BURP - BackUp and Restore Program

ABOUT
WHY
FEATURES
REQUESTS
CHANGELOG
PROTOCOL 1
PROTOCOL 2
NEWS
FAQ
DOCS
BURP-UI
DOWNLOAD
LICENCE
CONTRIBUTORS
DONATIONS
SPONSORS
CONTACT

Shuffling, hardlinked_archive, librsync
---------------------------------------

The following was originally posted on the mailing list, but possibly helpful
to people with more questions about the shuffling pase, librsync=[0|1], and
hardlinked_archive=[0|1].

You can set both those options globally in the server burp.conf, or for
individual clients in the server clientconfdir files.

On Thu, May 10, 2012 at 07:29:09PM +0200, Davide Bozzelli wrote:
> Just a simple question: what is the purpose of the shuffle operation at the
> end of backup ?

On Thu, May 10, 2012 at 9:54 PM, Graham Keeling wrote:
> Once all the files and bits of files have been transferred to the server,
> it needs to assemble them into the final backup.
> By default, this includes hardlinking files that haven't changed since the
> immediately previous backup into place, patching the ones that have changed
> with the new deltas, generating reverse deltas to be able to recreate the
> previous files that change, and deleting the files that changed from the
> previous backup.

On Thu, May 10, 2012 at 07:29:09PM +0200, Davide Bozzelli wrote:
> I see that a lot of time is taken by it (even much more that the time taken
> for transfer file).
>
> Is there a way to speedup this process ?

On Thu, May 10, 2012 at 9:54 PM, Graham Keeling wrote:
> Yes, you can stop it from generating the reverse deltas and deleting the
> original files by setting hardlinked_archive=1, which means that a complete
> copy of every version of each file will be kept. This also has the benefit
> of speeding up restores for older backups, because no reverse deltas need
> to be applied, but it has the disadvantage of taking up more disk space.

On Tue, May 15, 2012 at 09:20:50PM +0200, Davide Bozzelli wrote:
> Does it means that on every backup run ALL the files need to be trasferred
> ? or onlu the changed ones ?

On Tue, May 15, 2012 at 10:35:10PM +0100, Graham Keeling wrote:
> Only the changed ones.



On Wed, May 16, 2012 at 11:12:50AM +1200, Jason Haar wrote:
> Just to clarify this for myself and others, can you confirm burp works
> like this
>
> client -> I have one file that is different -> server
> server -> server copies old version (from last backup) into temp dir
> client -> sends whole file using rsync tricks [compression, deltas] ->
> server pointing at temp dir
>
> server then compares this temp file against last backup version, creates
> delta, deletes temp file, creates hardlink to old version and documents
> that file == last backup + delta. This is the "shuffling" phase. Does
> that sum it up?

No.
In the source, there is a file called 'backup_phases.txt' that explains it:
------
backup_phase1_client: Scan the client filesystem and send stats to the server.
backup_phase1_server: Receive the stats from the client.

backup_phase2_server: Request and receive changes from the client and create
        an unchanged list and a changed list.
backup_phase2_client: Send the changes that the server requests. The work of
        the client is now finished.

backup_phase3_server: Generate the new manifest list for the backup out of
        the unchanged list and the changed list.

backup_phase4_server: Finish off the backup by jiggling the received data
        around and putting everything in the correct place. Need to
        generate reverse deltas in order to save space for the previous backup
        (unless hardlinked_archive is turned on).
------
(phase4 is the shuffling bit)

> One last question about "hardlinked_archive=1". This means for each
> backup, if the file hasn't changed since the last backup, burp hardlinks
> the old file to the new?

Yes, and it doesn't generate reverse deltas for the files that changed and it
doesn't delete the original files.

> If "hardlinked_archive=0", then burp... does what?

It generates reverse deltas for the files that changed and it deletes the
original files.

> The documentation says that "hardlinked_archive=1" uses more
> diskspace - but I can't understand how. Surely "hardlinked_archive=0"
> would mean burp is keeping a zero-sized delta plus metadata referring to
> the original backup - that must take up about as much diskspace as a
> hardlink?

No, for at least three reasons:

a) If a file didn't change and 'hardlinked_archive = 0', burp will just delete
the original and keep the new version, and not keep a delta at all (zero-sized
deltas actually have a few bytes in them).

b) A complete file plus a reverse delta takes up less space than two complete
files. Imagine you have a 1MB file, back it up, then add a byte to it, then
back it up again. With 'hardlinked_archive = 1', the server stores both
complete versions of the files. With 'hardlinked_archive = 0', the server
stores the most recent version, plus a reverse delta (which contains a small
block that has the effect of removing a byte from the recent version)
so that it can re-generate the older version if it has to.

c) Your filesystem doesn't have unlimited inodes. A hardlink uses an inode.


You can also set librsync=0, which turns off librsync so that, if a file has
changed, the whole file is transferred from the client. This means that
deltas don't have to be applied to the previously backed up files. So, if the
speed of your network beats the speed of being able to apply deltas, you might
want to set librsync=0.

See the man page for more information on these options.


Donate with Bitcoin

Burp is open and free software. I work on it in my spare time. If you would like this work to continue, please consider making a small donation.


Burp, don't suck. Last updated: March 2017
By Graham Keeling
Hosted by 6sync