Re: Node Requirements: Elevating DHCPv6 from MAY to SHOULD

[Christopher Morrow <[e-mail]>]

On Mon, May 23, 2011 at 7:48 PM, Thomas Narten <[e-mail]> wrote:
>
>> ok, so ... as a thought experiment, in v4 you wake up, decide you have
>> no address and are supposed to dhcp for that..
>> in v6, you wake up decide you have no address (and don't know if v4/v6
>> are available)... if you are configured for v6 dhcp, you make that
>> request and get all the 'right' data.
>
>> Essentially, spec dhcpv6 host actions to be the same as v4?
>
> yes.
>
> Have DHCPv6 and SLAAC run independently. If you get stuff via one or
> the other or both, just use them.
>
> The mistake of the M&O bits was that you needed an RA to tell you to
> use DHCP. But if the bits weren't set right, or something, you
> wouldn't run DHCP in cases where you should have. Just decouple the
> two protocols completely.
>
> The one downside is that you run DHCP even if there are no DHCP
> servers. In some environments, that is extra traffic the operator
> might not want. I recall many long threads about how the cost of those
> extra DHCP pacekts on a wireless network were unacceptable...

I guess I was thinking that today you have a device, it either is
configured to do dhcp or is manually configured or just is broken. In
the v6 world you could just forget M&O and require someone to
configure (via os config tweaks that already exist for v4 anyway)
dhcpv6 if anything more complex than 'subnet + random address + defgw'
are required (or if they just like DHCP over SLAAC).

(are we saying the same thing? I'm --coffee so...)
-chris