Roofnet

Ah! A recent paper, Mobicom '05. This was a design and deployment of a multi-hop ad-hoc network around MIT. It what is expected, where there are routers for each house and the routers collude to create paths to network gateways. I think it's quite clever and pretty useful. It doesn't quite make sense in developed areas where there are gateways in almost all houses, but it would be nice as a way to backup your connection in case of a local failure.

The paper itself had a few confusing points, particularly on the benefits of single-hop vs multi-hop. They argued each side of that a few times, based (seemingly) solely on what they implemented. Each scheme both increased and decreased throughput in different areas of the paper.

I do think this makes sense in areas with less wired infrastructure. For instance, a user may purchase this to get limited bandwidth through the roofnet system. Then, as they scale up, they can get a deeper wired connection and have it subsidized by other roofnet users.

It's also interesting that they did not hit the wall with scaling the system out. There should be a physical limit to the density of nodes allowed, as determined by the hardware channels available for communication.

There is likely more here, as I found this to be a particularly neat paper, but I'm tired and fairly out of it, so this ends here for now.