4
Indexing and Querying Spatial Data

4.1.1 Creating R-Tree Indexes

If you create a spatial index without specifying any quadtree-specific parameters, an R-tree index is created. For example, the following statement creates a spatial R-tree index named territory_idx using default values for parameters that apply to R-tree indexes:

CREATE INDEX territory_idx ON territories (territory_geom)
   INDEXTYPE IS MDSYS.SPATIAL_INDEX;

For detailed information about options when creating a spatial index, see the documentation for the CREATE INDEX statement in Chapter 8.

R-tree indexes can be built on two, three, or four dimensions of data. The default number of dimensions for an R-tree index is two, but if the data has more than two dimensions, you can use the sdo_indx_dims parameter keyword to specify the number of dimensions on which to build the index. However, if a spatial index has been built on more than two dimensions of a layer, the only spatial operator that can be used against that layer is SDO_FILTER (the primary filter or index-only query), which considers all dimensions. The SDO_RELATE, SDO_NN, and SDO_WITHIN_DISTANCE operators are disabled if the index has been built on more than two dimensions.

If the rollback segment is not large enough, an attempt to create an R-tree index will fail. The rollback segment should be 100*n bytes, where n is the number of rows of data to be indexed. For example, if the table contains 1 million (1,000,000) rows, the rollback segment size should be 100,000,000 (100 million bytes).

To ensure an adequate rollback segment, or if you have tried to create an R-tree index and received an error that a rollback segment cannot be extended, review (or have a DBA review) the size and structure of the rollback segments that are available to the schema that owns the table with the geometries. Create a public rollback segment of the appropriate size, and place that rollback segment online. In addition, ensure that any small inappropriate rollback segments are placed offline during large spatial index operations. For information about performing these operations on a rollback segment, see the Oracle9i Database Administrator's Guide.

The system parameter SORT_AREA_SIZE affects the amount of time required to create the index. The SORT_AREA_SIZE value is the maximum amount, in bytes, of memory to use for a sort operation. The optimal value depends on the database size, but a good guideline is to make it at least 1 million bytes when you create an R-tree index. To change the SORT_AREA_SIZE value, use the ALTER SESSION statement. For example, to change the value to 20 million bytes:

ALTER SESSION SET SORT_AREA_SIZE = 20000000;

The tablespace specified with the tablespace keyword in the CREATE INDEX statement (or the default tablespace if the tablespace keyword is not specified) is used to hold both the index data table and some transient tables that are created for internal computations.

• The R-tree index data table requires approximately 70*n bytes (where n is the number of rows in the table)

• The transient tables require up to approximately 200*n bytes (where n is the number of rows in the table); however, this space is freed up after the R-tree index is created.

For large databases (over 1 million rows), a temporary tablespace may be needed to perform internal sorting operations. The recommended size for this temporary tablespace is 100*n bytes, where n is the number of rows in the table.

4.1.2 Determining Index Creation Behavior (Quadtree Indexes)

With a quadtree index, the tessellation algorithm used by the CREATE INDEX statement and by index maintenance routines on insert or update operations is determined by the SDO_LEVEL and SDO_NUMTILES values, which are supplied in the PARAMETERS clause of the CREATE INDEX statement. They are interpreted as follows:

An explicit commit operation is executed after the tessellation of all the geometries in a geometry column.

By default, spatial index creation requires a sizable amount of rollback space. To reduce the amount of rollback space required, you can supply the SDO_COMMIT_INTERVAL parameter in the CREATE INDEX statement. This will perform a database commit after every n geometries are indexed, where n is a user-defined value.

4.1.3 Spatial Indexing with Fixed-Size Tiles (Quadtree Indexes)

If you choose quadtree indexing for a spatial index, you should use fixed indexing for most applications, except for the rare circumstances where hybrid indexing should be considered. (See Appendix B for information about hybrid indexing. However, you should also consider using R-tree indexing before deciding on hybrid indexing.)

The fixed-size tile algorithm is expressed as a level referring to the number of tessellations performed. To use fixed-size tile indexing, omit the SDO_NUMTILES parameter and set the SDO_LEVEL value to the desired tiling level. The relationship between the tiling level and the resulting size of the tiles depends on the domain of the layer.

The domain used for indexing is defined by the upper and lower boundaries of each dimension stored in the DIMINFO column of the USER_SDO_GEOM_METADATA view, which contains an entry for the table and geometry column to spatially index. A typical domain could be -180 to 180 degrees for longitude, and -90 to 90 degrees for latitude, as represented in Figure 4-1. (The transference of the domain onto a sphere or other projection is left up to an application, unless a coordinate system is specified, as explained in Chapter 5.)

