JPH0496837A - Index constituting method for data base processing device - Google Patents

Abstract

PURPOSE:To simultaneously execute many retrieval processings and update processings without complicating the constitution of a balance tree by constituting a device of a function part, which takes a key value to be retrieved as the input to output function values, and a tree part having tree structures corresponding to respective outputted function values. CONSTITUTION:The key to be retrieved is inputted to a partition function part 2, and a directory 3 is referred to by the output value to determine a pointer to corresponding trees 4-1 to 4-3. This pointer determines one of plural trees constituting the tree part which the entry of the key to be searched is stored in one of leaf nodes 5-1 to 5-3 belonging to, and the tree pointed by the pointer is searched to search the entry having the objective key value. Consequently, highly simultaneous executability is realized though many retrieval processings and update processings exist together and are simultaneously executed.

Description

Detailed Description of the Invention [Field of Industrial Application] The present invention relates to a method of configuring an index for speeding up database search processing and update processing in a database processing system. This invention relates to an index configuration method for increasing concurrency when processes are processed in parallel.

[Prior Art] As a conventional technique for speeding up database search or update processing, secondary information called an index is generated in advance for columns specified in search/update conditions.
A method of searching for the storage position of a target row using the index is widely known. In this method, the storage position of each row stored in the data base and a collection of entries with the key value as the key value are indexed in advance, using the value of the column specified as the search/update column '1' as the key value. information indicating the storage position of the row is obtained from the entry with the key value that matches the specified condition, and the target row is searched or updated. Since the row to be processed can be directly identified from the index, database search and update processing can be speeded up.

In general, an index employs a balanced tree structure in order to search for a pointer to a target row in the shortest possible time, and to search for any pointer in a uniform amount of time. The index construction method and search/update processing method using a balanced tree are described by Jeffrey
It is described in detail in ``Principles of Database Systems J'' by D. Ulman, translated by Toshiyasu Kunii and Nobuo Oyasu.

FIG. 2 shows a conceptual diagram of a database processing device equipped with an index. The terminal 100 inputs inquiries and outputs results. Usually, a plurality of terminals 100 are used, each accepting independent processing requests. The execution control unit 110 controls the execution order so that there are no inconsistencies in the search and update processes input from the plurality of terminals 100. The database access unit 120 executes execution control gI! Based on the processing request input from llo, the database search and update processing are actually executed. The database access unit 120 uses an index depending on the search/update processing request, or uses an index if processing can be completed in a shorter time. The index storage W l 30 stores an index that is secondary information for speeding up database access. The database storage unit 140 permanently stores a database to be searched/updated, and is usually realized by a secondary storage medium such as a disk device.

Note that index storage section 130 and database storage section 14.0 may be physically integrated. less than,
The conventional method of configuring an index will be explained.

FIG. 3 is a diagram showing a conventional index structure for 14 row data. In the figure, 1 is the search starting point, 5-
10 to 5-17 hold entries for the 14 rows that make up the database; 11-1 to 11;
-4 is a branch node for searching for leaf nodes. Each entry of the leaf nodes 5-10 to 5-17 consists of a pair of a pointer ($a to $n) indicating the storage position of the row and a key value of the row (] ] 3.-88. Numerical values in branch nodes 11-1 to 11-4 represent threshold values.Search starting point and branch node, branch node and branch node/
Leaf nodes are connected by pointers as shown by arrows in the figure. The illustrated example shows a case where one branch node can branch to a maximum of three branch nodes or leaf nodes. Which node to branch to is determined by comparing the threshold value stored in the branch node with the value of the key to be searched. Note that the branch node 11-1 pointed to by the search starting point 1 is particularly referred to as a root node.

The example in Figure 3 is a third-order balanced tree that can branch in a maximum of three directions from one branch node, and the number of pairs that can be stored in a single branch node is limited to three. Generally, in an m-th order parasis tree where a branch is waiting in the m direction,
If the height of the tree is h, a maximum of In' rows can be managed in the entire tree. In an m-th order parasis tree, the root node has at least two branches, and each other branch node can take m/2 or more and m or less branches. If the number of rows is managed using an m-th balanced tree, the number of searches will be 10g,\, and even in a database with a large number of rows, the desired row can be searched for with a small number of searches. For example, m is 10
When it is 0, even if there are 1,000,000 1,000,000 entries, the entry for the target row can be obtained by searching three times.

In FIG. 3, the process of searching for (te) with a key value of 28 searches for the root node 11-1 from the search starting point 1 and compares it with the threshold value 30 that the root node 11-1 has. , search for branch node 112 using a pointer to a node smaller than 30. Search for branch node ] -2 in the same way. Compare key value 28 with thresholds (15 and 22), and from the comparison result leaf node 5 Find -12.

Then, the storage position $f of the target row is obtained from the entry of the key value 28 stored in the leaf node 5-12. In this example, the storage position of the target row is obtained from 14 rows through three searches.

