Re: [Autoconf] Using DHCPv6 without link-local? Support only EUI-64interfaces?

[Teco Boot <[e-mail]>]

Op 2 aug 2010, om 18:55 heeft Ulrich Herberg het volgende geschreven:

> Teco,
> 
> On Sat, Jul 31, 2010 at 3:56 PM, Teco Boot <[e-mail]> wrote:
>> Fred,
>> 
>> Do you mean DHCP relay can be used on a node, that request an address
>> for itself?
> 
> I have tried that a while ago. It works with some limitations (see below).
> 
>> 
>> I think it could work this way:
>> 1) Node queries with link-local to All_DHCP_Relay_Agents_and_Servers.
>> 2a) Node acts as also relay and queries with ULA (site-local) to All_DHCP_Servers.
> 
> Do you mean that a node is DHCP client and relay in the same time?
> That is not possible according to RFC3315, which says (i) in section
> 15.13 "clients MUST discard any received Relay-forward messages" and
> (ii) section 15.3 "servers and relay agents MUST discard any received
> Advertise messages".

I don't think there is such limitation.
A node that combines client and relay would not send out a packet that does
not conform to the spec.


> Also, the relay would need to have a direct unicast connection to the
> central node or use other relaying mechanisms such as SMF (as you
> mentioned below), because multiple relaying is not really feasible in
> DHCPv6 itself: Relaying uses encapsulation, so packets would be
> encapsulated at every hop, quickly increasing overhead.

Yes, relaying and encap on every hop is a bad idea, I think.
But we don't have a standard track multicast protocol for MANETs.
We also don't have a protocol for service discovery, for learning the
DHCP server address.
This makes our work experimental.


>    And I also
> don't think that DHCP relaying allows duplicate packet detection.

Why not?
SMF hashing should work. Each packet has a random transaction ID.
Or include a SMF-DPD header option.


> 
>> 2b) If node is provisioned with DHCP server unicast address, it could use that
>>    instead of All_DHCP_Servers.
> 
> Sure, that is possible if a unicast routing protocol is used.
> 
>> I think this is in line with your RFC 5558.
>> 
>> Drawback of 1: it can result in high number of relayed DHCP packets, in case
>> of many neighbors.
> 
> True.
> 
>> Another drawback of 1: there is a timeout delay when there is no relay or server
>> at one hop.
> 
> But I guess this timeout can be set dynamically?

When 1 is used as a first try, looking for a DHCP-server at one hop, the node 
should wait some time for a response. If no response arrives, it could go to step 2.
The timeout would be a configured parameter, I think.
Maybe it is better to skip such a mechanism. Needs more thoughts.


>> For 2a: the network needs multicast support. Could be SMF.
> 
> Yes, that could be a possibility.
> 
> 
>> 
>> For both 2a and 2b: a temporally used unicast address must be routable. So this
>> DHCP mechanism can only be used as a second step, moving from the self-generated
>> address to a centrally managed address.
> 
> Yes, that seems possible (but I have to re-read the DHCPv6 RFC after
> my vacations ;-)

Not _during_ your vacation ??  :-))


Teco