Thanks for the feedback, I did subscribe to the zfs-discuss list.
amount for backup is 600MB on Server A and 900MB on Server B.
Both servers have tape stackers.
speeds up data restoration. I have found that I can restore any
2hours. I would expect with the 5TB disk subsystem this may increase to
site. The cost of implementing a high speed data connection 150 miles
would be very expensive.
Post by Al Hopper Post by John Martinez
No, unfortunately. Of course you can always do a tar - assuming that
you are OK with the zfs extended ACLs not being saved. The only ZFS
mechanism available right now are "zfs send" and "zfs receive".
Interesting topic. I'll bite.
Using tar for large amounts of data is unrealistic. Many of us in
"enterprise" situations already have a backup product in-house. These
solutions are run by another group of people who have invested
millions into not only software, but also hardware solutions such as
tape libraries as well as disk-based backup solutions.
Post by Russell Aspinwall
Data restore tends to be two different types :-
1) Selective file or directory restore
2) File system restore due to hardware failure
( 3 disk failed in a 13 disk RAID5 set + HS)
I like the ufsrestore due to the interactive mode allowing for complete
or selective restores.
a) re-arrange your data so that, instead of using sub directories to
I like this solution, which is why I am looking into migrating to
ZFS. But I hear Russell's request above. It is essential to have the
functionality to restore sections of file systems. I would have to
restore a user's entire file system just because he mistakingly blew
away one little directory in his file system.
Another point: Ask yourself, with 5TB of data, do you really want to
mess with tapes - or build another ZFS box as the backup/restore
facility. Don't forget that you can turn on compression for zfs
Why yes. We have strict requirements for not only migrating data to
tape for off-site storage, we also move data around to other
geographic locations via the backup software. We are well into the
hundreds of terabytes.
Understood - but my point is that for smaller datasets, in Russells'
case, 5 terabytes, ZFS, in combination with cost effective, high
capacity SATA disk drives, may very well have already obsoleted tape
based backups. Obviously, there is some user-scenario-specific
storage capacity (translation number-of-terabytes) where tape will be
the only economically viable backup storage medium currently
The other point I did'nt make, is that, in many situations,
time-to-restore is much more important than the cost of the backup
subsystem. I know many people on this list have wasted years (OK
exaggeration) of their lives, waiting for a tape to "find" the
required dataset, before it even *starts* to stream off the required
Everyone goes down at some time or another - for various reasons.
And downtime, while becoming increasingly untolerated by a typical
user community, can be explained and is understood and (somewhat)
accepted. But there is far less user community tolerance when it
takes, for example, a business day to restore a service. The user
community expects the service provider to restore a service in a
matter of "N" hours - where N is often (in todays environment) less
than 2. Down the road, it is easy to imaging that N will continue to
Post by John Martinez
We also have situations where we do off-host backups for database
applications. We freeze the database on the production server, take a
snapshot on the SAN storage array, unfreeze the database, mount the
file systems on a different host, create a database checkpoint and
back it up. This results in about <1 minute interruption for our
users, where it could take hours otherwise. I have ideas on how ZFS
would work in this scenario, and I don't think it would be different
from how our current solution works.
Moving to ZFS requires that you re-think how you use storage. But
it's well worth the effort IMHO.
Indeed it does. I think it still has a bit to mature for "enterprise"
applications, but it's getting there.
100% agreed. On zfs-discuss you can see the long running thread where
users have been asking for the ability to shrink a pool - for various
reasons. In one scenario, it's because they need to reallocate
physical disk drives due to changing business requirements, in other
scenarios, it's because rocket-scientist-turned-manager changes
his/her mind, after making a really bad decision, and needs to
reallocate physical disk resources. Or the unfortunate sysAdmin makes
a couple of "typos".... :)
There were also a few posts in the early days of zfs-discuss where
people thought that ZFS should never have been released before a
comprehensive and fully featured backup/restore feature set was
available. Personally, I'm glad those people did'nt get to stall the
release of ZFS! :)
I think its very healthy to *always* list the current deficits of a
product or technology - rather than sending the message that
technology "A" is the greatest thing since sliced bread - and then
many of the early adopters discover, to their frustration, the
limitations already known by the developers/proponents of it.
The really good news, is that team ZFS knows all this. And the
underlying technology has already been designed in such a way that the
currently missing functionality can be added with minimal disruption
to the zfs user community. So, your point that ZFS is still shy of a
fully featured enterprise storage subsystem is accurate - but there is
so much functionality and features currently available that it would
be shameful to not deploy it today, where it fits, and simply work
around the currently missing features/facilities.
In Russells scenario, he could build a backup server, with, say 4
terrabytes of storage for a reasonable cost. Store his snapshots
there, and be able to "render" a zfs filesystem to the same backup
server storage pool very, very quickly (at 400 to 500Mb/Sec) and
restore a specific user subdirectory structure in minutes. Way before
a tape based system would be able to even stream to the required
 can anyone say docs.sun.com (or 24 hours)
Post by John Martinez
Feel free to email me offline if I can help.
Thanks for your insight, Al. Always helpful.
Does anybody know if ZFS works with the following list?
IBM Tivoli backup
Yep - please subscribe to zfs-discuss. I've been on this list from
Voice: 972.379.2133 Fax: 972.379.2134 Timezone: US CDT
OpenSolaris Governing Board (OGB) Member - Apr 2005 to Mar 2007
This email has been scanned by the MessageLabs Email Security System.
For more information please visit http://www.messagelabs.com/email