A high-throughput path metric for multi-hop wireless routing

The main contribution of this paper is in recognizing that choosing a route based on minimum hop-count may not always be the right thing to do, since it does not take into account the link loss rates. Furthermore, doing so might also tend to pick hops with long distances, which is likely to be correlated with high losses. Instead, the authors suggest using a metric that is based on minimizing the expected number of transmissions along the route. This is an additive metric, where the metric of each link is the expected number of transmissions (and re-transmissions) along that link, given by 1/(df*dr), where df and dr are the delivery ratios for the forward and backward directions. Both ratios can be estimated by using regularly broadcasted probe packets from each node.

What I thought was missing from the paper was an intuition of why this expected transmission count (ETX) is the right thing to do, under the assumptions that the radio uses a fixed power, and that the link layer uses re-transmissions as an error control mechanism. Ideally, one would want to choose a path that maximizes throughput. However, ETX seems to be minimizing the expected link latency (assuming no queueing delay). Even then, ETX does not take into account the turn-around time at each hop, although that probably is a second-order consideration. And why not use the expected transmission count of the bottleneck link as the metric? My best guess is that ETX accounts for both the number of hops (in a high hop count route) and the bottleneck (in a low hop count route) in a single easy to describe/implement package.

The use of the experimental wireless testbed is quite convincing, first in motivating the need to move away from a hop-count metric in routing. I'm suprised by the significant assymetry in links; I wonder how "omni-directional" the antennae really is. The CDF also paints a convincing picture of the improvement provided by using ETX.