KR20040066942A - Extract operation technique for the indices based on the R-tree - Google Patents

Abstract

PURPOSE: A method for operating the extraction of an R-tree based index is provided to use the arranged information, when a query is entered, by making all nodes arranged along an axis in order as rearranging the node information to the highest priority axis. CONSTITUTION: From the highest priority axis in the node overlapped with a query window, a projection of the axis is overlapped with the query window. The lower nodes included in the overlapped projection are searched. In case that a leaf node and the query window are overlapped, it is checked that the line segment of a leaf node entry is overlapped with a range of the query window by using a linear interpolation method for an 'xt' and a 'yt' plane, and a result is returned. Until the query is completely processed, the previously steps are repeated.

Description

Extract operation technique for the indices based on the R-tree}

The present invention is to improve the search efficiency and to reduce the storage space in most indexes based on the R-tree. R-trees, popularly known as spatial indices, are height balanced trees that allow objects to be completely contained within overlapping k-dimensional rectangular regions without cutting or transforming k-dimensional spatial objects. Because of its dynamic structure, extended indexes in R-trees are used not only in spatial databases but also in many other applications such as time and space-time databases. However, most indexes that extend R-trees have problems with R-trees, such as overlap that prevents effective searches due to overlapping dead spaces and nodes that make unnecessary searches, and these problems become larger depending on the application. Also In order to solve this problem, the present invention proposes an indexing technique for adding an extraction operation to an R-tree. As shown in FIG. 4, the extraction operation is an operation of arranging subnodes of each node in order with respect to each axis, and storing this information in Projeciton for each axis. Therefore, each projection has the sorted pointers of the subnodes included, and this information is expected to improve the search performance by reducing the number of unnecessary node searches due to overlap and dead space when searching for information as shown in FIG. Since it has sorted node information, it can be expected to utilize partitioning cost and take up less storage space.

In order to solve the above problems, the present invention first prioritizes each axis according to an application, performs an extraction operation of obtaining aligned information of lower nodes for each non-leaf node of an R-tree, and then priorities. The technical task is to rearrange the node information around the highest axis so that all nodes are ordered according to a specific axis so that the sorted information can be used when a query comes in.

1 is a hardware configuration diagram applied to the present invention

2 is a comparison of the number of node searches of an R-tree using an R-tree and an extraction operation

3 is a structure of an R-tree-based moving point object index using an extraction operation

4 is an example of an extraction operation and a query to generate a projection for each axis

* Explanation of symbols for the main parts of the drawings

1. Central Processing Unit 2. Main Memory

3. I / O controller 4. Auxiliary memory device

5. I / O Device 6. T Projection

7. X Projection

8.Y Projection

9. Projection for each axis of the node obtained through the extraction operation

10. You leaf node

11. Leaf Node

12. Boundary of upper node and lower node used in extraction operation

13. Information of child nodes stored in the projection for each axis

14. Point-in-Time Queries

In order to achieve the above object, the present invention is to perform efficient retrieval using the extraction operation in indexing, storing and retrieving information in an R-tree-based index under a computer hardware environment. We present a mechanism that has a projection that stores the information of the lower nodes for each axis by performing the extraction operation on the y, t, and t axes. To this end, first, the extraction operation is performed on each axis to construct a projection, and the division and retrieval steps using the projection.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

1 is a hardware configuration diagram of a computer to which the present invention is applied. Various data are used to store and execute a program to be executed in a central processing unit 1 and a central processing unit 1 for controlling and managing the overall operation of the computer. And an input / output controller (3) connected to an internal connection bus and controlling inputs and outputs to the auxiliary storage device (4) and the input / output device (5). The auxiliary memory device 4 serves to store a large amount of data, and the input / output device 5 includes a general keyboard, a display device, a printer, and the like.

In the computer configured as described above, the main memory device 2 stores a moving point object indexing program for efficient searching using the extraction operation to which the present invention is applied, and is performed under the control of the central processing unit 1.

Fig. 2 shows an example of node search between an R-tree and an R-tree utilizing an extraction operation proposed in the present invention, and it can be seen that an R-tree searches more nodes even with the same query. The R-tree using the extraction operation was named Moving Point R-tree (MPR-tree) because it was applied to moving point object indexes that deal with moving objects such as cars and people to understand its efficiency. If you search for a node first, the existing R-tree searches for nodes that overlap with the range you want to search, and then searches for all child nodes that overlap with the search range until the search is completed. At this time, even if there is no node that satisfies the query, if there is a node overlapping with the search range, all nodes corresponding to the search range are searched and the query processing is completed, so that unnecessary search may be performed. On the other hand, the R-tree using the extraction operation has a Projection (9) that stores the information about the subnodes in order in each node. Therefore, only the nodes belonging to the corresponding Projection are searched by finding the Projection of each axis that overlaps the query range. This reduces the cost of searching more effectively than searching all nodes. In FIG. 2, only one projection (with respect to the t-axis) has been described and described. As the number of lower nodes increases, the deeper the level of the tree, the higher the search efficiency becomes.

