[Olsr-users] Question about hidden node behavior

[Mitar]

Hi!

So we have found a bug in our monitoring code which was parsing LQ in
ILQ swapped. Hurray for the theory.

OK. So the story is now like this: once we added trzaska-12 node
(which has a good link in both direction to glinska-9 node), packets
from fri node to glinska-9 node start getting lost. Probably
trzaska-12 and fri node does not see each other transmitting.

The question is why only in one direction packet loss is increased. So
why packets from trrzaska-12 to glinska-9 are also not lost when fri
transmits at the same time.

I added RTS/CTS of 200 bytes on all three nodes and there is no
perceived improvement of users using it to access the Internet. ETX
has not changed (as it uses broadcast/multicast). Ping packet loss
from the nodes to the central server seems to have even increased
(maybe just for larger than 200 bytes packets, I will need more data
to see if it is just a temporary fluctuation).

Maybe we could also try fragmentation.


Mitar

On Thu, Jan 6, 2011 at 8:23 PM, Mitar <(spam-protected)> wrote:
> Hi!
>
> We have such setup. Three nodes in a line: fri - glinska-9 -
> trzaska-12, on same channel, ad-hoc, where fri and trzaska-12 does not
> see each other. Between fri and glinska-19 nodes there is ETX 1.0 and
> between glinska-9 and trzaska-12 nodes there is ETX 1.0, when those
> nodes operate independently from the third node. But thing gets
> interesting because fri and trzaska-12 does not see each other.
>
> So what we noticed is, that link from fri to glinska-9 started to
> deteriorate once we added trzaska-12 node:
>
> https://nodes.wlan-lj.net/graphs/1559/year
>
> This is graph of link quality to fri node from a perspective of
> glinska-9. As you see ILQ started to drop, that is the ratio of
> packets that are successfully sent from the glinska-9 to fri node.
>
> If we look at graph of how much WiFi traffic does trzaska-12 produce,
> we can see a correlation:
>
> https://nodes.wlan-lj.net/graphs/3842/year
>
> So trzaska-12 produces more and more traffic (users get used to it,
> more users find it available, more users are using it...).
>
> And now the question. Why is ILQ deteriorating and not LQ. I would
> assume that what is happening is that both fri and trzaska-12 (because
> they do not see each other) send packets at the same time to glinska-9
> node. So that LQ from fri on glinska-9 node should be low. Why is
> opposite? Why would fri node have problems receiving from glinska-9
> because of the trzaska-12 traffic?
>
> What is interesting is that ETX between glinska-9 and trzaska-12 is
> all the time around 1.0. That means that trzaska-12 does not have
> problems sending anything to glinska-9 node, even if (by my
> assumption) there are packets conflicts between fri and trzaska-12.
>
> (I am long suspecting that in nodewatcher we have ILQ an LQ turned
> around, but after many attempts to find a bug, it seems that code is
> correct.)
>
>
> Mitar
>