--L+ofChggJdETEG3Y
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Tue, Dec 04, 2007 at 08:40:28PM +0000, Michael van Elst wrote:
> tron@zhadum.org.uk (Matthias Scheler) writes:
>=20
> >> Now guess what happens when you temporarily bypass softdep's ordering.
> >Bad things(TM)
>=20
> Why would you want to stop disk writes? The kernel can easily _read_
> data (when it has free buffers) even when it pushes out dirty
> buffers in a tight loop. The problem is that you don't have
> free buffers because all are occupied by data that needs to
> be written. Stopping the writes won't free any buffer.

The problem is that our experience does not agree with your assertion=20
about the kernel being able to read while writing. I think you're correct=
=20
that the kernel can easily issue reads while it is writing in a tight=20
loop. The problem is getting the reads actually serviced by the drive.

The reason to stop writing is to let other access happen. Also, part of=20
the idea is to avoid the avalanche by starting to write sooner and pacing=
=20
ourselves.

Take care,

Bill

--L+ofChggJdETEG3Y
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (NetBSD)

iD8DBQFHVcSUWz+3JHUci9cRApL3AJ9j1lnU59MkQj2384FpyUNWd4o3IQCeMeix
+V9a/bj9FayliU4ncXc52d0=
=E0fx
-----END PGP SIGNATURE-----

--L+ofChggJdETEG3Y--