If the SDO_LEVEL column is set to 1, then the tiles created by the indexing mechanism are the same size as tiles at the first level of tessellation. Each tile would be 180 degrees by 90 degrees as shown in Figure 4-2.

The formula for the number of fixed-size tiles in a domain is 4ⁿ where n is the number of tessellations, stored in the SDO_LEVEL column. In reality, tiles are only generated where geometries exist, and not for the whole domain. Figure 4-3 shows fixed-size tiling at level 2. In this figure, each tile is 90 degrees by 45 degrees.

The size of a tile can be determined by applying the following formula to each dimension:

length = (upper_bound - lower_bound) / 2 ^ sdo_level

The length refers to the length of the tile along the specified dimension. Applying this formula to the tiling shown in Figure 4-3 yields the following sizes:

length for dimension X = (180 - (-180) ) / 2^2
                       = (360)  / 4
                       = 90
length for dimension Y = (90 - (-90) ) / 2^2
                       = (180) / 4
                       = 45

At level 2, the tiles are 90 degrees by 45 degrees in size. As the number of levels increases, the tiles become smaller and smaller. Smaller tiles provide a more precise fit of the tiles over the geometry being indexed. However, because the number of tiles generated is unbounded, you must take into account the performance implications of using higher levels.

Besides the performance aspects related to selecting a fixed-size tile, tessellating the geometry into fixed-size tiles might have benefits related to the type of data being stored, such as using tiles sized to represent 1-acre farm plots, city blocks, or individual pixels on a display. Data modeling, an important part of any database design, is essential in a spatial database where the data often represents actual physical locations.

In Example 4-1, assume that data has been loaded into a table called ROADS, and the USER_SDO_GEOM_METADATA view has an entry for ROADS.SHAPE. You can use the following SQL statement to create a fixed index named ROADS_FIXED.

CREATE INDEX ROADS_FIXED ON ROADS(SHAPE) INDEXTYPE IS MDSYS.SPATIAL_INDEX
  PARAMETERS('SDO_LEVEL=8');

The SDO_LEVEL value is used while tessellating objects. Increasing the level results in smaller tiles and better geometry approximations.

4.1.4 Indexing Geodetic Data

To take full advantage of Spatial features, you must index geodetic data using a geodetic R-tree index. Geodetic data consists of geometries that have geodetic SDO_SRID values, reflecting the fact that they are based on a geodetic coordinate system (such as using longitude and latitude) as opposed to a flat or projected plane coordinate system. (Chapter 5 explains coordinate systems and related concepts.) A geodetic index is one that provides the full range of Spatial features with geodetic data. Thus, it is highly recommended that you use a geodetic index with geodetic data.

Only R-tree indexes can be geodetic indexes. Quadtree indexes cannot be geodetic indexes. If you create an R-tree or quadtree index and specify 'geodetic=false' in the CREATE INDEX statement, the index is non-geodetic. The following notes and restrictions apply to non-geodetic indexes:

• If you create a non-geodetic index on geodetic data, you cannot use the unit parameter with the SDO_WITHIN_DISTANCE operator or the SDO_NN_DISTANCE ancillary operator with the SDO_NN operator.

• If you create a non-geodetic index on projected data that has a projected SDO_SRID value, you can use the full range of Spatial features.

• If you create a non-geodetic index on projected data that has a null SDO_SRID value, you cannot use the unit parameter with the SDO_WITHIN_DISTANCE operator or the SDO_NN_DISTANCE ancillary operator with the SDO_NN operator.

For additional information, see the Usage Notes about the geodetic parameter for the CREATE INDEX statement in Chapter 8.

4.1.5 Constraining Data to a Geometry Type

When you create or rebuild a spatial index, you can ensure that all geometries that are in the table or that are inserted later are of a specified geometry type. To constrain the data to a geometry type in this way, use the layer_gtype keyword in the PARAMETERS clause of the CREATE INDEX or ALTER INDEX REBUILD statement, and specify a value from the Geometry Type column of Table 2-1 in Section 2.2.1. For example, to constrain spatial data in a layer to polygons:

CREATE INDEX cola_spatial_idx
ON cola_markets(shape)
INDEXTYPE IS MDSYS.SPATIAL_INDEX
PARAMETERS ('layer_gtype=POLYGON');

The geometry types in Table 2-1 are considered as a hierarchy when data is checked:

• The MULTI forms include the regular form also. For example, specifying 'layer_gtype=MULTIPOINT' allows the layer to include both POINT and MULTIPOINT geometries.

