JPH07230402A - Data base management device - Google Patents

Abstract

PURPOSE:To lighten the load of a lock process for the management of a data base consisting of both short and long records and improve the parallelism in a parallel process when at the time of the lock process. CONSTITUTION:This data base management device is equipped with an allocating process means 4 which allocates long data, requiring plural pages beyond the size of a page as the fixed-length storage allocation unit of the data base, to a page containing the substantial data of the long data and a page including the header indicating the position of the page containing the substantial data, and a lock control means 5 which locks the page containing the header for the whole long data and also locks a specific part of the long data for the page containing the substantial data; and the data base that plural users share is accessed with efficiency.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a database management apparatus capable of reducing the load of lock processing at the time of accessing a database in which short records and long records coexist and improving parallel processing of accesses.

[0002]

2. Description of the Related Art Conventionally, in a database management system, there are many cases where a plurality of programs that access a database are executed simultaneously (parallel processing) or a plurality of the same programs are executed. Here, the execution of one set of processing in a program is called a transaction, and the data in the database is accumulated as a plurality of items.

Therefore, the execution of a set of processes by the program for the database is performed serially at each transaction step, but in the case of parallel processing, when transactions for a plurality of programs are executed in a mixed manner, Depending on the execution order, the programs may not be executed serially. Even in the case of this parallel processing, it is possible to serialize that the execution result is equivalent to that when a plurality of programs are executed independently.

In order to guarantee this serializability, when parallel processing is performed, an item of data in the database is locked, and by this lock, one transaction is
Other transactions are prohibited from accessing the item. In addition, the prohibition of access is unlocked by unlocking the item. And
Appropriate use of the lock processing of the lock and unlock can guarantee serializability during parallel processing.

For example, the first phase is a lock phase, the second phase is an unlock phase,
By using an algorithm based on a two-phase locking method in which all locks precede all unlocks in any transaction, serializability during parallel processing is guaranteed.

Locking by an algorithm such as this two-phase locking method includes a reference lock (or a shared lock) for prohibiting update such as writing of another transaction at the time of reference such as reading by a certain transaction, and writing by a certain transaction. There are two types of locks, an update lock (or an exclusive lock) that prohibits references such as reading of other transactions when updating, and prohibits updates such as writing of other transactions.

Therefore, at the time of the shared lock, an arbitrary number of transactions can simultaneously hold the shared lock on the shared locked item. Also, if a transaction holds an exclusive lock on an item, no other transaction can acquire a shared or exclusive lock on this item.

In order to guarantee serializability during parallel processing of such a plurality of programs, the lock processing described above is performed as one of the transactions.

For details, refer to "Principles of Database Systems" (translated by Jeffery D. Ullman, translated by Toshiyasu Kunii and Nobuo Ohbo,
Japan Computer Association [Original published in 1982, translated published in 1985
Year]), Chapter 11, "Parallel Operations on Databases".

Here, as a unit (granularity) of the data to be the above-mentioned lock processing, a variable length record unit for performing the lock processing for each individual record, and a plurality of records belonging to a certain page are collectively locked. There is a fixed length page unit.

The former lock processing in units of variable-length records has an advantage that lock processing can be performed only on truly necessary data and parallelism at the time of parallel processing is improved. However, in database management that processes many short records,
Since it is necessary to perform the lock processing for each record, there is a disadvantage that the time required for the lock processing increases. In particular, the deadlock detection time increases as the number of records increases. Therefore, the lock processing in units of variable-length records is suitable for a transaction processing a small number of short records.

The latter locking process in units of fixed-length pages has a drawback in that the locking process is performed even for data that belongs to the page but is not an access target, and the concurrency decreases during parallel processing. However, when processing a large number of short records on the same page, there is an advantage that it is not necessary to perform lock processing for each record, and only the time required for lock processing for one page is required. Therefore, the lock process in units of fixed length pages is suitable for a transaction that processes many short records on the same page.

By the way, in recent years, not only numerical data but also
The database manages large data such as image data and large character string data that exceeds a fixed page size.

As an efficient storage method of a database for such long data, for example, Japanese Patent Laid-Open No. 3-62 is used.
As described in Japanese Laid-Open Patent Publication No. 137, there is a method of storing long data by a variable length block group in which a long record is divided into a header and a group of variable length blocks and stored.

