GTT helps performance - one day to be several minute

Question

I have a a case like this:
for rec (select *
from gl.interface
where accounting_date between...
order by c1, c2)
loop
   -- concate some columns of rec
   -- do update gl_interface
end loop;it takes about a day to generate notepad report from this query (the records is about 6.000.000).
And after along day, i resolve it and it takes only several minutes to generate the report.
What i am doing is these steps:
1. create global temporary table that delete records after commit. And create some indeks, like rn that is row_number() over (order by c1, c2)
2. ITAS the gtt with select * , row_number() over (order by c1, c2)
from gl.interface
where accounting_date between
3. change the cursor query above in my oracle form:
for rec in(
   select *
from gtt
where accounting_date between...
order by rn
loop
end loop;
update gl_interface
set...
where exists(
  select ..
from gtt...
where col1 =...
)I still wonder how it works? and what is the drawbacks of using this gtt? tx.
Edited by: infotools on Jan 7, 2013 3:35 AM
Edited by: infotools on Jan 7, 2013 3:35 AM
Edited by: infotools on Jan 7, 2013 3:36 AM

Answer 1

infotools wrote:
No, i don't want to use bulk for i don't want to create arrays. And for my case, i am satisfied with the results. But, i still wonder how it decrease the time so much using gtt...and can be ITAS...and may be the
select a||b||c
from
instead of
for ()
loop
s = cc.a || cc.b||cc.c....
end loop;
is it because i put much of the pl-sql statements in my select-sql?Likely because you have reduced context switching between the SQL and PL/SQL engines. By selecting rows in a loop and updating or inserting for each row selected, you are constantly switching back and forth between PL/SQL and SQL, and that is known to be poor for performance. Using a GTT will help improve things because you are reducing the amount of context switches, but then if the data is already coming from database tables, you shouldn't need to use a GTT in the first place... as just doing it directly from the database tables, using something like MERGE will do it all in the SQL engine anyway.