[arin-ppml] Stepping forward, opening my mouth and removing all doubt about

[Iljitsch van Beijnum]

On 28 aug 2008, at 16:12, William Herrin wrote:

> My guess puts it at 3% to 5% but let's use 10% for our calculations
> just to be on the safe side. That's the maximum number of folks choose
> to use applications which either fail or function suboptimally when
> behind a typical NAT firewall.

Give them all NAT and magically the number of people running NAT- 
incompatible applications shrinks to 0%. We'd be doing those misguided  
individuals who try to use these apps today a favor, really.

> A liberalized transfer policy doesn't force anyone to do anything.
> What it does do is enable an ISP to reap a benefit from making the
> effort to use RFC 1918 addresses.

You mean: give them an incentive to break connectivity for their  
existing customers.

ISP-provided NAT is a lot worse than the NAT people run themselves  
because opening up ports is much harder or may even be impossible.

> What we are saying is
> that it is essentially impossible to achieve sufficiently ubiquitous
> IPv6 deployment in the next 3 years as to allow IPv6-only deployments
> to customers. Ain't gonna happen.

Spoken like a true poster from the top level domain of one of only  
three places in the world that haven't embraced the metric system yet.

No matter how much time you allot, IPv6 deployment will never be  
"sufficiently ubiquitous" because there are always holdouts.

With NAT-PT-like mechanisms there is no need for all current IPv4  
users to switch to IPv6, but new users can be given IPv6 and be  
translated by their ISPs with only a very small number of IPv4  
addresses.

I see no reason why this couldn't be put in place within 3 years.

> we're going to need an interim solution

You mean on top of current address conserving technologies such as:

- ethernet switching
- VLSM
- CIDR
- NAT

At some point you'll have to make peace with the fact that IPv4 can't  
power the internet anymore.

> We went through this pretty extensively last year. Control of IPv4
> addresses can be legitimately traded now using The Ruse and The
> Container Sale. No one is proposing that we suddenly make IPv4
> tradable; for all practical purposes it already is. One point of a
> liberalized transfer policy is to give ARIN better control over the
> trading process so that the community can avoid the more egregious
> abuses (like heavy disaggregation).

Oh right, because ARIN is in charge of aggregation. I forgot that.

Actually the RIR policies that try to promote aggregation are harmful.  
See this presentation that I did at the LACNIC meeting (the prefix  
growing stuff):

http://www.bgpexpert.com/presentations/lacnic-ivb-ipv6-routingtable.pdf

What needs to happen is that people return address space when they are  
no longer using it in accordance with the relevant policies so it can  
be given to others. About 10 million addresses per year are returned,  
which is enough to keep giving out > /16 prefixes for years to come.  
The big ISPs that need /12s etc won't be able to get those for prices  
that they can't afford if there is trading anyway, because freeing up  
these amounts of legacy space requires expensive audits to make sure  
that it's really no longer in use so the minimum price will be  
sufficiently high that ISPs can't afford millions of those kinds of  
addresses.

Address trading would do nothing for us except waste a lot of time and  
resources and make a few people rich who don't deserve it. And it  
could lock up otherwise useful address space.