Thanks a lot.

I knew that LACP cannot do on different nodes…mine was a rethoric question! J

Anyway: so we have the old same issue and mitigation requires a lot of work to do.

As you certainly have understood the case I report does not involve “common” users and their office files on some share…(they can loose their connection if adviced!)…

…but here we have hundreds of Windows desktop on VMware View platform and their related VDI users that can loose not the last doc they’ve created…but their entire desktop!

Regards,

Da: Parisi, Justin [mailto:Justin.Parisi@netapp.com]
Inviato: mercoledì 29 luglio 2015 17:24
A: Milazzo Giacomo <G.Milazzo@sinergy.it>
Cc: Toasters <toasters@teaparty.net>; Basil <basilberntsen@gmail.com>
Oggetto: RE: NDO & CIFS

No, can’t do different nodes. Same nodes. Just mitigates port failures/lif migrations a bit further.

With SFO/CFO, the disruption does not come on takeover, but on giveback, as the locks/handles have to be returned. And auto-giveback will not perform with active locks, so the only way to create a disruption in that scenario is cf giveback –f/-override-vetos; essentially forcing the giveback to break the locks.

Thus, you have time to procure a “maintenance window” to minimize issues. Vol moves, stop services, etc. Whatever you need to do.

As for risks, SMB2/3 have much better NDO functionality that SMB1. And once CA shares get vetted on normal workloads, this conversation changes quite a bit.

From: Milazzo Giacomo [mailto:G.Milazzo@sinergy.it]
Sent: Wednesday, July 29, 2015 11:19 AM
To: Parisi, Justin
Cc: Toasters; Basil
Subject: R: NDO & CIFS

LACP between ports on different nodes? …

I’m interested on how to mitigate this issue –if possible- in case of to/gb operations for normal o unexpected ha maintenance in cDOT.

Assume that there’s no conditions to perform a vol move (and againg, moving the volume on another aggregate/node will mean to change ethernet ports too).

It seems that using clients with SMB 2.x capabilities (and this is the case, all Windows 7 and later versions) this is possible but I read at the same time that there are a lot of risks.

But if I can do a lif migration with no risk before the sfo in an HA pair I could solve this

Regards

Da: Parisi, Justin [mailto:Justin.Parisi@netapp.com]
Inviato: mercoledì 29 luglio 2015 17:13
A: Milazzo Giacomo <G.Milazzo@sinergy.it>
Cc: Toasters <toasters@teaparty.net>; Basil <basilberntsen@gmail.com>
Oggetto: RE: NDO & CIFS

LIF migrations will only be NDO in SMB2 and 3 if the client has established a durable handle. But SMB1 clients will always have a disruption. One way to build in additional resiliency to CIFS is to use LACP ifgrps to eliminate port failure as a potential NDO issue.

Luckily, only Windows XP and prior default to SMB1. Win2k8 and beyond use SMB2 or later by default. The NDO capability depends on client, SMB version, etc.

SFO/CFO scenarios will always have a disruption unless you use SMB3 CA Shares, which currently have limited use cases.

From: Milazzo Giacomo [mailto:G.Milazzo@sinergy.it]
Sent: Wednesday, July 29, 2015 11:06 AM
To: Parisi, Justin
Cc: Toasters; Basil
Subject: R: NDO & CIFS

Yes, I know that’s related to CIFS itself, but I though that using failover groups that move LIF from one node to another the dialog between storage and clients could remain active.

Reading the NDO scenarios TR you’ve linked me it seems that the “nondisruptive” operation related to the LIF migrate (I presume before the takeover on the remaining node) could have issue and clients risk to loose their connection.

I ask this for I’m facing a particular environment where hundreds of VMware View desktop have their folder profile on a CIFS share created on a NetApp FAS (now 7Mode flexvols and in the next weeks on a cDOT SVM).

With customer were discussing some operations to remove or to add some PCI card in the HA first pair of the new cluster and we must schedule this together with a stop of View clients.

This mean that NDO operations, always on and so on is not true if there’s CIFS.

Regards,

Da: Parisi, Justin [mailto:Justin.Parisi@netapp.com]
Inviato: mercoledì 29 luglio 2015 16:57
A: Basil <basilberntsen@gmail.com>; Milazzo Giacomo <G.Milazzo@sinergy.it>
Cc: Toasters <toasters@teaparty.net>
Oggetto: RE: NDO & CIFS

Note that this is an issue with the protocol itself, rather than the storage. This happens across all vendors.

We have a TR that covers CIFS NDO scenarios.

http://www.netapp.com/us/media/tr-4100.pdf

From: toasters-bounces@teaparty.net [mailto:toasters-bounces@teaparty.net] On Behalf Of Basil
Sent: Wednesday, July 29, 2015 7:49 AM
To: NGC-g.milazzo-sinergy.it
Cc: Toasters
Subject: Re: NDO & CIFS

Using anything but the newest version of SMB (which, as far as I know, is windows 2012 only), no. All LIF migrations and node HA failovers entail a reset of the CIFS sessions.

On Wed, Jul 29, 2015 at 6:26 AM, Milazzo Giacomo <G.Milazzo@sinergy.it> wrote:

Hi all,

with cDOT is there a way to avoid that CIFS sessions fall during a takeover?

In 7 Mode when an NDU operation is performed CIFS must is interrupted and open CIFS sessions fall if not closed before.

Is there a way using failover groups of the data LIF to avoid this?

Regards,

_______________________________________________
Toasters mailing list
Toasters@teaparty.net
http://www.teaparty.net/mailman/listinfo/toasters