[c-nsp] Default route pointed to an interface
Justin Shore
justin at justinshore.com
Thu Aug 16 15:41:08 EDT 2007
This has been talked about before on this list but I'd like to bring it
up again for my particular situation.
I have a pair of border routers, 1 with 2 upstream connection and the
other with a single connection. I'm pulling down full tables from each
provider and have iBGP between the border routers and the core routers.
Each border router has a static default pointed to the outgoing
interface for the local upstream connections (2 on one router and 1 on
the other). As expected the router with 2 default routes flops back and
forth between the 2 interfaces every second or two.
35578126: Aug 16 10:20:22.385 CDT: RT: SET_LAST_RDB for 0.0.0.0/0
OLD rdb: is directly connected, GigabitEthernet0/3
NEW rdb: is directly connected, ATM1/0.1
35578127: Aug 16 10:20:23.645 CDT: RT: SET_LAST_RDB for 0.0.0.0/0
OLD rdb: is directly connected, ATM1/0.1
NEW rdb: is directly connected, GigabitEthernet0/3
I know that this forces an ARP query every time a packet falls back on
the default route. That isn't desirable of course. However if I don't
hardcode the default route each border router will learn the default
from the other border router thanks to the IGP (both borders originate
the default for the benefit of the rest of the network behind it).
Given that these routers have full tables the only time I can think of
that a packet should not find its destination in the RIB is if it's for
a route that I'm dampening or it's a bogon (which I'm catching with
ingress/egress filters anyway). However routing these packets to the
other border router results in less than optimal routing. I would like
to implement uRPF on the upstream interfaces though so the default
routes will have to be removed for that to work.
So a question would be how I remove the static default without learning
the default from the IGP (distribute-list?)? What would be the proper
configuration for this scenario? Besides the frequent ARPs and my uRPF
desires, is this really a big problem? Or am I missing something
obvious again? :-)
Thanks
Justin
More information about the cisco-nsp
mailing list