Further, as described in Japanese Patent Laid-Open No. 1882749/1992, there is a relational expression database storage system in which a long record is divided into a fixed length block list and a variable length block list and stored.

[0016]

However, as long data, such as long image data such as still image data, it is necessary to access all data at the time of access, and long data such as character string data. For example, there is data that may access only a specific part of long data at the time of access.

Therefore, when the short record and the above-mentioned long record are mixed, there is the following problem when performing the lock process for guaranteeing the serializability during parallel processing.

That is, when the lock processing adopting the variable length record unit described above is performed, the lock processing for one long and large data need only be performed once, so that the lock processing does not have a load and does not take much time. In addition, since all data is always accessed when accessing long data such as still image data, parallelism during parallel processing does not matter. However, the lock processing when accessing only a specific part of long data such as character string data is also performed for the part other than the specific part, so the parallelism during parallel processing decreases. There was a problem.

On the other hand, if the lock processing adopting the fixed length page unit described above is performed, parallel processing is performed even when it is necessary to access all the data for long data or when only a specific portion is accessed. In some cases, since only the necessary data is locked, concurrency is not a problem, but the locking process is overloaded because multiple pages must be locked, and as a result, it takes time. There is a point.

Therefore, the present invention eliminates such a problem, reduces the load on the lock processing in the lock processing of the database management in which short records and long records are mixed, and reduces the concurrency during parallel processing. An object is to provide a database management device that can be improved.

[0021]

According to the present invention, in a database management apparatus for managing a database shared by a plurality of users, a large-scale data is stored over a plurality of pages exceeding a page size which is a fixed-length storage allocation unit in the database. Allocation processing means (4 in FIG. 1) for allocating data to a page including substantive data of the long data and a page including a header indicating the position of the page including the substantive data, and the entire long data. Lock control means (5 in FIG. 1) for controlling the page including the header and the lock processing for the specific portion of the long data to be performed on the page including the actual data. It is characterized by

[0022]

According to the present invention, when storing predetermined data in a database shared by a plurality of users, the allocation processing means extends the plurality of pages beyond the size of the page, which is the storage allocation unit of the database. If the data is long data, a process of allocating the page including the substantial data of the long data and the page including the header indicating the position of the page including the substantial data is performed. Thereafter, when a plurality of users access the database, the lock control means performs lock processing for ensuring serializability of parallel processing in accessing the database allocated by the allocation processing means. In the case of locking the entire data, the page including the header is performed, and in the case of locking the specific portion of the long data, the page including the actual data is performed.

Therefore, when the access target is the entire long data, only the page including the header is locked, and when it is a specific portion of the long data, the locking process is performed only on the minimum necessary page. Since the minimum number of times lock processing is required according to each access target is performed, the load on the lock processing is reduced, and only the minimum access target area is locked, so the same Since it is possible to access other pages in long data, concurrency during parallel processing is also improved.

[0024]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a diagram showing the configuration of a database management apparatus which is an embodiment of the present invention.

In FIG. 1, the database management system includes a storage unit 6 that stores database files, a database management unit 3 that performs access management including lock processing of the database files stored in the storage unit 6,
It has an input unit 1 for instructing access to a database file stored in the storage unit 6 and an output unit 2 for outputting a result of access to the database file.

The database management unit 3 further has a data allocation unit 4 and a lock control unit 5.

The data allocating section 4 performs a record allocating process when storing data in the database based on the page size of each database file designated in advance.

The data allocating process of the data in the database file by the data allocating unit 4 is based on the page size of the database file. The page size is determined when the database is generated. . Based on the determined page size, the data allocating unit 4 determines whether the data exceeds the page size, that is, whether it is allocated as a long record or a short record, and performs data allocation processing.

The lock control section 5 stores a lock admissibility criterion corresponding to the lock type in the lock table 5.
a, and controls the lock process based on the current lock type and the lock table when accessing the database.

The lock control performed by the lock control unit 5 basically performs a lock process in units of fixed length pages. Then, whether the partial lock process at the time of accessing partial data in the long data or the whole lock process at the time of accessing all data of the long data is determined by the user's designation. Therefore, the lock control unit 5 controls the partial lock process and the entire lock process.

Next, the allocation layout of the records in the storage unit 6 that has been allocated by the data allocation unit 4 will be described.

This record allocation arrangement includes data that can be stored in short record units within the page size, and
Allocation is executed by dividing the page size into long data (long record) that cannot be divided by the page size.

