Thanks! It looks like there were a bunch of academic papers about the mesh performance -- if you search for anything by (or citing) Ricardo Carrano, he was one of the more academic engineers who worked on the problem.
For example, here he proposes a hack on top of 802.11s to switch between reactive and adaptive mesh topologies based on measuring the network density -- if you're a small group of kids in a village you want a reactive network, which comes with higher bandwidth saturation costs, but if all the kids come into the same classroom at once you need to quickly switch to adaptive to avoid using up all the spectrum with routing detection:
The 802.11s spec is pretty useless for high density areas without such a hack, it seems. I suppose the takeaway message is that your use of mesh networking needs to be extremely tailored to your specific use case, quite possibly including writing custom routing software.
For example, here he proposes a hack on top of 802.11s to switch between reactive and adaptive mesh topologies based on measuring the network density -- if you're a small group of kids in a village you want a reactive network, which comes with higher bandwidth saturation costs, but if all the kids come into the same classroom at once you need to quickly switch to adaptive to avoid using up all the spectrum with routing detection:
http://wiki.laptop.org/go/MAD_Heuristics
http://wiki.laptop.org/go/MAD_Architecture
The 802.11s spec is pretty useless for high density areas without such a hack, it seems. I suppose the takeaway message is that your use of mesh networking needs to be extremely tailored to your specific use case, quite possibly including writing custom routing software.
John Gilmore also wrote a summary of the OLPC mesh experience here: http://lists.alioth.debian.org/pipermail/freedombox-discuss/...