• COLLECTION allows the layer to include all types of geometries.

4.1.6 Creating a Cross-Schema Index

You can create a spatial index on a table that is not in your schema. Assume that user B wants to create a spatial index on column GEOMETRY in table T1 under user A's schema. User B must perform the following steps:

1. Connect as user A (or have user A connect) and execute the following statement:

   GRANT select on T1 to B;
   

1. Connect as user B and execute a statement such as the following:

   GRANT create table to A;
   CREATE INDEX t1_spatial_idx on A.T1(geometry)
     INDEXTYPE IS mdsys.spatial_index;
   

4.1.7 Using Partitioned Spatial Indexes

You can create a partitioned spatial index on a partitioned table. This section describes usage considerations specific to Oracle Spatial. For a detailed explanation of partitioned tables and partitioned indexes, see the Oracle9i Database Administrator's Guide.

A partitioned spatial index can provide the following benefits:

• Reduced response times for long-running queries, because partitioning reduces disk I/O operations

• Reduced response times for concurrent queries, because I/O operations run concurrently on each partition

• Easier index maintenance, because of partition-level create and rebuild operations

  Indexes on partitions can be rebuilt without affecting the queries on other partitions, and storage parameters for each local index can be changed independent of other partitions.

The following restrictions apply to spatial index partitioning:

• The partition key for spatial tables must be a scalar value, and must not be a spatial column.

• Only range partitioning is supported on the underlying table. Hash and composite partitioning are not currently supported for partitioned spatial indexes.

To create a partitioned spatial index, you must specify the LOCAL keyword. For example:

CREATE INDEX counties_idx ON counties(geometry)
   INDEXTYPE IS MDSYS.SPATIAL_INDEX LOCAL;

In this example, the default values are used for the number and placement of index partitions, namely:

• Index partitioning is based on the underlying table partitioning. For each table partition, a corresponding index partition is created.

• Each index partition is placed in the default tablespace.

If you do specify parameters for individual partitions, the following considerations apply:

• The storage characteristics for each partition can be the same or different for each partition. If they are different, it may enable parallel I/O (if the tablespaces are on different disks) and may improve performance.

• Any Oracle Spatial parameters (relating to R-tree or quadtree indexing) should be the same for each partition.

To override the default partitioning values, use a CREATE INDEX statement with the following general format:

CREATE INDEX <indexname> ON <table>(<column>) 
  INDEXTYPE IS MDSYS.SPATIAL_INDEX 
     [PARAMETERS ('<spatial-params>, <storage-params>')] LOCAL 
     [( PARTITION <index_partition> 
        PARAMETERS ('<spatial-params>, <storage-params>') 
     [, PARTITION <index_partition> 
        PARAMETERS ('<spatial-params>, <storage-params>')] 
     )] 

For example, if the COUNTIES table has two partitions, P1 and P2, you can create a quadtree index as follows:

CREATE INDEX counties_idx ON counties(geometry) 
 INDEXTYPE IS MDSYS.SPATIAL_INDEX
    PARAMETERS ('sdo_level=6 tablespace=def_tbs') 
LOCAL 
    (PARTITION ip1 PARAMETERS ('sdo_level=6 tablespace=local_tbs1'), 
     PARTITION ip2 PARAMETERS ('sdo_level=6 tablespace=local_tbs2'),
     PARTITION ip3); 

In the preceding example:

• IP1 is the index partition that corresponds to partition P1 of the COUNTIES table, and IP2 is the index partition that corresponds to partition P2.

• The tablespace parameters are specified for each of the local index partitions as LOCAL_TBS1 and LOCAL_TBS2, respectively.

• If you omit the PARTITION ip2 PARAMETERS... clause (as is done for partition IP3), the default parameters specified before LOCAL are used. Specifically, the DEF_TBS tablespace is used for storing the index partition (which will have the same name as the second partition of the COUNTIES table, that is, P2).

Queries can operate on partitioned tables to perform the query on only one partition. For example:

SELECT * FROM counties PARTITION(p1)
    WHERE ...<some-spatial-predicate>;

Querying on a selected partition may speed up the query and also improve overall throughput when multiple queries operate on different partitions concurrently.

