JPH04355818A - Data input/output controller - Google Patents

Abstract

PURPOSE:To reduce the number of times in I/O issue and to increase the I/O processing speed by collectively issuing following read requests at the time of I/O issue and temporarily storing data in a buffer and successively transferring the data to read request sources thereafter. CONSTITUTION:I/O requesting blocks 6 are queued in a queue management table 5 in the order of intre-disk relative position, and the first I/O requesting (read request) block 6 is referred; and when the capacity of all I/O requesting (read request) blocks following this first I/O request block does not exceed the prescribed capacity of a buffer 4, they are revised to a collected I/O requesting (read request) block 6 and are issued, the data read from a disk device 9 is stored in the buffer 4, and the data from the buffer 4 is successively transferred to access requesting sources.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data input/output control device for controlling input/output to a disk device.

0002

[Prior Art] Conventionally, for magnetic disk devices, etc.
A device driver that inputs/outputs read/write requests by queuing them sorts I/O requests in order of relative position within the disk in order to reduce seek time, which is the time it takes for the magnetic head to move to the magnetic disk. I was trying to access them in this sorted order. A brief explanation will be given below using FIG. 8.

(1) As shown in FIG. 8(a), an I/O request (read request) block A is sent to the device driver.
, B, and C and are sorted and queued. (2) When I/O request (read request) block D is notified in the state of (1), as shown in (b) of FIG. 8, I/O request (read request) block B and I/O request(
The I/O request (read request) block D is sorted and queued between the read request) block C and the I/O request (read request) block D.

(3) In the queued state shown in FIG. 8(b), I/O requests (read requests) blocks A, B, D, and C are issued sequentially from the beginning to perform seek operations and rotation. The system waits, reads data from the magnetic disk, and transfers it to the access request source.

[0005]

[Problem to be Solved by the Invention] As described above, sorting by relative position within the disk and queuing in order, issuing I/O corresponding to a read request from the beginning, performing a seek operation and waiting for rotation. Because data was read from the magnetic disk and transferred to the access request source, I/O was issued sequentially even if queued read requests and write requests existed in adjacent or nearby blocks (sectors). , the number of I/O issues increases, and I/O can be performed quickly.
There was a problem that processing could not be performed.

[0006] The present invention issues subsequent read requests all at once when I/O is issued, and after storing the data in a buffer, sequentially transfers the data to the read request source, thereby reducing the number of I/O issuances and increasing I/O. The purpose is to speed up O processing.

[0007]

[Means for Solving the Problems] Means for solving the problems will be explained with reference to FIG. In FIG. 1, an I/O control unit 3 queues I/O request (read request/write request) blocks 6 from access request sources in a queue management table 5 in the order of their relative positions within the disk. (Read request) When the combined capacity of blocks 6 is within the predetermined capacity of the buffer 4, the combined I/O request (read request) is changed to block 6 and issued, and the data read from the disk device 9 is sent to the buffer. 4, and transfers data from this buffer 4 to the access request source.

The buffer 4 is a buffer of a predetermined capacity that temporarily stores data read from the disk device 9. The queue management table 5 is for queuing I/O request blocks 6.

[0009]

[Operation] As shown in FIG.
The I/O request block 6 from the access request source is queued in the queue management table 5 in the order of relative position within the disk, and the I/O request (read request) block 6 that follows is referred to and the subsequent I/O request block 6 is queued. When there is an O request (read request) block 6, and the combined capacity is within the predetermined capacity of the buffer 4, it is changed to the combined I/O request (read request) block 6, issued, and read from the disk device 9. The obtained data is stored in a buffer 4, and the data is sequentially transferred from this buffer 4 to the access request source. Also, while the data read from the disk device 9 by issuing the combined I/O request (read request) block 6 is stored in the buffer 4, newly enqueued I/O
When the data of the request (read request) block 6 exists in the buffer 4, the data is transferred from the buffer 4 to the original enqueued access request source and this new access request source. In addition, the I/O request blocks 6 from the access request sources are queued in the queue management table 5 in order of their relative positions within the disk, and the
The predetermined capacity of the buffer 4 is the combined capacity of the I/O request blocks 6 of these read requests when there is a subsequent write request and the next read request by referring to the request (read request). I/O requests collected when
Read request) block 6 is changed and issued, the data read from the disk device 9 is stored in buffer 4, and the data of I/O request (write request) block 6 is overwritten in buffer 4. The data is sequentially transferred to the access request source and an I/O request (write) block 6 is issued to write the data to the disk device 9.

