Moamen
A few things about Fabric Failover (FF) to keep in mind before I try to address your questions.
FF is only supported on the M71KR and the M81KR.
FF is only applicable/supported in End Host Mode of Operation and applies only to ethernet traffic. For FC traffic one has to use multipathing software (the way FC failover has worked always). In End Host mode, anything along the path (adapter port, FEX-IOM link, uplinks) fails and FF is initiated for ethernet traffic *by the adapter*.
FF is an event which is triggered by vNIC down i.e a vNIC is triggered down and the adapter initiates the failover i.e it sends a message to the other fabric to activate the backup veth (switchport) and the FI sends our gARPs for the MAC as part of it. As it is adapter driven, this is why FF is only available on a few adapters i.e for now the firmware for which is done by Cisco.
For the M71KR (menlo's) the firmware on the Menlo chip is made by Cisco. The Oplin and FC parts of the card, and Intel/Emulex/Qlogic control that.
The M81KR is made by Cisco exclusively for UCS and hence the firmware on that is done by us.
Now to your questions -
>1- As I understand when I create a vNIC , it can be configured to use FI failover , which means if FI A is down , or the uplink from the FEX to the >FI is down , so using the same vNIC it will failover to the other FI via the second FEX ( is that correct , and is that the first stage of the failover ?).
Yes
> 2- This vNIC will be seen by the OS as 1 NIC and it will not feel or detect anything about the failover done , is that correct ?
Yes
>3- Assume that I have 2 vNICs for the same server (metal blade with no ESX or vmware), and I have configured 2 vNICs to work as team (by the >OS), does that mean that if primary FI or FEX is down , so using the vNIC1 it will failover to the 2nd FI, and for any reason the 2nd vNIC is down (for >example if uplink is down), so it will go to the 2nd vNIC using the teaming ?
Instead of FF vNICs you can use NIC teaming. You bond the two vNICs which created a bond interface and you specify an IP on it.
With NIC teaming you will not have the vNICs (in the Service Profile) as FF. So the FF will not kick in and the vNIC will be down for the teaming software to see on a fabric failure etc for the teaming driver to come into effect.
> B- FEX failover
> 1- As I understand the blade server uses the uplink from the FEX to the FI based on their location in the chassis, so what if this link is down, > >does that mean FI failover will trigger, or it will be assigned to another uplink ( from the FEX to the FI)
Yes, we use static pinning between the adapters and the IOM uplinks which depends on the number of links.
For example, if you have 2 links between IOM-FI.
Link 1 - Blades 1,3,5,7
Link 2 - Blades 2,4,6,8
If Link 1 fails, Blade 1,3,5,7 move to the other IOM.
i.e it will not failover to the other links on the same IOM-FI i.e it is no a port-channel.
The vNIC down event will be triggered. If FF is initiated depends on the setting (above explanation).
> C- Fabric Interconnect Uplink failover
> 1- Using static pin LAN group, the vNIC is associated with an uplink, what is the action if this uplink is down ? will the vNIC:
> a. Brought down , as per the Network Control policy applied , and in this case the OS will go for the second vNIC
If you are using static pin group, Yes.
If you are not using static pin groups, the same FI will map it to another available uplink.
Why? Because by defining static pinning you are purposely defining the uplink/subscription ratio etc and you don't want that vNIC to go to any other uplink. Both fabrics are active at any given time.
> b. FI failover to the second FI , the OS will not detect anything.
Yes.
> c. The FI A will re-pin the vNIC to another uplink on the same FI with no failover
For dynamic pinning yes. For static pinning NO as above.
>I found all theses 3 scenarios in a different documents and posts, I did not have the chance it to test it yet, so it will be great if anyone tested it and >can explain.
I would still highly recommend testing it. Maybe its me but I don't believe anything till I have tried it.
> Finally I need to know if the correct scenarios from the above will be applied to the vHBA or it has another methodology.
Multipathing driver as I mentioned before.
FF *only* applies to ethernet.
Thanks
--Manish
