Re: UFS2 support?

[Christopher Weimann <csw@xxxxxxxxx>]

On 02/03/2005-04:19PM, Oliver Fromme wrote:
> 
> Note that dd(8) probably works slower on a snapshot than on
> a normal device, because the read access is not completely
> sequential.  

Thanks.  I'm thinking it still must be better than what I am
getting out of dump.

> It is also worth noting that dump(8) now has a
> -C option in FreeBSD 5, which tells dump that the filesystem
> is read-only (which is the case with snapshots) and that it
> is OK to cache blocks freely.  That improves performance of
> dump considerably.
> 

Hmm... 4.10 seems to have -C as well.

     -C cachesize
             Specify the cache size in megabytes.  This will greatly improve
             performance at the cost of dump possibly not noticing changes in
             the filesystem between passes.  Beware that dump forks, and the
             actual memory use may be larger then the specified cache size.
             The recommended cache size is between 8 and 32 (megabytes).

I am not using this right now. Looks like the file system doesn't have
to be RO to use it either. I'll try this tonight. This still doesn't
solve the bigger problem of the time for restore though. A full backup
taking 12 hours is bad but the restore taking 2-3 days is the real
killer.

> By the way, snapshots don't require UFS2.
> 
> From reading Kirk McKusick's paper on soft-updates and snap-
> shots, I get the impression that implementing snapshots in
> FFS is pretty straight-forward and simple -- much simpler
> than the implementation of soft-updates.  Thus I think there
> is hope that someone might bring it to DragonFly some day,
> too.  :-)
> 

That would be nice.  There is a lot about snapshots that is very
cool.  NetApp filers have a snapshot facility and that does one
every hour so you can get back those files you deleted by accident.

In the long run Matt is right, this isn't really the right answer and
jounrnaling would be better.  It just doesn't exist yet.