[plug] Semi OT - OS X "Network Discovery" and SLP

[W.Kenworthy]

On Thu, 2006-11-09 at 10:11 +0800, Timothy White wrote:
> On 11/9/06, Paul Arch <paul at esidium.com.au> wrote:
> > Timothy White wrote:
> > > Ok, I have a little problem. I'm building a "NAS" box (Network Area
> > > Storage) that is designed to automate backups on a network. The target
> > > network is obviously small home networks, unix based of course :p
> > >

Have a look at "dirvish" rather than pure rsync for server initiated
pull backups (http://www.dirvish.org).  Advantages are that its space
and network efficient as well as secure (via ssh - it asks for a
password or you can use keys).  Instead of transferring just changes,
dirvish is a true backup system with multiple copies with programmable
retention times.

Its space efficient because if a file has not changed, it creates a
symlink (hardlink?) to the file, so it only uses inodes.  Changed files
are transferred as a whole (or rsynced).  A really smart thing about
this method is that when the last reference to a file is deleted, the
file goes as well.  Size (for me) has stablised at about 2.5 times the
backup size (using a daily 2 week retention, and a sunday 2 month
retention)  I have not tried, but rsync over ssh may be used as well
according to the FAQ.  I would expect more normal (i.e, those who are
not gentoo serial updaters!!!) users to do better.  I use a section of
an LVM used for mythtv (reiserfs, same directory as the myth recordings)
without problems.  See an occasional glitch in remote live TV, but
recordings are fine, even when multiple recordings and backups are going
on at the same time.

Apparently doesnt saturate a network (I suspect that this statement is a
bit of a stretch) so parallel backups should be possible (if the file
system can take it!)

The developer recommends against a journaled file system for speed, but
for me reiserfs works fine (full laptop backups > 40G in under an hour).
You do need an fs without restrictions (or with a massive possible
number) on inode numbers.

Downside? Absolutely hammers the backup disks at times - tried it once
on an ext2 loopback with the predictable results.  No mention of doze,
but with the right tools this may be possible.  Large files (i.e., 8G
vmware container files) impact performance, as will huge numbers of
changes on a disk (creates spikes in disk usage and blowouts in time
taken) - not unexpected.

BillK