[ppml] Comments on revised 2005-1 proposal of 2006-02-03

[Stephen Sprunk]

Thus spake <Michael.Dillon at btradianz.com>
>> > I do think that a /40 is about the smallest sized block that I would
>> > like to see given out as IPv6 PI space at this time. Just how you
>> > define who is large enough to justify such a assignment I do not
>> > know.
>> >
>> >   I personaly would rather see unique street addresses be considered
>> > as justification for space, more so then number of employees... but
>> > perhaps either or... or some combination of both, etc...
>
>> This is a dead end. The last 2005-1 was defeated at the last ARIN
>> meeting specifically because of the 100k host requirement.
>
> I beg to differ with you. Glenn is trying to define how
> to justify a specific size of IPv6 address block. This is
> something that we have not had to define before since the
> IPv6 rules that we now have, only cover allocations to
> providers who then make assignments to their subscribers.
>
> In this case, 2005-1, we have to define rules that ARIN
> would use to determine the size of a block to be assigned
> to a subscriber who comes directly to ARIN rather than
> going to an upstream network operator. We are already in
> broad agreement that it is justified to make such an
> application to ARIN when there is no single upstream provider
> to provide a single assignment.

I think some of the ISP folks would disagree but would rather support a 
somewhat reasonable policy than risk a completely unreasonable one.

> Now, how do we measure justification for IPv6. As Glenn has
> pointed out, it is more consistent with IPv6 addressing
> design and existing policy, if we count the number of physical
> locations. He suggests street addresses. Others might suggest
> that we count the number of buildings. In either case, the
> ARIN analysts could request backup information to confirm these
> numbers.

There is nothing "consistent with IPv6 addressing design and existing 
policy" about counting street addresses, buildings, etc.  The NRPM 
specifically defines an end site as a subscriber with no reference to the 
number of locations; RFC 3177 has similar wording.  If you don't like this 
aspect of _existing_ policy, you're welcome to propose a change, but you 
can't just ignore it because it doesn't suit your worldview.

In any case, what is wrong with giving a /48 by default to each org and 
using some HD ratio as the trigger for larger blocks?  Any org with a 
significant number of locations will necessarily have a large number of 
subnets, which in turn will justify a large allocation.

> When you say that the last meeting defeated the proposal due
> to the 100k host requirement, you seem to ignore the fact that
> Glenn has moved the discussion away from host counting and
> into location counting,

Right, but now Thomas has suggested 100k subnets -- a number that few, if 
any, orgs in the world have today, and definitely a step in the wrong 
direction.  Even 10k subnets gives you a pretty short list.

> with presumably, one /48 being assigned
> per location. In any case, if we need to justify the magnitude
> of the address block assigned, then we need to measure something
> or other in the applicant's situation.

The HD ratio for IPv6 PA allocations seems a reasonable metric to adopt 
until we have more experience.  Even then, existing policy requires LIRs to 
send applications for assignments larger than a /48 to ARIN for review; if 
we're going to make hard rules for PI assignments, we should amend the PA 
rules as well.  The lack of drive to do so for the latter should indicate 
the lack of need to do so for the former _at this point in time_.

> So the question is, what do we measure? And what is the threshold
> at which an end user orgnization qualifies?
>
> If anything is a dead end at this point, it appears to be the
> concept that measurement is unnecessary and a PI block is
> justified by the mere fact of multihoming. Everybody seems
> to want some measurement, some threshold.

I think there's consensus that IPv6 PI is justified for folks that have (or 
qualify for) IPv4 PI.  The matters of debate at this point seem to be: what 
justifies more than a /48, whether an "end site" should be redefined to mean 
location, and whether tunnels should count as "full-time connectivity".

Perhaps we could propose the part we seem to have consensus on and revisit 
the latter questions separately, if for no other reason than to get things 
moving in the meantime...

S

Stephen Sprunk        "Stupid people surround themselves with smart
CCIE #3723           people.  Smart people surround themselves with
K5SSS         smart people who disagree with them."  --Aaron Sorkin