JP2001188658A - Disk control system and data rearranging method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To remarkably enhance the reading performance by efficiently rearranging data in a disk control system in a log structuring writing system. SOLUTION: When a read request is generated to plural blocks (virtual addresses V2, V3) with continuous virtual addresses, physical addresses (p5, p81) of the blocks are calculated from an address mapping table 171 and read from a disk device 18. The blocks (data D5, D81) read from the physical addresses are transferred to a buffer 201 at a read requesting origin and written back in a bulk writing buffer 171. When a free space is not available in the bulk writing buffer 171, the contents of the bulk writing buffer 171 is written in the disk device 18 and pieces of data with continuous virtual addresses are stored in the disk device 18 again so that their physical addresses also becomes continuous.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a disk control system and a data relocation method, and more particularly to a disk control system using a log structured writing method and a data relocation method applied to the system.

[0002]

2. Description of the Related Art In recent years, RAID (Redundant Array) has been developed from the viewpoint of speeding up disks and improving reliability.
2. Description of the Related Art A computer system equipped with a technology of Independent Disks is mainly used. RAID
There are 0, 1, 10, and 5 types of technologies.

Recently, a log structured file system (LFS) technique has been considered as a technique for greatly improving such RAID write performance. The log-structured file system technology is a technique for improving performance utilizing the characteristics of a disk that "sequential access is much faster than random access". The performance degradation at the time of random access is caused by the seek and rotation wait of the disk, and is a part that should be eliminated if possible. Therefore, in the log-structured file system technology, a method of using a buffer for storing write data, converting the write data of a plurality of small blocks into one large chunk of data, and performing "collective writing". Used. Thus, random writing can be changed to sequential writing, and writing performance can be improved.

[0004]

However, when the log structured file system technique is used, data is arranged and written in the order of write requests. Therefore, when a random write occurs, data whose logical addresses are separated from each other are continuously written in one physical area. This results in continuous areas being written to physically distant areas on the file system. Sequential reading for such an area becomes random reading and is inefficient.

In order to correct such a state, it is necessary to periodically rearrange the data so that the data is physically continuously arranged. However, since this data rearrangement process is performed separately from the normal write and read processes, many I / O processes are required only for the purpose of rearrangement, and as a result, the system Performance degradation will be caused.

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and enables data to be rearranged efficiently by using read requested data or write requested data, and achieves low read performance and low load. It is an object of the present invention to provide a disk control system and a data relocation method that can achieve a significant improvement.

[0007]

In order to solve the above-mentioned problems, the present invention accumulates write data in a buffer, and stores a plurality of blocks of the write data accumulated in the buffer in a continuous storage area of a disk drive. A disk control system of a log-structured write system, which collectively writes data to the disk device in stripes composed of a plurality of virtual addresses, when a read request is issued to a plurality of blocks having continuous virtual addresses, And a control unit for writing back the plurality of blocks to the disk device so that physical addresses are consecutive or stored in the same stripe. .

In this disk control system, blocks having continuous virtual addresses can be stored in the disk device so that physical addresses are also continuous, so that read performance can be greatly improved. . In particular, since a block requested to be read is targeted, it is not necessary to read the block for this purpose, and the load for data relocation can be reduced.

Writing back to the disk device can be automatically performed later simply by writing a plurality of blocks for which reading has been requested as write data in the collective writing buffer. Of course, a buffer dedicated to writing back may be prepared, and writing back may be performed using the buffer.

Further, it is possible to improve the reading speed by rewriting a plurality of blocks having continuous virtual addresses so that they are stored in the same stripe without necessarily writing back such that physical addresses are continuous. Can be.

The write-back process is performed on the condition that the block requested to be read is not stored in the disk device in a state where the physical addresses are continuous or is not stored in the same stripe. By doing so, it is possible to prevent the occurrence of a problem due to execution of useless processing.

It is preferable to determine whether or not to perform a write-back process based on the number of valid blocks in a stripe in which a block requested to be read is stored. In this case, for example, by preventing writing back of blocks included in a stripe in which all blocks are valid, it is possible to effectively prevent an increase in the number of stripes including invalid blocks.

Further, if write-back processing is performed when the disk access frequently occurs, the performance may be degraded, so that the load situation regarding the disk device is determined, and the write-back is performed based on the load state. It is preferable to determine whether or not to perform the processing.

Further, according to the present invention, write data is stored in a buffer, and a plurality of blocks of write data stored in the buffer are collected in the disk device in stripe units formed of a continuous storage area of the disk device. This is a disk control system of a log structured writing system for writing, wherein when a write request occurs, the block and the virtual address for which the write request is continuous are 1
And a control unit that reads one or more blocks from the disk device and writes the read blocks together with the blocks requested to be written into the buffer.

Also in this disk control system, by writing blocks having continuous virtual addresses to a buffer, they can be written to the disk so that physical addresses are also continuous or stored in the same stripe. it can. Further, since the block for which the write request has been made is used as it is, the number of blocks to be read for the purpose of rearrangement can be reduced.

