[arin-ppml] Policy Proposal 107: Rework of IPv6 assignment criteria

[Owen DeLong]

On Jan 17, 2010, at 3:50 PM, William Herrin wrote:

> On Sun, Jan 17, 2010 at 1:04 PM, David Farmer <farmer at umn.edu> wrote:
>> I would like to find a replacement for HD-Ratios
>> too.  But I haven't figured that out just yet
> 
> My observation here is that IPv6 addressing seems to be LAN-centric
> rather than host-centric. That is, it's driven by the number of /64
> LANs deployed rather than the number of individual computers.
> 
Correct.  In current policy (and hopefully any future policy) host count
is irrelevant.  An IPv6 /64 allows for several undecillion hosts in
each network, so, the number of hosts becomes far less relevant.
> 
>>> 2. I'm concerned about assignments to non-connected networks where
>>> qualification is based on the promise that they won't ever connect to
>>> the Internet and therefore won't introduce a route into the IPv6
>>> backbone. If the promise is meant to be kept, I don't think such
>>> assignments should be made from address blocks within 2000::/3.
>>> 2000::/3 is intended to be the block used on the public Internet.
>> 
>> I understand the concern, I share it, I am open to suggestions.
> 
> Speaking off the cuff, I think I'd shape it like this:
> 
> 1. Ask IANA for a /16 delegation of of the existing ULA space, e.g.
> FC42::/16.Failing that, simply assert regiistration over a portion of
> ULA space e.g. FD42::/16.
> 
Personally, I would rather see us move in the direction of making no
distinction between numbers for connected and disconnected networks.
Unless you want to put ARIN in the role of gatekeeper to the routing
table (which I think is a bad idea), there's no need for such a distinction.

> 2. With a mostly automated web-based system, accept registration of
> /48's within the space.
> 
> 3. A registration account costs $10/year. No concept of organizations;
> just accounts each billed seperately.
> 
> 4. All /48's in the account must be contiguous to the maximum extent
> possible. Each /48 registered costs an additional $1/year. In ULA
> parlance, each /48 is "one Global ID."
> 
This pricing strategy, while interesting, isn't particularly relevant to
a policy discussion. If you want to talk about fees ARIN should
charge, I believe it is better suited to the arin-discuss list.

> 5. Private registration available if desired at no cost. If private,
> ARIN will publish a relay email address that can be used to contact
> the registrant's real email address. They'll publish no other
> information. After all, do we really need to know that the DOJ is
> using a particular range of private IP addresses privately inside
> their private system? I don't think we do.
> 
Depends on whether it leaks into the global routing table or not.
If it does, it's good to know who to call and say "Did you mean to
do this? If so, you're using the wrong prefix.  If not, review your
configuration."

> 6. RNDS delegation in the public DNS if desired. Let the registrants
> decide for themselves if they want leaky name lookups to lead back
> inside. Could be very helpful in a large private network where you
> don't want every participant to have to plug lots of exceptions into
> his DNS server.
> 
Yep... Alsouseful for the organization that built out a huge "non-
connected" network that later needs to connect and they'd rather
bribe their ISP than renumber.

> 7. Registration is non-binding. ARIN guarantees only that if both
> networks participate in registration then they won't have conflicting
> address use.
> 
Should there be inter-RIR cooperation on this such that if you participate
in ARIN registration, you're not going to conflict with APNIC registrants?
If so, this probably requires a global or globally coordinated proposal.

Owen