Using histograms with dbms_stats

Oracle Tips by Burleson Consulting
February 3, 2004
Don Burleson

Starting in Oraclerelease 2 we have several super-exciting ways to automate the tedious task of providing the cost-based SQL optimizer with index histograms:

Details on new method_opt features:

These automate the detection of columns that require histograms, and automatically create them:

method_opt=>'for all columns size skewonly'

method_opt=>'for all columns size repeat'

method_opt=>'for all columns size auto'

method_opt=>'for all columns for column (xxx(yyy))'

Remember, analyzing for histograms is time-consuming, and histograms are used under two conditions:

Table join order ? The CBO must know the size of the intermediate result sets (cardinality) to properly determine the correct join order the multi-table joins. This is normally performed for foreign key constraints, and non-constraint, non-unique columns that are used in table joins. This is a critical task for the CBO to determine the optimal sequence for joining many tables together as efficiently as possible.
Table access method ? The CBO needs to know about columns in SQL where clauses, where the column value is skewed such that a full-table scan might be faster than an index range scan. Oracle uses this skew information in conjunction with the clustering_factor columns of the dba_indexes view.

Hence, this is the proper order for using the dbms_stats package to locate proper columns for histograms:

1. Skewonly option - You want to use skewonly to do histograms for skewed columns, for cases where the value will make a difference between a full-table scan and an index scan.

2. Monitor - Next, turn-on monitoring. Issue an ?alter table xx monitoring? and ?alter index yyy monitoring? command for all segments in your schema. This will monitor workload against

3. Auto option - Once monitoring is in-place, you need to re-analyze with the "auto" option to create histograms for join columns within tables. This is critical for the CBO to determine the proper join order for finding the driving table in multi-table joins.

4. Repeat option - Finally, use the "repeat" option to re-analyze only the existing histograms.

Periodically you will want to re-run the skewonly and auto option to identify any new columns that require histograms. Once located, the repeat option will ensure that they are refreshed with current values.

Oracle guru Guy Harrison also offers this advice for 11g statistics collection for a function-based index column.

DBMS_STATS.gather_table_stats
   (ownname => USER,
    tabname => ?SALES?,
    method_opt => ?FOR ALL COLUMNS FOR COLUMNS
   (sale_category(amount_sold))? );

��