UCS Fabric Failover not really working

[p-blalock]

Hello,

I need a little help with a poblem I'm having with ethernet during fabric failover with the UCS 5108; I have a fairly vanilla infrastructure (attached image); 2 Nexus 5Ks layer2 northbound of the 62xx fabrics, the fabrics are port-channeled and vPCd with the Nexus 5Ks.

My setup is a host running HyperV (2012r2) on the UCS that boots from SAN (direct FC connects from the 62XXs to the SAN); The HyperC host is presented with 3 vNICs, one for management (tagged with native vlan), one for production vms (allows multi-vlans, no native tag), and one for live migration (allows for single valn, no native tag).  The vNICs are set to fabric A and fabric B, but the vNICs are also ticked off for "enable failover."

All works very well until I fail the i/o uplink from the UCS FEX to the 62XX.  When I do the fail of the A FEX, it takes about 5-minutes before the B side fabric learns the hyperV host MAC, and the vms running on the host are never learned by the B side fabric; bviously, all the servers are down during this failover period.  If I then re-enable the A side FEX uplink, it takes about two minutes before HyperV host is learned and packets resume. However, it takes several hours before the HyperV host's vms MACs to be re-learned by the Fabric A.  I think I have it properly configured, but maybe not. I'd like to know if anyone else has had this problem, maybe code issue v2.2.3d, or if I'm missing something critical.

Thanks

[Walter Dey]

Hi

- how many links do you have between IOM and FI ? (should at least be 2 or 4)

- have you configured pc between IOM and FI ?

my 2c

- the failover flag should be used with care ? it comes from the W2008 times, when MSFT didn't have their own NIC teaming.

Best practise with W2012 is:

- create 6 vnics instead of 3, connecting each type (mgt, live migration,....) to both fabrics. No hardware failover flag set.

- then configure multipathing in W2013r2

Therefore the Multipathing and Failover should be done and controlled by the OS, not the hardware.

Good luck

Walter.

[p-blalock]

Thanks Walter,

I have 2 uplinks from each i/o to the FIs and do have port-channels set.  I posted the physical topology diag along with the original post.

I'm getting the same responses from others "don't rely on fabric for redundancy," and since you point out it's best practices with 2012-r2, that's what I'll do.

thanks again