InteracCve WiFi ConnecCvity from Moving Vehicles

Transcription

1 InteracCve WiFi ConnecCvity from Moving Vehicles Aruna Balasubramanian, Ratul Mahajan Arun Venkataramani, Brian N Levine, John Zahorjan University of Massachuse2s Amherst Microso5 Research University of Washington MoCvaCon Increasing demand for network access from vehicles E.g., VoIP, Web, Cellular access expensive Ubiquity of WiFi Cheaper, higher peak throughput compared to cellular 2 1

2 Our work Given enough coverage, can WiFi technology be used to access mainstream applica9ons from vehicles? ExisCng work shows the feasibility of WiFi access at vehicular speeds focus on non interaccve applicacons. e.g., road monitoring Internet 3 Talk outline Can popular applicacons be supported using vehicular WiFi today? Performance is poor due to frequent disrupcons How can we improve applicacon performance? ViFi, a new handoff protocol that significantly reduces disrupcons Does ViFi really improve applicacon performance? VoIP, short TCP transfers 4 2

3 VanLAN: Our Vehicular Testbed Uses MS campus vans Base stacons(bses) are deployed on roadside buildings Currently 2 vans, 11 BSes 5 Measurement study Study applicacon performance in vehicular WiFi seang Focus on basic conneccvity Study performance of different handoff policies Trace driven analysis Nodes send periodic packets and log recepcons 6 3

4 Handoff policies studied PracCcal hard handoff Associate with one BS Current Ideal hard handoff Use future knowledge Imprac9cal 7 Handoff policies studied PracCcal hard handoff Associate with one BS Current Ideal hard handoff Use future knowledge Imprac9cal Ideal sod handoff Use all BSes in range Performance upper bound 8 4

5 Comparison of handoff policies DisrupCon Summary Performance of interac9ve applica9ons poor when using exiscng handoff policies Sod handoff policy can decrease disrupcons and improve performance of interaccve applicacons PracCcal hard handoff Ideal hard handoff Ideal sod handoff 9 Talk outline Can popular applicacons be accessed using vehicular WiFi? How can we improve applicacon performance? ViFi, a pracccal diversity based handoff protocol Does ViFi really improve applicacon performance? VoIP, short TCP transfers 10 5

6 Designing a pracccal sod handoff policy Goal: Leverage mulcple BSes in range Not straighgorward Constraints in Vehicular WiFi 1. Inter BS backplane oden bandwidth constrained 2. InteracCve applicacons require Cmely delivery 3. Fine grained scheduling of packets difficult Internet 11 Why are exiscng solucons inadequate? OpportunisCc protocols for WiFi mesh (ExOR, MORE) Uses batching: Not suitable for interaccve applicacons Path diversity protocols for enterprise WLANs (Divert) Assumes BSes are connected through a high speed back plane Sod handoff protocols for cellular (CDMA based) Packet scheduling at fine Cme scales Signals can be combined 12 6

7 ViFi protocol set up Internet A C Vehicle chooses anchor BS Anchor responsible for vehicle s packets D B Vehicle chooses a set of BSes in range to be auxiliaries e.g., B, C and D can be chosen as auxiliaries ViFi leverages packets overheard by the auxiliary 13 ViFi protocol (1) Source transmits a packet (2) If descnacon receives, it transmits an ack (3) If auxiliary overhears packet but not ack, it probabilis9cally relays to descnacon (4) If descnacon received relay, it transmits an ack (5) If no ack within retransmission interval, source retransmits C Source A B Dest D Downstream: Anchor to vehicle C Dest A B Source D Upstream: Vehicle to anchor 14 7

8 Why relaying is effeccve? Losses are bursty Independence Losses from different senders independent Losses at different receivers independent C A B C A B D Downstream D Upstream 15 Guidelines for probability computacon 1. Make a colleccve relaying decision and limit the total number of relays 2. Give preference to auxiliary with good conneccvity with descnacon How to make a colleccve decision without per packet coordinacon overhead? 16 8

9 Determine the relaying probability Goal: Compute relaying probability R B of auxiliary B Step 1: The probability that auxiliary B is considering relaying C B = P(B heard the packet). P(B did not hear ack) Step 2: The expected number of relays by B is ViFi: PracCcal sod handoff protocol uses probabiliscc E(B) = C B R B relaying for coordinacon without per packet Step 3: Formulate ViFi probability equacon, E(x) = 1, x 2 coordinacon cost auxiliary to solve uniquely, set R B propor9onal to P(des9na9on hears B) Step 4: B escmates P(auxiliary considering relaying) and P(desCnaCon heard auxiliary) for each auxiliary 17 ViFi ImplementaCon Implemented ViFi in windows operacng system Use broadcast transmission at the MAC layer No rate adaptacon Deployed ViFi on VanLAN BSes and vehicles 18 9

10 Talk outline Can popular applicacons be accessed using vehicular WiFi? Due to frequent disrupcons, performance is poor How can we improve applicacon performance? ViFi, a pracccal diversity based sod handoff protocol Does ViFi really improve applicacon performance? 19 EvaluaCon EvaluaCon based on VanLAN deployment ViFi reduces disrupcons ViFi improves applicacon performance ViFi s probabiliscc relaying is efficient Also in the paper: Trace driven evaluacon on DieselNet testbed at UMass, Amherst Results qualitacvely consistent 20 10

11 ViFi reduces disrupcons in our deployment PracCcal hard handoff ViFi 21 ViFi improves VoIP performance Use G.729 codec seconds > 100% ViFi PracCcal hard handoff 0 Length of voice call before disrupcon 0.5 DisrupCon = When mean opinion score (mos) is lower than a threshold 22 11

12 ViFi improves performance of short TCP transfers Workload: repeatedly download/upload 10KB files > 50% > 100% ViFi PracCcal hard handoff Number of transfers before disrupcon 0.5 Median transfer Cme (sec) 0.5 DisrupCon = lack of progress for 10 seconds 23 ViFi uses medium efficiently Efficiency: Number of unique packets delivered/ Number of packets sent 1 efficiency ViFi PracCcal hard handoff

13 Conclusions Our work improves performance of interaccve applicacons for vehicular WiFi networks InteracCve applicacons perform poorly in vehicular seangs due to frequent disrupcons ViFi, a diversity based handoff protocol significantly reduces disrupcons Experiments on VanLAN shows that ViFi significantly improves performance of VoIP and short TCP transfers hop://research.microsod.com/netres/projects/vanlan/ 25 PracCcal sod handoff AllBS itself is not a pracccal protocol How does diversity even work in the downstream? Internet 26 13

14 Distributed computacon B needs to compute for each auxiliary (1) contending probabilices (2) P(V will hear from the auxiliary) Can be computed using loss rates between the auxiliary and its neighbors C Compute loss rates using beacon recepcons V A D B Embeds its loss rates in beacons Learns loss rates of auxiliaries from their beacons EsCmates relaying probability of B, C and D 27 Distributed probability computacon V Compute consistent relaying probabilices C A D B Compute consistent relaying probabilices B need to know for C and D 1. Probability that C and D are making a relaying decision 2. Probability that vehicle will hear from C and D Both can be escmated using loss probabilices between C and D and their neighbors Nodes exchange 2 hop loss rates using beacons B computes relaying probabilices for B, C and D 28 14