Sunday, November 7, 2010

Using Database Hints

The application developer and end users know more about the data and how it is used than the optimizer does. Oracle provides a method known as HINTS to enable you to tell the optimizer the method to use for the SQL statement. Oracle recommends that HINTS not be used as the main method of controlling the optimization for SQL statements. Instead, the SQL statement should be appropriately rewritten for better performance.

You can use hints to specify:

  • The optimization approach for a SQL statement
  • The goal of the cost-based approach for a SQL statement
  • The access path for a table accessed by the statement
  • The join order for a join statement
  • A join operation in a join statement

A SQL statement can have only one comment containing HINTS. The HINT must be placed after the SELECT, UPDATE, or DELETE keyword in the SQL statement. It should be preceded by /*+ and followed by */ as seen in this example:

/*+ FULL(a) */

An alternate approach is to precede the HINT with --+. If multiple HINTS are used, they must be separated by spaces.

Here is an example of a statement using a hint of the first type:

SELECT /*+ FULL(a) */ *
WHERE empno > 1;

Note in this case that we are using the FULL hint, which indicates to Oracle that a full tablescan should occur on the EMP table. Note too that we have aliased the hint, which is a good practice but we could have used the table name as well. The other hint form could have been used in this manner:

SELECT --+ FULL(emp)
WHERE empno > 1;

Insights into HINTS Usage

  • If HINTS are incorrectly specified, Oracle will treat the HINT as a comment and will ignore it during SQL statement optimization. You will NOT receive an error message.
  • If multiple HINTS exist, Oracle will ignore those with syntax errors but will use those that are correctly included in the statement.
  • If any of the HINTS provide conflicting optimization requests, Oracle will not choose between them, and conflicting HINTS will be ignored.
  • The optimizer recognizes hints only when using the cost-based approach. If you include a hint (except the RULE hint) in a statement block, the optimizer automatically uses the cost-based approach.

HINTS and Their Meanings




Uses the Cost Based Optimizer if tables or indexes are analyzed. You should set this option if you have a database that is used predominantly for batch processing such as a Data Warehouse.


Uses the Cost Based Optimizer if tables or indexes are analyzed. You should set this option if you have a database that is used predominantly for OLTP processing.


If any of the tables or indexes in the statement are analyzed, use the Cost Based Optimizer otherwise use the Rule Based.


Use the rule based optimizer for this statement.


Used to join single column indexes. You must specify at least 2 indexes.


INTO table…causes the data being inserted to be placed at the end of the table. It does not use free space in the earlier blocks of the table. This is new with Oracle8.


Places blocks read into the most recently used end of the buffer cache which will retain the data in the buffer cache longer.


Uses a cluster scan. Clusters store each child related to a parent in the same physical address.


Disables cursor sharing if it is enabled.


Forces query execution to be done at a different site.


Allows you to control dynamic sampling.


Used for queries containing grouping sets (such as queries with GROUP BY GROUPING SET or GROUP BY ROLLUP). The hint forces a query to be transformed into a corresponding query with UNION ALL of individual groupings.


Denotes that the hinted table is a fact table when the transformation is a star transformation.


Tells Oracle to perform a full table scan on the table, even if there is an index in place. If you have the degree of parallelism set, it may also cause the table to be read using parallel query processors.


Uses a hash scan to access the specified table.


Uses a hash anti-join to speed up NOT IN.


Uses a hash semi-join to speed up EXISTS.


Informs the optimizer to use a specific index on the specified table.


Informs the optimizer to use a specific index on the specified table.


Informs the optimizer to use a specific index on the specified table. Used for bitmap indexes.


Informs the optimizer to use a specific index on the specified table.


Informs the optimizer to perform a fast full index scan instead of a full table scan. It is faster than a normal index scan.


Informs the optimizer to use an index join as the access path.


The LEADING hint causes Oracle to use the specified table as the first table in the join order. An ORDERED hint will override this hint.


Merge a view on a per-query basis.


