JPH0496818A - Collective disk device - Google Patents

Abstract

PURPOSE:To increase the transfer speed to improve the processing efficiency by writing matrix data corresponding to a disk drive in parallel and reading out data from the disk drive with one block as the unit. CONSTITUTION:Write data transferred from a CPU 1 at a high speed in serial is distributed as matrix data 10 corresponding to an ECC disk group 14 in a data buffer 3. Data 9 and data 10 consists of many blocks. After data 9 is expanded to data 10, it is collectively written on data disks 4 in a group 14 from the buffer 3. At the time of write to disks 4, blocks are transferred to an ECC generator 6 also, and ECC is generated by blocks and is recorded on an ECC disk 5. At the time of read, only disks 4 where data is stored are accessed to read out data. Thus, the speed of transfer from the CPU is increased to perform the processing at a high speed.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a computer system 11, and particularly to a disk file system that enables input/output operations with even higher performance than 1".

[Conventional technology]

In current computer systems, data required by higher-level devices such as CPt is stored in secondary storage devices.
Data is written to and read from the secondary storage device as required by the CPU. As this secondary storage device, a nonvolatile storage medium is generally used, and typical examples include a magnetic disk and an optical disk. When a database of patents, academic literature, etc. is used by a normal user, the data is only read from the secondary storage device, and the user does not directly rewrite the data. When changing or adding data, the person managing the database adds the data all at once.

Conventionally, a large number of relatively small-capacity magnetic disks are prepared, and the data transferred from the CPU is divided and stored simultaneously on multiple magnetic disks.When reading data, data is read from each magnetic disk at the same time, and the data is transferred to the CPU. There is a technology that performs parallel processing to transfer data to Products of such systems have been announced by companies such as Stridge Concepts. Furthermore, at the Information Processing Society of Japan's 39th Annual National Conference, Norio Sakurai et al. discussed processing data in parallel or processing it as serial data in a single study in ``A study on the configuration method of array control disks''. We report on a method for dynamically switching within the system.

[Problem to be solved by the invention]

In databases of patents, academic literature, etc., when used by normal users, only individual data are read from the secondary storage device, and five users do not directly rewrite the data. When changing or adding data, the person managing the database adds the data all at once. This additional work is performed on large amounts of serial data sent from the CPU in detail and sequentially for individual access units that make up the blocks.
Writing takes a lot of time. In addition, since the process is performed on each individual piece of data that is serially transferred as described above, the transfer speed of the CPU, which is the host device, is determined by the processing time of the magnetic disk. It is not possible to increase the transfer speed for faster processing. As a way to solve the above problems, the disk drive configuration shown in Figure 1 is used to
In Japanese Patent Application Laid-Open No. 1999-1, one data transferred from No. 1 is divided into n pieces and written to data disks #1 to #n in parallel at once, and when reading data, the data is read from data disks #1 to n at once. -250128. However, this method of writing in a -batch and reading in a -batch is good when handling a large amount of data at once, but when reading a small amount of data, data disk #1
Since all of #n is used, the processing efficiency of the system is poor.

Measures to solve problem C] To solve the above problem, when writing data in bulk, the unit of access when reading data is a large amount of data that is serially transferred from the CPU, which is the primary device. Based on the block, it is converted into a matrix data structure, and the data in the matrix data corresponding to each data storage disk drive is written in parallel. Furthermore, when reading data, it is possible to solve the problem by reading blocks from individual disk drives in units of blocks.

[Effect]

By creating matrix data and performing parallel write processing as in the present invention, the transfer speed of the system is lowered even though the transfer speed of each individual disk drive is the same. Furthermore, since the processing time on the magnetic disk is shortened, the transfer speed from the CP, which is a host device, can be increased. On the other hand, during reading, it becomes possible to access individual disk drives, thereby improving processing efficiency. [Embodiment] Embodiment 1 An embodiment of the present invention will be described below with reference to FIG. 1.

