[ppml] Policy Proposal: IPv4 Transfer Policy Proposal

[michael.dillon at bt.com]

> > The imperative today is for those organizations with 
> steadily growing 
> > networks at the heart of their business model (ISPs) to begin 
> > transitioning. Whether it is painful or not, they must do it or die 
> > because network growth is fundamental to their being.

> Given this diversity in cost of migrating to IPv6 
> (dual-stack) and reducing IPv4 demand, there is an 
> opportunity to allow organizations for whom the migration 
> cost is higher to delay migration until IPv6 technology is 
> better developed/deployed, and in the mean time get IPv4 
> addresses from other organizations for whom the migration 
> cost is lower.  Additionally, such a transfer policy would 
> provide an incentive to encourage organizations to migrate 
> their installed base to some form of IPv6 where it's easier 
> to do so, rather than requiring growing networks to do so if 
> it's more expensive for them.

I agree that there will be diverse scenarios playing out, but
as I said, the pioneers of IPv6 deployment, not migration, wil
be ISPs since network growth is at the core of their business
model. This does not mean that they will be shrinking their
IPv4 networks, and even if they do recover some IPv4 address
space, there is no benefit to them to transfer it to someone
else. In fact, the benefit to ISPs is to reuse those IPv4
address to continue selling IPv4 services to the market segment
which demands that.

Given that this whole thing plays out at a time when there is
a worldwide scarcity of IPv4 addresses, I can't see any reason
for anyone to give up the IPv4 addresses that they have until
they are absolutely sure that doing so does not risk serious
business losses. That will take years after IANA runs out of
IPv4 addresses.

An then there is the technical issue of having a usable aggregate
of IPv4 addresses to transfer and not scattered small blocks.
An ISP has a number of ways of using scattered small blocks,
including carrying many long prefixes in their iBGP, but in
the global Internet, only large blocks/short prefixes are usable.

> I suspect that after ARIN free pool exhaustion, all ISPs will 
> offer some form of IPv6 service.  To do so successfully, and 
> support dual-stack, however, there will be a continued need 
> for IPv4 addresses.

That's why I don't believe many ISPs will go to dual stack on
any kind of scale. It is simpler to use MPLS with 6PE at the
edge, or build an IPv6 mesh over GRE or PWE3 on an IPv4 core.
This isolates the impact of IPv6 to the edge, and only to those
edge routers which are needed for IPv6 services.

> > I disagree that this policy does what you say. In fact this 
> policy is 
> > trying to set up a market for buying and selling IP addresses.
> 
> I don't think we're in disagreement here.  This policy 
> proposal allows organizations to choose what's best for them 
> by setting up a market for transferring IP addresses.  

As I have said in the past, you can't have a market without
liquidity and a resource that is not only scarce, but has a 
cap on supply, simply cannot provide that liquidity. I have
no doubt that there will be a few sales of IP address blocks, 
regardless of whether or not any RIR implements policies to
allow it, but the prices will not follow a pattern, and
will not be predictable. In general asking prices will be
astronomically high, and some sales will actually take
place at very high prices. But other sales will go through at
a fraction of the asking price because the only reason the
address block is on the market is because the seller genuinely
has no use for them any more.

> Yes, it is clear and simple, but it is not sufficient.  If a 
> network has no incentive to go to the trouble of renumbering 
> out of IPv4 addresses, they won't return them, and there 
> won't be enough IPv4 addresses to meet the needs of those who 
> need more addresses. 

Since when is it our job to provide incentives to install NAT
or IPv6? People have been warned by the RIRs and ICANN that IPv4
is running out. There will be more warnings. Those who don't heed
the warnings will get what they deserve. IPv6 and IPv4 are amazingly 
flexible technologies and with the two to three years of advance
warning there is no need for anyone to suffer. If an organization
knows that they are absolutely dependent on IPv4 for some part
of their network where inflexible devices are attached, they should
begin now to make the rest of their network usable without IPv4.
In the worst case, where an org needs a steady supply of globally
unique IPv4 addresses, they can borrow a block like 126/8 by
ensuring that they filter out any route announcements from Softbank
Japan. Or similar with another /8.

It is not against the law to use IP addresses allocated to another
organization if you have no need to communicate with that other 
organization. If you talk to systems integrators you will learn that
this is common practice, not just in the enterprise, but you even run
across it inside telecom companies.

Running out of IPv4 addresses is not like hitting a brick wall.
It's more like colliding with a giant marshmallow and wondering how
to get out of the sticky mess.

--Michael Dillon