JPH06119227A - Distributed data base control system - Google Patents

Abstract

PURPOSE:To shorten the updating time of a distributed data base. CONSTITUTION:In the case of updating a data base 15 by a computer 1, a copy kind designating means 73 is referred to, and when it shows that the data base 15 is main, only the data base 15 is updated by an updating means 34, and a data base 25, etc., are not updated. Also, when the copy kind designating means 73 shows that the data base 15 is subsidiary, updating of its data base is requested to the computer in which the main data base is stored through a communication controller 31 by the updating means 34. The subsidiary data base 15 is not updated. On the other hand, when a synchronous time designated by a synchronous time designating means 74 comes, the computer in which the main data base is contained sends updating history of the data base to all the computers in which the subsidiary data base is stored, and requests updating.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a distributed database control system for controlling duplicate stored data in a distributed database.

[0002]

2. Description of the Related Art FIG. 2 is a block diagram showing an example of functions of a conventional distributed database control system (distributed DBMS). 3 is an explanatory diagram of the structure of the stored information management table in FIG. The distributed database management system shown in the figure has a plurality of computers 41 connected by a communication network 45,
Multiple distributed database management systems 50, 6 on 42
It consists of zero. For example, when a user requests a database search from the computer 41, the inquiry processing function 51 checks the contents of the stored information management table 54. Then, if the corresponding data is stored in the own database 55, using the search function 52, for example,
The contents of 6 are responded.

When a user requests a database update from the computer 41, the inquiry processing function 51 checks the contents of the stored information management table 54, updates the table 56 of its own database 55, and updates the corresponding table 66. The update is also instructed to the stored computer 42. Receiving this, the computer 42 uses the update function 63 to update the contents of the table 66 and responds to the computer 41 which has issued this command that the update has been completed. Even when there is an update request from the computer 42, the inquiry processing function 61 as in the computer 41.
Examines the contents of the stored information table 64, updates the table 66 of its own database 65, and also instructs the computer 41 storing the corresponding table 56 to update. In a database, a set of processes for a series of databases is called a transaction. For example, in the business processing of a bank or the like, the transfer of money from Account A to Account B is
And account B updates. That is, the remittance transaction consists of two processes, a withdrawal process from the account A and a deposit process into the account B. If either one of these two processes is unsuccessful, the deposit balance will be incorrect for the entire bank. Therefore, in the database, when one process cannot be executed successfully, the other process is rolled back to keep the state of the database normal. Therefore, when a transaction commit request is not received, the rollback request for invalidating a part of the update processing performed in the transaction can be issued when the notification of the normal end of the transaction is not received.

That is, for example, from the user on the computer 41,
When a transaction commit request is issued, the query processing function 51 commands the distributed DBMS 60 of the computer 42 related to the transaction to prepare for commit.
If all the responses to this commit preparation instruction are normal, the commit is issued to the distributed DBMS 60 related to the transaction. If even one response to this commit command is abnormal, the distributed DBMS 60 related to the transaction is commanded to roll back. In this way, when an update in one transaction is performed on a plurality of computers 41, 42, the distributed DBMSs 50, 60 exchange commit preparations and commits twice in order to synchronize the updates. What we do is called a two-phase commit system. The two-phase commit system will be described below.

FIG. 4 is an explanatory diagram of commitment control, FIG. 5 is an explanatory diagram of subtransactions in a distributed database, FIG. 6 is an explanatory diagram when the two-phase commit method is normal, and FIG. 7 is two-phase. It is explanatory drawing of the failure example of the sub-transaction by a commit method. First of all, I will explain about transaction management.
For example, for remittance processing for banking operations, from A account to B
The transfer to an account consists of two subtransactions: withdrawal from A account and deposit into B account. When executing such a transaction, assuming that only the withdrawal from the A account (“update 1” in FIG. 4) is completed and the deposit into the B account (“update 2” in FIG. 4) is not completed due to an obstacle, The contents of the database of the bank as a whole will go wrong. Therefore, from the start of the transaction, "Update 1",
The process of "update 2" is stored in another file for temporary storage, and after completion of both subtransactions, the database is updated by commit. This is called commitment control. When only "update 1" is completed and "update 2" becomes an obstacle, "update 1" is not reflected in the database.

In the above example, in the distributed database, the respective sub-transactions are executed by the respective computers installed in a distributed manner via the communication line. Thus, in the distributed database, as shown in FIG. 5, a transaction for the user is realized as a set of sub-transactions executed by each computer. If the commitment control described above is performed in the distributed database,
As shown in FIGS. 6 and 7, “sub-transaction 1” and “sub-transaction 2” will be committed by different computers. In this case, the master computer, which is the transaction requesting side, first issues a commit preparation instruction to both computers, and after receiving a response of completion of preparation from both computers, issues a commit instruction to each computer. This is called a two-phase commit method. When the two-phase commit method is normally performed, as shown in FIG. 6, the preparation completion response to the commit preparation instruction is obtained from both computers, and the commit instruction is issued. On the other hand, FIG.
As shown in, in the two-phase commit method, for example, when the computer that executes “sub-transaction 2” fails in commit preparation, an abort instruction is issued to each computer,
Stop execution on both computers.

[0007]

However, the above-mentioned conventional technique has the following problems. That is, as described above, in the conventional distributed database, when the same data is stored in a plurality of databases, a plurality of distributed DBMSs are used.
It is necessary to issue an update command to multiple distributed DBs by the two-phase commit system in order to synchronize the update.
I had to communicate the prepare for commit and the commit request to the MS. So in such a system,
This requires a lot of communication between computers, which causes a decrease in data processing performance. The present invention has been made by paying attention to the above points, and when storing the same data in a plurality of databases, it is not always necessary to update the same data in a plurality of databases in synchronization, The purpose of the present invention is to provide a distributed database control system with high performance.

[0008]

DISCLOSURE OF THE INVENTION A distributed database control system of the present invention stores a database in which the same data is redundantly distributed and stored in a plurality of computers connected by a communication line, and stores the same data in each of the databases. For the copy of the data, the copy type designating means for designating a part of the copy as the primary copy and the other copy other than the primary copy as the slave copy, and the update request for the data, An update unit that updates only the main copy on a computer having a database storing the main copy and stores the update history, and a computer that stores the main copy updates the update history with respect to the computer that stores the slave copy. Send
And a synchronization timing designating unit for designating a timing when the slave copy is updated to the same content as the primary copy according to the update history.

[0009]

In the distributed database control system of the present invention, when the distributed database is updated by the computer, when the data such as the table stored in each database are duplicated, the update unit makes the main copy to the copy type designation unit. Only the database specified as is updated. Therefore, when an update request occurs on the computer that stores the primary copy, only your database is updated, and when an update request occurs on the computer that stores the slave copy, the main copy is stored. You will be asked to update the calculator.
When a transaction consisting of multiple updates is committed (executed), only the main copy is updated for each update. When some of the multiple updates that make up a transaction have failed, only the update history of the main copy is rolled back when it must be rolled back. On the other hand, the synchronization timing designating means designates the synchronization timing for matching the primary copy and the secondary copy. When this synchronization time comes, the computer storing the primary copy sends the update history of the primary copy to the computer storing each slave copy. Upon receiving the update history, the computer storing each slave copy updates each slave copy by the updating means.

[0010]

Embodiments of the present invention will now be described in detail with reference to the drawings. FIG. 1 is a block diagram of an embodiment of the distributed database control system of the present invention. The illustrated system is composed of computers 1, 2 and so on. The computers 1 and 2 are connected by a communication network 5. Each of the computers 1 and 2 has processors 30, 35, communication control devices 31, 36, and a main storage device 3.
2, 37, magnetic disks 33, 38 and the like. The processors 30 and 35 perform update processing of the databases 15 and 25 stored in the magnetic disks 33 and 38, respectively. The processors 30 and 35 have updating means 34 for this purpose.
Is equipped with. The updating means 34, 39 are composed of an input / output program of a magnetic disk, etc. When issuing a request for updating the database of another computer, the updating means 34, 39 are instructed to that effect to the communication control devices 31, 36.

The communication control devices 31 and 36 control communication between the computers 1 and 2 via the communication network 5. When receiving a database update request for another computer from the update means 34, 37, the communication control devices 31, 36 transmit the update request to the corresponding computer. Main memory 32,
The numeral 37 is composed of a random access memory or the like, and temporarily stores the data stored in the magnetic disks 33 and 38. The stored information management tables 14 and 24 for managing the databases 15 and 25 are resident in the main storage devices 32 and 37. The stored information management tables 14 and 24 respectively include copy type designating means 73 and 83 and synchronization time designating means 74 and 84.

The copy type designating means 73, 83 designates a part of the duplicate copies of the data stored in the databases 15, 25 as the main copy, and copies other than the main copy. Is designated as a slave copy. The synchronization timing designating means 74, 84 designates the timing when the computer storing the primary copy sends the update history to the computer storing the secondary copy, and the secondary copy is updated to the same content as the primary copy by the update history. . Magnetic disks 33, 37
Stores various control programs and processing programs in addition to the databases 15 and 25. Next, the function and operation of the system of the present invention will be described.

