Re: [v6ops] How do you solve 3GPP issue if neither operator nor handset supports PD?

[Philip Homburg <[e-mail]>]

> > DHCPv6 has a very complex model. There can be multiple relays and multiple
> > DHCP servers. So a downstream router first has to collect offers and then
> > select one.
> 
> Its hard to see what you could remove. A RA solution would likely
> have to be implemented with some sort of database backend like
> RADIUS. That something isnt specified, doesnt make it simple.

Radius is an optional feature. Yes, you can use it if for example you have
static prefixes, but you can just as well have a pool of prefixes on the
router.

For SLAAC, a router also needs to know what prefix to use. You could use radius
for that, but I assume that local configuration is more likely.

> > Worse, we have trouble tying a DHCP PD to link state.
> 
> The goal of PD is _not_ to tie it to link state. I would expect
> that goal to be general regardless of wire-representation.

I think this is the key change we should make is to avoid renumbering problems.
The downstream router has to actively verify that the prefix is still valid.

> > RA has the advantage that all information can be in a single RA, ensuring
> > consistency. We can also add mechanisms that the downstream node quickly
> > and reliably notices flash renumbering independent of link state.
> 
> It might be worth noting that  RAs do not guarantee all information
> in a single RA. It can be split across multiple messages or come
> from different sources.

So that is something we should work out. I.e., specify that related information
has to go in a single message. Maybe a bit that says this a point-to-point
link. Maybe a counter to allow the host to verify that all parts have been
received.

> Lets ask the question differently. Would RS/RA be a good protocol
> for address assignment?

RA is widely used for SLAAC. Though SLAAC has a few renumbering issues.

On the other hand, in the mode DHCPv6 PD is commonly used, it is a bad protocol.