Optimizing RSync algorithm using techniques with Google used in courgette

Shachar Shemesh shachar at shemesh.biz
Tue Sep 8 06:27:03 MDT 2009


Hasanat Kazmi wrote:
> Hi,
> I am student at LUMS SSE (http://cs.lums.edu.pk) and an active RSync user.
> Just few days ago, Google wrote about Courgette*: an algorithm which 
> is specially written for syncing executables. By using Courgette, 
> Google made diff size 1/10th of previous techniques used.
> I was wondering if this (or something on same lines) can be used to 
> optimize RSync? I am senior and have to do a project. I am thinking to 
> implement this in RSync. I need input from developers. What do you 
> guys think?
>
> *http://dev.chromium.org/developers/design-documents/software-updates-courgette
>
> Hasanat Kazmi
>
Hi Hasanat,

Like you said in the subject, this is an optimization. A format specific 
optimization. In other words, it uses a known property of the file being 
synchronized in order to make the diff size smaller. If you were to try 
to use the courgette pre-processing on something which is not an 
executable, you would have gotten significantly worse results than 
merely running rsync.

At the moment, for better or for worse, rsync does not do format 
specific optimizations. As long as that is the case, rsync cannot be 
optimized using this algorithm.

Even if we (and by "we", I mean Wayne, or anyone else brabe enough to 
pick this task up) were to implement such a functionality, I can think 
of quite a few things that would have a lot more to gain than 
executables. In particular, something that would uncompress both source 
and destination, and apply the rsync algorithm to both files, and then 
make sure the recompression of the target produces the exact same result 
would, IMHO, be much more useful than the change you are suggesting.

Shachar

-- 
Shachar Shemesh
Lingnu Open Source Consulting Ltd.
http://www.lingnu.com

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.samba.org/pipermail/rsync/attachments/20090908/296a5007/attachment.html>


More information about the rsync mailing list