Release 8.0
A58227-01
 Library |
 Product |
 Contents |
 Index |
| Oracle8 Concepts Release 8.0 A58227-01 |
|
My object all sublime I shall achieve in time - To let the punishment fit the crime.
Sir William Schwenck Gilbert: The Mikado
This chapter discusses the different types of database objects contained in a user's schema. It includes:
Associated with each database user is a schema. A schema is a collection of schema objects. Examples of schema objects include tables, views, sequences, synonyms, indexes, clusters, database links, snapshots, procedures, functions, and packages.
Schema objects are logical data storage structures. Schema objects do not have a one-to-one correspondence to physical files on disk that store their information. However, Oracle stores a schema object logically within a tablespace of the database. The data of each object is physically contained in one or more of the tablespace's datafiles. For some objects such as tables, indexes, and clusters, you can specify how much disk space Oracle allocates for the object within the tablespace's datafiles.
There is no relationship between schemas and tablespaces: a tablespace can contain objects from different schemas, and the objects for a schema can be contained in different tablespaces. Figure 8-1 illustrates the relationship among objects, tablespaces, and datafiles.
|
Additional Information:
This chapter explains tables, views, sequences, synonyms, indexes, and clusters. Other kinds of schema objects are explained elsewhere in this manual, or in other manuals. Specifically:
|
Tables are the basic unit of data storage in an Oracle database. Data is stored in rows and columns. You define a table with a table name (such as EMP) and set of columns. You give each column a column name (such as EMPNO, ENAME, and JOB), a datatype (such as VARCHAR2, DATE, or NUMBER), and a width (the width might be predetermined by the datatype, as in DATE) or precision and scale (for columns of the NUMBER datatype only). A row is a collection of column information corresponding to a single record.
See Chapter 10, "Built-In Datatypes", for a discussion of the Oracle datatypes.
You can optionally specify rules for each column of a table. These rules are called integrity constraints. One example is a NOT NULL integrity constraint. This constraint forces the column to contain a value in every row. See Chapter 24, "Data Integrity", for more information about integrity constraints.
Once you create a table, you insert rows of data using SQL statements. Table data can then be queried, deleted, or updated using SQL.
Figure 8-2 shows a sample table named EMP.
When you create a non-clustered table, Oracle automatically allocates a data segment in a tablespace to hold the table's future data. You can control the allocation of space for a table's data segment and use of this reserved space in the following ways:
Oracle stores data for a clustered table in the data segment created for the cluster. Storage parameters cannot be specified when a clustered table is created or altered; the storage parameters set for the cluster always control the storage of all tables in the cluster.
The tablespace that contains a non-clustered table's data segment is either the table owner's default tablespace or a tablespace specifically named in the CREATE TABLE statement. See "User Tablespace Settings and Quotas" on page 25-8.
Oracle stores each row of a database table as one or more row pieces. If an entire row can be inserted into a single data block, Oracle stores the row as one row piece. However, if all of a row's data cannot be inserted into a single data block or an update to an existing row causes the row to outgrow its data block, Oracle stores the row using multiple row pieces. A data block usually contains only one row piece per row. When Oracle must store a row in more than one row piece, it is "chained" across multiple blocks. A chained row's pieces are chained together using the ROWIDs of the pieces. See "Row Chaining and Migrating" on page 2-10.
Each row piece, chained or unchained, contains a row header and data for all or some of the row's columns. Individual columns might also span row pieces and, consequently, data blocks.
Figure 8-3 shows the format of a row piece.
The row header precedes the data and contains information about
A non-clustered row fully contained in one block has at least three bytes of row header. After the row header information, each row contains column length and data. The column length requires one byte for columns that store 250 bytes or less, or three bytes for columns that store more than 250 bytes, and precedes the column data. Space required for column data depends on the datatype. If the datatype of a column is variable length, the space required to hold a value can grow and shrink with updates to the data.
To conserve space, a null in a column only stores the column length (zero). Oracle does not store data for the null column. Also, for trailing null columns, Oracle does not store the column length because the row header signals the start of a new row (for example, the last three columns of a table are null, thus there is no information stored for those columns).
Clustered rows contain the same information as non-clustered rows. In addition, they contain information that references the cluster key to which they belong. See "Clusters" on page 8-32.
The column order is the same for all rows in a given table. Columns are usually stored in the order in which they were listed in the CREATE TABLE statement, but this is not guaranteed. For example, if you create a table with a column of datatype LONG, Oracle always stores this column last. Also, if a table is altered so that a new column is added, the new column becomes the last column stored.
In general, you should try to place columns that frequently contain nulls last so that rows take less space. Note, though, that if the table you are creating includes a LONG column as well, the benefits of placing frequently null columns last are lost.
The ROWID identifies each row piece by its location or address. Once assigned, a given row piece retains its ROWID until the corresponding row is deleted, or exported and imported using the IMPORT and EXPORT utilities. If the cluster key values of a row change, the row keeps the same ROWID, but also gets an additional pointer ROWID for the new values.
Because ROWIDs are constant for the lifetime of a row piece, it is useful to reference ROWIDs in SQL statements such as SELECT, UPDATE, and DELETE. See "ROWID Datatype" on page 10-12 for more information.
A null is the absence of a value in a column of a row. Nulls indicate missing, unknown, or inapplicable data. A null should not be used to imply any other value, such as zero. A column allows nulls unless a NOT NULL or PRIMARY KEY integrity constraint has been defined for the column, in which case no row can be inserted without a value for that column.
