Subject: Re: what's it take to get a journaling filesystem?
To: rudolf <netbsd@eq.cz>
From: Thor Lancelot Simon <tls@rek.tjls.com>
List: tech-kern
Date: 12/14/2005 11:37:25
On Wed, Dec 14, 2005 at 05:29:17PM +0100, rudolf wrote:
> 
> Incidentally today I read some documentation for FreeBSD's GEOM 
> framework and it seems to me that besides the filesystem as such, 
> everything what Sun is calling 'ZFS' could be done within this BSD 
> framework.

I think that's very much wrong.  Much of the design of ZFS, it seems
to me, is a well-founded reaction *against* strictly eparating logical
volume management from the filesystem implementation.  Doing so discards
many opportunities to improve performance and reliability, as I think even
the introductory slides on ZFS that you can find on Sun's web site
pretty conclusively demonstrate.

We've speculated before about the possibility of enhancing our
filesystems to allow them to allocate blocks from multiple underlying
devices.  For both FFS and LFS, there would seem to be some
clear advantages to this model over the one in which a volume manager,
which knows nothing about the logical structure of the data it stores,
aggregates devices in a very simplistic way and hands homogenized "disk"
objects up to the filesystem layer.  The difference is that Sun didn't
just speculate about it -- they actually did something.

Thor