JP3387017B2 - File access control method and file relocation method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a computer system such as a PC, a workstation or a server, and a disk array system, and more particularly to a data access control method for data stored in a storage device, a computer system and a disk array system.

[0002]

2. Description of the Related Art Recent breakthroughs in CPU performance have significantly improved the computing performance of workstations and servers. However, compared with the improvement of the CPU performance, the improvement of the I / O performance, particularly the access speed of the file stored in the magnetic disk device, cannot be said to be sufficient. This is an access speed bottleneck of a secondary storage device such as a magnetic disk device,
And the speed bottleneck of the I / O bus interface is the main factor.

A disk array device (or array disk device) is known as one of the techniques for eliminating this neck and realizing high-speed file access. This is a device that incorporates multiple magnetic disk devices (hereinafter referred to as disk drives or simply disks for simplification), and aims to realize high-speed file access by operating each disk in parallel. is there.
Furthermore, by using a high-speed interface for connection with the I / O bus, it is trying to eliminate the speed bottleneck of the I / O bus interface.

FIG. 46 shows the configuration of the array disk device, and FIG. 47 shows a time chart for file reading. 46 and 47 are examples in which each stack is composed of four disks, but the number of disks may be increased or decreased. The operation of the array disk device will be described with reference to these figures.

The array disk device 300 is a host CPU
When a read command is received from the device 400, a controller (not shown) inside the device first causes the stack 0
Data blocks are read out from the disks hd0 to 3 at almost the same time. Each disk in the stack 0 reads the data block from the disk medium after waiting for seek and rotation. The data blocks 0 to 3 read out from the disk are different FIFOs (310 to 31).
It is stored in 3). Since this read operation is executed independently and in parallel on the disks hd0 to 3 of the stack 0, the speed of the entire read process is realized. Next, these data blocks 0 to 3 are sequentially and sequentially transferred to the buffer 330 via the high-speed bus inside the device, i.e., the i-Bus 320, in a time division manner.
The buffer 330 arranges the data blocks 0 to 3 read independently from each disk in the correct order. Then, the data blocks 0 to 3 are transferred to the host CPU device 400 via the high-speed SCSI bus 160. afterwards,
If the next read command is issued, the controller in the array disk device immediately activates the read operation for the disks in the stack from which the data block is to be read. In addition, a method is known in which parity is used for the data block in order to ensure reliability.

The array disk device realizes high speed file access by repeating such a series of operations.

Known examples related to the array disk device include JP-A-64-41044 "data transfer system" and JP-A 64-21525 "magnetic disk controller".

[0008]

As mentioned above, the array
Achieves high-speed file access using a disk device
can do. However, make up the stack
Controller with a built-in device that controls the operation of multiple disks
As a result, complicated and expensive hardware is required. This
Due to the low cost performance of the computer system
There is a problem of defeating. Also, increase the number of disks
If you do, you must add one stack
Therefore, the flexibility regarding total capacity is insufficient.
There is also the problem that

Therefore, an object of the present invention is to provide a computer system
Array data without degrading cost performance
Achieves a high-speed file access function that is comparable to ISK
High-speed file access method, and
It is to provide a computer system. In addition, other than the present invention
For the purpose of
File access method that enhances personality, and
It is to provide a computer system using this.

The present invention aims to solve the above problems.
And multiple units without complicated and expensive hardware.
Multiple control of access to magnetic disk device by software
Who realize high-speed file access by
Law and computer system.

FIG. 48 shows a principle diagram of the present invention, and FIG. 49 shows a time chart of file reading in the present invention. 48 and 49 show examples in which the number of disks is four, but the number of disks may be increased or decreased. The principle of the present invention will be described with reference to these drawings.

In the present invention, the computer system has an SC
An I / O bus 160 having a disconnect / reconnect function such as an SI bus is provided to provide the host CPU device 400.
Further, a plurality of magnetic disk devices 170-0 to 170-3 having internal cache memories 20-0 to 20-3 are connected to the I / O bus 160 to perform the following access control.

When writing a file, the host CPU device 400 divides the corresponding file into subfiles, and these are divided into magnetic disk devices 170-0 to 170-17.
The data is sequentially written in the internal cache memories 20-0 to 20-3 of 0-3. When the host CPU device 400 finishes writing the data blocks forming the subfile into the internal cache memory of each magnetic disk device, each magnetic disk device notifies the host CPU device 400 of the end of writing. Then, the data written in the internal cache memory inside each magnetic disk device is written in the disk medium after seeking and waiting for rotation. Of course, during this time, the host CPU device 400 can write another subfile to the internal cache memory of another magnetic disk device.

When reading a file, the host CPU device 400 issues an instruction to read a subfile storing the file to the corresponding magnetic disk device 1.
Issued one after another for 70-0 to 170-3. After issuing, the host CPU device 40 via the I / O bus 160
0 and the magnetic disk devices 170-0 to 170-3 are disconnected. On the other hand, upon receipt of the read command, each magnetic disk device reads data from the disk medium into the internal cache memory after waiting for seek and rotation, and issues a reconnect request to the host CPU device 400. From the magnetic disk device that received the reconnection request earliest, the host CPU device 4
00 starts reading data into the main memory. During this time, the other magnetic disk device also performs an operation of reading data from the disk medium to the internal cache memory based on the read command issued by the host CPU, and issues a reconnect request to the host CPU device 400 upon completion of reading. The subfiles thus read are combined on the host CPU device 400 side, and the original file is reproduced.

In the above file access operation,
Since access to the internal cache memories 20-0 to 20-3 of the magnetic disk devices 170-0 to 170-3 via the I / O bus 160 can be performed at a high I / O bus speed, seek and rotation waiting are performed. , And the data transfer time between the disk medium and the internal cache memory can be extremely short. As a result, the read operation and the write operation are performed almost independently in parallel by the respective magnetic disk devices, so that high-speed file access is realized in the entire system.

As described above, whether the file is read or written, the existence of the internal cache memory of the magnetic disk device and the high speed access to it from the I / O bus are utilized to Parallel operation of the magnetic disk device and efficient use of the I / O bus can be realized. This makes it possible to provide a computer system that enables high-speed file access.

In order to realize each of the above functions, the high-speed file access control method provided by the present invention has at least one input / output bus having disconnect and reconnect functions, each of which is provided with a bus interface device. In a computer system in which a plurality of external storage devices each having an internal cache memory, which can be identified as different devices, are connected to the input / output bus connected to the main storage device, (1) is connected to the input / output bus Select a desired external storage device from the multiple external storage devices, and refer to the disk management information that stores the configuration information of the external storage device that defines the external storage device group, and specify the specified external storage device group. From the disk configuration information reference step for obtaining the external storage device name that constitutes this from (2) access target Refers to the file descriptor management information in which the file descriptor of the original file and the file descriptors of the subfiles that make up this file are stored in association with each other. A subfile descriptor reference step for obtaining a file descriptor of a file, and (3) file management information storing storage location information of each subfile constituting the original file on each external storage device in the external storage device group. A file position information reference step of referring to the position information on the external storage device in which the subfile is to be stored or already stored from the specified subfile descriptor, and (4) the file position External storage device name and corresponding external storage device that stores each subfile obtained in the information reference step When writing a file based on the above storage location information, the magnetic disk device to be written is connected to the bus interface device, and the corresponding subfile from the main storage device is stored in the internal cache memory of the magnetic disk device. Transfer to
When the transfer is completed, without waiting for the completion of writing from the internal cache memory to the disk medium, the magnetic disk device is disconnected from the bus interface device, the magnetic disk device to be the next storage target is connected, and writing is performed. The operation of transferring the target subfile to the internal cache memory of the magnetic disk device and disconnecting it as soon as the transfer is completed is repeated one after another. When reading the file, the magnetic disk device storing the subfile to be read Connect the bus interface device, issue a read request to the magnetic disk device, and do not wait for the subfile on the disk medium in the magnetic disk device to be transferred to the cache memory in the device. , Disconnect the magnetic disk unit, and immediately While repeating the operation of connecting to the disk device and issuing a read request for the next subfile, reconnect from the magnetic disk device that notified that the reading of the subfile was completed on the internal cache memory, and then the corresponding cache memory It is composed of a file access control step for performing an operation of reading the above subfile onto the main memory.

The virtual disk device composed of a plurality of magnetic disk devices described above will be referred to as a virtual array disk, or simply a virtual array.

[0019]

In the present invention, the operation of each processing step described in the means for solving the above problems will be described.

First, in the processing step (1), a virtual array disk is allocated to a designated directory of the file system to make it available. At this time, the magnetic disk device constituting the virtual array disk is recognized by referring to the disk management information, and the magnetic disk device, which is a constituent element, is made available for use as a virtual array disk.

The disk management information defines which magnetic disk device a virtual array disk is composed of. Generally, since each magnetic disk device is divided into a plurality of partitions for use, the area to be used can be determined by specifying each magnetic disk device and its partition. The disk management information manages the magnetic disk device used by the virtual array disk and its partition.

Next, in the processing step (2), the file descriptor management information in which the correspondence between the file descriptor of the original file and the file descriptor of the subfile obtained by dividing the original file is registered is referred to Get the file descriptor of a subfile for accessing the stored magnetic disk device.

