Interactive WiFi Connectivity For Moving Vehicles

1. Interactive WiFi Connectivity For Moving Vehicles Presented by Zhou Yinggui

2. Background 1 the ubiquity of WiFi 2 Cellular networks tend to be expensive

3. Background Can WiFi deployments support common applications from moving vehicles ? Vi-Fi

4. Outlines 1 Introduction 2 Experimental platform 3 Handoff strategies 4 ViFi design & implementation 5 Evaluation 6 Conclusion

5. Introduction ViFi is a protocol that minimizes disruptions to support interactive applications. Hard Handoff --Clients communicate with only one basestation at any given time.

6. Hard handoffs are limited by 1.gray periods --- connectivity drops sharply and unpredictably; 2.the difficulty of estimating the continuously changing channel quality. Introduction • Hard handoff methods hasfrequentdisruptions even when clients are close to WiFi basestations.

7. Diversity--using multiple basestations simultaneously. Introduction • ViFi exploits diversity and opportunistic receptions to minimize disruptions.

8. Experiment Experiments are performed on two vehicular mobile network testbeds: VanLAN and DieselNet. DieselNet--Downtown of Amherst, Massachusetts trace-driven study VanLAN--Microsoft campus in Washington. live experiments.

9. Experiment VanLAN • Eleven basestations (BSes) • Two vehicles equipped with Atheros 5213chipset, omnidirectionalantennae , • GPS unit . • The box covers 828×559 m2 area. • in which at least onepacket is • received by vehicles from any BS. • Not all pairs of BSesare within wireless range of one another. • The vehiclesshuttle around the campusten times a day, witha speed of40 Km/h.

10. Experiment DieselNet • The equipment of vehicles are the same with VanLAN • log all beacons heard from nearby BSes • traces from Channels 1 and 6 • log more than 100,000 beaconsin 3 days • Only analysis to BSes in the core of the town and to BSes • that are visibleon all three days

11. Handoff strategies Using a trace-driven evaluation on VanLAN Each BS and vehicle broadcasts a 500-byte packet at 1Mbps every 100 ms. • criterions: • aggregate performance--total number of packets delivered and • thetotal time or distance… • periods of uninterrupted connectivity--contiguous time intervals • when the performance of an applicationis above a threshold.

12. Handoff strategies • hard handoff • RSSI:client associates to BSes with higher signalstrength. • BRR:client associates to the BS with the highest strength. • Sticky:client does not disassociate from the currentBS until connectivity is absent. • History:client associates to the BS that has historicallyprovided the best average performance. • BestBS:not practical, represents an upper bound of hard handoff. • diversity methods AllBSes:client opportunistically uses all BSes.Ideal method that represents an upper bound of any handoff protocol

13. Handoff strategies Aggregate Performance Average number of packets delivered per day in Van-LAN by various methods. pick BRR as representative

14. Handoff strategies Uninterrupted Connectivity adequate (uninterrupted)connectivity: at least 50% of the packets are received in a one-second interval. black dot (gray periods)：areas that connectivity drops sharply and unpredictably go to 34

15. Handoff strategies Uninterrupted Connectivity • connectivity is often undermined by gray periodseven close to BSes. • gray periods tend to be short-lived, and do not severely impact aggregate performance.

16. Handoff strategies Uninterrupted Connectivity other definitions of adequate connectivity

17. Handoff strategies The insight of losses • upstream direction lossesare roughly independent across BSes and a packet sent by the vehicleis received by at least one BS with a high probability. the conclusion has been shown previously by S. Biswas and R. Morris. ExOR: opportunistic multi-hop routing forwireless networks. In SIGCOMM, Aug. 2005.

18. Handoff strategies The insight of losses • downstream direction gray period:even when a vehicle is associated to a BS with a lowaverage loss rate, it can lose many packets in a small time period,hurting interactive applications.

19. Handoff strategies The insight of losses • downstream direction most burst losses are path dependent rather than receiver dependent.

20. ViFi protocol environment • Diversity: A packet sent by a moving vehicle can often be • heard by multiple BSes, and multiple BSes can oftendeliver • packets to a moving vehicle. • Bandwidth-limited inter-BS communication

21. ViFi protocol imposes minimal additionalload Motivated by AllBSes does not increase per packet latency can handle rapidly changing sets of BSes.

