KR100194589B1 - T-Tree Rollback Method in Distributed Main Memory Database Management System - Google Patents

Abstract

The present invention relates to a method for rolling back a T-tree index structure in a distributed main memory database management system (DBMS) of an ATM switch system. The present invention relates to a process in which a DBMS of an ATM switch, which is a distributed real-time system, processes a user's query request. Responds to a user's request to abort a transaction more quickly because the T-tree rollback process rolls back the meta-database and relations and then rolls back the meta-database and relations using the previously written log in reverse when a transaction aborts. This can improve the performance of the exchange application.

Description

T-Tree Rollback Method in Distributed Main Memory Database Management System (DBMS)

1 is a structural diagram of a main memory DBMS having a distributed structure of an ATM switch system applied in the present invention.

2 is a block diagram of a rollback queue having a structure of a circular queue in the present invention.

3 shows the configuration of the MMDB according to the present invention.

4 is a flowchart in which a process executing an application program that accesses a database of an ATM switch system of the present invention performs a rollback of an MMDB upon aborting a transaction.

5 is a flowchart illustrating a process of reconstructing a T-tree index structure in a database of an ATM switch system of the present invention.

* Explanation of symbols for main parts of the drawings

1: Meta Database Area 2: Relation Area

3: T-tree structure area 10, 221, 321: Local database (LDB)

20, 222, 322: Database Management System (DBMS)

21: DBK (Datebase Kernel Group) 22: DBBG (Datebase Backup Group)

23: DBA (Datebase Supporting Group) 24: DBTG (Datebase Transaction Group)

25: DBQG (Datebase Query Group) 100: Central Exchange Subsystem (ACS)

110: operation maintenance processor (OMP) 115: Inter-Module Interface Link

120: central switch network module (CSNM) 130: disk (disk)

200, 300: connection exchange subsystem (ALS / S0, ALS / S1) 210, 310: access switch network module (ASNM)

The present invention relates to a T-tree rollback method in a distributed main memory database management system (DBMS). In particular, the DBMS of an ATM exchange, which is a distributed real-time system, processes a user's query request. ), A method of maintaining a T-tree rollback process in a DBMS to effectively roll back the T-tree index structure after rolling back the meta database and relation by applying the previously written log in reverse.

A main memory database management system (DBMS) operating in a conventional distributed environment processes in a transaction unit in the process of supporting data manipulation of applications that perform call processing, operation, maintenance, billing and statistics functions. .

There is also an abort that causes a transaction to fail successfully and rolls back, meaning that all data operations within the transaction are returned to their pre-transaction state.

The general main database management system provides the T-tree structure as the most effective index structure to access the database.

When updating the contents in the database in the application program, all information about the changed contents of the main memory, including the T-tree structure, is stored in the main memory separately after the transaction is completed. Move the contents to disk and use the contents to roll back the database when the transaction aborts.

The storage of information about changes in the database within a transaction has been studied using various techniques such as using shadow pages or storing logs using stable memory.

However, the creation of information about data changes is common in that it does not distinguish between the T-tree structure and the actual user's meta database and relations.

Therefore, when the transaction is aborted, rolling back by reflecting the stored information is completed to finish the rollback of the T-tree structure, meta database, and relation.

Unlike the structure of a general distributed system, however, the ATM switch system does not use stable storage in each subsystem and the main memory area for the database is not large enough to use the shadow area and perform in-place update.

In addition, since there is a disk only in one subsystem because of the structural characteristics of the exchange, there is a burden that the processor of the subsystem receives a log from all the subsystems distributed in the entire system and stores the disk in the disk. Therefore, the DBMS for the ATM exchange has a main memory meta database, a relation and a T-tree area, and only a meta database and a relation on the disk.

Logs for changes in the main memory database are created only for changes in the meta database and relation area.

In the distributed database, whenever a transaction is stopped, the log is used to roll back only the meta database and the relation.

Therefore, since there is no log for the T-tree structure, an effective method for rolling back only the T-tree structure is urgently needed.

Accordingly, the present invention has been proposed to roll back the T-tree index structure when a DBMS of an ATM exchange, which is a distributed real-time system, processes a user's query request, thereby responding to the user's transaction suspension request more quickly. Its purpose is to provide a T-tree rollback method in a distributed main memory database management system (DBMS) that improves the performance of applications.

