JP2001142752A - Database management method - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To add a database storage area with an increase in database capacity, it is necessary to relocate data to the added database storage area. When m database storage areas are provided as storage areas of a database, one or more data items of the database are used as a partitioning key, and a hash function is applied to the database using the partitioning key. A hash map table that divides into n (m ≦ n) logical units of buckets and determines the correspondence of database storage areas that manage each bucket according to the given number of database storage areas;
The database is managed by a segment hash map table for mapping the distributed buckets to segments, which are storage units in each database storage area.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a database management system for increasing the size of a database by rearranging data that increases in time series into a newly added database storage area. .

[0002]

2. Description of the Related Art Generally, in a database system having a database, information stored in the database changes every moment. New information is added to the information over time, and the information may become larger than the capacity of a database storage area for storing a prepared database.

Therefore, when dealing with a large-scale database, a method is adopted in which the database is divided into a plurality of database storage areas and stored in advance. Means such as key range division, hash division, and equal division are prepared as division means for dividing and storing data in a plurality of database storage areas. A method is employed in which each partitioning means associates the divided partitions one-to-one with a designated database area, or associates a plurality of partitions with a certain database area. When new information is added over time, two database expansion methods are conceivable. One is a method of adding a new key range or dividing a certain key range when the key range division is adopted as the dividing means. This can be operated without adding a new database area. Another method is to add a new database area or extend an existing database area without changing the dividing means. This is a method of increasing the database storage area as the capacity increases. On the other hand, if hash division is adopted as the division means, the overhead for the division is small, and the flexibility to increase the data is high.

[0004] However, in the database storage area prepared in advance, not only new information is added with time, but also old data in the past may be deleted, and the capacity is not always larger than the capacity of the database storage area. . In such a case, a solution is made by creating a space by deleting old data by performing a database reorganization, and using it to store information to be newly added.

However, in a database system that accumulates past data endlessly, preparing a database storage area in advance in anticipation of the amount of data expected in the future requires resource management costs. A database area capable of storing up to the expected data amount is prepared, and an operation is performed in which a database storage area is added when a new database storage area becomes necessary as the data amount actually increases.

[0006] Adding a database storage area to a database involves changing the definition of the database. As the simplest method, there is a method of once backing up the contents of a database, performing a definition change process for adding a database storage area, and then reloading the backed up database. However, in this method, in the case of a large-scale database, the time required to acquire a backup and the cost of a backup medium are extremely high.
The process of reloading also takes an enormous amount of time.

As a first known example, US Pat.
There are 2,285. In this known example, a table is hash-divided, but a technique is disclosed in which the table is divided into a plurality of buckets in advance, and some buckets are managed in correspondence with virtual processors.

A second known example is disclosed in Japanese Patent Application Laid-Open No. H6-11391.
No. 19 discloses a technique for reorganizing data according to the access frequency of a database divided on a parallel database system including a plurality of processors.

A third known example is disclosed in Japanese Patent Application Laid-Open No. 6-3142.
No. 99 discloses a technique for hierarchically dividing a database on a parallel database system including a plurality of processors.

[0010] As a fourth known example, Japanese Patent Application Laid-Open No. 7-1413 is disclosed.
Japanese Patent Application Laid-Open No. 94-293006 discloses a fifth known example of a database divided on a parallel database system composed of a plurality of processors. A technique is disclosed in which data is not required to be rearranged when the number of divided databases is changed, and data is to be stored in a newly added storage area. However, in this known example, data rearrangement is not necessary, but all the divided storage devices are searched at the time of search.

[0011]

In the case of the former method in the prior art, a problem arises in that the database is enlarged. Relocation is required due to the increase in the size of the database when data is added to a table in the database area, so that a predetermined database storage area becomes full.

In the case of a database employing the above-described hash division, the hashing result depends on the number of divisions, that is, the given number of database storage areas. Therefore, when a new database storage area is added, all the data accumulated up to that point is updated. Hashing must be performed again under the number of divisions and restored. This wastes a huge amount of time and resources, and the system cannot be operated efficiently.

In the case where data is moved from each existing database storage area to another database storage area without re-storage as a result of rehash as a rearrangement process, all data in each database storage area is eventually returned. It must be read, and even if there is data remaining in the same database storage area even after rehashing, useless reading of data occurs.

[0014] A first object of the present invention is to add a database storage area to a hash-divided database at the lowest cost from an existing database storage area to a newly added database storage area. It is an object of the present invention to provide a database storage management method that enables reallocation.

[0015]

In order to achieve the above object, the present invention manages a database storage area composed of one or a plurality of external storage devices, and the database storage area has a physically fixed length. In a database management method for managing a plurality of segments composed of a plurality of pages, when m database storage areas are given as storage areas of a database, one or more data items of the database are used as a partitioning key. Applying the hash function to the partitioning key, dividing the database into n (m ≦ n) logical units of buckets, and dividing each bucket according to the given number of database storage areas. A hash map table that determines the correspondence between the database storage areas to be managed and the distributed buckets Providing a database management method characterized by managing the database by the segment hash map table for segmented and mapped a storage unit of database storage area.

