Re: [Softwires] 4rd Address Mapping - version-01

[Ole Troan <[e-mail]>]

Remi,

>>>> I would be tempted to suggest that we write a "an implementation MUST NOT make this assumption" in the MAP draft.
>>>> and that we in IETF documents write this as if there is a prefix reserved for address mapping.
>>>> in the general case there is no way to guarantee that a host doesn't pick the same IID. and we'll have to specify how this would work with DAD and proxy ND. I'd rather not. left up to the implementor is my suggestion...
>>> Taking the EUI-64 into account, a careful design of the 4rd IID can make the probability of conflicts quite low. Then as you mentioned, DAD would further reduce it.
>>> Anyway, that's not to say I'm in favor of this design, the reason I see, of doing this is just to accommodate the double translation while not affect at all the IPv6 addressing (a reserved prefix? no ...).
>> 
>> the DAD/ND redirect solution would fail in the routed home case. i.e. the case where there is multiple levels of routers in the network.
> 
> Could you document more why you think that DAD/ND would fail if IPv6 addresses used to reach 4rd CEs would have a 4rd-distinguishable IID?

this is the translation case for a routed home. DAD/ND wouldn't be done on any of the links the dIVI CE was connected to.
for 4rd this isn't a problem since the packets are destined to the 4rd CE. the dIVI CE has to intercept packets not destined for itself.
in that case it would make sense to reserve a prefix for that purpose.
same as the 4rd CE address comes out of a /64 used for numbering internal interfaces on the CE.

>> that's why I think the recommendation should be to use one /64 of the delegated prefix to the translation "pool".
> 
> This would defeat one important objective: possible operation without ANY impact of IPv4 on IPv6 prefixes delegated to CEs.
> 
> As such, this would be AFAIK a step backward.

I don't see that?

I thought this discussion started out with the case where IPv6 hosts were sharing the same prefix as is used for translated addresses.
I'm pointing out that this isn't a good idea. and I don't see why we need to describe that case in any of our documents.

cheers,
Ole