Re: [v6ops] [SUSPECTED SPAM] Google Alert - IPv6

[Dave O'Reilly <[e-mail]>]

> On 28 Oct 2017, at 03:21, Ole Troan <[e-mail]> wrote:
> 
>>> It is clear that if the number of subscribers that can be simultaneously
>>> behind a CGNAT is limited, the deployment of IPv6 increases, and the IPv6
>>> adoption in Belgium bears that out.
>> 
>> Yes, and I think you’re right, but the Belgian interpretation that address
>> sharing must not exceed 16:1 may not be generalizable. That is: Belgium is
>> one country, and we don’t know if their rule would work everywhere.
> 
> We are digressing from the draft in question, but since I'm sitting somewhere over Oregon at the moment I have ample time.
> It is not at all clear that the number of subscribers behind a CGN is "limited".
> 
> I'd certainly would have liked to see some research on this.

Some links I found (hardly research, but links anyway!):

https://www.intgovforum.org/multilingual/content/igf-2017-ws-214-how-can-we-limit-the-negative-impact-of-carrier-grade-nat-technologies-and
http://www.ipv6conference.ch/wp-content/uploads/2015/06/T04a-Vyncke-ch-council-iPv6-in-Belgium-June-2015.pdf

> If you are clever and use endpoint dependent connections where you can. You can stretch address sharing very very far.
> Since you can reuse source ports for multiple connections, assuming the destination port is largely constant at 443, the number of connections for a single port is bounded by the distribution of destinations on the Internet.
> 
> You will burn one source port for each concurrent connection to the _same_ destination address.
> 
> The bitter truth is that we can make IPv4 'scale' forever. ;-(
> 

Exactly the problem and this means the business case needs to be there for moving to IPv6. The upshot of this appears to be that, at least for the foreseeable future, CGNAT is going to be with us.

daveor