[0016]

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

FIG. 2 is a diagram showing an example of a hardware configuration of the computer system of the present embodiment. The computer system 10 includes a CPU 12, a main storage device 14, an external storage device 20 such as a magnetic disk device, and a large number of terminals 30.
It consists of. A database management system 40 is placed on the main storage device 14, and database areas 50 a, 50 b, 5 for storing databases managed by the database management system 40 using a plurality of external storage devices 20.
0c, 50d, a database definition information storage area 60 that manages database definition information, and a database log storage area 70 that manages update history information related to update operations on the database. . Further, a program for implementing the database management system 40 is also stored on the external storage device 20.

FIG. 1 shows a database management mode according to the present invention. The database definition information storage area 60 shown in FIG. 2 stores a database storage area and database definition information constituting a database system.
The database storage area management table 61 manages information on a database storage area defined as an area for storing a database in the database system. The database storage area configuration file management table 62 manages information related to the configuration of the file allocated on the external storage device shown in FIG. 2 specified at the time of defining the database storage area. The database management table 63 manages a table defined particularly in a case such as a relational database management system. The database configuration column management table 64 manages information related to columns configuring the defined table. The database division management table 65 has information relating to a table and a database storage area when data is divided and stored by designating a database storage area defined in the database system at the time of table definition. When the table defined in the database required by the present invention is divided into a plurality of database storage areas and stored, the hash map table 66 is designated when the hash division method is designated as the division method. It has a correspondence table for determining which database storage area stores each hash bucket when divided into hash buckets corresponding to the number of database storage areas.

Next, the database storage areas 50a, 50
The configurations of b, 50c, and 50d will be described. A management method for a hash-divided table in a database storage area will be described. When dividing a defined table into hash buckets, if the element number of the hash bucket is determined by the hash map table, a plurality of physically fixed-length pages that store the data of the hash bucket with the determined element number Segment hash map table 51 for managing the leading segment number when data is stored in a segment 54 composed of a segment and a segment bit map for managing the allocation of all segments secured in the database storage area as a bit map Table 52
And a page bitmap table 53 that manages the allocation of pages in the allocated segment 54 as a bitmap and manages the chain between segments when a plurality of segments are allocated with the same hash bucket element number. You.

FIG. 3 shows an embodiment of the database definition information stored in the database definition information storage area 60 shown in FIG. 2 in detail. DDICAREA, DBAREA1, D
The table shows information stored in the database storage area management table 61 and the database storage area configuration file management table 62 when BAREA2, DBAREA3, and DBAREA4 are defined. The database storage area management table 61
It consists of columns of database area name, type, number of configuration files, server name, page length, segment size and number of segments. Further, the database storage area configuration file management table includes columns of a database area name, a number, and a configuration file name. In the database, D
When a database storage area called DICAREA is defined, it is identified by information “D” indicating that it is a database definition information storage area, and the number of files constituting the database storage area is 1, 1, FIG.
Segment 54 indicated by the database storage area in
Is 409 as the page length which is the size of the page constituting
6 bytes, 5 as the number of pages making up the segment
Zero pages and the number of segments 50 allocated to the database storage area are stored as definition information.
Similarly, DBAREA1, DBAREA2, DBARE
When A3 and DBAREA4 are defined, 'U' for identifying the database storage area as the type of the database storage area, and information on the number of configuration files, server name, page length, segment size, and number of segments are stored. You. On the other hand, when the database storage area configuration file management table 62 is composed of a file name and a plurality of files configuring each database storage area, the serial number is stored as information.

Next, when a "ZAIKO" table is defined in the database, how the definition information of the table is stored in the database management table 63, the database configuration column management table 64, and the database division management table 65. Will be described. An example of the definition sentence of the "ZAIKO" table is shown.

CREATE TABLE ZAIKO (SCODE CHAR (10), SNAME CHAR (10), COL CHAR (4), TANKA INTEGER, ZSURYO INTEGER) HASH BY SCODE IN (DBAREA1, DBARE2, DBAREA3, DBAREA4); ZAIKO "table has four columns SCO
DE, SNAME, COL, TANKA, ZSURYO
And hash-divided using the SCODE column as a partitioning key. The hash-divided data is stored in database storage areas DBAREA1, DBAREA2, D
It is instructed to store the data separately in BAREA3 and DBAREA4. According to the contents of the definition, the database management table 63 has a table name, an owner, the number of constituent columns, a dividing method, a dividing number indicating the designated number of database storage areas, and a database for indicating that the partitioning key information is a SCODE column. The column number in the constituent column management table is stored. The database configuration column management table 64 includes “ZAI
As the definition information of the five columns constituting the “KO” table, column names,
The table name, owner, column number indicating the order of the columns constituting the table, data type, and data length are stored. Further, in the database division management table 65, a table name, a number indicating the order of definition of the database storage area, and a database storage area name are stored in order to store the “ZAIKO” table divided into four database storage areas. .

