[CUWiN-Dev] IPv6 -- TAKE II: NEED ANSWER BY TUESDAY NOON...

[Quantum Scientific]

On Monday 25 April 2005 4:55, David Young wrote:
> An IPv4/IPv6 host that is stuck behind a NAT firewall
> can make IPv6 connections using "Teredo."  Some of these protocols are
> showing up in mainstream operating systems: I believe Windows XP already
> has both 6to4 and Teredo capabilities, and Mac OS X has 6to4.

Many tunnel brokers do not support Teredo (I use Hurricane Electric), and 
those that do will require you to use the /64 IP block they assign (so they 
can route).  And being an M$ concoction, support for Teredo in Linux and BSD 
is 'eccentric' (Miredo), and even under their sanctioned config, performance 
isn't good.  However some NAT gateways do actually pass protocols 41, 43, 44, 
50, 51 transparently, making IPV6 available inside.  The only way to confirm 
is with experimentation, as info is spotty and of varying vintage.

Another alternative is if your server can be placed in the NATting gateway's 
DMZ (bridged), to be fully visible to the outside.  This is what I do.  I 
suspect most gateways will be DSL or cable, and many allow the user to set 
themselves in the DMZ, by web interface.  Or if a fiber PoP, it should have 
IPV6 native.

Maybe I don't understand the technical issues, but I'm surprised there's a 
problem with IP address conflicts.  Seems like as long as there's a gateway 
for a cloud (doesn't there -have- to be, for backhaul?), it will allocate IPs 
for the nodes, and the nodes will allocate IPs for their local LANs.  Even 
when a node is some hops from the gateway, its DHCP request should be passed 
along to/from the gateway.  If multiple gateways, the closest one would 
allocate, and the gateways would dispense different blocks in the same class 
C (or B?).


> There is such a thing as an "experimental" prefix.  You can get these for
> a limited time only, I think.  Microsoft has one for Teredo (see above).
> I think it is appropriate to experiment with 6to4 and Teredo before
> getting a new experimental allocation, though.

IPV6 has moved beyond the 3ffe::: stage, and that's only legacy now.  The 
public 2001:::'s are given out as a matter of course, and is what should be 
expected.


> To predict CUWiN's costs for provider-assigned IPv6 addresses might be 
> hard, since there are no IPv6 providers in Champaign-Urbana.

Surprisingly Comcast Cable automatically gives you a global IPV6 block if it 
gets the request, and it's possible that other providers do as well.  But 
can't expect them to know anything.  Experimenting with a known working 
config is best.

Best,

Carl Cook