JPH0784844A - Information processing system equipped with data base management system and method for updating data of data base - Google Patents

Abstract

PURPOSE:To efficiently reflect the contents of a data base buffer that the data base management system has on a data base file. CONSTITUTION:When the updated contents of the data base buffer 1 in the DBMS 1 need to be reflected on the data base, a block determination part 121 in a data base writer 12 determines object blocks. A data base and file determination part 122 sorts all the determined object blocks according to their file descriptor numbers and in-file positions to determine object block write order and a write destination data base file, which are held in a sorting result holding part 13. A writing and flashing part 123 writes and flashes the contents of the respective corresponding object blocks together through a buffering mechanism 22 in a file system 2 by determined data base files according to the sorting results.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information processing system having a database buffer and a database management system for performing a delayed write to the database, and database data for reflecting the updated contents of the database buffer in the database. Regarding the update method.

[0002]

2. Description of the Related Art In order to perform high-speed processing of a database stored in an external storage device such as a disk device, there is a database buffer in which a copy of a part of the database is placed, and a database management for writing a delay to the database. The system is known.

In this type of database management system, processing such as searching and updating of the database is usually performed on the database buffer. Therefore, in order to match the contents of the database buffer with the contents of the database, the data updated in the database buffer is stored in the corresponding database in the database.
Delayed writing to be reflected in the file is required.

Database buffers in database management systems are typically divided into operating system (OS) managed block sizes. Therefore, data transfer between the database buffer and the database file is performed in block units.

Generally, database files are managed by the OS file system. In addition, the file system has a buffer (file system buffer) similar to the database buffer in the database management system in order to perform database processing (access processing for database files) at high speed. However, the block size and the buffer size do not always match.

The database management system issues a read / write request to that effect to the file system when read / write (delayed write, etc.) is required for the database file. The file system is
When a read / write request is received from this database management system, the requested read / write processing is normally performed on the file system buffer. Therefore, even on the file system, in order to match the contents of the file system buffer with the contents of the database, the delay that reflects the updated data in the file system buffer to the corresponding database file in the database. Writing is required.

In the information processing system having the above database management system and file system, when the database buffer in the database management system becomes full, the contents of several blocks on this buffer are applied to the corresponding database. -It is necessary to reflect the file and free the block.

Therefore, conventionally, the following processing has been performed for each block to be reflected. First, the database management system issues a write request after the database management system requests the file system to open the file in order to write the contents of the target block to the corresponding database file. .

The file system opens a database file requested according to a file open request from the database management system, and then performs a write process according to a write request from the database management system. This write process is performed on the file system buffer, not on the open database file.

As described above, when the file system of the OS is performing buffering, the requested write operation is performed on the buffer in the file system by simply issuing a write request from the database management system to the file system. Therefore, the database file may not immediately reflect the updated contents of the block on the database buffer to be written.

Therefore, the database management system forces the file system to perform a flash for forcibly reflecting the contents of the block in the buffer in the file system corresponding to the target block in the database buffer in the target database file. Request for.

When the file system receives the flush request from the database management system, it writes the contents of the updated block on the file system buffer corresponding to the requested database file to the database file. . With this flash, the buffer in the file system and the database file can be synchronized (contents can be matched).

[0013]

As described above, the database management system having the database buffer and applying the delay writing method, and the file system having the file system buffer and applying the delay writing method are provided. In the conventional information processing system, for example, when the database buffer becomes full, the database management system uses (1) a database of the contents of the block for each target block to secure an empty block. • Writing to a file (actually writing to a block in the file system buffer), and (2) writing the contents of the updated block on the file system buffer corresponding to this database file to the database file. Forcibly write the flash and It had gone through yl system.

As described above, conventionally, each block to be secured is reflected in the database file, that is, writing / flushing is independently performed for the number of blocks to be secured. There has been a problem that the number of (and closing), the number of times of writing to the database file, and the number of times of flushing increase, which makes it impossible to speed up data transfer.

