This was a weird one, and I'm still not sure what happened. I'm going to try to run some more tests, but I won't be able to until next week. Knowing my luck, it'll be something totally obvious...
Scenario is the following:
- Unix machines, running two i40e-driver cards
- Cisco 6509E VSS
- one interface from each card in a multichassis EtherChannel
- Each EtherChannel interface is configured with a native VLAN plus three dot1q VLANs
-
Cisco side example config is:
interface Po42 switchport switchport trunk encapsulation dot1q switchport trunk native vlan 100 switchport trunk allowed vlan 500,600,700 switchport mode trunk switchport nonegotiate
-
Server side config matches, interfaces come up on both sides, traffic passes fine between the server and other similarly configured servers, etc
Now here's the fun part... some UDP packets, sent by only a couple of the other machines, are received twice on this server, with some delay (a few microseconds) - once on each physical interface of the EtherChannel. After I removed "switchport nonegotiate", there were no more duplicates.
I've looked at pretty much any PortChannel 'show' command I could think of and/or find before and after the change, nothing looked wrong on either side (switch & server). I genuinely cannot figure out why the duplication was happening... Has anyone seen this before?
(to be continued, once I manage to run some more tests!)
No comments:
Post a Comment