JPH0773107A - Control method for disk system - Google Patents

Abstract

PURPOSE:To set data transfer to be efficient and to improve a cache hit rate by directly storing request data in a main storage at the time of sequential access and storing request data in a disk cache and the main storage at time except for sequential access. CONSTITUTION:At the time of sequential access, a host computer 10-side stores request data D transmitted from a disk device with look-ahead buffer 20 into the main storage 13 without storing it into the disk cache 14. At time except for sequential access, the host computer 10-side stores request data D transmitted from the disk device with look-ahead buffer 20 into the disk cache 14 and stores it into the main storage 13. Thus, data transfer can be set efficient and the cache hit rate improves.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of controlling a disk system having a disk cache on the host computer side and a read-ahead buffer in a disk device.

[0002]

2. Description of the Related Art FIG. 3 is a diagram showing the overall configuration of a computer system. In the figure, 10 is a host computer, 11 is a central processing unit, 12 is a DMA controller, 13 is a main memory, 14 is a disk cache, 20 is a disk device with a read-ahead buffer, 21 is a read-ahead buffer, and 22 is a disk device. Shows.

The host computer 10 has a central processing unit 11, a DMA controller 12, a main memory 13, a disk cache 14, and the like. In the host computer 10,
In addition, a display device, a keyboard, a printer, etc. are connected. Disk device 20 with prefetch buffer
Has a prefetch buffer 21 and a disk device 22.

The disk cache 14 exists between the disk device 20 with a read-ahead buffer and the main memory 13,
The structure is such that the data from the disk device with read-ahead buffer 20 can be held therein. Of course, the data from the disk device with read-ahead buffer 20 can be directly transferred to the main memory 13 without using the disk cache 14. These controls are performed by an instruction to the DMA controller 12 of the disk control program.

The DMA controller 12 has a plurality of channels (for example, 8 channels), one of which is used to access the main memory 13 and one of which accesses the disk cache 14. Used for. Data transfer between the DMA controller 12 and the disk device with read-ahead buffer 20 is performed, for example, via a SCSI interface.

FIG. 4 is a diagram showing the structure of the disk cache. The disk cache is composed of a cache memory, a cache management table, an access history table and the like. 1 MB cache memory
It has a size of about 16 MB to 16 MB and stores a partial copy of the data in the disc. One block has the same size as the disk device (256 bytes or 512 bytes).

In the cache management table, one item is several bytes (for example, 2 bytes), and corresponds to the cache memory on a one-to-one basis. Currently, data in any part of the disk is stored in the cache memory. It is a table showing whether or not it is held. The access history table is a table that holds past access history. Details will be described later.

FIG. 5 is a diagram for explaining the block storage position in the disk cache. A hash method is used as a method for determining the storage location of data in the cache memory and the location of the cache management table that has a one-to-one correspondence with the storage location. Specifically, the cache
Since the memory is about 1MB to 16MB, which is considerably smaller than that of the disk, only a part of the data in the disk can be stored. However, if the block position (address) of a disk is known, the cache position is uniquely determined. It is a method to say.

As an operation, for example, assuming that the cache memory is 1 MB and one block of the disk is 512 bytes, 1 MB (1048576 bytes)
÷ 512 bytes = 2048 blocks of data can be stored in the cache, and this number is called the number of entries. It is assumed that the cache is used as shown in FIG. In the figure, the numbers are disk blocks.
Indicates an address. That is, block 0, block 204
8 and blocks 4096, ... Are stored in the A position. Similarly, the block 1, the block 2049, ... Are stored in the position B, and the block 2, the block 2050, ... Are stored in the position C and are used. this is,
When the block X of the disk is given, it is a method in which X is divided by the number of entries (2048 in this case) to determine the number of stages from a.

That is, X / (number of entries) = Y remainder n
Then, this n is a number indicating the number of stages from a. 20 if the block address of the disk is 2050
50 ÷ 2048 = 1 remainder 2, and the block of address 2050 is stored in the position C of the second stage counting from a in the cache memory. When the address is 4096, 4096 ÷ 2048 = 2, which is almost 0, and the block of the address 4096 is stored in the position A of the 0th row counting from a of the cache memory. Such a method is called a hash method, and the storage location of the cache can be determined by a simple algorithm.

