Subject: Re: Journaling for FFS
To: Michael van Elst <mlelstv@serpens.de>
From: Bill Studenmund <wrstuden@netbsd.org>
List: tech-kern
Date: 10/03/2006 10:50:38
--5vNYLRcllDrimb99
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Tue, Oct 03, 2006 at 10:36:28AM +0200, Michael van Elst wrote:
> On Mon, Oct 02, 2006 at 05:18:45PM -0700, Jason Thorpe wrote:
> >=20
> > >There should be a way to stripe the journal over several disks.
> >=20
> > ...and that should be done for a multi-disk-aware file system, but not =
=20
> > for UFS.
>=20
> Then you have to repeat the problem of Solaris and UFS logging.
>=20
> The filesystem doesn't need to know what disks are. You can avoid
> the problem by storing the journal on a separate device (ugly, how?)
> or by spreading it over the filesystem device, e.g. close to
> the inodes it is logging.

I do not like the idea of multiple journals in the file system. The
problems I see are: 1) which journal should be replayed first on recovery? =
=20
2) You're assuming that journal-near-inodes means journal-near-where-we-
write. Yes, that's true for writing the inode and most-likely true for=20
early block allocations, but for a large file, it's not necessarily true.=
=20
For a REALLY large file, it isn't.

How about we work on _a_ journaling file system, then worry about=20
optimizing it.

Take care,

Bill

--5vNYLRcllDrimb99
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.3 (NetBSD)

iD8DBQFFIqLuWz+3JHUci9cRAmTOAJ4tpjYDiL/1oMG1rxwad496dRGptwCfRolf
pKKy/hGsNTXqwOFVZtuWNHw=
=otJu
-----END PGP SIGNATURE-----

--5vNYLRcllDrimb99--