RE: Fabricpath spine learns tons of MAC, why?

Do you have an M card in the same VDC as the FabricPath Spine? If so it must proxy learn the MACs for the M card. It's kind of a bug so to speak. True spine should only have F ports and all ports should run FabricPath. If you have ports attaching to classical Ethernet or if you're doing SVIs (VLAN interfaces) you're technically not a spine, you're leaf, and the leaf must learn the MACs.

Send more details on your design and some of your outputs, such as...

show module
show vdc membership
show feature
show interface status
show fabricpath route
show fabricpath isis database detail (sp?)
show mac address-table dynamic (sp?)
show spanning-tree

-----Original Message-----
From: nobody_at_groupstudy.com [mailto:nobody_at_groupstudy.com] On Behalf Of Yuansen Chen
Sent: Saturday, July 19, 2014 10:51 AM
To: ccielab_at_groupstudy.com
Subject: Fabricpath spine learns tons of MAC, why?

This may be off topic. Just see anybody can shed some light on it.

We have 2 Nexus 7706 (running V6.2.8) at the core. The devices are the spine of our fabricpath topology. A bunch of other Nexus 7706 leaves connect to the core using fabricpath mode. Theoretically, the spine should not see individual MAC addresses. Unfortunately, the spine sees almost the same number of MACs as seen on the fabricpath leaves. The destinations/paths are correct. They use switch IDs, not CE or trunk ports.
It seems for some unknown reason, the spine strips off the fabricpath header and look into the Ethernet header. I double check the ports of the
spine: one management port in routed mode, all connected up/up ports are in fabricmode, and some notConnected ports are in trunk mode (does this matter?). Did I miss anything here?

PS: Did anybody successfully bring a 100G link (N77-F312CK-26) up between two Nexus 7706 boxes running v6.2.8a? We have to downgrade to v6.2.8 to get the speed.

Thanks,

--
Yuansen Chen
CCIE #20383
Blogs and organic groups at http://www.ccie.net