This example is cr-] 1. Data control unit (T:D
CU ) 2, data buffer (hereinafter referred to as buffer) 3. Data magnetic disk 4 (hereinafter referred to as data disk), ECC
Mi disk (hereinafter referred to as ECC disk) 5. ECC generator 6 + Read data buffer (hereinafter referred to as Read buffer) 7. It is composed of a data restoration section 8. ■/○ requests issued by CPU1 are II) CU2
issued to each magnetic disk through.

First, the general operation of the specific ■/○ processing in this embodiment will be briefly explained using FIGS. 1 and 2. In this embodiment, a large amount of data is stored or rewritten at once in units of groups for which ECC is created (this is referred to as an ECC disk group 14). The write data 9, which is serially transferred at a higher speed than the CPU, is stored in the data buffer 3 under the control of the DCU 2 shown in FIG.
The data is distributed as matrix data 10 corresponding to the CC disk group 14. Note that the write data 9 and the matrix data 10 are composed of a large number of blocks. This unit of block is one access unit from the upper level. In other words, when a Read request to read data is issued from the host (CPUI) to the data disk 4, a Re
The unit of ad is one block. In this way, after developing the write data 9 into matrix data 10, each block is transferred from the buffer 3 to the corresponding data disk group 1.
data is written to each data disk 4 in 4 at once.

Furthermore, when writing to the data disk 4, each block is simultaneously transferred to the ECC generator 6, and ECC is created from each block and recorded on the ECC disk 5 in the same manner. When reading, only the data disk 4 in which the data is stored in the ECC disk group 14 is accessed and the data is read. In this embodiment, Ilo is processed as described above.

Therefore, the work/○ processing operation will be explained in detail below.

As described above, in this embodiment, since data is written in batches, it is necessary to convert the write data 9 transferred from the CPU 1 as serial data into matrix data 10. Therefore, first, this conversion method will be explained with reference to FIG.

In this embodiment, each data disk 4 and ECC disk 5
The track capacity (capacity of data that can be stored in one round of the magnetic disk) is 40 KB, the amount of data in each block is 4 KB, and the number of data disks 4 is four.

As shown in FIG. 3, the write data from the CPU 1 is serially sent in large quantities. It is preferable that the amount of the write data 9 sent in bulk is always constant. The capacity of this write data 9 is D in advance.
Registered with CTJ2. Next, a large amount of write data 9 has already been registered in advance (1 at the time of reading).
n
It is divided into several areas. Next, each area is divided into several blocks on the data disk 4 in units of blocks. This four-part data is a collection of blocks to be stored in each data disk 4, and is stored in each data disk 4 in the data buffer 3 under the control of the DCU 2 as 0 or more blocks called CoJ, umn data. Matrix data 1 is a collection of blocks (Co]umn data) created corresponding to each data disk 4.
Set to 0. In this embodiment, when writing, the four data disks 4 in the ECC disk group 14 are considered as one logical volume. In other words, the logical track capacity is the physical track capacity (track capacity of each data disk 4 and ECC disk 5) of the data disk 4.
will be multiplied by the number of units.

Next, a method for writing the matrix data 10 created in this way into brackets will be explained using FIG. 4. Figure 4 shows data disk 4 and FCC disk 5.
The internal configuration of is shown. The data disk 4 and the ECC disk 5 are composed of ``) disks, an actuator 11 for moving the head, and .

