JPH0387941A - Decentralized data base processing system - Google Patents

Abstract

PURPOSE:To process the reference requests to a decentralized data base at a high speed by performing the communication of many produced control messages via a shared storage. CONSTITUTION:A decentralized data base control means 4 analyzes a data reference request and produces a message to be sent to the other site in order to attain the request. A communication control part 5 decides a message to be transmitted and writes a control message if decided to a shared storage means 2. Meanwhile a data message is sent to the corresponding site via a communication circuit 1. Thus the storage means 2 is set between both sites so that both sites refer to the information stored in the storage means 2 at a high speed. As a result, the communication of messages can be performed in a short time and the processing efficiency is improved with a decentralized data base.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial Application Field) The present invention relates to a distributed database processing method for data reference requests in a distributed database management system.

(Prior Art) Hereinafter, operations such as reading, writing, erasing, adding, and initializing data to a database will be collectively referred to as data reference. Furthermore, among the computers that make up the distributed database system, the computer that holds the database for that part is called a site.

In a distributed database management system, when a data reference request that spans multiple sites occurs, it is divided into parts, sends a partial reference request, control information for exclusive control, etc. to each site, and receives a response. Communication information at this time can be divided into control messages and data messages.

The former includes data lock information for parallel control, commitment information indicating completion of an operation, etc., and the latter includes data retrieved from a database upon request or data to be stored. Generally, the former has a small amount of data but a large number of data, and the latter has a large amount of data but the number of transfers is smaller than the former.

Currently, these messages are communicated using the same communication route without being differentiated, resulting in congestion on the communication route and a large amount of time being spent on communication.

In data reference processing in a distributed environment, communication requires more processing time than calculations and input/output of external storage devices, so communication time has been a major problem in speeding up data reference processing.

(Problem to be Solved by the Invention) As described above, the conventional method has the disadvantage that it requires a lot of processing time because control messages and data messages are communicated using the same communication path without distinguishing them. Ta.

Therefore, the present invention aims to improve processing efficiency by arranging shared storage means between sites so that information on the shared storage can be referenced quickly from each site, thereby allowing message communication to be carried out in a short time. .

[Structure of the Invention] (Means for Solving the Problems) In the present invention, each site is connected by a communication line. Furthermore, a shared memory created using semiconductor memory or the like that can be referenced at high speed is connected to each site, and data can be referenced from each site to the shared memory. Each site has data storage means such as a disk device for storing distributed databases. Distributed database management means, generally implemented in software, perform query processing, reference optimization, and communication message creation.

(Function) The distributed database management means of the present invention analyzes a data reference request and creates a message to other sites to accomplish the request. The communication management unit determines the message to be communicated, and if it is a control message, it is written to the shared memory, and if it is a data message, it is transferred to the corresponding site via the communication line.

(Example) An example of the present invention will be described below with reference to the drawings. FIG. 1 is a schematic diagram showing an example of the configuration of the present invention, and in this example, it consists of three sites A, B, and C. Each site is connected by a communication line 1, and the shared storage 2 is, for example,
It is connected to each site by a shared memory connection means 3 via a memory bus or the like.

The shared memory 2 is a device that can be referenced at high speed, such as a semiconductor memory.

FIG. 2 is a diagram showing an example of the internal configuration of one site.

One site alone can perform processing as a computer. Therefore, it consists of a CPU 7, a main memory 8, and an external storage device 6 such as a disk that stores a database and the like. Further, a distributed database management unit 4 that manages distributed databases performs database management such as so-called question processing and database reference control. Generally, the distributed database management section is often realized by software, and in this case, it exists on the main memory 8 as shown in FIG. The communication management section 5 is a section in charge of communication with the outside of the site, and sends and receives messages to and from the communication management sections of other sites in accordance with requests from the distributed database management section 4 and the like.

FIG. 3 is a flowchart showing an outline of processing for explaining the present invention.

The present invention will be explained with reference to these figures. For example, suppose that site A receives the following data reference request from a terminal device, etc.

'Subtract the value of data a stored in the database of site A from data b stored in the database of site B, and subtract the result from data C stored in the database of site C.

Furthermore, the result is written to the data b' of the B site.

The data reference request at this time is b'←C←c-(ba-a) It can be expressed as. - represents value assignment.

When this process is performed mainly at site A, the data reference procedure is described as shown in FIG. 4.

In this case, it is also possible to send the difference between b and a to the C site, calculate and write the new value of C at the C site, and write the result to b' at the B site, but this explanation will be based on the above example. I will explain.

Generally, multiple database reference requests often occur at the same time. Therefore, in order to process efficiently, instead of allocating CPU processing time to one reference request until it is completed, we divide the time into blocks for multiple reference requests.
Allocate PU processing time. Concurrency control is therefore required, and the data is referenced after locking to maintain data consistency. At this time, methods have been devised that give the same result as serially executing the reference requests, but here we will explain a method that uses the most common two-phase lock protocol. Locking shall be performed for both data reading and writing.

Furthermore, in database processing, it is necessary to maintain the database even when a failure occurs in a computer, external storage device, communication device, or the like. Particularly in the case of distributed databases, there are many devices that are subject to failure, so care must be taken when updating data. Therefore, this example will be explained assuming that a two-phase commit protocol is adopted.

