Re: Oracle memory allocation on Linux 2.6

From: <vitalisman_at_gmail.com>
Date: Wed, 16 Apr 2008 03:01:52 -0700 (PDT)
Message-ID: <0dc8ad33-92a9-4d17-bc45-429cc9f23c91@8g2000hsu.googlegroups.com>


On Apr 15, 6:58 pm, NetComrade <netcomradeNS..._at_bookexchange.net> wrote:
> On Tue, 15 Apr 2008 06:11:35 -0700 (PDT), vitalis..._at_gmail.com wrote:
> >On Apr 14, 9:39 pm, NetComrade <netcomradeNS..._at_bookexchange.net>
> >wrote:
> >> On Fri, 4 Apr 2008 23:24:54 -0700 (PDT), vitalis..._at_gmail.com wrote:
> >> >On 2 avr, 18:16, NetComrade <netcomradeNS..._at_bookexchange.net> wrote:
>
> >> >> >Do you have FILSYSTEMIO_OPTIONS set to either "SETALL" or "DIRECTIO"?
> >> >> >In my experience, that will significantly reduce the use of caches.
>
> >> >> Have you tried playing with Veritas/ODM?
>
> >> >I've never used ODM. But is it compatible with OCFS2?
>
> >> Sorry, no clue.. my understanding with OCFS, an oracle supplied FS,
> >> double buffering shouldn't be an issue... but i never used it... was
> >> 'fortunate' enough to work with ADM
>
> >Could you please explain why this "double buffering" is wanted? And
> >have you found a means of inhibiting it altogether (using Direct IO on
> >Linux/OCFS2 does not do it according to my tests and a few messages
> >found on the net)?
>
> If my reply read as 'double buffering' is a good thing, my apologies.
> With ODM, I am pretty sure it's not an issue... I've verified it with
> trussing/stracing in the past and proper system calls.. I haven't
> done anything beyond that (no clue into 'peaking' into buffers)
>
> I am not familiar with OCFS2, but why would Oracle build a file system
> that sucks similarly to others? I think w/o peaking into buffers, it's
> hard to tell if buffered data is Oracle's. There will also be local
> file systems.. Linux does not like to 'waste' memory.
> If you really think it's an issue in your case, go raw or ODM.. I'd be
> curious to see if you have links showing that directio is a 'scam'
> .......
> We run Oracle 9iR2,10gR2, 10g2RAC on RH4/RH5 and Solaris 10 (Sparc)
> We use RMAN and remote catalog for backups

According to the (few) tests I've done, datafile blocks on OCFS2 are still buffered by Linux with DIO. It seems to be the "expected" behaviour as evidenced by messages by other Net comrades ;-) such as http://tinyurl.com/6rehxw Received on Wed Apr 16 2008 - 05:01:52 CDT

Original text of this message