Each head is attached to one actuator 11, and when the actuator 11 moves, all the heads move in the same way. In other words, the movement of the actuator ]l determines the drag on the r disks corresponding to the r1 nodes. This set of tracks is called a cylinder. In a magnetic disk, tracks are set concentrically on the disk, so addresses are assigned to the tracks from the outer circumference to the inner circumference. This address is called a cylinder number. Therefore, the R/W circuit 12
selects the head, determines the track, and reads and writes data. The capacity of one round of each disk is the physical track capacity. Data disks 4 and E with such a configuration
The CC disk 5 constitutes an ECC disk group 14 . First, when writing area 1, matrix data 10 corresponding to area 1 as shown in FIG. 3 is created, and each column data is written to the data disk 4. As shown in FIG. 4, head #1 is selected on each data disk 4 to write the data in the matrix data 10 of area 1 in FIG. 3 into the shaded area of disk #l. Note that if heads #1 and #2 can be read/written at the same time, they are performed at the same time. If writing to disk #1 is completed but writing of data in area 1 is not yet completed, head #2 is selected by electrical switching and writes to the diagonally shaded portion of disk #2 in the same way as disk #1. Disk #3 below. #4...Write in the same way as #n. In this way, areas 2, 3, 4.5...
...Write in the same way as n. If a track on the disk surface cannot be used due to a defect on the disk surface, the head is currently moved to a spare track that is ready for use. The present invention ignores such switching to the spare track, skips disk #2, and writes to disk #3. In this way, each area is written to the cylinder, and even if one of the data disks 4 in the ECC disk group 14 has a disk #
When n is finished writing, writing to that cylinder is finished, and writing is moved to another cylinder.

On the other hand, at the same time as writing data to each data disk 4 in this way, the data is also transferred to the ECC generator 6,
Create an ECC and write it to the ECC disk 5. Therefore, the ECC creation method will be explained below. The ECC creation method converts the write data 9 into matrix data 10 and develops it into column data corresponding to each data disk 4 as shown in FIG. Therefore, regarding the matrix data 10, the parity creation group 13 (Ro
ν data) by the ECC generator 6.
(e.g. odd parity).

By creating ECC for the matrix data 10 during batch writing in this way, if a failure occurs in one data disk 4, the corresponding Column
If the data becomes unreadable, the remaining data and EC
It becomes possible to restore Column data corresponding to the failed data disk 4 from C.

In this way, matrix data 10 is created in each area, and Co1. Writing the umn data in parallel enables high-speed processing.

The method for creating and storing the matrix data 10 has been described above. As shown in FIG. 9, when the write data 9 from the CPU 1 is divided into areas and the matrix data 10 is created in the last area
The lengths of the umn data (a set of blocks stored corresponding to each data disk 4) may not be the same. Therefore, the processing in such a case will be described below. DCU
When creating the matrix data 10, No. 2 recognizes that the lengths of each Column data are not the same, as shown in FIG. Matrix data 10 in Figure 9
As can be seen, the length of the Column data corresponding to data disk #4 is two blocks shorter than that of other Column data. Therefore, when writing such matrix data 10 in parallel to each data disk #4, the Column corresponding to data disk #4
When the data transfer up to block 38 is completed, the transfer of the Column data ends. The ECC creation method at this time, that is, when the lengths of each Column data are not the same, will be explained. In creating the ECC, Column data corresponding to data disk #4 of matrix data 10 in FIG. 9 is used, and Row data from blocks 31 to 38 are Column data corresponding to data disks #1 to #4. The umn data is transferred in parallel to the ECC generator 6, and thereafter only data disks #1 to #3 are transferred in parallel to create an ECC. The DCU 2 controls the dynamic switching of the number of data disks 4 to be transferred in parallel as described above.

The matrix data 10 is created as described above, and the data and ECC are stored in the ECC disk group 14. However, regarding the writing timing at that time, the first
This will be explained with reference to Figure 6. Each C in matrix data 10
The column data is transferred to each data disk 4 in the ECC group 14 in parallel, and at the same time, the ECC
Data is similarly transferred to the generator 6 and an ECC is created. However, creating an ECC requires a certain amount of processing time, and a lag occurs between the start time of data storage on the data disk 3 and the start time of data storage on the ECC disk 5. At this time, each data disk 4 and ECC
There are two possible timings for recording onto the disc 5: synchronous and asynchronous. When synchronized, data storage and ECC storage are considered as a set, and storage of the next area is
It starts when the ECC storage is finished. On the other hand, in the case of asynchronous data storage and ECC storage are considered to be separate, and
Data storage in the next area is started without waiting for the end of CC storage. FIG. 16 shows the write time when data storage and ECC storage are handled synchronously and when they are handled asynchronously. From this figure, in the case of synchronization, when storing areas 1 and 2, the storage time = data write time + ECC creation deviation x 2. If up to n areas are stored, storage time = data writing time + ECC creation deviation Xn
becomes. On the other hand, when areas 1 and 2 are stored in valve synchronization, the storage time = data write time + ECC creation, and this is the same even when n areas are stored. From the above, the time required to write the write data 9 all at once is shorter if the data write operation and the ECC write operation are handled asynchronously. From the above, data and EC,
, C is preferably stored asynchronously.

