Links on the blink - what will/should mci & sprint do? you're kidding, right? that's right up there with in its wholesale wastage of .COM symbol table space.

i rant. anyway:

If we were to seriously adopt the strategy of forwarding even when the
destination appears down, it would be necessary to support this in the
routing protocols to insure loop avoidance. This is not out of the
realm of theoretical possibility. However, we're NOT there today.

the only reason this would be of value is if we think we're having a flap
and that this destination will in the near future be reachable over the
old path. to the extent that this "flappage assumption" is valid, we ought
to solve it with more and better flap dampening, or extended holddowns, or
something else at the routing protocol layer.

right now bgp's route withdrawal is implemented without any timer since
that's what the spec says to do. given the number of false withdrawals i
think we can declare that the protocol spec did not anticipate the diameter
of the net in 1995 or the average intelligence of NSP operators in 1995.
therefore let us retire to IDRP where we can take appropriate steps to
make sure that "BGP 5" (ne IDRP) takes our operational experience into

but please let's stop arguing about it on NANOG.