FIG. 4 shows the hash map table 66 shown in FIG. As an embodiment of the present invention, in the database, the maximum number of database storage areas when a database table is divided into a plurality of database storage areas is 512. At this time, the maximum number of hash buckets in hash division is set to 1,024. Hash map table 66
May be a method of indicating the correspondence between the element numbers of the hash buckets in each table and the database storage area. Has a correspondence table between the hash bucket element numbers obtained by the hash function and the values in the “number” column of the database division management table shown in FIG. 3 in accordance with all the specified numbers of database storage areas. The simplest example is a two-division example. In the case of two divisions, the first entry of the hash map table is referred to, indicating that the hash bucket element numbers 0 to 511 are stored in the designated number 0 of the database storage area, and the hash bucket element numbers 512 to 1023
Up to the designated number 1 in the database storage area. On the other hand, the database storage area is determined by referring to the numbers 0 and 1 of the database division management table of the corresponding table. Next, in the case of three-partitioning, when one database storage area is added from the two-partitioned database, the added database storage area is added to the three hash buckets of the existing two divided database storage areas. This is a hash map table in which the hash buckets are moved one by one. Therefore, 342 of hash bucket element numbers 0 to 341
Hash buckets are stored in the database storage area of the designated number 0 in the database division management table,
341 to hash bucket element numbers 512 to 853
Hash buckets are stored in the database storage area of the designated number 1 in the database division management table,
Hash bucket element numbers 342-511 and 854-
This indicates that a total of 341 hash buckets up to 1023 are stored in the database storage area of designated number 2 in the database division management table. If the data is divided into one-thirds or if the data cannot be divided, adjustment is made so that more hash buckets remain in the smaller designated number of the database division management table. As a result, the amount of data to be moved can be reduced even when data is rearranged. Further, in the case of four divisions, similarly, when one database storage area is added from the database in the three division state, the added database storage area is replaced by four hash buckets of the existing three divided database storage areas. This is a hash map table in which the hash buckets are moved one by one. In this case, hash bucket element numbers 0
This indicates that 256 hash buckets up to 255 are stored in the database storage area of the designated number 0 in the database division management table, and the hash bucket element numbers 512 to 512 are stored.
This indicates that 256 hash buckets up to 767 are stored in the database storage area of the designated number 1 in the database division management table, and the hash bucket numbers 342 to 46
9, hash bucket element numbers 854 to
This indicates that a total of 256 hash buckets up to 981 are stored in the database storage area of the designated number 2 in the database division management table, and 86 hash bucket element numbers 256 to 341 and hash bucket element numbers 470 to 470 are shown. A total of 25 pieces of 42 pieces up to 511, 85 pieces of hash bucket element numbers 768 to 853, and 43 pieces of hash bucket element numbers 982 to 1023
This indicates that six hash buckets are stored in the database storage area of designated number 3 in the database division management table. In this way, the correspondence table of the hash buckets up to the maximum division number of 512 is held. Therefore, the entry of the hash map table to be referred to may be determined based on the number of database storage areas determined at the time of defining the table of the database.

Next, the segment hash map table 51, the segment bit map table 52, the page bit map table 53, and the segment 54 in FIG.
FIG. 6 shows a schematic flow of a process of inserting data into a database table of FIG. When a data insertion processing request is made to a database table, "ZAIK
Taking the "O" table as an example, the following request is issued to the database management system.

