JPH077351B2 - How to end transaction processing - Google Patents

Description

The present invention relates to a transaction processing termination method for reducing the number of communications between a plurality of nodes in a distributed system connected by a network.

[Conventional technology]

Conventionally, a network line is connected from one node to each of the other nodes, and transaction processing is performed in a one-to-n relationship. In addition, when performing transaction processing of 1 node and 1 node, since the two-phase commit method is adopted in which the commit processing is executed while confirming each other's nodes, commit from 1 node to another 1 node After the processing is executed, on the contrary, the commit processing is executed from one other node to one node, and after the commit processing is completed with each other, one transaction processing is executed by the end processing from one node and vice versa. It was finished.

However, this requires communication between nodes for commit processing and end processing for each processing between nodes, and when data is not updated at all or when it is performed by only one node, other nodes Eliminates the need for communication of commit processing with.

As a technique for solving such a problem, for example, JP-A-63
There is a processing method disclosed in Japanese Patent Publication No.-259742.

This is because if there is no need for commit processing for the above reasons, a table that manages whether updates have been made beforehand is provided, and by referencing this table, only commit processing is executed without executing commit processing. Yes, the number of updates is reduced.

[Problems to be Solved by the Invention]

In a conventional distributed system using a network, a line is connected from one node to each of a plurality of other nodes, so transaction processing is performed in a one-to-n relationship, and the processing method is a two-phase commit method. The number of communications at two nodes (probably A node and B node) is A
Commit processing from node to B node, conversely from B node to A node, A after completion of commit processing
End processing from node to B node, conversely from B node to A
The end processing to the node is 4 times in total, and in the case of transaction processing in n nodes, the number of times of communication is n × 4 times.

Further, in the above-described processing method of the above-described known example that reduces the number of times of communication of n × 4 times, the commit processing is necessary only when the update is not performed at all or when the update is not performed by only one node. Since it was not there, the number of updates could be reduced.

The present invention is not a conditional update count reduction method as described above, but by changing the network configuration and the commit method, it is possible to reduce the update count with the same certainty as in the conventional case in any case, and to increase the transaction processing speed. It is to provide a method of ending the processing.

[Means for Solving the Problems]

In order to achieve the above-mentioned object, the present invention has a plurality of nodes, and in a distributed system using a ring network configuration, a node that performs transaction termination processing sends a commit preparation request to both directions of the ring network, The transaction end processing is performed when the responses to the two requests are returned.

[Action]

In a distributed system using a ring network,
The node that receives the request instruction performs processing for the request and sends the request instruction to the node in the opposite direction to the node that sent the request instruction to determine the partner node with which to communicate.

There are two directions in the network, a request flow direction and a response return direction. One node that receives a commit preparation request from the request direction node prepares for commit, and after completion of commit preparation, prepares for commit according to the request direction. Make a request.

Upon receiving the commit preparation requests from both the left and right nodes, one node immediately executes the commit process after completing the commit preparation. After the commit process ends, a commit process end response is returned according to the response direction. In this way, when one node is ready to commit, all other nodes do not judge whether or not they are ready to commit, execute transaction processing according to the response content of the node that has returned a response, and terminate it. As a result, the number of communications can be reduced, and transaction processing can be speeded up.

〔Example〕

 An embodiment of the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a configuration diagram of a distributed system according to an embodiment of the present invention, and is a diagram showing a situation in which a plurality of nodes can communicate with each other via a communication network 41.

Normally, in the node A system 1, if the data access request from the application program 5 is for the data in the own node, the transaction tree management unit 6 recognizes that the access is to the own node, and manages the transaction. The database 9 of the own node is accessed through the processing unit 7 and the data management processing unit 8.

If the access request is for another node 2, 3, 4
The transaction tree information generated by the transaction tree management unit 6 is transferred from the communication control processing unit 10 to
It transmits to the other nodes 2, 3 and 4 and requests access. In the node B system 2, the node C system 3, and the node D system 4, the transaction tree information and the data access request are received by the respective communication management processing units 10, 20, 30 and the respective transaction tree management units 16, 26, 26, 36 and the transaction management processing units 17, 27, 37.

Each transaction management processing unit 17, 27, 37 accesses its own database 19, 29, 39 through the data management processing unit 18, 28, 38. After that transaction management department 17,27,37
The node A system 1 that returns a response is recognized based on the transaction tree information recognized by the node A, and the response is returned by the respective communication management processing units 10, 20, and 30.