The present invention has been made in consideration of the above circumstances, and an object thereof is the number of file open (and close) times when the contents of the database buffer of the database management system are reflected in the database file, An object of the present invention is to provide an information processing system and a data updating method of a database that can speed up data transfer by reducing the number of times of writing to a database file and the number of times of flushing.

[0016]

According to the present invention, there is provided a database management system having a database buffer and applying a delayed writing method, and first determining means for determining a block in the database buffer to be reflected in the database. For all the blocks determined by the first determining means, the second determining means for determining the database file in which the content is to be written out, and the content of each block determined by the first determining means for the second Writing / flushing means for writing the data in a batch for each database file decided by the deciding means, and for flushing the written database file. To do.

[0017]

In the above structure, when an empty block is to be secured because the database buffer in the database management system is full, for example, the first determining means in the database management system first sets the target block. To determine the required number. An LRU (Least Recently Used) method or the like can be used to determine the target block.

Next, for all the blocks (target blocks) determined by the first determining means, the second determining means determines the database file whose contents should be written out. In order to determine the database file, for example, a management data section (management data holding means) including information (file descriptor number) designating a database file corresponding to the block is provided in each block. Just set it. Then, by acquiring the file descriptor numbers from the management data section of each of the target blocks and sorting the target blocks by the numbers, it is possible to determine the database file for batch writing / flushing. In addition, information (in-file position information) that specifies the in-file position of the corresponding block data is set in the management data section, and the above file descriptor number is used as upper-side information and the position information is used as lower-side information. By sorting the target blocks, it is possible to determine the order of the target blocks in addition to the database file for batch writing / flushing.

The writing / flushing means writes the contents of each target block decided by the first deciding means through the file system for each corresponding database file based on the sorting result of the second deciding means. Export in bulk. This export is the target database
This is done by issuing a write request from the writing / flushing means to the file system while the file is open. This write request is issued in units of consecutive blocks on the target database file. Such a block group can be easily discriminated from the sorting result.

When the file system receives the write request from the writing / flushing means, the delayed writing method is applied. Therefore, the writing of the contents of the requested block (block group) in the database buffer is performed by the file system. On system buffer.

Further, the writing / flushing means, after the above-mentioned batch writing, executes the above-mentioned each database
1 for each database file in order to reflect the contents of the updated block in the file system buffer corresponding to the file in the relevant database file
Each time, the file system is requested to flush the file.

Upon receiving this flush request, the file system performs a flush to write the contents of the updated block on the file system buffer corresponding to the requested database file to the database file.

[0023]

1 is a block diagram showing a configuration of a portion of an information processing system having a database management system according to an embodiment of the present invention, which is directly related to the present invention. In the figure, 1 is a database management system (hereinafter, D
2 is a file system of the OS.

The DBMS 1 is a database buffer 11 in which a copy of a part of database files 20-1 (# 1) to 20-4 (# 4) described later is placed in block units.
A database writer 12 for reflecting the contents of the buffer 11 in a corresponding database file, and a sort result holding unit 13.

The database buffer 11 is secured in, for example, a main memory (not shown), and processing such as search and update for the database files 20-1 to 20-4 is
It is performed on the buffer 11. For this reason, in the DBMS 1, the updated contents of the database buffer 11 are appropriately stored in the corresponding database file 20-k (k
= 1 to 4), the delayed writing is performed.

A management data section 111 for managing the block Bij is secured in a predetermined area of each block Bij (i = 0 to 4, j = 0 to 4 in the example of FIG. 1) on the DBMS 1. There is. The management data unit 111 corresponds to a block identifier (hereinafter, referred to as a block ID) for identifying the block Bij, a file descriptor number indicating the corresponding database file 20-k, and the content of the block Bij. Information indicating the position in the database file, a flag indicating the state of the block Bij, and the like are set. This flag D indicates that the content of the block Bij has been updated and that the content has not been reflected in the corresponding database file 20-k.
It is set to either (dirty) or C (clean) indicating that the updated contents of the block Bij have already been reflected in the corresponding database file 20-k or not updated. The management data section 111 does not have to be placed in each block, and may be placed in a memory area independent of the database buffer 11 so as to correspond to each block.

