>Routing flaps considered harmful...
A while ago Curtis proposed a scheme for holddowns that
has a property that a holddown time associated with
a route increases with the frequency of this route
flapping -- so that frequently flapping routes getting
progressively long timeouts.
This is a good idea, if a bit of a gross and ugly hack at
one aspect of the problem (frequent flaps of one specific
set of prefixes). There are also ways of helping out right
now that aren't dependent on implementation changes to
router software and are better general fixes for the sheer
number of flaps, imho.
Part of what needs doing is a decision about what should
give a host unreachable/network unreachable message. Right
now it tends to be a box very close to the originator; I
would tend to argue that this is no longer necessarily a
good idea, and that in the case of statically-routed
single-homed internets a box as close as possible to the
destination could do that job just as well (with side
effects like fewer globally-visible route-flaps).
This dovetails into the new religion that doing BGP or other
dynamic routing with singly-homed customers is a bad idea,
even for policy or political reasons. If it's *really*
necessary to do this (people argue that, among other
reasons, this is something that should be done so that the
CIX router can distinguish CIX member routes from
non-CIX-member routes belonging to non-CIX-member
resellers), it probably would be better to route them
statically and advertise a lie saying that the routes
originated in e.g. BGP or EGP with AS XXXX.