Re: direct IO

From: Dion Cho <ukja.dion_at_gmail.com>
Date: Mon, 29 Nov 2010 10:36:12 +0900
Message-ID: <AANLkTikhYEtBtxhdU4bimE=B8L_-=d2_v4LnutojORpO_at_mail.gmail.com>

I believe that you would need to increase the buffer cache size to compensate OS file cache vanished by direct I/O.

---

Dion Cho - Oracle Performance Storyteller

http://dioncho.wordpress.com (english)
http://ukja.tistory.com (korean)
http://sites.google.com/site/otpack (tpack)
================================

2010/11/28 Mohammad Rafiq <rafiq9857_at_hotmail.com>

> You did not mention oracle version. Try to get it set aio-max-nr to
> 1048576 (default is 65536). If it is 10.2.0.5 then it must be 1048576 as a
> work around.
>
> Regards
> Rafiq
>
> > From: jcmiranda_at_oesia.com
>
> > To: przemolicc_at_poczta.fm; oracle-l_at_freelists.org
> > Date: Fri, 26 Nov 2010 12:58:22 +0100
> > Subject: RE: direct IO
>
> >
> >
> > We are getting
> >
> > WARNING:io_submit failed due to kernel limitations MAXAIO for process=128
> pending aio=110
> > WARNING:asynch I/O kernel limits is set at AIO-MAX-NR=3145728
> AIO-NR=77074
> > WARNING:Oracle process running out of OS kernel I/O resources (1)
> > WARNING:Oracle process running out of OS kernel I/O resources (1)
> > WARNING:Oracle process running out of OS kernel I/O resources (1)
> >
> > on several systems with linux pppc64 and it seems there are no patch for
> this error on this platform.
> > I can´t tell if direct i/o is slower.
> >
> > Anyone know how to solve this problem?
> >
> > ________________________________________
> > De: oracle-l-bounce_at_freelists.org [oracle-l-bounce_at_freelists.org] En
> nombre de przemolicc_at_poczta.fm [przemolicc_at_poczta.fm]
> > Enviado el: viernes, 26 de noviembre de 2010 10:21
> > Para: oracle-l_at_freelists.org
> > Asunto: Re: direct IO
> >
> > On Thu, Nov 25, 2010 at 05:32:02PM +0200, Laimutis.Nedzinskas_at_seb.ltwrote:
> > > Looking for experiences/advices of others regarding one issue with
> direct
> > > IO.
> > >
> > > Recently we switched to another data center and some databases reported
> > > increased IO time.
> > > After investigating we found that IO is slower on file systems
> (Veritas,
> > > Solaris 10) mounted with direct io options.
> > > AWR shows more time on scattered reads but for the same number of
> blocks
> > > too.
> > > We are going to switch of direct IO and test but for the time beeing I
> am
> > > wandering why direct IO is bad in our case.
> > >
> > > We've got seriously increased times for statistics gathering which does
> > > full scan naturally - we've found 2 full scan sqls on the same table
> > > clearly comming from dbms_stats package . Full scan is not cached by
> > > oracle. But file system cache can help in that particular case.
> >
> > Do you have exactly the same storage configuration in both data centers ?
> >
> > We found that _usually_ switching from buffered filesystem to direct IO
> won't
> > help unless you increase also Oracle buffers (so Oracle uses memory which
> > had been used previously for filesystem buffering).
> > Also on Veritas/UFS I recommend to turn on direct IO in Oracle
> (FILESYSTEMIO_OPTIONS=SETALL)
> > and not in Veritas (we don't set any mount options in /etc/vfstab for
> Veritas filesystems).
> >
> > If you bought Veritas ODM - just use it.
> >
> > Regards
> > Przemyslaw Bak (przemol)
> > --
> > http://przemol.blogspot.com/
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > ----------------------------------------------------------------------
> > Najlepsza wyszukiwarka tanich lotów!
> > Sprawdż >>> http://linkint.pl/f284a
> >
> > --
> > http://www.freelists.org/webpage/oracle-l--
> > http://www.freelists.org/webpage/oracle-l
> >
> >
>

--
http://www.freelists.org/webpage/oracle-l

Received on Sun Nov 28 2010 - 19:36:12 CST