Further, according to the present invention, write data is stored in a buffer, and a plurality of blocks of write data stored in the buffer are collected in the disk device in stripe units formed of a continuous storage area of the disk device. A disk control system of a log structured write system for writing, wherein a defragmentation buffer for temporarily holding a copy of a block requested to be written or a block of a block requested to be read and a read request or a write request are generated Means for writing a copy of the requested block to the defragmentation buffer; and a block in which the virtual address in the defragmentation buffer is continuous so that the physical address is also continuous or stored in the same stripe. Control means for writing back to the disk device. The features.

In this disk control system, a buffer for defragmentation is used to write back a block in which virtual addresses are continuous to a disk so that physical addresses are also continuous or stored in the same stripe. It is provided as the write-back buffer described above. When a read or write request occurs, the requested block is written to the defragmentation buffer. Thereafter, a process of writing back to the disk device so that the physical addresses of the blocks in the defragmentation buffer where the virtual addresses are continuous is also continuous or stored in the same stripe. On the defragmentation buffer,
Since a block requested to be read or written a plurality of times can be held, it is possible to write back more blocks so that physical addresses are continuous. Since the defragmentation buffer can be used as a cache, the read performance can be further improved.

Further, when a read request is issued, the prefetch of the block whose virtual address is continuous with the read requested block is executed, and the prefetched block is stored in the defragmentation buffer together with the read requested block. By providing the read-ahead means for writing, it is possible to write back more blocks so that the physical addresses are continuous. In addition, the effect of read-ahead can be expected with respect to speeding up the read.

When a write request occurs, a plurality of blocks whose virtual addresses are continuous with the write-requested block are read from the disk device, and the read block is read out together with the write-requested block by the defragmenter. By providing a means for writing data in the buffer for use, more blocks can be written back so that physical addresses are continuous.

[0020]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows a configuration of a computer system to which a disk control system according to an embodiment of the present invention is applied. This computer system is a server (PC
Server 1) and a plurality of CPUs 1
1 can be mounted. Each of these CPUs 11 is connected to the bridge 12 via the processor bus 1 as shown. Bridge 12 is connected to processor bus 1 and PCI.
This is a bridge LSI for connecting the bus 2 bidirectionally, and also has a built-in memory controller for controlling the main memory 13. The main memory 13 is loaded with an operating system, an application program to be executed, a driver, and the like.

As shown, the PCI bus 2 has an RAI
A D controller 16 and a RAID booster card (RAID BOOSTER) 17 are connected.
The disk device 18 controlled by the RAID controller 16 is a disk array composed of a plurality of disk drives, and is used for recording various user data and the like.

The disk device 18 functions as a disk array having a RAID 5 configuration under the control of the RAID controller 16. This disk array has N + 1 disks (here, DISK0 to DISK) obtained by adding one disk device for parity storage to N disk devices for data storage.
K4 disk drives). These N + 1 disk drives are grouped and used as a single logical disk drive.

As shown in the figure, a stripe (parity group) composed of data (DATA) and its parity (PARITY) is assigned to the disk unit 18, and the parity position is N + 1 disk units for each stripe. Moved to That is, in the physical stripe S0, data (DA) on the stripe unit allocated to the same position of each of DISK0 to DISK3.
The parity (PARITY) of the TA) group is recorded on the corresponding stripe unit of DISK4. In the next physical stripe S1, the parity (PARITY) corresponding to the data of the stripe S1 is DISK3.
Is recorded on the corresponding stripe unit. By dispersing the parity in the stripe units among the N + 1 disk devices in this way, it is possible to prevent the concentration of accesses to the parity disk.

RAID booster card (RAID BO
The OSTER 17 is for improving the write performance to the disk device 18 by using the log structured file system technology, and is realized as a PCI expansion card that can be inserted into a PCI slot. Here, referring to FIG. 2, a RAID booster card (RAID BOOSTER) 17 and an RAI incorporated and used in an operating system (OS)
The principle of write control by the D booster card control driver will be described.

In the writing method using the log structured file system technology, the data writing position is not determined according to the logical address requested by the host to write, but by storing the writing data in the order requested by the host. A large data block composed of a plurality of blocks of write data is formed, and the large data block is collectively written in a free area of the disk device 18 in a batch. The unit of “collective writing” is a stripe. In other words, a stripe is generated each time “collective writing” is performed, and a plurality of blocks of write data are sequentially written therein. As a result, random access can be converted to sequential access.

FIG. 2 shows that the block data size of write data from the application program sent via the file system is 2 KB, the data size of one stripe unit is 64 KB, and the data size of one stripe (parity group) is 256 KB ( = 64 KB
・ This is an example of the case of 4). Basically, 256KB of data is stored in a RAID booster card (RAID BOOS
When the data is stored in the TER 17, writing to the disk device 18 is performed at once.

Even when the disk device 18 is composed of a single disk drive, random writing can be performed sequentially by setting a stripe as a unit of "collective writing" to one cylinder. Can be converted to light.

RAID booster card (RAID BO
OSTER) 17 is a controller for the above-mentioned "collective writing" control, and, as shown in FIG. 1, has a log memory (collective writing buffer) 171 and an address mapping table 172. ing.

The collective write buffer 171 is for storing write data from the application. When the write data for one physical stripe is stored in the collective write buffer 171, the collective write buffer 171 collectively writes data to the disk array 18. Is performed.

The address mapping table 172 is used to manage the correspondence between the logical address of the write data and the actual write position on the disk array 18. The management of addresses by the address mapping table 172 is performed in data block units of a predetermined size. That is, the virtual address (logical block number) and the corresponding physical address (physical block number) are managed by the address mapping table 172 for each block of the data requested to be written.

Here, a basic read / write operation using the RAID booster card 17 will be described.

When a read request is issued, the following operation is performed.

1) The read request is decomposed into blocks, which are data management units.

2) The physical address is obtained by searching the address mapping table 172 for each block.

3) Issue a read request to the disk for each block.

When a write request is issued, the following operation is performed.

1) The write request is decomposed into blocks, which are data management units.

2) Batch write buffer 17 for each block
Write to 1.

3) When there is no more free space in the batch write buffer 171, the contents are written to a continuous area of the disk. Further, the contents of the address mapping table 171 are updated.

In this writing method, random writes are collectively written in one continuous area. Therefore, in a randomly written area, a result in which blocks having consecutive virtual addresses are included in physically separated (separate) stripes is obtained. Become. Sequential reading for such an area is physically random reading and is inefficient. In order to correct such a state, it is necessary to rearrange the data so that the data is physically continuously arranged. This processing will be referred to herein as relocation processing.

In the rearrangement processing according to the present embodiment, the data requested to be read or the data requested to be written is used as it is for rearranging the data. Hereinafter, a specific description will be given.

(Relocation Processing at Read Time # 1) First, FIG.
The operation when a read request occurs will be described with reference to the flowchart of FIG.

1) The read request is decomposed into blocks, which are data management units (step S101). 2) Address mapping table 1 for each block
72 to determine a physical address (step S10).
2). 3) A read request is issued to the disk device 18 for each block (step S103). 4) If the read request can be divided into a plurality of blocks in 1), the data read from the disk device 18 is written back to the collective write buffer 171. At this time, writing is performed so that the virtual addresses of each block are continuous (step S104).

As a result, when the contents of the collective writing buffer 171 are written to the disk device 18, the block read this time and having continuous virtual addresses is also stored with the physical addresses continuously.

For example, FIG. 4 to FIG. 6 show a case where a single read request is issued to virtual addresses V2 to V3.
This will be described with reference to FIG. In the state of FIG. 4, virtual addresses V2 to V
When a read request is issued to the device No. 3, the process is performed as follows.

1) The read request is decomposed into two blocks, V2 and V3 (Vi: virtual address). 2) From the address mapping table 171, it can be seen that V2 is stored in the physical address p5 and V3 is stored in the physical address p81 (pi: physical address). 3) A request to read data stored in p5 and p81 is issued to the disk device 18. 4) As shown in FIG. 5, the read results of p5 and p81 are written in the buffer 201 that has issued the read request, and also written back to the collective write buffer 171. At this time, the virtual addresses of each block are made continuous. 5) In FIG. 6, since the data D100 and D101 of V100 and V101 have been written and there is no free space, the contents of the collective write buffer
8 is written. Also, the contents of the address mapping table 172 are changed. This allows V2 and V3
Is stored in the area where the physical addresses are continuous.

As described above, when a read request is issued for a plurality of blocks having continuous virtual addresses, the requested plurality of blocks are transferred to the read request source, and the disk device 1
8 so that the data having the continuous virtual address is changed so that the physical address is also continuous.
8 can be stored again. Also, there is no need to read data from the disk just for that purpose.

If data having continuous virtual addresses is not continuous at physical addresses, it is highly likely that the data is not stored on the same cylinder. In this case,
When read access is requested for a plurality of blocks, disk access becomes random access, and seek and rotation wait are required every time one block is read. For example, in the state of FIG. 4, if a read request is issued to the blocks V2 and V3, p5 and p81 are read. If these are stored in different cylinders, two seek times and a rotation waiting time are required to read p5 and p81.

On the other hand, if the physical addresses are consecutive,
It is more likely to be stored on the same cylinder of the disk. When performing read access to a plurality of blocks stored on the same cylinder, the number of seeks and rotation waits may be reduced. After the state shown in FIG. 6 according to the present invention, when a read request for the blocks V2 and V3 occurs, it is only necessary to read p100 and p101 continuously (sequential read), one seek time and rotation wait. It takes time.

In the disk device 18, assuming that the amount of data to be transferred is small, the operating time of the mechanical device such as the seek time and the rotation waiting time occupies a high ratio in one access time. Taking a disk device as an example, 2KByte
Data read time is 130
μs, the seek average time is 5.2 ms,
The average rotation waiting time is 2.99 ms. Therefore, the effect obtained by changing the state of FIG. 4 to the state of FIG. 6 according to the present invention is great.

(Relocation at Read # 2) FIG. 7 is a flowchart showing a second example of the relocation process by rewriting read data.

In this example, the following step S105 is performed instead of step S104 in FIG.

