch processing inside the disk array control device 3, and the control becomes simple. Further, in FIG. 15, a prefetch command may be newly set instead of the method of reading and discarding the read data described above. When the disk command activation / termination means 15 sends a prefetch command to the disk device 9, the command control means 24 judges this, and the data control means 23 transfers the necessary data to the data buffer 22 and ends the processing. By setting such a command, unnecessary data transfer is not performed, thereby reducing the traffic between the disk I / F 8 and the upper I / F 21. Therefore, throughput can be increased in a configuration in which a large number of disk devices are connected to the same disk I / F 8.

As described above, according to the above-mentioned two methods in which the prefetch buffer is provided inside the disk device 9, the above-mentioned effects can be obtained, and further the following effects can be obtained. That is, when a large number of disk devices are connected to one disk I / F 8, prefetching can be performed in each disk device, so that, for example, a plurality of sequential access requests are issued from a plurality of host computers 1. When these are operated in multiple, if they are accesses to different disk devices, the prefetch can be hit for any sequential access request from the host computer 1.

The effects of the above embodiments can be summarized as follows.

(1) When the prefetch buffer 20 is provided in the disk array device 3, if the prefetch hits at the time of writing, the read operation of the old data for parity generation becomes unnecessary and the processing speed is higher than that of the conventional method. About twice as fast.

(2) When the prefetch buffer 20 is provided in the disk array device 3, if the prefetch hits at the time of reading, access to the disk device 9 becomes unnecessary, and the read speed from the prefetch buffer 20 to the host computer 1 is read. The operation can be realized, and extremely high-speed processing can be realized as compared with the conventional method.

(3) When a prefetch buffer (data buffer 22) is provided in the disk device 9 and a false read command is issued from the disk array control device 3 to prefetch data into this prefetch buffer, the disk array control device A prefetch buffer is not required in the device 3, the cost of the disk array control device 3 can be easily controlled, and the prefetch operation having substantially the same performance as the above (1) and (2) can be realized.

(4) By providing the prefetch command as the disk command, the data can be prefetched into the buffer in the disk device, so that it is not necessary to transfer the data between the disk device and the disk array controller. It is possible to reduce the traffic of the disk I / F.

(5) In the cases (3) and (4) in which the prefetch buffer (data buffer 22) is provided in the disk device 9 to perform prefetch control, a plurality of sequential data access commands are issued from a plurality of host computers 1 at substantially the same time. Are issued, and if they are operated in multiple, if these are accesses to different disk devices,
For either access request from the host computer 1,
Prefetch can be hit.

Incidentally, each component of the embodiment described above is
It may be configured by either hardware or software.

The two embodiments of the command clustering control method and the sequential data prefetch control method have been described above, but both of these control methods can be performed simultaneously. At this time, the effects of both can be expected, and the effect of speeding up data transfer is great.

In the above description, the disk array device is a system in which the parity information is stored in a single disk device (parity disk), but the parity information is distributed to a plurality of disk devices or all disk devices. The same data control method as described above can also be adopted in the disk array device of the storage method, and the same effect can be obtained.

Further, although the command clustering control method has been described by taking the disk array control device as an example, it is adopted in all recording / reproducing devices such as a magnetic disk control device, a magnetic tape control device, an optical disk control device, and a semiconductor disk control device. can do. At this time, an arbitrary data length can be handled, and the same effect as described above can be obtained regardless of read or write.

Furthermore, the prefetch control system can be applied to all recording / reproducing devices having a large number of recording / reproducing media such as an optical disk library device and a magnetic tape library device, and the same effect as described above can be obtained. ..

[0107]

As described above, according to the present invention,
During sequential read in which n commands are continuously issued, the rotation wait does not occur between commands, so the maximum (n-1) rotation wait time can be reduced compared to the conventional method, and read It is very effective in speeding up data transfer. Also, during sequential write in which n commands that handle small data are issued consecutively,
Since the rotation wait does not occur between commands, the maximum (2n−2) rotation wait can be reduced as compared with the conventional method, and the effect of speeding up the write data transfer is great.

