JPH04115333A - Method for processing data base management - Google Patents

Abstract

PURPOSE:To make it possible to immediately access data in a splitted state by adding a page in the case of splitting data, updating the page IDs of a distributing page and a new added page, pointing out the distributing page from the new page by a pointer, and at the time of failing the detection of required data in the new page, retrieving also undistributed data left in the distributing page. CONSTITUTION:A split processing part 15 sets up a pointer indicating a distributing page in control information set up in a new page, and when necessary, sets up flags indicating the execution of split processing in both the pages and updates directory information or the like. Said processing is executed so as to inhibit a data base managing part 12 from accessing a data base in a state occupying the whole data base e.g., and at the time of completing the processing, access inhibition is released. At the time of executing proper data redistribution, the processing part 15 redistributes data in the distributing page and an overflow page connected to the former page to the distributing page and the new page, and when necessary, also to the overflow page.

Description

[Detailed Description of the Invention] [Summary] Regarding a database that performs area management using the dynamic hashing method in a computer, split processing using the dynamic hashing method is performed by minimizing interference with access to use the database. The purpose of the database management processing method is to divide the storage area into pages, store the data in each page group determined from the directory based on a predetermined key in the data, and store each page group in one or more pages. When a predetermined trigger is detected in the database composed of the linked pages, the new page is added as a new page group, and the source page group is added to the pointer provided corresponding to the new page group. , update the contents of the directory for the new page group and the distribution source page group, and at another predetermined opportunity, update all data stored in the distribution source page group. After completing the redistribution of all the data by redistributing it to the distribution source page group and the new page group, the pointer of the new page group is invalidated, and the data in the database is redistributed using the given key. When searching, the page group determined by the key and the page group pointed to by the valid pointer of the page group are configured to be searched.

[Industrial application field]

The present invention relates to a database management processing method for performing area management using a dynamic hashing method in a computer.

(Prior Art) FIG. 4 shows an example of the configuration of a computer system, in which a user who uses a computer database accesses the computer from, for example, a terminal 10 and starts a business processing unit 11 to perform a required business. When processing is performed and access to the database is required for that purpose, the database 13 is accessed via the database management unit 12 to search for necessary data or insert new data.Split processing The unit 15 performs split processing, which will be described later, on the database.

The data stored in the database 13 is stored in each page by dividing a large-capacity storage head such as a magnetic disk storage device into appropriately sized pages 14, and is identified by, for example, a page ID for each page. By using a function that converts a predetermined key in the data stored in that page into a page ID, you can search for the necessary data by key, obtain the page IO from the key, read the page, and retrieve the data of that page. scan the key to obtain the desired data.

In order to read a desired page in this way, page 10 is determined from the key, control information in the database 13 is accessed to determine the physical location of the page with that page ID, and a certain page When there are multiple pages to be accessed with an ID, they must be read out sequentially.

Therefore, in a database, the utilization efficiency of the storage area for storing data, that is, the ratio of valid data to the occupied storage area, and the processing efficiency of data search by key, that is, how many pages need to be accessed for one search. is the main index of database performance, and since the two generally have conflicting factors, various database management methods have been considered to obtain an appropriate trade-off between the two.

One such management method is the dynamic hashir method (for example, “Dynamic Hashir+5 S
chemesACM Computtng 5urve
ys, Vol, 20+No, 2. June 1988, p.
(See p. 85-113). This method divides the storage area into pages and stores each piece of data in a page determined from its key. The page ID is determined, and depending on the increase or decrease in the amount of data,
As the number of allocated pages increases or decreases, the bit length used as the bit string of the page ID is managed to expand or contract.

Various methods have been considered for this dynamic hashing method, but the most popular are the directory method and the linear hashing method.

Figure 2 is a diagram explaining the directory method, and Figure 2 (
al), each page consists of 2 pages.
For example, if the leaf node of the advanced tree structure is configured as shown in the figure, the directory at that time will be like (a2),
In this case, the page ID is 2 bits. The directory uses the first two bits of the binary number representing the key, and the diagram shows that the directory is configured such that, for example, when the directory is retrieved with "00", a pointer pointing to page A is obtained.

Next, for example, if page A becomes full, add page D to the position shown in Figure 2 (bl) and page I.
Expand D to 3 bits, divide the data on page A into page A and page D using the 3-bit "000" from the beginning of the key and rooI J, and create subsequent directories as shown in (b2) in the figure. Create a structure that indexes using 3-bit page IDs. The basic idea of the directory method is to increase or decrease the depth of the binary tree structure in accordance with the amount of data.

The operation of redistributing the data of page A to page A and page D is called page splitting, and the split processing unit 15 is activated and processed at an appropriate opportunity. In the case of the directory method, the trigger for splitting occurs when a certain amount of data accumulates on a certain page and the condition for splitting is reached.
You can also connect an overflow page to that page, temporarily storing data, and then delaying the time to expand the directory and perform a split.

The idea of the linear hash method is that for pages set in logically continuous areas, the page pointed to by the split pointer (SP) is the next page to be split, and each time a split is performed at an appropriate opportunity, the SP is sequentially advanced to the next page. The hash function to be applied differs depending on before and after SP.