The database writer 12 determines a block Bij in the database buffer to be reflected in the database file, and a management data section for all blocks Bij determined by the block determining unit 121. Files in 111
A database file determination unit 122 that determines the database file whose contents are to be written out by sorting by descriptor number and position in the file
And a writing / flashing unit 123. The writing / flashing unit 123 stores the contents of each block Bij determined by the block determining unit 121 in a database.
The database determined by the file determination unit 122
This is for collectively writing each file via the file system 2 and for flushing the written database file.

The sort result holding unit 13 is a database / database.
It is used to hold the sorting result by the file determining unit 122. On the other hand, the file system 2 includes a disk device 21 in which, for example, four database files 20-1 to 20-4 (consisting of) are stored, and a buffering mechanism 22. The buffering mechanism 22 has a file system buffer 23 in which a copy of a part of the database files 20-1 to 20-4 is placed in block units. The buffer size and block size of the buffer 23 do not necessarily match the buffer size and block size of the database buffer 11.

The buffering mechanism 22 receives a read / write request for the database file 20-k (k = 1 to 4) from the DBMS 1 or the application work processing unit (not shown), and executes the requested processing as a file. The updated contents of the buffer 23 executed on the system buffer 23 are stored in a corresponding database.
The delayed writing to write to the file 20-k is performed. In addition, the buffering mechanism 22 also performs a flush on the designated database file 20-k when receiving a flush request.

Next, the operation in the configuration of FIG.
When the contents of the database buffer 11 in S1 need to be reflected in the corresponding database file 20-k due to the reason that the database buffer 11 is full, etc., the flowchart of FIG. 2 and the operation of FIG. 3 will be described. It will be described with reference to the drawings.

First, the database writer 12 in the DBMS 1 is written from the application work processing unit that manages application work as shown in FIG.
It is assumed that the request 31 is entered.

Then, the block determining unit 121 in the database writer 12 is activated. Block determination unit 121
Determines a block (target block) Bij to be reflected in the database file 20-k from the database buffer 11 (step S1). This decision includes
For example, the LRU method is used, and the flag in the management data unit 111 is sequentially determined from the block used most recently among the blocks in the state D.

Here, in FIG. 3, the shaded area is 5
Two blocks, namely blocks B01, B12, B20, B32,
It is assumed that B43 is determined as the target block Bij.
Also, blocks B01 and B20 are database files 2
0-1 (# 1), blocks B12 and B43 are database
The file 20-2 (# 2) corresponds to the database file 20-3 (# 3), and the block B32 corresponds to the database file 20-3 (# 3).
It is assumed that 01 and B20 have a relationship of B01 <B20, and B12 and B43 have a relationship of B12 <B43.

When attempting to secure a free block because the database buffer 11 is full, the required number (predetermined number) is irrespective of the state of the flag, in order from the most recently used block. , Or the requested number) is determined, and the block whose flag status is D is determined as the target block Bij to be reflected in the database file 20-k.

Step S1 by the block determination unit 121
When the processing of step 1 is completed, the database file determination unit 1
22 is activated. Database file determination unit 12
2 is a sort for all target blocks Bij determined by the block determination unit 121 by the file descriptor number in the management data unit 111 and (the information of) the position in the file (the file descriptor number is on the upper side). The write order of the target block Bij and the write destination database file 20-k are determined (step S2). This sorting result is stored in the sorting result holding unit 1.
Held at 3.

The arrangement of the respective target blocks Bij by the sorting in step S2 is B01, B20, B12, B4
The order is 3, B32. The arrangement order determines the writing order of the target block Bij. Also, the file descriptor numbers of B01 and B20 and the file descriptors of B12 and B43
From the descriptor number and the B32 file descriptor number, the write destination database file 20-k
Are the database file 20-1 (# 1), the database file 20-2 (# 2), and the database file 20-3 (# 3).

