Re: [v6ops] New Version Notification for draft-yourtchenko-ra-dhcpv6-comparison-00.txt (fwd)

[Lorenzo Colitti <[e-mail]>]

On Sun, Jan 12, 2014 at 3:56 AM, Nick Hilliard <[e-mail]> wrote:

> Matt needs to run DHCPv6 anyway,


Because... ? If Matt works for the same billion-dollar enterprise he worked
for a while ago, then he runs a single-tenant datacenter, where it's
perfectly reasonable to configure the hosts via other protocols than DHCPv6.


> it requires the routers to maintain a pile of RADIUS state


Nope: RS in, RADIUS request out, RADIUS reply in, RA out. No state.


> which they don't need to do, opens them up to DoS attacks (requests from
> multiple v6 addresses),


Which they have to in DHCPv6, too - anyone on the link can send them a
request.


> Whether you want to refer to this as "insane" is of course your own
> choice, but I am glad that you have at least some appreciation that this
> suggestion hails from the Heath Robinson school of network design.
>

I don't know how you can say that, given that the number of packets is the
same:

In DHCPv6, a multicast solicit request gets sent. The router receives it
and sends a request to a server, the server sends an answer back to the
router, and the router relays it to the host.

In this scheme, a multicast RS gets sent. The router receives it and sends
a RADIUS request to the server, the server sends an answer back to the
router, and the router sends an RA to the host.

Regardless of that - it seems to me that in a single-tenant architecture,
ECMP over multiple RAs is a much better solution, and the only reason that
Matt has this design in the first place is that *you can't do that in
DHCPv4*.