[ale] Ext4 adoption anyone?

[Michael B. Trausch]

On Wed, 21 Jan 2009 18:49:49 -0500
Pat Regan <thehead at patshead.com> wrote:

> I only read the Wikipedia page [for Tux3].  I'm going to be lazy in
> the interest of keeping the discussion rolling.
> 
> If I read correctly about the versioning taking place at the inode
> level, does that mean I can snapshot a subdirectory without
> snapshotting the whole volume/subvolume?  If you can, that is a
> killer feature in my opinion and something I would REALLY like to
> have.  I imagine it also means you'd have to snapshot the inode of
> each individual file/directory, too, so it might be a slower
> snapshotting process.
> 
> I might be able to live with that for the extra flexibility.

The impression that I get from reading the high-level overview is that
it will be a true versioning filesystem where you can take any file
back in time, at least, that's what I think when I read about the trees
being used to organize the blocks of the file the way it does.  I'd
hope that it means that you can treat your filesystem as a sort of
primitive, yet global, version control system.  I expect that it would
need some special-purpose utilities to do that, too, but learning them
would be useful.

I say primitive because I would expect that it would actually delete
file when you delete them and not retain that sort of history without
an active snapshot.  There have certainly been times where I could have
benefited from having a snapshot around, but they're usually times when
I would have had to have the foresight to realize that anyway, and it
wouldn't have saved my bacon any better than usual.  That's kind of the
reason that I rsnapshot every couple of hours.  :)

> > Supposedly the FUSE module for ZFS is fast and reliable, have you
> > toyed with that?
> 
> Almost all my playing around has been with zfs-fuse.  It did work
> quite well, but I didn't feel I wanted to use fuse for something as
> important as my home directory.

I've used FUSE on and off for various things for a while; the only
plugin that I have found to be utterly unreliable (though more because
of the typical configuration of a remote machine) is the sshfs plugin.
It's great if you control both machines and don't get disconnected for
inactivity, but when you do, it's a pain.

That said, I am actually kind of surprised that more filesystem
development isn't happening that way.  I think it'd be interesting to
actually move to a model where most filesystem code resides outside of
the kernel, but only because I think that is probably better since FUSE
plugins can run on any platform that supports FUSE.  I like the idea of
being able to use a single filesystem "driver" on multiple UNIX-like
systems, ensuring compatibility between them.  Today, it's still a pain
to read media from FreeBSD on Linux, and vice versa---FreeBSD supports
ext2 only, last I checked, and Linux still makes you do some manual
tweaking to mount UFS media from any system that uses a variant of that
filesystem format.  *shrugs*

> fuse+zfs+raid-z was much faster on a stack of 4 usb 2.0 hard drives
> than md+raid5+ext3.  I think it was the journaling that was horking
> me up for raid 5, there.
> 
> Which got me thinking about something that is somewhat off topic that
> I've not yet gotten around to trying.  I wonder how much improvement
> in write performance you'd get on an ext3+raid5 if you use an external
> journal on a raid1 or raid10 device.

As I don't really play with RAID that much, I don't know.  The only
thing I could see using RAID for is for something like RAID5, or the
RAID-Z that ZFS has, and I just don't have that many discs in any
machine except my server --- and even there, I have four drives totally
devoted to space, using LVM, housing a single partition.

That actually reminds me that I need to get a BD burner so that I can
have an easier time of doing a whole-filesystem-data backup.  DVD is
just not big enough...

> > RCS?  Really?  Sounds kinda painful to me...  ;-)
> 
> In general, if I create and edit something it ends up in a Darcs
> repository.  If I didn't create it, I probably don't care about it too
> much.  In that case, it ends up on in a backup from time to time.
> 
> My home directory will definitely end up being whichever stable in
> kernel filesystem ends up giving me snapshotting at least as quick and
> easy as zfs.

Ahh.  I thought you meant RCS, the classic version control system.  I
use Bazaar (bzr) myself; I used Subversion for a long time before that,
but I don't think I could look back even for a very large purse of
money.  I can't imagine _not_ using bzr for VCS tasks any more.

I use git, but only for pulling things which already use it.  I like it
as well, though I like some of the ways that bzr does things better
since the branch model is much more distributed, and you don't *have*
to work with all of the branches in a given tree when pulling it.  My
understanding of git and other DVCS tools is that they are repositories
housing a collection of branches, whereas with bzr you just get the
branch.  You can have multiple branches in the same directory and pool
storage between them if you'd like, but it's not required.  Not sure
how Darcs works at its lower levels, but reading on WP leads me to
think it is probably closer to git's model than it is bzr's.

	--- Mike

-- 
My sigfile ran away and is on hiatus.
http://www.trausch.us/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 197 bytes
Desc: not available
Url : http://mail.ale.org/pipermail/ale/attachments/20090122/61be4c26/attachment.bin