IPv6 over unidirectional links?

[Gorry Fairhurst <gorry@erg.abdn.ac.uk>]

Here is an off-list thread that Bernhard introduced on the use of IPv6 
autoconfiguration with unidirectional links. More thoughts would be welcome.

Gorry


>>> Bernhard Collini-Nocker wrote:
>>>>
>>>> Dear Gorry,
>>>>
>>>> I´d like to add a point that I have raised a few times in the
>>>> past years without too much feedback/support: there is in my
>>>> opinion an issue with unidirectional
>>>>  links and IPv6 (neighbourhood discovery) addressing of
>>>> rx-only network adapters.
>>>>
>>>> --Bernhard
>>
>> Hi Gorry,
>>
>> Gorry Fairhurst schrieb:
>>>
>>> Here is what I recall: this is to do with IPv6 autoconf when
>>> you can't do DAD, NUD, etc. Is that right?
>>>
>> right, it is all kind of tricky when the link is not per se
>> bidirectional. The impacts may in many cases be neglectible,
>> but there may corner cases.
>>>
>>> I don't recall what you proposed to do in this case though,
>>> because on a unidirectional link, I'd expect the sender ND
>>> cache to be pre-populated with IPv6 addresses (since ND/SEND
>>> could not be used to query this). So is it right that the remote
>>> endpoints just rely on RA's to find their own on-net prefix
>>> and configure their IPv6 addresses without being able to
>>> validate their uniqueness (is this an issue though?
>>> - in receive-only applications, the endpoints never send
>>> with this prefix anyway). Maybe I have misunderstood?
>>
>> Well, my concern was/is that in an environment, where
>> destination IP addresses of rx-only devices are non-unique,
>> but are used as identifiers this might cause problems.
>> Also, some mechanisms of IPv6 might fail in what case such an
>> interface might be unavailable? Actually there are both
>> operational and conceptual aspects in that problematic.
>>>
>>> If we need work in this space, we'd need to decide if this
>>> were a 6man or ipdvb issue - but there is certainly no harm
>>> discussing this in ipdvb.
>>
>> I guess so.
>>
>> --Bernhard
>>