Oracle Clustering Factor
Version 11.2.0.3

Data Distribution Demo
conn uwclass/uwclass

-- create demo table
CREATE TABLE t1 (
doc_id    VARCHAR2(10),
person_id NUMBER(7),
doc_name  VARCHAR2(45));

-- create document id sequence
CREATE SEQUENCE seq_t;

-- load demo table
DECLARE
 t_docname dbms_sql.VARCHAR2_TABLE;

 CURSOR c IS
 SELECT object_name
 FROM dba_objects_ae;
BEGIN
  OPEN c;
  LOOP
    FETCH c BULK COLLECT INTO t_docname LIMIT 250;

    FORALL i IN 1..t_docname.COUNT
    INSERT INTO t1
    (doc_id, person_id, doc_name)
    VALUES
    (seq_t.NEXTVAL, MOD(seq_t.CURRVAL, 233), t_docname(i));

    EXIT WHEN c%NOTFOUND;
  END LOOP;
  COMMIT;
  CLOSE c;
END;
/

-- look at the first 300 rows of data
SELECT *
FROM t1
WHERE rownum < 301;

-- verify the data for each person is equivalent
SELECT person_id, COUNT(*)
FROM t1
GROUP BY person_id
ORDER BY 1;

-- begin tracing
set autotrace traceonly

-- select all documents belonging to person 221
SELECT doc_name
FROM t1
WHERE person_id = 221;
-- save the autotrace / note Oracle does an FTS (full table scan)

-- create a normal B*Tree index
CREATE INDEX ix_t1_person_id
ON t1(person_id);

exec dbms_stats.gather_table_stats(USER, 'T1', CASCADE=>TRUE);

set autotrace off

SELECT clustering_factor
FROM user_indexes
WHERE table_name = 'T1';

set autotrace traceonly

-- repeat the select ... note Oracle still does an FTS
SELECT doc_name
FROM t1
WHERE person_id = 221;

set linesize 121

-- force Oracle to use the index
SELECT /*+ INDEX(t1 ix_t1_person_id) */ doc_name
FROM t1
WHERE person_id = 221;
-- note that the cost went up

set autotrace off
 
Highly Clustered Data Distribution Demo
-- recreate sequence object
DROP SEQUENCE seq_t;
CREATE SEQUENCE seq_t;

-- create an new identical table and load it with each person's documents highly clustered
CREATE TABLE t2 AS
SELECT * FROM t1
WHERE 1=2;

DECLARE
 x PLS_INTEGER;
BEGIN
  INSERT INTO t2
  (doc_id, doc_name)
  SELECT rownum, object_name FROM dba_objects_ae;

  FOR i IN 1..235
  LOOP
    UPDATE t2
    SET person_id = i
    WHERE person_id IS NULL
    AND rownum < 234;

    SELECT COUNT(*)
    INTO x
    FROM t2
    WHERE person_id IS NULL;

    EXIT WHEN x = 0;
  END LOOP;
  COMMIT;
END;
/

-- look at the first 300 rows of data
SELECT *
FROM t2
WHERE rownum < 301;

-- verify the data for each person is equivalent
SELECT person_id, COUNT(*)
FROM t2
GROUP BY person_id
ORDER BY 1;

-- begin tracing
set autotrace traceonly

-- select all documents belonging to person 221
SELECT doc_name
FROM t2
WHERE person_id = 221;
-- save the autotrace / note Oracle does an FTS (full table scan)

-- recreate a B*Tree index
CREATE INDEX ix_t2_person_id
ON t2(person_id);

exec dbms_stats.gather_table_stats(USER, 'T2', CASCADE=>TRUE);

SELECT clustering_factor
FROM user_indexes
WHERE table_name = 'T2';

-- repeat the query without a hint
SELECT doc_name
FROM t2
WHERE person_id = 221;

-- note that Oracle uses the index and the cost is much lower
 
Another Index Quality Query SELECT dt.table_name, di.index_name, blocks, clustering_factor, (blocks/clustering_factor) CF_RATIO
FROM dba_tables dt, dba_indexes di
WHERE dt.owner NOT LIKE '%SYS%'
AND dt.owner = di.owner
AND dt.table_name = di.table_name
AND dt.blocks > 0
AND di.clustering_factor > (blocks*5)
ORDER BY 5;

Related Topics
Indexes
SYS_OP_LBID
SYS_OP_RPB

Morgan's Library Page Footer
This site is maintained by Dan Morgan. Last Updated: This site is protected by copyright and trademark laws under U.S. and International law. © 1998-2013 Daniel A. Morgan All Rights Reserved