Further, according to the present invention, at the time of sequential write in which n commands dealing with large data are continuously issued, the rotation wait does not occur between the commands and the parity data is generated. Since it is possible to significantly reduce the number of read times, it is possible to reduce the waiting time for rotation (2n-5) at the maximum as compared with the conventional method, and it is effective in increasing the speed of write data transfer.

Further, according to the present invention, by providing the prefetch buffer in the disk array device, when the prefetch hits at the time of writing, the read operation for parity generation becomes unnecessary, and the processing speed is higher than that of the conventional system. This has the effect of increasing the speed by about twice. Also,
When a prefetch buffer is provided in the disk array device, if the prefetch hits during reading, access to the disk device is not required, and the read operation can be realized at the transfer speed from the prefetch buffer to the host computer. The effect is that high-speed processing can be realized.

Furthermore, according to the present invention, when a prefetch buffer is provided in the disk device (or when it is realized by a data buffer), the prefetch buffer is not required in the disk array control device, and the cost is low. The control of the disk array control device can easily realize the prefetch operation, and the performance is equivalent. In this case, if multiple sequential data access commands are issued from multiple higher-level computers at approximately the same time and they are operated in multiple, if these are access to different disk devices, which higher-level computer will be Since the prefetch can be hit even for the access request of, and the above effect can be expected, high throughput processing can be realized.

[Brief description of drawings]

FIG. 1 is a block diagram showing a first embodiment of a disk array control system according to the present invention.

FIG. 2 is a block diagram showing an example of a disk array device using a disk array control system according to the present invention.

FIG. 3 is a diagram showing a logical address map of the disk array device as seen from the host.

FIG. 4 is a diagram showing a logical address map of the disk array device after command clustering in the first embodiment shown in FIG. 1;

5 is a diagram showing a logical address map of each disk device in the first embodiment shown in FIG. 1. FIG.

FIG. 6 is a timing chart showing a multiple read operation according to a conventional method.

FIG. 7 is a timing chart showing a multiple read operation according to the first embodiment shown in FIG.

FIG. 8 is a timing chart showing another example of the multiple write operation according to the conventional method.

9 is a timing chart showing another example of the multiple read operation according to the first embodiment shown in FIG.

FIG. 10 shows a second disk array control method according to the present invention.
It is a block diagram showing an example of.

FIG. 11 is a block diagram showing an arrangement of prefetch buffers in the disk array device using the second embodiment shown in FIG.

12 is a diagram showing a logical address map of each disk device according to the second embodiment shown in FIG.

FIG. 13 is a timing chart showing a write operation according to a conventional method.

FIG. 14 is a timing chart showing a write operation according to the second embodiment shown in FIG.

15 is a block diagram showing an example of the internal structure of the disk device according to the second embodiment shown in FIG.

FIG. 16 is a block diagram showing an example of a conventional disk array control method.

[Explanation of symbols]

 1 host computer 2 disk array device 3 disk array control device 4 host I / F 5 data control means 6 command control means 7 data buffer 8a-8e disk I / F 9a-9e disk device 10 parity generation means 11 host command input / output means 12 host command queuing means 13 host command interpreting means 14 command clustering means 15 disk command starting / terminating means 16 main control means 17 data transfer timing control means 18 disk command management table 19 prefetch control means 20a to 20e prefetch buffer

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Shoichi Miyazawa, 292 Yoshida-cho, Totsuka-ku, Yokohama-shi, Kanagawa Inside the Microelectronics Equipment Development Laboratory, Hitachi, Ltd. (72) Inventor Taka Oeda Totsuka, Yokohama-shi, Kanagawa 292 Yoshida-cho, Ward, within Hitachi, Ltd. Microelectronics Device Development Laboratory (72) Inventor Seiji Honda, 292, Yoshida-cho, Totsuka-ku, Yokohama, Kanagawa Prefecture Within Hitachi, Ltd. Microelectronics Device Development Laboratory

Claims (12)