3 illustrates an R-tree-based index structure using a projection obtained by performing an extraction operation. For the experiment, non-leaf nodes and leaf nodes are shown as the structure of the index used to process the moving point object information whose position changes over time. First of all, in the structure of these, the non-leaf node has the projection of each axis obtained through the extraction operation, and the projection for each axis has the information about the aligned lower nodes as shown in FIG. Leaf nodes represent the movement of a move point object as a line segment. The following describes how they are used when entering and retrieving moving point object information.

* Enter your information

I_1) Find a non-leaf node that overlaps the range of object information you want to enter.

I_2) It checks whether the borders of the retrieved non-leaf nodes' projecitons and the object information to be input overlap.

I_3) If you reach the leaf node level by repeating the above process, the object information is stored in the leaf node and pointed to the corresponding projection.

I_4) If the boundaries do not overlap, create a new Projection and connect it to the existing Projection.

I_5) The above process is repeated until all the information is entered.

Node split

Since all nodes of the tree are already sorted, there is little splitting cost, unlike an R-tree with a large splitting cost.

SL_1) Partitions the nodes so that they do not overlap so that the priority of the corresponding node is based on the information of the aligned lower nodes in the projection of the axis with the highest priority.

SL_2) If the lower nodes of the projection of the axis with the highest priority all have the same range, then the projection information of the axis with the highest priority is partitioned so that it does not overlap as much as possible.

SL_3) If all of the subordinate nodes of the projection to be divided have the same range, the process proceeds to the projection of the axis with the next priority and continues the division.

SL_4) Repeat the process of SL_3 until the division is completed.

* Information search

S_1) Check whether the overlapping query window starts from the projection of the highest priority axis in the node that overlaps the query window. Projections are also sorted in order, so you don't need to check any more if they fall outside the overlapping range of the query window.

S_2) Search for child nodes belonging to overlapping projections.

S_3) When the leaf node and the query window overlap, the linear interpolation method for the xt plane and the yt plane is used to check whether the range of the query window and the line segment information of the leaf node entry overlap and return the result.

S_4) The above process is repeated until the query processing is completed.

Through the processing described above, moving point object information input, division, and retrieval are performed. When retrieving moving point object information, by using the projection obtained through the extraction operation, it is possible to reduce unnecessary node search and splitting costs by searching only the sub-nodes of the projection that overlap with the query scope without having to search all the sub-nodes within one node. In addition, since partitioning is performed using the already sorted information, partitioning is easy and you reduce the number of leaf nodes. Therefore, if only one projection is used for one axis, it has less storage space than R-tree depending on the application, and seems to be the best form in terms of storage space and efficiency.

4 shows an extraction operation used to reduce dead space and overlap, which is a disadvantage of an R-tree in moving point object indexes. The extraction operation is to sort the subnodes of a node in order for each axis and store them in the projection for each axis. If there is node boundary information as shown in (b), lower nodes corresponding to a specific range are stored through an extraction operation as shown in (a). The information of subordinate nodes stored in the projection for each axis is as shown in (c). When searching for moving point object information by using this projection, it searches the projection overlapping with the query window first and searches the corresponding subnodes, so it is more effective than the general R-tree approach that searches all nodes. . When processing the point-in-time query 14, it is assumed that the t axis is the most important and the projection on the t axis is examined. In the case of Query 1, since the query point intersects with the Projection with the range of 2001/09/09/13/20 ~ 2001/09/09/13/35, the sub-nodes C and B belonging to this Projection have the x-axis and y Find out whether the query range for the axis is satisfied. The next projection with the range 2001/09/09/13/44 ~ 2001/09/09/13/50 is searched for intersecting the query point. The query is finished because it does not intersect. For query 2, we first look at the projection on the t-axis. Inquired point does not intersect with Projection which ranges from 2001/09/09/13/20 ~ 2001/09/09/13/35, 2001/09/09/13/44 ~ 2001/09/09/13 / It also does not intersect with the next Projection with a range of 50, thus completing the query. If we search with a generic 3D R-tree, we can find that no nodes satisfy the query after all nodes have been searched. In this way, Projection using the extraction operation can be used to reduce unnecessary node searches for efficient searches.

As described in detail above, the present invention is to improve the efficiency of search and storage space in most indexes based on R-trees. In particular, when there is a lot of information on a specific axis and the length is longer than other axes, the extraction operation is performed. You can expect better performance. Most of the indexes based on the existing R-trees searched all subnodes of nodes intersected with the query window when searching, but there was a problem of performing unnecessary searches due to dead space and overlap. However, when the extraction operation proposed in the present invention is used for the R-tree, not only unnecessary search is required but also the partitioning cost and storage space can be saved by using the sorted node information.

Claims (2)

The present invention prioritizes each axis according to the application when the information is retrieved under the computer hardware environment, inputs and retrieves the information according to the order, and uses the proejction of each axis obtained through the extraction operation to improve the performance. It presents an R-tree-based index that performs the search, and stores the information about the sorted subnodes obtained through the extraction operation in order in the projection of each axis, and uses this projection to process the effective retrieval. Mechanism to reduce space Moving Point R-tree model, a moving point object index that handles moving object information whose position changes over time, presented to experiment with the efficiency of the extraction operation.