[net2-wg] 3-node test at different rates

[Philip Levis]

On Wed, 2006-06-14 at 12:56, Omprakash Gnawali wrote:
> I haven't been able to use the testbed to do large scale test because
> there seems to be some problems with the testbed at the moment.
> 
> I testbed collection with 3 Tmotes on my desk using different rates. I
> am having issues with the serial toolset so I am not using
> Dissemination to change the value - I just recompile with different
> rates. I changed the rates by changing the argument to
> Timer.startPeriodic() in TestNetworkC.nc to the following eight rates:
> 
> Timer.startPeriodic(60)
> Timer.startPeriodic(50)
> Timer.startPeriodic(40)
> Timer.startPeriodic(30)
> Timer.startPeriodic(25)
> Timer.startPeriodic(20)
> Timer.startPeriodic(15)
> Timer.startPeriodic(10)
> 
> For each rate, I log the packets for 100 seconds. Then I plot the
> number of packets received from each node every second and also total
> packets for each second:
> 
> http://enl.usc.edu/~om_p/net2/
> 
> Two observations:
> 
> - The rates seem more or less fair even at high rates.
> 
> - The aggregate rate seems to be peaking at about 80-90 pkts. Is this
> expected behavior?

It might be that as it approaches 80 packets per second, packet losses
cause it to slow down (e.g., when a packet is not acked, the forwarding
engine backs off for 128-255ms). My only recommendation is to use a
packet sniffer and observe what's going on. Also, the sequence numbers
of the packets would be useful here. I'm going to modify TestNetworkC so
it generates sequence two sequence numbers; one is incremented on
successful sends, the other is incremented on each send. That way we can
see if packets are being throttled at the application, if there are
network-level retransmissions, etc.

Phil