In this way, when storing data and ECC, the EC
Each data disk 4 in the C disk group 14. EC
The rotation control of the C disk 5 will be described below. Each data disk 4 in the ECC disk group 14. ．． By synchronizing the rotation of the ECC disk 5, the average rotation waiting time (Read / 'Jri
The rotation time of the disk until the record reaches the bottom of the TE head is 172 rotations, which is the same as when using one disk.

From this, each data disk 4. ．． It is desirable to write data to the ECC disk 5 by synchronizing its rotation.

Next, the case of reading individual data in the data disk 4 will be described below with reference to FIG. In this embodiment, the size of a block (access unit at the time of reading) is 4 KB. At the time of reading, access is not made in units of ECC disk groups, but access is made to the data disk 4 that stores the record.

A Read request issued from CPU 1 is decoded as a command in DCU 2, and after performing address translation,
A read request is issued to the data drive 4 under the control of DCL'2. In the data drive 4, DCU2
As soon as it receives the command from the CPU, it disconnects from the CPU and starts reading the other data disks.
U2 and bus will be vacated.

In this case, on data disk 1, after opening DCU 2 and the bus, perform a seek to independently move the head to the track where the relevant record is stored, and wait for rotation until the relevant record comes under the head. Let's do it. After seeking and waiting for rotation, as soon as it becomes possible to read the record, acquire the DCC 20 and bus again and read
The data is transferred to the Data Buffer (RDB) 7 for the CPU, and from there the data is transferred at high speed to the CPU. In addition, when reading data disk 4, FC
C Even if the rotation of the disk S is synchronized or not, there is no horizontal crossbar).

Next, the address management method will be explained.

First, writing will be explained. When writing, CPUI
Assume that the IE CC disk group 14 is a logical volume I. The DCU 2 has an ECC disk group and a management table showing the usage status of the individual data disks 4 and ECC disks 5 within the group. This ECC disk group management table and the individual data disks 4. The management table of the ECC disk 5 has a hierarchical structure. If even one of the data disks 4 and ECC disks 5 in the ECC disk group is in use, a flag indicating that the data disks 4 and ECC disks 5 are in use is displayed in the management table that indicates their usage status. becomes "ON" (in use), and at the same time the flag indicating that the ECC disk group to which it belongs is in use becomes "ON" (in use).
become.

When writing, the CPU uses ECC as shown in Figure 19.
The management table showing the usage status of the disk group is looked at, and the flag is checked to see if the ECC disk group 14 is in use. If the ECC disk group 14 is in use due to the previous work/○ request, an I10 request is issued again, and if it is not in use, write processing begins. By preparing such a flag, it is not necessary to check the usage status of each data disk 4 in the ECC disk group 14 during batch writing. Further, the track capacity of the system in this embodiment is assumed to be four times the actual track capacity of each data disk 4 (when the number of data disks in the ECC group is four). In this embodiment, as shown in FIG. 2, sequential write data 9 is sent from the CPUI. When this write data 9 is stored in the ECC group 14, the write data 9 is converted into matrix data 10 in the data buffer 3 under the control of the DCU 2, and is converted into matrix data 10 for each data disk in the ECC disk group 14. are written simultaneously in parallel. At this time, the address conversion table and the catalog in each data disk 4 are simultaneously written as shown below. Normally, when writing, the CPU
assigns a data set name to the write data 9 and transfers it. Therefore, the DCU2 stores an address conversion table in the internal memory of the DCU2 as a correspondence table between the data set name held by the CPUI (○S) and the addresses (ECC group number, data disk number) actually managed by the DCU2. . The memory that stores the address conversion table may be volatile, but it is preferably non-volatile.