FIG. 6 is a diagram showing an example of the configuration of a disk device with a read-ahead buffer. In the figure, 20 is a disk device with a read-ahead buffer, 21 is a read-ahead buffer, 22
Indicates a disk device, and 23 indicates a controller. The prefetch buffer 21 has 256 KB to 512 KB.
It has a size of about KB. The controller 23
For example, it is composed of a microprocessor.

FIG. 7 is a diagram showing a control flow of the disk control program. First, the case of reading data from the disc will be described. When the software requests to read the disk data, the disk read start block address and the number of blocks are passed. Based on this, the disk control program searches the cache management table to check whether the corresponding data exists in the cache memory. If it exists, it is called a hit, and if it does not exist, it is called a mishit.

In the case of a hit, the data is transferred from the cache memory to the main memory and the disk device is not accessed. This makes the disk device look faster. In the case of a mishit, it is judged from the access history whether it is a sequential access (access to a sequential file, etc. corresponds to this), and in the case of a sequential access, the read including the read-ahead is added to the read-ahead buffer. The disk device is requested, and otherwise, only the requested read is requested to the disk device with the prefetch buffer.

The data transferred from the disk device with the read-ahead buffer is stored in the corresponding position of the disk cache, and at the same time, the requested data is transferred to the main memory. After that, the cache management table is updated (for the data newly transferred to the cache from the disk device with the prefetch buffer). This completes one-time activation from the software. By starting from the software after this,
If there is a request for the same data as before, or if there is a subsequent data request for sequential access, there is a high possibility of hitting the cache.

Next, the operation of writing data to the disk will be described. There is no hit / miss hit like a read when writing. When there is a write request (storage instruction) to the disk for the main memory data from the software, the write start block address and the number of blocks of the disk are passed as in the case of reading. Based on this,
The disk control program determines the storage position on the disk cache, transfers the data on the main memory to the disk cache, and at the same time performs the processing for writing the data to the disk device with the read-ahead buffer. When both transfers are completed, the corresponding item in the cache management table is updated. This completes one write activation from the software. After this, starting from the software,
If a read request for the same data as the one just written is made, there is a high possibility of hitting the disk cache.

FIG. 8 is a diagram for explaining the read-ahead operation of the conventional host-side disk cache. In the figure, indicates read including read-ahead from disk device to cache, transfer only request from disk cache to main memory, indicates read-ahead in disk cache, and indicates read-ahead in read-ahead buffer. ing.

Now, there is a disk data read request from the software, and the disk read start block
Address = X, block number = Y1, main memory address = Z
Suppose that was. The host computer 10 (actually the disk control program) checks whether or not the data specified by this read request exists in the disk cache 14, and if there is no data, it refers to the access history to determine whether it is a sequential access or not. To find out. In the case of sequential access, the read-ahead portion is set to Y2 and the start block
A read command having address = X and block number = Y1 + Y2 as parameters is sent to the disk device 20 with the prefetch buffer.

The disk device 20 with the read-ahead buffer is
When the read command is received, it is checked whether or not the block designated by the read command is in the prefetch buffer 21. In some cases, the (Y1 + Y2) block having the block address = X as the head is read from the prefetch buffer 21 and sent to the host computer 10. Host computer 1
0 (actually DMA controller 12) is a block
The (Y1 + Y2) block having the address = X as the head is written in the disk cache 14, and the Y1 block having the block address = X as the head is written in the area after the address Z of the main memory 13.

When the data specified by the read command does not exist in the read-ahead buffer 21, the disk device 20 with the read-ahead buffer stores a predetermined number (for example, 1000) of blocks including the data specified by the read command in the disk device 22. Read from the block and store it in the prefetch buffer 21 and at the beginning of the block address = X (Y1
+ Y2) Block is read and sent to the host computer 10. The subsequent operation is as described above.