Nulls are stored in the database if they fall between columns with data values. In these cases they require one byte to store the length of the column (zero). Trailing nulls in a row require no storage because a new row header signals that the remaining columns in the previous row are null. In tables with many columns, the columns more likely to contain nulls should be defined last to conserve disk space.
Most comparisons between nulls and other values are by definition neither true nor false, but unknown. To identify nulls in SQL, use the IS NULL predicate. Use the SQL function NVL to convert nulls to non-null values.
|
Additional Information:
See Oracle8 SQL Reference for more information about comparisons using IS NULL and the NVL function. |
Nulls are not indexed, except when the cluster key column value is null or the index is a bitmap index (see "Bitmap Indexes and Nulls" on page 8-27).
You can assign a column of a table a default value so that when a new row is inserted and a value for the column is omitted, a default value is supplied automatically. Default column values work as though an INSERT statement actually specifies the default value.
Legal default values include any literal or expression that does not refer to a column, LEVEL, ROWNUM, or PRIOR. Default values can include the SQL functions SYSDATE, USER, USERENV, and UID. The datatype of the default literal or expression must match or be convertible to the column datatype.
If a default value is not explicitly defined for a column, the default for the column is implicitly set to NULL.
Integrity constraint checking occurs after the row with a default value is inserted. For example, in Figure 8-4, a row is inserted into the EMP table that does not include a value for the employee's department number. Because no value is supplied for the department number, Oracle inserts the DEPTNO column's default value "20". After inserting the default value, Oracle checks the FOREIGN KEY integrity constraint defined on the DEPTNO column.
For more information about integrity constraints, see Chapter 24, "Data Integrity".
In the Oracle object-relational database, you can create a table with a column whose datatype is another table. That is, tables can be nested within other tables as values in a column. The Oracle server stores nested table data "out of line" from the rows of the parent table, using a store table which is associated with the nested table column. The parent row contains a unique set identifier value associated with a nested table instance.
A view is a tailored presentation of the data contained in one or more tables (or other views). A view takes the output of a query and treats it as a table; therefore, a view can be thought of as a "stored query" or a "virtual table". You can use views in most places where a table can be used.
For example, the EMP table has several columns and numerous rows of information. If you only want users to see five of these columns, or only specific rows, you can create a view of that table for other users to access. Figure 8-5 shows an example of a view called STAFF derived from the base table EMP. Notice that the view shows only five of the columns in the base table.
Since views are derived from tables, they have many similarities. For example, you can define views with up to 1000 columns, just like a table. You can query views, and with some restrictions you can update, insert into, and delete from views. All operations performed on a view actually affect data in some base table of the view and are subject to the integrity constraints and triggers of the base tables.
Unlike a table, a view is not allocated any storage space, nor does a view actually contain data; rather, a view is defined by a query that extracts or derives data from the tables the view references. These tables are called base tables. Base tables can in turn be actual tables or can be views themselves (including snapshots). Because a view is based on other objects, a view requires no storage other than storage for the definition of the view (the stored query) in the data dictionary.
Views provide a means to present a different representation of the data that resides within the base tables. Views are very powerful because they allow you to tailor the presentation of data to different types of users.
Views are often used
For example, Figure 8-5 shows how the STAFF view does not show the SAL or COMM columns of the base table EMP.
For example, a single view might be defined with a join, which is a collection of related columns or rows in multiple tables. However, the view hides the fact that this information actually originates from several tables.
For example, views allow users to select information from multiple tables without actually knowing how to perform a join.
For example, the columns of a view can be renamed without affecting the tables on which the view is based.
For example, if a view's defining query references three columns of a four column table and a fifth column is added to the table, the view's definition is not affected and all applications using the view are not affected.
For example, a view can be defined that joins a GROUP BY view with a table, or a view can be defined that joins a UNION view with a table.
For example, a query could perform extensive calculations with table information. By saving this query as a view, the calculations can be performed each time the view is queried.
Oracle stores a view's definition in the data dictionary as the text of the query that defines the view. When you reference a view in a SQL statement, Oracle merges the statement that references the view with the query that defines the view and then parses the merged statement in a shared SQL area and executes it. Oracle parses a statement that references a view in a new shared SQL area only if no existing shared SQL area contains an identical statement. Therefore, you obtain the benefit of reduced memory usage associated with shared SQL when you use views.
In evaluating views containing string literals or SQL functions that have NLS parameters as arguments (such as TO_CHAR, TO_DATE, and TO_NUMBER), Oracle takes default values for these parameters from the NLS parameters for the session. You can override these default values by specifying NLS parameters explicitly in the view definition.
Oracle determines whether to use indexes for a query against a view by transforming the original query when merging it with the view's defining query. Consider the view
CREATE VIEW emp_view AS SELECT empno, ename, sal, loc FROM emp, dept WHERE emp.deptno = dept.deptno AND dept.deptno = 10;
Now consider the following user-issued query:
SELECT ename FROM emp_view WHERE empno = 9876;
The final query constructed by Oracle is
SELECT ename FROM emp, dept WHERE emp.deptno = dept.deptno AND dept.deptno = 10 AND emp.empno = 9876;
In all possible cases, Oracle merges a query against a view with the view's defining query (and those of the underlying views). Oracle optimizes the merged query as if you issued the query without referencing the views. Therefore, Oracle can use indexes on any referenced base table columns, whether the columns are referenced in the view definition or the user query against the view.
In some cases, Oracle cannot merge the view definition with the user-issued query. In such cases, Oracle may not use all indexes on referenced columns.