That is, step S104 described above was as follows. 4) If the read request can be divided into a plurality of blocks in 1), the data read from the disk is written back to the collective write buffer 171. At this time, writing is performed so that the virtual addresses of each block are continuous (step S104). This is done as follows.

4) If the read request can be divided into a plurality of blocks in 1), the data read from the disk is written back to the collective write buffer 171. At this time, each block is written back in the same stripe (step S105).

Here, a case where a single read request is made to the virtual addresses V2 to V3 will be specifically described by way of example.

Data D5 and D of virtual addresses V2 and V3
Assume that 81 has been written back to the collective write buffer 171 as shown in FIG. This situation is, for example, that the data D81 is first read from the disk device 18 and the data D1000 of the V1000 is read before the reading of the data D5 is completed.
It occurs when 1000 is written.

Thereafter, the contents of the collective write buffer 171 are written to the disk device 18 as shown in FIG. 9 because the free space is exhausted, and the contents of the address mapping table 172 are also changed.

In the case of FIG. 9, unlike the case of FIG.
And V3 are not stored in the area where the physical addresses are continuous. However, even in this case, when a similar read request occurs, there is a possibility that the read access time can be shortened as compared with the state of FIG.

A) Case of storage in different disk drives For example, consider the case where the disk device 18 of FIG. 5 is a disk array composed of four disk drives 0 to 3. Each of the disk drives 0 to 3 stores a surplus block of (physical address / disk drive number). At this time, data D5 and D5 of V2 and V3
81 is stored in another disk drive. Therefore, by using the striping technique at the time of these read requests, there is a possibility that the disk access time can be reduced as compared with the state of FIG.

B) Case in which D81 and D5 are stored in the same disk drive of a disk array composed of a plurality of disk drives. Generally, a stripe element of a disk array (a continuous area composed of a plurality of blocks on one disk drive; a stripe is formed by a set of corresponding stripe elements of a plurality of disk drives) is the same for speeding up access. It is arranged on a cylinder. Therefore, D81 and D5 are also arranged on the same cylinder. If a read request for V2 and V3 occurs in this state, there is a possibility that a disk read can be performed with a single seek and rotation wait by the elevator sort technique. As a result, there is a possibility that the disk access time can be reduced.

In both cases a) and b), if the state is not a) or b) before write-back in the present invention, the access time may be improved by the present invention.

(Application to Relocation Processes # 1 and # 2 During Read) Next, an example of application to relocation processes # 1 and # 2 during read will be described. In the relocation processes # 1 and # 2 at the time of reading, the read data is written to the collective write buffer 171. In this example, the following conditions are set when performing this writing.

Condition a) If the physical address of the block requested to be read is not continuously stored, the block is written to the collective write buffer 171.

That is, as shown in the flowchart of FIG. 10, it is determined using the address mapping table 172 whether or not the read-requested block also stores the physical address continuously (step S).
111). Then, only when the physical addresses are not stored consecutively, write back to the collective write buffer 171 is performed (step S112).

For example, in the state shown in FIG. 6, when there is a request for virtual addresses V2 to V3 in one read request, the data is not written in the collective write buffer 171. As a result, only when the disk access time can be reduced, the relocation processes # 1 and # 2 at the time of reading can be performed. In particular, when the load on the disk is high, the relocation processes # 1 and # 2 need to be performed only when a sufficient effect is obtained, and the above condition a) can be used as a selection criterion.

Condition b) If the block requested to be read is not stored in the same stripe, write to the collective write buffer 171.

That is, as shown in the flowchart of FIG. 11, the physical address of the block requested to be read is searched using the address mapping table 172, and the storage position of each block on the disk device is determined from the result. A check is made to determine whether or not they are within the same stripe (step S113). Then, only when the data is not stored in the same stripe, write back to the collective write buffer 171 is performed (step S114).

For example, in the state shown in FIG. 9, if there is a request for virtual addresses V2 to V3 in one read request, these are stored in the same stripe.
It does not write to the batch write buffer 171.

As a result, only when the disk access time can be reduced, read rearrangement processes # 1 and # 2
It can be performed.

Condition c) Whether to write or not to write is determined based on the number of valid blocks in the stripe in which the block requested to be read is stored.

The block requested to be read is subjected to the batch write buffer 17 by the rearrangement processes # 1 and # 2 at the time of reading.
When it is written to 1, the area of the block stored until then becomes invalid. For example, p5 and p81 in FIG.
Block becomes invalid. The area of such a block cannot be used as a free stripe for a new batch write until all blocks in the stripe containing the block become invalid.

For example, if p80 to p83 are the same stripe, p81 cannot be used until p80, p82 and p83 become invalid. Further, the utilization rate of this stripe is 3/4. As the number of stripes including an invalid block with a low utilization rate increases, the number of empty stripes for performing a new write decreases, and eventually, the number of empty stripes disappears. As a result, RAID BOO
In STER17, a plurality (M) including invalid blocks
1 or more effective blocks in the stripe (N, N <N
A process (packing process) for creating an empty stripe by packing into the M) stripes is required. This repacking process is a process that should be avoided as much as possible because it involves disk access.

