[arin-ppml] IPv6 Non-connected networks

michael.dillon at bt.com michael.dillon at bt.com
Mon Mar 22 05:27:37 EDT 2010

> The ISP _will_ listen because the customer has money, 
> collectively if not enough individually.

You are wrong! Spammers have lots of money, but the vast
majority of ISPs refuse to accept customers who spam.
Money is not the only factor in an ISP's decision.

ULA-C is only intended to be routed between consenting
parties, which means that ISPs will likely accept a few
ULA-C routes for transit where two or more of their
customers need to interconnect their networks. That is
OK and is indeed what ULA-C is intended to facilitate.
And there is nothing bad about that because it does not
affect the global routing table or the public Internet.

> If this would really work, then why bother having a separate 
> database for ULA-C?  We could just put them in WHOIS like any 
> other addresses, just with a comment on the record that RFC 
> XYZ says they shouldn't be routed publicly.

Because ULA-C addresses don't come from a single RIR's supply
but from IANA's unique database, even if that database is 
operated by a couple of RIRs. And ULA-C addresses are not
intended for use on the public Internet so they should have
a whois entry similar to

And because if someone comes to an ISP and asks to route a
ULA-C block, the ISP can say that for the past 12 years our
policy has been to *NOT* route blocks which are not registered
to your organization in whois. The whois entry will back them
up, and when the customer points to the other database, the ISP
can legitimately say, but that is not a whois entry, that is 
some other database which also has a note saying that these
addresses are not for use on the pulic Internet.

> We're supposed to be considering the good of the entire 
> community here, not just ourselves, and IMHO ULA-C is bad for 
> the entire community.  It made a _small_ bit of sense back 
> before PIv6 was allowed, but that excuse is gone.  If GUAs 
> are too difficult for legitimate users to get, we need to fix 
> the GUA policy, not create a way for people to bypass the system.

Enterprise users, who are sorely underrepresented in the RIRs,
rather like to have internal networks addressed with blocks which
don't work on the Internet. It adds an additional layer of security
in case various people make mistakes in configuring things like
routers and firewalls.

> Also, by specifically saying that ULA-Cs are 
> unroutable, that implies that non-ULA-C addresses _are_ 
> somehow guaranteed to be routable, and that is a dangerous 
> implication for us to make.

The wording can be adjusted. For instance it could say that
unlike the global unicast addresses in 2000::/3, addresses
from the ULA-C range are not guaranteed to be routable on
the public Internet. Typically, any ULA-C routes will be 
blocked at AS boundaries.

--Michael Dillon

More information about the ARIN-PPML mailing list