[ppml] Proposed Policy: IPv6 Direct assignments to end sites

[Howard, W. Lee]

> ### * ###
> Policy Proposal Name: IPv6 Direct assignments to end sites
> Author: Kevin Loch
> Policy Statement:
> Changes to NRPM section 6:
> add new section 6.5.8:
> 6.5.8. Direct assignments to end sites.
>          6.5.8.1. To qualify for a direct end site assignment, an
>                   organization must:
>                   a) not be an LIR;

Have we agreed that in IPv6 we'll use the term "LIR"?  We inherited
this from RIPE and APNIC, and accepted it so we could have a 
harmonized IPv6 policy, but most IPv4 policies refer to "ISPs."
That's interesting, since "ISP" is not defined in the NRPM, but
"LIR" is (even in the IPv4 section).

>          6.5.8.3. Subsequent direct assignments to end sites
>                   Only one direct assignment may be made to 
> 			an end site organization.
> 
>                   End sites that require more than 65536 
> subnets should request space from an LIR or consider becoming
>                   an LIR.

So NRPM 6.2.4, would define "the users of the network services that
it [the LIR] provides" as intra-company departments, to whom it
would have to SWIP or equivalent?  Is that unusual?  


>                   Organizations receiving an LIR allocation must
>                   renumber into that allocation and return any direct
>                   assignments within 1 year.  Micro allocations made
>                   under section 6.10 are not subject to this 
> requirement.
> 
>                   An LIR allocation shall disqualify an 
> organization from
>                   receiving a direct end site assignment unless it
>                   agrees to return all LIR allocations within 1 year.
>                   Micro allocations made under section 6.10 are not
>                   subject to this requirement.

Do we have any teeth to the "return within 1 year" policy?  In the
past, the enforcement has been, "or you won't get any more address
space," which isn't much incentive in IPv6.


> Rationale:
> 
>         The lack of provider independent direct assignments is a
>         significant impediment to adoption of IPv6 by enterprises and
>         large content sites. 

In what way does the lack of PI assignments inhibit enterprise adoption?
Is multi-homing harder, is it less reliable, or is it fear of commitment
to an ISP?

For large content sites (web hosting providers?), is there any guidance 
on how they should assign IPv6 addresses?  Should each web server get a 
/48, or a /64, or a /128?  Or does each server get a /64 (per interface)

and each virtual host gets a /128?



> This policy proposal defines clear
>         verifiable requirements for receiving a direct assignment.
> 
>         Current IPv6 multi-homing was chosen as the key 
> requirement for the following reasons:
> 
>         a) it is reasonable to expect that those reqesting provider
>            independence would be connecting to two or more providers.
> 
>         b) the requirement of demonstrating current multi-homing will
>            promote active deployment of IPv6 by those seeking direct
>            assignments.
> 
>         It is possible that future technology developments will render
>         this policy unnecessary. At this time there are no viable
>         alternatives for IPv6 provider independence, other 
> than becoming an LIR.

I admit I haven't been following shim6, but would that be one of the
future technology developments that would render this policy
unnecessary?
If endpoint and network identifiers are abstracted, then provider
independent space actually muddies the routing table.

At this time, multi-homing in IPv6 isn't fully baked.  Does this 
proposed policy help?


> 
>         It is likely that this will help conserve IPv6 address space
>         as most organizations requiring provider independence could
>         easily qualify for an LIR allocation under current policy.
>         Allowing them to apply for the more appropriate /48 is
>         responsible resource management.

If most organizations could easily qualify for an LIR allocation, then
how is the lack of this proposed policy hindering adoption?

 
>         This policy can easily be adapted to increase requirements for
>         direct assignments if future conditions warrant.  For example,
>         the multihoming demonstration requirement could be increased
>         to three or four separate LIR's.  Additional verification
>         of active current multihoming could be used.  Or, as native
>         connectivity becomes widespread the option of tunnel based
>         connections for justification could be removed.
> 
>         It is extremely unlikely that this will result in a 
> "land rush"
>         of direct assignments.  The requirements in this 
> policy require
>         more effort than the current requirements for a /32.

If most organizations qualify as an LIR, and it's harder to qualify as
an end site, how will this proposal promote adoption?

Lee