[ppml] NANOG IPv4 Exhaustion BoF

[David Conrad]

John,

On Mar 7, 2008, at 5:44 AM, John Curran wrote:
> In a post-depletion scenario where customers want IPv4
> connectivity and bring their own blocks, ISPs have little
> choice but to accept the customers (and inject the route)
> rather than sending them to a competitor.  This includes
> customers showing up with just a /30 and NAT CPE...

Obviously, the fact that you announce a route does not imply your  
peers or their peers will accept that route.

Ignoring the fact that the vast majority of customers will not be  
bringing their own blocks (they will instead be connecting via a  
(perhaps multi-layered) NAT and thus  not contribute to the routing  
load), there will be back pressure on "transfers" of such long  
prefixes simply because they won't get you where you need to go.

For your nightmare scenario to be made reality the following would  
need to occur:

1) origin ISP accepts long prefix
2) origin ISP announces long prefix to direct peers
3) direct peers ISPs accept long prefix
4) direct peers ISPs announce long prefix to their peers
5) Peers of peers accept long prefix
6) Peers of peers announce long prefix
n) etc.

At any point, if one of these does not hold true, you get some  
unreachability, leaving the customer (or perhaps the origin ISP if the  
customer is paying enough) to go try to negotiate (pay for) acceptance  
of the long prefix by the ISPs that are refusing it.

All for the privilege of having only a single layer of NAT?

Riiiight.

Regards,
-drc