FIG. 3 is a diagram for explaining the linear hashing method. In the example shown in the figure, in order to simplify the explanation, the difference in the hashish functions described above is shown as a difference in bit length, and the hashish function is assumed to correspond to the page ID. explain. For example, if the page ID is 2 bits in the initial state, then the page ID in Fig. 3(a)
The pages corresponding to pages IDr00" to "11" are provided in a logically continuous area, and SP points to the first page a. In this way, when data is input and, for example, the data increases by a predetermined amount, the split processing unit 15 is started using this as a trigger for splitting, and the SP
Page a pointed to by is redistributed to page a and page A as shown in (b), and SP is updated to point to the next page b.

When splitting is performed in this way, each page ID is
The page pointed to by SP and the pages to the right are the original page I
In D, the pages to the left and the added pages have page IDs that are 1 bit longer, so the database management unit 12 needs to identify them when indexing the pages.

When the predetermined amount of data further increases in this state, page b pointed to by SP is changed to page b and page B as shown in (C) in the figure.
and SP points to the next page C. If the data further increases, page C is sequentially added in the same manner as above and page C is split (Fig. 3(d)), page d
(Fig. 3(e)), at this point all pages 10 are 3 bits, and SP is again page a.
Move to. During this time, the data overflowing from each page is stored by connecting the overflow pages (pages b, d, etc. in Figure 3 (d)), and when the page is split as described above, the data on the overflow page is stored. Split including.

[Question B to be solved by the invention] As explained above, when splitting pages in a database managed using the dynamic hashing method, it is necessary to change the directory or page ID, etc., which is the mechanism for indexing the pages. Therefore, in order to avoid conflicts between the page indexing mechanism and the page contents, for example, it is necessary to add pages and complete the redistribution of data between pages while prohibiting access to the split source page. However, after that, it is necessary to temporarily prohibit access to the database, update the page indexing mechanism, and then remove the access prohibition.

In that case, redistribution of data for splitting involves the processing of determining the page ID from the key of each data and transferring the data to the determined page, which requires a relatively long processing time and prohibits access. The period becomes longer. Therefore, there is a problem in that the dynamic hashing method cannot be used effectively in fields that require a relatively short response time, such as online application work.

An object of the present invention is to provide a database management processing method that enables split processing using the dynamic hashing method to be executed with as little interference as possible to access to use the database.

[Means to solve the problem]

FIG. 1 is a process flowchart showing the configuration of the present invention.

The figure shows the configuration of a database management processing method, in which a storage area is divided into pages, the data is stored in each page group determined from a directory based on a predetermined key in the data, and each page group is divided into one or more pages. In a database composed of the linked pages, when the split processing unit detects a predetermined trigger, it adds the new page as a new page group according to FIG. 1(a). Processing step 1): Process step 2): Set a value indicating the distribution source page group in a pointer provided corresponding to the group; Process step 3): update the contents of the directory for the new page group and distribution source page group. , At another predetermined opportunity, all the data stored in the distribution source page group is redistributed to the distribution source page group and the new page group (processing step 4), and all the data is redistributed. After completing the processing step 5), the database management section invalidates the pointer of the new page group, and when searching for data in the database using the given key, the database management section invalidates the pointer of the new page group according to FIG. 1(b). The page group determined by the key and the page group pointed to by the valid pointer of the page group are searched (processing steps 6 to 9).

[For production]

When performing a split using this processing method, add a page, update the page ID of the distribution source and the additional new page, and set a pointer that indicates the distribution source page from the new page. If the required data cannot be found on a page, the unredistributed data remaining at the distribution source can also be searched according to the pointer, making it possible to access it immediately after the split. . In addition, although the trigger for data redistribution is delayed, redistribution can be performed at an opportunity that takes into consideration CPU load, etc., so it is possible to avoid concentration of load between online business processing and data redistribution processing. can. Note that if data redistribution is delayed and it is necessary to free up a storage area in order to satisfy an insert request to the distribution source, at that time, only part of the data that should have been moved from the distribution source page to the new page will be moved. Therefore, there is no need to add a new overflow page to the distribution source page group.

〔Example〕

The trigger for the split is the same as in the conventional case, and when the split processing unit 15 in FIG. 4 starts up for split processing, in processing step 1 in FIG. I'll add one more thing.

In processing step 2, the split processing unit 15 sets a pointer indicating the distribution source page in the control information provided on the new page, and if necessary, sets a flag indicating that split processing is in progress on both pages, In processing step 3, directory information etc. are updated.

The above processing is performed, for example, in a state where the entire database is occupied so as to prohibit access to the database by the database management unit 12, and when the processing is completed, the access prohibition is lifted.

The split trigger, placement of new pages, directory update contents, etc. in the above differ slightly due to differences in management methods such as the directory method and the linear hash method, but in any case, as a result of the above processing, the distribution source is Control information that allows access to the page and the new page is prepared.

Thereafter, in response to an appropriate data redistribution execution opportunity, the split processing unit 15 re-distributes the data of the distribution source page and the overflow page connected to it into the distribution source page and new page, and if necessary, the overflow page connected to them in processing step 4. However, the split processing unit 15 accesses in a manner competing with the database management unit 12 and proceeds with processing by occupying only necessary pages, so that the database management unit 12 can access in parallel.

