JPH06243013A - Distributed data base system - Google Patents

Abstract

PURPOSE:To eliminate the need of excessive communication for a synchronous processing, to prevent an update stop at the time of a site fault and communication network interruption and to prevent the deterioration of the processing by dividing a primary copy into appropriate size and dynamically dispersing the places of the divided copies between sites. CONSTITUTION:A revision/primary management part 3 executes revision management and primary management. When its own site is not the primary copy and the present revision of the page differs from the latest revision of the page in the whole data base system, the announcement of effect that the page of its own site to be the primary copy is transmitted to the whole system by a high speed channel. At that time. the site where the page is the primary copy compares the revision in the middle of the announcement with the latest revision and transmits data of the difference by using the high speed channel. Data of the page of its own site is updated by the data.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a distributed database system, and more particularly to resource management of a plurality of databases provided at each database site.

[0002]

2. Description of the Related Art FIG. 6 is an overall system diagram for explaining a distributed data system. In the illustrated example, three database sites A, B and C are connected by a communication network 501 to construct a distributed database. ing. Each database site comprises a server 502, a database 7, and a terminal device 503. In a distributed database system, multiple sites can have the same data redundantly. When various sites in a distributed database system are searched for specific data, there is a demand to store the same data in multiple sites in order to improve response time and reduce communication costs. Also, to improve reliability,
There is also a demand for having the same data redundantly between multiple database sites and preparing for a database site failure. However, in the conventional technology, because of synchronization processing and equalization processing between a plurality of database sites, the communication amount (cost) increases, the processing efficiency (response) decreases, or the update is stopped at the time of failure, Such a problem occurs in reverse.

[0003]

The problems associated with these conventional techniques will be described below. As described above, when the same data is stored in a plurality of sites in duplicate, the data may be retrieved, but communication will occur between the plurality of database sites when the data is updated. That is, it is necessary to match the contents of the data that are redundantly stored in the plurality of database sites, and extra communication that synchronizes the update processing with each other is required. In addition, although the duplication of databases in the distributed database system is effective in improving the reliability of the system, it has a drawback that extra communication occurs from the viewpoint of processing efficiency and the processing efficiency decreases.
Duplication does not matter if it is a read-only database system. However, when there is a write, each transaction operates on an arbitrary copy, so transaction management such as concurrency control becomes more complicated than when there is no duplicate data. Although various methods have been conventionally used for controlling the simultaneous execution of duplicate data, the primary copy (ledger) method, which is the basis of the present invention, will be described here.

FIG. 7 is a diagram for explaining the primary copy method. This method is used among a plurality of duplicate databases.
Specially treat one specific as the primary copy. If this copy is locked for reading / writing, it is not necessary to lock all the copies.
In FIG. 7, for example, transaction B is written /
When updating, first check the lock status of the primary copy, and if it is not locked, lock it and write / update (continue the written / updated contents are reflected in the copy of each site. As for this, a database management system (DBM
It depends on the function of S). ). If the primary copy is locked, transaction B waits until the lock is released. In the case of reading, it is read from the copy of the own site. This method is (write / update frequency) <
This is effective in the case of (readout frequency).

In the distributed database system having duplicated data, the problem of updating is complicated. If you update all the copies at the same time, if the site fails, you will not be able to update at all. As a measure against this,
There is a method such as "Update the database site that can be updated and save the update history. The failed database site synchronizes with other copies according to this history before restarting processing." If the site that has the primary copy fails, this method will not work. Another problem is that a part of the communication network breaks down and is divided into separate sub-communication networks that cannot communicate with each other. FIG. 8 shows this case. In this case, since communication is possible within each sub communication network, the processing can be continued. However, updating the copy independently in each sub-communication network causes a contradiction. The primary copy method can also be used as one solution to this problem. That is, only the sub-communication network having the primary copy can update the data, and the other sub-communication networks can read the old value of the data. The present invention has been made in view of the above circumstances, and does not require extra communication for synchronization processing, does not stop updating at the time of site failure and communication network division, and is a distributed type that does not lower processing efficiency. The purpose is to provide a database system.

