[arin-ppml] 2008-3 Support

[michael.dillon at bt.com]

> Practically speaking, section 2.5 in RFC 4192 is not 
> achievable in a complex network with currently deployed 
> technologies. The IGP's don't support source+destination 
> based routing, they only support destination based routing. 
> The routers themselves can usually support source-based 
> routing statically, but often not in the fast path 
> (hardware-accelerated path) which in a large network has the 
> same effect as not supporting source-based routing at all.

So then, you agree with me that RFC 4192 is a reasonable way 
for community networks to handle IPv6 renumbering. They rely on
hand-me-down equipment and don't handle enough traffic to worry
about things like the fast path. They certainly don't run large
complex networks. Of course, their networks are complex, but
not in the way you describe. They tend to patch together things
that were never meant to be like pringle-antenna wifi links,
Gatorboxes and 10-base2 point-to-point to extend beyond the reach
of 10baseT cables. Not to mention old routers that don't even
have a fast path.

> Even if RFC 4192 offered a realistically implementable 
> strategy for renumbering your hosts, you'd still run afoul of 
> things like web browser DNS pinning, application-layer 
> handling of the name to address map and the lack of timeout 
> info in gethostbyname.

Run the two ISPs in parallel for a month.

> Meanwhile, RFC 2894 is sketchy at best. It calls for 
> communicating with a router via IPSec in order to change the 
> addresses via which you're communicating with the router. 

Nowadays we have ULA addresses for internal stuff so there
is no need to ever change the IP address that you are using
to access your router.

> Reality check: renumbering for IPv6 is no easier and no 
> closer to being solved than renumbering for IPv4.

You exaggerate. People with actual experience at renumbering
in IPv6 have written about their experiences on the web. 
With foresight, it is possible to set up a network so that it
can be renumbered with far less pain than an IPv4 network.

--Michael Dillon