Because a view is defined by a query that references other objects (tables, snapshots, or other views), a view is dependent on the referenced objects. Oracle automatically handles the dependencies for views. For example, if you drop a base table of a view and then recreate it, Oracle determines whether the new base table is acceptable to the existing definition of the view. See Chapter 19, "Oracle Dependency Management", for a complete discussion of dependencies in a database.
A join view is defined as a view with more than one table or view in its FROM clause and which does not use any of these clauses: DISTINCT, AGGREGATION, GROUP BY, START WITH, CONNECT BY, ROWNUM, and set operations (UNION ALL, INTERSECT, and so on).
An updatable join view is a join view, which involves two or more base tables or views, where UPDATE, INSERT, and DELETE operations are permitted. The data dictionary views ALL_UPDATABLE_COLUMNS, DBA_UPDATABLE_COLUMNS, and USER_UPDATABLE_COLUMNS contain information that indicates which of the view columns are updatable. Table 8-1 lists rules for updatable join views.
Views that are not updatable can be modified using INSTEAD OF triggers. See "INSTEAD OF Triggers" on page 18-11 for more information.
In the Oracle object-relational database, object views allow you to retrieve, update, insert, and delete relational data as if they were stored as object types. You can also define views that have columns which are object datatypes, such as objects, REFs, and collections (nested tables and VARRAYs).
|
Additional Information:
See Chapter 13, "Object Views" and the Oracle8 Application Developer's Guide. |
The sequence generator provides a sequential series of numbers. The sequence generator is especially useful in multi-user environments for generating unique sequential numbers without the overhead of disk I/O or transaction locking. Therefore, the sequence generator reduces "serialization" where the statements of two transactions must generate sequential numbers at the same time. By avoiding the serialization that results when multiple users wait for each other to generate and use a sequence number, the sequence generator improves transaction throughput and a user's wait is considerably shorter.
Sequence numbers are Oracle integers defined in the database of up to 38 digits. A sequence definition indicates general information: the name of the sequence, whether it ascends or descends, the interval between numbers, and other information. One important part of a sequence's definition is whether Oracle should cache sets of generated sequence numbers in memory.
Oracle stores the definitions of all sequences for a particular database as rows in a single data dictionary table in the SYSTEM tablespace. Therefore, all sequence definitions are always available, because the SYSTEM tablespace is always online.
Sequence numbers are used by SQL statements that reference the sequence. You can issue a statement to generate a new sequence number or use the current sequence number. Once a statement in a user's session generates a sequence number, the particular sequence number is available only to that session; each user that references a sequence has access to its own, current sequence number.
Sequence numbers are generated independently of tables. Therefore, the same sequence generator can be used for one or for multiple tables. Sequence number generation is useful to generate unique primary keys for your data automatically and to coordinate keys across multiple rows or tables. Individual sequence numbers can be skipped if they were generated and used in a transaction that was ultimately rolled back. Applications can make provisions to catch and reuse these sequence numbers, if desired.
|
Additional Information:
For performance implications when using sequences, see the Oracle8 Application Developer's Guide. |
A synonym is an alias for any table, view, snapshot, sequence, procedure, function, or package. Because a synonym is simply an alias, it requires no storage other than its definition in the data dictionary.
Synonyms are often used for security and convenience. For example, they can do the following:
You can create both public and private synonyms. A public synonym is owned by the special user group named PUBLIC and every user in a database can access it. A private synonym is in the schema of a specific user who has control over its availability to others.
Synonyms are very useful in both distributed and non-distributed database environments because they hide the identity of the underlying object, including its location in a distributed system. This is advantageous because if the underlying object must be renamed or moved, only the synonym needs to be redefined and applications based on the synonym continue to function without modification.
Synonyms can also simplify SQL statements for users in a distributed database system. The following example shows how and why public synonyms are often created by a database administrator to hide the identity of a base table and reduce the complexity of SQL statements. Assume the following:
At this point, you would have to query the table SALES_DATA with a SQL statement similar to the one below:
SELECT * FROM jward.sales_data;
Notice how you must include both the schema that contains the table along with the table name to perform the query.
Assume that the database administrator creates a public synonym with the following SQL statement:
CREATE PUBLIC SYNONYM sales FOR jward.sales_data;
After the public synonym is created, you can query the table SALES_DATA with a simple SQL statement:
SELECT * FROM sales;
Notice that the public synonym SALES hides the name of the table SALES_DATA and the name of the schema that contains the table.
Indexes are optional structures associated with tables and clusters. You can create indexes explicitly to speed SQL statement execution on a table. Just as the index in this manual helps you locate information faster than if there were no index, an Oracle index provides a faster access path to table data. Indexes are the primary means of reducing disk I/O when properly used.
Oracle provides several indexing schemes, which provide complementary performance functionality: B*-tree indexes (currently the most common), B*-tree cluster indexes, hash cluster indexes, reverse key indexes, and bitmap indexes.
The absence or presence of an index does not require a change in the wording of any SQL statement. An index is merely a fast access path to the data; it affects only the speed of execution. Given a data value that has been indexed, the index points directly to the location of the rows containing that value.
Indexes are logically and physically independent of the data in the associated table. You can create or drop an index at anytime without affecting the base tables or other indexes. If you drop an index, all applications continue to work; however, access of previously indexed data might be slower. Indexes, as independent structures, require storage space.
Oracle automatically maintains and uses indexes once they are created. Oracle automatically reflects changes to data, such as adding new rows, updating rows, or deleting rows, in all relevant indexes with no additional action by users.
Retrieval performance of indexed data remains almost constant, even as new rows are inserted. However, the presence of many indexes on a table decreases the performance of updates, deletes, and inserts because Oracle must also update the indexes associated with the table.