The file descriptor management information manages the correspondence between the file descriptor of the original file and the file descriptor of the sub file stored in the virtual array disk. This makes it possible to access a plurality of magnetic disk devices in which subfiles are stored. When reading or writing a file, the file descriptor of the original file given by the application program is converted into the file descriptor of the sub file by referring to the file descriptor management information. Also, when creating a new file and writing it, newly allocate the file descriptors of the original file and the sub file,
These correspondences are registered in the file descriptor management information. After that, access to the subfile, which is the divided entity of the file, can be performed by the file descriptor of the original file.

Next, in the processing step (3), from the file descriptor of the subfile obtained in the processing step (2),
Get file management information of subfile. This is used to access the magnetic disk device storing the subfile.

In the file management information, the magnetic disk device storing the subfile, the striping block number, and the position information indicating the position where the data block forming the subfile is stored are registered. By referring to the file management information of each sub-file, it is determined which magnetic disk device and where the data block stored in order is to be accessed in order. According to this, the original file is accessed by sequentially accessing the sub files.

Finally, in processing step (4), the file stored in the virtual array disk is accessed. In the actual processing, the access is to the magnetic disk that stores the original file.

File access may be read or write, but in any case, the file management information of the subfile obtained in the processing step (3) is used. The file management information manages the storage position of the subfile on the magnetic disk. Therefore, the storage position of the data block to be accessed is obtained from the file management information of each subfile, and access requests are issued to the magnetic disk devices that store each subfile one after another. At this time, the order of accessing each sub-file is obtained from the file descriptor management information. In addition, the number of blocks that are continuously accessed for each magnetic disk device that stores a subfile is registered in the file management information as a striping block.

In the case of file reading, the magnetic disk device in which the subfile to be read is stored and the I / O bus interface device are connected and a read request is issued. When the request is issued, the magnetic disk device positions the head and transfers the data from the disk medium to the internal cache memory. This causes latency. Therefore, the magnetic disk device and the I / O
Disconnect the bus interface device once.
During this time, another magnetic disk device and the I / O bus interface device can be connected and a read request can be issued. In this case as well, a waiting time for positioning the head and transferring data to the internal cache memory still occurs, so that a read request can be issued to another magnetic disk device. A data block is read from the magnetic disk device that is ready for reading after head positioning is completed, and immediately after that, a read request for the next data block to be read is issued, and the magnetic disk device and the I / O bus interface device are disconnected. Control to connect. In this way, read requests are issued to each magnetic disk device one after another, and the subfile is read into the main memory, thereby performing the read process of the original file.

In the case of file writing, the magnetic disk device to be stored and the I / O bus interface device are connected and a write request is issued. After issuing the request, the data block is transferred to the magnetic disk device. The transferred data block is written in the internal cache memory of the magnetic disk device, and when the positioning of the head is completed, it is written in the disk medium. At this time, when the writing of the data block to the internal cache memory is completed, the magnetic disk device and the I / O bus interface device are disconnected, and the writing to the magnetic disk device is completed. Therefore, it is possible to issue the write request and transfer the data block by connecting the next magnetic disk device and the I / O bus interface device without waiting for the completion of writing from the internal cache memory to the disk medium. In this way, write requests are issued to each magnetic disk device one after another, and the original file is divided into subfiles and written from the main memory, thereby performing the original file write processing.

In the file access in each of the above processing steps, in the case of file reading, a read request can be issued to the next magnetic disk device during data transfer from the disk medium of the magnetic disk device to the internal cache memory. In addition, it is possible to perform multiplex control of read processing across a plurality of magnetic disk devices. By this, seek and rotation waiting in each magnetic disk device,
And data transfer to the internal cache memory is performed in parallel. Therefore, when a subfile read request is issued to each magnetic disk device, when a data block is read from each magnetic disk device in response to a reconnect request, the internal cache memory stores a subfile that can be read into the main memory. The data block will always be present. While reading these in order, issue the next read request, and then control to move to the processing of the next readable magnetic disk device, thereby achieving high-speed file reading with high I / O bus utilization efficiency. Can be realized.

In the case of file writing, a write request can be issued to the next magnetic disk device without waiting for its completion during data transfer from the internal cache memory of the magnetic disk device to the disk medium. It is possible to perform multiplex control of write processing across a plurality of magnetic disk devices. As a result, seek and rotation waiting in each magnetic disk device, and data transfer from the internal cache memory to the disk medium are executed in parallel. Therefore, high-speed file writing can be realized by sequentially issuing sub-file writing requests to each magnetic disk device.

Moreover, in the above file access,
The application program need only be aware of the file descriptor of the original file, and need not be aware of the fact that it is divided into subfiles and stored. Therefore, it is possible to access with the same application interface as a file stored in a single magnetic disk device.

As described above, according to the present invention, a high speed file access is realized by sequentially accessing a plurality of magnetic disk devices and operating them in parallel by the above processing steps.

[0034]

FIG. 1 shows the first embodiment of the present invention. The configuration of this embodiment has a CP connected to a common data bus 101.
U100, memory 110, CRT display 120,
Keyboard 130, network interface 14
0 and an I / O bus interface 150.

The I / O bus interface 150 has an I
I / O bus 160 is connected to this I / O bus 160.
Magnetic disk devices 170-0 to 170-6 are connected. The number of magnetic disk devices may be other than seven. Each magnetic disk device 170-0 to 170-6
Between the I / O bus 160 and the I / O bus 160 and in the I / O bus interface 150.
0-0 to 180-7, and the magnetic disk device 170-
When 0 to 170-6 do not transmit / receive data to / from the I / O bus interface 150 via the I / O bus 160, the electrical connection between them can be released (disconnected). When electrical connection is needed again, it can be reconnected. These controls are performed in cooperation with the CPU and the magnetic disk device.
The disconnect / reconnect device described above may be built in the magnetic disk devices 170-0 to 170-6 together with a disk controller (not shown).

The network interface 140 is connected to the network 191, and enables remote access from the workstation 190 or the like at a remote place connected to the network 191.

Next, the operation of this embodiment will be described by taking the case of writing a file on a disc as an example.

In the computer system of this embodiment, the file 10 is managed by the OS (operating system) as a set of fixed-length data blocks. That is,
As shown in FIG. 1, A10-0, B10-1, C10-
2, D10-3, E10-4, F10-5, G10-
6, H10-7, I10-8, J10-9 ,. ． ． It consists of a series of multiple blocks.

In the case of file writing according to the prior art, one magnetic disk device is selected and the writing process is performed for that particular magnetic disk device. That is,
Data blocks A10-0, B10-1, C10-
2 ,. ． ． The same magnetic disk device, for example 170-0
Will be written to. At this time, the data block is first written in the internal cache memory (not shown) of the magnetic disk device 170-0, and after being sought and waiting for rotation, written on the disk medium (not shown). CPU
When 100 writes data to an internal cache memory (hereinafter, also referred to as an internal cache for simplicity), it enters a wait state (wait state) without proceeding to the next write processing until the writing to the disk medium is completed. . Therefore, in this case, the writing of the block A10-0 to the disk medium is completely completed, and then the next block B1
It becomes possible to execute writing of 0-1. This means that the block B10-1 is replaced by the magnetic disk device 170-0
This is because even if an attempt is made to write the data to the disk medium, the data cannot be written because the data on the internal cache has not been written to the disk medium.

On the other hand, in this embodiment, the file is divided and stored in a plurality of magnetic disk devices. For example, the memory 11 is composed of fixed-length data blocks.
In the file 10 stored in 0, 1 data block is used as an I / O unit, and 4 from id = 0 to 3 in FIG.
It is divided into the magnetic disk devices 170-0 to 170-3 and stored. At this time, first, the data block A10-0 is written to the magnetic disk device 170-0 with id = 0. The data block A10-0 is the magnetic disk unit 170-0.
First, the data is written to the internal cache of the disk, then the seek and the rotation are waited, and then the data is written to a predetermined position on the disk medium.
This completes the writing process. In this embodiment, when the writing to the internal cache of the magnetic disk device 170-0 is completed without waiting for the completion of writing to the disk medium from the internal cache, the block B10-1 which is the next data block is id = The process of writing to the first magnetic disk device 170-1 is performed.
In this case as well, the process of writing the next data block C10-2 to the magnetic disk device 170-2 with id = 2 is performed without waiting for the end of writing of the data block B10-1 from the internal cache to the disk medium. Similarly, the data block D10-3 is stored in the magnetic disk device 1 of id = 3 without waiting for the completion of writing of the data block C10-2 from the internal cache to the disk medium.
Write to 70-3. After the writing of the data block D10-3 to the internal cache of the magnetic disk device 170-3 is completed, the magnetic disk device 170-0 with id = 0
After confirming that the writing of the data block A10-0 from the internal cache to the disk medium is completed, the block E10-4 is transferred to the magnetic disk device 170-without waiting for the completion of the writing of the block D10-3.
Write to 0 internal cache. If sufficient time cannot be taken to complete writing to the magnetic disk device with id = 0, the number of magnetic disk devices as a whole may be increased. As a result, the file writing performance can be further improved. Hereinafter, the data block F10-
5, G10-6, H10-7, I10-8, J10-
9 ,. ． ． In the same manner, different magnetic disk devices 170-1, 170-2, 170-3, 170-
0 ,. ． ． Then, the data blocks are written in order.
When all writing is completed, the contents of the original file are divided into four, and the four subfiles are divided into different magnetic disk devices 170.
It is stored in -0 to 170-3.

