SQL Performance issue: Using user defined function with group by

Question

Hi Everyone,
im new here and I really could need some help on a weird performance issue. I hope this is the right topic for SQL performance issues.
Well ok, i create a function for converting a date from timezone GMT to a specified timzeone.
CREATE OR REPLACE FUNCTION I3S_REP_1.fnc_user_rep_date_to_local (date_in IN date, tz_name_in IN VARCHAR2) RETURN date
IS
tz_name VARCHAR2(100);
date_out date;
BEGIN
SELECT
to_date(to_char(cast(from_tz(cast( date_in AS TIMESTAMP),'GMT')AT
TIME ZONE (tz_name_in) AS DATE),'dd-mm-yyyy hh24:mi:ss'),'dd-mm-yyyy hh24:mi:ss')
INTO date_out
FROM dual;
RETURN date_out;
END fnc_user_rep_date_to_local;The following statement is just an example, the real statement is much more complex. So I select some date values from a table and aggregate a little.
select
stp_end_stamp,
count(*) noi
from step
where
stp_end_stamp
BETWEEN
to_date('23-05-2009 00:00:00','dd-mm-yyyy hh24:mi:ss')      
AND
to_date('23-07-2009 00:00:00','dd-mm-yyyy hh24:mi:ss')
group by
stp_end_stampThis statement selects ~70000 rows and needs ~ 70ms
If i use the function it selects the same number of rows ;-) and takes ~ 4 sec ...
select
fnc_user_rep_date_to_local(stp_end_stamp,'Europe/Berlin'),
count(*) noi
from step
where
stp_end_stamp
BETWEEN
to_date('23-05-2009 00:00:00','dd-mm-yyyy hh24:mi:ss')      
AND
to_date('23-07-2009 00:00:00','dd-mm-yyyy hh24:mi:ss')
group by
fnc_user_rep_date_to_local(stp_end_stamp,'Europe/Berlin')I understand that the DB has to execute the function for each row.
But if I execute the following statement, it takes only ~90ms ...
select
fnc_user_rep_date_to_gmt(stp_end_stamp,'Europe/Berlin','ny21654'),
noi
from
select
stp_end_stamp,
count(*) noi
from step
where
stp_end_stamp
BETWEEN
to_date('23-05-2009 00:00:00','dd-mm-yyyy hh24:mi:ss')      
AND
to_date('23-07-2009 00:00:00','dd-mm-yyyy hh24:mi:ss')
group by
stp_end_stamp
)The execution plan for all three statements is EXACTLY the same!!!
Usually i would say, that I use the third statement and the world is in order. BUT I'm working on a BI project with a tool called Business Objects and it generates SQL, so my hands are bound and I can't make this tool to generate the SQL as a subselect.
My questions are:
Why is the second statement sooo much slower than the third?
and
Howcan I force the optimizer to do whatever he is doing to make the third statement so fast?
I would really appreciate some help on this really weird issue.
Thanks in advance,
Andi

Answer 1

Hi,
The execution plan for all three statements is EXACTLY the same!!!Not exactly. Plans are the same - true. They uses slightly different approach to call function. See:
drop table t cascade constraints purge;
create table t as select mod(rownum,10) id, cast('x' as char(500)) pad from dual connect by level <= 10000;
exec dbms_stats.gather_table_stats(user, 't');
create or replace function test_fnc(p_int number) return number is
begin
    return trunc(p_int);
end;
explain plan for select id from t group by id;
select * from table(dbms_xplan.display(null,null,'advanced'));
explain plan for select test_fnc(id) from t group by test_fnc(id);
select * from table(dbms_xplan.display(null,null,'advanced'));
explain plan for select test_fnc(id) from (select id from t group by id);
select * from table(dbms_xplan.display(null,null,'advanced'));Output:
PLAN_TABLE_OUTPUT
Plan hash value: 47235625
| Id  | Operation          | Name | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT   |      |    10 |    30 |   162   (3)| 00:00:02 |
|   1 |  HASH GROUP BY     |      |    10 |    30 |   162   (3)| 00:00:02 |
|   2 |   TABLE ACCESS FULL| T    | 10000 | 30000 |   159   (1)| 00:00:02 |
Query Block Name / Object Alias (identified by operation id):
   1 - SEL$1
   2 - SEL$1 / T@SEL$1
Outline Data
  /*+
      BEGIN_OUTLINE_DATA
      FULL(@"SEL$1" "T"@"SEL$1")
      OUTLINE_LEAF(@"SEL$1")
      ALL_ROWS
      OPTIMIZER_FEATURES_ENABLE('10.2.0.4')
      IGNORE_OPTIM_EMBEDDED_HINTS
      END_OUTLINE_DATA
Column Projection Information (identified by operation id):
   1 - (#keys=1) "ID"[NUMBER,22]
   2 - "ID"[NUMBER,22]
34 rows selected.
SQL>
Explained.
SQL>
PLAN_TABLE_OUTPUT
Plan hash value: 47235625
| Id  | Operation          | Name | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT   |      |    10 |    30 |   162   (3)| 00:00:02 |
|   1 |  HASH GROUP BY     |      |    10 |    30 |   162   (3)| 00:00:02 |
|   2 |   TABLE ACCESS FULL| T    | 10000 | 30000 |   159   (1)| 00:00:02 |
Query Block Name / Object Alias (identified by operation id):
   1 - SEL$1
   2 - SEL$1 / T@SEL$1
Outline Data
  /*+
      BEGIN_OUTLINE_DATA
      FULL(@"SEL$1" "T"@"SEL$1")
      OUTLINE_LEAF(@"SEL$1")
      ALL_ROWS
      OPTIMIZER_FEATURES_ENABLE('10.2.0.4')
      IGNORE_OPTIM_EMBEDDED_HINTS
      END_OUTLINE_DATA
Column Projection Information (identified by operation id):
   1 - (#keys=1) "TEST_FNC"("ID")[22]
   2 - "ID"[NUMBER,22]
34 rows selected.
SQL>
Explained.
SQL> select * from table(dbms_xplan.display(null,null,'advanced'));
PLAN_TABLE_OUTPUT
Plan hash value: 47235625
| Id  | Operation          | Name | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT   |      |    10 |    30 |   162   (3)| 00:00:02 |
|   1 |  HASH GROUP BY     |      |    10 |    30 |   162   (3)| 00:00:02 |
|   2 |   TABLE ACCESS FULL| T    | 10000 | 30000 |   159   (1)| 00:00:02 |
Query Block Name / Object Alias (identified by operation id):
   1 - SEL$F5BB74E1
   2 - SEL$F5BB74E1 / T@SEL$2
Outline Data
  /*+
      BEGIN_OUTLINE_DATA
      FULL(@"SEL$F5BB74E1" "T"@"SEL$2")
      OUTLINE(@"SEL$2")
      OUTLINE(@"SEL$1")
      MERGE(@"SEL$2")
      OUTLINE_LEAF(@"SEL$F5BB74E1")
      ALL_ROWS
      OPTIMIZER_FEATURES_ENABLE('10.2.0.4')
      IGNORE_OPTIM_EMBEDDED_HINTS
      END_OUTLINE_DATA
Column Projection Information (identified by operation id):
   1 - (#keys=1) "ID"[NUMBER,22]
   2 - "ID"[NUMBER,22]
37 rows selected.