In Randy Bush's presentation at the plenary, he refers to the
multi-homing customer "grazing the commons" by having a long prefix
route carried into the default free zone and thus consuming resources
in routers in lots of networks which are not paid for that work.
It's a powerful metaphor, but flawed. If we are honest with
ourselves, we have to acknowledge that we put those sheep in that
meadow.
"New" ISPs have sold multihoming to customers as a way to gain market
share against incumbents whenever they could, and as is often the
case, the customers learned the lesson well. They realized that the
only way they can control their own routing policy right now is to
have networks announced in a way that lets them prepend, use
communities, agreed on MEDs and generally have the termerity to act
like they really run a network. As ISPs have had troubles, the
customers continue to have the message that they need to multihome
reinforced not just by sales folk meeting quotas but by mass media and
less technical folk giving their "Film at a Eleven"
Multihoming does present a real problem to routing table growth, don't
get me wrong. But let's not expect to be able to engineer solutions
which take away customer's ability to achieve what they believe
multihoming gets them. Not only won't the customers be happy, but
there might even be ISPs who want revenue enough to keep doing it. We
have to engineer solutions which keep that ability while allowing the
DFZ to survive the process. As Randy noted, ptomaine is the WG
working on the problem, and I won't belabor the potential solutions
here, but I think we have to start the work acknowleging that we have
met the enemy, and he is us.
regards,
Ted