Architecture and Evaluation of an Unplanned 802.11b Mesh Network

This paper presents the design and evaluation of an ad-hoc wireless mesh network between residential users. There are a lot of details, but the main conclusion is that it is possible to have reasonably good performance using an automatically managed network deployed by users in an ad-hoc fashion with no central oversight. The largest design challenge was probably the link quality measurement and routing metric.

I liked the ETT metric, which combines information about throughputs and loss probabilities for both big data packets and ACKs without making many assumptions about the link (e.g. constant error rate per bit). I also found the multihop interference very interesting - it seems like there should be a way to fix that. (But then, why have they disabled RTS/CTS?)

I also liked the fact that this is a real implementation, and not a simulation, like the previous papers! Some of the comparisons are still calculations based on simulation, but at least they are using measured values for throughput, latency, etc. I actually think this wireless portion of the course could be improved by having more experience papers about real systems, if these are available.

I had a number of questions about the paper that weren't really answered:
- Why is RTS/CTS disabled in the wireless nodes? This is especially strange because it may be a cause of the multihop self-interference effect seen.
- How much more powerful are the antennas on the Roofnet nodes compared to a "standard" wifi router? Can they interfere with peoples' WLANs at home?
- How does the placement of the directional antennas affect performance?
- Was Click performance good enough to never be a bottleneck?
- How often do route metrics change, and how often do they lead to changes in preferred paths?
- The system recovers from lost nodes or degraded link quality, but can it recover from a complete restart, as would happen for example with a power failure?
- I was very surprised that ISPs actually allowed wireless sharing of links. Is this still true today? I imagine ISPs would be okay with it if they could charge extra, and if we do that, what is a good pricing model to use? This also leads to a general question about what the "market demand" for systems like Roofnet is. Sure, Roofnet is very cool technology, but what is the economic problem that it solves? Could it be used to set up a lower-cost "local ISP" than standard ISP's?