When the data redistribution is completed in this way, the split processing flag is reset in processing step 5, and the new page pointer is reset to a predetermined value indicating that it has not been set, thereby invalidating it.

In the case of the search process, the database management unit 12 that accesses the database in parallel performs the process of searching for the page with the page ID determined by the key and the overflow pages connected to it in processing step 6, as shown in FIG. is not successful, processing step 7
If the page I is identified as unsuccessful in process step 8, the page I
It is determined whether the control information pointer of page D is valid.

If the pointer is valid, there is a possibility that data before redistribution remains on the page pointed to by the pointer, so in processing step 9, the page pointed to by the pointer is set as a new search target. Return to step 6 and perform the search.

FIG. 5 is a diagram illustrating an example of processing when implementing the present invention in the directory method. To simplify the explanation, it is assumed that one page has a capacity to store two pieces of data. Also, the page ID is the last bit of the key.

Initially, data for keys 002 and 003 are stored on a page pointed to by the directory as shown in Figure 5(a), and when you try to insert data for key 005, the page does not exist. As shown in (b), an empty page is added, a pointer pointing to the original page is set on the new page, and a flag indicating that split processing is in progress is set to "1".

In this state, the key of the data to be inserted is 005, so
As shown in (C), it is moved to page 10 r I J, and this completes the split, so the flag is reset to "0" and the pointer is invalidated (indicated by φ in the figure). For example, if it becomes necessary to insert data for key 008 while there is data 002.003 on the page, the split processing flag is "1", so the data to be moved during the insertion processing is Alternatively, the data may be searched for and moved.

FIG. 6 is a diagram illustrating an example of processing when implementing the present invention in the linear hash method, and this example also assumes that one page has a capacity to store two data. Also, the page ID is the last bit string of the key.

Initially, there are four pages with page IDs r00J to "11", and SP points to page ID rio. For example, each time three pieces of data are inserted, page IDrl
oo Pages to the right of J are added one by one, and S
P is moved one page to the right and split processing is performed, and as shown in FIG. 6(a), page 10 rlOJ
A case will be described in which a page ID rllOJ is added as a page for splitting a page.

Therefore, SP is advanced to page ID rllJ, the pointer of the new page rllOJ is set to point to the original page ro10J, and the flags of the new page, the original page, and its overflow page are set to "1". In this state, normal access by the database management unit 12 is possible, and if data is redistributed in response to another redistribution execution, the result will be as shown in (b).

Note that data redistribution is proceeded page by page from the original page ro10J, and when the data redistribution of page r010J is finished, the pointer of the new page is changed to roloJ.
Update the page to point to the overflow page that leads to it, and use rllo J during the split process.
It can improve the processing efficiency of retrieving page data.

In the above explanation, the directory method and linear hashing method have been described as dynamic hashing methods, but other methods such as the so-called partially expanded hashing method can also be used.
If necessary for each method, the present invention can be similarly applied by making changes such as providing a plurality of pointers pointing from the new page to the original page.

〔Effect of the invention〕

As is clear from the above description, according to the present invention, split processing can be performed in a database whose area is managed by a computer's dynamic hashing method with as little interference as possible to access to use the database. It has significant industrial effects.

[Brief explanation of the drawing]

Fig. 1 is a process flowchart showing the configuration of the present invention, Fig. 2 is an explanatory diagram of the directory method, Fig. 3 is an explanatory diagram of the linear hash method, Fig. 4 is a block diagram of an example configuration of a computer system, and Fig. 5 is an explanatory diagram of the directory method. FIG. 6 is an explanatory diagram of an embodiment of the directory method of the present invention. FIG. 6 is an explanatory diagram of an embodiment of the linear hash method of the present invention. In the figure, 1 to 9 are processing steps, 10 is a terminal, 11 is a business processing unit, 12 is a database management unit,
13 is a database, 14 is a page, (al) (a2) (bl) (b2) An explanatory diagram of the directory method. Figure 2. A process flow diagram showing the configuration of the present invention. Figure 1. An explanatory diagram of the linear hash method. Figure 3. A computer system. Configuration example bronc diagram Diagram explanatory diagram of an embodiment of the directory method of the present invention Diagram explanatory diagram of an embodiment of the linear hash method of the present invention

Claims (1)

[Claims] A storage area is divided into pages, the data is stored in each page group determined from a directory based on a predetermined key in the data, and each page group is composed of one or more linked pages. When a predetermined trigger is detected in the database, a new page is added as a new page group (1), and a value indicating the distribution source page group is added to a pointer provided corresponding to the new page group. (2), updates the contents of the directory for the new page group and the distribution source page group (3), and at another predetermined trigger, updates all data stored in the distribution source page group. is redistributed to the distribution source page group and the new page group (4), and after completing the redistribution of all the data, the pointer of the new page group is invalidated (5), and the database is When searching for data using a given key, the page group determined by the key and the page group pointed to by the valid pointer of the page group are searched (6 to 9). A database management processing method comprising: