Re: Mutliplexing control/redo files

From: <przemolicc_at_poczta.fm>
Date: Fri, 8 Oct 2010 09:10:22 +0200
Message-ID: <20101008071022.GA4434_at_host.pgf.com.pl>



On Thu, Oct 07, 2010 at 10:28:08PM -0400, Christo Kutrovsky wrote:
> To all using ZFS and Oracle and saying ZFS is memory hungry.
>
> ZFS is just as memory hungry as any other filesystem.

I disagree. If you use UFS/VxFS with direct mount option it doesn't use a lot of memory. In ZFS you don't have direct option.

To be more precise: by default ZFS is much more memory hungry then UFS/VxFS.

> It uses all available
> memory for caching. Exception to that is advanced features such as dedup,
> which can be 'memory hungry' in the true sense. And anyone using Oracle
> should NOT be relying on ZFS caching and giving all possible memory to
> Oracle.
>
> As per file and directory layout, why not use OMF (Oracle Managed Files).
> There is no need for one to be bothered with this information.
>
> In terms of physical design, the whole zfs concept is to forget about
> mountpoints, and instead fully rely on nested filesystems. Oracle does not
> need ZFS. As closer to raw as possible is ideal for Oracle (with ASM).
> Oracle also has most useful features ZFS offers such as snapshots (flashback
> database) and cloning (standby/dataguard/incremental backups).
>
> If you have to use ZFS, remember the following very important performance
> configurations
> - configure REDO and backup (including archivelogs) with default ZFS block
> size (128k)
> - configure DATA and controlfiles to match database block size - this is
> critical
> - Reduce the maximum amount of memory the ARC (zfs file cache) can use. Give
> as much as possible to Oracl. Don't cache in filesystem

This advice is common for all filesystems: give memory as much as possible to Oracle despite the fact what kind of filesystem you use. The problem with ZFS is that in order to limit its memory cache you have to change settings in /etc/system _and_ reboot the server (in UFS/VxFS you can remount filesystem online). When you have many applications working on your server it becomes serious limitation.

> There are other tuning aspects, but these are the main ones.
>
> And one thing to remember - RAIDZ is not RAID5 - you get the IO capacity of
> a single spindle, not the combined.
>
> --
> Christo Kutrovsky
> Senior Consultant
> Pythian.com
> I blog at http://www.pythian.com/blogs/
>
> On Thu, Oct 7, 2010 at 1:04 PM, girlgeek <girlgeek_at_live.com> wrote:
>
> > their use of 'large' and 'small' with no clue of what large or small
> > actually means.
> > -Claudia
> >
> > On 10/7/2010 9:48 AM, Rodd Holman wrote:
> >
> >> What does it mean "large db" ?
> >>
> > --
> > http://www.freelists.org/webpage/oracle-l
> > X-ecartis-version: Ecartis v1.0.0
> > Sender: oracle-l-bounce_at_freelists.org
> > Errors-to: oracle-l-bounce_at_freelists.org
> > X-original-sender: girlgeek_at_live.com
> > Precedence: normal
> > Reply-To: girlgeek_at_live.com
> >
> > List-help: <mailto:ecartis_at_freelists.org?Subject=help>
> > List-unsubscribe: <oracle-l-request_at_freelists.org?Subject=unsubscribe>
> > List-software: Ecartis version 1.0.0
> > List-Id: oracle-l <oracle-l.freelists.org>
> > X-List-ID: oracle-l <oracle-l.freelists.org>
> > List-subscribe: <oracle-l-request_at_freelists.org?Subject=subscribe>
> > List-owner: <mailto:steve.adams_at_ixora.com.au>
> > List-post: <mailto:oracle-l_at_freelists.org>
> > List-archive: <http://www.freelists.org/archives/oracle-l>
> > X-list: oracle-l
> >
> >
> >
>
>
> --
> Christo Kutrovsky
> Senior Consultant
> Pythian.com
> I blog at http://www.pythian.com/blogs/

-- 
Regards
Przemyslaw Bak (przemol)
--
http://przemol.blogspot.com/





























----------------------------------------------------------------
Zobacz ile kosztuje m2 w Twoim miescie. Szukaj w bazie ogloszen!
http://linkint.pl/f281f

--
http://www.freelists.org/webpage/oracle-l
Received on Fri Oct 08 2010 - 02:10:22 CDT

Original text of this message