Furthermore, it is desirable that this address translation table be simultaneously written to the address translation table disk 15 to increase reliability. Further, the write data 9 is expanded into matrix data 10, and each Co1. Since the data is stored on the data disk 4 as umn data, the head number, cylinder number, and record number of the data stored on each data disk 4 are recorded in the catalog within the data disk 4.

On the other hand, when reading, the CPUI (O8) specifies the data set name and checks whether it can be read. 19th
As shown in the figure, first, the ECC disk group 1
If the flag indicating that 4 is in use is '○FF' (not in use), reading continues and the flag is turned ON.
'' (in use), it is checked whether the data disk 4 is in use.

If the data set is usable, the DCU 2 allocates the address of the data disk 4 in which the data set is stored, based on the data set name specified by the CPUI, as shown in FIG. This allocation method will be explained below. As shown in FIG. 18, the ECC disk group 14 is found from the data set name specified by the CPU 1. As shown in FIG. 17, if the address where the data center is stored in the address conversion table is from a to the bottom, it corresponds to data disk #1, and b
to C corresponds to data disk #2, C to d corresponds to data disk #3, and d to e corresponds to data disk #4. In this way, the data set is stored,
Find the address of the data disk 4, then the fifth
As shown in the figure, the data is read out according to the catalog in the data disk 11 using the head number, cylinder number, and record number in which the data is stored. In the above address management method, the catalog is placed in each data disk 4 as shown in FIG. 5, but if high-speed access is required, this catalog can be placed in the D
It is preferable to place it in the address translation table in the CUQ.

By doing so, it is possible to immediately allocate a data disk number, head number, cylinder number, and record number to the data set name transferred from the CPUI, and access the data at high speed. In addition, instead of placing the catalog only in each data disk 4, each data disk 4 is provided with a buffer for the catalog, and the data disk 4 is used to read the catalog one by one.
It is also possible to quickly find the address of the data without accessing the data.

As described above, by using this embodiment,
The conversion algorithm when converting write data 9 into matrix data 10 becomes simple. Also, managing individual blocks is easy.

In this embodiment, it is clear that similar effects can be expected even if the track capacity, block data amount, and number of data disks are changed. Also, data disk 4
Although a magnetic disk is used as the ECC disk 5, the same effect can be achieved by using an optical disk, a floppy disk, or the like.

Embodiment 2 Another embodiment of the processing method when the lengths of the column data are not the same when the DC1J2 creates the matrix data 1o will be described below as Embodiment 2. In Figure 9, write data 9 from the CPUI is divided into areas,
When the matrix data 10 is created in the last area n, the lengths of each Column data (a set of blocks stored corresponding to each data disk 4) may not be the same. In this embodiment, as in the first embodiment, an ECC is created as shown in FIG. 7 during batch writing, and when a failure occurs, the Column data corresponding to the failed data disk 4 is restored from the remaining data and ECC. becomes possible.

Therefore, the ECC creation method in such a case will be explained next. When the DCTJ 2 creates the matrix data 10, it recognizes that the lengths of each Column data are not the same, as shown in FIG.

Therefore, in this embodiment, Co1. In order to equalize the length of the umn data, IIO11 or arbitrary data is written for the missing data, as shown in FIG. 6.8. In this way, the apparent customer Column
Matrix data 10 having the same length of n data is created, and an ECC is created using a normal ECC creation algorithm. As described above, DC1J2 allows A
It is better to write L L O data and make the lengths the same.
Since there is no need to prepare an algorithm for cases where the lengths are not the same in the ECC creation algorithm, it can be simplified.

