Use or not to use table compression in Oracle 11g (11.2)?

Question

Hi All,
I was trying to explore the difference between COMPRESS FOR ALL OPERATIONS, COMPRESS FOR DIRECT_LOAD OPERATIONS and NOCOMPRESS, for a table in Oracle 11.2.
I know, we can go thru documentation and make a decision.
Still I have run some very simple tests here.
Case 1. Create table with COMPRESS FOR DIRECT_LOAD OPERATIONS and then update few records
Case 2. Create table with COMPRESS FOR ALL OPERATIONS and then update few records
Case 3. Create table with NOCOMPRESS and update few rows
I know, Case 1 is a real dummy, but still I did that to see difference between Case1 and Case2.
--  ---------- CASE 1 --------
SQL> create table aaa
  2  nologging
  3  compress for direct_load operations
  4  as
  5  select * from all_objects ;
Table created.
Elapsed: 00:00:02.00
SQL> select count(*) from aaa ;
  COUNT(*)
     50317
Elapsed: 00:00:00.11
SQL> update aaa set created=sysdate where owner='SYS' and object_type='VIEW';
3485 rows updated.
*Elapsed: 00:00:05.43*
SQL> commit;
Commit complete.
Elapsed: 00:00:00.04
SQL>
--  ---------- CASE 2 --------
SQL>
SQL> create table bbb
  2  nologging
  3  compress for all operations
  4  as
  5  select * from all_objects ;
Table created.
Elapsed: 00:00:02.01
SQL> select count(*) from bbb ;
  COUNT(*)
     50318
Elapsed: 00:00:00.20
SQL> update bbb set created=sysdate  where owner='SYS' and object_type='VIEW';
3485 rows updated.
*Elapsed: 00:00:05.31*
SQL> commit;
Commit complete.
Elapsed: 00:00:00.04
SQL>
SQL>
--  ---------- CASE 3 --------
SQL> create table ccc
  2  nologging
  3  nocompress
  4  as
  5  select * from all_objects ;
Table created.
Elapsed: 00:00:01.84
SQL> select count(*) from ccc ;
  COUNT(*)
     50319
Elapsed: 00:00:00.15
SQL> update ccc set created=sysdate  where owner='SYS' and object_type='VIEW';
3485 rows updated.
*Elapsed: 00:00:00.06*Case1 and Case2 took 5.43 and 5.31 seconds respectively. Case 3 took 0.06 seconds.
Difference is drastic.
Am I doing wrong kind of test (lets be honest)?
Should we not use compression for OLTP systems (or any systems with reasonable updates)?
Apart from allowing to drop a column, what is the difference between COMPRESS FOR ALL OPERATIONS and COMPRESS FOR DIRECT_LOAD OPERATIONS ? where/how can I see that difference?
Thoughts please.
Thanks in advance.

Answer 1

Hi,
I have realised that I am using the syntax which is deprecated in 11.2.
So I am doing the same test with
COMPRESS BASIC
COMPRESS FOR OLTP
instead of
COMPRESS FOR DIRECT_LOAD OPERATIONS (deprecated)
COMPRESS FOR ALL OPERATIONS (deprecated)
But the results are same. Even if I do COMPRESS FOR OLTP, my update is taking 5.4 seconds which is not very different from COMPRESS BASIC
-- --------- CASE 1 ---------------
SQL> create table aaa
  2  nologging
  3  compress basic
  4  as
  5  select * from all_objects ;
Table created.
Elapsed: 00:00:02.46
SQL>
SQL> select count(*) from aaa ;
  COUNT(*)
     50318
Elapsed: 00:00:00.11
SQL>
SQL> update aaa set created=sysdate where owner='SYS' and object_type='VIEW';
3485 rows updated.
Elapsed: 00:00:05.48
-- ---------- CASE 2 ---------------
SQL> create table bbb
  2  nologging
  3  compress for oltp
  4  as
  5  select * from all_objects ;
Table created.
Elapsed: 00:00:02.01
SQL>
SQL> select count(*) from bbb ;
  COUNT(*)
     50319
Elapsed: 00:00:00.12
SQL>
SQL> update bbb set created=sysdate  where owner='SYS' and object_type='VIEW';
3485 rows updated.
Elapsed: 00:00:05.25
-- ---------- CASE 3 ---------------
SQL> create table ccc
  2  nologging
  3  nocompress
  4  as
  5  select * from all_objects ;
Table created.
Elapsed: 00:00:01.81
SQL>
SQL> select count(*) from ccc ;
  COUNT(*)
     50320
Elapsed: 00:00:00.10
SQL>
SQL> update ccc set created=sysdate  where owner='SYS' and object_type='VIEW';
3485 rows updated.
Elapsed: 00:00:00.04Any thoughts??