[Babel-users] fun with babeld, and tools for measuring ETX for Babel meshes

[Juliusz Chroboczek]

> Anyway, at 3:30pm (and several dips in the pool... it's 40C today) I was
> ready to head out, and my assistant and I went out around the block.  Please
> see picture gallery at:
> https://plus.google.com/photos/103865510556691933694/albums/6031203377916425729?authkey=CJSkqIOWkJ3t8QE

Heh.  Your assistant seems to be enjoying a pleasant work atmosphere.

> The 3800 has two radios: and b,g,n radio, and 802.11a radio.  I think of
> it as the 2.4Ghz radio and the 5Ghz radio, but my understanding is that
> 802.11n uses both 2.4Ghz and 5Ghz.

Not quite.  The 3800 has one bgn radio at 2.4, and one an radio at 5.
n is an extension to 802.11 that applies to both bg and a.

> One thing you can see in the screen shots is that at a certain point (slide
> 11 of above) the intermediate hop is no longer :82::1 (which is the 5Ghz
> interface) to the :8f::1 (the 2.4Ghz radio).  I find mtr unclear when/if it
> switches back.

I think that you missed an opportunity to run BabelWeb.  Your assistant
would love it.

> The best would be to get the RX power right from the radio.  Does babel
> currently get any info that way?

No.  Contrary to pretty much everyone's intuition, RSSI is not a good
predictor of packet loss.

Now the kernel is already measuring stuff in order to choose a rate.  The
current plan is to ask the gentle kernel for its conclusions, and pick
a metric based on that (idea due to Dave).  The data is already being
exported by netlink in recent kernels (code due to Antonio Quartulli), so
it should be a fairly simple thing to do.

> In the ROLL/RPL/contiki space, I have been considering a NETCONF/YANG
> way to get the mesh adjacencies (in particular, knowing the one which were
> not chosen by each node) via a CoAP interface.

An explanation would be welcome.

-- Juliusz