Interactive WiFi Connectivity For Moving Vehicles

This paper describes a protocol for vehicular WiFi (ViFi) based on utilizing multiple access points. A mobile host chooses a primary base station and several auxillary base stations. When the auxillary base stations overhear a packet to or from the mobile host that is not ACK'd, they probabilistically resend it. The probability of resending is based on the probability of other nodes deciding to resend the packet and the liklihood of reaching the ultimate destination, with a goal of usually resending the (apparently) unACK'd packet exactly once.

The authors evaluated their protocol using a trace-driven simulation (based on loss rate/location traces measured from two real testbed) which the authors apparently validated with measurements on the real testbed. Using their measurements, they compared ViFi to optimal handoff between base stations (with no backups to relay) and to optimal use of relaying (where the best possible base station always did the relying) and showed that they achieved substantially better utilization than using a single base station and nearly the optimal utilization for the relaying strategy. They also tested the application performance of some transport-layer protocols (TCP, VoIP), which was much better with their scheme than simple handoff though they ignored some TCP anomalies (e.g. apparently indefinately stalled connections).

A disappointing artifact of the paper was that the evaluation was apparently performed with very few simulated users (though it is unclear what the DeiselNet trace contained). The authors did not evaluate fairness between users despite a dubious decision to rely on carrier-sense and not attempt to reimplement exponential backoff which using broadcasts had disabled.