[olsr-dev] olsr results in intermittent rate interrution?
Thu Sep 1 08:40:03 CEST 2005
You don't describe your test set up at all, but my team did a lot of testing
of MANET protocols under stress (we used TBRPF) turned up a couple of
conditions that gave us the same reactions.
1) If you congest the radio link you're going to start dropping routing
protocol packets, drop enough and the route goes away. This is especially a
problem if your ingress ports are very fast - we tested on 266MHz Geodes
(Soekris 4801) with 100Mb/s full duplex Ethernet feeding 802.11a/b/g radios.
On a good day we would get maybe 20Mb/s through the radio, so ignoring a lot
of stuff, right off the bat you can see that a lot of packets aren't going
to make it through the radio.
2) If you're using low powered hardware (i.e. Atheros 802.11a running on
anything less than 600-800MHz) we also encountered a problem where under the
load of a full-bore Ethernet that the kernel spent all it's time throwing
packets away and *not* giving time to the routing daemon running in user
The above can be greatly aggravated by your test equipment - we used Ixia
which can do all manner of mean things like violate interframe spacing to
acheive utilization > %100. Doing this to an ether->802.11 link was very
hard on the routing protocol which was sending relatively few packets
compared to the test traffic. We got the best results by running the traffic
generator at some level less than the expected throughput of the 802.11.
This is a little unrealistic in that you can't really control the traffic,
except maybe by implementing the source quench message (I don't know if
Linux can do that).
Testing throughput is also complicated by the "tuning" of the protocol
parameters. So a network tuned for static stations would perform better
under heavy load than one tuned for high mobility. When I was looking for
good throughput numbers/characteristics I just set the protocol to be very
unrepsonsive. Networks tuned for high mobility are almost always going to
suffer from instability under load - that's just the way things work.
Ok, I just checked and you say you're using two ethernet interfaces, I think
the discussion still holds, but not as pronounced as when there's a great
disparity in speed between the interfaces.
On 8/31/05, Chang Jiang <(spam-protected)> wrote:
> Hi all,
> When I measure the end-to-end traffic rate of terminals in my ad-hoc
> with OLSR routing, I found the traffic rate suffers intermittent
> interruption. The phenomenon disappeared after I replace the OLSR routing
> with static routing (add route in the routing table).
> What's the reason? Thanks.
> Two nodes (with 2 NIC -eth0,eth1): Node1, Node2
> eth1 in each node runs the olsrd.
> eth0 works as the accessing NIC.
> Two terminals: terminal_1, terminal_2
> terminal_1 access to the eth0 of node 1;
> terminal_2 access to the eth0 of node 2;
> 1. Redhat Linux 9.0
> 2. olsrd 0.4.9
> terminal_1<--->|eth0 eth1|<------>|eth1 eth0|<------>terminal_2
> Node1 Node2
> olsr-dev mailing list
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Olsr-dev