FIG. 4 is a diagram showing the state of reconfiguration of the conventional balance tree when one entry is added to the index based on the balance tree shown in FIG.

The entry to be added has a key value of 8 and a storage location of $1. When the balance tree is searched using the key value 8 of the entry to be added, it is found that it is appropriate to add the entry to the leaf node 5-10 in FIG. However, the leaf node 5-10 already holds three entries and cannot add any more entries. Therefore, leaf node 5-1
o is the leaf node 5-21 and 5 as shown in FIG.
-22. Along with the division operation, the upper branch node is also updated. In the example of FIG. 3, there is no free space in the parent node 11-2 of the leaf node 5-10, so the branch note 11-2 is divided into branch nodes 11-13 and 1114 as shown in FIG. Ru. Furthermore, since there is no free space in the branch node (root node) 11-1, which is its parent, the branch node 11-1 becomes 11-11.
and 11-12, and a new root node 11-10 is added above it.

In this way, adding a new entry to the balance tree may require splitting a leaf node, which in turn requires splitting the branching node and reconstructing the balance tree, i.e., the entire index. may occur. When deleting a row from the database, the entry is deleted from the index notebook, but in this case as well, leaf nodes may be merged and the index may need to be reconfigured accordingly. While the index is being reconfigured, the pointer chain between nodes is uncertain, so other search or update processing cannot be executed.

If you reorganize the index when adding or deleting entries, the index will be reorganized frequently.
Other search and update processes are interrupted. As a conventional method to solve this problem in mass production, there is a method that does not reconfigure the index when adding or deleting an entry, but instead provides a separate opportunity to reconfigure the index.
n de J onge.

Ardie 5chijf: Concur
renL Access t.

B-trees, International
l Conferenced on Database
s, Parallel Architectures
, andTheir Applications,
I EEE Computer Society Pr
ess, pp. 312-320''.

In this method, it is proposed that a pointer chain that horizontally links leaf nodes and each branch node is formed in advance, and the index is reconfigured horizontally. Reorganize the index by adding entries/
Since this is not done when deleting entries, if entries are added and deleted repeatedly, the structure of the parasis tree will become disordered, which is explained above! &
Appropriate search time is no longer guaranteed. However, by reconfiguring the index when adding or deleting an entry, the increase in processing time due to interruption of other processes can be solved, so the process of adding or deleting entries can be speeded up. In this method, a balanced tree with a disordered structure is reconfigured at an appropriate opportunity to restore the balanced state. This reconstruction process is performed hierarchically using pointer chains that are horizontally linked between notes. The reason for this is that in the conventional method shown in Figure 4, indexes are reconfigured vertically, so the process of updating the root node occurs frequently, and the process of searching multiple indexes is executed simultaneously. This is to prevent other index search processes from being interrupted by the root node update process.

As mentioned above, the conventional method proposed above aims to speed up the index update process that involves adding and deleting entries, and to speed up the update process when multiple index search/update processes are running simultaneously. This attempt was made to resolve the factors that impede concurrency through the accompanying reconfiguration of the index structure.

[Problem to be solved by the invention] Database search and update processes can be sped up using indexes, but when multiple processes are attempted to be executed simultaneously, execution may be affected by other processes. There was a problem with waiting. In particular, in update processing that involves adding or deleting entries, it is necessary to reconfigure the index structure, and other search processing or update processing cannot be performed while the reorganization is being performed.

As a conventional technique for solving this problem, a method has been proposed in which the above-mentioned index reconfiguration is separated from the update process, and the index reconfiguration is not performed in the update process. However, this method has a problem in that the internal structure of the index becomes complicated because horizontal pointer chains are provided in the balanced tree-structured index and the index is hierarchically reconfigured.

An object of the present invention is to complicate the configuration of a balanced tree in a database processing device that is equipped with an index that obtains the storage position of a row from the value of a key that is a search condition for a database for a database consisting of a plurality of rows. An object of the present invention is to provide an index construction method that allows a large number of search processes and update processes to be executed simultaneously without any problems.

[Means for Solving the Problems] In order to achieve the above object, the index construction method of the present invention includes a function part that inputs a key value to be searched and outputs a function value, and a function part that outputs a function value by inputting a key value to be searched. The index is characterized in that an index is constructed with a tree part consisting of a plurality of trees in a tree structure corresponding to each function value. Furthermore, during update processing that involves adding or deleting entries,
Alternatively, the index may be reconfigured in units of one or more of the trees constituting the R+5 tree section based on the function value of the function, upon the occurrence of a need for index reconfiguration. It is something.

[Operation] In the index configuration method of the present invention, by configuring the index as a tree section consisting of a plurality of trees and a function section that branches to the tree section, multiple search processes and updates can be performed simultaneously. It is possible to distribute processing to multiple trees. Furthermore, when reconfiguring an index, it is possible to perform the reconfiguration in units of one or more of multiple trees, so search processing and update processing related to trees that have not been reconfigured can be executed at the same time. I can do it.

