Fabric Failover with VMware

[Sandeep Singh]

<<Original post by Erwin. Answered by Michael. Moving from document section to discussion>>

Hello everyone!

I am fairly new to this, so forgive me my possibly simple questions:

I have a UCS (v2.0) with 2 Fabric Interconnects running a ESXi 5.0 host on a B200M2 blade. FI A has a single 10G uplink to a 6500, FI B has a single 10G uplink to a second 6500, no VSS.

I define 2 vNICs on a Cisco VIC, one going to Fabric A, the other to Fabric B, which are both are presented to a plain vSwitch0 (no N1K) for the service console and vMotion as physical adapters.

I define another 2 vNICs the same way for a second vSwitch1 for VM traffic.

What would you experts recommend:

vSwitch0 active/active or active/standby pNICs?

vSwitch1 active/active or active/standby pNICs?

enable Fabric Failover for the vNICs or not (faster as VMware?)

I watched Brad Hedlund's (btw terrific) videos to learn the UCS networking, and i was wondering if i could/should use his recommended N1K design also for a plain vSwitch with 2 pNICs

Thank you in advance!

Best regards,

Erwin

[Sandeep Singh]

Hi Erwin,

To answer your question, enabling fabric failover on a bare-metal hypervisor, such as ESXi, can be done and will work, but generally it isn't enabled and the failover is left to the vSwitch/dVS/Nexus1000v. Prior to 1.4, enabling fabric failover in this type of scenario would cause issues. Also, I assume that the two 6500s are still a part of the same Layer 2 domain. That is, the same vlans are present and active on both 6500s, so that either Fabric Interconnect A and B has access to communicate via the same vlans. If not, enabling Fabric Failover can lead to traffic black holes.

If you look at the best practices guide for Nexus 1000v on UCS, it still recommends NOT to enable Fabric Failover for the vNICs. Similarly can be thought of for the VMware vSwitch.

http://www.cisco.com/en/US/prod/collateral/switches/ps9441/ps9902/white_paper_c11-558242.html

If you liked Brad's video, have a look at his write up for Fabric Failover:

http://bradhedlund.com/2010/09/23/cisco-ucs-fabric-failover/

Thanks,

Michael