This paper presents expected transmission count (ETX) as a metric for routing in wireless networks. ETX attempts to minimize the total number of transmissions (including re-transmits) required to deliver a packet. It also accounts for asymmetric links. The hope is that a minimum number of transmissions equates to maximum possible throughput.
The authors actually test this routing metric on a real setup in an office building. They compare ETX's performance to to that of min-hop count. In fact, they spend a significant amount of time trying to convince the reader that minimum hop count doesn't always do the best job, which I thought was handled succinctly enough by the first example they presented. I assume they were having to fight very hard to get min-hop count enthusiasts to come over to their side.
ETX works by having each node calculate the forward and reverse delivery ratios along each of its links. Nodes broadcast link probes once every set period of time, and nodes at the other end of said link keep count of received probes. There is a danger of probes not being able to be sent along the link, affecting the neighbors' ETX values.
The evaluation of ETX using the DSDV routing protocol shows that ETX does only a little better than min-hop count in the context of 1 mW radio with small packets. When larger packets (1,386 instead of 134 bytes) are used, ETX doesn't outperform min-hop count, apparently because it overestimates link quality because it uses small packets for the probes, and these smaller packets are more likely to be delivered successfully. Transmitting at a higher power also hurts ETX's relative performance compared to min-hop since nodes require fewer hops to reach one another. I wasn't particularly impressed by the performance benefits of ETX over min-hop count, and especially not since we already read about ETT which was developed by essentially the same group of authors after this metric and seemed to do well enough.
No comments:
Post a Comment