The optimizer can use an existing index to build another index. This results in a much faster index build.
Indexes can be unique or non-unique. Unique indexes guarantee that no two rows of a table have duplicate values in the columns that define the index. Non-unique indexes do not impose this restriction on the column values.
Oracle recommends that you do not explicitly define unique indexes on tables; uniqueness is strictly a logical concept and should be associated with the definition of a table. Alternatively, define UNIQUE integrity constraints on the desired columns. Oracle enforces UNIQUE integrity constraints by automatically defining a unique index on the unique key.
A composite index (also called a concatenated index) is an index that you create on multiple columns in a table. Columns in a composite index can appear in any order and need not be adjacent in the table.
Composite indexes can speed retrieval of data for SELECT statements in which the WHERE clause references all or the leading portion of the columns in the composite index. Therefore, the order of the columns used in the definition is important; generally, the most commonly accessed or most selective columns go first.
Figure 8-6 illustrates the VENDOR_PARTS table that has a composite index on the VENDOR_ID and PART_NO columns.
No more than 32 columns can form a regular composite index, and for a bitmap index the maximum number columns is 30. A key value cannot exceed roughly one-half (minus some overhead) the available data space in a data block.
Although the terms are often used interchangeably, you should understand the distinction between "indexes" and "keys". Indexes are structures actually stored in the database, which users create, alter, and drop using SQL statements. You create an index to provide a fast access path to table data. Keys are strictly a logical concept. Keys correspond to another feature of Oracle called integrity constraints.
Integrity constraints enforce the business rules of a database (see Chapter 24, "Data Integrity"). Because Oracle uses indexes to enforce some integrity constraints, the terms key and index are often are used interchangeably; however, they should not be confused with each other.
When you create an index, Oracle automatically allocates an index segment to hold the index's data in a tablespace. You control allocation of space for an index's segment and use of this reserved space in the following ways:
The tablespace of an index's segment is either the owner's default tablespace or a tablespace specifically named in the CREATE INDEX statement. You do not have to place an index in the same tablespace as its associated table. Furthermore, you can improve performance of queries that use an index by storing an index and its table in different tablespaces located on different disk drives because Oracle can retrieve both index and table data in parallel. See "User Tablespace Settings and Quotas" on page 25-8.
Space available for index data is the Oracle block size minus block overhead, entry overhead, ROWID, and one length byte per value indexed. The number of bytes required for the overhead of an index block is operating system dependent.
When you create an index, Oracle fetches and sorts the columns to be indexed, and stores the ROWID along with the index value for each row. Then Oracle loads the index from the bottom up. For example, consider the statement:
CREATE INDEX emp_ename ON emp(ename);
Oracle sorts the EMP table on the ENAME column. It then loads the index with the ENAME and corresponding ROWID values in this sorted order. When it uses the index, Oracle does a quick search through the sorted ENAME values and then uses the associated ROWID values to locate the rows having the sought ENAME value.
Although Oracle accepts the keywords ASC, DESC, COMPRESS, and NOCOMPRESS in the CREATE INDEX command, they have no effect on index data, which is stored using rear compression in the branch nodes but not in the leaf nodes.
Oracle uses B*-tree indexes that are balanced to equalize access times to any row. The theory of B*-tree indexes is beyond the scope of this manual; for more information you can refer to computer science texts dealing with data structures. Figure 8-7 illustrates the structure of a B*-tree index.
The upper blocks (branch blocks) of a B*-tree index contain index data that points to lower level index blocks. The lowest level index blocks (leaf blocks) contain every indexed data value and a corresponding ROWID used to locate the actual row; the leaf blocks are doubly linked. Indexes in columns containing character data are based on the binary values of the characters in the database character set.
For a unique index, there is one ROWID per data value. For a non-unique index, the ROWID is included in the key in sorted order, so non-unique indexes are sorted by the index key and ROWID. Key values containing all nulls are not indexed, except for cluster indexes. Two rows can both contain all nulls and not violate a unique index.
The B*-tree structure has the following advantages:
Creating a reverse key index, compared to a standard index, reverses the bytes of each column indexed (except the ROWID) while keeping the column order. Such an arrangement can help avoid performance degradation in indexes in an Oracle Parallel Server environment where modifications to the index are concentrated on a small set of leaf blocks. By reversing the keys of the index, the insertions become distributed across all leaf keys in the index.
Using the reverse key arrangement eliminates the ability to run an index range scanning query on the index. Because lexically adjacent keys are not stored next to each other in a reverse-key index, only fetch-by-key or full-index (table) scans can be performed.
Under some circumstances using a reverse-key index can make an OLTP Oracle Parallel Server application faster. For example, keeping the index of mail messages in Oracle Office: some users keep old messages around, and the index must maintain pointers to these as well as to the most recent.
The REVERSE keyword provides a simple mechanism for creating a reverse key index. You can specify the keyword REVERSE along with the optional index specifications in a CREATE INDEX statement:
CREATE INDEX i ON t (a,b,c) REVERSE;
You can specify the keyword NOREVERSE to REBUILD a reverse-key index into one that is not reverse keyed:
ALTER INDEX i REBUILD NOREVERSE;
Rebuilding a reverse-key index without the NOREVERSE keyword produces a rebuilt, reverse-key index. You cannot rebuild a normal index as a reverse key index; you must use the CREATE command instead.
|
Attention: Bitmap indexes are available only if you have purchased the Oracle8 Enterprise Edition. See Getting to Know Oracle8 and the Oracle8 Enterprise Edition for more information about the features available in Oracle8 and the Oracle8 Enterprise Edition. |
The purpose of an index is to provide pointers to the rows in a table that contain a given key value. In a regular index, this is achieved by storing a list of ROWIDs for each key corresponding to the rows with that key value. (Oracle stores each key value repeatedly with each stored ROWID.) In a bitmap index, a bitmap for each key value is used instead of a list of ROWIDs.