In this embodiment, LL O11 is written because the length is insufficient, but a specific pattern other than II OI+ may be written.

Further, this embodiment is applicable to all other operations described in the first embodiment.

Example 3 Another example of the matrix data creation method will be shown below as Example 3. In this embodiment, when writing data all at once, write data 9 is created in one direction in Ro (currently, IE CC disk group 14).
This is a method of converting into matrix data 10 (distributing the data sequentially to each data disk 4). In this embodiment, as in Embodiment 1, each data disk 4 and ECC disk 5 have a rank of 8 to 1 (capacity of one rotation of the magnetic disk).
40KB.

4. Data amount of each block. KB, and increase the number of data disks 4 by 1.

As shown in FIG. 10, a large amount of write data 9 transferred from the CPU J is converted into matrix data 1o under the control of the DCU 2. First, the 11th
As shown in the figure, the write data 9 is "
Divided into one area.

Next, each area is distributed to the number of two data disks 4. This distributed data becomes a collection of blocks stored on each data disk 4. The distribution method a- will be explained in detail below using FIG. 12. The write data 9 transferred from CY' U 1 is composed of a large number of blocks, and is transferred serially.

A set of one or more blocks (the size of the set should be equal) is created in this write data 9 and distributed as Column data corresponding to the data disk 4 to be written. Blockl Kara Blo with write data 9
ck up to L is stored on data disk #l in Figure 1 (Co1.umn is distributed to data #l).

B], ock L +1 to Block m should be stored in data disk #2 < Column data #2
Divide into sections.

Similarly, Block m is added to Column data 33.84.
Assign n from +1 and P from n+1. In this example, the number of data disks was four, so Colu
After assigning the mn data #4, the column data #1 is assigned lock p + 1 to q again, and the write data 9 is similarly expanded to the column data. As described above, the write data 9 is converted into matrix data 10.

By creating the matrix data 1o in this way, it becomes possible to write in parallel and at high speed by batch writing.

Next, when using this conversion method, a processing method will be described when the lengths of each Column data in the matrix data 10 are not the same. When matrix data 10 is created from write data 9 as shown in FIG.
There may be cases where the lengths of each Column data are not the same.

The data writing and ECC creation processing method in such a case may be performed as follows. In this embodiment, as in the first embodiment, an ECC is created as shown in FIG. 7 during batch writing.

When the DCU 2 creates the matrix data 10, the DCU 2 recognizes that the lengths of each Column data are not the same, as shown in FIG. Looking at the matrix data 10 in FIG. 13, the Column data corresponding to data disks #3 and 4 are shorter in length by one block than other Column data. Therefore, such matrix data 10 is stored on each data disk 4.
When writing to data disks in parallel, data disks #3 and 4
When the transfer of the Column data corresponding to blocks 35 and 36 is completed, the transfer of the Column data ends. When creating the ECC, the data disks I3 and 4 of the data disks I3 and 4 of
Column data corresponding to block 3 to 35
゜Row data from 4 to 33 is on data disk R
” or 644 in parallel with the Column data r=
c c Transfer to the generator 6, and thereafter only data disks #1 and #2 are transferred in parallel to create an ECC. Data disk 4 to be transferred in parallel as above
Dynamically switching the number of units is l') CU
2 shall be carried out.

Next, as in Example 2, we will explain how to make the lengths of Column data the same [j], and write '0' or arbitrary data for the missing data as shown in Figure 14. As in 1. below, D c t
: Each Co of 2 Nyorimato IJ Tsukusu Data 1o ],
Write the data of ALL “O” or NIN in the part where the length of um n is not enough, and it looks like Raku Co ],
, u m n data having the same length is created, and an ECC is created using a normal ECC creation algorithm. By aligning the lengths like this, F
There is no need to prepare an algorithm for when the lengths are not the same in the CC creation algorithm, which simplifies the process.

