[ppml] IPv4 wind-down

[michael.dillon at bt.com]

> IPv6 Internet routes (if everyone does dual stack and 
> advertises the same
> number of more specifics for TE, but discount the extra 
> routes from the
> assigments of dis-aggregates) and project the number of internal IPv6
> routes you end up with a scary number... My projection for 03/2014 if
> wide spread IPv6 adoption occurs by then is 1.3M IPv4 + IPv6 
> routes.

On the other hand, if ISP X has 30 IPv4 routes and one IPv6 /32, and ISP
X also has some magic inside their network to let IPv6 customers get
IPv4 traffic, and IPv4 customers get IPv6 traffic, then you can drop all
their IPv4 routes, gain 30 routing table slots and still have
connectivity to their IPv4 customers. Consider a 6/4 version of
hot-potato routing, i.e. send them the traffic on IPv6 and let them sort
it out. Multiply this by many ISPs and the routing table numbers don't
look so scary.

Of course its not quite as simple as that. You have to decide how to
route v4 traffic. Maybe maintain a v4 over v6 mesh of tunnels with
v4-only BGP mesh? This is where it helps for ISPs to actually start
building IPv6 networks in their labs and also build-out some
experimental WAN links with v6.

--Michael Dillon