Each bit in the bitmap corresponds to a possible ROWID, and if the bit is set, it means that the row with the corresponding ROWID contains the key value. A mapping function converts the bit position to an actual ROWID, so the bitmap index provides the same functionality as a regular index even though it uses a different representation internally. If the number of different key values is small, bitmap indexes are very space efficient.
Bitmap indexing efficiently merges indexes that correspond to several conditions in a WHERE clause. Rows that satisfy some, but not all conditions are filtered out before the table itself is accessed. This improves response time, often dramatically.
Bitmap indexing benefits data warehousing applications which have large amounts of data and ad hoc queries, but a low level of concurrent transactions. For such applications, bitmap indexing provides:
Fully indexing a large table with a traditional B*-tree index can be prohibitively expensive in terms of space since the index can be several times larger than the data in the table. Bitmap indexes are typically only a fraction of the size of the indexed data in the table.
Bitmap indexes are not suitable for OLTP applications with large numbers of concurrent transactions modifying the data. These indexes are primarily intended for decision support (DSS) in data warehousing applications where users typically query the data rather than update it.
Bitmap indexes are integrated with the Oracle cost-based optimization approach and execution engine. They can be used seamlessly in combination with other Oracle execution methods. For example, the optimizer can decide to perform a hash join between two tables using a bitmap index on one table and a regular B*-tree index on the other. The optimizer considers bitmap indexes and other available access methods, such as regular B*-tree indexes and full table scan, and chooses the most efficient method, taking parallelism into account where appropriate.
Parallel query and parallel DML work with bitmap indexes as with traditional indexes. (Bitmap indexes on partitioned tables must be local indexes; see "Index Partitioning" on page 9-22 for more information.) Parallel create index and concatenated indexes are also supported.
The advantages of using bitmap indexes are greatest for low cardinality columns: that is, columns in which the number of distinct values is small compared to the number of rows in the table. If the values in a column are repeated more than a hundred times, the column is a candidate for a bitmap index. Even columns with a lower number of repetitions (and thus higher cardinality), can be candidates if they tend to be involved in complex conditions in the WHERE clauses of queries.
For example, on a table with one million rows, a column with 10,000 distinct values is a candidate for a bitmap index. A bitmap index on this column can out-perform a B*-tree index, particularly when this column is often queried in conjunction with other columns.
B*-tree indexes are most effective for high-cardinality data: that is, data with many possible values, such as CUSTOMER_NAME or PHONE_NUMBER. A regular B*-tree index can be several times larger than the indexed data. Used appropriately, bitmap indexes can be significantly smaller than a corresponding B*-tree index.
In ad hoc queries and similar situations, bitmap indexes can dramatically improve query performance. AND and OR conditions in the WHERE clause of a query can be quickly resolved by performing the corresponding boolean operations directly on the bitmaps before converting the resulting bitmap to ROWIDs. If the resulting number of rows is small, the query can be answered very quickly without resorting to a full table scan of the table.
Table 8-2 shows a portion of a company's customer data.
Since MARITAL_STATUS, REGION, GENDER, and INCOME_LEVEL are all low-cardinality columns (there are only three possible values for marital status and region, two possible values for gender, and four for income level) it is appropriate to create bitmap indexes on these columns. A bitmap index should not be created on CUSTOMER# because this is a high-cardinality column. Instead, a unique B*-tree index on this column in order would provide the most efficient representation and retrieval.
Table 8-3 illustrates the bitmap index for the REGION column in this example. It consists of three separate bitmaps, one for each region.
| REGION='east' | REGION='central' | REGION='west' |
|---|---|---|
|
1 |
0 |
0 |
|
0 |
1 |
0 |
|
0 |
0 |
1 |
|
0 |
0 |
1 |
|
0 |
1 |
0 |
|
0 |
1 |
0 |
Each entry (or "bit") in the bitmap corresponds to a single row of the CUSTOMER table. The value of each bit depends upon the values of the corresponding row in the table. For instance, the bitmap REGION='east' contains a one as its first bit: this is because the region is "east" in the first row of the CUSTOMER table. The bitmap REGION='east' has a zero for its other bits because none of the other rows of the table contain "east" as their value for REGION.
An analyst investigating demographic trends of the company's customers might ask, "How many of our married customers live in the central or west regions?" This corresponds to the following SQL query:
SELECT COUNT(*) FROM CUSTOMER WHERE MARITAL_STATUS = 'married' AND REGION IN ('central','west');
Bitmap indexes can process this query with great efficiency by merely counting the number of ones in the resulting bitmap, as illustrated in Figure 8-8. To identify the specific customers who satisfy the criteria, the resulting bitmap would be used to access the table.
Bitmap indexes include rows that have NULL values, unlike most other types of indexes. Indexing of nulls can be useful for some types of SQL statements, such as queries with the group function COUNT.
SELECT COUNT(*) FROM EMP;
Any bitmap index can be used for this query because all table rows are indexed, including those that have NULL data. If nulls were not indexed, the optimizer would only be able to use indexes on columns with NOT NULL constraints.
SELECT COUNT(*) FROM EMP WHERE COMM IS NULL;
This query can be optimized with a bitmap index on COMM.