INSERT INTO ZAIKO VALUES ('101', 'Probe', 'red', 35000, 62) When a request is given, first, database definition information is obtained from the database definition information storage area 60 in FIG. (Step 410). From the acquired information, the “ZAIKO” table contains the partitioning key SCO
It turns out that the hash value is divided by the DE column. Therefore, a hash function is applied to the value '101' specified in the SCODE column (step 411). As a result of performing the hash function, the hash bucket element number in the hash map table in FIG. 4 is determined. Here, it is assumed that the hash bucket element number is 256. Since the “ZAIKO” table is given to be divided into four database storage areas, the hash bucket element number 256 of the hash map table having the number of divisions of 4 in the hash map table in FIG. It can be seen that the number 3 in the database division management table is the database storage area for storing data (step 412). By searching for a database storage area name in which the “number” column of the database division management table is 3, it is determined that the database storage area name is “DBAREA4”. Since the database storage area has been determined, the segment hash map table 51 of the “ZAIKO” table is referred to to determine the segment to be stored in the database storage area 'DBAREA4'. Segment hash map table 5
1, the entry of the array number 256 in the segment hash map table is referenced using the hash bucket element number 256 obtained earlier (step 41).
3). Here, as shown in FIG. 6, the array number 256 in the segment hash map table 66 holds the number of segments of the data having the hash bucket element number and the number of the head segment management table as the segment head pointer. In step 414, it is checked whether or not a segment is allocated by determining whether the number of segments is 0 or the number of the leading segment management table is 0. If it is 0, no segment has been allocated yet, that is,
Since no data corresponding to the hash bucket element number is stored, a process for securing a segment is performed with reference to the segment bitmap table in FIG. 1 (step 415). The segment bitmap table 52 in FIG. 1 is managed by bitmaps for all segments that can be secured in the database storage area.
An empty segment found by searching for an empty bit is secured as a new segment. The page bit map table 53 shown in FIG. 1 is used as a segment management table for managing the allocated pages of the segments.
From the segment hash map table
The number of segments is set to 1 in the 56 entries, and the secured segment number is set as the segment head pointer. In the secured segment management table,
The NEXT pointer is set to 0 to indicate that the next segment has not been allocated yet (step 4).
16). Next, a page for storing data is determined from the secured segments. In the case of a newly secured segment, all bits of the page bitmap area included in the segment management table are in an empty page state. It decides to use the first page, and updates the first bit to be used. If the segment of the corresponding hash bucket has already been secured in step 414, the chain of the segment management table is followed, and the last page number is accessed by referring to the page bitmap area of the segment management table of the last segment. The page having the last page number is input, and if there is a free area in which a record (row) to be inserted can be stored, the page is determined as an insertion page (step 417). If there is no free area, a free page is searched from the segment management table, and if there is a free page, the page is determined as an insertion page. Further, if there is no free page in the segment, a segment bitmap table is searched to secure a free segment. If there is no free segment, the insertion page is determined by tracing the chains of all the segment management tables corresponding to the hash bucket element numbers and searching for free pages from the core segment management table. Even if there is no free page, there is no free area in the database storage area, and insertion becomes impossible. Thus, the row insertion processing is completed by storing the row in the normally determined insertion page (step 418). When a row is inserted, it is, of course, necessary to acquire a log relating to the insertion processing in the database log storage area 70 in FIG.

When inserting a row as described above, a database storage area determined by a hash function is determined, and data is stored in a segment corresponding to a hash bucket element number in the database storage area. The data of the different hash bucket element numbers determined by the shocking key value are clearly stored in another segment. Further, when a request to insert a row is issued from a plurality of users, even in the same database storage area, the segment to be stored is different if the division key value is different, so that contention such as insertion into the same segment is largely suppressed. It becomes possible.

Next, FIG. 7 shows a flow of a search process when a search request is issued as a query to the database. When a data search processing request is issued to a database table, the following request is issued to the database management system using the "ZAIKO" table as an example.

SELECT SCODE, SNAME, COL, TANKA, ZSURYO FR
OM ZAIKO WHERE SNO = '101' When the above request is given, first, the database definition information is acquired from the database definition information storage area 60 in FIG. 2 (step 420). From the acquired information, the “ZAIKO” table contains the partitioning key SCO
It turns out that the hash value is divided by the DE column. Next, it is checked whether or not there is a condition designation for the SCODE column as a partitioning key in the search condition of the search request (step 421). In the case of this inquiry, S
Since the condition for the NO column is specified, the database storage area is determined by applying a hash function. As a result of performing the hash function, the hash bucket element number in the hash map table in FIG. 3 is determined.
Here, it is assumed that the hash bucket element number is 256. Since the “ZAIKO” table is given to be divided into four database storage areas, the hash bucket element number 256 of the hash map table having the number of divisions of 4 in the hash map table in FIG. It is found that the number 3 in the database division management table is the database storage area for storing data (step 423). "Number" column in the database partition management table is 3
By retrieving the database storage area name, it is determined that the database storage area name is 'DBAREA4'. Since the database storage area has been determined, the segment hash map table 51 of the “ZAIKO” table is referred to in order to determine the segment corresponding to the hash bucket element number in the database storage area 'DBAREA4'. When referring to the segment hash map table 51, the entry of the array number 256 in the segment hash map table is referenced using the hash bucket element number 256 obtained earlier (step 42).
4). When the head segment management table number set in the entry of the array number 256 of the segment hash map table exists, the search condition is evaluated by tracing the chain of the segment management table with the assigned pages of all segments as search targets ( Step 425). if,
If no segment number is set in the corresponding entry, it means that there is no search target data, and the search result is 0. If no condition is specified for the SNO column as the partitioning key in step 421, all database storage areas are to be searched. When all database storage areas are to be searched, search condition evaluation is performed while tracing the segment chain of each element number for those whose segment numbers set in the segment hash map table in each database storage area are not 0. (Steps 426 through 42
7).

As described above, in the database search, when a condition for the partitioning key is specified, not only the database storage area to be searched but also the segments in the database storage area can be narrowed. Segments that do not need to be searched can be excluded from the search.

