[Sugar-devel] ESSIDs and BSSIDs, NM and Sugar

[James Cameron]

On Tue, Aug 11, 2009 at 11:49:21AM -0300, Andr?s Nacelle wrote:
> First of all thanks James for taking your time and reading the stuff I
> send and giving me such a complete response. Getting into what you
> were saying, I'm aware of the high dynamics under the signal sensing
> and reception, because of that I managed to run the test in a big
> storehouse 65 meters long and 35 meters wide, completely empty (except
> because of me and the equipment). In addition is in a place where
> there's almost no interference from other wireless devises. This way I
> tried to minimise the influence of variables out of my management.

Good, thanks.

> The concept of the XO taking the decision to connect to the "best" AP
> is what we are chasing. But the definition of best is not that easy
> and the behavior of the XO is not that as desired. What I'm saying
> here is that best AP also means no overloading a single AP when there
> is another one nearby with a reasonable quality.

I wonder if the wireless firmware on the XO can determine the load on an
AP.

> In the other hand the XO doesn't stay in the same AP, it changes all
> the time even when there's 14 dBm power difference between each signal
> instead of staying connected to the same AP. There should be a way to
> make those connections more stable.

I'd be more interested in the samples of signal level captured by
iwconfig as an XO moves from one AP to another.

> I've been using two different ways for doing this. The first is
> running 15 times the iwlist eth0 sc, and doing the average. This I did
> it in different regions of the testing area. [...]

Okay.  I've compared the scan results against the associated results,
and they tend to be different and slower.  They are different in that
the scan results are optimistic, by up to 10 dBm.  They are slower
because the scan takes much longer.

It would be interesting to test whether the XOs migrate from one AP to
another if they were not regularly scanning.  To test that, on a unit,
manually stop Network Manager, associate with the AP, obtain IP, and
then monitor.

	# /etc/init.d/NetworkManager stop
	# iwconfig eth0 essid quozl.linux.org.au
	# dhclient eth0

(the last two steps take four seconds for me, but six to eight seconds
for Network Manager).

As an aside, the program ssm.py that I wrote yesterday can be used to
demonstrate an odd behaviour ... close the antennae, then place a hand
over one antenna then the other.  The variations in signal level and
link quality suggest that the wireless chipset is preferring the
left-hand antenna, which is coincidently the one with the shortest
internal RF cable.  It also less frequently tries the other antenna.  (I
wish more data was exposed).

-- 
James Cameron
http://quozl.linux.org.au/