FIG. 9 is a diagram showing the processing of the host computer when reading data from the disk device. When reading data from a disk device with a read-ahead buffer,
Three pieces of information are passed from the software: the read start block address of the disk, the number of blocks, and the memory address of the main memory. Of these, the information
It is passed to the disk device with the read-ahead buffer as a parameter of the read command. The corresponding data flows from the disk device with the read-ahead buffer to the host computer via the interface bus. At this time, the information is stored in the main memory by the DMA controller.

FIG. 10 is a diagram showing the structure of the prefetch buffer of the disk device with the prefetch buffer. The total size of the prefetch buffer is about 256 KB to 512 KB, and this is divided and held by the number of divisions called the number of segments. The number of segments is usually around 4 to 8, so 32 to 12 per segment
It will be about 8 KB.

This read-ahead buffer has a two-port structure, and has a port for reading from the disk medium and storing it in the read-ahead buffer, and a port for sending out from the interface bus to the outside, while reading data from the medium. It can be sent to the outside. In addition, it will be used by overwriting any one,
LRU (Least Recent Used)
) Is used to make the decision. In other words, the buffer used in the oldest (oldest) access is crushed and used, and this time the buffer is registered as the latest one, so it is possible to keep new data in the buffer as much as possible. is there.

The size of one read-ahead buffer is 32 KB.
Although it is about 128 KB, it is assumed that it is 32 KB. If the size of one block of the disk is 512 bytes, 32 KB (32768 bytes) ÷ 512 bytes = 64 blocks of data can be held in one prefetch buffer. Although not shown, there is a management table corresponding to each prefetch buffer, and each management table holds information indicating which block address data is stored in each block storage area of the corresponding prefetch buffer. There is.

FIGS. 11 and 12 are diagrams showing the processing of the disk device with the prefetch buffer at the time of reading. When a read command is sent from the host, the block address and the number of blocks are passed, so the controller in the disk device with the read-ahead buffer moves the head to the corresponding position on the disk medium (called seek). And
Find the target block and store the data in one of the prefetch buffers. At the same time, data is sent to the outside from the external sending port.

As described above, the corresponding data is transferred based on the read command from the host computer. For example, it is assumed that the requested block of the host computer is 10 blocks. Since one prefetch buffer (one segment) can hold 64 blocks of data, 54 blocks of data are continuously prefetched even after the data transfer to the host computer is completed, and the prefetch buffer It will be stored in. If the subsequent host access is the subsequent data, it hits the prefetch buffer.

When a new read command is received from the host during pre-reading, a check mark is entered at the last part marked with a star in the flow of FIG. 12, and if the requested data of the command is the subsequent data described above. The data that is being entered into the prefetch buffer is transferred to the host computer as it is, and if it is not the subsequent data, the prefetch operation is immediately stopped and a new read command is processed. This prevents performance deterioration.

FIG. 13 and FIG. 14 are views showing the sequential access judgment processing. In the control flow of FIG. 7, the access history is referred to determine whether the access is sequential access or not. This is called an “access history table” because the access history up to the past 16 times is stored. I have a table. When a read instruction is issued from software, 16 (4 bytes) of the start block address and the number of blocks are remembered, which is a 64-byte table. 13 and 14 show a flow of sequential access determination processing. By using the algorithm shown in the figure, whether the current access is a sequential access or not is determined by the ON / O of the SEQ flag.
It turns out by FF. Furthermore, this access status will be newly added to the table.

FIG. 15 is a diagram showing an example of transition of the access history table. FIG. 15 (a) shows the initial state. In the initial state, each item (row) in the access history table has F
FFFFFFF is stored. This value is set because FFFFFFFF cannot exist due to the maximum capacity of the disk. If set to 00000000, 0000
An access whose start block is 0000 is always mistakenly recognized as a sequential access.

When there is a read request from block 00000000 to block 10 in the access history table shown in FIG. 15 (a), the contents of the access history table change as shown in FIG. 15 (b). When the algorithm for determining sequential access is used, since 00000000 does not exist in the item of the table, it is determined not to be sequential access. The first access is always judged not to be sequential access,
The frequency is minimal as it is only the first time.