FIG. 8 is a functional block diagram of the distributed database control system of the present invention, and FIG. 9 is an explanatory diagram of the structure of the stored information management table in FIG. In addition, FIG.
11 is a flowchart illustrating a database updating procedure, and FIG. 11 is a flowchart illustrating a database synchronizing procedure. When the user makes a search request to the database, the same process as the conventional process is performed. In the present invention, as shown in FIG. 9, a column of copy type and a column of synchronization time are added to the stored information management table 14. These constitute copy type designating means 73 and synchronization time designating means 74. (A) When the user makes an update request to the database (step S1 in FIG. 10), the inquiry processing function 11 checks the contents of the stored information management tables 14 and 24 (steps S2 and S3). (1) If the copy type column 73 of the table to be updated has the same rating, the same process as the conventional process is performed (step S4).

(2) If the copy type column 73 of the table to be updated is "main", the database managed by the distributed DBMS is updated according to the request (step S5). Then, which record of which table is updated is recorded in the table update history 17 (step S6). (3) If the copy type column 73 of the table to be updated is “subordinate”, the database update request is transferred to the “main” distributed DBMS (steps S7 and S8). The "main" distributed DBMS processes the forwarded update request in the same way as an update request coming from a user. That is, the database 15 is updated (step S5), which record of which table is updated is recorded in the table update history 17 (step S6), and the result is returned to the distributed DBMS that requested the update. on the other hand,
Distributed DB that received update results from the "main" distributed DBMS
The MS responds to this with the user.

(B) When the user requests the transaction to be committed, the same processing as the conventional system is performed. That is, the inquiry processing function 11 commands the distributed DBMS of the computer related to the transaction to prepare for commit. If all responses to this commit preparation command are normal, the distributed DB related to the transaction
Command the MS to commit. For this commit command,
The database is updated only for the "main" database (steps S5 and S8). If even one response to this commit command is abnormal, the distributed DBMS related to the transaction is instructed to roll back. In this case, the database is updated only for the "main" database in response to this rollback command (steps S5 and S8).

(C) When the update function 11 reaches the time written in the synchronization time 21 of the stored information management table 14 (see FIG. 1).
1 step S11), according to the contents of the table update history 17, all the history of updates to the table made so far,
Distributed DBMS of the computer whose copy type 73 is "subordinate"
To (steps S12 and S13). Upon receiving this, the inquiry processing function of the distributed DBMS updates the database by the update function (step S9 in FIG. 10), and responds that it has normally completed. When all the "subordinate" copies are completed, the update history of the table is deleted from the table update history.

[0017]

As described above, according to the distributed database control system of the present invention, when the update request is received from the user, only the database storing the main copy is updated and the subordinate copy is updated. Since the stored databases are collectively and periodically updated, even if the databases are redundantly stored, the databases can be updated in the same time as when they are not redundantly stored. Further, when it is desired to keep the values of duplicated stored data always the same as in the conventional distributed DBMS, the update can be synchronized as in the conventional case.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of a distributed database control system of the present invention.

FIG. 2 is a functional block diagram of a conventional distributed database control system.

FIG. 3 is an explanatory diagram of a structure of a conventional stored information management table.

FIG. 4 is an explanatory diagram of commitment control.

FIG. 5 is an explanatory diagram of sub-transactions in the distributed database.

FIG. 6 is an explanatory diagram when the two-phase commit method is normal.

FIG. 7 is an explanatory diagram of a failure example of a sub-transaction in the two-phase commit method.

FIG. 8 is a functional block diagram of the distributed database control system of the present invention.

FIG. 9 is an explanatory diagram of a structure of a stored information management table of the present invention.

FIG. 10 is a flowchart illustrating a database update procedure.

FIG. 11 is a flowchart illustrating a database synchronization procedure.

[Explanation of symbols]

 1, 2 Computers 14, 24 Storage information management table 15, 25 Databases 34, 39 Update means 73 Copy type designation means 74 Synchronization timing designation means

Claims (1)

[Claims] 1. A database in which the same data is redundantly distributed and stored in a plurality of computers connected by a communication line, and a part of the copy of the data stored in each of the databases is duplicated. For a copy type designating unit that designates a main copy and designates a copy other than the main copy as a slave copy, and for an update request for the data, a computer that has a database storing the main copy is used for the master copy. An updating unit that updates only the copy and saves the update history, and a computer that stores the primary copy sends the update history to the computer that stores the slave copy, and the slave copy uses the update history as the master copy. A distributed database control system, comprising: a synchronization timing designating means for designating a timing of updating to the same content as a copy.