Stephen Sprunk wrote:
If you start with the solution, one can always find problems that it
solves. However, if you start with a defined problem, the solution may
not be what you expect. Every example you've given so far is solved
at least as well by PI space as by ULA space, and in some cases better
by PI.
Stephen makes a good point, so maybe I'll try defining what I see as the
problem space here. First, I'll outline the current forms of IPv6
addresses and some of their characteristics:
* Provider Assigned or Provider Aggregatable (PA) addresses are
assigned by IPv6 transit providers to their customers. These
addresses are *readily available* to pretty much any sized
network, and *globally registered* (can be looked up via WHOIS
and DNS), but are *temporary* (assigned roughly for the duration
of the contract with the provider). Because these addresses come
out of a provider aggregate, they are covered by an aggregate
route (usually a /32), ensuring that they are *globally routable*
(in the DFZ).
* Provider Independent (PI) addresses are assigned by a Regional
Internet Registry (RIR) directly to end user organizations,
usually as /48's. Recipients of such addresses expect to be able
to announce their /48 to their transit providers, and expect that
announcement to be accepted by most/all other networks providing
Internet transit service to ensure they are *globally routable*.
Because of this expectation, the rules for getting IPv6 PI space
require an organization to be running a network of several hundred
machines and multihomed with BGP to two or more transit
providers. As a result, PI addresses are *not readily available*
to all networks, but because they are not limited to the duration
of a contract with a transit provider, they are more or less
*permanent*. As with PA addresses, they are *globally registered*.
* Locally assigned Unique Local Addresses (ULA-L) are *not globally
routable* and *not globally registered*, because they are randomly
assigned and used by an organization. They are extremely *readily
available*, and *permanent* (except in the rare case of a
collision, which is more likely to be caused by failure to follow
the recommended assignment algorithm than by random chance).
To put it another way, here are the four pairs of characteristics, and
the list of which kinds of space fall into each:
Readily available: PA and ULA-L
Not readily available: PI
Globally registered: PA and PI
Not globally registered: ULA-L
Globally routable: PA and PI
Not globally routable: ULA-L
Permanent: PI and ULA-L
Temporary: PA
Now an ideal new type of IP space would be *readily available*,
*globally registered*, *globally routable*, and *permanent*. However,
due to the current routing architecture of the Internet, no such
solution currently exists. Therefore, the next best thing would be to
have three types of IP space that each met three of the four desired
characteristics. (I leave out the possibility of having globally
routable space that is not also globally registered.) We have two of
the three already: PA space is *readily available*, *globally
registered*, and *globally routable*, and PI space is *globally
registered*, *globally routable*, and *permanent*. However, we don't
have any sort of space that is *readily available*, *globally
registered*, and *permanent*, because ULA-L is *not globally registered*.
As concisely stated by Paul Vixie, "ad-hoc local networking depends on
the same global framework as public networking, for example, in-addr
services, RPKI, and whois." I would add that there are numerous other
forms of private networking and internetworking that also depend on
those services.
So, that's the problem statement: we need IPv6 addresses that are
readily available to all applicants without discrimination, permanently
assigned as long as the recipient continues to need the space, and that
come with the kind of in-addr, whois, and possibly RPKI services
provided by global registration. I can think of at least three ways to
solve that problem, but I'll save that for another message.
-Scott
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