An embodiment in which data is rearranged using the database division storage method according to the present invention will be described with reference to FIGS. As the database system configuration, the configuration shown in FIG. 2 is used, and as the definition of the database, the “ZAIKO” table shown in FIG. 3 is used, and a new database storage area “DBAREA5” is added to the “ZAIKO” table. FIG. 8 shows the flow of the database rearrangement process in the case of adding "." The database relocation processing is performed by the database management system 4 as shown in FIG.
It operates as the database relocation processing unit 430, which is one processing program in 0. First, as a database relocation processing request, the database storage area “DBAREA
When the addition of “5” is requested, the hash map table with the number of divisions including the newly added database storage area is displayed in FIG.
(Step 431). Here, since one database storage area is added to the database divided into four parts, a hash map table in the case of five parts is obtained.
Next, a list is created by extracting the hash bucket element numbers managed by the additional database storage area from the obtained hash map table (step 432). In other words, the array number with the number 4 in each array of the hash map table in the case of division into five may be extracted as the hash bucket element number. Since the hash bucket element number to be managed by the database storage area added in this way has been found, the hash bucket of the hash bucket element number extracted earlier is currently managed by referring to the hash map table of four divisions, which is the current number of divisions. The number of the stored database storage area is acquired (step 433). If the number of the database storage area is known, the database storage area name can be obtained from the database division management table 65 in the database definition information storage area 60 in FIG. A correspondence table between the obtained hash bucket element numbers to be moved to the added database storage area thus obtained and the existing database storage areas is created, and the movement target hash bucket element numbers are sorted for each existing database storage area. If a correspondence table is created in advance, processing can be performed for each existing database storage area. First, processing is started from the database storage area “DBAREA1”. This will be described with reference to FIG. From the database storage area “DBAREA1”, hash bucket element numbers 204 to 255, which are 1/5 of hash bucket element numbers 0 to 255, are to be moved.
Movement target segment extraction processing (step 43)
4), the database storage area “DBA” in FIG.
REA1 "segment hash map table 51 in 50a
With reference to a, the segment 54a is read into the segment input / output buffer 42 by tracing the chain of the segment management table in the page bitmap table 53a while extracting the segment head numbers from the hash bucket element number 204 to 255 in order. The segment I / O buffer is read in segments, but is not used by referring to the page bitmap area of each segment management table, that is, free pages are not read or the destination database is read. If the page bitmap information is also copied when the corresponding segment of the storage area is assigned, there is no problem if the page bitmap information is read in segment units. In this way, a process of reflecting the segment read into the segment input / output buffer 42 to the additional database storage area "DBAREA5" 50e is performed (step 435). The reflection process is performed in the following procedure.

(1) Since processing is performed on a segment basis, an empty bit is searched from the segment bit map table in the additional database storage area "DBAREA5" 50e.
Allocate one new segment.

(2) When a new segment for the hash bucket is newly secured, the first segment head pointer of the segment hash map table 53a is updated. At this time, 1 is added to the number of segments for the hash bucket.

(2) A segment management table for managing the secured segments is secured from the page bitmap table 53e. As the page bitmap in the secured segment management table, the contents of the page bitmap before the movement may be copied. At this time, if the segment for the hash bucket has already been allocated, the segment is chained to the end of the segment management table.

(3) The secured segment is written from the segment input / output buffer 42.

After the completion of the reflection process according to the above procedure, a process of deleting the segment of the existing database storage area "DBAREA1" 50a whose movement has been completed is performed (step 43).
6). The process of deleting the moved segment is performed in the following procedure.

(1) Segment hash map table 51a
And updates the bits of the corresponding segment in the segment bitmap table 52a to empty bits while referring to the segment head pointer of the corresponding hash bucket element number and tracing the NEXT pointer in the segment management table of the page bitmap table 53a.

(2) When the bits of the segment bitmap table 52a are updated to empty bits for all the segments of the corresponding hash bucket, the number of segments of the corresponding hash bucket element number in the segment hash map table is updated to 0 (step 437). ).

It is checked whether or not all the hash buckets to be moved in each existing database storage area have been moved in the above procedure (step 438), and steps 434 to 437 are performed up to the database storage area "DBAREA4". When the processing of all the database storage areas is completed, the database definition information storage area 6
Update the database definition information in 0 (step 43)
9). The content to be updated is obtained by adding the information of the added database storage area to the database area management table 61 and the database storage area configuration file table, and clicking “ZAIKO”.
By updating the number of divisions in the database management table 63 to 5 as information on the database storage area added to the table, and adding the added database storage area “DBAREA5” and information of 4 as a number to the database division management table 65. is there.

As described above, in the database rearrangement process, it is not necessary to read data one record at a time from each existing database storage area, but it is sufficient to read data in segments and write data to the added database storage area. Also, any hash buckets that are not to be moved,
No need to read.

Next, a data rearrangement process in a parallel database system will be described as a second embodiment relating to data rearrangement according to the present invention.