SELECT COUNT(*) FROM CUSTOMER WHERE GENDER = 'M' AND STATE != 'CA';
This query can be answered by finding the bitmap for GENDER = 'M' and substracting the bitmap for STATE = 'CA'. If STATE may contain null values (that is, if it does not have a NOT NULL constraint), then the bitmaps for STATE = 'NULL' must also be subtracted from the result.
Like other indexes, you can create bitmap indexes on partitioned tables. The only restriction is that bitmap indexes must be local to the partitioned table - they cannot be global indexes. (Global bitmap indexes are supported only on nonpartitioned tables).
For information about partitioned tables and descriptions of local and global indexes, see Chapter 9, "Partitioned Tables and Indexes".
An index-organized table differs from a regular table in that the data for the table is held in its associated index. Changes to the table data, such as adding new rows, updating rows, or deleting rows, result only in updating the index.
The index-organized table is like a regular table with an index on one or more of its columns, but instead of maintaining two separate storages for the table and the B*-tree index, the database system only maintains a single B*-tree index which contains both the encoded key value and the associated column values for the corresponding row. Rather than having row ROWID as the second element of the index entry, the actual data row is stored in the B*-tree index. The data rows are built on the primary key for the table, and each B*-tree index entry contains <primary_key_value, non_primary_key_column_values>.
Index-organized tables are suitable for accessing data via primary key or via any key that is a valid prefix of the primary key. There is no duplication of key values because only non-key column values are stored with the key.
Applications manipulate the index-organized table just like a regular table, using SQL statements. However, the database system performs all operations by manipulating the corresponding B*-tree index.
Table 8-4 summarizes the differences between index-organized tables and regular tables.
|
Additional Information:
See Oracle8 Administrator's Guide for information about how to create and maintain index-organized tables. |
Because rows are stored in the index, index-organized tables provide a faster key-based access to table data for queries involving exact match and/or range search. The storage requirements are reduced as key columns are not duplicated in the table and the index. The data row stored with the key only contains non-key column values. Also, placing the data row with the key eliminates the additional storage required for ROWIDs that link key values to corresponding rows in the case of an index for a regular table.
B*-tree index entries are usually quite small since they only consist of the pair <key, ROWID>. In index-organized tables, however, the B*-tree index entries can be very large since they consist of the pair <key, non_key_column_values>. If the index entry gets very large then the leaf nodes may end up storing one row or row-piece thereby destroying the dense clustering property of the B*-tree index.
Oracle provides a Row Overflow Area clause to overcome this problem. You can specify an overflow tablespace as well as a threshold value. The threshold is specified as percentage of the block size.
If the row size is greater than the specified threshold value, then the non-key column values for the row that exceeds the threshold are stored in the specified overflow tablespace. In such a case the index entry contains a <key, rowhead> pair, where the rowhead contains the beginning portion of the rest of the columns. It is like a regular row-piece, except it points to an overflow row-piece that contains the remaining column values.
Index-organized tables are especially useful for the following types of applications:
Information Retrieval (IR) applications support content-based searches on document collections. To provide such a capability, IR applications maintain an inverted index for the document collection. An inverted index typically contains entries of the form <token, document_id, occurrence_data> for each distinct word in a document. The application performs a content-based search by scanning the inverted index looking for tokens of interest.
You can define a regular table to model the inverted index. To speed-up retrieval of data from the table, you can also define an index on the column corresponding to the token. However, this scheme has the following shortcomings:
In some cases, retrieval performance can be improved by defining a concatenated index on multiple columns of the inverted index table. The concatenated index allows for index-organized retrieval when the occurrence data is not required (that is, for Boolean queries). In such cases, the ROWID fetches of inverted table records is avoided. When the query involves a proximity predicate (for example, the phrase "Oracle Corporation"), the concatenated index approach still requires the inverted index table to be accessed. Furthermore, building and maintaining a concatenated index is much more time consuming than using a single column index on the token. Also, the storage overhead is higher as multiple columns of the key (token) are duplicated in the table and the index.
Using an index-organized table to model an inverted index overcomes the problems described above. Namely:
In addition, since index-organized tables are visible to the applications, they are suitable for supporting cooperative indexing where the application and database jointly manage the application-specific indexes.
Spatial applications can benefit from index-organized tables as they use some form of inverted index for maintaining application-specific indexes.
Spatial applications maintain inverted indexes for handling spatial queries. For example, a spatial index for objects residing in a collection of grids can be modeled as an inverted index where each entry is of the form:
<grid_id, spatial_object_id, spatial_object_data>
Index-organized tables are appropriate for modeling such inverted indexes because they provide the required retrieval performance while minimizing storage costs.
On-line analytical processing (OLAP) applications typically manipulate multi-dimensional blocks. To allow fast retrieval of portions of the multi-dimensional blocks, they maintain an inverted index to map a set of dimension values to set of pages. An entry in the inverted index is of the form:
<dimension_value, list_of_pages>
The inverted index maintained by OLAP applications can easily be modeled as an index-organized table.
Clusters are an optional method of storing table data. A cluster is a group of tables that share the same data blocks because they share common columns and are often used together.
For example, the EMP and DEPT table share the DEPTNO column. When you cluster the EMP and DEPT tables (Figure 8-9, "Clustered Table Data"), Oracle physically stores all rows for each department from both the EMP and DEPT tables in the same data blocks.
Because clusters store related rows of different tables together in the same data blocks, properly used clusters offer two primary benefits:
Therefore, less storage might be required to store related table and index data in a cluster than is necessary in non-clustered table format. For example, in Figure 8-9 notice how each cluster key (each DEPTNO) is stored just once for many rows that contain the same value in both the EMP and DEPT tables.
