[ppml] Summary of Trial Balloons for Dealing with IPv4 Address Countdown

[Iljitsch van Beijnum]

On 30-mrt-2007, at 23:34, Jim Weyand wrote:

> I find myself struggling with how to convert the suggestions and
> comments on this list into actual policy proposals.

Perhaps it would be a good idea to see if we can agree on what our  
goals and assumptions are. An important question is whether it's a  
good idea to try and postpone the moment when the RIRs have to turn  
down requests for lack of free address space. Assuming we still have  
around five years, and that a great deal can be accomplished in that  
time, is it worth going through a lot of trouble to buy us a limited  
number of additional years?

> 4)       Several similar informal proposals to encourage recycling by
> empowering ARIN to more actively police the use of IPv4 addresses by
> various means
> 6)       An informal proposal to ask holders of unused address IPv4
> addresses to voluntarily return the addresses

A "use it or lose it" policy would make sense. Large amounts of  
address space aren't visible on the global internet, so either people  
aren't using it at all, or they're only using it internally. If we  
set a deadline by which address space must be "in use" (hard to  
define) or it will be put at the end of the free list, this gives  
people who are using it for internal purposes a reasonable amount of  
time to move to something else.

> 7)       Several variants of informal proposals to start assigning  
> IPv6
> space with IPv4
> 8)       An informal proposal to get endusers to demand access to IPv6
> networks by creating a media storm similar to Y2K.

The depletion of IPv4 and the adoption of IPv6 are largely orthogonal  
in the short term. Having IPv6 doesn't mean you don't need IPv4 any  
more, not having IPv4 doesn't make IPv6 more useful.

This is what I suggest:

In my opinion, it's a problem that the RIRs are giving out extremely  
large blocks of address space to the world's largest ISPs. For  
instance, Softbank has a /8, Comcast got a /8 in two installments and  
French, Deutsche and British Telecom all have multi-million sized  
blocks. Even very large ISPs need some time to put these amounts of  
address space to use, so what happens is that at various intervals,  
new large blocks are requested, so the number of addresses given out  
in any particular year varies widely because one request can be as  
much as 5 to 10 % of the yearly world-wide use. So giving out large  
blocks makes making predictions harder. Another problem is that if  
and when business stalls, a good part of a large block will go  
unused. For both of these reasons, it's a good idea to limit the  
maximum block size that is given out *today*.

When we start to run out of address space for real, this only gets  
worse, and we run the risk that a large request clears out the  
remaining address space in one go. To avoid this, we should adopt a  
policy where there is a maximum block size, and a minimum interval  
between obtaining address blocks. As the number of addresses left  
gets smaller, the maximum block size is reduced.

For instance, we could make the maximum block size 2 million and the  
minimum interval 2 months. So if an ISP thinks they need 16 million  
addresses in a year, they'll have request 2 million, wait 2 months,  
request another 2 million and so on.

In 3 or 4 years, the limit could be half a million, so someone  
needing 16 million addresses would only be able to get 6 x 1/2  
million = 3 million. (Note that people who need smaller blocks still  
get what they need.) The effect is that an ISP who signs up 16  
million new users each year will then have to share an IPv4 address  
over several users, where the number of users per address increases  
every year, rather than that in year X every user can get their own  
address and in year Y there's nothing left.

The maximum block size could each year be set to (for instance) the  
next higher CIDR boundary of 0.1% of the remaining IPv4 address space.

This policy has the important property of being predictable so people  
can plan rolling out new technologies to deal with the IPv4 address  
shortage in ways that fit their business.

A problem would be that this works per-organization, so it favors  
smaller organizations over larger ones.