Re: American Registry for Internet Numbers?

[stewarts@ix.netcom.com]

At 08:23 AM 3/24/97 -0800, Robert Cannon <cannon@DC.NET> wrote:
>Does anyone know anything about ARIN (see www.arin.net).  This is a
>proposal to make IP numbering allocation seperate from domain name
>registration.  Currently NSI is paying for IP numbering paid for with
>its domain name fees.  Thus NSI is a big advocate of getting rid of this
>work which it is not receiving profit from.  But I find it difficult to
>discern from the ARIN web site how credible this proposal is. 

I'm not sure how credible it is.  However, a proposal to pay for
IP addresses now would be really bad timing.  The reason is that
sometime soon the net is going to change to IPv6.  It's not clear when,
given some of the stopgap measures to get by with existing IP space,
but it will probably be within the next 2-3 years.
There are three main differences between IPv6 and the current IPv4
1) It's different, so you need to upgrade your routers and machines,
	which is expensive and annoying.
2) IPSEC security feature support is mandatory, so various government
	agencies are footdragging
3) Addresses are 128 bits long instead of 32 bits.   
	
32 bits is 	4 billion, which would be enough for almost 
everybody on the planet to have one (except that the addressing 
structure wastes part of the space.)
64 bits is almost enough for everyone on the planet to have
a network number with a host number for everyone else.
It's almost enough for every human neuron on the planet to have an address.
128 bits is far, far larger than that.  Even if we waste lots of space
by using an inefficient structure, which we will to make routing easier,
there's still mindbogglingly more space than anyone needs.

There are two reasons to charge for Internet numbers:
1) Even if you've got nearly infinite supplies, somebody still
needs to coordinate it to prevent collisions, and this has non-zero cost.
2) Today, numbers are in short supply, so charging for them can
help reduce waste and bring in revenue for the lucky contractor,
especially if you charge by the address or size of address block.
Tomorrow, there are so many numbers that charging by the number
doesn't make sense - all you need is an initial coordination cost.
So if you want to make big bucks on this business tomorrow,
you'd better get the market locked in today.

There are alternative approaches that don't even require a registry.
For example, divide the address space into a 48-bit host part
(Ethernet numbers are unique) and an 80-bit network part,
and just pick a random network part.  On the average, the probability
of collisions is very low unless 2**40 other people are playing,
and you can put all the machines you want on that network.
If that's not guaranteed enough for you, you could use an
Ethernet board number as the network number (even if your
network isn't Ethernet based, you can buy a $50 board :-)

To put costs in perspective, the last time I checked, the Ethernet
industry charged manufacturers $1000 for a 24-bit chunk of numbers,
and there are 2**23 chunks available; there are very few manuacturers
who'll ever need to buy a second chunk, and not many manufacturers,
so this relatively high price is reasonable.  On the other hand,
if somebody wanted to set up a web-based registry of IPv6 numbers,
they could dole them out for a few dollars per chunk, most of which
would be spent on billing.  Perhaps the Ethernet folks could sell
IPv6 registry space for $1000/registry?

A bit of initial coordination can set aside different parts of the
number space for different strategies, and let the market settle it. 



#			Thanks;  Bill
# Bill Stewart, +1-415-442-2215 stewarts@ix.netcom.com
# You can get PGP outside the US at ftp.ox.ac.uk/pub/crypto/pgp
#     (If this is a mailing list, please Cc: me on replies.  Thanks.)