Hello Meik,

Tuesday, December 16, 2008, 12:58:49 PM, you wrote:

MH> On Tue, 16 Dec 2008 00:12:55 -0500
MH> Bob Doolittle <[email protected]> wrote:


>> I can't argue there :)  Plus you get the awesome ZFS Time Slider.  Might 
>> be a while before you see that in opensuse...

MH> How network-transparent is the timeslider? Does it work with NFS-mounted 
homes
MH> (the NFS server using ZFS)? 
MH> Is a configuration with 100 users making individual snapshots of their
MH> homes really feasible?
Not that it's a Sun Ray topic, but...

I've checked on a test server (sol10u6), the snapshots
are visible to a NFS client (sol10u4) as a tree of dirs
under ZFSMOUNTDIR/.zfs/snapshots/SNAPSHOTNAME

You enable this by making snapshots visible and creating
a few:
[r...@server /]# zfs set snapdir=visible pool/export/home
[r...@server /]# zfs create snapshot pool/export/home/usern...@20081216-test

[r...@client /]# ls -la 
/net/server/export/home/username/.zfs/snapshots/20081216-test/
...

Snapshot views are "implemented" within the ZFS filesystem
dataset and don't incur an additional mountpoint, neither
in the original server's filesystem hierarchy nor in the
NFS client's one. They are added and removed as soon as
I create and destroy the ZFS snapshots on the server.

Note that I did not check the ZFS timeslider GUI, but only
the underlying technology - it's functional.

Since for the client the homedir is an NFS mount, I'm not
sure whether you can send a request from the client system
to the NFS server to make a ZFS snapshot on-demand. I guess
you can do that via ssh and password-less keys, but I wonder
if there's any RPC call via NFS directly. I didn't actually
try to research this, though.

I think (but I'm not in any way authoritative) that the
Windows server allows its "shadows" to be managed over
CIFS protocol (perhaps RPC inside).

Finally, one more thing to consider when you give out
ZFS homes and implement automatic snapshots is quotas.

Originally ZFS had attributes "quota" and "reservation"
which act on the ZFS dataset *including* its snapshots.
By making automatic snapshots you can overrun the user's
quota (i.e. if the snapshots reference files which were
deleted afterwards) while the user's active dataset is
not overflowing his quota.

Recent ZFS (actually zpool v9 as in builds of OpenSolaris
Nevada snv_77 and Solaris 10 Update 6 at least) include
"refquota" and "refreservation" properties which act on
*referenced* usage of the active dataset, not including
its clones and snapshots.

Probably you'd want a big "quota" on pool/export/home
and not inherit it to users' home datasets, and set a
smaller "refquota" on each pool/export/home/username
dataset. This way all of your homes together would still
consume no more than predefined space, but each user
would know that if he deletes a file, he does in fact
back off from his personal "ref"quota overflowing.

-- 
Best regards,
 Jim Klimov                            mailto:[email protected]

_______________________________________________
SunRay-Users mailing list
[email protected]
http://www.filibeto.org/mailman/listinfo/sunray-users

Reply via email to