[ppml] IPv4 "Up For Grabs" proposal

[John Curran]

At 7:38 AM -0500 7/10/07, James Hess wrote:
>In that case, ISPs ultimately reclaim public addresses not used for servers,
>make customers pay dearly for each public IP, and resolve the problem of IPv4
>exhaustion by reducing the number of public IP addresses that are justifiable
>for any user of address space, to a small number of hosts that are used for
>operating well-known services to the public.

James -
 
   Your suggestion (just continue to use IPv4, with smaller and
   smaller assignments to end-sites) works fine, at least for the
   immediate future.  It not only delays depletion of IPv4, it also
   reduces the routing entries per new end-site.

   The challenge is that once there is not readily available new
   blocks of IPv4 space for the ISP's, they will need to explore
   new avenues to obtain new IPv4 to connect new customers.
   Some approaches (such as nicely asking your own customers
   with extra PA space to return it, or mining your network for
   unused 'stranded' space) work just fine and don't cause global 
   impact.  Some of the approaches (getting really big presently
   unannounced IPv4 address blocks from parties which forgot
   they were supposed to return them) also work with effectively
   the same global routing impact as today's system.

   However, there will be a natural tendency for providers of such
   big address space to make it into smaller blocks, since many
   smaller sales (particularly as scarcity increases) could be far
   more lucrative than the one big transfer.  Further, there will
   be a tendency to start mining IPv4 space from areas with
   even smaller potential return (such as unused space in ARIN
   PI or other ISP PA end-site assignments).   Unfortunately,
   as the pressure to continue to connect customers increases,
   these approaches become inevitable, and result in enormous
   load on the global routing system, leading eventually to nearly
   one to one ratio in new global routes to new customers.  At
   that point, it really doesn't matter if super backbone routers
   can do 500,000, 1M, 5M, or 10M routes, they're not going
   to keep up with a one-customer/one-global-route scenario.

   If you've got a way to keep IPv4 running, and still maintain
   the enough hierarchy to keep global routing running, then
   it's time to enter the spotlight and share the secret.  There
   is no doubt that its so much easier for us all to stay on IPv6
   then to move to IPv4, we just don't know how to do it, and
   still keep the Internet running.

/John