[0006]

A distributed database system according to [Claim 1] of the present invention will be described with reference to FIG. Note that FIG. 1 is a configuration diagram of one database site. In FIG. 1, 1 is a high-speed communication channel and 2 is a low-speed communication channel, which are linked to other database sites. 3 is a function for performing revision management and primary management for each page (primary copy if the own site name and primary site name are equal), 4 is for data equalization (transmission of update history: low-speed communication channel is used) Mechanism 5, when accessing the database,
A mechanism for immediate data equalization (transmission of update history: use of high-speed communication channel) when there is no latest data, 6 performs primary copy switching (declaration to all sites: use of high-speed communication channel) at the time of writing / updating It is a mechanism.
Reference numeral 7 is a database, 8 is a storage unit for storing revision information and primary information, 9 is a storage unit for the data itself, 10 is a database management unit, and 11 is a channel control unit. The distributed database system according to [Claim 2] of the present invention is shown in FIG. 5, and is different from [Claim 1] in that the page access counting unit 12 is added.

[0007]

The distributed database system according to [Claim 1] of the present invention is based on the conventional method of locking the primary copy as a method for synchronizing duplicate data. Rather than having a fixed number at one site, the primary copy is divided into appropriate sizes (a management unit called a page), and its location is dynamically distributed among the sites. Then, when writing / updating occurs, a copy of the page of the site is declared to the entire system as a primary copy (after being updated to the latest state). First, the format of revision / primary information will be described with reference to FIG. 2. The revision / primary management unit 3 performs revision management and primary management based on this information (for each page). For revision / primary information, page # section 21, own site revision section 22, latest revision section 23
And a primary site instruction unit 24. Each page contains the current revision 22 of the page, the latest revision 23 of the page in the entire database system, and information indicating whether the page is currently the primary copy (and the site information of the primary copy) 24. have.
Each site of the database system is connected by two high-speed and low-speed communication means (channels) 1 and 2. The latest revision and primary copy switching information in the entire system are transmitted to the entire system by this high-speed channel. Further, the latest data itself is transmitted by the low speed channel (in some cases, the exception is the use of the high speed channel.
See below. ).

Now, when reading / writing / updating,
First, check if the page on your site is currently the primary copy. Case (a) If the own site is the primary copy, the read / write / update may be performed as it is. However, when writing / updating, the latest revision of the page after writing / updating is transmitted to the entire system through the high-speed channel (at this time, the writing / updating data itself is not transmitted immediately.
It is transmitted periodically using the low speed channel. In addition, when there is a primary switching request from another site, it is transmitted using a high speed channel. ). Case (b) If the own site is not the primary copy, the current revision of the page is compared with the latest revision of the page in the entire database system. In case of coincidence and writing / updating (reading as it is), a declaration (a kind of lock: any locking method) that the page of the own site is the primary copy is transmitted to the entire system through the high-speed channel. . After the declaration is complete,
Write / update as is. If the procedure after writing / updating was the primary copy from the beginning, case (a)
Is the same as. Case (c) If the own site is not the primary copy and the current revision of the page is different from the latest revision of the page in the entire database system, the page of the own site is set as the primary copy. A declaration to that effect is transmitted to the entire system via a high-speed channel (when writing / updating (reading only receives data)).
However, in this case, the site whose page is currently the primary copy compares the revision being declared (corresponding to the site that is about to become the primary) with the latest revision, and the difference data is sent to the high-speed channel. I will use it to transmit. The operation after the data on the page of the own site is updated by this data is the same as the case (b). In this way, data is transmitted not only when there is a request for primary switching, but also regularly using a low-speed channel or at a time when communication costs are low, such as at night, so that all sites are updated.

The frequency is optimized by the following ideas. Target average response> Data transmission amount (average) when case (c) occurs ÷ transmission speed of high-speed channel Target maximum response> Data amount of 1 page ÷ transmission speed of high-speed channel (in worst case, 1 page of data transmission Total communication cost = (probability of occurrence of case (c) x case (c))
(Data transmission amount at the time of occurrence x high-speed channel usage charge) + (transmission frequency of regular data x transmission amount at the time of periodic data transmission x low-speed channel usage charge (or special charge such as nighttime)) (Note: Occurrence probability and transmission frequency The unit is the same. Here, the probability of occurrence of case (c) and the amount of data transmission at the time of occurrence of case (c) become smaller as the frequency of transmission of regular data increases. However, above a certain frequency, the usage fee for the entire slow channel will exceed that for the fast channel. Therefore, from the perspective of communication costs only,
The point that the total communication cost is the lowest (the usage fee for the entire low speed channel is the same as that for the high speed channel) is the optimum point. The transmission frequency of the periodic data is determined by adding the request in the response to this. Also, especially when the reliability of the database system is highly demanded, the frequency of transmission of the regular data is determined so that the probability of occurrence of case (b) is high. However, in the present invention, the primary copy is moved to the site where the write / update occurs, and as a result, the primary copy tends to be automatically relocated to the site that needs the write / update for the data of the page. As such, the frequency of transmission may be fairly low (and thus economical). In the distributed database system according to claim 2 of the present invention, the primary copy is switched only when the number of accesses for each page counted by the page access counter reaches a predetermined number.
Other functions are the same as those in [Claim 1].

[0010]

Embodiments will be described below with reference to the drawings. Figure 3
FIG. 3 is a diagram for explaining the operation of the embodiment of the present invention, and each database site is composed of the constituent elements shown in FIG.
Then, the server 502 at each site includes the revision / primary management unit for each page 3, the regular data equalization unit 4, the immediate data equalization unit 5, the primary copy switching unit 6, the database management unit 10, and the channel. It is composed of the control unit 11. Here, in order to mainly describe the operation, the action line of each mechanism is represented by another symbol. That is, a. High speed communication channel. b. Low speed communication channel. c. An action line showing revision management and primary management (primary copy if own site name and primary site name are equal) for each page. However, although the symbol c is shown at the end of the leader line in FIG. 3, it is in the database. d. A line of action that shows periodic data equalization (transmission of update history: low-speed communication channel is used). e. A line of action showing immediate data equalization (transmission of update history: using high-speed communication channel) when there is no latest data at the time of access. f. Action line showing primary copy switching (declaration to all sites: use high-speed communication channel) at the time of writing / updating. Is. Further, FIG. 4 is a diagram showing the transition of the revision / primary information of FIG. 3, which is shown in parallel by associating those at the respective sites of A, B, and C with each other.

Next, the operation will be described with reference to FIGS. Now, a case where the data of a certain page is being read in the site A will be described. In this example,
By comparing the own site revision (value 90) and the latest revision (value 91) of the revision / primary management section for each page, it can be seen that the revision of the own site is not the latest. Therefore, the latest data is requested from the primary site B via the high-speed communication channel a. That is, when there is no latest data at the time of access, immediate data equalization (transmission of update history) is performed as indicated by the action line e. The fact that the primary site is B can be known from the information (value B) in the primary site instruction section (if it is originally the latest, read as it is). Next, at the site C, the same page is being written / updated. In this example, the revision (value 89) of that page on the own site is also older than the latest revision (value 91). Therefore, the latest data is requested to the primary site B via the high-speed communication channel a. That is, similarly to the above, if there is no latest data at the time of access, immediate data equalization (transmission of update history) is performed as indicated by an action line e. At the same time as the data is received, the page of the own site is declared as the primary copy to all sites. As a result, the value of the primary site designator of this page of each site is marked with a circle (if it is originally the latest, only the primary copy declaration is made. If it is the original, it is written / updated as it is). Further, regardless of access, data equalization (update history transmission d) is regularly performed for all sites (using the low-speed communication channel b). As a result, the revision / primary information of all sites match (values 92, 92, C).
The frequency of this regular data equalization is as described above.