The function of issuing a data block write request to the next magnetic disk device without waiting for the end of writing the data block from the internal cache of the magnetic disk device to the disk medium is called asynchronous write. .

When a subfile is being written to a magnetic disk device by asynchronous writing, each magnetic disk device 170-0 to 170-6 and I / O bus interface 1
The 50 is electrically connected via the I / O bus 160 only while the data block is transferred to the internal cache of the magnetic disk device. That is, the host CPU
When a write request is issued from 100, the magnetic disk device is electrically connected to the I / O bus interface 150, and data blocks can be transferred. When the writing of the data block to the internal cache of the magnetic disk device is completed, the electrical connection between the magnetic disk device and the I / O bus interface 150 is released. After that, data writing from the internal cache to the disk medium is independently performed inside the magnetic disk device. Therefore,
Since the host CPU 100 can access the next different magnetic disk device during this period, as soon as it finishes writing the data block to the internal cache of the magnetic disk device, it immediately issues a write request to the next magnetic disk device, Data blocks can be transferred after establishing an electrical connection between the disk device and the I / O bus interface 150.

Also in the case of file reading, this embodiment operates in the same manner as in the case of file writing in FIG. That is, the magnetic disk device 170-0 with id = 0
The data blocks are sequentially read from. Host CP
U100 sends a read command to the magnetic disk device 170-0.
To 170-1, 170-2 ,. ． ． Issue in order. The magnetic disk device that has received the command temporarily releases the electrical connection with the I / O bus interface 150, except when the data block can be immediately read from the disk medium without waiting for seek and rotation. Then, after seeking and waiting for rotation, a sequence for reading data from a predetermined position on the disk medium is started.
Therefore, the host CPU 100 is the magnetic disk device 1
Immediately after issuing the read command to 70-0, it becomes possible to issue the read command to the next magnetic disk device 170-1 to be read.

The function of issuing a data block read request to the next magnetic disk device without waiting for the completion of reading the data block from the magnetic disk device is called asynchronous read.

Thereafter, similarly, the host CPU 100 issues read commands to the magnetic disk devices 170-2 and 170-3 one after another. On the other hand, each magnetic disk device 170
In the inside of −0 to 170-3, when data is read from the disk medium to the internal cache, the host CPU 100
A reconnect request to the host CP after establishing an electrical connection with the I / O bus interface 150.
Data transfer to U100 is performed. The reading of data from the disk medium to the internal cache can be performed independently and in parallel with the respective magnetic disk devices. That is, even in the case of reading a file, in the present embodiment, since the divided files are read from a plurality of different magnetic disk devices, the internal cache of the magnetic disk device can effectively operate to enable the parallel operation of the magnetic disk devices. Become. This enables high-speed file reading.

The disconnect / reconnect function via the I / O bus 160 described above is performed by, for example, the SCS.
I interface (Small Computer System Interface,
ANSI × 3.131-1986 standard), and the I / O bus of this embodiment can also be realized by using a SCSI interface device.

FIG. 2 shows a time chart in the case of file reading (sometimes referred to as read). In the case of using one magnetic disk device, which is a conventional technique, in (1), (2)
The respective cases of the present embodiment in which the virtual array disk of the present invention is configured and used with four magnetic disk devices are shown in FIG. The axis of the I / O bus in the figure is a time chart of the I / O bus 160, and the axis of id is a time chart of the magnetic disk device. On the I / O bus axis, the host CPU1
00 and the magnetic disk devices 170-0 to 170-3 show read request and data transfer timings. On each id axis, the timing of data transfer from the disk medium to the internal cache and the timing of data transfer from the internal cache to the I / O bus 160 are shown.

First, the operation in the case of one magnetic disk device of (1) will be described. When a read command is issued from the host CPU 100 to the magnetic disk device and the magnetic disk device is activated, command analysis is performed by the controller inside the magnetic disk device, and software overhead T1 of the magnetic disk device occurs. Then, the head is positioned to read the data block from the disk medium,
(Seek time + rotation waiting time) T2 occurs. During this period, the magnetic disk device is in a disconnected state. After the head positioning is completed, the magnetic disk device is reconnected, and while reading the data block from the disk medium to the internal cache, at the same time, I / O from the internal cache is performed.
Output to the bus 160. Therefore, the data transfer time T3 occurs.

On the other hand, in the case of (2) the virtual array disk composed of four magnetic disk devices, the host CPU 100 issues read commands to the four magnetic disk devices of id = 0 to 3 in sequence. Starts up.

First, the magnetic disk device 170 with id = 0-
When a read command is issued to 0, the magnetic disk device 170-0 waits for seek and rotation after command analysis, and is simultaneously disconnected. This allows the host CPU1
00 can issue a read command to the next magnetic disk device 170-1 with id = 1. id =
Similarly to the magnetic disk device with id = 0, the magnetic disk device 170-1 with No. 1 enters into seek and rotation waiting after command analysis, and is simultaneously disconnected. At this time, id = 0
The magnetic disk device 170-0 of No. 1 and the magnetic disk device 170-1 of id = 1 are operated independently in parallel. Further, the host CPU 100 issues read commands one after another to the magnetic disk device 170-2 with id = 2 and the magnetic disk device 170-3 with id = 3 in the same procedure. Similarly, these magnetic disk devices enter the seek and rotation wait after command analysis and are disconnected at the same time. At this point, the four magnetic disk devices operate in parallel independently of each other.

After the magnetic disk device 170-3 with id = 3 is disconnected, the magnetic disk device 170-0 with id = 0 completes positioning of the head and data can be read, or If the data is read from the disk medium and stored in the internal cache, a reconnect request is issued to the host CPU 100. Upon receiving the reconnect request, the host CPU 100 reads data from the magnetic disk device 170-0 with id = 0. At this time, if the data is stored in the internal cache in advance, the data can be read at high speed in a time T4 (T3> T4) shorter than the read time T3 from the disk medium. When the reading is completed, the next read command is issued to the magnetic disk device 170-0 with id = 0. The magnetic disk device 170-0 with id = 0 enters the seek and rotation waiting states after the command analysis and is disconnected at the same time as in the previous case. This time,
If the magnetic disk device 170-1 with id = 1 is readable, the magnetic disk device 170-1 with id = 0
As in the case of 0, the host CPU 100 accepts the reconnect request from the magnetic disk device 170-1 and reads the data. When the reading is completed, the next read command is issued to the magnetic disk device 170-1 with id = 1. Then, the present magnetic disk device 170-1 also enters the seek and rotation waiting after command analysis, and is simultaneously disconnected. Further, the host CPU 100 accepts the reconnect request to the magnetic disk device 170-2 having the next id = 2 and the magnetic disk device 170-3 having the id = 3 in the same procedure to read the data and to issue the next read command. Issue. Thereafter, this sequence is repeated as shown in FIG. In this figure, (seek time +
The rotation waiting time) T2 is assumed to be a constant, but in reality, each read command is often issued differently. In such a case, the data is read in order from the magnetic disk device that has issued the reconnection request earliest. Is also possible. At that time, the order of selecting the magnetic disk device from which data is read will be changed, but the above-mentioned procedure other than that is not changed. In FIG. 2, only the magnetic disk device 170-0 of id = 0 directly transfers data from the disk medium, and the magnetic disk devices 170-1 to 170-1 of other ids are used.
Reference numeral 70-3 is a form in which data is transferred from the internal cache. Therefore, there is a difference in data transfer time. This means that the data transfer time is shortened when the data is read into the internal cache before the host CPU 100 reads the data, and is read from the disk medium when the data is not read and the data transfer time is short. This is because it does not happen.

As is clear from FIG. 2, in the present invention, four magnetic disk devices are always operating in parallel, and the amount of data read by the host CPU 100 per unit time is effectively equal to the number of magnetic disk devices. The effect of doubling is obtained. That is, according to the present invention, the file reading speed is increased by about the number of magnetic disk devices.

In the file read shown in FIG.
A more generalized control flow (1) to (8) is shown in FIG. It is assumed that the n magnetic disk devices to be used are designated in advance and files are divided and stored in them. Information necessary for reading these files is given from, for example, an application program.
The host CPU 100 controls the file access sequence and asynchronous read.

Hereinafter, each step of the control flow will be described.

(1) A read request is sequentially issued to the magnetic disk devices with id = 0 to (n-1).

(2) Wait until one of the magnetic disk devices becomes ready for reading.

(3) If there is a readable magnetic disk device, proceed to the next step (4), otherwise return to step (2).

(4) i of readable magnetic disk device
Check and remember the d number.

(5) id = k checked in step (4)
Data is read from the magnetic disk device.

(6) If the reading is completed, the flow is ended, and if the reading is not completed, the process proceeds to the next step (7).

(7) If more data is to be read from the magnetic disk device with id = k, proceed to the next step (8). If no more data is to be read, step (2)
Return to.

(8) A read request is issued to the magnetic disk device with id = k, and step (2)
Return to.

By the control flow described above, high-speed file reading from the virtual array disk can be performed.

FIG. 4 shows a time chart in the case of file writing (sometimes referred to as write). In the case of using one magnetic disk device, which is a conventional technique, in (1), (2)
Each of the cases where the virtual array disk of the present invention is configured and used by four magnetic disk devices is shown in FIG. The meanings of the I / O bus axis and the id axis in the figure are the same as in FIG.

