Re: OCFS2, blockdev and readahead

From: Jeremy Schneider <jeremy.schneider_at_ardentperf.com>
Date: Sun, 08 Feb 2009 15:05:39 -0600
Message-ID: <498F4923.8010603_at_ardentperf.com>



Vladimir Barac wrote:
>
> Oracle 10.2.0.2, OCFS2, single instance database, NetApp FCP disks
>
> filesystemio_options=setall
>
> db_file_multiblock_read_count=128
>
> Do we need to set readahead to greater than 0 for OCFS2? What would be
> ideal value? How to measure impact positive/negative?
>
> Storage admins would prefered "blockdev -setra" to be set to maximum
> value, to speed up filesystem backups. On the other hand, since oracle
> will use O_DIRECT on OCFS2, I believe we don't need readahead to be set.
>

It's (as always) a little application dependent, although my experience has been that very few databases benefit from read-ahead. (Just some warehouses/datamarts.) NetApp has published best practices and they have a decent paragraph about this - as of May 2008 they generally recommend not using readahead.
http://www.netapp.com/us/library/technical-reports/tr-3369.html

These read ahead recommendations would apply the same on OCFS2 or ASM or NFS or a regular filesystem. Even though read ahead doesn't usually kick in right away anyway, few databases perform many large sequential reads and it would probably start doing read ahead right when Oracle stops reading data.

Furthermore, you certainly shouldn't tune your storage system to perform fast backups and do everything else a little more slowly. :) And if someday you move toward incremental or even updated image copy backups, then the readahead won't benefit at all for backups.

-Jeremy

-- 
Jeremy Schneider
Chicago, IL
http://www.ardentperf.com

--
http://www.freelists.org/webpage/oracle-l
Received on Sun Feb 08 2009 - 15:05:39 CST

Original text of this message