FIG. 10 is a diagram showing an example of a hardware configuration of a parallel database system. One computer system 10 includes a CPU 12, a main storage device 14, and an external storage device 20, such as a magnetic disk device, as in FIG. In the parallel database system. A plurality of computer systems and a plurality of terminals 30 are connected to a network. Each processing configuration unit is arranged in the database management system 40 configuring the parallel database system. The front-end server 41 (hereinafter, referred to as FES) is a component that receives an inquiry from the terminal 30 and generates a processing procedure for the inquiry. The FES has a database log storage area 70 for recording logs related to transactions and the like. Back-end servers 43a, 43b, 43c, 43d (hereinafter, referred to as
BES) receives the processing procedure from the FES, and stores the database storage areas 50a, 50b, 50c, and 50d managed by BES1, BES2, BES3, and BES4, respectively.
And returns the processing result to the FES. At this time, logs related to database changes and transactions related to the respective database storage areas are stored in the database log storage area 70.
a, 70b, 70c, and 70d. The data dictionary server 42 (hereinafter referred to as DIC)
The database manages definition information related to the database, and the definition information is stored in the database definition information storage area 70. Logs related to changes in the database definition information and logs related to transactions are stored in the database log storage area 70. In the FES, the definition information of the database accessed by the requested query is acquired from the DIC. FIG.
0, as shown in FIG.
A plurality of ESs can be arranged. This can increase the parallel processing capability by placing multiple BESs in one computer system. here,
The configuration is such that two BESs are arranged in each computer system. Using the hardware configuration of the parallel database system, "ZAIKO"
Database definition information storage area 70 when a table is defined
FIG. 11 shows the management information of. There is no difference in basic management information from the database definition information storage area 70 of FIG. 3 shown as an embodiment in the case of a non-parallel database system. The difference is the server name included in the database area management table 61. Information for managing which server manages each database storage area is added. In other words, the “ZAIKO” table has four database storage areas DBAREA1, DBAREA2, DBAREA
3, DBAREA4 and DBAREA1 is BE
S1, DBAREA2 is BES2, DBAREA3 is B
In ES3 and DBAREA4, the FES distributes the processing so that the database processing is performed in the BES4.

Next, a new computer system is added to the parallel database system as shown in FIG.
A description will be given of how the tactical data of the "ZAIKO" table is rearranged when the database storage area "DBAREA5" is added to the added BES5.

First, the BES 5 executes data reception processing as rearrangement processing. FIG. 13 shows a schematic processing flow of the data reception processing in the BES5. In BES5,
The hash map table in five divisions including the database storage area DBAREA5 to be added is extracted from the database definition information storage area of the DIC (step 441). Next, a list is created by extracting the hash bucket element numbers managed by the additional database storage area from the obtained hash map table (step 442). In other words, the array number with the number 4 in each array of the hash map table in the case of division into five may be extracted as the hash bucket element number. Since the hash bucket element number to be managed by the database storage area added in this way has been found, the hash bucket of the hash bucket element number extracted earlier is currently managed by referring to the hash map table of four divisions, which is the current number of divisions. The number of the database storage area and the server name are acquired from the DIC (step 443). If the number of the database storage area is known, the database storage area name can be obtained from the database division management table 65 in the database definition information storage area 60 in FIG. Further, the name of the server managing each database storage area can be obtained from the database area management table. A correspondence table between the obtained hash bucket element numbers to be moved to the added database storage area thus obtained and the existing database storage areas is created, and the movement target hash bucket element numbers are sorted for each existing database storage area. If a correspondence table is created in advance, processing can be performed for each existing database storage area. Next, from BES5, database storage area D
B that manages from BAREA1 to DBAREA4
The moving object segment extraction processing request is transmitted to the ES1 to the BES4 using the hash bucket element number to be moved and the database storage area as input information (step 444). FIG. 13 shows the flow of the data extraction processing in each BES that has received the message of the movement target segment extraction processing request. BES1 to BES4 receive the message (step 461). In each BES, the contents of the message are analyzed, and the extraction process is started with the combination of the database storage area and the hash bucket element number from which the data is extracted in the relocation. For example, in BES1, processing is started from the database storage area DBAREA1 50a. This will be described with reference to FIG. From the database storage area “DBAREA1” 50a, the hash bucket element numbers 204 to 255, which are 1/5 of the hash bucket element numbers 0 to 255, are to be moved, and the database storage area “DB
Segment hash map table 5 in “AREA1” 50a
1a, while extracting the segment head numbers from the hash bucket element number 204 to 255 in order from the hash bucket element number 204,
The segment 54a is read into the segment input buffer 45a while following the chain of the segment management table in the page bitmap table 53a (step 462).
Then, it is checked whether or not all the segments to be moved from the database storage area have been read (step 463). If the search has not been completed, the hash bucket element is sent from the segment input buffer 45a to the BES5 which transmitted the data extraction processing request. Transmission is performed together with the number (step 464). The data transmitted from BES1 is received by BES5 in the process of receiving data from each BES (step 445 in FIG. 13). At this time, it is checked whether or not the data transmission from all the BESs has been completed (step 446).
The process of reflecting the segment to be moved to the additional database storage area "DBAREA5" 50e is performed. The reflection process is performed in the following procedure.

(1) Since processing is performed on a segment basis, an empty bit is searched from the segment bit map table 52e in the additional database storage area "DBAREA5" 50e, and one new segment is secured.

(2) When a new segment for the hash bucket is newly secured, the first segment head pointer of the segment hash map table 53e is updated. Further, 1 is added to the number of segments for the hash bucket.

