Architecture and Evaluation of an Unplanned 802.11b Mesh Network

This paper evaluates the performance of a mesh network deployed in Cambridge, Massachusetts. The network had access points in locations that were convenient for the ‘community’ participants to used and routed between them using a link-state protocol. Using broadcasted probes, loss rate of every link was evaluated at each of the 802.11b supported bit rate (also taking into account the loss rate for ACKs on the reverse flow). From this information, routes (including the 802.11b bitrate at each hop) are chosen to maximize the end-to-end throughput across the network. Experimental evidence showed that the bandwidth prediction from the probes slightly overestimated multihop routes due to interference.

The authors showed that this network did, in fact, benefit in both connectivity and bandwidth from making extensive use of multihop paths because using several shorter links provided much better throughput (for TCP connections) than the (typically longer) paths with a lower hop count. A vast majority of nodes in the network had more than two neighbors in the routed network. Though the link quality (in terms of achievable throughput) was correlated with physical distance, there were plenty of relatively short links with low quality and some highly beneficial long links with high quality.

The authors notably evaluated (802.11b) RTS/CTS versus in their environment. Even though they had observed self-interference when a flow spanned multiple hops, RTS/CTS did not improve throughput.

As a case for creating wireless mesh networks without planning, this paper's result are disappointing. In the multi-hop TCP test 10% could not communicate (despite the nodes of the network being physically stationary), multi-hop routes were much slower than single-hop routes (even considering the penalty of switching off), and adding nodes at random appeared (from simulations) to have seriously diminishing returns.