Re: [dhcwg] DHCPv6 router option

[Thomas Narten <narten at us.ibm.com>]

Iljitsch van Beijnum <iljitsch at muada.com> writes:

> The ability to put a random address in DHCPv6 that is supposed to be a  
> router, but which in actuality may or may not be a working router is a  
> huge step backwards from the current situation where routers  
> themselves announce their existance and therefore there is a  
> reasonable expectation that they are, in fact, working routers.

When ND was designed, it was felt that even routers that advertised
themselves could not be trusted. I.e., they might well say they are a
default router, but then blackhole packets sent to them. Hence, ND
includes Neighbor Unreachability Detection. And the notion of a
Default Router List.

All the RA does, is place a router on the client's Default Router
List. When the client needs a router, it selects one from the list. If
it turns out that the chosen router doesn't actually work, NUD takes
over and a new (different) router is selected. That is how you get
better robustness with IPv6 and routers.

The DHCPv6 option that we are proposing does nothing more than add a
router to the client's Default Router List. If that router doesn't
actually work, it is intended that client will behave just like above
and select another router. Thus, if the client has at least one
working router, it will use that one, regardless of how it learned of
it. It will not use routers that do not deliver packets.

Thus, I don't quite get the hysteria that this option is going to
"break" your network.

> What I suggest is that you change the draft such that router  
> advertisements are still required, but the DHCPv6 option indicates  
> which of the routers that advertise their presence a host should
> use.

As others of said, this would undermine a key operational requirement,
namely, the site doesn't want to use RAs at all.

> I'm also missing any discussion about dead neighbor detection, which  
> would normally make a host select a different default router.

It was 100% intended by the authors that NUD be done with DHC-learned
default routers just as with RA-learned default routers. We will make
that more clear in the next version.

Thomas