2011-02-17 03:40 AM
is there a way to shorten the time it takes ONTAP to do an NFO failover? I notice that the VIF goes down, and then about 60 pings later the CF happens and the networking carries on...
60 pings is OK for CIFS i would suppose, but for NFS it's a different matter.
2011-02-17 09:57 AM
There is no setting to change this viewable options, but I would be interested if some of the hidden options would modify this (and what other things those options might affect) or if there is a hidden setflag or other method to modify the timeout. We use nfo often (since 7.1) and haven't had nfs issues with the nfs client timeo settings set to handle the nfo event. I could see some issues if the network has a quick drop and the cluster fails over (both nodes affected for example) if the timeout is shorter than 60 seconds...but still would be nice to be able to tune the setting.
We also had one case where a customer enabled NFO and also autogiveback. So the nfo failed over, then the partner node saw the other node was up and did an autogiveback, then the nfo did another failover after giveback. It only did this 3x before ONTAP gave an alert and said it would stop the automated takeover/giveback which was good, but another interaction we ran into once.
2011-02-17 04:20 PM
It would be good, it¹s more about the time ONTAP takes to think about moving
over once all the interfaces have gone down.. They are posted as down on
the console, then it takes a little bit for the controller to realise it
should be failing over... The actual takover takes no time at all, just the
arbitration bit I guess....
I can imagine the flippy flopping over controllers was a pain, it used to
happen a lot on the older FC SANS if and ISL went down or was trippin...