dhcrelay over wireless mesh

Luca Tavanti luca.tavanti at iet.unipi.it
Wed Jul 11 14:06:20 UTC 2012


Randall, Simon: thanks for your replies and sorry for not having 
provided a better description of my "scenario".

Actually, the "mesh" network I'm working with is more appropriately a 
multi-hop wireless network (a 802.11 network working in ad-hoc/ibss mode).
In short, each node can communicate only with its neighbours, and data 
packets are "routed" towards far nodes by means of intermediate nodes. 
To route the data a proper layer-3 routing protocol is at work on every 
node (OLSR).
However, OLSR does not seem to route BOOTP packets.
I could expect this, as in the README of the ISC DHCP distribution I 
found that "we completely bypass the Linux IP stack" (btw, does this 
means also that iptables won't work with the ISC dhcp suite?).

Now, my setup is as follwos:

((( 0 ))) ((( 1 ))) ((( 2 ))) ((( 3 )))

I put the dhcp server on node 0, and both a dhclient and a dhcrelay on 
nodes 1,2,and 3.
In this way both nodes 1 and 2 gets an IP from the server (directly for 
node 1, through the relay on node 1 for node 2).
Yet, node 3 never gets an IP.

I hope now I've made the problem clearer...

Luca


On 11/07/12 15:42, Simon Hobson wrote:
> Randall C Grimshaw wrote:
>> The mesh network should not be introducing any routing, so putting a
>> relay on each of the nodes is going to introduce a magnifying effect
>> that will eventually contribute to a DOS. You only need one relay per
>> layer-2 network and that is usually done on the router with the
>> layer-3 interface.
>
> What I know about mesh networks could be written on the back of a stamp,
> without using exceedingly small text ...
>
> But I think they are not a layer 2 network in the sense that we are used
> to dealing with it. For that to work in the massively interconnected
> mesh, it would involve some very inefficient packet routing by the time
> redundant links had been disabled. In fact I doubt that something like
> STP would even reach stability on many networks.
> I think they tend to do some sort of hybrid where the network "sort of"
> looks like a layer 2 network to devices, but is in fact more like a
> layer 3 network in terms of how it routes packets.
>
> So I assume they probably filter broadcast traffic in some way, hence
> running relay agents.
>
> Another thought is that perhaps once the request goes through more than
> one node, a second agent picks it up and incorrectly mangles it -
> meaning that the reply doesn't go back to the correct node.
>


More information about the dhcp-users mailing list