Next, the write data 9 in the case where the present invention is not applied and in the first embodiment are stored at one end in the data bath sofa 3, and then divided into four equal parts to each data disk 4 of the ECC disk group 14.
How to write data 9 to data buffer 3
In the method of writing to each data disk 4 of the ECC disk group 14 while writing to the data disk, a comparison of the respective write times will be explained using FIG. As shown in FIG. 15, the transfer rate of the data disk 4 is set to 3MB.
/s, rotation speed is 3600r, p, m, track capacity is 11u40KB/TRK, and CC disk group 14 has four data disks 4. The amount of data transferred to the ECC disk group is 40KB/4 TRKs = 160 Ni B/G
roup, a four-rotation disk would have to rotate to store this data unless the present invention is applied. The writing time at this time is 66.4 ms (open time (16,6 m5) x 4 rotations for one rotation of the disk). In the method of Example 1, when the write transfer rate to the data buffer 3 is 12 MB/s, the total writing time (160 MB) is 30 m s.
This is a reduction of 45% compared to the case where the present invention is not applied. Furthermore, when the write transfer rate to the data buffer 3 is set to 24.48 MB/s, the total write time is further reduced to 23.3.20 m5. On the other hand, when writing to each data disk 4 of the ECC disk group 14 while writing to the data buffer 3 as in this embodiment, if the write transfer rate to the data buffer 3 is set to 12 M B/s or more, the data buffer Since the writing time to data disk 3 is performed during the writing time to data disk 4, the total writing time is 16.6 m s, which is reduced to 25% compared to the case where the present invention is not applied. do.

From the above, by applying the present invention, the write time is reduced, and this embodiment is effective when the write transfer rate to the data buffer 3 is low.

It is clear that the matrix data creation method shown in this embodiment can be used in the same way even if the track capacity, block data amount, and number of data disks are changed. Furthermore, this embodiment is applicable to all other operations described in the first and second embodiments.

〔Effect of the invention〕

As explained above, in data storage where data is written in bulk and read out separately, the writing time can be shortened by writing in bulk and in parallel at high speed as in the present invention. Furthermore, during batch writing, the data is divided into areas, converted into matrix data, and written in H in parallel. At this time, if the data lengths corresponding to each data writing magnetic disk are not the same, high-speed processing is possible by using the data without narrowing the difference.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of the overall configuration of the present invention, FIG. 2 is a block diagram for explaining the data processing operation of the first embodiment, and FIG.
Figure 3 is a conceptual diagram of data conversion of the first embodiment of the present invention;
The figure shows a data disk of the present invention. An internal configuration diagram of an ECC disk, FIG. 5 is a conceptual diagram of address conversion of the present invention, FIG. 6 is a conceptual diagram of data conversion of the first embodiment, FIG. 7 is a conceptual diagram of ECC creation of the present invention, and FIG. 8 is a conceptual diagram for explaining ECC creation in the second embodiment, FIG. 9 is a conceptual diagram for explaining data conversion in the first embodiment, and FIG. 10 is a data processing operation in the third embodiment. A block diagram for explanation, No. 11171 is a conceptual diagram for explaining data conversion of the third embodiment, FIG. 12 is a conceptual diagram showing details of data conversion of the third embodiment, and FIG. 13 is a conceptual diagram for explaining data conversion of the third embodiment. FIG. 14 is a conceptual diagram explaining data conversion of the third embodiment. FIG. 15 is a comparison diagram of write processing time of the present invention. FIG. 16 is a diagram of data of the present invention. Comparison time chart of write times, FIG. 17 is a block diagram explaining the address conversion method of the present invention, FIG. 18 is a block diagram explaining the address determination method of the present invention, and FIG. 19 is a diagram explaining the address conversion method of the present invention. access flowchar]. 1,-CPji (Central Processing Unit), 2-I)
CL' (Ddj, 1 control unit), 3 data buffer,
・1 data disk (magnetic disk for data), 5■
・: C(: Disk (Magnetic for ECC (Disk), 6
G: CC generator, 7. Read data buffer, 8. Pig restoration section, 9. Input data, 10. MA (
-Rinox Data 211 Actuator, 12-R/
W circuit, 13 parity creation group, l 4-ECC
Disk group, 15-Address translation table ■ zu figure// z A74. LL-7 Vw Different Diagram Kafrono'□ Space Diagram E (C Tizfufu IL-). Fig. No. q Fig. No. ku Tomi/ρ Fig. Z Fig. times ■ Fig. Mo 2 Tei 77 (+1Lk-1 3HδIs /lρKf31 Sae and 0g Fig. No. 6 Tomi/7 Fig. No. 7)

