We experienced an issue after migrating from a 3750 to an N9k where the ESXi hosts connected downstream were no longer able to reach their local gateway. The hosts were connected via access ports, and used Lo0 on the switch as their gateway. After the cutover to the N9K, the hosts were no longer able to ping Lo0. Nor was Lo0 able to ping the hosts. source-interface Vl100 needed to be specifically appended to the ping command in order for ICMP to work. After changing the hosts' DG from the loopback to the SVI, traffic started flowing.
Can anyone elaborate exactly why we experienced this behavior with Nexus, but not with Catalyst? My guess is that Nexus uses a different switching mechanism than CEF on Catalyst. Perhaps CEF on Nexus doesn't forward frames based on existing L2 adjacencies? It seems to me that the Nexus wasn't internally routing traffic from VLAN to loopback. But the Nexus is routing. It has RIB entries and running EIGRP.
No comments:
Post a Comment