thank you, sir! Your opinion is very important for me!
I have a couple more finds about deterministic functions and ssc and I will post them asap.

Thanks for the link. Interesting difference in the caching effects.

thank you for your explaining comment. In Cost Based Oracle (chapter 9) I saw that in 10g the size of the hash table was controlled by _query_execution_cache_max_size – and obviously that is still true in 11.2.0.1. With values above the default (65536) the number of Buffers drops to the value I expected:

alter session set "_query_execution_cache_max_size"=262144;

-----------------------------------------------------------------------------------------------------------------
| Id  | Operation           | Name | Starts | E-Rows | A-Rows |   A-Time   | Buffers |  OMem |  1Mem | Used-Mem |
-----------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT    |      |      1 |        |      0 |00:00:00.09 |   29820 |       |       |          |
|   1 |  MINUS              |      |      1 |        |      0 |00:00:00.09 |   29820 |       |       |          |
|   2 |   SORT UNIQUE       |      |      1 |   2500 |    250 |00:00:00.01 |      77 | 13312 | 13312 |12288  (0)|
|   3 |    TABLE ACCESS FULL| T1   |      1 |   2500 |   2500 |00:00:00.01 |      77 |       |       |          |
|   4 |   SORT UNIQUE       |      |      1 |   2500 |    250 |00:00:00.08 |   29743 | 13312 | 13312 |12288  (0)|
|   5 |    TABLE ACCESS FULL| T2   |      1 |   2500 |   2500 |00:00:00.01 |      77 |       |       |          |
-----------------------------------------------------------------------------------------------------------------

alter session set "_query_execution_cache_max_size"=2097152;

-----------------------------------------------------------------------------------------------------------------
| Id  | Operation           | Name | Starts | E-Rows | A-Rows |   A-Time   | Buffers |  OMem |  1Mem | Used-Mem |
-----------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT    |      |      1 |        |      0 |00:00:00.06 |   19425 |       |       |          |
|   1 |  MINUS              |      |      1 |        |      0 |00:00:00.06 |   19425 |       |       |          |
|   2 |   SORT UNIQUE       |      |      1 |   2500 |    250 |00:00:00.01 |      77 | 13312 | 13312 |12288  (0)|
|   3 |    TABLE ACCESS FULL| T1   |      1 |   2500 |   2500 |00:00:00.01 |      77 |       |       |          |
|   4 |   SORT UNIQUE       |      |      1 |   2500 |    250 |00:00:00.06 |   19348 | 13312 | 13312 |12288  (0)|
|   5 |    TABLE ACCESS FULL| T2   |      1 |   2500 |   2500 |00:00:00.01 |      77 |       |       |          |
-----------------------------------------------------------------------------------------------------------------

With the difference of 98 blocks (19348 – 19250) I can live …

Martin

perhaps it’s not too closely related to the topic of the article – but playing a little bit with your example I got a surprising “Buffers” value. My modifications of your example were the use of MOD to limit the number of different IDs and a DETERMINISTIC keyword added to the function definition:

create table t1
as
select
    mod(rownum, 250)          id,
    lpad(rownum,200)    padding
from    all_objects
where   rownum <= 2500
;
 
create table t2
as
select  * from t1
;
 
exec dbms_stats.gather_table_stats(user, 't1')
exec dbms_stats.gather_table_stats(user, 't2')
 
create or replace function f (i_target in number)
return number deterministic
as
    m_target    number;
begin
    select max(id) into m_target from t1 where id <= i_target;
    return m_target;
end;
/

select  /*+ gather_plan_statistics */
    id
from    t1
minus
select
    f(id)
from    t2
;
 
-----------------------------------------------------------------------------------------------------------------
| Id  | Operation           | Name | Starts | E-Rows | A-Rows |   A-Time   | Buffers |  OMem |  1Mem | Used-Mem |
-----------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT    |      |      1 |        |      0 |00:00:00.16 |   54747 |       |       |          |
|   1 |  MINUS              |      |      1 |        |      0 |00:00:00.16 |   54747 |       |       |          |
|   2 |   SORT UNIQUE       |      |      1 |   2500 |    250 |00:00:00.01 |      77 | 13312 | 13312 |12288  (0)|
|   3 |    TABLE ACCESS FULL| T1   |      1 |   2500 |   2500 |00:00:00.01 |      77 |       |       |          |
|   4 |   SORT UNIQUE       |      |      1 |   2500 |    250 |00:00:00.16 |   54670 | 13312 | 13312 |12288  (0)|
|   5 |    TABLE ACCESS FULL| T2   |      1 |   2500 |   2500 |00:00:00.01 |      77 |       |       |          |
-----------------------------------------------------------------------------------------------------------------

Column Projection Information (identified by operation id):
-----------------------------------------------------------

   1 - STRDEF[22]
   2 - (#keys=1) "ID"[NUMBER,22]
   3 - "ID"[NUMBER,22]
   4 - (#keys=1) "F"("ID")[22]
   5 - "ID"[NUMBER,22]

I expected the Buffers for step 4 to be 77 * 250 = 19250: with one FTS for each distinct value in t2. But I got 54670 (= 77 * 710) (and I got the same results with clustered data when using TRUNC instead of MOD to create the test tables). Can you explain the number of buffers?

Martin