xfs
[Top] [All Lists]

Re: Curious about mount options and Direct IO.

To: Austin Gonyou <austin@xxxxxxxxxxxxxxx>
Subject: Re: Curious about mount options and Direct IO.
From: Chris Wedgwood <cw@xxxxxxxx>
Date: Thu, 24 Oct 2002 13:37:24 -0700
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <1035490218.1472.11.camel@ubergeek>
References: <20021024200521.GA17994@xxxxxxxxxxxxx> <1035490218.1472.11.camel@ubergeek>
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mutt/1.4i
On Thu, Oct 24, 2002 at 03:10:18PM -0500, Austin Gonyou wrote:

> I agree, but I guess my question was a general question about XFS's
> capabilities to actually do anything in a DIO mode.I didn't see any
> mount options for DirectIO, just osyncisdsync.

XFS under Linux supports O_DIRECT.  For some applications, it's a bug
win, for others, it's not.

> Our app is Oracle, but I can't see that anything will really benefit
> beyond relieving the logging(logbufs=8).

O_DIRECT means

         reduced memcpy when doing a read

         reduced memcpy when doing a write

         *much* lower vm pressure when doing lots of IO
         (eg. a table-scan of 2GB of data won't evict all your cached
         pages on 1GB box)

I use O_DIRECT mostly because of the last issue... when writing
multi-GB files, I don't want my desktop to start swapping or evicting
useful data from cache.  All there help with performance, but to what
degress depends on what you are doing and how well implemented it is.
I would think Oracle already uses O_DIRECT and is pretty well tuned
for it, but I don't know for sure, you'll have to ask.



  --cw


<Prev in Thread] Current Thread [Next in Thread>