[arin-ppml] A modest proposal for IPv6 address allocations

[William Herrin]

On Sat, May 30, 2009 at 7:49 PM, Owen DeLong <owen at delong.com> wrote:
> Uh, since the first assignment from an ISP to a customer could well
> be a /48, how exactly are they supposed to accomplish that and number
> their own infrastructure for even a single POP out of that first /48?

Hi Owen,

Maybe we should discourage that behavior.

A /48 is 16 /52's or 256 /56's or 4096 /60's.

If we assume, as I do in this plan, that any multihomed entity will
get its IP addresses directly from ARIN then only single-homed
entities will be downstream from you. A /52 is 4000 subnets. Can you
identify many single-homed entities who use or would use more than
4000 subnets? Neither can I. In fact, I'd be willing to bet that only
a fraction of a percent of single-homed customers would ever use more
than the 16 subnets provided in a /60.

Let me put it another way: an IPv6 /48 contains more subnets than all
but the largest dozen ISPs have ever used in IPv4. Surely they can
find a way to make do with their very first /48 long enough to justify
the upgrade to a /32.

That having been said, I wouldn't be morally opposed to allowing
entities which hold at least a /16 of IPv4 space to skip the /48 step
and go straight to /32 if they want to and have been doing a
responsible job with SWIP or RWHOIS.


On Sat, May 30, 2009 at 3:51 PM,  <bmanning at vacation.karoshi.com> wrote:
>  can you say,,,  A. B. C....

Bill,

Would classful addressing have led to the same problems if there were
16 million A's in the pool so that we didn't have to give anybody
multiple C's or B's?

The fellas at the IETF have told us that our best shot at restraining
the DFZ BGP table size is to standardize the prefixes that go into the
table in a way that limits single-entity control of disaggregated
address space and discourages the propagation of disaggregated routes.

The particular way they suggested for limiting disaggregation doesn't
look very promising, but they're not entirely wrong.


On Sat, May 30, 2009 at 8:17 PM, Garry Dolley <gdolley at arpnetworks.com> wrote:
>  1. RFC 3177, "IAB/IESG Recommendations on IPv6 Address Allocations to Sites"
>     http://tools.ietf.org/html/rfc3177
>
>  2. RFC 5375, "IPv6 Unicast Address Assignment Considerations"
>     http://tools.ietf.org/html/rfc5375
>
> Everyone who is participating in these policy debates should read
> these.

Garry,

Agreed, however...

> This is because all their downstream
> assignments will be /48's (RFC 3177).

The RFCs provide useful knowledge but we would be mistaken to treat
their guidance on addressing dogmatically.

I have participated in and attended meetings for both ARIN and the
IETF. In particular, I've been a participant of the IRTF Routing
Research Group for years. I'll tell you straight up: the center of
expertise on the address assignment process is here on the PPML and in
the similar forums at the other RIRs.

In other words, /48 as the standard downstream assignment is not the
gospel truth. It was based in part on the proposition that we should
try to allocate all the addresses an organization will ever need up
front the first time. Should we find that doesn't work as well as,
say, assigning a small block first and then assigning a second much
larger one when you run out, then the advice on /48's downstream no
longer holds.

A better idea might be a /60 for single-homed downstreams, add a /56
when needed and add a /52 after that. Increments, mirroring my crazy
notion for the RIR process. But I'd expect service providers to treat
that notion with all the reverence I'm giving to RFC 3177. And that's
okay too.

Regards,
Bill Herrin


-- 
William D. Herrin ................ herrin at dirtside.com  bill at herrin.us
3005 Crane Dr. ...................... Web: <http://bill.herrin.us/>
Falls Church, VA 22042-3004