Re: 10g sqlldr and PARALLELism

  • From: Tanel Põder <tanel.poder.003@xxxxxxx>
  • To: <oracle-l@xxxxxxxxxxxxx>
  • Date: Thu, 20 Jan 2005 01:09:11 +0200


Given your server, I assume that your IO is fast enough for these parallel
loads as well?

If you have logging on and lousy IO for redologs, then this might be a
reason why parallel jobs get even slower. Although such drastic performance
drop probably isn't due logging bottleneck.

So I guess the logical next step is to run your parallel sqlldr sessions
again and check from v$session_event (and v$sesstat) where most of the time
is going. If it doesn't ring a bell, then you could also run the normal
sqlldr job and check whether there's a difference in proportions of time
spent. Also you could use sql trace with waits for getting this information,
but I think in this particular case identifying the session id's and using
v$session_event would be easier.


> I am running Oracle 10g on an IBM P690 with AIX.  For a particular job I
> load 8GB using sqlldr, direct=yes in two hours.  I wanted to speed this
> up, so I broke the job up into four jobs, each loading 2GB each with the
> parallel=true option turned on.  These four jobs now take six hours
> each.   Why is there a time increase?
> I can't find a parameter that will solve the apparent contention that is
> going on.
> Any ideas where I may be going wrong?
> Thanks ... Roger


Other related posts:

  • » Re: 10g sqlldr and PARALLELism