Therefore, before writing back to the collective write buffer 171, first, as shown in the flowchart of FIG. 12, the number of effective blocks of the stripe storing the block requested to be read is checked ( Step S115). This can be done by searching the address mapping table 172 from the physical address. For example, the quotient obtained by dividing the value of the physical address of each block (valid block) managed by the address mapping table 172 by the number of blocks constituting one stripe is obtained, thereby obtaining the physical address in which each block is stored. Calculate the stripe number. Then, the number of valid blocks can be calculated by checking the number of blocks that match the physical stripe number in which the read requested block is stored.

Of course, a dedicated table for managing the value of the number of valid blocks is prepared for each stripe, and the table is updated every time writing is performed so that the latest number of valid blocks can always be obtained. It may be.

Then, it is determined whether or not the number of effective blocks is a stripe having a predetermined value X or more (step S11).
6) Only when the number of valid blocks is a stripe smaller than the predetermined value X, the blocks read from the stripe are written back to the collective write buffer 171 (step S117). As a result, for example, it is possible to set a condition such as not to write back a block included in a stripe in which all blocks are valid, and it is possible to suppress the occurrence frequency of the repacking process.

Condition d) Write according to the load status of the system.

A means is provided for observing conditions such as CPU load, disk read load, disk write load, disk read / write load, and the like. If these are equal to or more than a certain value, the rearrangement processes # 1 and # 2 at the time of reading are not performed.

That is, as shown in the flowchart of FIG. 13, the status of the system load is determined based on, for example, whether or not the load of disk read / write is high (step S118). The write-back of the read data to the write buffer 171 is executed (Step S119). If the relocation processes # 1 and # 2 are performed during frequent disk accesses, the performance may be degraded. However, by applying the above condition d), this problem can be avoided. .

In the relocation processes # 1 and # 2, the read-requested data is written to the collective write buffer 171. However, a buffer dedicated to writing back to the disk device 18 is provided, and the disk device is connected via the buffer. The write-back to 18 may be performed.

(Relocation Process at Write Time # 3) Next, FIG.
The relocation process # 3 at the time of writing will be described with reference to the flowchart of FIG. When a write request occurs,
It works as follows. 1) The write request is decomposed into blocks, which are data management units (step S201). 2) Write each block to the batch write buffer 171 (step S202). 3) Issue a disk read to read one or more blocks whose virtual addresses are continuous with the block for which the write request was made (step S203). 4) A block in which the virtual address is continuous and which is written by the write request and a block in which the read is requested in 3) are
Batch write buffer 17 so that physical addresses are continuous
Write 1 (step S204).

As a result, when the contents of the collective write buffer 171 are written to the disk device 18, in addition to the block for which the write request has been made this time, the block in which the newly read virtual address is continuous has the physical address. It will be stored continuously.

For example, a case where there is a request for virtual addresses V2 to V3 in one write request will be described with reference to FIGS. The virtual address V in the state of FIG.
When there is a write request for 2 to V3, the processing is performed as follows.

1) Decompose a write request into two blocks, V2 and V3. 2) Write the contents of the write request source buffer 301 to the collective write buffer 171. 3) The physical addresses of the V4 and V5 blocks, which are continuous with the virtual address of the data requested to be written, are found from the address mapping table 171. V4, V5
Are p3 and p83, respectively, and issue their disk reads.

4) D1000, D1001 and D3, D8
3 is written to the collective write buffer 171 so that the physical addresses are continuous as shown in FIG. 5) The contents of the collective write buffer 171 are written to the disk device 18 as shown in FIG. Also, the contents of the address mapping table 171 are changed. As a result, V2 to V
4 is stored in an area where the physical addresses are continuous.

According to this processing, data with continuous virtual addresses can be stored again on the disk so that physical addresses are also continuous. Further, for that purpose, data for which a write request has been made is used, so that it is not necessary to read all data from the disk. The advantage of storing data having continuous virtual addresses on a disk so that physical addresses are also continuous is as described above.

Next, relocation processing using write data #
An example of application to No. 1 will be described. a) In the above-described example, the data is collectively written into the buffer 171.
When writing back, the physical address is written so as to be continuous, but relocation processing using read data # 2
Similarly to the above case, each block may be written back in the same stripe.

B) It is determined under the following conditions whether or not to issue a disk read for reading one or more blocks in which a write request and a virtual address are continuous.

B-a) When a write request is issued, attention is paid to a block in which a block to be written and a virtual address are continuous. A disk read is issued when the physical address of the block is not consecutively stored with another block having a virtual address continuous with the block. That is, by using the address mapping table 172, one or more blocks in which the write-requested block and the virtual address are continuous are checked, and it is determined whether the checked block stores the physical address continuously. Then, a disk read is issued only when the physical address is not stored continuously.

Bb) When a write request is issued, attention is paid to a block in which a write address and a virtual address are continuous. If the block is not stored on the same stripe as another block whose virtual address is continuous with the block, a disk read is issued. In other words, using the address mapping table 172, the block where the write request is made and the virtual address
The above blocks are checked to determine whether the checked blocks are stored in the same stripe. Then, a disk read is issued only when the data is not stored in the same stripe.

