EIGRP over GRE

[cprice]

I run multiple tunnels over a providers MPLS network. For a while I would have my tunnels flap because of hold-down timer messages. After finding Cisco document 21324 and configuring keepalives and ip tcp adjust-mss the tunnels have stabilized. How do these commands interact with the EIGRP process? EIGRP is not TCP and at the same time I also don't experience high CPU utilization on the hub or spokes in question. Thanks in advance, EIGRP experts.

[jamie.grive]

The TCP Adjust-MSS command is not effecting the tunnel itself, it is effecting the hosts that are using the tunnel for transport. By lowering the MSS the hosts negotiated a lower TCP segment size and as such their traffic does not require fragmentation. This is because of the extra overhead introduced by GRE (24 Bytes (20 IP + 4 GRE)

Excessive fragmentation would cause your CPU resources to be used heavily.

The fragmentation process wouldn't effect the EIGRP and cause for any lost EIGRP messages but if your CPU is so overutilised from the normal fragmentation process it could cause it not to be able to process the EIGRP messages to refresh the hold time.

Another possibility is that potentially during the EIGRP update phase it might have tried to send an update with a 1500MTU and maybe it has the DF-Bit set for EIGRP updates, this could cause the EIGRP update to not get through and then get dropped.

[bjornarsb]

Hi.

Have you tried: tunnel path-mtu-discovery on your tunnel interface?

We are using this in our DMVPN configuration.

regards

bjornarsb

[cprice]

@Jamie - Are you saying that EIGRP, by default has the DF-bit set or the DF-bit can be configured, in later I dont have an overly complicated EIGRP setup and I have not set the DF-bit.

@Bjornasb - I would have used tunnel path-mtu-discover, if the keep-alive and ip tcp adjust-mss did not fix the issue.

I still haven't got my head around why this works.

[Vishesh Verma]

Could you post the hold down timer log message?

Your EIGRP neighborship was flapping because your tunnel was. MSS can't fix the eigrp, nor does the tunnel keepalive, it is something else.

Have you been able to identify the cause the tunnel flaps?

If you have some logs/debugs of when tunnel was flapping, may be we could predict something.

[Caleb Yates]

If the tunnel path-mtu-discovery does not work, try manually  unsetting the DF-Bit with a route-map on both side's tunnel interface. I  have had a similar issue with this before, even with the MSS set.

route-map df_unset

match ip address ACL#

set ip df 0