When queries use a partitioned spatial index, the semantics (meaning or behavior) of spatial operators and functions is the same with partitioned and nonpartitioned indexes, except in the case of SDO_NN (nearest neighbor). With SDO_NN, the requested number of geometries is returned for each partition that is affected by the query. For example, if you request the 5 closest restaurants to a point and the spatial index has 4 partitions, SDO_NN returns up to 20 (5*4) geometries. In this case, you must use the ROWNUM pseudocolumn (here, WHERE ROWNUM <=5) to return the 5 closest restaurants. See the description of the SDO_NN in Chapter 10 for more information.

4.1.8 Exchanging Partitions Including Indexes

You can use the ALTER TABLE statement with the EXCHANGE PARTITION...INCLUDING INDEXES clause to exchange a spatial table partition and its index partition with a corresponding table and its index. For information about exchanging partitions, see the description of the ALTER TABLE statement in the Oracle9i SQL Reference.

This feature can help you to operate more efficiently in a number of situations, such as:

• Bringing data into a partitioned table and avoiding the cost of index re-creation.

• Managing and creating partitioned indexes. For example, the data could be divided into multiple tables. The index for each table could be built one after the other to minimize the memory and tablespace resources needed during index creation. Alternately, the indexes could be created in parallel in multiple sessions. The tables (along with the indexes) could then be exchanged with the partitions of the original data table.

• Managing offline insert operations. New data can be stored in a temporary table and periodically exchanged with a new partition (for example, in a database with historical data).

To exchange partitions including indexes with spatial data and indexes, the two spatial indexes (one on the partition, the other on the table) must be of compatible types. Specifically:

• Both indexes must be either R-tree indexes with the same dimensionality (sdo_indx_dims value) or quadtree indexes with the same SDO_LEVEL and SDO_NUMTILES values.

• Both indexes must be either geodetic or non-geodetic. (Geodetic and non-geodetic indexes are explained in Section 4.1.4.)

• Neither index can have a status of deferred updates. (Deferred update status is set by specifying 'index_status=deferred' with the ALTER INDEX statement, as described in Chapter 8.)

If the indexes not compatible, an error is raised. The table data is exchanged, but the indexes are not exchanged and the indexes are marked as failed. To use the indexes, you must rebuild them.

4.2.1 Spatial Query

An important concept in the spatial data model is that each geometry is represented by a set of exclusive and exhaustive tiles. This means that no tiles overlap each other (exclusive), and the tiles fully cover the object (exhaustive).

Consider the following layer containing several objects in Figure 4-4. Each object is labeled with its geometry name (geom_1 for the polygon with a hole, geom_2 for the triangular polygon, geom_3 for the line, and geom_4 and geom_5 for other polygons). The relevant tiles are labeled with Tn (T1, T2, and so on)

A typical spatial query is to request all objects that lie within a defined fence or window. A query window is shown in Figure 4-5 by the dotted-line box. A dynamic query window refers to a fence that is not defined in the database, but that must be defined before it is used.

SDO_FILTER(geometry1 MDSYS.SDO_GEOMETRY, geometry2 MDSYS.SDO_GEOMETRY,

  params VARCHAR2)

SELECT A.Feature_ID FROM TARGET A

 WHERE sdo_filter(A.shape, mdsys.sdo_geometry(2003,NULL,NULL,
                                       mdsys.sdo_elem_info_array(1,1003,3),
                                       mdsys.sdo_ordinate_array(x1,y1, x2,y2)),
                        'querytype=window') = 'TRUE';

SELECT A.Feature_ID FROM TARGET A

 WHERE sdo_filter(A.shape, :theWindow,'querytype=window') = 'TRUE'; 

SELECT A.Feature_ID FROM TARGET A, WINDOWS B

 WHERE B.ID = 'WINS_1' AND

  sdo_filter(A.shape, B.shape,'querytype=window') = 'TRUE';

SDO_RELATE(geometry1  MDSYS.SDO_GEOMETRY,
            geometry2  MDSYS.SDO_GEOMETRY,
            params     VARCHAR2)

SELECT A.Feature_ID FROM TARGET A
   WHERE sdo_relate(A.shape, mdsys.sdo_geometry(2003,NULL,NULL,
                                       mdsys.sdo_elem_info_array(1,1003,3),
                                      mdsys.sdo_ordinate_array(x1,y1, x2,y2)),

                        'mask=anyinteract querytype=window') = 'TRUE';

SELECT A.Feature_ID FROM TARGET A, WINDOWS B

 WHERE B.ID= 'WINS_1' AND

        sdo_relate(A.shape, B.shape,
          'mask=anyinteract querytype=window') = 'TRUE'; 

