MAC vs DUID issue

sthaug at sthaug at
Wed Apr 2 20:26:48 UTC 2014

> > There's been some spirited discussion about this on Nanog recently,
> > see for instance
> Douglas Owen's response was correct.   The choice of wording in 3315 with respect to opaqueness was unfortunate - the intended goal of that text was to require that DHCP servers consider the _whole_ DUID as the unique identifier, not just the link-layer portion, but it wasn't intended that the link-layer portion couldn't be used for other reasons.

That may well be the case - but unfortunately this doesn't change the
language in the RFC - which states in no uncertain terms that you're
not allowed to peek inside the DUID. It also doesn't help me identify
beforehand which of the three DUID formats a client will use.

And here again we see the problem - some us at times *very explicitly*
need to use MAC address and only the MAC address as the basis for IP
address selection. This applies both to IPv4 and IPv6. For IPv4 there
is the client identifier, which *must* be used if it is provided. For
IPv6 it's the whole DUID mess.

I can give an IPv4 real life example - thousands of DHCP clients (DSL
modems) with *all* of them sending the same client id. I needed to
explicitly patch ISC DHCP to ignore the client id in order to make
this work.

Steinar Haug, Nethelp consulting, sthaug at

More information about the dhcp-users mailing list