Re: ** import tuning

  • From: Egor Starostin <egorst@xxxxxxxxx>
  • To: ajoshi977@xxxxxxxxx
  • Date: Tue, 7 Jun 2005 14:51:02 +0700

>  I have a export taken with direct=3Dy. This is done setting a high recor=
dlengh =3D 64K.
Good.
> When importing using this export sould I set recordlength or buffer?
You may set buffer.
> How high is Ok for buffer?
They say, it depends. There is a chance that you will see no time
difference between buffer=3D100m and buffer=3D1m.

> I am on 817 on sun 5.8. I need to do this operation frequently and need t=
o tune it.
Just don't forget about direct=3Dy and recordlength=3D64k. And try to
parallelize export. If you need to export user's data then export its
schema (rows=3Dn) and run several exports in parallel (one for each
table). I personally (as a python addicted user :) ) created small
python script for such task http://www.oracledba.ru/python/parexp.py).

>  Basically is export taken with direct option any different from regular =
export?
Yes. Direct export is faster. Direct export with recordlength=3D64k is
even faster.
Direct path export should be always faster (two-three times) than
conventional path export (no matter of buffer=3D settings).

As for settings to speedup import, I don't recommend you to regularly
use "_disable_logging"/"_wait_for_sync" options. Leave them only for
migration tasks. As an option you can increase priority of lgwr
process during import (renice -20 <lgwr's_pid>).

When I played with export/import speed issue I enabled system trigger like =
this:
***
create or replace trigger traceimpexp after logon on database
begin
  for tr in (
    select * from (
    select sid,serial# serial
    from v$session
    where username =3D user and (substr(program,1,4)=3D'imp@' or substr(pro=
gram,1,4)
=3D'exp@')
    order by logon_time desc
    ) where rownum < 2
  ) loop
    dbms_system.set_ev(tr.sid,tr.serial,10046,8,'');
  end loop;
end;
/
***
and analyzed trace files with my profiler (OraSRP).
For example, I played with redo logs/log buffers and when I saw that
redolog-related events consumed in sum less than 3-5% of all the time,
I stopped to tune them.

Reading of Notes 93763.1 and 155477.1 also might be helpful.

--=20
Egor
http://www.oracledba.ru/orasrp/
Free Oracle Session Resource Profiler
--
//www.freelists.org/webpage/oracle-l

Other related posts: