[Olsr-users] Flooding IPv6 Prefix with olsrd2

[Martin Garbe]

On 11/13/2015 02:53 PM, Juliusz Chroboczek wrote:
>> currently I am looking for a possibility to distribute IPv6 prefixes
>> with olsrd2 and auto configure interfaces. I searched for solutions of
>> this problem and found a discussion from 2005. I was assuming that there
>> must be an implemented solution today but did not find it.
> 
> Both AHCP[1] and HNCP[2] are routing-protocol agnostic: while they were
> both originally deployed with Babel, they work just fine with OLSRv2.  (I
> happen to believe that's good design, unlike the autonomous configuration
> protocols that people have in the past tried to build into OSPF, IS-IS and
> OLSRv1.)
> 
> We're still running AHCP, but we're deprecating it in favour of HNCP.
> HNCP is not designed for mesh networks, and there are some reasons why
> Henning and I think it will perform suboptimally in unstable meshes.
> However, I'm opposed to extending HNCP until we get more experimental
> data.
> 
> So please use HNCP in your mesh, and let us know if it works well enough
> (do let us know even if the answer is just "it works fine").  I'm probably
> biased, but I would be grateful if you could deploy shncpd [3] (the other
> implementation of HNCP is called hnetd).

We discussed HNCP in a small group here. At the end we agreed that HNCP
should be used on user nodes which are comparable to CPE devices [1].
In the first place we discarded to use HNCP in mesh because it looks
like overkill for our special problem. The following points were discussed:

- How many messages are generated in a mesh with 500 nodes where the
infrastructure is not static? As far as I know every node knows the
status of all other nodes. Whenever the status of one node changes (e.g.
new neighbour) all other nodes are informed. But especially this is the
task of OLSR and overhead of HNCP here.
- Do we need all this? We only need prefix and address assignment.
- For every local link a /64 net is choosen in Homenet. In our mesh
network we plan to use one /64 for all backbone mesh nodes including all
interfaces. Currently we do not see the advantages to waste that much
address space in our mesh network.
- We plan to use one IP per node. Every node has a loopback interface
with a public IP. Physical interfaces only have local link addresses. Is
this configuration supported by homenet?

Please correct me if anything above seems to be wrong or misunderstood.

Best regards,
Martin

[1] https://en.wikipedia.org/wiki/Customer-premises_equipment