Revising buffering in multihop CSMA/CA wireless networks, Olivier Dousse, in Proc. of IEEE SECON'07.
This paper is more a position paper than a technical paper: it presents some new directions to investigate, and some ideas to get thoughts rolling.
It considers a wireless multi-hop tandem network and considers the achieved throughput when using CSMA/CA. The paper observes that for a long line, the packets are accumulated at the beginning, then sparely spaced over the rest of the line. The author then suggests to impose some rules, which are in essence a rate limiting mechanism.
The imposed rules force a structure on the problem which is tractable as a Markov chain. Some of the additional assumptions to optimize the problem are a bit unrealistic (all packets of equal size, all buffers of size one, if a flow is bi-directional, then exactly the same number of packets in each direction) but just bringing up these ideas and deriving the consequences is refreshing.
I have some doubts about the derivations (actually, the maths are only sketched) but the first proposition states that, for N links in series, the average occupancy at the i-th buffer is equal to 1 - the average occupancy at the (N-i)-th buffer. That would force the average occupancy at 1/2 for a buffer where i = N-i. However, if l links interfere, then only 1 buffer is occupied, and the next l-1 are empty, so the average occupancy is 1/l. Anyhow, it is a paper which thinks outside of the box, so this is not very consequential.