In order to achieve the above object, the present invention provides a T-T when a transaction is interrupted by an application program accessing a database management system (DBMS) having an area of a meta database, a relation, and a T-tree in a main memory in an ATM exchange system. A T-tree rollback method for rolling back relation information having a tree index structure to a rollback queue, wherein the DBMS recovers a meta database and a relation to a state before the transaction occurs by using a log created during a transaction of a user application. Storing the restored relation in the rollback queue if the recovered relation is relation information having a T-tree index structure, and preventing other applications from accessing the process until the T-tree rollback is completed for the relation. There is a characteristic consisting of steps.

Another feature of the present invention is to stop the rollback process for a predetermined period of time when there is no relation information to perform a rollback in the rollback queue, and then examine the rollback queue again, and if the relation information exists in the rollback queue, Fetching, rolling back the T-tree index structure of the relation using the retrieved information, and terminating access to the relation.

As such, the present invention can increase the transaction processing speed of an application program and thereby increase the performance of the exchange system.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

1 is a configuration diagram of a main memory DBMS having a distributed structure of an ATM switch system.

The ATM switch has two subsystems: an ATM Centrl Switching Subsystem (ACS) 100 and an ATM Local Switching Subsystem 200 (ALS).

The central switching subsystem 100 performs a function of operating, maintaining, and managing the system, and maintaining various traffic information, statistics, and billing information, and includes an operation and maintenance processor (OMP) 110. have.

The operation maintenance processor (OMP) 110 includes a local database (LDB) 10 and a main memory database DBMS 20 to be described later.

Access exchange subsystem (200, 300) is a processor that performs the access processing of the general subscriber, and performs the access control according to the subscriber service request together with the subscriber matching circuit, traffic control such as access acceptance control, congestion control, etc. Perform the function. In addition, the connection exchange subsystems 200 and 300 are installed in plurality according to the subscribers, and include call connection control processors (CCCPs) 220 and 320.

On the other hand, the central switch network module (CSNM) 120 provided in the central switching subsystem 100 is connected to each access switch network of each of the connection switching subsystems (ALS / S0, S1; 200, 300). It is responsible for the switching for the module ASNM (Acess Switch Network Module) (210, 310).

And, the access switch network module (ASNM, 210, 310) is responsible for the subscriber switch in each ALS (210, 310), the operation maintenance processor 110 is connected to one port of the CSNM (120) and each call connection control processor ( 220 and 320 are connected to one port of each access switch network module (ASNM) 210 and 310 through an inter-module interface (IMI) link 115 to transmit and receive messages.

At this time, the communication between the call connection control processor (220, 320) is a three-stage structure of ASNM-CSNM-ASNM, the communication between the operation maintenance processor (OMP, 110) and the call connection control processor (220, 320) is CSNM-ASNM It is a two-stage structure.

ATM DBMS is distributed to each processor to manage data distributed to multiple processors, resides in main memory for real time processing, retrieves and modifies data, remote data processing, duplicate data processing, data backup It provides features, data recovery, concurrency control, and transaction management.

Among the above-described processors, the DBMS 20 in the operation preservation processor (OMP) 110 may include a Database Kernel Group (DBKG) 21, a Database Backup Group (DBBG) 22, and a Database Supporting Group (DBSG) 23. 5 blocks of DBTG (DataBase Transaction Group) 24 and DBQG (DataBase Query Group) 25, and DBMSs 22 and 322 of call connection processors 220 and 320 are DBKG (21). ), DBSG 23, and DBTG 24 are provided with three blocks.

The difference between the operation preservation processor 110 and the call connection control processor 220 or 320 is that the disk is mounted and performs a function of receiving an input from an operator. Therefore, only the operation maintenance processor is provided with a DBBG 22 block for accessing the disk 130 and a DBQG 25 block for interfacing the operator terminal with the ATM switch system.

Each block 20, 222, 322 of the DBMS is responsible for receiving a request to manipulate data from requests input directly from an application program or a user, and returning the result.

The DBKG 21 is actually responsible for retrieving, changing, concurrency control, generating log records, and the like, of data in the database according to the retrieval and change of the data for which the processing request is received.

The DBSG 23 performs a function of requesting the DBKG 21 to handle retrieval and change requests for remote data in an application, and a communication function between DBMSs of other processors for remote data processing, and the DBBG 22 If the data requesting a change requires a backup of the disk 130, it is responsible for backing up the database on the disk and recovering the database upon system restart.

The DBTG 24 performs a transaction management function, and the DBQG 25 performs an on-line interactive query processing function occurring on an operator terminal or a PC.

2 shows the configuration of a rollback queue having a structure of a circular queue.

