Deleting 3 millon duplicated records from 4 million records

Question

Hi,
I want to delete 3 million duplicate records from 4 million records in production. We dont have the partitions. I cann't recreate table using CTAS Option. So I have to delete the data in batch because of shared environment and people acceessing the data.
Is there any fastest way to delete the data in stead of using bulk delete by using max rowid.
please help me out.
Regards,
Venkat.
Edited by: ramanamadhav on Aug 16, 2011 8:41 AM

Answer 1

After deletion of data( with suggestion given by Justin), make sure that latest statistics are also taken.
After such a heavy delete there is always a mismatch between num_rows and actual count.
the best way is to shrink space and then gather table statistics, other wise optimizer will get confused and it will not take into consideration correct statistics.
For e.g.
16:41:55 SQL> create table test_tab_sm
as
select level col1, rpad('*888',200,'###') col2
from dual connect by level <= 100000;16:42:05   2  16:42:05   3  16:42:05   4 
Table created.
Elapsed: 00:00:00.49
16:42:08 SQL> exec dbms_stats.gather_table_stats(ownname => 'KDM', tabname => 'TEST_TAB_SM', cascade => true);
PL/SQL procedure successfully completed.
Elapsed: 00:00:00.51
16:42:17 SQL> select table_name,num_rows,blocks,avg_row_len,chain_cnt from user_tables where table_name = 'TEST_TAB_SM';
TABLE_NAME                       NUM_ROWS     BLOCKS AVG_ROW_LEN  CHAIN_CNT
TEST_TAB_SM                        100000       2942         205          0
1 row selected.
Elapsed: 00:00:00.01
16:42:28 SQL> delete from TEST_TAB_SM where mod(col1,2) =1;
50000 rows deleted.
Elapsed: 00:00:01.09
16:42:39 SQL> select table_name,num_rows,blocks,avg_row_len,chain_cnt from user_tables where table_name = 'TEST_TAB_SM';
TABLE_NAME                       NUM_ROWS     BLOCKS AVG_ROW_LEN  CHAIN_CNT
TEST_TAB_SM                        100000       2942         205          0
1 row selected.
Elapsed: 00:00:00.01
16:42:47 SQL> exec dbms_stats.gather_table_stats(ownname => 'KDM', tabname => 'TEST_TAB_SM', cascade => true);
PL/SQL procedure successfully completed.
Elapsed: 00:00:00.26
16:42:55 SQL> select table_name,num_rows,blocks,avg_row_len,chain_cnt from user_tables where table_name = 'TEST_TAB_SM';
TABLE_NAME                       NUM_ROWS     BLOCKS AVG_ROW_LEN  CHAIN_CNT
TEST_TAB_SM                         50000       2942         205          0
1 row selected.
Elapsed: 00:00:00.01
16:43:27 SQL> alter table TEST_TAB_SM move;
Table altered.
Elapsed: 00:00:00.46
16:43:59 SQL>  select table_name,num_rows,blocks,avg_row_len,chain_cnt from user_tables where table_name = 'TEST_TAB_SM';
TABLE_NAME                       NUM_ROWS     BLOCKS AVG_ROW_LEN  CHAIN_CNT
TEST_TAB_SM                         50000       2942         205          0
1 row selected.
Elapsed: 00:00:00.03
16:44:06 SQL> exec dbms_stats.gather_table_stats(ownname => 'KDM', tabname => 'TEST_TAB_SM', cascade => true);
PL/SQL procedure successfully completed.
Elapsed: 00:00:00.24
16:44:17 SQL> select table_name,num_rows,blocks,avg_row_len,chain_cnt from user_tables where table_name = 'TEST_TAB_SM';
TABLE_NAME                       NUM_ROWS     BLOCKS AVG_ROW_LEN  CHAIN_CNT
TEST_TAB_SM                         50000       1471         205          0
1 row selected.
Elapsed: 00:00:00.01
16:44:24 SQL> We can see how the no. of blocks changes. It changed from 2942 to 1471 which is half and in your case it will be 1/4th times of what it was using.
There are other options for shrinking space as well other than MOVE.
alter table <table_name> shrink space;Try it out and you can see the difference yourself.