Subject: Re: Extension of fsync_range() to permit forcing disk cache flushing
To: Jason Thorpe <>
From: Daniel Carosone <>
List: tech-kern
Date: 12/19/2004 11:14:27
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Sat, Dec 18, 2004 at 11:38:10AM -0800, Jason Thorpe wrote:
> On Dec 17, 2004, at 2:06 PM, Thor Lancelot Simon wrote:
> >With newer SATA disks, *if we supported tagged command queueing*, which
> >we don't, and only with controllers that actually supported it, we=20
> >could do the right thing here.
> Don't assume that SATA NCQ is a panacea.  It is still not as good as=20
> SCSI tagged queueing (where you have mode pages that allow you to tune=20
> various aspects of how the drive's queue is managed).

Indeed, and with only 16 tagged commands, I'm skeptical that its a
deep enough queue to allow the drive to usefully optimise ordering of
writes to platters.  I'm sure my model will allow much better
concurrency of committed writes than NCQ; its an open question whether
our filesystem will generate enough of that concurrency in the BUFQ
for any of this to matter. I'll hopefully soon have an answer to that
from my testing.

Content-Type: application/pgp-signature
Content-Disposition: inline

Version: GnuPG v1.2.6 (NetBSD)