When there is a read request from the block 00000010 to the block 10 in the access history table shown in FIG. 15 (b), the contents of the access history table changes as shown in FIG. 15 (c). Since there is an item “00000010” in the items of the access history table, it is determined as the sequential access. Figure 15
From the table state of (b), the number of 00000010 is updated to be 00000010 + 10 = 00000020, which is as shown in FIG. 15 (c).

After that, it is assumed that there are various accesses and the contents of the access history table are as shown in FIG. 15 (d). When there is a read request for eight blocks from block 02222222 under the state of the access history table of FIG. 15 (d), the contents of the access history table changes as shown in FIG. 15 (e). Since there is an item called 00222222 in the items of the access history table, it is determined to be sequential access. The location of 00222222 is updated from the state of the table of FIG. 15 (d) so that 00222222 + 8 = 0022222A,
Moreover, the item becomes the top (latest part), and the old item shifts backward.

When there is a read request for two blocks from the block 0000ABCD under the condition of the access history table of FIG. 15 (e), the contents of the access history table changes as shown in FIG. 15 (f). Since there is no item called 0000ABCD in the items of the access history table, it is determined that the access is not sequential access. In the table item of FIG. 15 (e), 0000ABCD + 2 = 00
The item 00ABCF is added, the item becomes the head (latest position), and the old item shifts backward. At this time, the oldest 00000020 is evicted (deleted from the access history table).

In the conventional technique, i) the data of the block requested to be accessed is held in the disk cache. (Use the software's "data reuse" property) ii) Read ahead and retain the subsequent data of the block for which an access is requested. (Using the property called "locality of reference" of software), the disk device appears to operate at high speed.
However, recent disk devices have a read-ahead buffer in the device, and the function of ii) is meaningless.

[0034]

The above-mentioned conventional technique has the following problems. (a) Subsequent read requests are expected to hit the read-ahead data in the disk cache. At this time, the time to transfer the disk cache data to the main memory and the read-ahead buffer disk unit The time to transfer the data in the prefetch buffer to the main memory without passing through the cache on the host side is the same, and the effect of the disk cache is lost. (b) Not only that, because the old data in the disk cache is lost when the read-ahead data is transferred from the disk device with the read-ahead buffer, the disk cache is inefficiently used, resulting in an overall hit. The rate drops. That is, during sequential access, the performance does not change with or without the host-side cache.
Also, when the transfer speed from the disk device side becomes high,
A performance reversal phenomenon occurs in which it is slower when the disk cache is added.

Next, consider the case where the sequential access is not performed. (1) The buffer on the disk device side is mainly designed for prefetching. (2) (Host side cache capacity)> (Disk side prefetch buffer capacity). (3) When accessing the disk, use the read-ahead buffer of the disk device with the read-ahead buffer in any case. For this reason, it can be said that the prefetch buffer has low data retention. That is, in cases other than the sequential access, the effect of the prefetch buffer on the disk device side as a cache cannot be expected. In the case of sequential access, data used once is not used again in many cases, so there is no problem even if the data retention is low.

The present invention was created in view of this point, and provides a disk system control method capable of efficiently performing data transfer and improving the cache hit rate. It is an object.

[0037]

FIG. 1 is a diagram for explaining the principle of the present invention. A method of controlling a disk system according to claim 1, wherein a main memory (13), a disk cache (14), a disk device (20) with a read-ahead buffer having a read-ahead buffer (21) and a disk device (22). A method of controlling a disk system comprising: a disk cache (14) for requesting data sent from a disk device (20) with a read-ahead buffer during sequential access.
The request data sent from the disk device (20) with a read-ahead buffer is stored in the main memory (13) without being stored in
It is characterized by being stored in the main memory (13) as well as being stored in the cache (14).

