[arin-ppml] IPv6 Non-connected networks

[Matthew Kaufman]

(Keeping all context for clarity)

David Farmer wrote:
> Matthew Kaufman wrote:
>> David Farmer wrote:
>>> I want people to realize that current IPv6 policy allows someone who 
>>> could justify a non-connected network under IPv4 policy and to get 
>>> globally unique IPv4 addresses per 4.3.5 to get an globally unique 
>>> IPv6 addresses too.
>>>
>>> I believe we should do one of the following;
>>>
>>> 1. Implement PP#107 as written allowing non-connected network 
>>> assignments from common blocks with Internet connected assignments, 
>>> or;  (I believe this is the status-quo of the current convoluted 
>>> IPv6 policy)
>>
>> Right. You can already do this, but they look just like connected 
>> ones. These are the kind you should get if you're not connected *now* 
>> but plan to connect *later* and don't want to renumber.
>
> Should they have to meet basically the same criteria as a Internet 
> connected assignments them?
It sure seems that way. From ARIN's POV, there's no difference between 
an address you've chosen to route and one you've chosen to not route 
yet. The operator probably charges you less to not have service, but 
that's an operator issue.

It should be possible to justify such space, however, just as it was 
possible in the past to justify IPv4 space for networks that were not 
yet connected but had some probability of being connected in the future.

The real problem is if you try to keep the split where there's no such 
thing as PI addresses unless you have some crazy criteria like that 
you're actively multihomed... especially this early in the adoption 
cycle. A Fortune 500 company that wants to get their IPv6 now but 
doesn't intend to actually hook up for another year shouldn't need to 
lie on their application to get the space.
>
>>> 2. Define a separate blocks of address space for non-connected 
>>> networks from the space ARIN has already or get more space from 
>>> 2000:/3 for this. By directing ARIN in PP#107 to make assignments 
>>> for non-connected networks from separate defined and published 
>>> blocks, or;
>>
>>  This is a bad idea.
>
> To maybe help develop a consensus, could you expand on why you think 
> this is a bad idea?
Because either the addresses are in a set for which there is worldwide 
IETF and/or IANA-level agreement that the prefix WILL NEVER BE ROUTED 
EVER, or the addresses are in a set which is suspiciously similar to 
allocated-to-RIR, will-be-routed-some-day address space. The latter 
*will* be routed, for one reason or another... which just makes them 
otherwise identical to the blocks from choice 1, only the space is 
strangely sparse in the routing table for a while.

Imagine if we'd known that CIDR was coming, and that the Internet was 
going to be a big popular thing, and we'd said "well, we're going to 
reserve 192.0.0.0/8 and just give those addresses out to people who want 
a class C to play with internally but who don't plan to spend all that 
money it takes to hook up to the Internet". How many of those addresses 
would be in the global routing table at this point?
>
>>> 3. Implement ULA-Central or a similar proposal, either through the 
>>> IETF or the RIR global policy process, as make assignments from a 
>>> block within fc00::/7.  In this case I would suggest pull 
>>> non-connected networks out of PP#107 and starting a whole new policy 
>>> for this.
>>
>> If there's going to be a central registrar for never-to-be-connected 
>> networks (something for which I believe there are good arguments), 
>> this is what would need to happen, and the ARIN PDP wouldn't be 
>> involved until late in this process, and then only if ARIN was going 
>> to be the world's or this region's registrar for that kind of space 
>> (something which I don't believe should be a given).
>
> Are you suggestion we should do both #1 and #3?  It kind of seems to 
> me that you are, and I hadn't thought of that option.  I think I hear 
> you saying there are three types of network in question;
>
> 1. Internet Connected
> 2. Non-Connected
> 3. Never to be Connected
>
> And, never to be connected networks should be served by ULA or 
> ULA-Central, and that filtering them is probably OK.  But, that 
> non-connected networks should be served out of 2000::/3 and that 
> making them easy to filter them is not OK.
That's what I'm saying.

Clearly there is a need (or, I should say, "customer demand") for blocks 
of type 3 which are guaranteed-unique with a central registrar, and not 
simply likely-unique. Whether or not ARIN should be in that business, I 
don't know... there's no reason, really, that the registrars for that 
space must necessarily be the same as the registrars for the other 
space. But it should be possible to get that kind of block.

Several *years* ago I was at a big company that wanted to have a whole 
lot of equipment in things like test labs assigned IPv6 addresses, and 
wanted to be sure those would never overlap the space of companies they 
might merge with, and wanted them "officially assigned", but didn't want 
to pay *too* much for them because, well, they'd *never* be externally 
accessible. Likely-unique local addresses wasn't good enough for them, 
and getting an IPv6 ISP and some PA addresses was exactly the opposite 
of what they needed. I think this would make this set of potential IPv6 
numbering customers happy.

Matthew Kaufman