MAC flap on Flex-10 stacking links on external switch

[ChristianWickham]

Our networking team are complaining to me that they are seeing MAC flap alerts from our blade chassis. The annoying thing is that this MAC address is shown in the Virtual Connect manager as being the Stacking link that is BETWEEN our two Flex-10 (horizontally adjacent) modules.

The Cisco switch complains of;
Aug 11 10:20:08 AEST: %SW_MATM-4-MACFLAP_NOTIF: Host 0026.5558.2903 in vlan 1 is flapping between port Gi2/0/51 and port Gi1/0/49

and in the Virtual Connect manager it is listed as;
Port X7 Stacking Link OK Linked 10 Gb Internal 26 00:26:55:58:29:03(X7)

How can I stop this MAC address from going out to the outside network and raising alerts?
Why is this internal stacking link being presented on the uplinks to the Cisco switch?
Has anyone else seen this?

[WFHC-WI]

Hi Christian,

In our environment this sort of error has been resolved by changing individual blade's NIC teams from Load Balanced to Network Fault Tolerant Only.

I'm curious, are your VC modules configured as Active/Active or Active/Standby?

good luck!

[Frank Gessner]

Yes, we saw it.
Are the corresponding cisco ports set to edge? (Best practice!)
And: don't use LACP-Protocol with the current
IOS-Releas of Cisco to VC. There is a known
incompatibility regarding lagp-timer, which can lead to such effects.

regards, Frank

[ChristianWickham]

Thanks for your replies.

Our environment is 100% ESX 4.1 and we do not have any teaming set at the OS level. We do have links from each Flex-10 module in the same vNet, and it shows active links on both Flex-10 modules.
I have asked our Cisco guys about LACP and their config - what would be the best setting to use?

[Frank Gessner]

if there is a setup VC<->Cisco using LACP (channels) which mean active/active you should set the redundant uplinks from auto
-> failover on VC (disabeling LACP)!
on Cisco site the channel will then inactive.
The diasdvantage = loss of Bandwith but stable.
Cisco IOS-Release NX-OS version 4.2(1)N2 will fix the timing-issue for Nexus5000.
verify connected Cisco Ports are setup as edge-ports. (Best Practice Guide HP and Cisco)
VC-FW should be V2.33

regards, Frank

[The Brit]

You should note however, that the failover between VC modules in the event of a failure, takes ~25-35 secs.

If you do not run NIC Teams, it is possible that your systems might not survive a VC failover.

Our experience is that some clusters/systems will fail. You might be able to adjust the "sensitivity" of your systems to Network blackouts. (OS dependent I expect)

Hope this is useful.

Dave