Bc) When a write request is generated, pay attention to a block in which the write request and the virtual address are continuous.
A disk read is issued according to the number of effective blocks in a stripe in which blocks having consecutive virtual addresses are stored. That is, the address mapping table 1
Using 72, a block in which the write-requested block and the virtual address are continuous is checked, and the number of valid blocks in the stripe in which the checked block is stored is determined. Then, for example, if all the blocks are valid stripes, no disk read is issued.

Bd) Issue a disk read according to the status of the system load.

That is, the state of the system load is determined based on, for example, whether the load of the disk read / write is high, and the disk read is issued only when the load is small.

Under these conditions of ba) to bd),
The same effect as in the case of each condition described in the rearrangement processing using read data can be obtained.

When it is necessary to write the data requested to be written to the disk, it is preferable to write back only the read data of the block for which the read request has been completed to the disk device 18. In other words, by the time writing from the collective write buffer 171 to the disk device 18 is required, the disk device 1
Only the blocks for which read processing from step 8 has been completed are written back. This can prevent the completion of the requested write processing from being delayed due to the read processing. This is particularly effective when the load on the disk device 18 is high.

(Relocation processing using defragmentation buffer #
4) Next, the relocation processing # 4 using the defragmentation buffer will be described. In this example, a buffer for temporarily holding a plurality of blocks is prepared in order to write back a block in which virtual addresses are continuous to a disk so that physical addresses are also continuous. Here, it is called a defragmentation buffer. This is shown in FIG. As shown in FIG. 18, the RAID BOOSTER 17 is provided with a defragmentation buffer 173. The defragment buffer 173 is used as a cache for speeding up read processing as well as relocation processing.

The operation when a read and write request has occurred will be described below with reference to the flowchart of FIG.

1) When a read request or a write request occurs, a copy of the read data or the write data is written to the defragment buffer 173 (step S21).
1). The defrag buffer 173 is also used as a cache as described above. 2) When a fixed number or more of valid blocks are stored in the defragment buffer 173 as a trigger, the valid blocks in the defrag buffer 173 where the virtual addresses are continuous are written back to the disk device 18 so that the physical addresses are also continuous. (Step S212).

Hereinafter, an example in which a write request is issued to V2 and V3 in the state of FIG. 18 will be described with reference to FIG. 1) Break the write request into two blocks, V2 and V3. 2) Write the contents of the write request source buffer 301 to the collective write buffer 171 and the defrag buffer 173. 3) The following processing is performed as a trigger when a certain number or more of effective blocks are stored in the defragmentation buffer 173 or the like.
Blocks V2, V3 and V100, V1 in the defragment buffer 173 where virtual addresses are continuous.
01 is written back to the disk device 18 so that each physical address is also continuous as shown in FIG. Also, the address mapping table 172 is updated.

According to this example, data having continuous virtual addresses can be stored again in the disk device 18 so that physical addresses are also continuous. Also, relocation processing # 1
Since # 3 uses the defragmentation buffer 173,
The effect on the original collective write operation of the RAID BOOSTER 17 is small. Further, since the defragment buffer 173 can store data requested to be read and written a plurality of times, the continuity of the virtual address can be seen over a plurality of accesses. This is particularly effective when there is locality in data access or when sequential access is performed continuously for backup of a disk or the like.

(Relocation processing using defragmentation buffer #
Next, an example of application to relocation processing # 4 using a defragmentation buffer will be described. a) When a read request is issued, a prefetch of a block in which a virtual address is continuous with a requested block is issued. The read-requested block and the pre-read block are written in the defrag buffer 173. That is,
As shown in the flowchart of FIG. 22, when a read request occurs (YE in step S221).
S): A block in which the read request and the virtual address are continuous is checked with reference to the address mapping table 172, and the read-requested block is read, and the prefetch of the block in which the virtual address is continuous is executed (step S). S222). Then, the read-requested block and the pre-read block are stored in the defragmentation buffer 173.
(Step S223).

As a result, it is possible to store a larger number of blocks having consecutive virtual addresses with the physical addresses also being consecutive. Also, a look-ahead effect can be expected.
Note that the pre-reading process can be applied to the rearrangement process at the time of reading.

B) When a write request occurs, a plurality of blocks that are continuous with the virtual address of the block to be written are read. The block to be written and the read block are written to the defragment buffer 173. That is,
As shown in the flowchart of FIG. 23, when a write request occurs (YE in step S231).
S) The physical address of each block in which the write request and the virtual address are continuous is stored in the address mapping table 17.
2 and check those blocks and store them in the disk drive 18.
(Step S232). Then, the write requested block and each read block are written to the defragment buffer 173 (step S234).

As a result, more blocks with continuous virtual addresses can be stored with their physical addresses also being contiguous. In addition, a cache effect can be expected.

C) When writing back a block in the defragmentation buffer 173 where virtual addresses are continuous,
Write back in the same stripe. As mentioned above,
Even if the physical addresses are not consecutive, there is a possibility that the disk access time can be reduced.

D) Attention is paid to the block requested to be read in the case of a) and the block where the virtual address is continuous with the block. On the condition that each of these blocks is not stored in such a way that the physical address and the block in which the virtual address is continuous are also stored in the same stripe, or that these blocks are not stored in the same stripe as the block in which the virtual address is continuous. , Read ahead, and defrag buffer 17
Write to 3.