In the case where these methods are adopted, the processing is performed mainly at the A site, and the message is processed smoothly without any failures or data lock conflicts with other reference requests during the processing. The flow is shown in Figure 5.

LOCK also regarding reference to data a.

Although the COMMIT operation is necessary, there is no need to send or receive messages because the processing is centered on the A site.

WRITE(e) represents writing to data named C. Further, the type indicates whether the message is a control message or a data message.

From FIG. 5, it can be seen that 20 message communications are required. Previously, all of these message communications were sent to communication port 1.
! It was transferred via 1. The data reference request described here may also occur at sites B and C, and such data reference requests may occur several thousand times per second in a large-scale information system. Communication became a major bottleneck, reducing the throughput of the entire system.

In the present invention, the control message is written into the shared memory 2 without going through a communication line. The shared memory 2 is made of a storage medium such as a semiconductor that can be referenced at high speed, and when connected to a computer via a memory bus, it can be transferred at a much higher speed than when control messages are transferred via the communication line 1. can.

Explain the steps. First, the data reference request is analyzed by the distributed database management unit 4 and a processing procedure is determined. the result,
It can be seen that messages such as those shown in Table 1 must be exchanged. forward messages in that order,
Perform calculations.

When it becomes necessary to send/receive a message, the distributed database management section 4 passes the reception request and the message to the communication management section 5 along with information for distinguishing whether it is a control message or a data message.

It is also possible to include discriminating information inside the message and use the communication management unit 5 to make the determination. The communication management unit 5 communicates data messages via a communication line operator, and writes control messages in the shared memory 2.

When writing to the shared memory 2, an identifier such as the number of the site where the message should be received is added. Each site periodically refers to whether there is a message addressed to the site in the shared memory 2. Alternatively, the moment the data is written to the shared memory 2, it is possible to interrupt the site to be read and notify that the control message has arrived.

Once the receiving site reads the information in shared memory 2, the message is no longer needed and is deleted from shared memory 2.

There are various ways to manage data in shared memory 2. For example, there is a method of writing the data into the shared memory 2 in a variable length, or a method of dividing the data into fixed-length records. Methods for determining whether or not a message has been read include a method of erasing the message after it has been read, and a method of attaching a mark (flag) indicating whether the message has been used or not. If you have multiple sites receiving messages, for example, if you have multiple copies of your data and want to send a control message such as a lock to all those sites, share the message with the number of sites you want it to read. Write to shared memory 2. Then, reduce the number of sites you read by one.

After that, the number is checked, and if it becomes 0, it means that the message has been read by all the sites to which it was sent, so it is invalidated.

In the above example, the number of times the communication line 1 has been used is 20 times in the past, but by communicating the control message using the shared memory 2 according to the present invention, the number of times the communication line 1 is used is four times. For example, the effect is calculated by focusing only on the communication time, assuming that the communication time via the communication line 1: the time written to the shared memory 2 - 100:1.

・Conventional...20 communication line usage time ・This invention...
・4 communication line usage time + 16 shared memory writing time + I
B Shared memory read time As a result, the conventional two inventions - 20: (4416/100 + 16/1
00)-20:4.32 Therefore, the speed can be increased by about 4.8 times.

Currently, communication takes more time than CPU processing and disk input/output, and the present invention can be expected to dramatically speed up the processing of distributed database references.

The example discussed here has duplicate data (
data safety,
Copies may be placed on other sites to speed up loading. If a copy exists, there will be more messages about locks and commitments. therefore,
This is especially effective in distributed databases where copies exist.

In addition, various methods have been proposed for optimizing data reference requests, but the present invention is more effective when adopting a strategy that minimizes the number of data transfers, although the number of control messages may be large. It is true.

Note that the present invention is not limited to the embodiments described above, and the method of configuring the processing section within the site can be modified as appropriate without departing from the gist thereof. For example, in FIG. 1, the shared memory is connected from all sites and can be directly referenced, but even if the shared memory is not connected from all sites, a certain effect can be obtained by the present invention.

Furthermore, the present invention is independent of the query optimization method, exclusive control method, and commitment method, and regardless of which method is adopted, processing can be performed faster than when the present invention is not used.

[Effects of the Invention] As is clear from the above description, the present invention enables high-speed processing of reference requests to a distributed database by communicating control messages that occur in large numbers using shared memory in distributed database processing. It becomes possible to do so.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a diagram showing an example of the internal configuration of one site in the present invention,
Fig. 3 is a flowchart showing an outline of the processing flow of only the communication processing part when implementing the present invention and realizing high-speed communication for referring to a distributed database, and Fig. 4 explains an example of a request for referring to a distributed database. Figure 5 is the 4th diagram for
This is a list of examples of messages that should be communicated when processing the data reference request shown in the figure. DESCRIPTION OF SYMBOLS 1... Communication line, 2... Shared memory, 3... Shared memory coupling means, 4... Distributed database management section, 5...
・Communication management unit, 6...external storage device, 7...CPU
, 8...main memory.

Claims (1)

[Claims] In a distributed database system that consists of multiple computer sites containing databases that are connected to each other by a communication path, and that handles data groups scattered on each site, it can be shared and referenced from each site without going through the communication path. 1. A distributed database processing system comprising a shared storage means and transmitting messages regarding database reference control via the shared storage means.