[0010] Therefore, when issuing an I/O, subsequent read requests are issued all at once, data read from the disk device 9 is temporarily stored in the buffer 4, and then sequentially transferred to the read request source, and write requests are also mixed. At times, data read from the disk device 9 is stored in the buffer 4,
After overwriting the write request data, the data is sequentially transferred to the read request source and I/O is issued to write the data to the disk device 9, thereby reducing the number of I/Os issued to the disk device 9 during reading. Reduce I
It is possible to speed up /O processing.

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, the structure and operation of an embodiment of the present invention will be explained in detail using FIGS. 1 to 7. In FIG. 1, a main memory 1 is a memory that stores data read from a disk device 9 and loads programs.

The device driver 2 receives read/write requests from the access request source, issues an I/O request block 6 to the disk device 9, reads data, and transfers it to the access request source. It performs control such as writing data transferred from an access request source to the disk device 9, and is composed of an I/O control unit 3 and the like.

The I/O control unit 3 accepts read requests/write requests from access request sources, and queues them as I/O request (read request/write request) blocks 6 in the queue management table 5 in order of relative position within the disk. or
Queued I/O requests (read requests) I/O requests (read) that are grouped together when there are multiple blocks 6 and their combined capacity is within the predetermined capacity of the buffer 4
The data read from the disk device 9 by changing to the block 6 or issuing the I/O request block 6 is transferred to the buffer 4.
and transfer data from the buffer 4 to the access request source (see the specific examples in FIGS. 2 to 6 and the flowchart in FIG. 7).

The buffer 4 is a buffer that temporarily stores data read from the disk device 9. The queue management table 5 stores read requests/read requests from access request sources.
This is a table for queuing and managing write requests as an I/O request (read request/write request) block 6.

The I/O request (read request/write request) block 6 is a block in which read requests/write requests from access request sources are queued in the queue management table 5. The intra-disk relative position register 7 indicates the relative position within the disk apparatus 9 of the data read from the disk apparatus 9 and stored in the buffer 4 (in-disk relative position).
This is a register that stores .

The block number register 8 is stored in the disk device 9.
This register stores the number of blocks (for example, sectors) of data read from the buffer 4 and stored in the buffer 4. The disk device 9 is a storage device that reads and writes data by a seek operation, and is a randomly accessible storage device such as a magnetic disk device, floppy disk device, optical disk device, magneto-optical disk device, CD-ROM device, etc. It is a very good device.

Next, the operation of a specific example of the present invention will be explained in detail using FIGS. 2 to 6. 2 and 3 show illustrations of specific examples of the present invention. This shows a specific example where multiple read requests are combined into one. In (a) of FIG. 2, an I/O request (read request) is stored in the queue management table 5.
A state in which blocks A, B, and C are sorted and queued in the order of disk relative positions within the magnetic disk 10 is shown. Here, the I/O request (read request) of the first I/O request (read request) block A is
/O processing is in progress, that is, I/O is issued and magnetic disk 1
After reading data from 0 and temporarily storing it in the buffer 4, a process is being executed in which data is transferred to the main memory 1 and copied. Here, data is transferred directly from the disk to the main memory unless the request is combined with a subsequent I/O request block.

FIG. 2B shows a state in which I/O request (read request) block A has been completed, dequeued from the queue management table 5, and I/O processing has been completed. (c) in FIG. 2 shows the next I/O request (read request) block B and I/O request (read request).
O request (read request) block C is combined into one I
Change the /O request (read request) block (B+C) and issue it, and transfer data B from the magnetic disk 10 to the buffer 4.
This shows the state in which C is stored as shown. This is because when executing the I/O process for the first I/O request (read request) block B in the queue management table 5, the next I/O request (read request) block C was queued. , Comparing the capacities of both of these and the predetermined capacity of buffer 4, the former is smaller than the latter, so both are combined into one (B+C) I/O request (read request) block. The data B and C are read from the magnetic disk 10 and stored in the buffer 4 as shown in the figure, and the disk relative position (■) and block number (■) of the magnetic disk are stored in the disk relative position register 7.
Set.

