Subject: vnd vs. direct I/O (was Re: CVS commit: src)
To: Thor Lancelot Simon <tls@netbsd.org>
From: Chuck Silvers <chuq@chuq.com>
List: tech-kern
Date: 10/06/2006 09:19:27
On Thu, Oct 05, 2006 at 06:06:26PM +0000, Thor Lancelot Simon wrote:
> On Thu, Oct 05, 2006 at 02:48:33PM +0000, Chuck Silvers wrote:
> > 
> > Log Message:
> > add support for O_DIRECT (I/O directly to application memory,
> > bypassing any kernel caching for file data).
> 
> Like Rui said, thanks!
> 
> How hard would it be to convert vnd to use direct I/O?  It seems like
> that would fix some annoying double-caching problems (and a lot more
> people actually use vnd on production systems now because it is a
> convenient backing store for Xen virtuals).

it probably wouldn't be very hard, but the caching behaviour would be
the same as vnd is today (ie. metadata is cached, data is not).
it would still be a good change to make, since it would remove knowledge
of vnode locking from the vnd driver, but it wouldn't solve all the
problems with vnd.

-Chuck


> If it's just mechanical changes, let me know what they are, and I will
> have a shot at it if I can find the time this weekend.
> 
> Thor