If you want that control, and I agree that some circumstances and
desires warrant that, then the manual process is great if your
application is allowed to be brought down. But the process of doing
that can sometimes kill an SLA, especially if policies of the business
demand that, say, the page file be wiped upon host system shutdown- a
process that takes forEVER.
Some systems simply don't have the luxury of downtime. I'm sure there
are other ways of accomplishing the uptime for the server, but an
automated process that is transparent to the application is a great
example of a benefit of storage virtualization- accomplishing all tasks
at the storage layer of the infrastructure without any work or knowledge
of the operation at the host layer. It's a religion thing. ;) If
you've ever experienced the rush of Vmotion, which moves virtual servers
from one physical device to another with no interruption of service, you
know what I mean.
I understand that GX alreadyu has a functionality like this....I'd like
to see it in regular ONTap!
Glenn (the other one)
-----Original Message-----
From: Callaway, Michael [mailto:Mcallaway@ntta.org]
Sent: Tuesday, June 20, 2006 11:06 AM
To: Glenn Dekhayser; Carl Howell; toasters(a)mathworks.com
Subject: RE: moving a flex vol to a new aggr ?
This is easily accomplished with what Glen wrote. I know I would rather
have control of the copy of data, make sure it's over there and then to
a one-time reconnect of the lun than just start the process up and have
it complete without me knowing. SnapMirror accomplishes this move
pretty easily, and while the data is live.
When you look at it, whether your data is on 1 head or separate head, it
doesn't matter. The point is your data is on separate disks and you are
going to have to copy it over one block at a time. Moving your
"enterprise" data of a few terabytes is not something you should be
taking lightly. Which is why I don't mind the SnapMirror, update,
quiesce, break. You are in control of the move and most importantly,
you decide when your data is in and your lun can be re-homed.
Thanks, Michael
Server Administrator
NTTA/IT
214-461-2033
-----Original Message-----
From: owner-toasters(a)mathworks.com [mailto:owner-toasters@mathworks.com]
On Behalf Of Glenn Dekhayser
Sent: Tuesday, June 20, 2006 9:11 AM
To: Carl Howell; toasters(a)mathworks.com
Subject: RE: moving a flex vol to a new aggr ?
I would think that this would be a fairly common need, especially for
larger shops with big Netapps.
Given the new 6000 series that can have thousands of drives, and that
different volumes (especially ones containing luns) have vastly
different I/O requirements, you would definitely want to be able to
transparently move a flexvol from a smaller aggregate with fewer
available iops to a larger aggregate where the flexvol can benefit from
more or different RPM spindles. Currently this operation requires
unacceptable levels of application downtime for a 24/7 global
enterprise.
One can extend this feature by having multiple filers communicate with
one another, handing off FC/iSCSI target responsibility for migrated
luns to another head altogether. This of course would be trickier and
require coordination with the host initiator, but hey, isn't that why we
have SnapDrive? :) Within a single filer head I would imagine this
would not be required. Across heads it would.
I smell a great RFE!
Glenn (the other one).
-----Original Message-----
From: Carl Howell [mailto:chowell@uwf.edu]
Sent: Tuesday, June 20, 2006 9:57 AM
To: Glenn Dekhayser; toasters(a)mathworks.com
Subject: RE: moving a flex vol to a new aggr ?
What Glenn is suggesting below is what I'm looking for. We have quite a
few FlexVol's that live in FC aggregates today that I would like to move
to SATA aggregates and don't want to have to SnapMirror or copy them to
get it done.
- Carl
-----Original Message-----
From: owner-toasters(a)mathworks.com [mailto:owner-toasters@mathworks.com]
On Behalf Of Glenn Dekhayser
Sent: Monday, June 19, 2006 9:08 PM
To: toasters(a)mathworks.com
Subject: RE: moving a flex vol to a new aggr ?
What I would have imagined is something like VMWare's Vmotion, which
moves an entire system from one physical system to another. Since the
aggregates are on the same system, we wouldn't have the issue of
snapmirror making a 'new' volume. The process would be as such:
1) Make baseline copy to new volume on new aggregate with separate name
(snapshot will suffice);
2) Do a once/min snapshot until lag < 60 sec.
3) Change to a semi-sync snapmirror
4) Change to a sync mirror
5) Once in sync, in the OS 'swap' the handles for the primary and mirror
volumes (since they are in effect the same data now) so that data is now
served from the mirror.
6) Once swap is complete, break the mirror and (optionally) remove the
old volume.
I think this is what everyone wants! Heck, combine this with FlexShare
and you've got automatic load balancing!
Glenn (the other one)
-----Original Message-----
From: owner-toasters(a)mathworks.com [mailto:owner-toasters@mathworks.com]
On Behalf Of George, Andrew
Sent: Monday, June 19, 2006 8:00 PM
To: John Clear; Fox, Adam; Carl Howell; Greg Wilson
Cc: toasters(a)mathworks.com
Subject: RE: moving a flex vol to a new aggr ?
Err...I would have thought syncmirror would have been the way to go in
that case I'd rather manually move the connections anyway rather than
rely on an automated process And I'd especially not like to see the old
copy getting deleted before I've had a chance to make sure the customers
are happy
-----Original Message-----
From: owner-toasters(a)mathworks.com [mailto:owner-toasters@mathworks.com]
On Behalf Of John Clear
Sent: Tuesday, 20 June 2006 9:06 AM
To: Fox, Adam; Carl Howell; Greg Wilson
Cc: toasters(a)mathworks.com
Subject: RE: moving a flex vol to a new aggr ?
What I'd like to see is a way to move a live volume between aggregates.
This would be especially useful on my R200s, since the aggregates and
volumes are quite large. Ideally, something like 'vol move volname
aggr0 aggr1' would move the volume without any interruptions on the
client side.
My understanding is that vol copy takes a snapshot, and then transfers
that, which means any changes that occur after the copy is started are
not transferred. The only way around that is to make sure the volume
isn't mounted/accessed, which is often an unacceptable amount of down
time.
Between snapmirror and vol clone/copy, the code to do this appears to be
mostly there already. The only piece missing is doing the live cutover
between the old volume and new volume.
John
-----Original Message-----
From: owner-toasters(a)mathworks.com [mailto:owner-toasters@mathworks.com]
On Behalf Of Fox, Adam
Sent: Monday, June 19, 2006 8:43 AM
To: Carl Howell; Greg Wilson
Cc: toasters(a)mathworks.com
Subject: RE: moving a flex vol to a new aggr ?
Perhaps the toaster community could better define what you want from a
proposed vol move command that you aren't getting currently from vol
copy?
-- Adam Fox
adamfox(a)netapp.com
-----Original Message-----
From: Carl Howell [mailto:chowell@uwf.edu]
Sent: Monday, June 19, 2006 9:35 AM
To: Greg Wilson
Cc: toasters(a)mathworks.com
Subject: RE: moving a flex vol to a new aggr ?
I would like to see a move operation added to the vol command. Does
anyone know if this will ever be available?
Thanks,
- Carl
-----Original Message-----
From: owner-toasters(a)mathworks.com [mailto:owner-toasters@mathworks.com]
On Behalf Of Greg Wilson
Sent: Thursday, June 15, 2006 12:23 AM
To: toasters(a)mathworks.com
Subject: moving a flex vol to a new aggr ?
hey all
i need to move a flex vol from one aggregate to another aggregate on
the same filer
is there a really simple command like
vol move <volname> aggr1 aggr2
if not what would be the easiest way to move the flex vol to a new
aggregate?
vol copy ?
thanks
gw
--
Greg Wilson Senior System Administrator greg.wilson(a)aapt.com.au