[Olsr-users] TCP flow in a 5-hops wifi network - experiment
Krzysztof Szuster
(spam-protected)
Thu Jul 19 10:21:03 CEST 2012
Dnia czwartek 19 lipca 2012 07:21:35 Marco Mezzavilla
<(spam-protected)> pisze:
>
> Dear all,
>
> I've correctly deployed my multi-hop wifi network, and I put OLSR into
> running.
> I am evaluating TCP transmission flows by means of IRPEF+JPERF (such useful
> tool!).
>
> My question is: have any of you ever experienced the following behaviour?
> The TCP throughput falls to zero because OLSR TC messages collide, thus
> provoking IP entries losses.
> Hence, the path is corrupted, and the TCP transmission stops for few
> seconds, until new TC messages are correctly received by all the nodes..
Exactly. Actually the fenomenon is much more likely if you use UDP traffic. You
also don't need 5 radios in line, two is enough. The problem is that if a
radio router has much more packets to send than wireless link can consume,
some packets are discarded by the router. If an olsr machine is flooded from
ethernet at, say, 100Mb/s and radio output is 10Mb/s, then 90% of packets will
be discarded. Not only iperf packets, but olsrd packets too, since they all
share that same path. olsrd will definitely drop routes if the packet loss is
so high.
My solution was to send olsr packets with higher priority, than other traffic.
Use Traffic Control with prioritized queues on radio interface. You can configure
olsrd to set TOS for outgoing packets.
Krzysztof Szuster
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.olsr.org/pipermail/olsr-users/attachments/20120719/5cc1676a/attachment.html>
More information about the Olsr-users
mailing list