On 4/01/2013 8:30 PM, tonyalbers wrote:
> Ok, it seems to me that there is some confusion of which kind of dedupe I'm
> referring to.
>
> What I'm talking about is variable-length block-based target dedupe, that is,
> the DXi or DataDomain box receives all data from the client via the backup
> server (bacula SD) and then handles the deduplication and compression of
> these data. There is no dedupe or compression on the client or backup server
> itself of any sort.
Just to be sure, the targets are file volumes on LUNs or the like? The
storage is not behaving like some sort pseudo-tape device?
And compression is not enabled anywhere in any Bacula configuration file
and the entire Bacula process chain has been restarted?
Spooling is turned off as well?
> This does not work in practice right now, since bacula does something to the
> backup stream that makes most of the blocks or segments unique.
Bacula was initially designed to be as efficient as possible when
writing to *tape*, it chunks up data before it is written, anything
could cause a shift that reduces similarity.
> I've tried this on a DXi and on a fileserver using SDFS, and both had only 5%
> data reduction after running three 6GB backups of the same filesystem on a
> client. In theory that should give us at least 66% data reduction ratio.
I am surprised that it is so low, I would have expected more.
Oh, what happens when you just "tar" the filesystem to the storage
several times?
> Please bear in mind that target dedupe is pretty much the industry standard,
> since RTO plays a huge role in deciding what type of backup and storage to
> use.
There are *no* standards, just whatever the vendor is peddling.
Cheers,
Gary B-)
------------------------------------------------------------------------------
Master HTML5, CSS3, ASP.NET, MVC, AJAX, Knockout.js, Web API and
much more. Get web development skills now with LearnDevNow -
350+ hours of step-by-step video tutorials by Microsoft MVPs and experts.
SALE $99.99 this month only -- learn more at:
http://p.sf.net/sfu/learnmore_122812
_______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users
|