[arin-ppml] Global Uniqueness vs Global Reachability

[David Farmer]

I've mostly been sitting back and watching the discussion 
ensuing from the Open Access to IPv6 Proposal.

In much of the discussion I've seen a tension between two 
separate camps pushing the dominance of one of two 
properties Global Uniqueness vs Global Reachability.

Maintaining a scalable routing table is essential for Global 
Reachability, and given current technologies this is best 
insured by maintaining a routing hierarchy.  Multihoming 
seems to be the key discriminator if you should have a spot in 
that hierarchy or exist under another part of the hierarchy.  This 
is by and large the primary use that most people have for 
Global Unique IP addresses, in this case IPv6 addresses.

However, there are several uses that Global Reachability is not 
entirely necessary but Global Uniqueness is necessary or at 
least highly desirable.  I believe that the IPv6 address space is 
more than large enough to meet both needs, that's not the 
issue.  I believe the issue is one of assumptions made by 
some operators and most users too, that all Globally Unique 
IPv6 addresses should be or need to be Globally Reachable.

Now ULA (RFC 4193) exists and can probably meet some of 
these needs, but ULA is not quite the same as real Globally 
Unique IPv6 addresses, the primary differences are the RIRs 
and the Registries they provide, authoritative reverse DNS, 
and explicit uniqueness vs. the statistical uniqueness of ULA.

So currently ARIN assignes from two pools of IPv6 addresses 
one for IPv6 Allocations and IPv6 Assignments.  There are 
several other micro-allocation pools too, but I'll ignore those for 
now.  These two pools are split to allow for easy filter 
assumptions to be made, one /32 or less and the other /48 or 
less.  Further, I believe most people assume these two pools 
to be for Globally Reachable allocations and assignments.

What if we made the Global Reachability assumption explicit 
and created a separate pool without an explicit assumption of 
Globally Reachability.

In this way we could create policies which reinforce the routing 
hierarchy for the pools that explicitly have the Global 
Reachability.  While at the same time, we can provide the full 
benefit of Global Uniqueness to those that don't necessarily 
need that Global Reachability.  Creating a separate pool for 
this later purpose allows network operators to easily filter 
blocks that don't necessarily require Global Reachability if they 
so desire.

There should be more that enough IPv6 address space to 
allow this even if everyone had two /48s one from the Global 
Reachability pools, either an allocation from a LIR or an 
assignment direct from ARIN and a separate NON-Globally 
Reachable assignment.  (I know, I need a better name for it, 
but right now it is the idea that is important).

This isn't fully fleshed out, but if people like the idea I'll work on 
it.  If we can really get an ID/Locator split going in the future the 
distinction my be come a mute point, or maybe this new pool is 
used for those that need IDs and the Global Reachability 
Allocations become Locators.  But for now I think this could 
work as a way to make both camps happy.

What do you think?

===============================================
David Farmer                                      Email:farmer at umn.edu
Office of Information Technology
Networking & Telecomunication Services
University of Minnesota		       Phone: 612-626-0815
2218 University Ave SE		       Cell: 612-812-9952
Minneapolis, MN 55414-3029	       FAX: 612-626-1818
===============================================