On 18 August 2010 14:59, William Kenworthy <billk@iinet.net.au> wrote:
On Wed, 2010-08-18 at 14:09 +0300, Nganon wrote:
>
>
> On 17 August 2010 22:34, Enrico Weigelt <weigelt@metux.de> wrote:
>         For things I'd like to keep an history (eg. /etc) I'm using
>         git, and
>         pushing the repo to a remote server (denying non-fastfoward
>         updates
>         there, so an theorectical highjacker cannot destroy my
>         history)
>
>
> Using git for /etc is a great idea.
> Thanks.
>
Another option is:
*  app-backup/dirvish
     Latest version available: 1.2.1
     Latest version installed: 1.2.1
     Size of downloaded files: 47 kB
     Homepage:    http://www.dirvish.org/
     Description: Dirvish is a fast, disk based, rotating network
backup system.
     License:     OSL-2.0


Works by first creating a copy (--init) and then hard-linking subsequent
versions of files/directories back to the original original if its
identical.  If a file is changed/new, it is copied instead of linked so
actual space usage quickly stabilises even with a varying number of
versions.  Backup over the network (this is how I have configured mine)
uses rsync over ssh with keys and is "pull" from a cron job on the
backup server or manual on demand (i.e., server initiated).

Version management is by a reasonably sophisticated date of version
scheme where by running "dirvish-expire" deletes out of date versions
(runs in a cron job).  The smart part is that once the last hard link to
file is deleted, its gone, otherwise its kept in the remaining
versions :)

Restore is a simple matter of identifying the version you want and
copying it back - Ive restored individual files through to complete
systems after total disk failure.

Can do includes/excludes, whole systems or just directories such as /etc
and can be easily automated.

Doesnt use compression, but most backup regimes (every day for a weekly
rota + a Sunday kept for 6 months) stabilise at about 2x the original
(gross) copy size, no matter how many copies with average changes
between versions.  Though large scale changes such as an "emerge -e
world" will take more as it will generate new copies of most files.

Downside is it will hammer the destination file system - reiserfs3 works
well, ext2/ext3 have been hopeless everytime I've tried - mass
corruption.  The file system will need a large number of inodes (for
links) if there are an excessive number of files x versions - again
reiserfs3 scores well here.

Highly recommended!

BillK




Thanks. It sound just it is made just for this. It even call itself 'time machine'. 
Obviously compression is left out by using links but it sounds kind of 
overwhelming to me. I don't have a reiserfs partition and cannot afford to 
have one at the mo..