[Claims] 1. A means for receiving a command from a host computer and notifying the end of the command, a means for interpreting a command issued by the host computer and converting it into a command for each disk device, and each disk. A means for activating a command to the device and detecting the end of read and write data transfer by the command, a means for controlling data transfer between the host computer and each disk device, and each of these means. In the control method of the disk array control device comprising control means, there are provided means for temporarily storing commands issued by the host computer in a queue, managing the commands, and a plurality of commands stored in the queue. When the command requires access to one continuous area, these plural commands are grouped into one command, Means for converting one looped command into a command for each of the disk devices, conversion information when converting one grouped command into a command for each of the disk devices, and one of the grouped commands By providing a means for storing and managing information data for the command, and by considering a plurality of commands issued by the host computer to the continuous area as one sequential access command inside the disk array control device, The data handled by the command is treated as one continuous data,
A disk array control system characterized in that the entire data is controlled so that a command is distributed and issued to each of a plurality of disk devices for access. 2. The plurality of read commands to the continuous area issued by the host computer according to claim 1, regarded as one read command inside the disk array control device, and the plurality of read commands based on the one read command. The disk array control method is characterized in that all the data handled by the plurality of read commands to the continuous area is collectively read by distributing and issuing the read command to each of the disk devices. 3. The write command according to claim 1, wherein a plurality of write commands issued by the host computer to the continuous area are regarded as one write command inside the disk array control device, and first, based on the one write command, By distributing and issuing a read command to each of the plurality of disk devices, the data and parity information already recorded in the area from which all the data handled by the plurality of write commands is written Are collectively read, new parity information is generated from the read data, the parity information, and the data to be written, and then a write command is issued to each of the disk devices based on the one write command. All of the data to be distributed, issued, and written and the new parity information generated are generated. The disk array control method and controls to write and Batch. 4. The write command according to claim 1, wherein a plurality of write commands issued by the host computer to the continuous area are regarded as one write command inside the disk array control device, and data of the write command is to be written. Cut out a data area from which data is written to all stripes starting from the same logical address in all disk devices, and write the data area to the data area without reading the old data from the data area. By generating parity data from the data and writing the data and the parity data together, it is possible to control the data area so that reading of old data for generating the parity data is eliminated. Characteristic disk array control method. 5. The disk array according to claim 1, further comprising: first control means for controlling so that continuous data following the data handled by the command issued by the host computer is prefetched from the disk device in advance. Data storage means for storing the data prefetched by the first control means inside the control device or each of the disk devices, and second control for controlling input / output of the prefetched data to the data storage means A disk array control system characterized by comprising means and. 6. The determination unit according to claim 5, further comprising: a determination unit that determines whether or not the data handled by the read command issued from the host computer is stored in the data storage unit, and the determination unit stores the data. When it is determined that the data handled by the read command is stored in the means,
A disk array control system characterized in that the data is read from the data storage means and sent to the host computer. 7. The disk array control system according to claim 5, wherein the parity information generated for the write command issued from the host computer is stored in the data storage means. 8. The method according to claim 5, wherein the old data and old parity information already recorded in the area where the new data handled by the write command issued from the host computer should be written are stored in the data storage means. Determining means for determining whether the old data and the old parity information are stored in the data storing means when the old data and the old parity information are stored in the data storing means. A disk characterized by generating new parity information from the new data and the old read data and old parity information, and controlling the new data and the generated new parity information to be written to a disk device. Array control method. 9. The read command according to claim 5, 6 or 8, wherein a read command is issued to a disk device that has not been an access target for storing data in an area subsequent to a data area requested by the host computer, A disk array control system characterized in that the read data thus obtained is stored in the data storage means, and the data in the area following the data area of the access request from the host computer is preread. 10. The data read from the disk device according to claim 9, the data is read out and discarded, the data is transferred to a data buffer inside the disk device in advance, and a read access request to the area is made next time. In the disk array control method, the data is read from the data buffer. 11. The disk device according to claim 5, 6 or 8, wherein a disk device which is not an access target for storing data in an area subsequent to a data area requested by the host computer is A disk array control method comprising a data read-ahead command for controlling to read data to a certain data buffer, and pre-reads data in an area following a data area requested by an access from the host computer. 12. The read-ahead method according to claim 9, 10 or 11, when a request is made to access the disk device from the host computer during read command processing for the read-ahead to the disk device. A disk array control method characterized by controlling so as to stop the read processing.