Re: Comments on draft-yourtchenko-colitti-nd-reduce-multicast

[Erik Nordmark <[e-mail]>]

On 2/26/14 5:44 AM, Ralph Droms wrote:
>
> Or, the ND messages could be relayed to the DHCP server...
Ralph,

To support DHCP relays with off-link servers and handle legacy hosts 
doing DAD, then we need a relay-to-server extension. One way to do that 
would be to relay the ND messages. But we might need some additional 
info with those. At least we'd want a response with "OK" vs. "duplicate" 
from the server, which doesn't exist in today's ND. And we'd want to 
have DUIDs for the legacy hosts to fit in with what the server does - 
might make sense creating the MAC address-based DUIDs on the relay and 
passing them along to the server.

Supporting relays is solvable - but it requires some new protocol 
mechanisms.
>
> The relay could even snoop the assigned addresses (which is already done now [?] for other purposes.
The relay would be required to build "registered NCEs" based on the 
assigned/verified addresses from the server, and use those plus the 
associated MAC addresses in the forwarding plane. (And somehow ensure 
that other routers on the link have the same info.)
>
> So, legacy hosts can be assigned addresses w/o hints while registering hosts send in the hint to indicate address registration.
Yep - just requires code change on the hosts.
>
> ...and DHCP already has tackled the problem of restoring state through leasequery.
But here the state needs to to be distributed differently. The servers 
(plural) and all the routers (which might be all the DHCP relays) need 
to have consistent state. The leasequery mechanism allows a relay to 
rebuild its gleaned state from a server.

Would we be comfortable mandating that all the routers on a link also be 
DHCP relays, so they can use leasequery?
> I don't see how you can support legacy hosts without that. We can't assume that legacy hosts do DHCP address registrations for RFC 4941 temporary addresses when the A flag is set. Clearing the A flag means that some legacy hosts will not work on those links.
> Hosts with no DHCP implementation at all?
Even if they host implementation contains DHCP, the DHC client code 
(like ISC's) doesn't have a way to request specific addresses (but has 
the ability to renew IA_TA addresses already part of a lease).
Unless other DHC client code has very different support, I don't see how 
deployed host code can handle RFC 4941 addresses using DHCP.

> There may be something to learn about the sync protocol from work on DHCP sync.  We thought it was pretty straightforward until we got to the problem of simultaneous updates between different host/server pairs.

For address registration it would be better to look at software like 
Zookeeper which does quorum election. The would outside of the IETF 
isn't about simple active/standby or active/active network elements.

Regards,
    Erik