Interactive WiFi Connectivity For Moving Vehicles

Summary
This paper looks at a scheme for improving WiFi performance when the clients are in motion, for example cars or buses that will pass by a number of base stations. The authors develop ViFi, a protocol that achieves good performance in these situations.

The paper begins by explaining why the mobile environment is difficult for WiFi applications. Handoff, in which a client changes base stations, is the crux of the problem. Typically clients must perform a 'hard handoff', which means picking a new base station and leaving the old one. The authors evaluate a number of methods for switching but show that none perform as well as a scheme that can opportunistically use multiple base stations simultaneously.

From this observation the paper develops ViFi. The basic idea is that clients pick one base station to be an anchor in much the same way a base station is picked normally. 802.11 broadcast is used, which means no link-layer ACKs, but ViFi introduces its own ACKs. If a base station that is not the anchor overhears a packet but not the ACK for that packet it will relay the packet. The relay is performed over the 'base station backplane', which is either a mesh-network or wired, depending on the deployment.

The authors run a number of experiements in a real deployment at Microsoft, and also simulations based on data from a system in Amherst MA. These show that ViFi can perform much better than standard 'hard-handoff' schemes.

Comments
I enjoyed this paper. The idea seems a fairly obvious one, but the authors present a rigorous analysis showing why using multiple BSs is a good idea, and how much better their system can perform.

Their workload did seem a bit arbitrary however, and it would have been nice to see a more varied workload approach. They also dismiss the problem of packet re-ordering by saying it would be easy to buffer and order packets at the anchor, but we've seen that this can wreck havoc with TCPs RTT estimates and it's not clear how well this would actually work.

Still, a nice approach and a good paper that I would keep on the reading list.