Bill Simpson claims that this idea is ~10 years old, but if so perhaps
it's time to air it again. I thought of this during the route flap BOF at
NANOG, mentioned it, and nobody came up with any immediate reasons why
it wouldn't work, so I'm sending it to the list for further consideration.
It's dead simple, really: Assign address blocks to pairs of providers.
Both providers announce those blocks all the time, and assign addresses
out of those blocks to customers who multihome between those two
providers.
Now, clearly this won't help all multihomed nets, because you can't
possibly provide netblocks for all provider pairs- you'd have n*n-1
blocks instead of n, for n providers- but you can optimize for common
cases. In particular, any two providers who will state that they are
going to coordinate multihoming get a netblock. You can also limit this,
at least in the beginning, to some smallish group of providers (such
as the big six/seven/whatever, though that's probably a poor criterion).
Any provider who wants to start doing this today can, of course, simply
by declaring that part of its aggregate is now jointly owned by another
provider, and that other provider announcing that part.
This isn't really a change in technology so much as it is a change in
bookkeeping: by putting all the multihomed nets together, both of the
providers of those multihomed nets can aggregate announcements. But
the benefit is real: fewer routes, and no route flaps when any single
multihomed customer falls off the net. You reach breakeven on # of
routes as soon as any pair of providers has two customers doing
multihoming. That makes this scheme beneficial both to customers and
to providers.
So, any takers?
/a