E) In the case of b), attention is paid to the block whose virtual address is continuous with the block for which the write request is made. On the condition that this block is not stored in such a manner that the physical address and the block in which the virtual address is continuous are also stored in a continuous manner, or that the block is not stored in the same stripe as the block in which the virtual address is continuous. Read and write to the defrag buffer 173.

F) Write back blocks in the defragmentation buffer 173 having continuous virtual addresses to the disk device 18 according to the load of the system. When the load is high, when the defragmentation buffer 173 is full, the data is appropriately discarded. Further, the processes a) and b) are not executed when the system is in a high load state.

G) The size of the defragment buffer 173 is dynamically changed according to the memory usage of the system. This is particularly suitable when the defragment buffer 173 is configured using the storage area of the main memory 13. Of course, R
Since the operation of the AID BOOSTER 17 is controlled by the built-in processor, the RAID BOO
The size of the defragment buffer 173 may be dynamically changed according to the memory usage in the STER 17. In either case, if there is a lot of free memory, defrag buffer 1
By increasing the size of 73, the effect can be enhanced. If there is little free memory, defrag buffer 1
By reducing the size of 73, it is possible to give priority to the processing of other parts.

[0109]

As described above, according to the present invention,
By using the data requested to be read or the data requested to be written, data can be efficiently rearranged, and the read performance can be greatly improved with a small load.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a computer system according to an embodiment of the present invention.

FIG. 2 is an exemplary view for explaining the principle of a writing process to a disk device provided in the computer system of the embodiment.

FIG. 3 is an exemplary flowchart illustrating the procedure of a first data relocation process applied to a disk device provided in the computer system of the embodiment.

FIG. 4 is a first diagram for explaining a data rearrangement operation by the data rearrangement processing of FIG. 3;

FIG. 5 is a second diagram for explaining a data rearrangement operation by the data rearrangement processing of FIG. 3;

FIG. 6 is a third diagram for explaining a data rearranging operation by the data rearrangement processing of FIG. 3;

FIG. 7 is an exemplary flowchart illustrating the procedure of a second data relocation process applied to the disk device provided in the computer system of the embodiment.

FIG. 8 is a first diagram for explaining a data rearrangement operation by the data rearrangement processing of FIG. 7;

FIG. 9 is a second diagram for explaining a data rearranging operation by the data rearrangement process of FIG. 7;

FIG. 10 is a flowchart showing a procedure of a determination process for determining whether or not each data relocation process of FIGS. 3 and 7 is to be performed;

FIG. 11 is a flowchart showing another procedure of the determination processing for determining whether or not each data relocation processing of FIGS. 3 and 7 is to be executed;

FIG. 12 is a flowchart showing still another procedure of a determination process for determining whether or not to execute each data relocation process of FIGS. 3 and 7;

FIG. 13 is a flowchart illustrating another procedure of a determination process for determining whether or not each data relocation process of FIGS. 3 and 7 is to be performed;

FIG. 14 is an exemplary flowchart explaining the procedure of a third data relocation process applied to the disk device provided in the computer system of the embodiment.

FIG. 15 is a first diagram for explaining a data rearrangement operation by the data rearrangement processing of FIG. 14;

FIG. 16 is a second diagram for explaining a data rearranging operation by the data rearrangement processing of FIG. 14;

FIG. 17 is a third diagram for explaining the data rearranging operation by the data rearrangement process of FIG. 14;

FIG. 18 is an exemplary block diagram showing a functional configuration for performing a fourth data relocation process applied to a disk device provided in the computer system of the embodiment.

FIG. 19 is an exemplary flowchart explaining the procedure of fourth data relocation processing applied to the disk device provided in the computer system of the embodiment.

20 is a first diagram for explaining a data rearranging operation by the data rearrangement processing of FIG. 19;

FIG. 21 is a second diagram for explaining the data rearranging operation by the data rearrangement process of FIG. 19;

FIG. 22 is a flowchart showing the procedure of a determination process for determining whether or not to execute the data relocation process of FIG. 19;

FIG. 23 is a flowchart showing another procedure of the determination processing for determining whether or not to execute the data relocation processing of FIG. 19;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 ... CPU 13 ... Main memory 16 ... RAID controller 17 ... RAID booster 171 ... Log memory (batch writing buffer) 172 ... Address mapping table 173 ... Defrag buffer

Claims (19)

