I've always lumped this under the "too idealistic" startup destined to be a nonprofit. Rural America is broke, for the most part, and our margins were going to be razor thin -- even using commodity wifi gear. I only ever got one angel interested, mainly because they all thought the market was bad. We never really got it off the ground. Perhaps skipping the backhaul altogether, (like the linked article), would have made it viable. But honestly I'm not sure I could have sold faux-internet, which is what most people are going to think of this. As it was the only way we were able to get a town interested was to uncover grant money that would have covered most of the costs and let us subsidize prices (see non-profit thing above).
But there were other challenges. Mainly wifi is just not that great over big distances. And it's absorbed by water -- so trees are a big problem. Also one of the most requested apps was VOIP, and the latency on a mesh network makes VOIP kinda crummy. As we built out the financial model, we learned that there was a lower limit to the density of a community that would support a profit. It ruled out most markets. And to make matters even worse, when we did find communities that met our already ridiculous criteria we usually had problems finding a backhaul. For our pilot town we were going to create a point to point wireless connection over 20 miles from the nearest major population center. Yeah, we gave up, graduated, and got jobs.
It's also worth knowing that wifi in major urban centers doesn't really work. There just isn't enough frequency to go around, so the interference makes it pointless. At the time Clearwire (now just Clear) was buying up spectrum in city markets. I thought it was genius, but I always doubted the wimax standard would live up to the hype. I remembered we had a point-to-multipoint wireless 56K connection back in the mid 90's and that was shit. I figured things would get better, but the basic problem remains the same: you can only really jam so many connections into one point, and you need LOS. Of course cell phones seem to work fairly well, so I probably don't know what I'm talking about. But Clear doesn't exactly have a stellar reputation.
Alright, so here are some links to other mesh networking projects that have been around for a long time and have software that is actually deployed:
http://pdos.csail.mit.edu/roofnet/doku.php (seems to be slightly broken, but working mostly)
http://www.cuwireless.net/ (interned here, really smart guys built this and they have a running network in Champaign-Urbana).
Also recently the feds opened up a bunch of spectrum, so I'm really hopeful about what we can do with it. 2.4 just isn't enough.
Edit: Highly recommend following @saschameinrath He's a genius and has been a champion of community wireless internet since the term was invented. He also started CUWIN.
I'm not sure how you dredged this up, but I'm flattered. Maybe someone taking a whack at a similar project can learn from our approach.
- Michael, OmegaSDG
But fundamentally, they are using VBR codecs that don't even live up to the standard 3.1 KHz PCM bearer spectrum of a fixed-line 64kbps DS0 channel. This is what, maybe 8 kbps (extrapolating from the relatively high quality to bandwidth ratio of something like G.729A, though they're obviously not using that)? That's hugely different than providing multi-megabit access to every endpoint, even with severe oversubscription and statistical packet multiplexing / low contention ratios on the table. This is one of the reasons why 3G network operators are so freaked out about tethering and data usage in general.
Like you said, there's only so much frequency. This "the world's gonna go wireless" stuff is a pipe dream. The enthusiasm for all forms of fixed-line communication will return with great fanfare once application-level bandwidth requirements increase an order of magnitude or two beyond where they are now, because fixed-line transmission is the only thing that can keep up at that point.