FIG. 2(d) shows a state in which the data of I/O request (read request) block B is copied from the buffer 4 to the main memory 1 and dequeued from the queue management table 5. In (e) of FIG. 3, when executing the I/O process for the next I/O request (read request) block C, the in-disk relative position 7 and the block number register 8 are checked, and if valid data exists here, Since it is clear, data D is copied from buffer 4 to main memory 1, and queue management table 5 is
Indicates the dequeued state.

[0020] Through the above processing, when the I/O request (read request) blocks are consecutive, the I/O request (read request) block is
By changing the I/O request (read request) block and issuing it, storing data from the magnetic disk 10 in the buffer 4, and sequentially copying the data from this buffer 4 to the main memory 1, the number of I/O issues can be reduced. This makes it possible to quickly perform I/O processing.

FIG. 4 shows a diagram illustrating a specific example of the present invention. This shows a specific example where there is an additional read request during I/O processing in which multiple read requests are combined into one I/O issue. (A) in FIG. 4 shows one I/O request (read request) block (B+C
) and issue it from magnetic disk 10 to buffer 4.
While data B and C are being stored as shown in the figure, an I/O request (read request) block D between them is queued at a relative position within the disk. In this state, the intra-disk relative position register 7 and the block number register 8 store the intra-disk relative position (■) and the block number (■) illustrated on the magnetic disk 10.

FIG. 4B shows a state in which data corresponding to I/O request (read request) block B is copied from the buffer 4 to the main memory 1 and dequeued from the queue management table 5. (c) in FIG. 4 shows that when executing the next I/O request (read request) I/O processing of block D, the relative position 7 in the disk and the block number register 8 are checked,
Since it is determined that valid data exists here, data D is copied from buffer 4 to main memory, dequeued from queue management table 5, and similarly, the next I/O request (read request) I/O of block C is When executing the process, the disk relative position 7 and block number register 8 are checked, and it is found that valid data exists here, so data C is copied from the buffer 4 to the main memory and dequeued from the queue management table 5. shows.

[0023] Through the above processing, when I/O request (read request) blocks are consecutive, I/O
While changing the I/O request (read request) block to an O request (read request) block and issuing it, and storing data from the magnetic disk 10 in the buffer 4, the I/O request (read request) block existing between these blocks is enqueued. By sequentially copying data from buffer 4 to main memory 1 when
It becomes possible to reduce the number of O issuances and perform I/O processing quickly.

FIG. 5 shows a diagram illustrating a specific example of the present invention. This shows a specific example where a plurality of read requests and write requests coexist. FIG. 5A shows a state in which an I/O request (write request) block D is enqueued between an I/O request (read request) block B and an I/O request (read request) block C.

FIG. 5(B) is the first I of FIG. 5(B).
When /O request (read request) block B is issued, the capacity of the subsequent I/O request (read request) block C, ignoring subsequent write requests, is the predetermined capacity of buffer 4. Since it is smaller than the capacity, both of these (B+C) are changed into an I/O request (read request) block (B+C) and issued, and the data is read from the magnetic disk 10 and stored in the buffer 4 as shown in the figure. In addition, a state in which an intra-disk relative position (■) and a block number (■) are stored in the intra-disk relative position register 7 and the block number register 8 is shown.

FIG. 5C shows a state in which data corresponding to I/O request (read request) block B is copied from the buffer 4 to the main memory 1 and dequeued from the queue management table 5. In (d) of FIG. 5, when executing the I/O process for the next I/O request (write request) block D, the in-disk position register 7 and the block number register 8 are checked, and if valid data is present, I figured it out, so after copying and updating it to the same location in buffer 4, I/O
A state in which a request (write request) block D has been issued and data has been written to the magnetic disk 10 is shown. Here, since data D partially or completely overlaps with data C, data C to be copied to main memory 1 next will be updated with data D. If the data D overlaps the data B, the data D is overwritten and updated in the buffer 4 before copying the data from the buffer 4 to the main memory 1 in (c) of FIG.

FIG. 5(E) shows a state in which the I/O request (write request) is dequeued from the queue management table 5 since the I/O processing of the I/O request (write request) block D has been completed in FIG. 5(D). In (f) of FIG. 5, when executing the I/O processing of the next I/O request (read request) block C, the in-disk relative position register 7 and the block number register 8 are checked, and valid data exists here. Since this has been determined, the state is shown in which the data is copied from the buffer 4 to the main memory 1 and dequeued from the queue management table 5.