FIG. 2 is a diagram showing a conventional transaction tree recognized by the transaction tree management units 6, 16, 26 and 36.

Node B system 2, node C system 3 and node D system 4 are in communication with node A system 1, and node B system 2, node C system 3 and node D system 4 can communicate only with node A system 1. On the contrary, the node A system 1 has a one-to-n correspondence that the node B system 2, the node C system 3, and the node D system 4 can communicate with each other, and the request is from the node A system 1 to the node B.
System 2, node C system 3, node D system 4,
The response is node B system 2, node C system 3, node D
When transmitted from the system 4 to the node A system 1, the transaction tree management units 6, 16, 26 and 36 recognize it.

FIG. 6 shows the transaction tree management unit 6 of each node.
FIG. 6 is a diagram showing a transaction tree according to the present invention recognized by 16, 26 and 36.

Node A system 1, node B system 2, node C system 3, node D system 4 are lined up in a ring, and the request is node A
System 1 to Node B system 2, Node D system 4
To the node C system 3 from the node B system 2 and the node D system 4.

On the other hand, when the response is transmitted from the node C system 3 to the node B system 2 and the node D system 4, and further from the node B system 2 and the node D system 4 to the node A system 1, the transaction tree management units 6, 16, 26,36 recognize.

FIG. 3, FIG. 5, FIG. 7, FIG. 8, FIG. 9 and FIG. 9 are diagrams showing an example of transaction tree information.

FIG. 10 is a flow chart of one transaction using the conventional transaction tree (FIG. 2).

Upon receiving the transaction start request (1000) from the application program 5, the transaction tree management unit 6 of the node A system recognizes the transaction tree as shown in FIG. 2, and the transaction management processing unit 7 generates a data processing request. , The transaction tree information and the data processing request at the same time, the communication control processing unit 10
Is transmitted to the node B system 2, the node C system 3, and the node D system 4 by using (1001, 1006, 1011).

In the node B system 2, the node C system 3, and the node D system 4, the transaction tree information and the data processing request are received by the respective communication control processing units 10, 20, 30 and the transaction tree information is stored in the transaction tree management unit 16, 26. , 36, the data processing request is recognized by the transaction management processing units 17, 27, 37 (1002, 1007, 1012). In each transaction management processing unit 17, 27, 37, processing is executed in each database 19, 29, 39 through the data management processing unit 18, 28, 38 (1003, 1008, 1013). After the execution is completed, the response is sent from each transaction management processing unit 17, 27, 37 using the respective communication control processing units 20, 30, 40 to the node A system 1
Returned to (1004,1009,1014).

When the communication control processing unit 10 receives the response (1005,1010,101
5) The transaction management processing unit 7 performs data processing on the database 9 through the data management processing unit 8 (101
6).

Next, when a transaction end request is received from the application program 5 (1017), the transaction management processing unit 7 makes a commit preparation in the database 9 through the data management processing unit 8 and if the commit preparation processing succeeds, the transaction tree management unit 6 A commit preparation request is generated and transaction tree information and a commit preparation request as shown in FIG.
Is transmitted to the node B system 2, the node C system 3 and the node D system 4 by using (1018, 1023, 1028).

The node B system 2, the node C system 3, and the node D system 4 respectively transmit the transaction tree information and the commit preparation request transmitted by the communication control processing units 10, 20, 30.
The transaction tree information is received by the transaction tree management units 16, 26, 36, and the commit preparation request is recognized by the transaction management processing units 17, 27, 37 (1019, 102).
4,1029).

The transaction management processing units 17, 27, 37 of the respective nodes are connected to the databases 19, 29, 39 through the data management processing units 18, 28, 38.
Execute the commit preparation in (1020,1025,1030).

After execution is complete, the response is the transaction management processing unit
Node A from 17, 27, 37 using communication control processing units 20, 30, 40
It is returned to the system 1 (1021, 1026, 1031).

When the communication control processing unit 10 of the node A system receives the response (1022, 1027, 1032), the transaction management unit 7 receives a commit preparation completion response from all of the node B system 2, the node C system 3, and the node D system 4. It is judged whether or not it has returned (1033), and if all nodes are ready for commit, commit processing is performed in the database 9 through the data management processing unit 8 (1035).

