Recently I had to solve a serious problem I came across while migrating an enterprise network to a new infrastructure (1812 routers + DMVPN without spoke-to-spoke communication). When there were just a few devices in the new network, everything worked fine, but when it grew up to about 10 devices (which is nothing compared to EIGRP’s capabilities), EIGRP tunnel interfaces started to flap (log messages Interface Goodbye Received, Route A.B.C.D/EF stuck-in-active state, holding time expired, followed by new adjacency), effectively disrupting any long-term communication in the new network. What was interesting is that all spoke routers were starting to break down at the same time and restarting the hub wasn’t helping.
It turns out that EIGRP limits itself as for the amount of generated traffic. This limit is set to 50% of interface bandwidth (definable with the
bandwidth command), which for a tunnel interface is by default set to a value of… 9 kbit/s. That’s why after connecting a greater number of devices into the network such a bandwidth allocation simply wasn’t satisfactory. Adding the command
bandwidth 2000 to the tunnel interfaces’ configuration solved the problem.