First, the operation in the case of one magnetic disk device of (1) will be described. When a write command is issued from the host CPU 100 to the magnetic disk device to activate it, command analysis is performed by the controller inside the magnetic disk device, and software overhead T1 of the magnetic disk device occurs. At this time, the host CPU 100 transfers the data to be written to the magnetic disk device, and the data is immediately stored in the internal cache. At this time, the data transfer time T4 occurs. With this, it is considered that the data writing process from the host CPU 100 is completed.
After that, in the magnetic disk device, the positioning of the head for writing the data block to the disk medium is performed,
(Seek time + rotation waiting time) T2 occurs. During this time, the magnetic disk device is in a busy state. After positioning the head, the magnetic disk device writes the data block from the internal cache to the disk medium. Therefore, the data transfer time T3 occurs.

On the other hand, in the case of the virtual array disk composed of (4) four magnetic disk devices, from the host CPU 100 to the four magnetic disk devices 170-0 to 170-3 with id = 0 to 3 , And write commands are issued in sequence, respectively, to activate them.

First, the magnetic disk device 170 with id = 0
A write command is issued to 0, and the data to be written is subsequently transferred and stored in the internal cache of the magnetic disk device. As a result, it is considered that the data writing process from the host CPU 100 is completed. The magnetic disk device waits for seek and rotation after command analysis.

When the head positioning is completed, the data in the internal cache is written to the disk medium. on the other hand,
The host CPU 100 can issue a write command to the next magnetic disk device 170-1 with id = 1 and transfer the data to be written subsequently. Similarly to the magnetic disk device 170-0 with id = 0, the magnetic disk device 170-1 with id = 1 waits for seek and rotation after command analysis, and writes data from the internal cache to the disk medium. At this time, the magnetic disk device 170-0 with id = 0 and the magnetic disk device 170-1 with id = 1
Are operating independently in parallel. Further, the host CPU 100 uses the next magnetic disk device 17 with id = 2.
The write command issuance and the subsequent data transfer are sequentially executed in the same procedure for the magnetic disk device 170-3 of 0-2 and id = 3. Similarly, these magnetic disk devices enter into seek and rotation waiting after command analysis, and when head positioning is completed, write data in the internal cache to the disk medium. At this point, the four magnetic disk devices 170-0 to 170-3 operate independently in parallel.

After the write command is issued and the write data is transferred to the magnetic disk device 170-3 with id = 3, if the data writing of the magnetic disk device 170-0 with id = 0 is completed, Data can be written. Therefore, the next write command issuance and data transfer are executed. Magnetic disk unit 170-0 with id = 0
As before, after entering command, seek and turn wait. At this time, if the data writing of the magnetic disk device 170-1 with id = 1 has been completed, the next data can be written. Therefore, the magnetic disk device 1 with id = 0
As in the case of 70-0, the host CPU 100 executes the next write command and data transfer to the magnetic disk device. After command analysis, seek and rotation wait. Further, when the host CPU 100 confirms the completion of the data writing previously issued to the magnetic disk device 170-2 having the next id = 2 and the magnetic disk device 170-3 having the id = 3, the next write command Issue. Thereafter, this sequence is repeated as shown in FIG. In this figure, (seek time + rotation waiting time)
It is assumed that T2 is a constant, but in reality, it often differs each time each write command is issued. In such a case, it is also possible to write the data in order from the magnetic disk device which is released from the busy state and becomes ready for writing.

The order of selecting the magnetic disk device for writing the data will be changed, but the above procedure other than that is not changed.

As is clear from FIG. 4, in the present invention, four magnetic disk devices are always operating in parallel, and the amount of data written by the host CPU 100 per unit time is effectively doubled by the number of disks. The effect is obtained. That is, according to the present invention, the file writing speed is increased by the number of magnetic disk devices.

File write shown in FIG.
FIG. 5 shows the more generalized control flows (1) to (7). It is assumed that the n magnetic disk devices to be used are designated in advance and files are divided and stored in these. The information necessary for writing these files is given from, for example, an application program.
Further, the host CPU 100 controls the division of the file into a plurality of magnetic disk devices, the sequence of file access, and the asynchronous writing.

Each step of the control flow will be described below.

(1) The following processes are sequentially executed for the magnetic disk devices with id = 0 to (n-1). 1. Issue a write request.

2. Transfer write data.

(2) Wait until one of the magnetic disk devices becomes writable.

(3) If there is a writable magnetic disk device, proceed to the next step (4), otherwise return to step (2).

(4) i of writable magnetic disk device
Check and remember the d number.

(5) If more data is to be written to the magnetic disk device with id = k, proceed to the next step (6). If no more data is to be written, proceed to step (2).
Return to.

(6) The following processes are sequentially executed for the magnetic disk device with id = k.

1. Issue a write request.

2. Transfer write data.

(7) If the writing is completed, the flow is terminated, and if the writing is not completed, the process returns to step (2).

By the above control flow, high speed file writing to the virtual array disk can be performed.

As described above, in the present embodiment, by dividing a file and writing to and reading from a plurality of different magnetic disk devices, the magnetic disk devices can operate in parallel and a high speed file access can be achieved. Will be realized. Further, it is possible to connect up to seven magnetic disk devices with one adapter board by using, for example, a SCSI bus without requiring dedicated hardware like the array disk device. For this reason,
A very low cost and high speed file access function can be realized.

It is also clear that seven or more magnetic disk devices can be connected by expanding the specifications of the SCSI interface. Furthermore, in the present embodiment, the case where fixed-length data blocks are distributed and stored in each magnetic disk device has been described as an example, but the same processing can be performed for variable-length data blocks.

FIG. 6 shows a second embodiment of the present invention. Second
In this embodiment, the file can be accessed without the application program being aware that the file is divided into a plurality of subfiles and stored in different magnetic disk devices.

The following tables and programs are used to specify a magnetic disk device to be used as a virtual array disk, divide a file into a plurality of subfiles and associate them with the magnetic disk device, and collect a plurality of subfiles. Access control to be shown in the file, and control of asynchronous read and write.

A disk management table 210 for storing disk management information, a file management table 220 for storing file management information, and a file descriptor correspondence table 23 for storing file descriptor correspondence information.
Hold 0 in memory. Access control program 200
Performs access control by referring to each of the above tables when writing and reading a file for a virtual array disk.

The above tables and programs are stored in the memory of the computer system.

Either of the above-mentioned tables can be given by an application program or can be managed as an internal management table of the OS.

The specific configuration of each table will be described below.

FIG. 7 is an explanatory diagram of the disk management table 210. The disk management table stores information regarding which partition (divided area) of which magnetic disk device is used to configure the virtual array disk.

An id number used for identifying the magnetic disk device by the SCSI interface and a magnetic disk device name used by the application software for identifying the magnetic disk device are assigned to each magnetic disk device. Further, since the magnetic disk device is usually divided into a plurality of partitions for use, each partition is assigned a partition number. Therefore, it is possible to specify the unit area when using the magnetic disk device in the system by the combination of the id number or the magnetic disk device name and the partition number. This is called a "logical disk device". In the example shown in this figure, each partition of the disk device with the disk device name hd6 is defined as a logical disk device, and for example, the fifth partition is a logical disk device with the name hd65. By combining multiple logical disk units,
Used as a "logical virtual array disk device".

The disk management table 210 is a matrix of id numbers or names of magnetic disk devices and partition numbers, and indicates which combination of these logical disk devices is defined as a logical virtual array disk device.

Four magnetic disk devices having names hd0 to hd3 are defined as virtual array disk devices va0, and each 0th partition thereof is used as a logical virtual array disk device va00 as shown in FIG. 1 or FIG. In case of va000 to va
Writing is performed as shown in FIG. 7 using up to 003 as identifiers.
As a result, the 0th partition of the magnetic disk devices hd0 to hd3 is transferred to the logical virtual array disk device va.
It is defined as 00.

Here, the identifier vaXYZ (where X,
(Y and Z are single-digit numbers) indicate that this logical disk device is a constituent element of the virtual array disk device vaX and that the logical virtual array disk device va is composed of the Yth partition of the virtual array disk device vaX.
It is a logical disk device which is a constituent element of XY and is a Zth constituent element of the logical virtual array disk device vaXY.

The disk management table 210 is set by the system administrator at the time of defining the system configuration of this computer system. This is because the setting contents of the disk management table 210 indicate the usage status of each logical disk device,
That is, it must be consistent with whether it is used for normal file storage or subfile storage.

When the mount process described later is performed on the logical virtual array disk device va00 by the above setting of the disk management table 210, hd0 to hd
The 0th partition of the four magnetic disk devices 3 can be used as the logical virtual array disk device va00.

FIG. 8 is an explanatory diagram of the file management table 220. The file management table 220 is allocated for each subfile, and has a file attribute area and
It is divided into disk block index areas. The file descriptor of each subfile is the file management table 2.
The file management table 220 is associated with the file descriptor by pointing to the head of 20-0 to 220-n.

The file attribute area comprises a file type, a file size, a storage device, and an entry for storing a striping block. The file type stores an identifier indicating whether the file managed by this table is a subfile stored in the virtual array disk device or a file stored in a normal magnetic disk device. The file size stores the capacity of the sub file. The storage device stores the name of the logical disk device, which is the substance of the magnetic disk device in which the subfile is stored. The striping block stores the number of data blocks used as a unit when the file is divided into the subfiles. That is, each subfile is created by dividing the file from the beginning for each number of data blocks.