After that, when 6 control is returned to the transaction management processing unit 7, the communication control processing unit 10 is used to generate a commit request to the node B system 2, the node C system 3, and the node D system 4, and as shown in FIG. And send it along with detailed transaction tree information (1036,1041,1046).

In the node B system 2, the node C system 3, and the node D system 4, the transaction tree information and the commit request are received by the respective communication control processing units 10, 20, 30 and the transaction tree information is stored in the transaction tree management unit.
16, 26, 36, the commit request is the transaction management unit 1
Recognize at 7,27,37 (1037,1042,1047).

Each transaction management processing unit 17, 27, 37 executes a commit in its own database 19, 29, 39 through the data management processing unit 18, 28, 38 (1038, 1043, 1048). The commit execution completion response is returned to the transaction management processing units 17, 27, 37.
Is returned to the node A system 1 using the communication control processing units 20, 30, 40 (1039, 1044, 1049).

In the node A system 1, the communication control processing unit 10 receives the response (1040, 1045, 1050), and the transaction ends (1051).

As can be seen from FIG. 4, the number of communications between the node A system, node B system, node C system, and node D system at the end of the transaction is (the number of partner nodes).
× 4 = 3 × 4 = 12 times.

FIG. 11 is a flow chart of one transaction in which the two-phase commit protocol which is the basis of the present invention is optimized.

In the case of executing a transaction between two nodes, after the node that initiated the transaction (called node A) completes the commit preparation process, it sends a commit preparation request to the other node (called node B), and the node After executing the commit preparation process, B executes the commit process if successful and sends the commit completion response, and if it fails, executes the rollback process and sends the rollback completion response to node A, thereby ensuring transaction consistency. It is characterized by reducing the number of communications while maintaining.

Although the transaction start request (1000) to the data processing execution (1016) are the same processing, the transaction management processing unit 7 that has received the transaction end request (1017) from the application program 5 prepares the commit processing and prepares the data management processing unit 8. If the commit preparation process succeeds in the database 9 through the transaction tree management unit 6, the transaction tree management unit 6 sends a commit preparation request from the communication control processing unit 10 to the node B system 2 (1102).

In the node B system 2, the communication control processing unit 20 receives (1103), the commit preparation request is recognized by the transaction management processing unit 17, and the commit preparation is executed in the database 19 through the data management processing unit 18 (1104) and the commit is made. When the preparation process is successful, that is, the commit is executed (1105), and when the control returns to the transaction management processing unit 17, the commit completion response is sent to the node A system 1 (1106).

In the node A system 1, the communication control processing unit 10 receives (1107), and the transaction management processing unit 7 determines whether the response from the node B system 2 is a commit completion response (1108).

If the transmitted response is the commit completion response, the commit is executed in the database 9 through the data management processing unit 8 (1109), and the transaction ends (111).
0).

As you can see from Figure 11, at the end of the transaction,
The number of communications between node A system and node B system is
(Number of partner nodes) × 2 = 2 times, which is half the number of communications when the two-phase commit protocol is not optimized.

FIG. 12 shows the transaction tree of the present invention (FIG. 6).
It is a flow chart of one transaction using the.

The process from the transaction start request (1000) to the transaction end start request of the node A system 1 to the commit preparation process (1017) is the same, but next, the transaction management processing unit 7 sends a commit preparation request to the transaction tree information (see FIG. 7). ) Together with the communication control processing unit 10
Node B subject to transaction tree information using
It is sent to the system 2 and the node D system 4 (1201).

In the node B system 2 and the node D system 4, the transaction tree management units 16 and 36 receive the transaction tree information received by the respective communication control processing units 20 and 40 (1202 and 1205).
The commit preparation request is issued by the transaction management processing unit 17, 37.
Recognizes each of them, and executes commit preparation in each database 19, 39 through each data management processing unit 18, 38 (120
(3, 1206), when control returns to the transaction management processing unit 17, 37, the transaction tree information (Fig. 7)
And the commit preparation process together using the respective communication control processing units 20 and 40, the node C according to the transaction tree information.
Send to system 3 (1204, 1207).

Since the commit preparation request has already been received for the other end, the node A system 1, the commit preparation request is not transmitted again.

