[arin-ppml] v4 to v6 obstacles

[Matthew Kaufman]

Lee Howard wrote:
>>> IPv6 is cheaper (for the ISP, who doesn't have to pass the expense on to you) 
>>>       
>
>   
>> than large-scale NAT.
>>     
>>>  
>>>       
>> But since they need to provide it *anyway* in order to give me dual-stack (which 
>>     
>
> Large-scale NAT is only required after runout, and only for connections over IPv4.
>   
Agreed, assuming that enough IPv4 addresses can't be found via transfer 
"after runout". I think the real answer is that it becomes needed after 
the *real* runout, which is after holders of IPv4 who don't really need 
it (see all the companies with legacy class A and class B space for 
their internal nets who NAT all of them to the outside world anyway) 
sell it to people who can use it more effectively.
> More IPv6 traffic means few NAT boxes.
>   
Agreed. Which reduces ISP cost... something most users don't care about. 
(And it doesn't help that $19 multi-megabit service and $0 dialup are so 
common that the users are trained to not cover ISP costs anyway)
>   
>>>> On those days when, for whatever reason, the IPv6 
>>>> is down... I don't even notice until I happen to see it in the logs.
>>>>         
>>> That's the goal!  If you can't tell the difference, then that's a success!
>>>       
>> Actually that's the *problem*. If I can't tell the difference, then there's no 
>> reason for me to demand IPv6 *or* for my provider to give it to me.
>>     
>
> Maybe we have different goals.  My goal is to provide connectivity to my
> customers.  IPv6 is a tool to do that, not an end in itself.
>   
Sure. But you'll still need to provide them with IPv4 (NAT or not), 
perhaps forever.
>   
>> IF it is possible to continue indefinitely providing dual-stack (NAT or 
>> otherwise for the IPv4), THEN it is possible to continue indefinitely *not* 
>> providing IPv6 at all.
>>
>> Until there's an application for which I *can* tell the difference, in a big 
>> way.
>>     
>
> Does anything break when it goes through two layers of NAT (i.e., 
> home gateway NAT and CGN, a.k.a. NAT444)?  Several of the 
> widely used NAT traversal systems assume a single NAT layer; not
> sure how VoIP, online gaming, p2p will work with NAT444 (or, if
> the other endpoint is also NAT444, you have NAT44444).  
>   
All NAT traversal systems currently fail in some NAT scenarios and must 
resort to relaying the traffic. The percentage might change as more NAT 
is deployed, but end-users don't usually care about this problem either. 
(When was the last time you used Skype and actually checked to see 
whether your VoIP was going over TCP vs. UDP, much less whether it was 
being relayed via some far-away university network or not?)

Matthew Kaufman