In the disk block index area,
The logical block number on the disk is stored as the index information of the data block. In this figure, the data blocks that make up the file are logical block numbers 100 and 2,
00, 300 ,. ． ． It is stored in 900.

The file management table 220 is used when a file to be stored in the virtual array disk is created.
Created for each subfile by the access control program. After that, the file management table remains until this file is erased, and the access control program refers to this when accessing each sub-file that constitutes the file. When the file stored in the virtual array disk is erased, each subfile is erased, and at the same time, the file management table is also erased.

FIG. 9 shows the file descriptor correspondence table 230.
FIG. The correspondence between the file descriptor vfd of the original file stored in the virtual array disk and the file descriptor fd of the subfile after division is shown.

In the example of the figure, the original file with vfd = 4 consists of four subfiles, and the file descriptors of each are fd0 = 5, fd1 = 6, fd2 = 7, f.
d3 = 8. Similarly, for the original file with vfd = 20, the file descriptor of the subfile is fd0.
= 21, fd1 = 22, fd2 = 23, and fd3 = 24.

When accessing a file stored in the virtual array disk, first open the file.
I have to en. At this time, the file descriptor vfd of the original file is allocated by the access control program. Next, the name of the subfile which is the entity is generated from the name of the original file, and the subfile is opened using this subfile name. At this time, the file descriptors fd0 to fd3 are assigned to each subfile. After that, when accessing this file from the application program, the file descriptor vfd is used. The access control program uses the file descriptor correspondence table 230, converts it into the file descriptors fd0 to fd3 of the subfile, and accesses each subfile. The entries in the file descriptor correspondence table are released when the file is closed, and the contents become invalid.

The disk management table 21 described above
0, the file management table 220, and the file descriptor correspondence table 230 will be used to explain the general procedure for controlling a virtual array disk.

FIG. 10 shows the overall control flow. Each step represents a process activated by a command from a user or a function call from a program. Each process is executed by the access control program stored in the memory. The contents will be described below according to the control flow of this figure.

In the mount process, the virtual array disk is allocated to a designated location in the directory managed by the OS and made available to software as a device under the designated directory name. That is, the process of allocating the logical disk devices constituting the logical virtual array disk device to the designated location of the directory by referring to the disk management table 210 is performed.

In the open process, the file descriptors are assigned to the original file and the subfiles, and are registered in the file descriptor correspondence table 230 so that the correspondence can be established.

In read / write processing, read
When the processing is designated, the file is read from the virtual array disk, and when the write processing is designated, the file is written to the virtual array disk. In either case, the file descriptor of the subfile is obtained from the file descriptor of the original file by the file descriptor correspondence table 230. Further, the file management table 220 is obtained from the file descriptor, and the logical disk device in which the sub file is stored is accessed.

In the close process, the original file and the subfile are closed, that is, the file descriptor assigned to the original file and the file descriptor assigned to the subfile are released.

In the umount process, each logical disk device assigned to the specified directory as a logical virtual array disk device is removed from the specified directory.

Therefore, it becomes impossible for software to recognize the device under the designated directory name.

The detailed control flow of each process is shown in FIG.
1, FIG. 12, FIG. 13, FIG. 14, FIG. 15, FIG. 16, and FIG.

Control flow of mount processing (1)-
(3) will be described with reference to FIG. The device name starting with / dev is the full path name used in the mount process and the umount process.

(1) The name and partition number of the magnetic disk device corresponding to the logical virtual array disk device to be mounted, that is, the name of the logical disk device is obtained from the disk management table 210.

(2) A mount directory is created in the directory where the logical virtual array disk device is mounted. For example, if the name of the logical virtual array disk device is / dev / va00 as shown in FIG. 12, a directory with the name ".va00" is first created under the mount directory / data, and further below that, Names under "/data/.va00" are ".va000", ".va001", ".va00"
Create directories "2" and ".va003".

(3) Referring to the disk management table 210, each directory ".va000", ".va00"
1 "," .va002 ", and" .va003 "corresponding to logical disk devices / dev / hd00 to / dev /
Mount hd03.

FIG. 12 shows a tree structure formed by the result of the mount process.

As described above, in the mount process, the virtual array disk is mounted under the directory having the tree structure, and the files in the virtual array disk are managed by the tree structure directory. In this example, the logical virtual array disk device "/ dev / va00" is mounted in the "/ data" directory. A directory having the same structure is formed in each magnetic disk device, and each subfile is stored under the directory. Here, the subfile name and the directory name are distinguished by adding a subscript.

Control flow of open processing (1) to (5)
Will be described with reference to FIG.

(1) A subfile name is generated by adding a subscript from the name of the original file designated as the open target.

For example, if / data / file is specified as the original file in the directory structure of FIG. 12, the file names of the subfiles will be /data/.va00/.va000/file0 and /data/.va00/.va001/file1. , /Data/.va00/.va002/file2, /data/.va00/.va003/file3, are generated.

(2) Each subfile is opened using the file name of the subfile, and the file descriptor f
Allocate d0 to fd3. As a result, the file descriptors fd0 to fd3 of each sub-file are associated with the file management table 220, and access using the file descriptor from the application program becomes possible.

(3) File descriptor vfd in the original file
Assign

(4) File descriptor vfd of the original file
And the file descriptors fd0 to fd3 of the sub files are paired and registered in the file descriptor correspondence table 230. (5) Replace the file descriptor vfd of the original file with this o
It is returned to the application program as a result of the pen processing, that is, as a return value of the open call.

Next, read / write processing
FIG. 14 regarding control flows (1) to (5) for executing d processing
Will be explained.

(1) The application program issues an access request with the file descriptor vfd of the original file as an argument. File descriptor vf that is an argument
The file descriptor correspondence table 230 is searched by using d as a key, and the file descriptors fd0 to fd3 of the subfiles corresponding to vfd are obtained. As shown in FIG. 9, for example, when the file descriptor vfd of the original file is 4, the file descriptors fd0 to fd3 of the subfiles are 5, 6, 7, 8
Is obtained.

(2) Next, the file management table 220 of these subfiles is obtained from the file descriptors fd0 to fd3 of the subfiles.

(3) The subfiles are accessed in the order of the subscripts of the descriptors of the disk management table 210. File management table 2 for subfiles in this order
Referring to 20, the logical block number of the data block to be read is obtained. In the example shown in FIG. 7, logical disk device / d
ev / hd00, / dev / hd01, / dev / hd
02, / dev / hd03 are accessed in this order. When the number of data blocks designated by the striping block written in the file management table 220 is accessed, the next subfile is accessed. That is,
Refer to the file management table of the following sub file,
Access the number of data blocks specified by the striping block. According to this procedure, the logical block numbers of the data blocks to be read are obtained in order.

(4) The data block is read according to the logical block number of the data block read from the magnetic disk device and the logical disk device written in the storage device entry of the file management table 220 of the subfile. The magnetic disk device that you actually access is
By referring to the disk management table 210, it can be obtained as a magnetic disk device including the relevant logical disk device as a part thereof.

(5) Finally, it is judged whether all the data blocks to be read have been read. If there is a data block to be read, return to (3) and repeat.

Next, wri in read / write processing
te processing execution control flow (1) to (7) FIG.
This will be described using 5. The outline of the procedure for accessing the data block of the subfile is the same as that at the time of executing the read process.

(1) The application program issues an access request with the file descriptor vfd of the original file as an argument. File descriptor vf that is an argument
The file descriptor correspondence table 230 is searched by using d as a key, and the file descriptors fd0 to fd3 of the subfiles corresponding to vfd are obtained. As shown in FIG. 9, for example, when the file descriptor vfd of the original file is 4, the file descriptors fd0 to fd3 of the subfiles are 5, 6, 7, 8
Is obtained.

(2) Next, the file management table 220 of these subfiles is obtained from the file descriptors fd0 to fd3 of the subfiles.

(3) When updating the contents of the existing file, the contents of the data block may be overwritten. However, when additional writing is performed beyond the file size, it becomes necessary to allocate a new area for writing. Therefore, it is determined whether or not a new disk block needs to be allocated, and if necessary, the procedure proceeds to (4), and if not, the procedure proceeds to (5).

(4) When allocating a disk block, it is determined which subfile is to be allocated from the logical disk device in which it is stored. If the total number of data blocks of the subfile that is the final part of the file is an integer multiple of the number of data blocks written in the striping block entry of the file management table, the data from the logical disk device that stores the next subfile Allocate blocks. If not, allocation is performed from the logical disk device in which the subfile of the last part of the file is stored. After allocating the data block, the logical block number of the data block is obtained. Next, proceed to (6).

(5) As described in the case of read processing, the original file is divided into sub-files, and the access order for synthesizing with the original file is determined by this. Therefore, according to the procedure, the logical block number of the data block in which the file is written is obtained from the file management table 220 of the subfile corresponding to the writing location.

(6) The data block is written according to the logical block number of the data block to be written to the magnetic disk device and the logical disk device written in the storage device entry of the file management table 220 of the subfile. The magnetic disk device to be actually accessed can be obtained as a magnetic disk device including the relevant logical disk device as a part thereof by referring to the disk management table 210.

(7) Finally, it is judged whether all the data blocks to be written have been written. If there is a data block to be written, return to (3) and repeat.

Control Flow of Close Processing (1)-
(3) will be described with reference to FIG.