In the node C system 3, the communication control processing unit 30 receives (1208), the transaction tree information is recognized by the transaction tree management unit 26, and the commit preparation request is recognized by the transaction management processing unit 27. The transaction management processing unit 26 determines whether the request from both nodes is a commit preparation request (1209). If the requests from both nodes are commit preparation requests, the transaction management processing unit 26 prepares a commit preparation in the database 29 through the data management processing unit 28. It is executed (1210), and when the preparation for commit is completed, that is, the commit process is performed (1211). When the control returns to the transaction management processing unit 27, the transaction tree information (FIG. 7) and the commit completion response are used together with the communication control processing unit 30 to use the node B system 2 and the node D according to the transaction tree information. Send to system 4 (1212).

In the node B system 2 and the node D system 4, each communication control processing unit 20, 40 receives (1213, 1216), and the transaction tree information is transaction tree management unit 16, 36.
However, the transaction management processing units 17 and 37 recognize the response from the node C system 3, and if the response is the commit completion response, the commit is executed in the respective databases 19 and 39 through the data management processing units 18 and 38. (1214,121
7), the control is performed by each transaction management processing unit 17,
Returning to 37, transaction tree information (7th
Figure) and commit completion response together, each communication control processing unit 2
It is sent to the node A system 1 according to the transaction tree information using 0, 40 (1215, 1218).

Since the commit completion response has already been received for the other end, the node C system 3, the commit completion response is not transmitted again.

In the node A system 1, the communication control processing unit 10 receives (1219), and if either the response from the node B system 2 or the response from the node D system 4 is transmitted to the node A system 1, that is, the response. Is a commit completion response by the transaction management processing unit 7 (12
30).

If the transmitted response is the commit completion response, the commit processing is executed in the database 9 through the data management processing unit 8 (1231), and if the commit is completed in the own node according to the transaction tree information, the transaction is completed. Since we know it is good, the transaction ends (1232). Here, an example in which the response from the node B system 2 is returned early is shown.

As you can see from Figure 12, at the end of the transaction,
The number of communications between the node A system 1, the node B system 2, the node C system 3, and the node D system 4 is (the number of partner nodes + 1) × 2 times = 8 times, which is smaller than the example of FIG.

FIG. 13 shows the transaction tree of the present invention (FIG. 6).
It is a flow chart of one transaction using the. However,
A case where the node D system 4 receives a commit preparation request from both nodes will be described.

The processing from the transaction start request (1000) to the transaction end start request reception and commit preparation processing (1017) of the node A system 1 is the same, but next the transaction management processing unit 7 sends a commit preparation request to the transaction tree information (seventh). (Fig. 2) together with the node B system 2 and node D system 4 according to the transaction tree information using the communication control processing unit 10 (1301).

In the node B system 2, the communication control processing unit 20 receives (13
02) The transaction tree management unit 16 recognizes the transaction tree information, and the transaction management processing unit 17 recognizes the commit preparation request, and the database 19 is executed through the data management processing unit 18 to execute commit preparation (1303). When returning to 17, the transaction tree information (FIG. 7) and the commit preparation request are sent together to the node C system 3 according to the transaction tree information using the communication control processing unit 20 (1304).

Since the commit preparation request has already been received for the node A system at the other end, the commit preparation request is not transmitted again.

In the node C system 3, the communication control processing unit 30 receives (13
05), the transaction tree management unit 26 recognizes the transaction tree information, and the transaction management processing unit 27 recognizes the commit preparation request, and executes commit preparation in the database 29 through the data management processing unit 28 (1306). When returning to the section 27, the transaction tree information (FIG. 7) and the commit preparation request are sent together to the node D system 4 according to the transaction tree information by using the communication control processing section 30 (130).
7).

Since the commit preparation request has already been received, the commit preparation request is not transmitted again to the node B system at the other end.

In the node D system 4, the communication control processing unit 40 receives (1310), the transaction tree information is recognized by the transaction tree management unit 46, and the requests from the node A system 1 and the node C system 3 are recognized by the transaction management processing unit 47. . The transaction management processing unit 46 determines whether the request from both nodes is a commit preparation request (131
1) If the request from both nodes is a commit preparation request, the data management processing unit 48 executes commit preparation in the database 49 (1312). When the commit preparation is completed, that is, commit is executed (1313). Transaction management processing unit
When the control returns to 47, the transaction tree information (Fig. 7) and the commit completion response are put together with the communication control processing unit.
40 is sent to the node A system 1 and the node C system 3 according to the transaction tree information (1314).

