clumsyoik Wrote: 
> Have *you*? 
> 
> I have never used rdiff-backup, but I have used xdelta.
> 
> Here is a simple non-scientific benchmark, although still fairly
> representative of the kind of diffs that would be generated in this
> application. 
> 
> This is using xdelta, which I believe to be pretty good at generating
> binary diffs.
No need to get huffy.

Have I tried binary diffs?  Indeed I have.  I've done some pretty
extensive evaluation of such, having been a consultant to a company for
which I provided version control software for hardware designers (who
can have VERY large binary files).

Your test case is a trivial one, and does not show the problems
associated with larger changes in a file, but is probably
representative of what will occur with music files.  Since only
metadata is being changed at the beginning of the file, binary diffs
are fine.  However, running such an expensive set of operations to
simply obtain and backup the tags seems wasteful.

> Clearly binary diffs can be very efficient indeed.Sure, but rarely.  As the 
> change set increases, the time required to
generate deltas goes up almost expoentially.

> This took about 5 seconds, would probably be 20 if they weren't cached.
> So, yes, it would be a long time for a large library, but it would be
> no worse than rsync. (In practice, you could selectively ignore files
> based on the last-modified time and filesize)
Agreed, you could be more clever about selecting which files to backup.
But your method would always be worse than rsync if you tell rsync not
to bother doing a diff, and to just copy.  As Jim earlier pointed out,
this is not constantly changing binary data in general - only the tags
change, so all you need is the tag information and one set of music
files.


-- 
MrC
------------------------------------------------------------------------
MrC's Profile: http://forums.slimdevices.com/member.php?userid=468
View this thread: http://forums.slimdevices.com/showthread.php?t=18029

_______________________________________________
Discuss mailing list
Discuss@lists.slimdevices.com
http://lists.slimdevices.com/lists/listinfo/discuss

Reply via email to