As can be seen from the above embodiment, writing / writing
The primary copy is automatically relocated to the site that needs the update. Therefore, the optimal primary copy arrangement can be obtained according to the mode of operation. As a result, it is possible to reduce the communication cost for synchronization and to realize a response in which the primary copy is always at hand in spite of the duplicate database. In addition, the primary copy tends to be collected at the site according to the write / update frequency for each page at that site, so there is a possibility that the primary copy of the required page is at hand even in the event of a failure or communication network disconnection. It is high, and it hardly causes any trouble in operation.

FIG. 5 is a block diagram of another embodiment. The same parts as those in FIG.
The difference from FIG. 1 is that a page access counting unit 12 is provided. Then, the primary copy switching unit 6 does not switch the primary each time writing / updating is performed, but only when the number of accesses for each page counted by the page access counting unit 12 reaches a predetermined number of times, Switch. As a result, it is possible to prevent the optimal arrangement from being destroyed due to temporary access. The present embodiment prevents a decrease in efficiency in the case that it happens to be updated.

[0014]

As described above, according to the present invention, the primary copy of a page with many writes / updates at its own site is automatically relocated to its own site. That is, the primary copy is automatically relocated to the site where writing / updating is required, and the optimal primary copy allocation can be obtained according to the mode of operation. Therefore, in spite of the duplicate database, it is possible to realize a response (processing efficiency) such that the primary copy is always at hand from any user, and the effects listed below can be obtained. Reduced access to other sites. Since the primary copy tends to be relocated to sites that need to write / update data on the page,
The probability of having to stop the update is low, and operation is unlikely to occur. Since the primary copy tends to be automatically relocated to the site that most needs to write / update the data of the page, there is a high probability that the primary copy of the required page will be at hand even when the communication network is cut off. The probability of having to stop is low. That is, the operation is less likely to be hindered.

[Brief description of drawings]

FIG. 1 is a block diagram of the principle of the present invention.

FIG. 2 Revision / primary information format.

FIG. 3 is a diagram showing the overall configuration and operation of the embodiment.

FIG. 4 is a diagram showing revision / primary information transitions for the embodiment of FIG. 3;

FIG. 5 is a block diagram of another embodiment.

FIG. 6 is an overall system diagram illustrating a distributed data system.

FIG. 7 is a diagram illustrating a primary copy method.

FIG. 8 is a diagram illustrating division of a communication network.

[Explanation of symbols]

 1 High-speed communication channel 2 Low-speed communication channel 3 Revision / Primary management unit 4 Periodic data management unit 5 Immediate data management unit 6 Primary copy switching unit 7 Database 8 Revision / Primary information storage unit 9 Data storage unit 10 Database management unit 11 Channel control Department

Claims (2)

[Claims] 1. A high-speed and low-speed communication means for connecting a plurality of databases, a means for performing revision management and primary management for each page, a means for periodically equalizing data, and a means for accessing at the time of access. A means for performing immediate data equalization when there is no latest data, and writing / writing
It consists of a means to switch the primary copy at the time of updating. When reading, if the revision of the relevant page on the local site is older than the latest revision, the latest data is requested to the primary site via a high-speed communication channel, and at the time of writing / updating the local site. If the revision of the relevant page is older than the latest revision, the latest data is requested to the primary site through the high-speed communication channel, and at the same time the latest data is received, the relevant page of the own site is declared as the primary copy to all sites, and access A distributed database system characterized by regularly equalizing data for all sites, regardless of. 2. The distributed database system according to claim 1, further comprising means for switching the primary copy only when the number of accesses for each page reaches a predetermined value.