Clusters can reduce the performance of INSERT statements as compared with storing a table separately with its own index. This disadvantage relates to the use of space and the number of blocks that must be visited to scan a table; because multiple tables have data in each block, more blocks must be used to store a clustered table than if that table were stored non-clustered.
To identify data that would be better stored in clustered form than non-clustered, look for tables that are related via referential integrity constraints and tables that are frequently accessed together using a join. If you cluster tables on the columns used to join table data, you reduce the number of data blocks that must be accessed to process the query; all the rows needed for a join on a cluster key are in the same block. Therefore, performance for joins is improved. Similarly, it might be useful to cluster an individual table. For example, the EMP table could be clustered on the DEPTNO column to cluster the rows for employees in the same department. This would be advantageous if applications commonly process rows department by department.
Like indexes, clusters do not affect application design. The existence of a cluster is transparent to users and to applications. You access data stored in a clustered table via SQL just like data stored in a non-clustered table.
|
Additional Information:
For more information about the performance implications of using clusters, see Oracle8 Tuning. |
In general, clustered data blocks have an identical format to non-clustered data blocks with the addition of data in the table directory. However, Oracle stores all rows that share the same cluster key value in the same data block.
When you create a cluster, specify the average amount of space required to store all the rows for a cluster key value using the SIZE parameter of the CREATE CLUSTER command. SIZE determines the maximum number of cluster keys that can be stored per data block.
For example, if each data block has 1700 bytes of available space and the specified cluster key size is 500 bytes, each data block can potentially hold rows for three cluster keys. If SIZE is greater than the amount of available space per data block, each data block holds rows for only one cluster key value.
Although the maximum number of cluster key values per data block is fixed by SIZE, Oracle does not actually reserve space for each cluster key value nor does it guarantee the number of cluster keys that are assigned to a block. For example, if SIZE determines that three cluster key values are allowed per data block, this does not prevent rows for one cluster key value from taking up all of the available space in the block. If more rows exist for a given key than can fit in a single block, the block is chained, as necessary.
A cluster key value is stored only once in a data block.
The cluster key is the column, or group of columns, that the clustered tables have in common. You specify the columns of the cluster key when creating the cluster. You subsequently specify the same columns when creating every table added to the cluster.
For each column specified as part of the cluster key (when creating the cluster), every table created in the cluster must have a column that matches the size and type of the column in the cluster key. No more than 16 columns can form the cluster key, and a cluster key value cannot exceed roughly one-half (minus some overhead) the available data space in a data block. The cluster key cannot include a LONG or LONG RAW column.
You can update the data values in clustered columns of a table. However, because the placement of data depends on the cluster key, changing the cluster key for a row might cause Oracle to physically relocate the row. Therefore, columns that are updated often are not good candidates for the cluster key.
You must create an index on the cluster key columns after you have created a cluster. A cluster index is an index defined specifically for a cluster. Such an index contains an entry for each cluster key value.
To locate a row in a cluster, the cluster index is used to find the cluster key value, which points to the data block associated with that cluster key value. Therefore, Oracle accesses a given row with a minimum of two I/Os (possibly more, depending on the number of levels that must be traversed in the index).
You must create a cluster index before you can execute any DML statements (including INSERT and SELECT statements) against the clustered tables. Therefore, you cannot load data into a clustered table until you create the cluster index.
Like a table index, Oracle stores a cluster index in an index segment. Therefore, you can place a cluster in one tablespace and the cluster index in a different tablespace.
A cluster index is unlike a table index in the following ways:
If you drop a cluster index, data in the cluster remains but becomes unavailable until you create a new cluster index. You might want to drop a cluster index to move the cluster index to another tablespace or to change its storage characteristics; however, you must recreate the cluster's index to allow access to data in the cluster.
Hashing is an optional way of storing table data to improve the performance of data retrieval. To use hashing, you create a hash cluster and load tables into the cluster. Oracle physically stores the rows of a table in a hash cluster and retrieves them according to the results of a hash function.
Oracle uses a hash function to generate a distribution of numeric values, called hash values, which are based on specific cluster key values. The key of a hash cluster (like the key of an index cluster) can be a single column or composite key (multiple column key). To find or store a row in a hash cluster, Oracle applies the hash function to the row's cluster key value; the resulting hash value corresponds to a data block in the cluster, which Oracle then reads or writes on behalf of the issued statement.
A hash cluster is an alternative to a non-clustered table with an index or an index cluster. With an indexed table or index cluster, Oracle locates the rows in a table using key values that Oracle stores in a separate index.
To find or store a row in an indexed table or cluster, at least two I/Os must be performed (but often more): one or more I/Os to find or store the key value in the index, and another I/O to read or write the row in the table or cluster. In contrast, Oracle uses a hash function to locate a row in a hash cluster (no I/O is required). As a result, a minimum of one I/O operation is necessary to read or write a row in a hash cluster.
A hash cluster stores related rows together in the same data blocks. Rows in a hash cluster are stored together based on their hash value.
When you create a hash cluster, Oracle allocates an initial amount of storage for the cluster's data segment. Oracle bases the amount of storage initially allocated for a hash cluster on the predicted number and predicted average size of the hash key's rows in the cluster.
Figure 8-10 illustrates data retrieval for a table in a hash cluster as well as a table with an index. The following sections further explain the internal operations of hash cluster storage.
