[arin-ppml] V6 address allocation policy

michael.dillon at bt.com michael.dillon at bt.com
Wed Jan 20 10:37:40 EST 2010

> The word "end site"  in no way implies a single physical location  
> any more than the term "web site"  implied a single physical location,

> or that the term  "web farm"  suggests  an actual farm  with  
> webservers installed in the barn, to keep the animals company;

> the dictionary
> is for providing  common-use definitions  only.  

Just so that we are speaking a common language here, let's check
Merriam-Webster online...

1 a : the spatial location of an actual or planned structure or set of
structures (as a building, town, or monuments) b : a space of ground
occupied or to be occupied by a building
2 a : the place, scene, or point of an occurrence or event <a picnic
site> b : one or more Internet addresses at which an individual or
organization provides information to others <an FTP site>; especially :
web site

That is where 99% of people will begin, including technical people
because most people get their definitions of terms from common usage,
not from RFC lawyering. If ARIN policy conflicts with common usage,
then eventually ARIN policy will have to change because we do not have
a body of ARIN case law to expound upon the meaning of word and phrases,
nor do we have a panel of ARIN judges to ponder the matter and write
learned opinions.

>   The dictionary
> does  not answer for technical or subject-matter-specific 
> definitions like the IPv6-specific word  "end-site"

The dictionary only reflects common usage as you will note in
inclusion of the term "web site".

> It's the intent of the design,  and  good technical practice that
> The usage of "End site"  in  IPv6  documents  has a very similar
> to the word  "autonomous system"  or what ARIN NRPM calls an "end user

No it does not. If it did, then 2.10 of the NRPM would say so instead of

2.10. End site

An end site is defined as an end user (subscriber) who has a business
relationship with a service provider that involves:

   1. that service provider assigning address space to the end user
   2. that service provider providing transit service for the end user
to other sites
   3. that service provider carrying the end user's traffic.
   4. that service provider advertising an aggregate prefix route that
contains the end user's assignment

> "End site"  doesn't mean  "physical place"...   it  means
> "IP-specific site",   as in the logical presence  in  the  IPv6
> address  space  created  by that end-user's  network.

An end site is like the physical place at the end of a driveway. Of
we all know that it is not physical but the analogy of a packet
network with a path that ends somewhere, is understood by all. 

> It is common for companies with several sites to have them all 
> connected to the Internet via a gateway at a central site. 
> Nevertheless, it would be ridiculous for ARIN to treat this a single

It's not that ridiculous.

Yes it is. ARIN does not need to nickel and dime everything to the nth
degree with IPv6 addressing because it would create a two-tier system
of IPv6 addressing. Inside an LIR with their /32 or bigger, they can
follow a liberal model of assigning a /48 to anything that might be an
end site, and still meet their HD ratio targets to get more. But for
the lowly peasants who get an assignment from ARIN, they have to suffer
under a strict regime designed to squeeze every spare ounce of blood
from them.

That is a ridiculous state of affairs and one which we must avoid
in crafting IPv6 policy. Owen's example of rounding up to the nearest
hexadecimal digit, is a good example of the way in which we should
approach the problem.

> I  believe  /48  is  selected  on the assumption that  all the  end
> subnets  would be taken from their  one /48.

All of the end site's subnets...  And end user or customer, could well
more than one site, and could connect those sites together using an 
IP-MPLS VPN from another provider. Nevertheless, every end site is still
a site, even if you only provide a single 1G circuit to that customer
at their central site.

> If  each physical location receives its own allocation, then there's
> really  no reason to  pick /48  over /56.      A large number of
> subnets at a single physical location is quite rare.

Sure there is, the RFCs and the NRPM allow for assigning a /48 to every
site. Using a /56 is optional and should only be considered for end
which are private residences. The choice of a /48 was so that each end 
site receives so many /64 subnets that they are highly unlikely to ever
run out, no matter how much their network expands internally. Clearly, 
even a small commercial site has much more potential for growth than
a large private residence, thus the distinction.

LIRs should never judge whether or not an end-site needs all the /64
that they are assigned. Regardless of the nature of the private
they should all get the same assignment size which would be /56 or /48
depending on your overall architecture and plans. And non-residential
should always get a /48. There will be a very few exceptions to this
in the case of things that look a bit like an LIR/ISP but aren't such as
a large corporate's data center where a single site needs more than one
Generally, you can expect these things to look like a single end site
the outside, but have some kind of internal partitioning that makes it
more like a bunch of sites bundled together, rather like a hosting ISP
rents cages to other companies.

We have to avoid making the mistake of imposing a scarcity regime in the
land of plenty because to do so will create inequities and lead to legal
challenges of ARIN's activities.

--Michael Dillon

More information about the ARIN-PPML mailing list