Network association algorithm

[Aaron Kaplan]

On May 2, 2007, at 3:04 PM, James Cameron wrote:

> I just did a half hour test in mesh mode, with access point off, using
> http://wiki.laptop.org/go/Mesh_Network_Details#Usage and found at  
> least
> four distinct levels detected in this way, depending on the distance
> between two units, with one unit doing a "ping -f" of the other.  The
> thresholds between each level seemed to be logarithmic by distance ...
> in that there were more thresholds the closer the units got to each
> other.

Logarithmic or 1/n^2 ?

Because radio signals get weaker by the square of the distance.
Anyway.

Wanted to say that from the experience from wireless community mesh  
networks - you really want to have
auto power gain. This is essential if you want to keep the mesh "cell  
size" (i.e. the collision radius) halfways at reasonable levels.
The XOs seem to have a very nice good range. But shouting out your  
packets to everybody in that range is like asking for trouble.

 From my experience: i would go for auto power gain. If I am allowed  
to hack into the FW driver, I will.
Well, ok... pseudo code (idea):


  old_num_neigh = 0;

  while(1) {
   if (number_of_stable_neighbors != old_num_neigh) {
     txpow = f(num_neighbors)
   }
  }

where f has to maximize a clique in which each node hears each other  
very well (good SNR) but connects each clique via strong signals to a  
few other cliques.
I would start with f(x) as something like if(x <= 5) { return  
MAXTXPWR; } else return MAXTXPWR*1/(x-4)^2
A minimum of 4 neighbors is always a good thing to have.

Which brings me to another point: do we have a 802.11s simulator  
somewhere?




ciao,
aaron.

---
there's no place like 127.0.0.1