Interactive WiFi Connectivity For Moving Vehicles

A. Balasubramanian, R. Mahajan, A. Venkataramani, B. N. Levine, J. Zahorjan, "Interactive WiFi Connectivity For Moving Vehicles", ACM SIGCOMM Conference, (August 2008)

This paper talked about ViFi (vehicular WiFi), a protocol designed for providing cheap WiFi connectivity from moving vehicles for common applications such as Web browsing (short TCP transfers) and VoIP.  It was motivated by the fact that clients can overcome many disruptions by communication with multiple base-stations(BS) simultaneously (a concept which they refer to as macrodiversity). The idea is that base-stations that opportunistically overhear a packet but not its ACK, probabilistically relay the packet to the next intended hop thereby minimizing wasted transmissions. This called for designing a new handoff policy and hence required the evaluation of 6 (4 practical and 2 theoretical) handoff policies viz. RSSI, BRR, Sticky, History, BestBS and AllBSeS. ViFi was aimed to be somewhere between BRR and AllBSeS, with the concept of choosing BSeS with high exponentially averaged beacon reception ration and not limiting itself to only one BS. Having multiple BSeS was considered a good strategy mainly because of 2 reasons:

• The vehicles were often in range of multiple BSeS.
• Packet losses were bursty and roughly independent across senders and receivers i.e. bursts were path dependent (due to multipath fading) rather than receiver dependent.

Protocol Design

1. Each vehicle designates one of the nearby BSeS as the anchor (using BRR).
2. All other BSes that the vehicle can hear are designated as auxiliaries.
3. Each vehicle periodically broadcasts its anchors and auxiliary BSeS in beacon packets.
4. Now whenever there is a conversation between the vehicle and its anchor BS, the auxiliary BS overhears it and if the packet is not ACKed for some time, it probabilistically relays the overheard packet with a relaying probability p.
5. If the vehicle moves out from the range of current anchor before it can deliver packets from the internet, ViFi enables the newly designated anchor to salvage packets by contacting previous anchors over a backplane. 

The authors evaluated this over 2 experimental platforms: VanLan (11 BSeS and 2 vehicles in Redmond, WA) and DieselNet (10 BSeS and (?) vehicles in Amherst, MA). Short TCP transfers and VoIP were evaluated since the authors argued that web browsing and VoIP are 2 most commonly used uses of WiFi. ViFi undoubtedly performed better than BRR on all the fronts in both the cases. However, comparing it only with the worst-case BRR leaves many questions about further optimization of this protocol unanswered.

Critiques

Overall, the paper was very interesting. It tackled a real problem and had appropriate implementation to back up the proposal. However, I  am a bit skeptical about some of the following points:

1. I might have misunderstood something, but since the vehicle is moving at 40 Km/h (~11 metres/second), I find it a bit intriguing that exchanging huge control information regarding probabilities of transfer between vehicle and BSes were not a problem. Computing relaying probabilities uses these values continuosly to make decisions. Isn't that supposed to result in a huge overhead of periodic beacons?
2. The authors didn't really consider the effects of reordering of packets and claimed that it would be easy to order packets using a sequencing buffer at anchor BSes and vehicles; however how is this going to affect the auxiliary BSeS (which in turn depends on whether the buffered packets are ACKed or not) remains unanswered.
3. ViFi's coordination mechanism depended on making the expected number of packets relayed to be 1, however, it makes much more sense to use a formulation that makes the expected number of packets received by the destination to be 1.