extending at the edge - Yes!

Thomas Schäfer thomas at cis.uni-muenchen.de
Fri Oct 12 22:59:22 CEST 2012

Am Freitag, 12. Oktober 2012 um 22:02:13 schrieb Philipp Kern:
> Thomas,

> having suffered this with a German server provider… we really don't want
> this as a general solution. Especially given the fact that a static table
> of addresses won't help you with privacy addresses and such.

Of course dhcpv6pd with greater networks would better and every kind of NAT 
would be worse.

It is not a general solution, but it is one solution.

I did the test with a IPv6-enabled test-SIM from a german provider. 

> Sure, it works for a very limited set of machines,

I don't think so. All components are available.
vanilla - linux -kernel
iproute2 (ip) tools
ndppd is new and makes the management of the ndp-proxy-entries easier, so 
privacy extensions should still work.

This could be easily integrated in (3G/4G)routers or smartphones. 

/64 per user seems to become common use. 

Fighting for /56 or /48 per user at mobile-devices could end in a disaster and 
we get hostroutes, not shareable without NATPT.

> but it's pretty ugly.

It is not nice. But it is a routing solution and stable. (unlike todays NAT)
As long IPv6-enabled hardware (e.g. some surfsticks I have) do only provide 
/64 ( I get the net-prefix-info via SLAAC, not via DHCP) I see no chance and 
no need for a change. 

The existing surfsticks work very easily with slaa-configuration for everyday 
life, the special sharing situation needs a little effort.

In the case of dhcp-pd the everday life is more complicated, and you have only 
small advantages for the case of sharing.

Wasting 2⁶⁴-addresses for one device is very much but delegating /48 is much 
more wasting - most people are probably not sharing their net. But it is 
important that they could!

> PS: The documentation prefix is 2001:db8::/32. ;-)

You are right, I saw that after the mail was already sent.


More information about the ipv6-ops mailing list