Causes NOT IN to be processed using a Merge Join. It is often significantly faster than standard NOT processing.


Transforms a correlated EXISTS subquery into a merge semi-join to access the specified table.


INTO table…overrides APPEND, which is used by default with parallel inserts.


Places the data into the least recently used end of the buffer cache, which is standard behavior. It also overrides the CACHE setting on a table.


Prevents the cost-based optimizer from considering OR-expansion for queries having OR conditions or INLISTS in the WHERE clause.


Denotes that the hinted table is not a fact table when the transformation is a star transformation.


Explicitly disallows a set of indexes for the specified table.


Causes Oracle not to merge views specified in the FROM clause.


Prevents a join predicate from being pushed into the view.


Delays evaluation of non-merged subqueries until the last step in the execution plan.


Overrides the degree of parallelism against a table to run in non parallel mode.


Overrides a PARALLEL attribute setting on an index.


Prevents pushing a join predicate into the view.


Prevents unnesting for specific subquery blocks.


Disables query rewrite for the query block.


Causes the SQL to be driven by the tables in the order left to right.


Causes the order of predicate evaluation to be preserved by the optimizer.


Sets the number of parallel processors to scan a table.


Will use parallel query processes for fast full index scans for indexes which have PARALLEL set.


Improves parallel join operations.


Forces pushing a join predicate into the view.


Forces a join predicate to be pushed into the view.


Place this hint in a non merged subquery if the subquery performs little processing.


Use with or without a view list to select the materialized view to be used.


Uses a table scan by rowid.


Causes Oracle to merge the reference tables together and join them to the central table using a nested loop


Causes Oracle to use a star query. It does not always use cartesian product of the reference tables, unlike the STAR hint.


Merges nested subqueries into the body of the statement that contains them. The optimizr then considers them together as it evaluates access paths and joins.


Causes all ORs in the statement to be transferred to UNION ALLs.


Tables are joined to the row resulting from using a hash join.


Selects a row from one row and then returns the associated row from another table using an index. Use for OLTP.


Will sort each table and merge the rows together. Use for batch processing.

RULE, CHOOSE, ALL_ROWS and FIRST_ROWS can be applied at the instance level using the INIT.ora OPTIMIZER_MODE parameter. If set in the INIT.ora, the setting will apply for all SQL statements. The INIT.ora is usually the preferred method to apply the hints unless one particular statement ran well under the Rule Based and is performing unacceptably using the Cost Based.

The CHOOSE and RULE HINTS specify whether the CBO or the RBO should be used. Correctly specified HINTS will override the optimizer mode specified. If an optimization approach is specified, that approach will be used regardless of the initialization parameter setting for OPTIMIZER_MODE or the session setting for OPTIMIZER_GOAL. Even if statistics are present for one table and the HINT specifies CHOOSE, the optimizer will use the CBO. If no statistics are available, the optimizer will use the RBO. The use of the RULE HINT will cause the optimizer to ignore any other HINTS specified.

Both the ALL_ROWS and the FIRST_ROWS use the CBO. If no statistics are available, the optimizer will use whatever storage information is available. The ANALYZE command should be used to provide statistics before using either the ALL_ROWS or the FIRST_ROWS HINT. If a HINT specifying an access path or join operation is also specified, it will be given precedence over the ALL_ROWS and FIRST_ROWS HINTS.

The ALL_ROWS HINT concentrates on the best throughput with the minimum total resource consumption. FIRST_ROWS optimizes with the goal of the best response time with the minimum resource usage necessary to return the first row. The FIRST_ROWS HINT will be ignored for DELETE and UPDATE statements.

If the described access method requires an index that does not exist, the HINT will be ignored. The table must be specified in the HINT the same as it is in the SQL statement. If an alias is used for the table, the table specified in the HINT must use the table alias instead of the table name. You cannot use the schema name for the table, even if the table is fully qualified in the FROM clause.


No comments:

Post a Comment