In the node C system 3, the communication control processing unit 30 receives (13
19), the transaction tree information is recognized by the transaction tree management unit 26, the response from the node C system 3 is recognized by the transaction management processing unit 27, and if the response is a commit completion response, the database 29 is committed through the data management processing unit 28. (1320), and when control returns to the transaction management processing unit 37, the transaction tree information (FIG. 7) and the commit completion response are put together,
It is sent to the node B system 2 according to the transaction tree information using the communication control processing unit 30 (1321).

Since the commit completion response has already been transmitted to the other end, the node D system 4, the commit completion response is not transmitted again.

In the node B system 2, the communication control processing unit 20 receives (13
22), the transaction tree management unit 16 recognizes the transaction tree information, the transaction management processing unit 17 recognizes the response from the node C system, and if the response is the commit completion response, the data management processing unit 18 executes the commit in the database 19. Then (1323), when the control returns to the transaction management processing unit 17, the transaction tree information (FIG. 7) and the commit completion response are put together, and the communication control processing unit 20 is used to comply with the transaction tree information. Send to 1 (1324).

In the node A system 1, if the communication control processing unit 10 receives (1315) and either the response from the node B system 2 or the response from the node D system 4 is transmitted to the node A system 1, that is, The transaction management processing unit 7 determines whether the response is a commit completion response (1316).

If the sent response is the commit completion response, the commit process is executed in the database 9 through the data management processing unit 8 (1317), and if the commit is completed in the own node according to the transaction tree information, the transaction is completed. Since we know it is good, the transaction ends (1318). Here, an example in which the response from the node D system is returned early is shown.

As you can see from Figure 13, at the end of the transaction,
The number of communications between node A system 1, node B system 2, node C system 3, node D system is (number of partner nodes +
1) × 2 times = 8 times, which is smaller than the example in FIG.

FIG. 14 is a transaction tree of the present invention (FIG. 6).
The flow of one transaction by recognizing the above is a diagram showing how to maintain transaction consistency when commit preparation fails in the node C system 3.

The process from the transaction start request (1000) to the commit preparation request from both nodes by the node C system 3 (1209) is the same as in FIG. 12, but in the node C system 3, the database 29 is executed through the data management processing unit 28. If the commit preparation processing fails in (1410), rollback processing is performed next (1411).

When control is returned to the transaction management processing unit 27,
The transaction tree information and the rollback completion response are sent together to the node B system 2 and the node D system 4 according to the transaction tree information using the communication control processing unit 30 (1412).

In the node B system 2 and the node D system 4, each communication control processing unit receives (1413, 1416), and the transaction tree information is transaction tree management units 16, 36.
However, the transaction management processing units 17 and 37 recognize the response from the node C system, and the response is a rollback completion response. Therefore, the rollback processing is executed in the databases 19 and 39 through the data management processing units 18 and 38, respectively. (1414,1417) and control returns to the respective transaction management processing units 17,37, the transaction tree information and rollback completion response are sent together from each communication control processing unit 20,40 to the transaction tree information. It is sent to the following node A system 1 (1415, 1418).

Since the rollback completion response has already been received for the other end, the node C system 3, the rollback completion response is not transmitted again.

In the node A system 1, the communication control processing unit 10 receives (1419), and either the response from the node B system 2 or the node D
If any response from the system 4 is transmitted to the node A system 1, that is, the transaction management processing unit 7 determines whether the received response is a commit completion response (1430). If the response sent is a rollback completion response, the data management processing unit 8 executes rollback processing in the database 9 (1431), and if the rollback is completed in the own node according to the transaction tree information, the transaction ends. Since it recognizes that it is okay, the transaction ends (1432). Here, an example is shown in which the response from the node B system 2 is returned early.

FIG. 15 shows the transaction tree of the present invention (FIG. 6).
It is a flow of one transaction by recognizing the above, but is a diagram showing how to maintain the transaction consistency when the commit preparation fails in the node B system 2.

From the transaction start request (1000), the node A system sends a commit preparation request and transaction tree information to the node B system 2 and node D system 4, and in the node B system 2 and node D system 4, each communication control processing unit receives them. (1202, 1205), the transaction tree information is the same until the transaction tree management units 16 and 36 recognize the commit preparation request, and the transaction management processing units 17 and 37 recognize the transaction preparation information. The commit preparation process was executed in the database 19 through the management processing unit 18, but it failed here (150
3) That is, the rollback process is performed (1504).