FIG. 2 is a diagram showing the layout of short records in the storage unit 6.

In FIG. 2, the short data that can fit in one page is allocated as a short record. That is, the database file has a plurality of pages P1 to Pn.
Each page is allocated in short record units. For example, in page P2, a plurality of short records R1 to Rn and their respective records R
Tables H1 to Hn that hold in-page offsets corresponding to 1 to Rn are arranged separately. The in-page offset tables H1 to Hn are an array in which positional information within a page of a plurality of short records R1 to Rn is used as an element.

FIG. 3 is a diagram showing the layout of long records in the storage unit 6. This long record is a unit that is stored across multiple pages. In this case,
It is composed of a plurality of pages for storing real data of long data and a page for storing a header storing allocation information of the real data of long data.

FIG. 3 shows an example of arrangement of long records over four pages. These four pages are a page Pa that stores a header and a page Pb that stores actual data.
.About.Pd.

The page Pa has the header of the long record and the in-page offset Ha. Then, in this header, the page offset indicating the page position of the actual data allocated to the page Pb and the page Pb
Offset information H about page Pb consisting of an in-page offset indicating the position of actual data from a fixed position in
Db, offset information HDc relating to page Pc, which is a page offset indicating the page position of the actual data allocated to page Pc, and an in-page offset indicating the position of the actual data from a fixed position within page Pc, and allocation to page Pd It has offset information HDd about the page Pd, which includes a page offset indicating the page position of the obtained actual data and an in-page offset indicating the position of the actual data from a fixed position within the page Pd.

The page P to which the actual data is allocated
b to Pd have respective real data and page offsets Hb to Hd for the respective real data.

Next, the lock processing at the time of accessing the data thus allocated will be described. This lock processing is necessary to guarantee serializability during parallel processing. And in this embodiment,
As described above, this lock processing is basically performed in fixed-length page units.

Here, as described above, there are two types of access to long data, one is to access the entire data without fail, such as still image data, and the other is to access partly with character string data. Therefore, two lock processes, that is, the whole lock process and the partial lock process are performed on the long data.

When the lock control unit 5 performs the lock process, the lock control is performed based on the lock table 5a. Here, there are a shared lock and an exclusive lock as the types of lock processing, and the shared lock prohibits updates such as writing of other transactions at the time of reference such as reading by one transaction. Any number of transactions can hold a shared lock on the same item. On the other hand, the exclusive lock is
It prohibits references such as reading of other transactions when updating such as writing by one transaction, and prohibits updating such as writing by other transactions. In this case, the other transactions share a lock on this item. And exclusive lock cannot be performed.

In the lock processing of the lock control section 5, it is determined by referring to the lock table 5a whether or not the lock to be controlled is possible.

FIG. 4 is a diagram showing the contents of the lock table.

In FIG. 4, the lock table 5a includes
The relationship between the current lock type, which is the locked state of the page stored in the storage unit, and the designated lock type that the lock control unit 5 is going to lock-control is stored. There are three types of locks: no lock, shared lock, and exclusive lock.

The relationship between the current lock type and the specified lock type is that the exclusive lock cannot be specified when the current lock type is the shared lock, and the shared lock and the exclusive lock are specified when the current lock type is the exclusive lock. I can't. In other combinations, the lock can be designated. For example, when the current lock type is a shared lock, the shared lock can be designated. By using this lock table 5a, the lock process is properly controlled.

Using such a lock table 5a,
The lock control unit 5 performs lock control, the details of which will be described below.

In the lock processing, in addition to the lock type such as the shared lock or the exclusive lock, it is necessary to determine the lock range, and there are the whole lock processing and the partial lock processing described above.

First, the whole lock process will be described.

FIG. 5 is a flow chart showing the processing procedure of the overall lock processing by the lock control unit 5.

In FIG. 5, first, an access request including designation of whether the user accesses the entire data or the data portion is input from the input unit 1 to the database management unit 3. . Less than,
The case where the access request for the entire data is included in the access request will be described.

First, the lock control unit 5 of the database management unit 3 determines whether the input data to be accessed is a short record data or a long record data based on the page size. A judgment is made (step 101).