[Example] An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a diagram showing an index structure according to an embodiment of the present invention. In the figure, l is the search starting point for starting the index search, 2 is the distribution function part, 3 is the directory, 4-1 to 4-3 are the trees, and 5-1 to 5-3 are the rows corresponding to the key values. This is a leaf node consisting of entries that hold storage locations.

The distribution function part 2 is a part that calculates a function that allocates index search destinations during search processing or update processing to multiple trees from a specified key value, and a remainder function is used as an example of such a function. be able to. Directory 3 calculates pair 1. from the values obtained by distribution function section 2. It holds a pointer to the tree to be edited. , 1-1 to 4
The tree of -3 has the same structure as the balanced tree shown in the conventional example of FIG. The leaf nodes 5-1 to 5-3 are also similar to those shown in the conventional example shown in FIG. 3. Below, the search processing procedure for the index having the configuration shown in FIG. 1 will be described. First, the key to be searched is input to the distribution function section 2, and the directory 3 is checked from the output value of the distribution function section 2 to find the corresponding tree 4-1 to 4-4.
- Determine the pointer to 3. This pointer indicates which leaf node 5-1 to 5 of the multiple trees that make up the tree part contains the entry of the key to be searched.
(By searching the tree pointed to by Haboishita in the same way as the conventional method, you can search for an entry with the desired key value. For example, if the I'TI diagram In the example, the key values 1, 3, 12, 15°22.28 are the same tree 4 to
If these key values are input to the distribution function unit 2, the same value will be output from the distribution function unit 2, and if you refer to the directory 3 with the output value of 2, the tree 4-1 will be displayed. You will get a pointer to .

For update processing that involves adding or deleting entries,
As in the past, in addition to the above-mentioned search process, it is necessary to reconstruct the index to restore the state of the balanced tree that has been disrupted due to the addition or deletion of entries. Here, the fourth
As in the example shown in the figure, consider adding an entry with a key value of 8. The key value 8 is set to tree 4- by the distribution function part 2.
When it is determined that the entry is under 1, the tree 4-
Reconfigure 1. In this manner, in the index configuration shown in FIG. 1, updating of the tree by reconfiguration is limited to one tree forming the tree section.

Therefore, even if tree 4-1 is reconfigured to add an entry with key value 8, search processing and update processing for entries under other trees 4-2, 4-3, etc. can be executed simultaneously.

In the above detailed description of the present invention, the trigger for reconfiguring the index due to the addition or deletion of an entry is not clearly stipulated. It is also easy to do it independently without synchronizing.

In addition, in the embodiment shown in FIG. 1, the function part is composed of the distribution function part 2 and the directory 3, but it is possible to directly calculate pointers to the trees 4-1 to 4-3. It is also possible to implement an allocation function section. In this case, the directory 3 becomes unnecessary, and a pointer chain is formed directly from the distribution function section to each tree forming the tree section.

Furthermore, in the embodiment shown in FIG. 1, for ease of explanation, the tree part is divided into trees 4-1 to 4-3 and leaf nodes 5-1 to 5-3 connected to them. ,
A leaf node is a part of a tree, and it is easy to configure a tree with a mixture of entries. Also,
It is also easy to form a chain of pointers directly from the directory to the leaf nodes without a tree.

In short, the feature of the present invention is that the index, which was conventionally formed as a single unit such as a balance tree, is configured as a tree section consisting of a function section that calculates a distribution function and a plurality of trees pointed from the function section. [Effects of the Invention] As explained above, according to the present invention, the index can be divided into the function part and the tree. By configuring it as a single section, high concurrency can be achieved even when a plurality of search processes and update processes are mixed and executed simultaneously. In particular, the problem with conventional index structures, in which index reconfiguration accompanying update processing inhibits the execution of other search and update processing, is greatly improved, and the throughput of the entire system can be improved.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an index configuration according to an embodiment of the present invention, FIG. 2 is a conceptual diagram of a database processing installation targeted by the present invention, and FIG. 3 is a configuration example of a conventional index.
FIG. 4 is a diagram for explaining the conventional index reconfiguration in the index configuration example of FIG. 3. Index search starting point, 2 Distribution function section, 3 Directory, 4-1 to 4-
:3 Tree 5-1 to 5-3 Leaf note.

Claims (2)

[Claims] (1) For a database consisting of N (an integer of N≧2) rows, in a database processing device equipped with indexing means for obtaining the storage position of a row from the value of a key that is a search condition for the row, a key value k (k≧
An index construction method characterized in that the index is constructed from a tree section consisting of (an integer number of 2) trees. (2) When it becomes necessary to reconfigure the tree section due to the addition, deletion, or update of a row to the database, the addition, deletion, or update of the row may trigger Claim (1) characterized in that, when configuration is required, one or more of the trees constituting the tree section is reconfigured based on the function value of the function section.
) index configuration method described.