Re: [DNA-BOF] BoF Last Call on DNA WG Charter.

[Bernard Aboba <aboba@internaut.com>]

> Whether we need to do work like this in DNA
> is up for discussion.  If we have an 'authoritive hint'
> from the link-layer, obviously this wouldn't
> be needed.
>
> Personally, I like to rely on the link layer as little
> as possible.

In general, the argument for doing this in the link layer depends on how
costly it is to do at L3.  In the case of IPv4 there is considerable cost
because we need to do DHCPv4 for address assignment and thus existing
implementations typically don't use IRDP.  So one might make a case that
there is some value at L2, although most of the current L2 link-Id
proposals don't seem to adequately take into account the dynamic nature of
VLAN assignment today.  This means that an L2-advertised Link-Id is only a
"strong" hint if it is advertised after the VLANID has been assigned.
Prior to that point, it is just a "weak" hint because the assigned VLANID
could turn out to be different than the one being announced at L2.

In IPv6, the case for L3 linkIds is much stronger because we need to do
RS/RA anway, so adding a 'linkId' seems quite natural.  Of course there is
nothing that says that a dual stack implementation can't make use of an
IPv6 facility as a "hint" for IPv4.