v6 & DSL / Cable modems [was: Private use of non-RFC1918 IP space

[iljitsch at muada.com (Iljitsch van Beijnum)]

On 5 feb 2009, at 20:06, Joe Abley wrote:

> 4) Obtain PA space and do what you're doing with v4.

> 5) Obtain PI space and do what you're doing with v4.

> (4) is problematic because filtering long prefixes in v6 seems to be  
> more energetic than it is in v4. (5) is problematic if you don't  
> qualify for PI space.

Better hope the RRG work (LISP, maybe) works out, then.

I'm sure some people will relax their filters but I'm also convinced  
that a lot of people won't, at least not until a consensus on a good  
prefix length filtering strategy emerges. The RIR policies are such  
that if you allow /48s you're dead in the water if someone tries to  
inject a large number of those on purpose or it happens by accident in  
a particular unfortunate way.

The reason I think people won't accept long prefixes is because of the  
above, or because (like me) they feel IPv6 PI was a mistake, or, the  
main contributor to routing table bloat, laziness. And the reason they  
won't care is that if an IPv6 destination returns !N applications that  
try both IPv6 and IPv4 fall back on IPv4 without a noticeable delay so  
outgoing sessions aren't affected. (Incoming sessions have to time out  
though, no ICMPs back to the originator for those.)