When the process of step S2 by the database / file determining unit 122 is completed, the writing / flashing unit 123 is activated. Writing / Flash section 1
23 first selects the first write destination database file 20-k based on the sort result held in the sort result holding unit 13 (step S3).

Next, the write / flash unit 123 requests the file system 2 to open the selected database file 20-k (step S4). Then, when the write destination database file 20-k is opened, the write / flash unit 123 selects the first write target block or block group based on the sort result held in the sort result holding unit 13. Then, (step S5), a write request for writing the contents of the selected block or blocks to the database file 20-k selected in step S3 is issued to the file system 2 (step S6).
Here, the condition for selecting a plurality of blocks, that is, a professional group, is that the positions in each file corresponding to these block groups are continuous.

When the write request is issued from the write / flash unit 123 (in the DBMS 1), the buffering mechanism 22 in the file system 2 is specified by the request because the delay write method is applied. The writing process of the contents of the block or block group in the existing database buffer 11 is performed not on the opened database file 20-k but on the file system buffer 23.

If there is no write destination block or block group on the file system buffer 23, it is necessary to secure the corresponding block or block group. Also, if the content of the secured block or blocks is updated and the same content is not reflected in the corresponding database file,
After reflecting the same contents in the file, it is necessary to write to the secured block or blocks.

As described above, in the buffering mechanism 22 (in the file system 2) (in the DBMS 1)
The writing process requested by the writing / flushing unit 123 is the open database file 2
The writing / flushing section 12 is executed not for 0-k but for the file system buffer 23.
From 3, it appears to be done on the database file 20-k.

When the contents of the block or block group selected in step S5 are written, the writing / flushing unit 123 is currently open based on the sorting result held in the sorting result holding unit 13. It is determined whether or not there is a remaining target block to be written in the database file 20-k (step S
7).

If there are remaining target blocks,
The writing / flashing unit 123 returns to step S5 and selects the next writing target block or block group. On the other hand, if there are no remaining target blocks,
The writing / flushing unit 123 determines that the writing of all target blocks to (the file system buffer 23 corresponding to) the currently open database file 20-k is completed, and closes the file 20-k. Request to the file system 2 (step S
8).

Next, the writing / flushing unit 123 determines whether or not there is a remaining database file (which should write out the contents of the database buffer 11) based on the sorting result held in the sorting result holding unit 13. It is determined whether or not (step S9).

If there are remaining database files, the writing / flashing unit 123 executes the step S.
Return to step 3 and select the next destination database file.

On the other hand, if there are no remaining database files, the writing / flashing unit 123 determines that the block writing to all the writing destination database files determined in step S2 has been completed, and Flush each database file by the procedure described below.

That is, the writing / flushing unit 123 first selects the first flush target database file 20-k based on the sorting result held in the sorting result holding unit 13 (step S10), and makes this selection. The file system 2 is requested to flush the created database file 20-k (step S11). As a result, the buffering mechanism 2 in the file system 2
2 is a flash that writes the contents of the blocks in the file system buffer 23 corresponding to the database file 20-k that have not been reflected in the database file 20-k to the file 20-k. I do.

When the requested flush is performed, the writing / flushing unit 123 determines whether or not there is a remaining database file to be flushed based on the sorting result held in the sorting result holding unit 13. It is determined (step S12).

If there are remaining database files, the writing / flushing unit 123 determines in step S
Returning to 10, select the next flush target database file.

On the other hand, if there is no remaining database file, the writing / flushing section 123 judges the end of a series of writing / flushing. From the above, in the above step S1, the blocks B01, B20 corresponding to the database file 20-1 (# 1),
Blocks B12 and B43 corresponding to the database file 20-2 (# 2) and the database file 20-3
In the present embodiment in which B32 corresponding to (# 3) is determined as the target block Bij, the series of processes described above is performed in the order described below.

First, the top database file 20
-1 (# 1) is opened, the contents of blocks B01 and B20 are written, and the file 20-1 (# 1) is closed. At this time, if the positions in the database file 20-1 corresponding to B01 and B20 are continuous, the writing of B01 and B20 can be specified by one write request.

