[BBLISA] Recommendation for NAS appliance?

[John Stoffel]

>> And backups are sucky unless you do NDMP directly to tape.  But then
>> restores suck suck suck because NDMP is just an encapsulation of the
>> native dump/restore tools, which means scanning the tapes bit by bit
>> to find and restore the file(s) you want to restore.  So in that case,
>> snapshots make a huge difference.  Except when they expire and the
>> user comes the next day with a restore request.

Edward> This was a huge win for us, in favor of ZFS.

No major arguements.  

Edward> Previously we had the NetApp with scsi attached tape library,
Edward> controlled via NDMP.  A total approx 1Tb storage used.  The
Edward> backup window was approx 10 hrs per night, and 30 hours for
Edward> full backup on the weekend.  The scsi bus would sometimes mess
Edward> up and cause the system to crash or reboot, approx 2-3 times a
Edward> year, which should of course be zero.  Very annoying.

Sounds like you had both crappy tape drives, and just poor performance
over the SCSI bus.  From another email, I see you had a StoreVault
thingy, which is NOT enterprise class Netapp hardware.  

Edward> When we got the sun fileserver, we also bought some cheap Dell
Edward> and installed solaris on it too.  Hook up the tape autoloader
Edward> to the Dell.  ZFS send to the Dell, and from there backup to
Edward> tape.  Now our backup window (zfs send) is about 7 minutes per
Edward> night, and about 10 hrs to do the full.  After that, I don't
Edward> know and don't care how long it takes to write the tapes,
Edward> because whatever the load is and however long it lasts, it's
Edward> not happening on my production system.  Even if the scsi bus
Edward> crashes, no big deal because it's just the secondary backup
Edward> server, not the production server.

Nice setup.  How many snapshots can you store on the Dell or the Sun
and how often have you had to restore from Tape?  

Edward> I can't emphasize this enough, it was a huge win to send snaps
Edward> server-to-server-to-tape.

Sure, but you now need to have double (if not more!) storage, and
twice as many systems to manage.  So you went from a single Netapp
box, to a pair of boxes running Solaris with ZFS.  Of course you got
better performance!

Edward> Yes you could do the same thing with NetApp, but you would
Edward> have to buy two units, plus the SnapMirror licensing.  We are
Edward> talking *way* more expensive to do the same thing with the
Edward> NetApp.

Sure, you'd need to buy two sets of hardware for the same type.  No
arguement.  But I'll also argue that a FAS3140 will have the
horsepower to push your SCSI tape drives all day and night and you'd
never notice it.

But restores from Tape are going to suck if you only need/want a
single file or a couple of files.  And you haven't addressed that
usage scenario at all.

Sure, restores from snapshots are trivial.  Never argued they
weren't.  And I personally *like* snapshot restores.  But when an
engineer creates a 500+gb file during a simulation run, it will simply
*kill* your snapshot reserve, and reduce the usefulness of snapshots
remarkably.   

This is the hidden killer issue of snapshots, single large chunks of
temporary data which get put into snapshots by accident are painful.
Suddenly you need to clean out snapshots more quickly than you
planned, and then you're back to tape.

Remember, users do NOT care about backups.  Ever.  They only care
about restores.  

John