How cluster will tune the query

Question

Hi All,
Please help me in understanding the autotrace Regarding Clusters.
I created 2 tables
tbl_master (col1 number, col2 number) with 800 rows AND unique col1.
and tbl_detail(col1 number, col2 number) with 80000 with repeated col1.
Now I created the replica of both tables with cluster on col1. as
tbl_master1 (col1 number, col2 number) with 800 rows AND unique col1. --selected from tbl_master
and tbl_detail1(col1 number, col2 number) with 80000 with repeated col1.--selected from tbl_detail
I traced the out put of the non cluster table joins ans clustered tables joins.
I set autotrace trace
and set timing on
I found that time elapsed is almost same even when COST is less. could anybody pls explain.
select a.* from tbl_master a, tbl_detail b where a.col1 =b.col1
80000 rows selected.
Elapsed: 00:00:00.67
Execution Plan
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=98 Card=80000 Byte
s=1040000)
1 0 HASH JOIN (Cost=98 Card=80000 Bytes=1040000)
2 1 TABLE ACCESS (FULL) OF 'TBL_MASTER' (TABLE) (Cost=55 Car
d=800 Bytes=8000)
3 1 TABLE ACCESS (FULL) OF 'TBL_DETAIL' (TABLE) (Cost=42 Car
d=80000 Bytes=240000)
Statistics
0 recursive calls
0 db block gets
5724 consistent gets
0 physical reads
0 redo size
1066812 bytes sent via SQL*Net to client
59175 bytes received via SQL*Net from client
5335 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
80000 rows processed
select a.* from tbl_master1 a, tbl_detail1 b where a.col1 =b.col1
80000 rows selected.
Elapsed: 00:00:00.65
Execution Plan
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=980 Card=72000 Byt
es=1656000)
1 0 NESTED LOOPS (Cost=980 Card=72000 Bytes=1656000)
2 1 TABLE ACCESS (FULL) OF 'TBL_MASTER1' (CLUSTER) (Cost=180
Card=800 Bytes=8000)
3 1 TABLE ACCESS (CLUSTER) OF 'TBL_DETAIL1' (CLUSTER) (Cost=
1 Card=90 Bytes=1170)
Statistics
0 recursive calls
0 db block gets
7749 consistent gets
0 physical reads
0 redo size
1065475 bytes sent via SQL*Net to client
59175 bytes received via SQL*Net from client
5335 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
80000 rows processed

Answer 1

There is a lot of information missing from your question.
1. What version of Oracle?
2. Cluster by what? Index or hash? The DDL would have helped.
and if hash was the number of has buckets determined correctly?
3. Did you use DBMS_STATS to gather relevant optimizer statistics?
4. What you posted is not an explain plan. In the future use dbms_xplan.display
http://www.psoug.org/reference/explain_plan.html
as demonstrated at the above link
All of that said ... explain plan says nothing about how long a query takes. The cost is a measure of disk i/o and, if done properly, CPU, based on Oracle's assumption of what it might do if it actually ran the query. It might do what EXPLAIN PLAN indicates and it might not.
Run a real explain plan and it might help understand what is happening but, quite frankly, with such a small number of rows the timing difference might be invisible.