Subject: Re: Journaling for FFS
To: Michael van Elst <email@example.com>
From: Bill Studenmund <firstname.lastname@example.org>
Date: 10/03/2006 16:13:53
Content-Type: text/plain; charset=us-ascii
On Tue, Oct 03, 2006 at 08:05:18PM +0200, Michael van Elst wrote:
> On Tue, Oct 03, 2006 at 10:50:38AM -0700, Bill Studenmund wrote:
> > I do not like the idea of multiple journals in the file system.
> I don't see this as multiple journals, it is just one spread over
> the disk. The problem I am talking about is that a single small
> journal area becomes the bottleneck for the whole filesystem.
The problem is that you spoke of the journal being near the inodes it is=20
logging. That means that where we write in the journal depends on what=20
inode we're writing.
The thing is that's not how a journal works. At least not the ones I'm=20
familiar with. To do this, we'd really need separate journals. Each one=20
would individually have traditional semantics, we just use each one for a=
subset of the inodes on the disk. That's separate journals.
However I think we're rushing ahead of ourselves. Exactly how the journal=
is layed out is not the hard part of this work; the hard part is all of=20
the transaction operations in the file system. I think we'll be best=20
served to actually get a working journal, then work from there. Yes, we=20
should look at performance issues others have seen, but we also need=20
something to tune before we can tune it. :-)
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.3 (NetBSD)
-----END PGP SIGNATURE-----