|
Home > Archive > Data Storage > December 2006 > CIFS Seamless Failover
You are viewing an archived Text-only version of the thread.
To view this thread in it's original format and/or if you want to reply to
this thread please [click here]
| Author |
CIFS Seamless Failover
|
|
| ron.lindman@gmail.com 2006-12-14, 7:16 pm |
| Hello All,
I'm trying to come up with a definitive answer regarding how NetApp
handles failover for CIFS sessions. I wonder if someone could help with
the following:
Assume Filer A and Filer B are in an Active-Active configuration. A
client makes a CIFS connection which ends up at Filer A. Some time
later, Filer A has a failure (or gets taken offline for an upgrade). At
this point, does the client need to reestablish a brand new TCP
connection and CIFS session which ends up at Filer B?
Scouring Usenet, I've gathered some evidence that the failure is backed
up to the client and a full reconnection takes place. However, I
haven't found anybody say so definitively. It would seem to me that a
truly seamless failover would require Filer A and Filer B to have
replicated TCP and CIFS state. Is there any way to configure a NetApp
cluster to achieve this? If not, do any filer vendors accomplish this?
As a bonus, if you could answer the same for NFS, that would be a
tremendous help as well.
Thanks for your answers!
Ron
| |
| Faeandar 2006-12-15, 1:15 am |
| On 14 Dec 2006 14:54:59 -0800, "ron.lindman@gmail.com"
<ron.lindman@gmail.com> wrote:
>Hello All,
>
>I'm trying to come up with a definitive answer regarding how NetApp
>handles failover for CIFS sessions. I wonder if someone could help with
>the following:
>
>Assume Filer A and Filer B are in an Active-Active configuration. A
>client makes a CIFS connection which ends up at Filer A. Some time
>later, Filer A has a failure (or gets taken offline for an upgrade). At
>this point, does the client need to reestablish a brand new TCP
>connection and CIFS session which ends up at Filer B?
A failover for both CIFS and NFS is truly transparent. No
re-establishing of sessions or connections.
I am not a protocol guru so can not go into the gory details of how it
works but I can confirm without doubt that it does indeed work.
Failback is transparent for NFS as well but CIFS takes a hit.
A failover happens within a CIFS session timeout (~40 seconds) but the
failback is a boot. This boot can be shortened tremendously depending
on the model (disk boot or flash boot) but either way it's still
longer than 40 seconds. So all CIFS connections will have to be
re-established.
About the only time this is an issue is when data is transferring.
Users will never notice otherwise as the OS or App handle the
re-connection automatically.
>
>Scouring Usenet, I've gathered some evidence that the failure is backed
>up to the client and a full reconnection takes place. However, I
>haven't found anybody say so definitively. It would seem to me that a
>truly seamless failover would require Filer A and Filer B to have
>replicated TCP and CIFS state. Is there any way to configure a NetApp
>cluster to achieve this? If not, do any filer vendors accomplish this?
NetApp does do this. It's the failback that bites CIFS protocol. NFS
is stateless but file handles are picked up. And since NFS's timeout
value is configurable and very high by default (15 minutes) it's not
an issue.
For questions like this you are best served by talking with the vendor
directly. They have no incentive to lie to you about something like
this since you'd find it during an eval and they would have spent time
and hardware for nothing.
~F
|
|
|
|
|