totd

The biggest problem with using non-routable ip addresses on numbered interfaces
whether point to point or frame or atm or whatever, is that you lose outside
connectivity from those interfaces. We tried this, but the essential
traceroutes from our core routes are too important when debugging BGP
problems to the outside.

Robert Bowman
Exodus Communications Inc.

Exactly. Especially when you have downstream customers who only announce
routes via BGP to you and/or other providers, it can be important for them
to be able to trace out with a source address that has global connectivity.

We usually use unnumbered interfaces, though, for singly-connected customers
(unless their routers can't support unnumbered interfaces). The only
major gotcha with that is that if they're using a Real Router (that deleted
routes associated with interfaces that aren't available), you can't
get to their router if their ethernet is down...

Avi

==>We usually use unnumbered interfaces, though, for singly-connected customers
==>(unless their routers can't support unnumbered interfaces). The only
==>major gotcha with that is that if they're using a Real Router (that deleted
==>routes associated with interfaces that aren't available), you can't
==>get to their router if their ethernet is down...

Actually, with release 11.0(x) on 1000-series (can't confirm on other
routers), unnumbered will still work with downed ethernet.

We use ip unnumbered on all remote ISDN and frame routers for cisco's
employees. Even if the remote Ethernet interface is up/down or
admin-down/down, we can still reach the router over the BRI interface or
serial interface.

I know that older IOS releases wouldn't let you, though.