On 12 dec 2007, at 19:00, James Kempf wrote:
I also don't buy the argument: "When the RIRs stop giving out IPv4
addresses in 2 years, then everybody will *finally* start to deploy
IPv6". What will happen is that someone will set up a Web site where
globally routable IPv4 addresses are auctioned.
So when I buy a block of four IPv4 addresses, what exactly can I do
with those?
The price of a globally routable IPv4 address will go from zero to
market driven. In other words, a market will develop in globally
routable IPv4 addresses. If and when the incremental cost of
deploying an IPv6 address exceeds the market price of a globally
routable IPv4 address by a sufficient amount (maybe 5x, maybe 10x,
whatever amount matches the disutility to the service provider of
having to deploy IPv6), then IPv6 may get deployed.
Please note that ARIN explicitly says that trading in IPv4 space isn't
allowed. Although of course you can easily _rent_ address space, and
the people providing you with that address space will throw in some
connectivity to the internet with that for good measure.
The problem with the IPv6 transition isn't the cost side: starting
from zero, it's probably cheaper already to deploy IPv6 than IPv4, and
if that isn't the case today, it will be soon. The problematic part is
the benefits side of the equation: with IPv4 you can talk to everyone
connected to the internet, with IPv6 you can't talk to a statistically
significant portion of the people connected to the internet.
Hopefully we can still address this omission by making available a
better version of NAT-PT before the lack of new IPv4 addresses starts
becoming a problem.
But even without that, what will happen after there is no longer
enough free IPv4 space left for the RIRs to grant all requests that
are in accordance with their policies, will be something like the
following:
First of all, there is a large distribution of allocation sizes. Some
large ISPs now get blocks of a million or more IPv4 addresses, while
the minimum size is in the 256 - 1024 range. Even when there are no
longer any blocks of a million, there will still be a lot of smaller
blocks and so far, something in the order of 10 million addresses is
returned to the RIRs each year. So the large ISPs will hit a wall, but
the small content guys will probably be fine for at least another year
or possibly many years.
Apparently it's hard to free up legacy IPv4 space, because this year
we saw the return of a class A block for the first time in many years.
So it's probably a mistake to assume that a little money will
magically make all legacy IPv4 space available, although a lot of
money might and a medium-sized amount of money will almost certainly
be able to free up part of this space. The question is whether it will
be possible to trade this address space. Remember that you can't use
addresses without exposing that use.
So you're a large ISP and you find yourself unable to get the IPv4
address space that you need to connect another million customers. You
can probably scrape that address space together if you invest enough
time and money--the first time. But what about a year later, and the
year after that? I'd say that most ISPs in that situation will prefer
to invest in technology that allows them to continue to do business
without needing to obtain new IPv4 addresses by putting multiple
customers behind a single IPv4 address.
The interesting part here is that there are three ways to do this:
1. IPv4 NAT
2. Dual stack IPv4 NAT + IPv6
3. IPv6 with some form of NAT-PT
Obviously, 2. is more expensive than 1., so unless there is something
pushing ISPs to do 2., they'll be doing 1. unless 3. is more
attractive than 1. However, 1. pretty much means the end of peer-to-
peer applications unless consumers get access to NAT traversal
technology that can work through four NAT devices (source CPE, source
ISP, dest ISP, dest CPE). ISPs have a love-hate relationship with p2p:
on the one hand, Skype means they can sell fewer voice minutes (if
they're also in that business) and BitTorrent clogs their networks,
but on the other hand, these services sell more and bigger
connections. It will be interesting to see whether the love or hate
part wins out.
For IPv6 to be widely deployed based on market forces, the economic
advantage to the service provider must be sufficiently attractive to
overcome the large sunk cost of the current IPv4 network. IPv6
deployment is a good example of how large sunk costs in
infrastructure essentially throttle the deployment of a new
technology, so that it becomes cheaper to continue developing the
old technology.
You mean like DSL and cable internet access rather than new fiber
infrastructures?
Nonmarket forces, such as government mandates, are a different
matter. They could possibly force deployment (and similarly and more
importantly for nonpetrolum powered transport).
If a technology isn't good enough to win the market without help from
governments, do we really want it? There is no evidence that
government technology selection is better than market technology
selection.
_______________________________________________
Int-area mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/int-area
