[Tinyos-devel] Re: from longer CTP/LQI experiments

[Rodrigo Fonseca]

Hi Om,

the results look coherent to me, and it seems that because there were
very few failures, alpha doesn't really matter as the inputs to the
EWMA are almost always the same.
It's interesting from the graphs to look at the settling time of the
network across the runs, from the goodcost over time graphs
(http://enl.usc.edu/~om_p/net2/ctp4bitle/mirage-12212007/goodcostt.html)

This is just qualitative, by looking at node 74 (the black checkered square)

a5   300s
a6 1300s
a7  900s
a7b 1100s
a9   4000s

At alpha 9, it takes a much longer time for the node to settle to a
good path, as the estimator is much more resilient to change. On
TutorNet, however, the agility of the estimator kills on lower alphas.

Another interesting observation stems from the comparison between the
goodcost(t) versus the thl(t) graphs. Take for example a6. THL is
always low, meaning that the high goodcost is due to retransmissions.
This is especially true in the beginning, when most nodes seem to
choose very short paths that are not so great (low THL, high
goodcost). Later they choose slightly longer paths with higher THL but
much lower cost (good links with few retransmissions). On a6, at
around 4000 seconds, there is a sharp rise in THL with a small rise in
goodcost, meaning that longer paths were chosen due to some network
phenomenon, but these were still fairly good (not so many
retransmissions).

>From the THL graphs one can conclude that there is little evidence of
persistent loops.

I shall try to run these tests with lower radio power to see if we
find a more challenging environment.

Cheers,
rodrigo



On Dec 23, 2007 8:26 PM, Omprakash Gnawali <gnawali at usc.edu> wrote:
>
> On Dec 23, 2007 8:10 PM, Philip Levis <pal at cs.stanford.edu> wrote:
> >
> > On Dec 23, 2007, at 7:54 PM, Omprakash Gnawali wrote:
> > >>
> > >>
> > >
> > > Here are the graphs that describe the outcome of the experiments on
> > > Mirage. Rodrigo did experiments with a5,a6,a7 (two runs), and a9:
> > > http://enl.usc.edu/~om_p/net2/ctp4bitle/mirage-12212007/figures.html
> > >
> > > The usual detailed plots are accessible from the links at the bottom
> > > of the page. The subdirectories for time series plots are named
> > > as usual as well.
> > >
> > > It seems like alpha does not change the metrics on Mirage: note the
> > > narrow range on the y-axis for most of the metrics graphs. My
> > > conclusion
> > > is they work equally well and for lack of a trend, if we try to
> > > draw more
> > > conclusion, it might be reading too much into the data.
> >
> > Make sense. When the network is stable, alpha doesn't make a big
> > difference, as the past is like the present. Om, it might be
> > interesting to figure out exactly what's going on in Tutornet that's
> > so challenging about it.
> >
> > Phil
> >
>
> Based on the results thus far, high alpha (9) is good because it works on
> two testbeds. That is the alpha used in 4bitle in the CVS.
>
> We might want to do some tests on motelab too when someone gets
> some time. Maybe just two or three alphas.
>
> Can we say something from the 1Khz RSSI samples from Tutornet
> other than higher noise than on Mirage?
>
> - om_p
>