22. ViFi protocol Protocol overview • the vehicle designates one of the nearby BSes as • theanchor(by BRR). • The vehicle designates other nearby BSes as • auxiliary. • The vehicle embeds the identity of the current • anchor and auxiliary into the beacons which are • broadcasted periodically. • The vehicle also embeds the identity • of the previous anchor for salvaging.

23. ViFi protocol Protocol overview 1. src transmits the packet P. 2. If dst receives P, it broadcasts an ACK. 3. If an auxiliary overhears P, but within a small window has not heard an ACK, it probabilistically relays P. 4. If dst receives relayed P and has not already sent an ACK, it broadcasts an ACK. 5. If src does not receive an ACK within a retransmission interval, it retransmits P.

24. ViFi protocol Protocol overview 1. src transmits the packet P. 2. If dst receives P, it broadcasts an ACK. 3. If an auxiliary overhears P, but within a small window has not heard an ACK, it probabilistically relays P. 4. If dst receives relayed P and has not already sent an ACK, it broadcasts an ACK. 5. If src does not receive an ACK within a retransmission interval, it retransmits P. Upstream packets are relayed on the inter-BS backplane and downstream packets on the vehicle-BS channel why relaying is better than a retransmission ?

25. ViFi protocol Computing relaying probability balance the trade-off between false negative(no diversity) andfalse positive(excessive load) • The guidelines of probability computation • G1: Account for relaying decisions made by other potentially • relayingauxiliaries. • G2: Prefer auxiliaries with better connectivity to the destination. • G3: Limit the expected number of relayed transmissions.

26. ViFi protocol Computing relaying probability The overall strategy is to compute relaying probabilitieslocally and the expected number of packets is equal to 1. Ci is the Bi’s probability that Bi has heard the packet but not anacknowledgment ri is Bi’s relay times （relay probability）which is less than one.

27. ViFi protocol Computing relaying probability Pab represents the probability that b correctly receives a transmissionfrom a

28. ViFi protocol Computing relaying probability Ci =Psrc,Bi(1 − Psrc,dst Pdst,Bi ) Ci is the Bi’s probability that Bi has heard the packet but not anacknowledgment(assumed independent)

29. ViFi protocol • Opportunistic receptions provide a low-overhead but • unreliable means. • With probabilistic relaying, each BS relays based onan • independently computed relaying probability, which avoids • theneed for explicit coordination messages between BSes. • The resulting protocol is lightweight , decentralized ,simple • and works well.

30. ViFi protocol Salvaging • Sometimes a vehicle moves out of range before the • anchor BS can deliver packets from the Internet. • newly designated anchors salvage packets by contacting • the previous anchor over the backplane. • the old anchor transfers any unacknowledgedpackets • within a certaintime threshold(based on the minimum • TCP retransmission timeout,3s).

31. ViFi protocol Estimating packet reception probabilities using beacons Pab:the number of beacons received in a given time interval divided by the number that must have been sent. Beacons packet reception probability from them to other nodes(learn from other nodes) current incoming reception probability

32. Evaluation link-layer close to ideal interactive application coordination mechanism two-fold compared to current methods low false positive and false negative rates

33. Evaluation Link-layer performance

34. Evaluation Link-layer performance back to 14

35. Evaluation application performance TCP most of ViFi’s gain is a result from diversity. Given that only 1.2% of the packets are salvaged, this benefit of salvaging is disproportionate（10%）.

36. Evaluation application performance VoIP Thestandard for evaluating a voice call is theMean Opinion Score (MoS) VoIP ismore challenging than TCP because quality is sensitive to both lossand delay.

37. Evaluation application performance VoIP interruption:the MoS value drops below 2 for a three-second period. ViFiaverageMoS is 3.4 compared with 3.0 of BRR.

38. Evaluation Effectiveness of coordination

39. Conclusion • current WiFi is unsuitable for vehicular client • firstly studyed basestation diversity，then designed ViFi • key to its effectiveness is a decentralized probabilistic algorithm • excellent link-layer performance • doubled the number of successful TCP transfers and the • length of disruption-free VoIP calls

40. Appendix • Extent of diversity • Efficiency of medium usage • Limitations • Deployment