Through the above processing, when I/O request (read request) blocks and I/O request (write request) blocks coexist, the I/O request (which is a collection of read requests) is created.
A read request) block is issued, data is stored from the magnetic disk 10 in the buffer 4, data of a write request is overwritten in this buffer 4, and an I/O request (write request) is issued.
By issuing a block and writing it to the magnetic disk 10, and copying it from the buffer 4 to the main memory 1 of the read request source, the number of I/O issues and reads from the magnetic disk 10 can be reduced and I/O can be performed quickly. It becomes possible to perform processing.

Next, the operation of the configuration shown in FIG. 1 will be explained in detail in accordance with the order shown in the flowchart shown in FIG. In FIG. 7, S1 determines whether or not it is a read request. For example, it is determined whether the I/O request (read request) block A queued at the head of the queue management table 5 in FIG.
(identified as S). If YES, it is a read request, so the processes of S2 to S8 are performed. On the other hand, in the case of NO, since it is a write request, the processes of S9 and S10 are performed.

In step S2, it is determined whether read data exists in the buffer 4 or not. Specifically, this determination is made by comparing the disk relative position and block number of the I/O request (read) block with the respective values set in disk relative position register 7 and block number register 8. Then, it is determined whether the read data exists in the buffer 4 or not. If YES, magnetic disk 1
Since the data was already stored in buffer 4 after reading from 0, the data is copied from buffer 4 to main memory 1 in S7, this I/O request (read request) block is separated from the queue in S8, and the next Proceed to processing. On the other hand, NO
In this case, there is no read data in the buffer 4, so the process advances to S3.

[0031] S3 is a queue (I/O request) at the back.
Check whether the read request/write request block) is checked or not. If yes, proceed to S4. If there is no other I/O request (read request/write request) block queued in the queue management table 5, the first queue is executed (I/O processing is executed).

[0032] In S4, it is determined whether or not there is a read request. Y
In the case of ES, it is determined in S5 whether or not it can be stored in the buffer 4 (determined whether the number of blocks in the combined read request is less than the predetermined capacity of the buffer 4), and if YES, the read request is executed in S6. The process proceeds to S3, and when the answer is NO, the first queue is executed. On the other hand, in the case of NO, the process advances to S3.

By the above processing of YES in S1, NO in S2, YES in S3, YES in S4, YES in S5, and YES in S6, the I queued in the queue management table 5 is
/O request (read request) blocks are collected sequentially from the beginning and the capacity is within the predetermined capacity of buffer 4 as one I.
/O request (read request) block, and issues this combined leading I/O request (read request) block to read data from the magnetic disk 10 and store it in the buffer 4.

[0034] Also, YES in S1, YES in S2, S7
, the corresponding data is copied to the main memory 1 from the data stored in the buffer 4 from the magnetic disk 10 by issuing one combined I/O request (read request) block through the process of S8. do. This makes it possible to eliminate I/O issues and speed up I/O processing.

[0035] Also, NO in S1, YES in S9, S10
Through the processing, when multiple I/O request (read request) blocks and I/O request (write request) blocks are mixed, an I/O request (read) block is issued and transferred from the magnetic disk 10 to the buffer 4. After storing the data and overwriting the write data in the buffer 4 during I/O processing of the I/O request (write request) block, the I/O request (write request) block is issued and the magnetic disk 10 is
By writing to , it is possible to omit I/O issuance during I/O processing of the next grouped I/O request (read request) block, and the number of I/O issuances can be reduced.

Next, S11 executes the head queue (I/
After issuing an O request (read request) block to read data from the magnetic disk 10 and storing it in the buffer 4, the I/O request block is queued (queue management table 5).
). Check whether the separated I/O request block is a read request or a write request. S12 is S1
As a result of the check in step 1, it is determined whether or not it is a read request. If YES, it was a lead request, so
In S13, data is copied from buffer 4 to main memory 1,
Proceed to the next process. If NO, it is a write request, so the process proceeds to the next step without doing anything. Here, S
Between 12 and S13, it is determined from registers 7 and 8 whether the request is included in buffer 4, and if YES, the process goes to S13, and if NO, the next process is performed without executing S13. .

Next, the I/O processing times of the conventional and the present invention will be calculated for an actually common disk device. (1) Disk specifications: Average seek time: 20ms Average rotational wait time: 10ms Data transfer speed between disk and main memory: 1MB/s (2
) When a 4KB read request A from disk relative position 0 (512B relative) and a 4KB read request B from disk relative position 16 are continuously queued in the queue management table 5: (3) Conventional I/O processing time: I/O processing time for read request A = 20ms + 1
0ms+(4KB×1ms)
=34ms
I/O processing time for read request B = 10ms + (
4KB x 1ms) = 14ms Therefore, (read request A+
The I/O processing time for read request B) is 34ms+14ms=48ms.

(4) I/O processing time of the present invention: Read request A and read request B are issued together as one I/O, so (read request A + read request B) = 2
0ms + 10ms + (12KB x 1ms) = 42msHere, 12KB is calculated as the transfer time because 4KB worth of data from relative positions 8 to 16 in the disk is empty read, so this time is added. It is.

(5) Therefore, the conventional I/O processing time is 48
Subtract the I/O processing time of the present invention, 42 ms, from ms,
6ms faster I/O processing time, i.e. 12.5% I
/O processing time could be increased. When read requests occur frequently and the load on the disk increases, the I/O processing time can be further increased compared to the conventional method.

[0040]

[Effects of the Invention] As explained above, according to the present invention,
When issuing an I/O, subsequent read requests are issued all at once, the data read from the disk device 9 is temporarily stored in the buffer 4, and then sequentially transferred to the read request source, or when write requests are mixed, the data is read from the disk device 9. A configuration is adopted in which the data read from is stored in the buffer 4, and after overwriting the data of the write request, the data is sequentially transferred to the read request source and I/O is issued to write the data to the disk device 9. Therefore, the number of I/O issues to the disk device 9 during reading is reduced, and the number of I/O issues that occur with write requests when read requests and write requests are mixed is reduced, increasing the speed of I/O processing. can be achieved.

[Brief explanation of drawings]

FIG. 1 is a configuration diagram of one embodiment of the present invention.

FIG. 2 is a diagram illustrating a specific example of the present invention (at the time of a read request).

FIG. 3 is a diagram illustrating a specific example of the present invention (a continuation of FIG. 2).

FIG. 4 is a diagram illustrating a specific example of the present invention (when a read request is made during read execution).

FIG. 5 is a diagram illustrating a specific example of the present invention (in case of a mixture of read requests and write requests).

FIG. 6 is a diagram illustrating a specific example of the present invention (a continuation of FIG. 5).

FIG. 7 is a flowchart explaining the operation of the present invention.

FIG. 8 is an explanatory diagram of the prior art.

[Explanation of symbols]

1: Main memory 2: Device driver 3: I/O control section 4: Buffer 5: Queue management table 6: I/O request block 7: In-disk relative position register 8: Block number register 9: Disk device 10: Magnetic disk

Claims (3)

[Claims] Claim 1: In a data input/output control device that controls input/output to a disk device, an I/O request block (6
), and a buffer (4) with a predetermined capacity to temporarily store data read from the disk device (9). 6) in the queue management table (5) in the order of their relative positions within the disk, refer to the first I/O request (read request) block (6), and then process subsequent I/O requests (read requests). When there is a block (6) and the combined capacity of these is within the predetermined capacity of the buffer (4), the combined I/O request (read request) is changed to block (6) and issued, and the disk device The data read from (9) is transferred to the above buffer (
4) and sequentially transfer data from the buffer (4) to an access request source. 2. While the data read from the disk device (9) by issuing the summarized I/O request (read request) block (6) is stored in the buffer (4), a new enqueue is generated. I/O request (read request)
When block (6) is a read request for data existing in buffer (4), the data is transferred from this buffer (4) to the original enqueued access request source and this new access request source. 2. A data input/output control device according to claim 1, characterized in that said data input/output control device is configured as follows. 3. In a data input/output control device that controls input/output to a disk device, an I/O request block (6
), and a buffer (4) with a predetermined capacity to temporarily store data read from the disk device (9). 6) in the above queue management table (5) in the order of relative position within the disk, refer to the first I/O request (read request) block (6), and when there is a subsequent write request, further When there is a next read request, the capacity of the I/O request block (6) of these read requests will be the capacity of the buffer (4).
When the capacity is within the specified capacity, the I/O request (read request) block (6) is changed and issued.
) is stored in the buffer (4), and data from the I/O request (write request) block (6) is overwritten in the buffer (4).
4) Transfer data sequentially to the access request source and I/
A data input/output control device characterized in that it is configured to write data to a disk device (9) by issuing an O request (write) block (6).