DragonFly users List (threaded) for 2007-02
Re: Why not FUSE ZFS?
On 2/20/07, Simon 'corecode' Schubert <email@example.com> wrote:
So the choice is:
1. port FUSE as a new filesystem
2. port ZFS as a new filesystem
notice that only the name differs? It is both work, so why not go for the real thing? FUSE does not have a particularly optimized interface, so performance won't be sexy.
Am I wrong that FUSE is both easier to (sup)port and going to bring a
lot more interoperability with other systems? This feels like Linux
emulation to me - it's not beautiful and it's not a real solution, but
its utility in the field is high enough to make or break an OS'
adoption. The code to support Linux emulation is greater than many
individual programs you'll run on it, but it scales because afterwards
you can run many such interesting programs, even ones you can't
possibly port because they're black boxes.
In fact, given there are going to be proprietary file systems out
there that use FUSE, and those will likely be Linux binaries, having a
high compatibility there will mean DragonFly can run Linux FUSE
binaries. I don't know if FreeBSD can do that yet, or if anybody has
even tried it, but it's certainly practical.
That leaves us (about) with: YOU (plural).
Great, so somebody can undertake the much lesser effort to support
FUSE, and ZFS comes for free. FUSE is coming sooner or later, one way
or another, because its utility is too high to ignore.
In fact a short-term approach might be to port puffs from NetBSD, and
then refuse can work on top of that. Later if SYSLINK is found to be a
better approach, puffs and FUSE can be replaced with wrappers for
If it was me doing the work (and someday it might be, I'm interested
too) I'd rather take the sub-optimal approach that favors
functionality and interoperability. It's nice that it also seems to
take less effort.
Just out of curiosity I'll investigate the work necessary to support
puffs within DragonFly's VFS sometime soon.
Centre for Synchrotron Science
Victoria 3800, Australia