(2) A segment management table for managing the secured segments is secured from the page bitmap table 53e. As the page bitmap in the secured segment management table, the contents of the page bitmap before the movement may be copied. At this time, if the segment for the hash bucket has already been allocated, the segment is chained to the end of the segment management table.

(3) The secured segment is written from the segment output buffer 43a.

In this data receiving process, segment output buffers 46a to 46d are prepared so that data can be received from each BES. Each time one data reception process is completed, the transmission destination is notified of the completion of the received segment reflection process (step 448). On the other hand, the BES1 receives the notification by receiving the segment reflection completion notification in step 465 of FIG. When the segment reflection completion notification is received, the moved segment is deleted (step 466). The process of deleting the moved segment is performed in the following procedure.

(1) Segment hash map table 51a
And updates the bits of the corresponding segment in the segment bitmap table 52a to empty bits while referring to the segment head pointer of the corresponding hash bucket element number and tracing the NEXT pointer in the segment management table of the page bitmap table 53a.

(2) When the bits of the segment bitmap table 52a are updated to empty bits for all the segments of the corresponding hash bucket, the number of segments of the corresponding hash bucket element number in the segment hash map table is updated to 0 (step 467). ).

In the above procedure, the BES on the data extraction side repeats steps 462 to 467, and completes transmission of all segments to be moved (step 46).
3), B indicates that the data transmission from the BES has been completed.
Notify ES5 (step 468).

On the other hand, the step 445 is also performed in the BES5.
To 448 are repeated, and when the data transmission completion notification is received from all the BESs from BES1 to BES4 (step 446), the database definition information of the DIC is updated (step 449). The contents to be updated are obtained by adding the information of the added database storage area to the database area management table 61 and the database storage area configuration file table, and dividing the database management table 63 as information of the database storage area added to the “ZAIKO” table. The number is updated to 5, and the added database storage area “DBAREA5” and information of 4 as a number are added to the database division management table 65.

As described above, in the parallel database system as well, it is not necessary to read data one record at a time from each existing database storage area of each BES. And write processing may be performed. Also, there is no need to read any hash buckets that are not to be moved.

Next, a second embodiment of the database dividing method according to the present invention will be described with reference to FIGS.

FIG. 15 shows a state in which key range partitioning is performed as a first partitioning method of a table, and partitions partitioned by each key range are hash partitioned.
The above-described "ZAIKO" table is taken as an example, and the dividing method is changed. FIG. 16 shows the management information in the database definition information storage area at this time. The difference from the management information of the database definition information storage area shown in FIG. 11 is a database management table 63 and a database division management table 65. As the definition of the table of "ZAIKO" table,
It is assumed that the condition for dividing the key range into the “SCODE” column is given, and that the “SCODE” column or another column may be hash-divided.
The database management table 63 holds a value of “RANGE” indicating that the partitioning method is the key range partitioning, and “SCODD” as the first partitioning key.
A column number 1 for indicating a column "E" is held, and a column number 1 for indicating a column "SCODE" is held as a second sub-partitioning key. Then, the serial number of the database storage area corresponding to the first partition and the second
Indicates that the serial number of the database storage area corresponding to the sub-partition is held, and that each division condition is held. For example, when the division condition is “C1> 100”, the serial number of the database storage area is 0, and “DBAREA1” is set as the database storage area in that partition.
"1", "DBAREA12", and "DBAREA13" are designated and further assigned as sub-numbers 0, 1, and 2 by a hash function. This can be considered similarly to the case where the first division condition is hash division. Therefore, if the first partition is divided into hash buckets in advance, when adding a database storage area to each partition, a hash map table corresponding to the number of divisions in each partition may be selected. become.

As described above, according to this embodiment,
Since the database data is divided into a plurality of fixed hash buckets in advance and the hash buckets are stored in physically independent segments in the database storage area, data is relocated to the newly added database storage area Sometimes, only the data of the hash bucket to be moved can be the relocation target.

Further, since the segment moved by the relocation processing is released, it can be reused as a storage segment of another hash bucket, and there is no waste in the space of the database storage area.

[0058]

As described above, according to the present invention,
Since the database data is divided into a plurality of fixed hash buckets in advance and the hash buckets are stored in physically independent segments in the database storage area, data is relocated to the newly added database storage area Sometimes, only the data of the hash bucket to be moved can be the relocation target.

[Brief description of the drawings]

FIG. 1 shows a configuration of a database storage area which is a feature of the present invention.

FIG. 2 shows a configuration diagram of a database management system for implementing the present invention.

FIG. 3 shows information of a table for managing database definition information.

FIG. 4 shows a hash map table which is a feature of the present invention.

FIG. 5 illustrates a schematic flow of a process of inserting data into a table according to the embodiment.

FIG. 6 shows a relation between a segment map table and a page bitmap table according to the present embodiment.

FIG. 7 shows a schematic process flow of a table search process according to the embodiment.

FIG. 8 shows a schematic processing flow of a database relocation processing according to the present invention.

