Performance issue with context search

Question

We have a performance problem with a table tise with about 10 mio rows
and a text column tise_desc with short descriptions (about 300 characters per row).
We currently use mixed queries of the form (here very simple)
SELECT /*+ FIRST_ROWS(10) */ * FROM tise
WHERE reg_id = 'REGI0000000000000132'
AND contains(tise_desc, '(employment)' ) > 0
When the structured query part (here reg_id) and the fulltext part (contains) is not selective some queries take about 30 sec to 90 sec to process.
When we repeat the query it only takes about 1-3 sec (may be due to caching).
We are not interested in scoring nor in sorting the data.
Until now we have tried to use different hints, use different index types (btree versus bitmap), use different sql query syntax, use one fulltext index with xml and within syntax without any real progress.
Has any body addtional ideas?

Answer 1

Since you are doing a combination of structured queries and simple text queries on only one column, a ctxcat index with a catsearch may be better than a context index with a contains search. The catsearch has fewer additional features available than contains, but some context features can be done with catsearch through a query template. If you only do simple searches and don't need those extra features then it doesn't matter.
Make sure you have current statistics, so that the optimizer can use them to select the best execution plan.
Use bind variables instead of literal values in your searches. When you use bind variables, the previous query in the SGA can be reused for different variable values without reparsing, so that every query with new values has the same effect as rerunning the previous query. This alone could account for the different between 30 to 90 seconds on the first run and 1-3 seconds on the second run.
If you are not using scoring or sorting, then you might be better off without the first_rows hint, which chooses the best path for sorting. Try it both ways and see which works best for you.
Once you have tested with a ctxcat index, current statistics, and bind variables, if it is still slow you may be able to use some tracing to determine the cause of the slowness and adjust some memory settings or some such thing. You can find more tuning hints for Oracle Text here:
http://download.oracle.com/docs/cd/B28359_01/text.111/b28303/aoptim.htm#i1007227
Please see the demonstration below that implements the recommendations above.
SCOTT@orcl_11g> CREATE TABLE tise
  2    (reg_id        VARCHAR2 (20),
  3       tise_desc  VARCHAR2 (300))
  4  /
Table created.
SCOTT@orcl_11g> INSERT INTO tise VALUES ('REGI0000000000000132', 'employment')
  2  /
1 row created.
SCOTT@orcl_11g> INSERT INTO tise SELECT object_id, object_name FROM all_objects
  2  /
68770 rows created.
SCOTT@orcl_11g> -- try a ctxcat index instead of a context index
SCOTT@orcl_11g> -- (make sure the index is not fragmented by periodically
SCOTT@orcl_11g> --  dropping and recreating or altering and rebuilding or optimizing)
SCOTT@orcl_11g> BEGIN
  2    CTX_DDL.CREATE_INDEX_SET ('tise_iset');
  3    CTX_DDL.ADD_INDEX ('tise_iset', 'reg_id');
  4  END;
  5  /
PL/SQL procedure successfully completed.
SCOTT@orcl_11g> CREATE INDEX tise_id_desc_idx ON tise (tise_desc)
  2  INDEXTYPE IS CTXSYS.CTXCAT
  3  PARAMETERS ('INDEX SET tise_iset')
  4  /
Index created.
SCOTT@orcl_11g> -- make sure you have current statistics:
SCOTT@orcl_11g> EXEC DBMS_STATS.GATHER_TABLE_STATS (USER, 'TISE')
PL/SQL procedure successfully completed.
SCOTT@orcl_11g> -- use bind variables so that the query in the sga can be reused
SCOTT@orcl_11g> -- for different variable values without reparsing:
SCOTT@orcl_11g> VARIABLE search_reg_id VARCHAR2 (20)
SCOTT@orcl_11g> VARIABLE search_desc VARCHAR2 (2000)
SCOTT@orcl_11g> EXEC :search_reg_id := 'REGI0000000000000132'
PL/SQL procedure successfully completed.
SCOTT@orcl_11g> EXEC :search_desc := 'employment'
PL/SQL procedure successfully completed.
SCOTT@orcl_11g> -- query using catsearch with a ctxcat index and bind variables:
SCOTT@orcl_11g> COLUMN tise_desc FORMAT A30 WORD_WRAPPED
SCOTT@orcl_11g> SET AUTOTRACE ON EXPLAIN
SCOTT@orcl_11g> SELECT *
  2  FROM   tise
  3  WHERE  CATSEARCH (tise_desc, :search_desc, 'reg_id=''' || :search_reg_id || '''') > 0
  4  /
REG_ID               TISE_DESC
REGI0000000000000132 employment
Execution Plan
Plan hash value: 409728589
| Id  | Operation                   | Name             | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT            |                  |  3439 |   100K|   102   (0)| 00:00:02 |
|   1 |  TABLE ACCESS BY INDEX ROWID| TISE             |  3439 |   100K|   102   (0)| 00:00:02 |
|*  2 |   DOMAIN INDEX              | TISE_ID_DESC_IDX |       |       |            |          |
Predicate Information (identified by operation id):
   2 - access("CTXSYS"."CATSEARCH"("TISE_DESC",:SEARCH_DESC,'reg_id='''||:SEARCH_REG_ID|
              |'''')>0)
SCOTT@orcl_11g> SET AUTOTRACE OFF
SCOTT@orcl_11g>