To find or store a row in a hash cluster, Oracle applies the hash function to the row's cluster key value. The resulting hash value corresponds to a data block in the cluster, which Oracle then reads or writes on behalf of an issued statement. The number of hash values for a hash cluster is fixed at creation and is determined by the HASHKEYS parameter of the CREATE CLUSTER command.
The value of HASHKEYS limits the number of unique hash values that can be generated by the hash function used for the cluster. Oracle rounds the number you specify for HASHKEYS to the nearest prime number. For example, setting HASHKEYS to 100 means that for any cluster key value, the hash function generates values between 0 and 100 (there will be 101 hash values).
Therefore, the distribution of rows in a hash cluster is directly controlled by the value set for the HASHKEYS parameter. With a larger number of hash keys for a given number of rows, the likelihood of a collision (two cluster key values having the same hash value) decreases. Minimizing the number of collisions is important because overflow blocks (thus extra I/O) might be necessary to store rows with hash values that collide.
The maximum number of hash keys assigned per data block is determined by the SIZE parameter of the CREATE CLUSTER command. SIZE is an estimate of the total amount of space in bytes required to store the average number of rows associated with each hash value. For example, if the available free space per data block is 1700 bytes and SIZE is set to 500 bytes, three hash keys are assigned per data block.
Although the maximum number of hash key values per data block is determined by SIZE, Oracle does not actually reserve space for each hash key value in the block. For example, if SIZE determines that three hash key values are allowed per block, this does not prevent rows for one hash key value from taking up all of the available space in the block. If there are more rows for a given hash key value than can fit in a single block, the block is chained, as necessary.
Note that each row's hash value is not stored as part of the row; however, the cluster key value for each row is stored. Therefore, when determining the proper value for SIZE, the cluster key value must be included for every row to be stored.
A hash function is a function applied to a cluster key value that returns a hash value. Oracle then uses the hash value to locate the row in the proper data block of the hash cluster. The job of a hash function is to provide the maximum distribution of rows among the available hash values of the cluster. To achieve this goal, a hash function must minimize the number of collisions.
When you create a cluster, you can use the internal hash function of Oracle or bypass the use of this function. The internal hash function allows the cluster key to be a single column or composite key.
Furthermore, the cluster key can consist of columns of any datatype (except LONG and LONG RAW). The internal hash function offers sufficient distribution of cluster key values among available hash keys, producing a minimum number of collisions for any type of cluster key.
In cases where the cluster key is already a unique identifier that is uniformly distributed over its range, you might want to bypass the internal hash function and simply specify the column on which to hash.
Instead of using the internal hash function to generate a hash value, Oracle checks the cluster key value. If the cluster key value is less than HASHKEYS, the hash value is the cluster key value; however, if the cluster key value is equal to or greater than HASHKEYS, Oracle divides the cluster key value by the number specified for HASHKEYS, and the remainder is the hash value; that is, the hash value is the cluster key value mod the number of hash keys.
Use the HASH IS parameter of the CREATE CLUSTER command to specify the cluster key column if cluster key values are distributed evenly throughout the cluster. The cluster key must be comprised of a single column that contains only zero scale numbers (integers). If the internal hash function is bypassed and a non-integer cluster key value is supplied, the operation (INSERT or UPDATE statement) is rolled back and an error is returned.
You can also specify any SQL expression as the hash function for a hash cluster. If your cluster key values are not evenly distributed among the cluster, you should consider creating your own hash function that more efficiently distributes cluster rows among the hash values.
For example, if you have a hash cluster containing employee information and the cluster key is the employee's home area code, it is likely that many employees will hash to the same hash value. To alleviate this problem, you can place the following expression in the HASH IS clause of the CREATE CLUSTER command:
MOD((emp.home_area_code + emp.home_prefix + emp.home_suffix), 101)
The expression takes the area code column and adds the phone prefix and suffix columns, divides by the number of hash values (in this case 101), and then uses the remainder as the hash value. The result is cluster rows more evenly distributed among the various hash values.
As with other types of segments, the allocation of extents during the creation of a hash cluster is controlled by the INITIAL, NEXT, and MINEXTENTS parameters of the STORAGE clause. However, with hash clusters, an initial portion of space, called the hash table, is allocated at creation so that all hash keys of the cluster can be mapped, with the total space equal to SIZE * HASHKEYS. Therefore, initial allocation of space for a hash cluster is also dependent on the values of SIZE and HASHKEYS. The larger of (SIZE*HASHKEYS) and that specified by the STORAGE clause (INITIAL, NEXT, and so on) is used.
Space subsequently allocated to a hash cluster is used to hold the overflow of rows from data blocks that are already full. For example, assume the original data block for a given hash key is full. A user inserts a row into a clustered table such that the row's cluster key hashes to the hash value that is stored in a full data block; therefore, the row cannot be inserted into the root block (original block) allocated for the hash key. Instead, the row is inserted into an overflow block that is chained to the root block of the hash key.
Frequent collisions might or might not result in a larger number of overflow blocks within a hash cluster (thus reducing data retrieval performance). If a collision occurs and there is no space in the original block allocated for the hash key, an overflow block must be allocated to hold the new row. The likelihood of this happening is largely dependent on the average size of each hash key value and corresponding data, specified when the hash cluster is created, as illustrated in Figure 8-11.
If the average size is small and each row has a unique hash key value, many hash key values can be assigned per data block. In this case, a small colliding row can likely fit into the space of the root block for the hash key. However, if the average hash key value size is large or each hash key value corresponds to multiple rows, only a few hash key values can be assigned per data block. In this case, it is likely that the large row will not fit in the root block allocated for the hash key value and an overflow block is allocated.
