Re: High db file sequential reads during imp?

From: Joerg Jost <joerg.jost_at_unitrade.com>
Date: Fri, 09 May 2008 11:30:49 +0200
Message-Id: <1210325449.7471.19.camel@localhost>

Am Freitag, den 09.05.2008, 08:22 +0100 schrieb cam:
> Hello all,
>
> Relative newcomer to performance analysis. I was surprised, upon
> investigating a very slow import job, to find very high 'db file
> sequential read's. This activity was mostly associated with an index
> that was being updated while around 11 million rows were inserted to a
> normal heap table. I'm clearly missing something obvious here but why
> would imp be reading (and not writing?) so heavily from this index?
>
> Regards,
> cam

Hi,

i think, you give the answer with the question, if i am right. Imp will insert a lot of rows. Depending on the configuration of the imp (commit=?) it will upgrade the indexes associated with the table you're inserting.

To do this, imp has to read the table. This are the db file sequential reads you have discovered i think.

You should try to disable commit with the above mentioned configuration parameter. If the import is still slow, try indexes=n and afterwards another imp - job with indexfile=/bla/blub.sql

Then you will get a file with the statement(s) to build the indexes after inserting every row into the table. This can be considerably faster.

hth

Jörg

--
http://www.freelists.org/webpage/oracle-l
Received on Fri May 09 2008 - 04:30:49 CDT

Original text of this message