[0038]

The operation during the sequential access will be described with reference to FIG. When receiving the read request from the host computer, the disk device with read-ahead buffer 20 reads the request data D and the subsequent read-ahead data D ′ from the disk device 22, and reads the request data D and the read-ahead data D ′. The request data D is stored in the prefetch buffer 21 and transferred to the host computer side. The host computer side stores the received request data D in the main memory 13 without storing the request data D in the disk cache 14 at the time of sequential access.

The operation other than the sequential access will be described with reference to FIG. When receiving the read request from the host computer, the disk device with read-ahead buffer 20 receives the request data D from the disk 22,
The subsequent read-ahead data D ′ is read, the request data D and the read-ahead data D ′ are stored in the read-ahead buffer 21, and the request data D is transferred to the host computer side. The host computer stores the request data D in the disk cache 143 and the request data D in the main memory 13 except for the sequential access.

According to the present invention, the data is not stored in the disk cache at the time of sequential access, and only the requested amount is stored in the disk cache at the time of non-sequential access. Therefore, the old data in the disk cache is lost. Can be minimized and the data in the disk cache can be
Does not disappear even when accessed. If the same data is requested again after the disk is accessed several times, a hit can be expected from this disk cache.
If there is no such disk cache and the cache is configured only by the read-ahead buffer in the disk, if the same data is requested again after several times of disk access, several times of disk access will be performed. You can't expect a hit because the data will be lost.

[0041]

FIG. 2 is a diagram showing an operation flow in the present invention. The hardware configuration for implementing the present invention is the same as the conventional example. The difference between the present invention and the conventional example is only a part of the disk control program. That is, in FIG. 7, the process of "update relevant access history"->"start disk with prefetched data stored in cache"->"update cache management table"->"transfer main memory from request only cache" Although there is a flow, the present invention is a modification of this process flow. In step S1, a read request from software is received. In step S2, the past several access histories are referred to. In step S3,
Check whether it is access. If YES, step S4
If NO, proceed to step S7.

In step S4, the corresponding access history is updated. In step S5, read activation is sent to the disc. In this case, the host side disk cache is not used. In step S6, the data from the disk is transferred to the main memory as it is.

In step S7, the access history is added. In step S8, read activation is sent to the disc.
In step S9, only the request from the software is stored in the cache. In step S10, data is transferred from the cache to the main memory.

[0044]

As is apparent from the above description, according to the present invention, the cache hit rate can be improved and
It becomes possible to transfer data efficiently.

[Brief description of drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is a diagram showing an operation flow in the present invention.

FIG. 3 is a diagram showing an entire computer system.

FIG. 4 is a diagram showing a configuration of a disk cache.

FIG. 5 is a diagram for explaining a block storage position in a disk cache.

FIG. 6 is a diagram showing a configuration of a disk device with a prefetch buffer.

FIG. 7 is a diagram showing a control flow of a disk control program.

FIG. 8 is a diagram illustrating a read-ahead operation of a conventional host-side disk cache.

FIG. 9 is a diagram showing processing of the host computer when reading data from the disk device.

FIG. 10 is a diagram showing the structure of a read-ahead buffer in the disk device.

FIG. 11 is a diagram showing processing of the disk device at the time of reading.

FIG. 12 is a diagram showing processing (continuation) of the disk device at the time of reading.

FIG. 13 is a diagram showing a sequential access determination process.

[Figure 14] Sequential access judgment processing (continued)
FIG.

FIG. 15 is a diagram showing an example of transition of an access history table.

[Explanation of symbols]

 10 host computer 11 central processing unit 12 DMA controller 13 main memory 14 disk cache 20 disk device with read-ahead buffer 21 read-ahead buffer 22 disk device 23 controller

Claims (1)

[Claims] 1. A disk system comprising a main memory (13), a disk cache (14), a disk device (20) with a read-ahead buffer having a read-ahead buffer (21) and a disk device (22). A control method, in the case of sequential access, the main memory (13) does not store the request data sent from the disk unit (20) with a read-ahead buffer in the disk cache (14).
The request data sent from the disk device (20) with read-ahead buffer is stored in the disk cache (14) and stored in the main memory (1
A method for controlling a disk system characterized by storing in 3).