FIG. 9 shows a data flow of a database relocation process according to the present invention.

FIG. 10 is a configuration diagram of a parallel database system for implementing the present invention.

FIG. 11 shows information of a table for managing database definition information in a parallel database system.

FIG. 12 shows a data flow of a database relocation process in the parallel database system.

FIG. 13 shows a schematic processing flow of data reception processing for database relocation according to the present invention.

FIG. 14 shows a schematic processing flow of data transmission processing for database relocation according to the present invention.

FIG. 15 shows an example of a table in the case of key range division according to the present invention.

FIG. 16 shows information of a table for managing database definition information of a table divided into key ranges.

[Explanation of symbols]

10: Computer system, 12: CPU, 14: Main storage device, 20: External storage device, 30: Terminal, 40: Database management system, 50: Database storage area, 51: Segment hash map table, 52: Segment bit map table , 53: Page bitmap table, 5
4: segment, 61: database storage area management table,
62: database storage area configuration management table, 63: database management table, 64: database configuration column management table, 6
5: database division management table, 66: hash map table, 60: database definition information storage area, 70: database log storage area.

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Ryuichi Hoshino 5030 Totsuka-cho, Totsuka-ku, Yokohama-shi, Kanagawa Prefecture Within Hitachi Software Works, Ltd. (72) Inventor Hideaki Kasao 6-81 Onoecho, Naka-ku, Yokohama-shi, Kanagawa Address Hitachi Software Engineering Co., Ltd. In-house F-term (reference) 5B075 NK45 NR03 5B082 BA09 CA11 EA01 EA02

Claims (7)

[Claims] 1. A database management system for managing a database storage area composed of one or a plurality of external storage devices, wherein the database storage area manages a plurality of segments composed of a plurality of physically fixed-length pages. In the method, when m database storage areas are provided as database storage areas, one or more data items of the database are used as a partitioning key, and the database is applied with a hash function on the partitioning key. Then n
(M ≦ n) buckets, which are logical units, are divided into buckets, and a hash map table that determines correspondence between database storage areas that manage each bucket according to the given number of database storage areas; A database management method comprising: managing a database by a segment hash map table for mapping a set bucket to a segment which is a storage unit in each database storage area. 2. The database management method according to claim 1, wherein: when a key range division is designated for a partitioning key; when a database storage area is provided for each key range; Using the same data item or a data item different from the partitioning key as a second partitioning key, applying a hash function to the database for the second partitioning key, and n buckets which are n logical units And managing the database using a segment hash map table for mapping the buckets to segments that are storage units in each database storage area. 3. The database management method according to claim 1, wherein when the addition of the database storage area is requested, the total number of buckets is calculated by adding the existing database storage area and the added database storage area to the total number of database storage areas. Based on the number of buckets in each database storage area when buckets are equally divided, a bucket larger than the number of buckets is selected from each existing database area, and the number of buckets in the database storage area to which the selected bucket has been added is selected. Updating the hash map table so as to relocate so as to satisfy, the bucket selected as a migration target from each database area is referred to the segment hash map table, and transferred to the database storage area added in segment units; Data transfer to the added database area is completed And deleting the transferred segment group in the transferred database storage area. 4. The database management method according to claim 1, wherein the number of buckets divided by a hash function can be set by a user's specification. 5. The database management method according to claim 3, wherein when data is relocated by adding a database storage area, when a group of segments for which data transfer has been completed is deleted, a copy of said segment is updated to a pre-update log. Without acquiring as, the bit of the deleted segment is rewritten to an empty state with respect to the segment bitmap table for managing the segment in the database storage area, and the pre-update state and the post-update state of the bit are obtained as an update log. A database management method for recovering by deleting a pre-update log for the deleted segment into the segment bitmap table, if the failure occurs before the data relocation is completed. 6. The database management method according to claim 1, wherein when the addition of the database storage area is requested, the total number of segments of the hash bucket included in each existing database storage area is calculated. By calculating the average number of segments in the database storage area of the database and calculating the average number of segments in the total number of database storage areas including the number of added database storage areas, the calculated average number of segments in all the existing database storage areas is calculated. And calculate the difference in the average number of segments with the total number of database storage areas including the added number of database storage areas, and add a hash bucket from each of the existing database storage areas so as to approximate the calculated number of segments. Data to a new database storage area Database management method characterized in that the Rukoto updates the hash map table by the hash bucket as a target of rearrangement. 7. The database management method according to claim 2, wherein when a database storage area is requested to be added to a certain key range partition, the total number of buckets in said key range partition is changed to said key range partition. Based on the number of database storage areas to be written when buckets are equally divided by the total number of database storage areas including the existing database storage area and the added database storage area, Select a large number of buckets, update the hash map table so that the selected buckets are relocated so as to satisfy the number of buckets in the database storage area that has been added for a thousand years, Referring to the segment hash map table, Database management method characterized by transferred to the database storage area added in Units, it deletes the transferred segment group of database storage in a region where the added data transfer in the database area is the transfer to be completed.