Subject: Re: i/o scheduling (was Re: NEW_BUFQ_STRATEGY)
To: YAMAMOTO Takashi <yamt@mwd.biglobe.ne.jp>
From: Thor Lancelot Simon <tls@rek.tjls.com>
List: tech-kern
Date: 12/17/2003 00:36:52
On Wed, Dec 17, 2003 at 02:19:59PM +0900, YAMAMOTO Takashi wrote:
> > On Wed, Dec 17, 2003 at 01:34:02PM +0900, YAMAMOTO Takashi wrote:
> > > 
> > > it isn't so easy.
> > > "delayed writes" can't be delayed so much because e.g. synchronous
> > > read requests can happen on the delayed-writed page and we have no way
> > > to look up corresponding buf from a page.
> > 
> > What keeps that from leading to data-integrity issues *right now*?
> 
> how do you think data-integrity issues can happen?

I must not understand what you're saying.  You say "synchronous read
requests can happen on the delayed-writed page and we have no way to
look up the corresponding buf from a page".  That suggests to me that
the read will return the wrong data; am I incorrect?  If not, what is
the problem, exactly, if synchronous read requests happen on pages
that have delayed writes in the I/O queues?

Is is it simply that the read will block indefinitely waiting for the
write to happen, with no way to cause the delayed-write queue to be
flushed?

Thor