65-node simple mesh test (and counting... ;-)

[Marcus Leech]

Marcus, this is indeed an interesting idea. However it has a significant
problem: wiring up more than 60 XOs onto a switch requires equipment,
time and space that OLPC cannot presently provide. Such a testbed though
is absolutely necessary not only as a proof of concept for your
suggestion, but also for doing large scale mesh network testing in
general.

Indeed, it does require more infrastructure than 60 Xos in simple mesh.
My hope is that the
  XO lab we're building here will have enough infrastructure in it to
make such scenarios
  viable.


>
> The common, but erroneous, assumption is often made that a wireless 
> network is just like a wired network, but with the wires removed.
>   

So very true!

> On a wireless network, broadcasts are successfully received with much 
> lower probability.  RF is mysterious and magical, and all sorts of 
> connection asymmetries, near-field effects, and radiation lobe 
> patterns conspire to make it unlikely that *everyone* can hear you 
> equally at once -- and then you get into remote collisions and other 
> mechanisms that make you unaware that not everyone heard you.  And 
> there is not 'ack' mechanism for 802.11 broadcast.
>   

All these are true also, but I think we're mystifying things a little
bit here. The wireless medium is unpredictable mainly because its
properties are also a function of time (a non-issue in wired networks),
but at least (thank God!) it [the wireless medium] does not discriminate
between broadcast and unicast frames! Adding an ack scheme to broadcasts
should yield equal (or even better due to lowered speed) reliability
using broadcast frames. Even without the ack scheme, I noticed that, on
average, some 95% of the data transmitted over broadcast are
successfully received on all nodes. We are throwing this away by
discarding it on our wireless interfaces.

Pol


I was playing in packet-radio circles long before 802.11 was even a
gleam in anyones eye :-)

We had to deal with hidden-terminal issues, non-uniform propagation,
etc.  The purpose of the
  experiment I proposed above (measuring the ratio between a Cerebro
network "equilibrating"
  over both a wired and a wireless network).  Yes, there will be
differences, but if they're
  *large* compared to the raw bandwidth ratios, then something isn't
working right, particularly
  if all the Xos are in the same room.  You shouldn't have
hidden-terminal issues.  Yes,
  there will be laptops that are in the "null" of another laptops
radiation pattern, but
  in terms of absolute received power, even being in a "null" (unless
it's a *very* deep null
  indeed) shouldn't dehance the SNR so as to not be able to coherently
receive bits.  The
  other thing that I wonder about is the collision behaviour in real
life of an 802.11 network.
  I understand that the network uses a Collision Avoidance (CA) scheme,
but I wonder how
  effective it is in real life.  Back in my packet-radio days, we moved
from a pure
  CSMA scheme to one that used P-persistant CSMA, with static
determination of P
  values.  This vastly improved overall throughput, and made collisions
more rare (not
  zero, but a lot better).

How does the collision model/scheme change between AP mode and
ad-hoc/mesh modes?

Cheers
Marcus