(1) The application program issues a processing request with the file descriptor vfd of the original file as an argument. The file descriptor correspondence table 230 is searched using the file descriptor vfd as a key, and the file descriptors fd0 to fd3 of the subfiles corresponding to vfd are searched.
To get

(2) Release the file descriptors fd0 to fd3 corresponding to the subfiles.

(3) Release the file descriptor vfd corresponding to the original file.

Finally, control flows (1) to (3) of the umount process will be described with reference to FIG.

(1) By referring to the disk management table 210, the name and partition number of the magnetic disk device corresponding to the logical disk device constituting the unmounted logical virtual array disk device are obtained.

(2) The logical disk device is unmounted. That is, in the example of FIG. 12, the directory “.
va000 ”,“ .va001 ”,“ .va002 ”, and“ .va003 ”to / dev / hd00 to / de
Remove v / hd30.

(3) Erase the directory created for mounting the virtual array disk. That is, FIG.
Example, ".va00", ".va000", ".va"
001 ”,“ .va002 ”, and“ .va003 ”are deleted.

As described above, the virtual array disk is defined by the disk management table 210, the original file and the subfile are associated with each other by the file descriptor correspondence table 230, and the subfile is referenced by referring to the file management table 220. By accessing, virtual users can use virtual array disks as if they were accessing a single file, without having to be aware of accessing multiple subfiles. .

File striping processing will be described below with reference to FIGS.

In the embodiment shown in FIG. 1, a file is divided into sub-files, that is, striping is performed in units of one data block. In this case, B, C, and D are stored in order from the data block A, from the id number 0 of the magnetic disk device to the first, second, and third numbers in order.
Then, from the next data block E, F, G, H are
Similarly, from the magnetic disk device id number 0 to 1 in order,
Store in numbers 2 and 3. FIG. 18 shows how the file is written and FIG. 1 shows how the file is read in this case.
9 shows.

FIG. 18 shows a case where files are written in order from the beginning. Data blocks are sequentially sent from the host CPU 100 via the I / O bus interface 150. Each magnetic disk device is sequentially connected to the I / O bus interface 150 by the disconnect / reconnect function, and data blocks 0, 1, 2, 3,
4. ． ． Have id numbers 0, 1, 2, 3, 0
No. ． ． Is stored in the magnetic disk device. The order of the magnetic disk devices to be stored does not necessarily have to be the order of the id numbers, and it is possible to start from a magnetic disk device other than 0. Further, the order of the id numbers may be other than ascending order or descending order.

FIG. 19 shows a case where the files are read in order from the beginning. Data blocks are sequentially sent to the host CPU 100 via the I / O bus interface 150. The data blocks are stored as shown in FIG. 18, and the magnetic disk devices are sequentially connected to the I / O bus interface 150 by the disconnect / reconnect function, and the data blocks 0, 1, 2, 3,
4. ． ． Have id numbers 0, 1, 2, 3, 0
No. ． ． Read from the magnetic disk device. In addition,
The order of the magnetic disk devices to be stored does not necessarily have to be the order of the id numbers, and it is also possible to start from a magnetic disk device other than 0. Further, the order of the id numbers may be other than ascending order or descending order.

On the other hand, it is possible to use a plurality of data blocks as a striping unit. As an example, FIG. 20 shows how files are written when striping is performed in units of four data blocks, and FIG. 21 shows how files are read.

FIG. 20 shows a case where files are sequentially written from the beginning in units of four data blocks. I
Host CPU via I / O bus interface 150
Data blocks are sequentially sent from 100. I / O for each magnetic disk unit by disconnect / reconnect function
The data blocks 0, 1, 2, 3 are connected to the O-bus interface 150 in order, and the data blocks 0, 1, 2, 3 have the id number 0, the data blocks 4, 5, 6, 7 have the id number 1, and the data block 8,
9, 10, 11 have id number 2, data block 12,
13, 14, and 15 have id number 3, data block 1
6, 17, 18, and 19 have id numbers 0. ． ． Are stored in the respective magnetic disk devices. The order of the magnetic disk devices to be stored does not necessarily have to be the order of the id numbers, and it is possible to start from a magnetic disk device other than 0. Further, the order of the id numbers may be other than ascending order or descending order.

FIG. 21 shows a case where files are read in order from the beginning in units of four data blocks. I
Host CPU via I / O bus interface 150
Data blocks are sent to 100 in sequence. The data blocks are stored as shown in FIG. 20, and each magnetic disk device is I / O-controlled by the disconnect / reconnect function.
The data blocks 0, 1, 2, 3 are connected to the O-bus interface 150 in order, and the data blocks 0, 1, 2, 3 have the id number 0, the data blocks 4, 5, 6, 7 have the id number 1, and the data block 8,
9, 10, 11 have id number 2, data block 12,
13, 14, and 15 have id number 3, data block 1
6, 17, 18, and 19 have id numbers 0. ． ． Read from the magnetic disk device. The order of the magnetic disk devices to be stored does not necessarily have to be the order of the id numbers, and it is possible to start from a magnetic disk device other than 0. Further, the order of the id numbers may be other than ascending order or descending order.

It goes without saying that the number of data blocks as a striping unit can be any number other than four.

A combination of logical disk devices used as a logical virtual array disk device will be described with reference to FIGS.

In this embodiment, the disk management table 21
A logical disk device to be used is defined by specifying the id number or name of the magnetic disk device and the partition number with 0. As described in the description of FIG. 7, a series of identifiers vaXY0 and vaXY are assigned to a plurality of logical disk devices.
1 ,. ． ． By adding, it is possible to define a combination of logical disk devices used as a logical virtual array disk device. The meaning of the suffix of the identifier is as described in the description of the disk management table in FIG. In this embodiment, the degree of freedom in the physical arrangement of subfiles is extremely high, and the following configuration is possible. Note that, for example, in FIG.
Areas that are not defined as logical virtual array disk devices, such as the disk devices with disk device names hd4 and hd5, can be used as normal logical disk devices without particular notice.

FIG. 22 shows the case where all the partitions of the four magnetic disk devices with id numbers 0, 1, 2, 3 are used as the logical virtual array disk device.

Disk management table 21 corresponding to this
FIG. 23 shows an example of setting 0. Four magnetic disk devices are defined as a virtual array disk device va0, and the whole is one logical virtual array disk device va0.
This corresponds to the case of using as 0. The three magnetic disk devices with id numbers 4, 5, and 6 are used as normal magnetic disk devices.

FIG. 24 shows the case of using specific partitions of four magnetic disk devices with id numbers 0, 1, 2, and 3.

FIG. 25 shows a setting example of the disk management table corresponding to this. This is a case where the 0th partition of the virtual array disk device va0 defined similarly to FIG. 23 is used as the logical virtual array disk device va00. The three magnetic disk devices with id numbers 4, 5, and 6 are used as normal magnetic disk devices.

In FIG. 26, three partitions of four magnetic disk devices with id numbers 0, 1, 2, and 3 are used.
This is a case where three sets of logical virtual array disk devices are defined.

The disk management table 21 corresponding to this
FIG. 27 shows an example of setting 0. The 0th, 1st and 2nd virtual array disk devices va0 defined in the same way as in FIG.
This is a case where the partitions are used as logical virtual array disk devices va00, va01, va02, respectively. The three magnetic disk devices with id numbers 4, 5, and 6 are used as normal magnetic disk devices.

FIG. 28 shows specific partitions of four magnetic disk devices with id numbers 0, 1, 2, and 3, and ids.
Two logical virtual array disk devices are used by using specific partitions of two magnetic disk devices of numbers 4 and 5.
This is the case when a pair is defined.

Disk management table 21 corresponding to this
An example of setting 0 is shown in FIG. Four magnetic disk devices with id numbers 0, 1, 2, and 3 are defined as a virtual array disk device va0, and two magnetic disk devices with id numbers 4 and 5 are defined as a virtual array disk device va1, respectively. This is a case where the partitions are used as the logical virtual array disk devices va00 and va10, respectively. The magnetic disk device with id number 6 is used as a normal magnetic disk device.

FIG. 30 shows three partitions of four magnetic disk devices with id numbers 0, 1, 2, and 3, and i.
This is a case where six sets of logical virtual array disk devices are defined by using three partitions of two magnetic disk devices of d numbers 4 and 5, respectively.

Disk management table 21 corresponding to this
An example of setting 0 is shown in FIG. This is a case where the 0th, 1st, and 2nd partitions of the virtual array disk devices va0, va1 defined similarly to FIG. 29 are used as the logical virtual array disk devices va00, va01, va02, and va10, va11, va12, respectively. The magnetic disk device with id number 6 is used as a normal magnetic disk device.

FIG. 32 shows three partitions of two magnetic disk devices with id numbers 0 and 2, two partitions of two magnetic disk devices with id numbers 4 and 6, and two partitions with id numbers 1 and 3. This is a case where three sets of logical virtual array disk devices are defined by using one partition of each magnetic disk device.

Disk management table 21 corresponding to this
An example of setting 0 is shown in FIG. The four magnetic disk devices with id numbers 0, 2, 4, and 6 are defined as virtual array disk device va0, and the four magnetic disk devices with id numbers 0, 1, 2, 3 are defined as virtual array disk device va1, respectively. The 0th and 2nd partitions of va0 are logical virtual array disk devices va00 and v, respectively.
As a02, the first partition of va1 is used as the logical virtual array disk device va11. Two magnetic disk devices with id numbers 0 and 2 are va
Although 0 and va1 are defined in duplicate, no problem occurs if the partitions to be used do not overlap. The magnetic disk device with id number 5 is used as a normal magnetic disk device. In each of the above configuration examples, the number of magnetic disk devices and the id number can be realized in combinations other than the above.