If the data is a short record, the database management unit 3 acquires the page number of the designated record (short record) in which the data is stored (step 10).
2). After that, the lock control unit 5 specifies the type of access, for example, the type of lock corresponding to read or write, in order to guarantee serializability during parallel processing. At this time, the lock control unit 5 Lock table 5a
It is judged whether the designated lock is possible or not based on (Step 1
03), if not possible, the designated lock is disabled (step 104), and the designated lock process is not performed, and this process ends. As a result, the specified data is not accessed. On the other hand, if the designated lock is possible in step 103, the designated lock process is permitted (step 10
5) Then, the designated lock process for the designated page of the page number acquired in step 102 is executed (step 106), and this process is terminated.

On the other hand, when it is determined in step 101 that the access is to a long record, that is, data exceeding the page size, the database management unit 3 acquires the page number in which the header of the designated record (long record) is stored. (Step 107). After that, the lock control unit 5 performs the lock process, and at that time, the lock table 5
It is determined whether or not the designated lock is possible by referring to a (step 108). If the designated lock is not possible, the designated lock is disabled (step 109), and the designated lock processing is not performed, and this processing ends. On the other hand, if the designated lock is possible, the designated lock processing is permitted (step 105),
The designated lock process to the page of the page number of the header of the long record acquired in step 107 is executed (step 106), and this process ends.

After that, by executing the transaction related to the whole lock process and the transaction of the access process itself, the access is executed in the state where the serializability of the parallel process is guaranteed, and the execution result is output. Is output to.

As a result, the lock processing for the short record and the lock processing for accessing the entire data for the long record are performed only for one page. In particular, the locking process for multiple pages is not performed for long records, the specified locking process is performed only for the page that has the header of long records, and the specified locking process is performed for the pages that have other actual data. Since there is no lock, the load on the lock processing is reduced.

Next, the partial lock processing will be described.

FIG. 6 is a flowchart showing the processing procedure of the partial lock processing by the lock controller 5.

In FIG. 6, first, an access request is input from the input unit 1 to the database management unit 3 including the designation whether the user accesses the entire data or the data portion. . Less than,
The case where the access specification for the data part is included will be described.

First, the lock control unit 5 of the database management unit 3 determines whether the input access target data is a short record data or a long record data based on the page size. A judgment is made (step 201).

If the data is a short record, the database management unit 3 acquires the page number of the designated record (short record) in which the data is stored (step 20).
2). After that, the lock control unit 5 specifies the type of access, for example, the type of lock corresponding to read or write, in order to guarantee serializability during parallel processing. At this time, the lock control unit 5 Lock table 5a
It is judged whether the specified lock is possible or not based on (Step 2
03), if not possible, the designated lock is disabled (step 204), and the designated lock process is not performed, and the process ends. As a result, the specified data is not accessed. On the other hand, if the designated lock is possible in step 203, the designated lock processing is permitted (step 20).
5) Then, the designated lock process for the designated page of the page number acquired in step 202 is executed (step 206), and this process ends.

On the other hand, when it is judged in step 201 that the access is to a long record, that is, data exceeding the page size, the database management section 3 acquires the page number in which the header of the designated record (long record) is stored. (Step 207). After that, the lock control unit 5 performs the lock process. In this case, since the partial lock process is performed, it is first determined by referring to the lock table 5a whether or not the shared lock to the page having the header of the long record is possible. (Step 208). As a result, if the shared lock is not possible, the designated lock is disabled (step 213), the designated lock process is not performed, and the process ends. As a result, the data designated by the user is not accessed. On the other hand, if the shared lock process for the page of the header well is possible, the shared lock process for the page of the header is executed (step 209), and then the database management unit 3 stores the actual data of the portion specified by the user. Get the page number of the page that has (step 21)
0). Further, the lock control unit 5 refers to the lock table 5a to determine whether or not it is possible to lock the page having the page number including the actual data acquired in step 210 (step 211). As a result of this determination, if the designated lock processing is possible, the designated lock processing is permitted, the designated lock processing for the page having the actual data is executed (step 206), and this processing is ended. On the other hand, if the designated lock to the page having the actual data is not possible in step 211, the shared lock release process to the page of the header already executed in step 209 is executed (step 212),
The designated lock is disabled (step 213), and this processing ends. The actual data page determined in step 211 may have a plurality of pages depending on the user's designation. In this case, it is determined whether or not the designated lock can be performed on the plurality of pages.