Claims (1)

[Scope of Claims] 1. In a system consisting of a disk drive that stores the data and a control device that manages the disk drive in response to a data input/output request from a higher-level device, data is serially transferred from the higher-level device. Error detection and ECC are generated from the data, and the divided data and ECC are stored in parallel on separate disk drives at once to create a group, and when reading, individual data is A collective disk device characterized by accessing only the disk. 2. In the collective disk device, if the size of the write data transferred from the host device is constant, the write data is divided into predetermined capacities and divided into multiple areas, and each Claim 1, characterized in that the area is first stored in a semiconductor memory, and then converted into matrix-like data by dividing the area into the same number of disk drives that store data in parallel at once. aggregate disk device. 3. In the collective disk device, when data and ECC are stored in the data storage disk drive and the ECC storage disk drive, storage of the next data is started without waiting for the end of ECC storage. A collective disk device according to claim 1. 4. In the collective disk device, at least when writing data and ECC in parallel, and when a failure occurs in a disk drive, in order to restore the data therein,
2. The collective disk device according to claim 1, wherein when reading the remaining data and the ECC data, the rotations of the data storage disk drive and the ECC storage disk drive are synchronized. 5. In the collective disk device, each data storage disk drive is provided with a buffer, and by storing a catalog of data stored on the data disk in both the disk drive and the buffer, data can be transferred from the data storage disk drive. 2. The aggregate disk device according to claim 1, wherein when reading the address, the address is determined by looking at the buffer. 6. When the data serially transferred from the host device in the collective disk device is converted into matrix data and divided into the number of disk drives to be stored in parallel, the length of the data stored in the disk drive is 2. The aggregated disk device according to claim 1, wherein when the lengths are not equal, certain special data is embedded to compensate for the missing lengths to make the lengths equal. 7. In the collective disk device, the data serially transferred from the host device is sequentially distributed in a certain unit to the disk drives to be stored in parallel, creating matrix-shaped data, and transferring the data in parallel as soon as the data is completed. A collective disk device according to claim 1, characterized in that: 8. The collective disk device according to claim 1, wherein a group formed by the data storage disk drives is treated as a logical volume in the collective disk device. 9. In the collective disk device, if the track in the disk drive is unwritable during batch writing, writing to that track is not performed and the head is moved to a spare track on the disk. and then skip that track without writing it.
2. The collective disk device according to claim 1, wherein writing is performed on the next track by electrically switching the head to another track within the same cylinder. 10. When performing bulk writing in the collective disk device, in a group composed of the relevant disk drives, when writing to the cylinder of one disk drive is completed, writing to other disk drives in the group is completed. 2. The aggregated disk device according to claim 1, wherein the cylinders are moved in all disk drives in the group when the group is completed. 11. In the collective disk device, if at least one of the individual disk drives in the group is being used in a group that is a unit that performs batch writing made up of multiple disk drives, the "ON" ” A flag indicating whether the group is in use is provided as a management table in a part that controls the collective disk device, and when performing bulk writing, it is determined whether writing is possible by looking at the flag. The collective disk device according to item 1. 12. In the collective disk device, in a group that is a unit for batch writing made up of a plurality of disk drives, if each disk drive in the group is used, an in-use flag that is set to "ON" is set. , the claim is characterized in that an in-use flag that is set to "ON" when at least one disk drive in the group is in use is provided in a part that controls the collective disk device as a hierarchical management table. The collective disk device according to item 1.