Rollback queues should be placed in the critical section to ensure exclusive access from each processor. The rollback queue consists of an array of data structures containing information of relations to be rolled back, an insertion location information field indicating its position when inserted into the queue, and a retrieval location information field indicating information first inserted into the queue when information is retrieved from the queue. .

3 is a structural diagram showing the configuration of the MMDB residing in the main memory of each subsystem (ACS and a plurality of ALS) of the ATM switching system, which is a distributed real-time system.

The MMDB is divided into a meta database area 1, a relation (tuple database) area 2, and an area 3 of the T-tree index structure.

This T-tree index structure is an index structure that DBMSs 20, 222, and 322, rather than data inserted by a user, separately configure to search for tuple data faster, and provides the most effective search speed in the main memory database.

In the present invention, the T-tree index structure is rolled back as in the accompanying drawings.

4 is a flowchart illustrating a process of performing a rollback of an MMDB by a processor executing an application program that accesses a database of an ATM switch system.

First, the log created during the transaction of the user application in the DBMS is applied in reverse to restore the meta database and the relation to the state before the transaction occurs (S1).

If the relation updated in the transaction has a T-tree index structure (S2), if the relation is a T-tree relation, the T-tree structure must also be rolled back. Therefore, the relation information is stored by inserting it into the rollback queue (S3).

Thereafter, the relation is marked so that no application program can access it until the T-tree rollback is completed (S4).

However, if it is not relation information having a T-tree index structure that has been updated in the transaction by the investigation S2, the execution is completed.

5 is a flowchart illustrating a process of performing a rollback process to reconstruct a T-tree index structure in a database of an ATM exchange system.

The rollback process first accesses the rollback queue and checks whether information exists (S11).

If there is no information on the relation to perform the rollback in the rollback queue, the process is paused (sleeped) for several seconds (S12) and the process of examining the rollback queue again is repeated.

On the other hand, if information exists in the rollback queue, the first inserted information is extracted (S13). The T-tree structure of the relation is rolled back using the retrieved information (S14). Thereafter, the state that the access is prohibited to the relation is terminated (S15).

As described above, the present invention does not create a log for the T-tree index structure when processing a transaction, and thus the processing speed is fast, and the log processing speed when a transaction is completed is also fast. However, if the process that executes the transaction is stopped, it will not be able to satisfy the real-time execution requirement if it is responsible for rolling back the T-tree.

This process of executing transactions is a sure way to increase the transaction processing speed of the application program and thereby increase the performance of the exchange system by completing the execution by storing only the relevant information without directly rolling back the T-tree.

In addition, the rollback of the T-tree index structure is a dedicated process, and the rollback process is responsible for rolling back the T-tree index structure independently of the application executing the transaction, thus minimizing the resources such as main memory required during rollback. You can do a rollback effectively.

Claims (5)

Rollback queue of relation information with a T-tree index structure in the event of aborting a transaction performed by an application program accessing a database management system (DBMS) having a database of meta databases, relations and T-trees in main memory in the ATM switch system. A method for rolling back a T-tree, the method comprising: recovering a meta database and a relation to a state before the transaction occurs by using a log created during a transaction of a user application in the DBMS; In the case of relation information having a T-tree index structure, a second step of storing the information in a rollback queue and a third step of prohibiting access of other application programs until the T-tree rollback is completed for the corresponding relation are included. When managing distributed main memory database T-Tree Rollback Method in DBMS. The distributed main memory database management system (DBMS) according to claim 1, wherein the T-tree index structure maintains relational information to be rolled back using the rollback queue in a critical region where mutually exclusive access to processes is possible. T-tree rollback method). At the ATM switch, roll back relation information with a T-tree index structure in the event of an interruption of a transaction performed by an application program accessing a database management system (DBMS) that has a meta database, a relation, and a T-tree area in main memory. In the T-Tree rollback method for rolling back to the first step, if there is no relation information to roll back to the rollback queue, the first step of stopping the rollback process for a predetermined time and examining the rollback queue again, the relation information in the rollback queue A second step of retrieving the information, if present, a third step of rolling back the T-tree index structure of the relation using the retrieved information, and a fourth step of terminating access to the relation; T-tree rollback method in a distributed main memory database management system (DBMS), characterized in that. 4. The T-tree rollback method of claim 3, wherein the second step retrieves relation information rolled back from a rollback queue. 4. The distributed main memory database management system (DBMS) according to claim 3, wherein the T-tree index structure maintains relational information to be rolled back using the rollback queue in a critical area that can be mutually exclusively accessed between processes. T-tree rollback method).