When the control returns to the transaction management processing unit 17,
According to the transaction tree information, a rollback completion response is sent to the node A system 1 together with the transaction tree information, and a rollback processing request is sent to the node C system 3 (1505). Since the commit preparation request is received, the rollback completion response is transmitted instead of the rollback processing request.) In the node A system 1, the communication control processing unit 10 receives (15
06), the transaction management processing unit 7 determines whether the received response is a commit completion response (1507). Since it is a rollback completion response, rollback processing is performed in the database 9 through the data management processing unit 8. (150
8). Since the transaction tree information recognizes that the transaction may be terminated if the rollback is completed in the own node, the transaction is terminated (1509).

On the other hand, the node D system 4 is similar to the node A in FIG.
The system 1 sends a commit preparation request (1205) and prepares for commit (1207), and sends a commit preparation request to the node C system 3 (1206).

Since the commit preparation request has already been received for the other end, the node A system 1, the commit preparation request is not transmitted again.

In the node C system 3, the communication control processing unit 30 receives (1513), the transaction tree information is recognized by the transaction tree management unit 26, and the requests from the node B system 2 and the node D system 4 are recognized by the transaction management processing unit 27. . Both nodes judge whether or not they have transmitted the commit preparation request (1514). Since the rollback request is received from the node B system and the commit preparation request is received from the node D system, that is, the rollback process is executed. The operation is performed (1515), and the rollback completion response is transmitted only to the system 4 of the node D (1516). (Since the rollback request has been received from the node B system, the rollback completion response is not transmitted.) In the node D system 4, the communication control processing unit 40 receives (1517), and the transaction tree information is transaction tree management unit 36. However, the rollback completion response is recognized by the transaction management processing unit 37, and the data management processing unit 38
Rollback processing is performed on the database 39 through (15
18) When the control returns to the transaction management processing unit 37, the transaction tree information and the rollback completion response are sent together to the node A system 1 according to the transaction tree information using the communication control processing unit 40 (151).
9).

Since the rollback completion response has already been received for the other end, the node C system 3, the rollback completion response is not transmitted again.

Since the node A system 1 has already received the rollback completion response from the node B system 2 and executed the rollback process, no process is performed.

FIG. 16 uses a transaction tree (transaction tree in FIG. 3 and transaction tree information in FIG. 8) of the present invention, and further determines a transaction node (commit processing depending on the result of commit preparation processing,
Or showing a node that performs rollback processing and sends a commit and rollback completion response) is a flow chart of one transaction in which a transaction completion response is broadcast and sent to all nodes related to the transaction.

From the transaction start request (1000), the node C system 3 executes the commit preparation in the database 29 (121
0), when the commit preparation is completed, that is, until the commit process is executed (1211), the process is the same as that of FIG. 12, but when control is returned to the transaction management processing unit 27, the committer completion response and the transaction information are displayed. Using the transaction tree information, the communication control processing unit 30 is used to transmit to the systems (node A system 1, node B system 2, node D system 4) of all nodes related to the transaction (1601).

In the node A system 1, the node B system 2, and the node D system 4, each communication control processing unit 10, 20, 40 receives (1604, 160
2,1603), the transaction tree information is provided by the transaction tree management units 6,16,36, and the commit completion response from the node C system is provided by the transaction management processing units 7,17,3.
7 recognizes each, and if the response is a commit completion response, commit processing is executed in each database 9, 19, 39 through each data management processing unit 8, 18, 38 (1610, 1605, 160).
6), control is performed by each transaction management processing unit 7,1
When returning to 7,37, the node A system 1 recognizes from the transaction tree information that the transaction may be completed if the commit is completed in the own node, and the transaction is completed (1611).

Since the node B system 2 and the node D system 4 can recognize that the commit completion response has been received by broadcasting in the transaction tree information, the node A system 1 assumes that the transaction has already been completed and the comic completion response is not sent to the node A system 1. Do not send.

The node A system 1 can quickly complete the transaction processing without waiting for the transmission of the commit completion response from the node B system 2 and the node D system 4.

Also, the number of communications is clearly reduced compared to the cases of Figs.

This is also applicable to the cases of FIGS. 13, 14, and 15.

FIG. 9 shows that the transaction tree information of FIG. 7 is registered in advance in the transaction tree management unit of each node, and the transaction tree information of FIG. It is transaction tree information for transaction pattern recognition that enables data to be transmitted and received by the information as shown in FIG. As a result, it is not necessary to transmit / receive transaction tree information each time data is transmitted / received.

