[ppml] Policy Proposal: IPv4 Transfer Policy Proposal

[Scott Leibrand]

michael.dillon at bt.com wrote:
>> Is there a scarcity of land on Manhattan?  There certainly is 
>> a fixed pool of land there, which is fully utilized.  
>> However, that doesn't stop developers from buying land and 
>> putting it to more productive use.  No one selling land on 
>> Manhattan is risking serious business losses, because they 
>> can lease or buy the land they need on the real estate market.
>>     
>
> But there are many companies in Manhattan who are *NOT* selling
> land because they can't afford to take the risk to their business
> of not being in Manhattan. Sometimes events change the level of risk 
> such as 9-11 making it clear to the financial services industry
> that there was a lower risk if they moved SOME of their operations
> away from Manhattan. But NYSE is still there and many companies 
> stay in Manhattan to arbitrage the advantage of a few microseconds
> less network latency to NYSE.
>   

Does that stop a new financial services firm from opening a Manhattan 
office, or a new employee of that firm from buying a Manhattan condo?  
An effective market does not require many/most asset holders be willing 
to sell, just enough to meet the demand at the market price.

> Moving from an extremely illiquid market to a very
> illiquid market really doesn't change things. In either case
> you will find it hard to sell IP addresses, hard to figure
> out what price to ask, and hard to know if an offered price
> is reasonable.
>   

I acknowledge that there is a risk of an IPv4 transfer market being 
insufficiently liquid to make a significant difference.  However, in 
that scenario, we're right back to the way things would be without this 
policy.

Do you foresee any risks of this policy making things worse, or do you 
just not think it will make things much better?

>>  One such behavior, 
>> IMO, will be address holders freeing up additional addresses 
>> to transfer as a rising price makes it worth their while.  
>>     
>
> How many hundred thousand dollars per /24 does the price have
> to reach before it is worthwhile?
>   

I think that can be calculated based on the cost of setting up NAT in 
your network, changing your DHCP pools to use private addresses, and the 
support costs of giving static public IPs to whatever percentage of your 
customers have problems due to NAT and want to opt out.  If this was 
done in the context of enabling IPv6, I would give a hand-wavy estimate 
that it would cost somewhere in the single-digit thousands of dollars to 
renumber a /24's worth of customers.  That would indicate that at a 
price of $20,000 per /24, significant supply would be available.  (In 
the initial period, I would expect the price to be much lower, as unused 
legacy space comes out of the woodwork.)  If we compare that with 
Marty's estimate of $200,000 on the current (black) market, that means 
that a liberalized transfer policy would be at least 10x better than the 
black market from the perspective of someone needing new IPv4 addresses.

>> Our job is to provide responsible stewardship of Internet 
>> number resources.  In my opinion, denying organizations 
>> access to IPv4 resources so they will "get what they deserve" 
>> is the antithesis of stewardship.
>>     
>
> You're right. We should be denying them access to IPv4
> address blocks because we have run out of them. It's not
> our fault that they are running out and we have essentially
> zero influence on the decision makers that might return 
> unused addresses to the pool, regardless of whether they
> can earn 6 figures by doing so.
>   

If we enable decision makers to earn 6 figures for transferring address 
space, I believe there will be significant numbers of addresses 
available at that price.  In fact, I don't foresee the price getting 
anywhere near that high, as demand will significantly fall at lower 
prices, until supply and demand come into balance.

> We need to start with
> a basic understanding of the technology here. An IPv4 connected
> host can communicate with an IPv6 connected host in the usual
> way, through one or more intermediaries. In the IPv4 Internet
> we call these intermediaries routers, load-balancers, NAT-boxes. 
> Adding IPv6 to the mix also adds NAT-PT boxes, Teredo tunnel servers,
> 6to4 tunnel servers, tunnel brokers, ISATAP and ALGs (Application
> Layer Gateways). These are all things that network operators 
> provide for their customers whether in the enterprise or in the
> telecom space. So it will be possible for IPv6-connected servers
> to provide a service to the general Internet, as well as vice versa.
>   

Where can I buy these NAT-PT and Teredo boxes and ALGs?  How much do 
they cost, particularly at scale?  And where can I get all of the 
middleboxes (firewalls, network management systems, VPN concentrators, 
load balancers, etc.) to support native IPv6?  It's pretty clear to me 
that we are *not* ready to do IPv6-only networking without dual-stack.


> If an ISP can't make the Internet a basically seamless service,
> regardless of IPv6 or IPv4, then they simply won't survive against
> their more nimble competition. 
>   

You are arguing that some networks will run IPv6, some will run IPv4, 
and there will be devices that happily translate between the two.  I am 
arguing that no one will go IPv6-only right away, everyone who does IPv6 
will want do dual-stack, and that they'll need IPv4 addresses to do so.  
Both approaches would satisfy the requirement of providing seamless 
service to the entire Internet, but I don't believe that ISPs will be 
able to make such a seamless service work cost-effectively at scale for 
IPv6-only clients before IPv4 exhaustion.

> One thing that ARIN staff could do to help this process would be
> to run a v6/v4  agnostic and fully transparent meeting network. By
> this I mean that you should be able to bring a pure IPv4 laptop or
> a pure IPv6 laptop to the meetings, and get equivalent service
> including access to any v6 or v4 Internet sites. By fully transparent
> I mean that all the technical details of how this service is
> supplied should be on display for people to copy and adapt to 
> their own networks. Yes, it would mean some development work
> on NAT-PT and ALGs to finish the job so that they actually work
> in the wild, but this is not an insurmountable task and it fits
> in with ARIN's education mandate.
>   

Sounds like an admirable goal, as is a similar exercise to turn off IPv4 
at the upcoming IETF plenaries and try to get people to communicate with 
the IPv4 Internet from an IPv6-only network.  However, until someone 
succeeds at demonstrating how this can be done cost-effectively at ISP 
scale, I think we need to be able to implement the dual-stack transition 
plan as originally specified, and I think that a transfer policy is a 
necessary prerequisite to that.

-Scott