[ppml] dual-stack (was Re: Expand timeframe of Additional Requests )

[David Conrad]

Hi,

On Aug 16, 2007, at 8:54 AM, Paul Vixie wrote:
> this seems to argue that we should have been deploying dual-stack  
> for some
> years now,

Well, yes.  It would be depressing if this were a surprise.

> and that david conrad's soft landing proposal doesn't go far
> enough since it doesn't REQUIRE dual stack, merely asks to hear a  
> plan for it.

Err, no.

"Phase:	     	3
  Threshold:    	10 /8s
  Requirements:
  ...
  * Demonstrate availability of production IPv6 infrastructure and
    connectivity services."

Well, OK.  It doesn't say "dual stack" as I made the implicit  
assumption that people aren't going to turn off IPv4 to get IPv6.

> if we're going to need new V4 addresses for a long time,

Yes.

> but we only have a supply that will last a short time,

Where "we" is IANA/RIRs, yes.

> then the pressures toward NAT, toward
> black marketeering and piracy, toward deaggregation and routing  
> table bloat,
> and toward RIR shopping and IP address globalization, give me a  
> serious case
> of the heebee jeebees.

It would be depressing if this were a surprise.  For my own sanity, I  
must assume you are merely restating the obvious.

> can we talk about ways to use our last years of new IPv4 to get  
> dual stack
> running everywhere, or at least running in enough places that a V6- 
> only client
> in the post-V4-depletion era will have services they can talk to?   
> should the
> conrad "soft landing" proposal require dual-stack for V4  
> allocations starting
> years before projected depletion?  or maybe just for access-side or  
> content-
> side?

For the sake of argument, assume IPv6 does not take off and that the  
IPv4 free pool becomes depleted to the point where an ISP is unable  
to obtain the address space they request.  Pragmatically speaking, I  
see the following scenarios:

a) the ISP cannibalizes the IPv4 addresses used for their internal  
infrastructure, either renumbering into private IPv4 address space or  
(preferabbly) into IPv6.
b) the ISP negotiates a private deal with some entity that has more  
address space than they are using.  This could be:
    1) acquiring an entity that has address space,
    2) offering services to customers who have address space if those  
customers give/lease their address space to the ISP,
    3) or buying/leasing fragments of address space registered to  
another entity.

Note that only b.3 implies additional routing system load, assuming  
the ISP doesn't announce new more specifics for traffic engineering.   
Also note that only b.1 and (maybe) b.3 imply involvement of the RIRs  
and this assumes the ISPs continue to use the RIRs as the means to  
determine address registration (the increased use of the black market  
for addresses will likely mean the RIR registration databases become  
less and less accurate over time).

However, in all cases, there will be exceedingly strong pressure on  
the ISP to provide to the ISP's customers the absolute minimum  
address space possible, most likely single IP addresses to be used  
for the public side of NAT endpoints. I would imagine this will most  
likely be implemented as _significantly_ increased costs for PA  
space,likely coupled with increased hosting services as a business  
offering.

Regards,
-drc