Next, the database file 20-2 (#
2) is opened, the contents of blocks B12 and B43 are written, and the file 20-1 (# 1) is closed. At this time, if the positions in the database file 20-2 corresponding to B12 and B43 are continuous, this B12,
Writing of B43 can be specified by one write request.

Next, the last database file 20
-3 (# 3) is opened and the contents of block B32 are written out. Next, the database file 20-1 (#
Flush for 1), then for database file 20-2 (# 2), and then for database file 20-3 (# 3).

The above situation is shown in FIG. In this example, the number of file opens / closes and the number of flushes are all 3 which is equal to the number of database files to be written. On the other hand, conventionally,
Opening / closing and flushing must be performed five times, which is a number matching the number of target blocks to be written. In this example, the write request also
Database file 20-1 position corresponding to B01 and B20, database file 2 corresponding to B12 and B43
If the positions within 0-2 are continuous, it is enough to do three times. On the other hand, conventionally, even if such a condition is satisfied, write requests for the number of target blocks (5 times) are required.

In the above-described embodiment, for all the target blocks determined in step S1, the batch writing for each database file determined in step S2 is completed, and then the batch flushing for each database file is performed. However, the present invention is not limited to this. For example, the batch flush may be performed on the database file every time the batch writing of all the target blocks to one database file is completed.

[0056]

As described above in detail, according to the present invention,
Determine the block (target block) in the database buffer to be reflected in the database, and write the contents of this target block to the database
A file is determined, and the contents of the corresponding target blocks corresponding to each of the determined database files are collectively written / flushed via the file system. The number of file opens (and closes), the number of writes to the database file, and the number of flushes can be reduced compared to what was used for flushing, thus speeding up data transfer (data update) and system performance. Is improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a portion of an information processing system having a database management system according to an embodiment of the present invention, which is directly related to the present invention.

FIG. 2 is a flowchart showing an operation procedure of the database writer 12 in the embodiment.

FIG. 3 shows a database buffer 1 in the embodiment.
FIG. 6 is an operation explanatory diagram when it is necessary to reflect the contents of 1 in the corresponding database file 20-k.

[Explanation of symbols]

1 ... Database management system (DBMS), 2 ... File system, 11 ... Database buffer, 12 ...
Database writer, 13 ... Sort result holding unit, 20
-1 to 20-4 ... Database file, 21 ... Disk device, 22 ... Buffering mechanism, 23 ... File system buffer, 111 ... Management data section (management data holding means), 121 ... Block determining section (first Determining means), 122 ... Database file determining section (second determining means), 123 ... Write / flash section.

Claims (3)

[Claims] 1. A database management system having a database buffer in which a copy of a part of a database in which various database files are stored is placed in block units, and a copy of a part of the database is a block. In an information processing system having a file system buffer placed in units, and a file system to which a delayed writing method is applied, in the database management system, determining a block in the database buffer to be reflected in the database. 1 deciding means, 2 deciding means for deciding a database file whose contents are to be written out for all blocks decided by the 1st deciding means, and said deciding means decided by the 1st deciding means The contents of each block are And writing / flushing means for writing the data in a batch for each database file decided by the deciding means through the file system and for flushing the written database file. Information processing system. 2. The database management system further comprises management data holding means for holding management data including designation information of a database file corresponding to each block in the database buffer, The determining means acquires the corresponding database / file specifying information for each block determined by the first determining means based on the contents of the management data holding means, and sorts by the same specifying information. The information processing system according to claim 1, wherein the database file to be collectively written / flushed is determined by this. 3. A database management system having a database buffer in which a copy of a part of a database in which various database files are stored is placed in block units, and a copy of a part of the database is a block. In a data updating method of a database used in an information processing system having a file system buffer placed in units and a file system to which a delayed write method is applied, a block in the database buffer to be reflected in the database is determined. A first step, a second step of deciding a database file whose contents are to be written out for all the blocks decided in the first step, and each block decided in the first step The contents of the second A third step of collectively writing out, via the file system, each of the database files determined in the step, and performing a flush to the written database file. Data update method.