As a result, the lock process when accessing partial data for a long record is only the shared lock process for the page having the header of the long record and the lock process for a specific page having the necessary actual data part. Since the access processing to the actual data in the necessary long data is executed, the load on the lock processing is reduced, and the access processing to other real data of the long data can be performed in parallel. Concurrency will be improved.

The access process to the database is
Execute by transaction T, which is one step of the access processing program. As described above, the lock processing for this transaction is performed when the entire lock processing is performed for the record R (long record or short record) to be accessed and the record R
The process is divided into the case of performing the partial lock process of processing a part of the above, and the process of the whole lock process is actually executed by the next instruction (LOCK). That is, LOCK (T, R, TYPE). Here, T indicates a designated transaction which is a target of lock processing, and R indicates a designated record which is a target of lock processing. Further, TYPE indicates a lock type that is a lock type of a shared lock or an exclusive lock depending on whether the access processing is read or write. By this instruction, all areas of the designated record R are locked with the designated lock type TYPE until the designated transaction T is completed.

When the partial lock processing is performed, it is actually executed by the next instruction (PART_ROCK).
That is, PART_ROCK (T, R, TYPE, OFFSE
T, LENGTH). Here, T indicates a designated transaction which is a target of lock processing, R indicates a designated record which is a target of lock processing, and TYPE indicates a lock type as in the above. Further, OFFSET is a designated offset, indicating the offset position of the access target record, and LENGTH is a designated length, indicating the offset amount from the offset position of the access target record. This instruction causes the specified offset O of the specified record R until the specified transaction T is completed.
The lock of the designated lock type TYPE is applied to the area from FFSET to the designated length LENGTH.

Incidentally, these instructions correspond to the execution of the lock phase in the two-phase locking system.

By the way, in the above-mentioned embodiment, the database management unit 3 is connected to the input unit 1 and the output unit 2 for the sake of convenience of explanation.
Is often connected to a network, and access processing from a plurality of terminal devices connected to the network is often executed. In this case, the parallelism at the time of parallel processing by the database management unit 3 can be further improved. At the same time, the load on the lock processing is reduced.

[0068]

As described in detail above, according to the present invention, when storing predetermined data in a database shared by a plurality of users, the allocation processing means allocates the predetermined data to a fixed-length storage allocation in the database. In the case of long data that exceeds a page size as a unit and spans a plurality of pages, a page including substantial data of the long data,
The process of allocating to the page including the header indicating the position of the page including the entity data is performed. Thereafter, when a plurality of users access the database, the lock control means performs lock processing for ensuring serializability of parallel processing in accessing the database allocated by the allocation processing means. In the case of locking the entire data, the page including the header is performed, and in the case of locking the specific portion of the long data, the page including the actual data is performed.

Therefore, if the access target is the entire long data, only the page including the header is locked, and if it is a specific part of the long data, only the minimum necessary pages are locked. Since the minimum required number of lock processes according to the access target of each is required, the load on the lock process is reduced, and the lock process is performed only on the access target area that is the minimum required. Since it is possible to access other pages in the same length data,
It has an advantage that parallelism at the time of parallel processing is also improved.

Particularly, it is possible to simultaneously satisfy the reduction of the load on the lock processing and the improvement of the parallelism during the parallel processing in the database management in which the short record and the long record including the accessibility to the specific portion are mixed. It has the advantage that it can.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of a database management apparatus that is an embodiment of the present invention.

FIG. 2 is a diagram showing an allocation layout of short records in a storage unit 6.

FIG. 3 is a diagram showing a layout of long records in a storage unit 6;

FIG. 4 is a diagram showing the contents of a lock table.

FIG. 5 is a flowchart showing a processing procedure of overall lock processing by a lock control unit 5.

FIG. 6 is a flowchart showing a processing procedure of partial lock processing by the lock control unit 5.

[Explanation of symbols]

1 input part 2 output part 3 database management part 4
Data allocation unit 5 Lock control unit 5a Lock table 6 Storage unit

Claims (1)

[Claims] 1. A database management device for managing a database shared by a plurality of users, wherein long data of a plurality of pages exceeding a size of a page, which is a fixed-length storage allocation unit in the database, is a substance of the long data. An allocation processing unit that performs a process of allocating to a page including data and a page including a header indicating the position of the page including the substantive data, and locking processing for the entire long data is performed on the page including the header, A database management apparatus comprising: a lock control unit that controls a lock process for a specific portion of the long data with respect to a page including the actual data.