Using GTT

Question

Hi,
In my procedure i am using normal tables with insert and delete dml operations. My code sample is as below:
create or replace procedure proc_name(parameter1....param6) as
variables declaration;
ran_num= random();/* random number generator function*/
cursor c1 as select * from t1 where rep_id=ran_num;
cursor c2 as select * from t2 where rep_id=ran_num;
cursor c3 as select * from t3 where rep_id=ran_num;
insert into t1 values(x,y,z,ran_num);
insert /*+ append parallel(t2,10) */into t2 (select /*+ append parallel(t1,10) */x,y,z,rep_id from t1 where rep_id=ran_num);
insert /*+ append parallel(t3,10) */into t3 (select /*+ append parallel(t2,10) */x,y,z,rep_id from t2 where rep_id=ran_num);
finally using cursors i am inserting data into final tem table
insert into t4 values(x,y,z,rep_id);
output ref_cursor;/* which throws output to report*/
end;
which is taking 3 hrs to run bcz t1 from which data is inserted is of huge volumes of data.
my senior suggested to use gtt tables instead of normal temp tables of which automatically data is deleted at the end of session and is independent for each session.
is it worth ful using gtt?
As i am not aware of the gtt concept and after reading from goole about that i am also having some doubts.
can any one clarify along with suggestion of using gtt.
1. Can i completely avoid using random number generation which is used to identify each session independentness after using GTT.
2. Can i usie them in place of t2 and t1 in select queries simply like select x,y,z from t1;
3. Some where read that gtt cannot be used in select statements that contain parallel hints.

Answer 1

Hi,
Statistics calculated at the table level are associated with the table even with a temp table. For theses tables (temp) : the statistics are shared between the different sessions so it means that a session may use inaccurate statistics.
If you work with temp tables sometime you need to hint your DML to avoid potenital issues (use /*+ cardinality(yourtemptable 10000) */ or dynamically specify the cardinality of your table (this is only an example, there's plenty of others hint that can be used)
If you call DBMS_STAT on a temp table with 2 different sessions, the last call will override the previous stats (unless locked).
I hope that one day Oracle will provide a mean to maintain statistics on temp at session level.
Note about parallelization:
It's posssible to use the parallelization over temp tables:
CREATE GLOBAL TEMPORARY TABLE TEMP_TEST
  COL1  NUMBER(8)
ON COMMIT PRESERVE ROWS
NOCACHE
EXPLAIN PLAN  FOR SELECT /*+ PARALLEL(R) */ * FROM TEMP_TEST R
SELECT * FROM  TABLE(DBMS_XPLAN.DISPLAY);
PLAN_TABLE_OUTPUT
Plan hash value: 10615466
| Id  | Operation            | Name      | Rows  | Bytes | Cost (%CPU)| Time     |    TQ  |IN-OUT| PQ Distrib |
|   0 | SELECT STATEMENT     |           |     1 |    13 |     2   (0)| 00:00:01 |        |      |            |
|   1 |  PX COORDINATOR      |           |       |       |            |          |        |      |            |
|   2 |   PX SEND QC (RANDOM)| :TQ10000  |     1 |    13 |     2   (0)| 00:00:01 |  Q1,00 | P->S | QC (RAND)  |
|   3 |    PX BLOCK ITERATOR |           |     1 |    13 |     2   (0)| 00:00:01 |  Q1,00 | PCWC |            |
|   4 |     TABLE ACCESS FULL| TEMP_TEST |     1 |    13 |     2   (0)| 00:00:01 |  Q1,00 | PCWP |            |
Note
   - dynamic sampling used for this statementEdited by: user11268895 on Jul 22, 2010 2:37 PM
Edited by: user11268895 on Jul 22, 2010 2:39 PM