[Claims] 1. A log structure for accumulating write data in a buffer, and collectively writing a plurality of blocks of the write data accumulated in the buffer to the disk device in stripe units composed of a continuous storage area of the disk device. When a read request is issued to a plurality of blocks having continuous virtual addresses, the requested plurality of blocks are transferred to a read request source, and the plurality of blocks are transferred to the read request source. A disk control system comprising control means for writing back to the disk device so that physical addresses are continuous or stored in the same stripe. 2. The disk control according to claim 1, wherein said control means includes means for writing said plurality of blocks to said buffer, and writing back to said disk device is performed via said buffer. system. 3. The method according to claim 1, wherein the control unit performs the write-back process when the read-requested block is not stored in the disk device in a state where the physical addresses are continuous. Item 2. The disk control system according to Item 1. 4. The system according to claim 1, wherein the control unit performs the write-back process when the block requested to be read is not stored on the same stripe of the disk device. Disk control system. 5. A method for detecting the number of valid blocks in a stripe in which a block requested to be read is stored, and determining whether or not to perform a write-back process by the control unit according to the detected number of valid blocks. 2. The disk control system according to claim 1, further comprising means. 6. The apparatus according to claim 1, further comprising: means for judging a load condition of the disk device, and deciding whether or not to perform a write-back process by the control means according to the judgment result. A disk control system as described. 7. A log structure for accumulating write data in a buffer, and collectively writing a plurality of blocks of the write data accumulated in the buffer to the disk device in stripe units composed of a continuous storage area of the disk device. A disk control system of a generalized write system, wherein when a write request is issued, a read process for reading from the disk device one or more blocks in which the write-requested block and the virtual address are contiguous. Control means for executing and writing the read block into the buffer together with the block requested to be written. 8. The control means writes the read block and the write-requested block to the buffer such that physical addresses are continuous or written in the same stripe of the disk device. The disk control system according to claim 7, wherein 9. The controller according to claim 1, wherein the block in which the write request and the virtual address are continuous is not stored in the disk in a case where the physical address and the other block in which the virtual address is continuous and the physical address are not stored continuously. 8. The disk control system according to claim 7, wherein read processing from the device is performed. 10. The disk device according to claim 1, wherein the block in which the write request and the virtual address are consecutive is not stored in the same stripe as another block in which the virtual address and the virtual address are consecutive. 8. The disk control system according to claim 7, wherein read processing is performed from a disk. 11. The number of valid blocks in a stripe in which a block whose virtual address is continuous with the block requested to be written is detected, and the read processing by the control means is performed according to the detected number of valid blocks. 8. The disk control system according to claim 7, further comprising means for determining whether or not to perform. 12. The apparatus according to claim 7, further comprising: means for judging a load condition of the disk device, and deciding whether to perform a read process by the control means according to the judgment result. Disk control system. 13. The method according to claim 1, wherein the control unit writes back only blocks for which read processing from the disk device has been completed to the disk device by the time when writing from the buffer to the disk device is required. The disk control system according to claim 7, wherein 14. A defragmentation buffer for rearranging data of the disk device, and when a read request is generated, prefetching of a block whose virtual address is continuous with a block requested to be read is executed. And a read-ahead means for writing the read block together with the read-requested block into the defragmentation buffer, wherein the virtual address in the defragmentation buffer is continuous so that the physical address is also continuous, or the same stripe is used. 2. The disk control system according to claim 1, wherein data is written back to said disk device so as to be stored in the disk drive. 15. A defragmentation buffer for rearranging data of the disk device, and, when a write request occurs, a plurality of blocks in which a write-requested block and a virtual address are continuous are read from the disk device. And a means for writing the read block together with the write-requested block to the defragmentation buffer, so that the virtual address in the defragmentation buffer is continuous so that the physical address is also continuous or the same. 8. The disk control system according to claim 7, wherein data is written back to said disk device so as to be stored in a stripe. 16. Writing data is accumulated in a buffer,
A data relocation method applied to a disk control system of a log-structured writing method in which a plurality of blocks of write data accumulated in the buffer are collectively written in a stripe unit composed of a continuous storage area of a disk device. When a read request is issued for a plurality of blocks having consecutive virtual addresses, the plurality of requested blocks are transferred to a read request source, and the plurality of blocks are arranged so that physical addresses are consecutive, or A data relocation method, wherein data is written back to the disk device so that the data is stored in the same stripe. 17. Writing data is stored in a buffer,
A data relocation method applied to a disk control system of a log-structured writing method in which a plurality of blocks of write data accumulated in the buffer are collectively written in a stripe unit composed of a continuous storage area of a disk device. When a write request is issued, one or more blocks whose virtual address is continuous with the write-requested block are read from the disk device, and the read block is written together with the write-requested block. A data relocation method characterized by writing to a buffer. 18. When a read request is issued, a read-ahead is performed for a block whose virtual address is continuous with a block for which a read request is made, and the read-ahead block and the read-requested block are read out of the disk device. A prefetching step of writing data to a defragmentation buffer prepared for rearranging data, wherein a block in the defragmentation buffer where virtual addresses are continuous is stored so that physical addresses are also continuous or in the same stripe. 17. The data relocation method according to claim 16, wherein the data is written back to the disk device so that the data is rewritten. 19. When a write request is issued, a plurality of blocks in which a virtual address is continuous with a block for which a write request is made are read from the disk device, and the read block is read together with the block for which the write request is made. The method further comprises a step of writing data in a defragmentation buffer prepared for rearranging data of the disk device, wherein a block in which virtual addresses are continuous in the defragmentation buffer is set so that physical addresses are also continuous, or in a same stripe. 18. The data relocation method according to claim 17, wherein the data is written back to the disk device so as to be stored in the disk device.