SDO_WITHIN_DISTANCE(geometry1  MDSYS.SDO_GEOMETRY, 
                    aGeom      MDSYS.SDO_GEOMETRY,
                    params     VARCHAR2);

SELECT A.Feature_ID
  FROM TARGET A
  WHERE SDO_WITHIN_DISTANCE( A.shape, :theWindow, 'distance=1.35') = 'TRUE';

CREATE TABLE cosine_bufs UNRECOVERABLE AS
   SELECT SDO_BUFFER (A.SHAPE, B.DIMINFO, 1.35)
     FROM COSINE A, USER_SDO_GEOM_METADATA B
     WHERE TABLE_NAME='COSINES' AND COLUMN_NAME='SHAPE';

SELECT a.gif, b.gid FROM parks A cosine_bufs B
  WHERE SDO_Relate(A.shape, B.shape, 'mask=ANYINTERACT querytype=JOIN') ='TRUE';

SDO_NN(geometry1  MDSYS.SDO_GEOMETRY, 
       geometry2  MDSYS.SDO_GEOMETRY,
       param      VARCHAR2
       [, number  NUMBER]);

SELECT /*+ INDEX(cola_markets cola_spatial_idx) */
 c.mkt_id, c.name  FROM cola_markets c  WHERE SDO_NN(c.shape,
   mdsys.sdo_geometry(2001, NULL, mdsys.sdo_point_type(10,7,NULL), NULL,
   NULL),  'sdo_num_res=2') = 'TRUE'; 

4.2.2 Spatial Join

A spatial join is the same as a regular join except that the predicate involves a spatial operator. In Spatial, a spatial join takes place when you compare all the geometries of one layer to all the geometries of another layer. This is unlike a query window that only compares a single geometry to all geometries of a layer.

In a spatial join, all tables must have the same type of spatial index (that is, R-tree or quadtree) defined on the geometry column; and if they have quadtree indexes, the SDO_LEVEL value must be the same for all the indexes.

Spatial joins can be used to answer questions such as, Which highways cross national parks?

The following table structures illustrate how the join would be accomplished for this example:

PARKS(    GID VARCHAR2(32), SHAPE MDSYS.SDO_GEOMETRY)
HIGHWAYS( GID VARCHAR2(32), SHAPE MDSYS.SDO_GEOMETRY) 

The primary filter would identify pairs of GID values from the PARKS and HIGHWAYS tables that interact in their index entries. The query that performs the primary filter join is:

SELECT A.GID, B.GID

  FROM PARKS A, HIGHWAYS B

  WHERE sdo_filter(A.shape, B.shape, 'querytype=join') = 'TRUE';

The original question, asking about highways that cross national parks, requires the secondary filter operator to find the exact relationship between highways and parks.

The query that performs this join using both primary and secondary filters is:

SELECT A.GID, B.GID 
  FROM parks A, highways B
  WHERE sdo_relate(A.shape, B.shape,

           'mask=ANYINTERACT querytype=join');

4.2.3 Cross-Schema Operator Invocation

You can invoke spatial operators on an indexed table that is not in your schema. Assume that user A has a spatial table T1 (with index table IDX_TAB1) with a spatial index defined, that user B has a spatial table T2 (with index table IDX_TAB2) with a spatial index defined, and that user C wants to invoke operators on tables in one or both of the other schemas.

If user C wants to invoke an operator only on T1, user C must perform the following steps:

1. Connect as user A and execute the following statements:

   GRANT select on T1 to C;
   GRANT select on idx_tab1 to C;
   
   

2. Connect as user C and execute a statement such as the following:

   SELECT a.gid 
     FROM T1 a
     WHERE sdo_filter(a.geometry, :theGeometry, 'querytype=WINDOW') = 'TRUE';
   
   

If user C wants to invoke an operator on both T1 and T2, user C must perform the following steps:

1. Connect as user A and execute the following statements:

   GRANT select on T1 to C;
   GRANT select on idx_tab1 to C;
   
   

2. Connect as user B and execute the following statements:

   GRANT select on T2 to C;
   GRANT select on idx_tab2 to C;
   
   

3. Connect as user C and execute a statement such as the following:

   SELECT a.gid
     FROM T1 a, T2 b
     WHERE b.gid = 5 AND 
           sdo_filter(a.geometry, b.geometry, 'querytype=WINDOW') = 'TRUE';