Re: Solaris Direct I/O

From: <przemolicc_at_poczta.fm>
Date: Mon, 15 Feb 2010 10:10:03 +0100
Message-ID: <20100215091003.GB15497_at_host.pgf.com.pl>

On Tue, Feb 09, 2010 at 12:44:46AM -0600, Taral Desai wrote:
> Hi All,
>
> We are using sparc 64bit v 10 with oracle 10.2.0.3. Above that we use SAN
> storage.
>
> 1. How to find if storage is mounted for direct or async io

It depends on filesystem.
Example using ufs:
mount | grep d8
(cached filesystem) /d8 on /dev/md/dsk/d8 read/write/setuid/devices/intr/noforcedirectio/largefiles/logging/xattr/onerror=panic/dev=1540008 on Thu Aug 6 15:52:53 2009 (direct mount) /d8 on /dev/md/dsk/d8 read/write/setuid/devices/intr/forcedirectio/largefiles/logging/xattr/onerror=panic/dev=1540008 on Thu Aug 6 15:52:53 2009

Using vxfs:
mount | grep <some filesystem>
(direct mount) ...... mincache=direct/convosync=direct ....

> 2. We had set filesystem option to setall. How to verify it's doing direct
> or async. As with truss i see many kaio failure and then it invokes lwp
> process which do i/o in chunks. So, is there a different call for SAN.

Look for 'directiostat' (for ufs - I don't know if it works with vxfs)

> 3 Is there any benefit using veritas ODM over oracle ODM ?

It all depends on your workload.
In our case when we switched to Veritas ODM and did benchmarks (up to 7000 transactions - reported from statspack) we observed much smoother workload. Also without Veritas ODM there was significant drop in number of transactions during checkpoints. With Veritas ODM - very little drop.

Regards
Przemyslaw Bak (przemol)

--
http://przemol.blogspot.com/


----------------------------------------------------------------------
Walentynkowy konkurs dla singielek.
Sprawdz >>> http://link.interia.pl/f25cb

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

Received on Mon Feb 15 2010 - 03:10:03 CST