Karl Garcia (from Google) discussed the project in rolling out WiFi across Mountain View. It sounds like a fascinating project; the idea is to bring information to people by making a public free access network (albeit with a Google login as required by the legal department) in Mountain View. The purpose of the network rollout is as a trial to find out what works and the way in which such networks may be rolled out in the future (for example, when rolling out a network across San Francisco), as well as to investigate other things such as location-based services.
The network is based around three main base stations, which are connected via fibre and point-to-point wireless network (for redundancy), which also have a point-to-multipoint wireless connection to 60 gateways distributed across the Mountain View. Those gateways are then connected (directly or indirectly) to another 320 mesh radios across the area. Given that the mesh radios are self configuring, they can be added in places where radio connections are weaker, and the will automatically configure to route back to a gateway (and from there, back to the base stations). The network uses 802.11g to communicate; each of the mesh radios have encrypted connections, but the end points to computers are open.
The mesh routers are located on top of light poles; primarily because they are high up (therefore have better views for radio signals) and also because they've got power sources nearby. Unfortunately, it's not always clear who owns the street furniture; for example, El Camino (the main road running through Mountain View) is a state highway, and as such, the street furniture is owned/managed by the state of California. So there's a different group of people to speak to in order to install that WiFi furniture there.
In order to test the wireless network, there were a handful of test devices based on either the Linksys router or Sharp Zarus PDAs that can have Linux flashed onto it. With built in batteries, extended battery packs and possibly solar panels, it was possible to set up miniature Linux clients that would connect back to a central Google server and receive commands to execute. The data was processed with LiveHTTPHeaders to find out what the connectivity was like; also, using performance measuring tools like
nuttcp); unfortunately, the commercial tools (ixChariot) didn't support some of the devices. These OSes then connected back and execute server, as well as heartbeat and extra command channels for remote processing and recovery. The devices were then moved throughout Mountain View, and to see what the behaviour was like with devices in buildings (at least, near to the edge of walls).
In all, it's a fascinating experiment and the next time that I'm out in CA, I'll be trying it out. It will also be interesting to see what happens with the San Francisco roll-out, and whether anything from that will be used to work within other locations around in the world (e.g. London :-)