UCS C-series standalone server NIC teaming configuration problem

[goranpilat]

Hi all,,

Customer has standalone rack mountable UCS-EZ-C220M3S server with VMWare 5.5. Server is connected through 4 physical interfaces to two 6800s in VSS (two cables to one and two cables to other 6800).
On 6800s these ports are grouped into port channel , mode ON (no LACP). On UCS side in VMWare ports are set to load balance with IP hash algorithm, all ports are 1000 mbps.

The other day we noticed that some phones are malfunctioning. Strange behavior, we realised that the ones not working are all registered to CUCM on this UCS. Further investigation showed, as seen on VMWare side, that out of 4 physical NICs, three were on 1000 Mbps and one was on 100Mbps (seems that one UTP cable was bad, when we replaced it everything went back to normal).

On 6800 side, this link was put to line protocol down state.

my concerns:

-6800 proclaimed link as unusable, while VMWare still considered it as active (though on 100Mb),

-all the servers on this UCS started malfunctioning (though if I remember correctly, they were still pingable and reachable)

-redundancy as concept failed miserably probably because of single disconnected PIN, despite VSS, 4 physical connections, double NICs etc

So it seems this is either the case of bad configuration which can be done better, or there is no way to do it better, which is bad by itself

Seems LACP is not an options here as it requires fabric interconnects, which we don't have

Any suggestion is welcomed

Thnx a lot

Goran Pilat

[Wes Austin]

Please review the following documentation:

https://www.cisco.com/c/en/us/support/docs/servers-unified-computing/ucs-c-series-rack-servers/117637-technote-UCS-00.html

[Yasser Al-Alem]

try to modify  Network failure detection to becon probing this way ESXi will be able to detect when a link is down from SW side.

when ink fails to 100 this will take it out of PO, but from esxi perspective, its still up so no failover will be done 

[goranpilat]

Hi Yasser,

thanks for your answer! However I am not suer if it would work, as per documentation:

https://kb.vmware.com/s/article/2006129

is is stated :

• Beacon probing is not supported with IP Hash. Only link status can be used as a failure detection method. If a link fails without the link state going down, there is no way to avoid network communication issues on the vSwitch.

And AFAIK, IP hash is way to do load balancing...

Regards,

Goran