Architecture and evaluation of an unplanned 802.11b mesh network

As its title suggest, this paper discusses an implementation and evaluation of a wireless mesh network. I like the overall presentation; its clear in explaining what was done, actually carried out the system experiments and show plenty of experimental results. Despite that, I do have some concerns.

First, the assumption that "community wireless networks typically share a few wired Internet connections" is a little dubious. Perhaps there might be good reasons to back up such an assertion (maybe in a rural or undeveloped setting??), but these were not made clear. On the other hand, a deployed mesh node requires power at the very least, and in the urban setting where experiments were carried out (deployment in buildings), it is hard to imagine that the nodes cannot have access to a wired internet connection. I suppose there could be a cost pressure in having wired internet connection, so it could make sense to reduce that number. In that case, it should be make clear, and the trade-off between cost of wired connections and internet access throughput could be studied.

Second, and related to above, the experiments measure throughput between pairs of mesh nodes. While the results are well and good, it doesn't answer what users would care most about, that is, how fast internet access is. Given that wired connections are limited, the gateways represent a bottleneck to the internet. This point is important because the key assumption of the study is the limited number of nodes with wired internet access.

Third, it was not clear if throughput was measured simultaneously or one at a time. Section 3.7 mentions inter-hop interference, but that seems to be due to intermediate nodes performing hops on the same path between two sender/receiver nodes. Given that a typical use case would have at least some number of nodes trying to communicate with the gateway, it would have been nice to see how average throughput varies with the number of simultaneous communications.

A minor note on the effect of density (section 3.4). It would have been interesting to see how the empirical density for full connectivity compares with known theoretical results for full-connectivity in random graph theory, given the maximum range of the communications.

Now, to end on a positive note. I think the routing protocol presented here is useful not just for wireless mesh networks, but would also be useful for an overlay network. I'm not sure how routing for overlay networks work, but I'm guessing it won't be too different.