WiLDNet: Design and Implementation of High Performance WiFi Based Long Distance Networks

Original paper here: http://www.cs.washington.edu/education/courses/590l/07sp/wildnet.pdf

Okay, my initial impression of this paper was that it was kind of cute, but not really groundbreaking. After reading some of the background, namely the 2P paper, I think most of the cool work was done there, and it's not entirely clear what new and interesting things happened here. (Side note: I really liked the 2P paper if for no other reason than they presented the idea very clearly. I got it almost immediately.)

The environment they're working in is long-haul wifi. We're talking about standard 802.11 used in a long-distance environment with point-to-point directional antennas. Specifically, a node has a different antenna for each communication partner. The problems in this environment are:

• the different antennas at a node can interfere with each other
• the propagation delay can be large enough to trigger link-layer timeouts
• the 802.11 specification mandates stop-and-wait semantics for frame acks (which take forever on those long links)
• external interference can wreak havoc
  

Note that in contrast to short-hop mesh nets like RoofNet, these guys assert that the cause of interference in their system is not multi-path but instead simply other 802.11 transmissions.

The solutions to these are fairly straightforward. Like 2P, these guys use time-slicing (TDMA) to eliminate collisions. They implement a sliding window ack mechanism to deal with the long propagation delay. Finally, they add erasure coding to cover loss. An interesting tweak to the erasure coding mechanism is that they send the packets they want to get there in the clear first, and then they send the "backup" packets afterwards. This eliminates the necessity of decoding the packets if there is no loss.

That's about it...they do a nice examination of the design tradeoffs of things like overhead versus loss masking for both erasure coding and retransmissions.