The flow of one transaction is shown in Fig. 11, Fig. 12, Fig. 13,
It is completely the same as Fig.14 and Fig.15.

〔The invention's effect〕

According to the present invention, in a distributed system using a ring network configuration, a node that performs transaction termination sends a commit preparation request in both directions of the ring network, and when the responses to these two requests are returned, transaction end processing is performed. By doing so, the number of communications can be reduced, which is effective in improving system performance.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a distributed system having four nodes, FIG. 2 is a diagram showing a conventional transaction tree having a one-to-n relationship, and FIG. 3 is a diagram recognized by a conventional transaction tree management unit. FIG. 4 is a diagram showing transaction tree information of a one-to-n relationship, FIG. 4 is a diagram showing transaction tree by two nodes of one embodiment of the present invention, and FIG. 5 is two nodes of one embodiment of the present invention. Showing transaction tree information recognized by each transaction tree management unit of FIG. 6, FIG. 6 is a diagram showing a circular transaction tree of one embodiment of the present invention, and FIG. 7 is each node of one embodiment of the present invention. FIG. 8 shows transaction tree information recognized by the transaction tree management unit of FIG.
FIG. 9 is a diagram showing transaction tree information transmitted by broadcasting according to an embodiment of the present invention, FIG. 9 is a diagram showing transaction tree information which is pattern-recognized according to one embodiment of the present invention, and FIG. A transaction flow chart, FIG. 11 is a flow chart of one transaction in two nodes using the transaction tree of FIG. 4 of one embodiment of the present invention, and FIGS. 12 and 13 are of the embodiment of the present invention. Flow chart of one transaction using the transaction tree of Fig. 6, Figs. 14 and 15 are flow charts of one transaction when a node fails the commit preparation process using the transaction tree of Fig. 6, Fig. FIG. 16 is a flow chart of one transaction of the broadcasted processing according to the embodiment of the present invention. 1,2,3,4 …… Distributed system of each node, 5 …… Application program, 6,16,26,36 …… Transaction tree management unit, 7,17,27,37 …… Transaction management processing unit, 8 , 18,28,38 …… Data management processor, 9,19,29,39…
… Database, 10, 20, 30, 40 …… Communication control processing unit, 41
...... Communication network.

Front Page Continuation (56) References Information Processing Society of Japan Research Report 88 "84" (1988-11-17) (Database System Research Group Report No. 68-5) Sadayuki Hikita "COM Series Graphical Distributed Database System Introduction "(1989-5-25), Ohmsha, P99-107 Atsushi Matsushita" Information Processing Series 5 Computer Network "(1983-10-25) P.F. 220

Claims (3)

[Claims] 1. In a distributed system having a ring network configuration having a plurality of nodes, a) a commit preparation request is sent from a node which performs transaction termination processing to both directions of the ring network, and b) a commit preparation request. The node which has received the commit preparation process sends a commit preparation request of a node in the opposite direction to the node which has sent the commit preparation request, and c) the above b until reaching the node which receives the commit preparation request from the nodes on both ends adjacent to each other. ) Is repeated, and d) the node that receives the commit preparation request from the nodes on both ends adjacent to each other performs the commit preparation processing and the commit processing,
If these processes are successful, a commit request is sent to the adjacent nodes, and if these processes are unsuccessful, a rollback request is sent to the adjacent nodes. E) The node that receives the request is designated. Commit process or rollback process, send the next request to the node in the opposite direction to the node that sent the request, and f) repeat the process of e) until the node that performs the transaction end process of a) above is reached, g) A method of terminating transaction processing, characterized in that a node that performs transaction termination processing performs commit processing or rollback processing according to a received request to terminate transaction processing. 2. The method according to claim 1, wherein in the step b) above, if the commit preparation process fails in the node that receives the commit preparation request,
Send a rollback request instead of the commit preparation request to the node in the opposite direction to the node that sent the commit preparation request. I) In step d) above, rollback is performed instead of the commit preparation process and the commit process, and they are adjacent to each other. A method for terminating transaction processing, characterized by sending a rollback request to the nodes at both ends. 3. In the step i) of claim 2,
A method for terminating transaction processing, characterized in that instead of sending a rollback request to the nodes on both ends adjacent to each other, the rollback request is broadcast and the rollback request is sent to all other nodes related to the transaction processing.