Re: RFC6085 update to rfc2464bis

[Bob Hinden <[e-mail]>]

Ole,

> On Jan 11, 2017, at 1:48 AM, [e-mail] wrote:
> 
> Bob, Jinmei,
> 
> As one of the authors of RFC6085 let me try to clarify how it would typically be used.
> Scenario: Wireless AP that already knows the L2 unicast address of all stations on a link.
> Some APs try to improve on IPv6 multicast on wireless by sending the RAs as L2 unicast to each individual station.
> The AP then runs through it's list of L2 unicast addresses and sends the multicast RA with the given L2 unicast mapping.
> 
> 2464bis says:
>   An IPv6 multicast packet may also be mapped to a unicast Ethernet
>   Link layer address as defined in Section 6.
> 
>   An IPv6 node receiving an IPv6 packet with a multicast destination
>   address and an Ethernet link-layer unicast address must not drop the
>   packet as a result using of this form of address mapping.
> 
> 
> As Jinmei also says, referring to section 6 is then wrong. That implies that the 6085 address mapping uses address resolution. Which is not the case.
> 
> 6085 is indeed very underspecified in stating how the mapping is done, from 6085:
>   The determination of the unicast Ethernet link-layer
>   address and the construction of the outgoing IPv6 packet are out of
>   scope for this document.
> 
> 
> Either do (Jinmei): 
>   An IPv6 multicast packet may also be mapped to a unicast Ethernet
>   Link layer address as described in RFC6085.
> 
> Or something like:
>  An IPv6 packet with a multicast destination address may also be mapped to an 
>  Ethernet link-layer unicast address [RFC6085].
>  E.g. when it is clear that only one address is relevant on the link and that the
>  mapping between an IPv6 multicast destination address and an Ethernet link-layer
>  unicast address is already known.

I agree with Jinmei and your comments that referring to Section 6 is not correct.  Adding the reference to RFC6085 is good, but as you note above, it doesn’t describe how to do this mapping.  The text you suggest may be the best we can do.  It’s not exactly an algorithmic mapping, as the node doing it needs to know something about the environment it is running in.  I guess this is the down side of using “Ethernet” on a variety of media that isn’t a real multicast domain.  

Any other suggestions?

Bob