[Olsr-dev] Same-IP on multiple interfaces problem replication

[Markus Kittenberger]

Am 15.09.2009, 22:42 Uhr, schrieb Jernej Kos <(spam-protected)>:

did you try with assymetric links?

i means bad lq but good nlq
and on the second link good lq and bad nlq

and also attach a 3rd node (which sees both other nodes per wirelless for  
example), and check if its topology view is always correct

if at least the 3rd node think the etx cost between the 2 nodes is for  
example 1 / good lq * good nlq, u have got a problem

but i think current olsrd code will even mix up nlqs of it`s neighbour in  
an 2 nodes only setup

Markus

> Hi,
>
> after this "same-IP on multiple interfaces discussion" I was trying to
> replicate these problems in my test setup but was unable to do so. Two  
> nodes
> (each has same IP on both wired ethernet and wifi) connected by wired
> ethernet and visible via wifi.
>
> By using the netem qdisc I simulated different packet losses so to  
> simulate
> different LQs. In each case both nodes correctly detected the better  
> link and
> properly set up the routes. Since olsrd binds to interfaces and not IPs  
> this
> is understandable as the incoming interface can be uniquely identified
> regardless of its IP address.
>
> So how can I replicate this problem here in order to better understand  
> what
> the problem is exactly ? I understand that 2-hop neighbours (and more  
> distant
> nodes) will not be able to see which interface exactly is the better one
> (because there is no interface identifier in olsr packet format), but I  
> don't
> see why this information is relevant as nodes will simply emit  
> "aggregated"
> information (in this case the ETX of the best link between the two). I  
> assume
> that this latter issue is the real problem here ? In what case will this
> information cause problems for routing and how can I replicate this in my
> setup ?
>


-- 
lg Markus