Leo Vegoda wrote:
On 20 Jun 2007, at 7:22pm, Scott Leibrand wrote:
So am I right in reading your answer as saying that the advantage of
ULA-C is that it solves the same problem that ARIN's IPv6 PI policy
solves but better. In effect, developing ULA-C helps side-step
ARIN's policy development process?
No, it solves a similar problem for a different (though possibly
partially overlapping) set of networks, and reduces the pressure to
apply a hammer when a screwdriver is what's really needed.
I am not sure I understand what you mean by applying a hammer. How is
ULA-C a screwdriver?
Fair enough. Currently, ARIN PI policy requires that to qualify for
IPv6 PI space, you meet the requirements for IPv4 PI space, which is
that you must be multihomed (running BGP) and qualify for a /22, or you
must qualify for a /20. The justification for this is that, since such
a PI allocation usually ends up in the Internet routing tables (the
DFZ), we should have a standard that limits the number of blocks given
out, and gives them out to networks that somehow justify a slot in
everyone's routing tables.
There are a number of smaller networks that wish to avoid "provider
lock-in" by avoiding the perceived difficulty of renumbering their
infrastructure when switching providers (and hence PA blocks), and who
perceive the current policy as unfair. As a result, there are proposals
on ARIN's PPML at least once a year to liberalize ARIN's PI policy in
some fashion. Some of these have succeeded, such as the move to change
the smallest block allocated down to a /22 (from a /20 before, IIRC).
Others have not, due to pushback from operators concerned about
increasing the rate of growth of the routing table.
For many smaller networks, the real goal is stable addressing, not the
ability to route their netblock across the entire DFZ. Therefore, I
believe some of those networks would be well served by ULA-C space.
Since such space does not impose a cost on third party operators, it can
be given out freely.
If ULA-C is not available, I believe many of those networks will instead
push for PI space. Once they get it, the path of least resistance is
announcing it globally, so most recipients of PI space will do so,
increasing pressure on the routing table and requiring more rapid
upgrades (to replace all Supervisor2's in Cisco 6500/7600s running BGP,
for example).
If ULA-C is available, I believe many networks will be able to use it to
number their internal infrastructure, and simply use some sort of NAT to
translate that into their current PA block when traffic sourced from
their routers passes out onto the Internet. They can dynamically assign
their customers/hosts PA space with DHCP, so that end-to-end
connectivity never needs to be NAT'd.
Much of this can also be done with ULA-L, but I believe that the lack of
"ownership" and the inability to do proper reverse DNS makes that
solution not quite as good as ULA-C for many networks.
-Scott
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