34 to 39 are examples of realizing the mirror mode in this embodiment. That is, the primary (primary) and backup (secondary) sets are defined as a set as a logical virtual array disk device, and the same file is stored in both. As a result, when a failure occurs in the primary, the backup can be used instead to improve reliability.

In the case of writing a file, it is written in both the primary and backup.

First, a data block is written to the logical virtual array disk device defined as the primary by asynchronous writing. Subsequently, the same data block as that of the primary is written by the asynchronous write to the logical virtual array disk device defined as the backup. After issuing the write command to the backup, the process moves to the writing process of the next data block, and the data block is written to the primary and the backup by the same procedure as above.

Regarding the method of writing data to the primary and the backup, for each corresponding disk device constituting each logical virtual array disk device, that is, h
It is possible to write as d0 and hd2, hd1 and hd3, and conversely, to write as hd0 and hd3, and hd1 and hd2. This is performed by using the value of Z of the identifier vaXYZ set in the disk management table as a key to establish correspondence. A concrete example is shown in FIG.
It will be described in the description of 5.

As a result, the mirror mode can be realized while the user hardly feels the write time to the backup.

When reading a file, only the primary is read. That is, the file is read from the primary in the same manner as the file stored in the logical virtual array disk device for which the mirror mode is not specified. However, if the primary fails, it will use the files stored in the backup instead.

This makes it possible to improve the reliability of the magnetic disk device. Also, in the normal usage state, files are read from only the primary, so there is no penalty due to the existence of backup,
Files can be read at high speed.

A combination of logical disk devices used as a logical virtual array disk device in the mirror mode will be described below with reference to the drawings. Similar to the definition of the logical virtual array disk device which is not in the mirror mode, the logical management is performed by adding a series of identifiers to the logical disk device to be used by the disk management table 210. Furthermore, whether it is primary or backup and its combination are indicated by an extension. For example, vaXYZ_p0 is defined as a component of the primary logical virtual array disk device in the mirror mode in the system where the logical disk device with the identifier vaXYZ is defined, and a logical disk device of vaXYZ_b0 defined as a component of the backup logical virtual array disk device. It shows that it is paired with. An area that is not defined as a logical virtual array disk device can be used as a normal logical disk device without special notice.

In FIG. 34, one partition of each of the two magnetic disk devices with id numbers 0 and 1 is used as a primary, and one of each of the two magnetic disk devices with id numbers 2 and 3 is used.
This is a case where one set of logical virtual array disk devices in mirror mode is defined as a backup using one partition.

Disk management table 21 corresponding to this
An example of setting 0 is shown in FIG. Two magnetic disk devices with id numbers 0 and 1 are connected to the virtual array disk device va.
Two magnetic disk devices of 0 and id numbers 2 and 3 are defined as virtual array disk device va1, respectively.
The 0th partition of each logical virtual array disk device va00_p0 (primary), va
This is the case of using as 10_b0 (backup).
In the example of the figure, the logical disk devices va000-p0 and va0011-p0 forming the primary are connected to the logical disk devices va100-b0 and va1 forming the backup.
01-b0 correspond to each. The magnetic disk devices with id numbers 4, 5, and 6 are used as normal magnetic disk devices.

In FIG. 36, two partitions of four magnetic disk devices with id numbers 0, 1, 2, 3 are used.
This is a case where one set of logical virtual array disk devices in the mirror mode is defined with the first partition as the primary and the second partition as the backup.

The disk management table 21 corresponding to this
An example of setting 0 is shown in FIG. Four magnetic disk devices with id numbers 0, 1, 2, 3 are defined as virtual array disk device va1, and the first and second partitions are logical virtual array disk device va11_p, respectively.
1 (primary) and va12_b1 (backup). The magnetic disk devices with id numbers 4, 5, and 6 are used as normal magnetic disk devices.

In FIG. 38, one partition of four magnetic disk devices with id numbers 0, 1, 2, and 3 is used as a primary, and one of the four magnetic disk devices with id numbers 3, 4, 5, and 6 is used. This is a case where one set of logical virtual array disk devices in the mirror mode is defined as a backup using the partition.

The disk management table 21 corresponding to this
An example of setting 0 is shown in FIG. Four magnetic disk devices with id numbers 0, 1, 2, 3 are defined as virtual array disk device va2, and four magnetic disk devices with id numbers 3, 4, 5, 6 are defined as virtual array disk device va3, respectively. The third and fourth partitions are respectively assigned to the logical virtual array disk device va23_p2.
(Primary) and va34_b2 (backup).

FIG. 40 shows the third embodiment of the present invention. The id numbers 0, 1, 2, in the first embodiment shown in FIG.
The magnetic disk devices 170-0 to 170-3 of No. 3 are replaced with array disk devices 300-0 to 300-3. Since the array disk device has a higher file transfer performance as a single device than the magnetic disk device, the present embodiment can obtain the effect of further speeding up file access as compared with the first embodiment.

The operation and control system is exactly the same as in the first embodiment of FIG.

Array disk device 30 corresponding to this
FIG. 41 shows a setting example of the disk management table of the virtual array disk including 0-0 to 300-3. Four of the id numbers 0, 1, 2, 3 are array disk devices. Here, four array disk devices ad0, ad
The virtual array disk device va0 is composed of 1, ad2, and ad3.

Also in this embodiment, the logical virtual array disk devices va01_p0 and va02_b0,
It is also possible to support the mirror mode like va21_p1 and va22_b1.

In the above-described embodiments, the case where seven magnetic disk devices are connected has been described as an example, but the number of magnetic disk devices is 2, 3, 4 ,. ． ． It is also possible to increase sequentially. Here, for example, when the number of four magnetic disk devices is increased to five, it is necessary to relocate the files stored in the four disks to the five magnetic disk devices.

According to the present invention, by copying a file to a new partition after adding a magnetic disk device, it is possible to automatically increase the striping number of the file. That is, when reading and writing a file stored in the virtual array disk device, the user can perform processing without being aware of the physical division of the file, so by simply copying the file from the copy source to the copy destination, The user can convert the striping number without paying particular attention to this. This case will be described below.

FIG. 42 shows a setting example of the disk management table 210, and FIG. 43 shows a control flow of file copy when adding disks. In FIG. 42, the logical virtual array disk device va00 constituted by the 0th partition with the magnetic disk device id numbers 0, 1, 2, and 3 is the copy source, and the magnetic disk device id
Logical virtual array disk device va1 configured by the first partitions of numbers 0, 1, 2, 3, 4
1 is a copy destination.

According to the control flow shown in FIG. 43, va0
The file is divided into fixed-length data and copied from 0 to va11.

(1) The copy destination logical virtual array disk device va11 is defined in the disk management table 210.

(2) Fixed-length data obtained by dividing a file is read from the copy-source logical virtual array disk device va00 according to the file read control flow of FIG. 14, and this is read to the copy-destination logical virtual array disk device va11. Writing is performed according to the file writing control flow in FIG.

(3) It is judged whether or not copying is completed, and if there is any remaining copy, the process returns to (2) and is repeated. If copying is completed, proceed to (4).

(4) Invalidate the copy source logical virtual array disk device va00 on the disk management table 210.

As described above, it is possible to easily change the striping number due to the addition of magnetic disk devices. At this time, an arbitrary number of blocks can be defined as the fixed-length data which is a unit of reading and writing.

FIG. 44 is an embodiment of the present invention having a plurality of I / O buses. In the configuration of the first embodiment of FIG. 1, I /
Although there is only one O bus, FIG. 44 has a configuration having four I / O buses. Therefore, I / O bus interfaces 150, 151, 152, 153 and I / O
Bus 160, 161, 162, 163 and each I / O
Seven magnetic disk devices 170-0 to 170 for each bus
-6, 171-0 to 171-6, 172-0 to 172-
There are four sets of disk subsystems 6 and 173-0 to 173-6. Except that the access control program schedules to a plurality of I / O buses, the control method of each disk subsystem can be applied as it is to the above-mentioned respective embodiments. For example, it is possible to perform multiple access control as in the first embodiment for each I / O bus, and further perform multiple access control on one hierarchy between I / O buses. It is also possible to perform multiple access control in parallel for each I / O bus.

Each I / O bus 160, 161, 162, 1
In the data transfer between 63 and the common data bus 101, the speed of the common data bus is higher than four times the speed of the I / O bus, and the partner of the data transfer is specified by using the control method described above. In the case where there is no concentration on the I / O bus, the embodiment of this figure can achieve the maximum data transfer rate. In that case, I /
In the embodiment shown in the figure, the data transfer speed is four times as high as that in the first embodiment having one O-bus. Therefore, according to this embodiment, it is possible to further improve the high speed of the virtual array disk. It should be noted that the case where there are four I / O buses is shown in this figure, but it is of course possible to adopt a configuration other than four.

Further, when the system is expanded by adding the magnetic disk device, it is possible to add the magnetic disk device in each I / O bus and the I / O bus unit. For this reason, compared to array disk devices, which require additional magnetic disk devices for each stack,
It has an advantage that it can respond to the expansion of the system more flexibly.

As described above, according to the present embodiment, there is an effect that it is possible to further speed up file access using the virtual array disk device and realize flexible system expandability at the same time.

As shown in the third embodiment of FIG. 40, the magnetic disk device of the embodiment of FIG. 44 may be replaced with an array disk device. This embodiment is shown in FIG. In FIG. 45, the magnetic disk device of the 0th I / O bus is the array disk device 300-0 to 30.
It is 0-6. This makes it possible to further speed up file access. At this time, in principle, the array disk device can be replaced with an arbitrary magnetic disk device, and a configuration other than that shown in this figure can be easily realized.

In the embodiment having a plurality of I / O buses shown in FIGS. 44 and 45, one I
Obviously, the various functions of the invention described in the description of the embodiment with an / O bus can be implemented.

According to the present invention, in a computer system having an I / O bus having a disconnect / reconnect function, files are divided and stored in a plurality of disk devices connected to the I / O bus and multiplexed by software. -By performing parallel access control, there is an effect that a high-speed file access comparable to an array disk device can be realized without using expensive hardware. Moreover, the user can access the files divided into a plurality of disks as if using one disk device. As a result, it is possible to obtain an effect that it is possible to realize a computer system capable of high-speed file access at a much lower cost than before.

Further, since no special hardware is used, it is possible to add one disk unit at a time when adding disk devices, and there is also an effect that the system can be flexibly expanded.

[0208]

According to the present invention, it is possible to realize a computer system capable of high-speed data access.

[Brief description of drawings]

FIG. 1 is a diagram showing a first embodiment.

FIG. 2 is a time chart of file reading in the first embodiment.

FIG. 3 is a flowchart of file reading according to the first embodiment.

FIG. 4 is a time chart of file writing in the first embodiment.

FIG. 5 is a flowchart of file writing in the first embodiment.

FIG. 6 is a diagram showing a second embodiment.

FIG. 7 is a diagram showing a disk management table.

FIG. 8 is a diagram showing a file management table.

FIG. 9 is a diagram showing a file descriptor correspondence table.

FIG. 10: Overall flowchart of file access

FIG. 11 is a control flowchart of mount processing.

FIG. 12 is a diagram showing a directory structure.

FIG. 13 is a control flowchart of open processing.

FIG. 14 is a control flowchart of read processing.

FIG. 15 is a control flowchart of write processing.

FIG. 16 is a control flowchart of close processing.

FIG. 17 is a control flowchart of umount processing.

FIG. 18 is a diagram showing striping in units of one data block.

FIG. 19 is a diagram showing striping in units of one data block.

FIG. 20 is a diagram showing striping in units of four data blocks.

FIG. 21 is a diagram showing striping in units of four data blocks.

FIG. 22 is a diagram showing a used area of a virtual array disk.

FIG. 23 is a diagram showing a setting example of a disk management table.

FIG. 24 is a diagram showing a used area of a virtual array disk.

FIG. 25 is a diagram showing a setting example of a disk management table.

FIG. 26 is a diagram showing a used area of a virtual array disk.

FIG. 27 is a diagram showing a setting example of a disk management table.

FIG. 28 is a diagram showing a used area of a virtual array disk.

FIG. 29 is a diagram showing a setting example of a disk management table.

FIG. 30 is a diagram showing a used area of a virtual array disk

FIG. 31 is a diagram showing an example of setting a disk management table.

FIG. 32 is a diagram showing a used area of a virtual array disk

FIG. 33 is a diagram showing a setting example of a disk management table.

FIG. 34 is a diagram showing a use area of a mirror mode.

FIG. 35 is a diagram showing a setting example of a disk management table in the mirror mode.

FIG. 36 is a diagram showing a use area of a mirror mode.

FIG. 37 is a diagram showing an example of setting the disk management table in the mirror mode.

FIG. 38 is a diagram showing a use area of a mirror mode.

FIG. 39 is a diagram showing a setting example of the disk management table in the mirror mode.

FIG. 40 is a diagram showing a third embodiment of a configuration including an array disk.

FIG. 41 is a diagram showing a setting example of a disk management table including array disks.

FIG. 42 is a diagram showing an example of setting a disk management table for file copy when adding disks.

[Fig. 43] Flowchart for file copy control when adding disks

FIG. 44 is a diagram showing an embodiment having a plurality of I / O buses.

FIG. 45 is a diagram showing an embodiment including an array disk having a plurality of I / O buses.

FIG. 46 is a diagram showing the configuration of an array disk.

FIG. 47 is a time chart of an array disk

FIG. 48 is a principle diagram of the present invention

FIG. 49 is a time chart of file reading according to the present invention.

[Explanation of symbols]

10 ... File, 10-0 to 10-9 ... Data blocks forming file, 20-0 to 20-3 ... Internal cache memory of magnetic disk device, 100 ... CPU, 10
1 ... Common data bus, 110 ... Memory, 120 ... CRT
Display, 130 ... Keyboard, 140 ... Network interface, 150, 151, 152, 153
... I / O bus interface, 160, 161, 16
2, 163 ... I / O bus, 170-0 to 170-6 ... Magnetic disk device, 171-0 to 171-6 ... Magnetic disk device, 172-0 to 172-6 ... Magnetic disk device,
173-0 to 173-6 ... Magnetic disk device, 180-
0-180-7 ... disconnect / reconnect means, 1
81-7, 182-7, 183-7 ... Disconnect /
Reconnecting means, 190 ... Workstation, 191
... LAN, 200 ... Access control program, 210 ...
Disk management table, 220 ... File management table, 230 ... File descriptor correspondence table, 300, 3
00-0 to 300-6 ... Array disk device, 310 ...
FIFO0, 311 ... FIFO1, 312 ... FIFO
2, 313 ... FIFO3, 320 ... Internal bus, 330 ...
Buffer, 340 ... SCSI bus interface, 40
0 ... Host CPU device.

Continued Front Page (72) Hiroyoshi Suzuki, 5030 Totsuka-cho, Totsuka-ku, Yokohama-shi, Kanagawa 5030, Hitachi Ltd. Software Business Department (72) Toshiyuki Maki, 1 Horiyamashita, Hadano-shi, Kanagawa Hitachi Computer Engineering Co., Ltd. (56) References JP-A-4-49413 (JP, A) JP-A-5-250099 (JP, A) JP-A-2-236714 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G06F 3/06 G11B 20/10 G06F 12/00 G06F 13/00

Claims (4)

(57) [Claims] 1. A plurality of disk devices are connected to a bus interface device via a unique disk connect / reconnect device, the bus interface device is connected to a bus, and a CPU and a main storage device are connected to the bus. A method for accessing a file in a computer system, wherein the main memory stores a correspondence relationship between a file descriptor of an original file specified by an application program as an argument in a read request and file descriptors of a plurality of subfiles forming the original file. It is stored in the device file descriptor correspondence table, and for each file descriptor of each subfile, the capacity of the subfile, the name of the theoretical disk device in which the subfile is stored, and the original file is divided into the subfiles. Sometimes indicates the number of data blocks in units A file management table including a triping block is held, and a relationship between the name of the disk device and the name of the logical disk device and the relationship of the partition of the disk device assigned to the logical disk is held in the disk management table. In response to a request to read the original file from the application program using the file descriptor as an argument, the file descriptor correspondence table is referenced to correspond to the file descriptor of the original file specified by the argument. Obtain the file descriptors of multiple subfiles, refer to the multiple file management tables corresponding to the obtained file descriptors of multiple subfiles, and determine the size of the subfiles that correspond to the multiple obtained subfiles. , Get the name of the logical disk, and the striping block, By referring to the disk management table, the partition of the disk device corresponding to the obtained logical disk name is obtained, and one disk device is stored for each of the plurality of disk devices storing the sub-files forming the original file. A process of connecting the device and the bus interface device, issuing a subfile read request from the CPU to the partition of the disk device obtained by referring to the disk management table, and then connecting the disk device. By sequentially repeating the above, during the waiting time caused by the positioning of the head of the disk device and the data transfer from the disk medium of the disk device to the cache memory, a sub file read request to another disk device is issued from the CPU. Issue and read subfile A process for issuing a reconnect request to the CPU from the disk device that is enabled to start reading the subfile from the disk device that has received the reconnect request to the main storage device, and performing a disk connect as soon as the read process is completed. Is executed for a plurality of disk devices that have issued a read request. 2. The mount process refers to the disk management table to allocate a logical disk device to a directory managed by the OS so that it can be used by software, and the unmount process sets the logical disk device to the OS.
2. The file access control method according to claim 1, wherein the file access control method makes the software unusable by removing it from the directory managed by. 3. A file descriptor is allocated to an original file and a subfile in the open process, a correspondence relationship is registered in the file descriptor correspondence table, and a file descriptor and a subfile are allocated to the original file in the close process. The file access control method according to claim 2, wherein the allocated file descriptor is released from the file descriptor correspondence table. 4. Under the control of the CPU, two sets of primary and secondary logical virtual array disks are defined by using the plurality of disks, and writing of a file is performed after writing to the positive logical virtual array disk. Write to the logical virtual array disk and read the file from the positive logical virtual array disk if there is no failure in the positive logical virtual array disk, and the secondary logical virtual array disk if there is a failure in the positive logical virtual array disk. The file access method according to claim 1, wherein the file is read from