FleaNet : A Virtual Market Place on Vehicular NetworksUichin Lee, Joon-Sang Park Eyal Amir, Mario Gerla

Network Research Lab, Computer Science Dept., UCLA

Advent of VANETs Emerging VANET applications

Safety driving (e.g., TrafficView)Content distribution (e.g., CarTorrent/AdTorren

t)Vehicular sensors (e.g., MobEyes)

What about commerce “on wheels”?

Flea Market on VANETs Examples

A mobile user wants to sell “iPod Mini, 4G”A road side store wants to advertise a special

offer How to form a “virtual” market place using

wireless communications among mobile users as well as pedestrians (including roadside stores)?

Outline FleaNet architecture FleaNet protocol design Feasibility analysis Simulation Conclusions

FleaNet Architecture-- System Components Vehicle-to-vehicle communications Vehicle-to-infrastructure (ad-station) communications

Inter-vehic lecommunications

Private Adstation

Vehic le-to-adstationcommunications

* Roadside stores (e.g., a gas station)

FleaNet Architecture -- Query Formats and Management Users express their interests using formatted qu

eries eBay-like category is provided

E.g., Consumer Electronics/Mp3 Player/Apple iPod

Query management Query storage using a light weight DB (e.g., Berkeley D

B) Spatial/temporal queries Process an incoming query to find matched queries (i.

e., exact or approximate match) E.g. Query(buy an iPod) Query(sell an iPod)

FleaNet Protocol Design FleaNet building blocks

Query disseminationDistributed query processing Transaction notification

Seller and buyer are notified This requires routing in the VANET

VANET challengesLarge scale, dense, and highly mobile

Goal: designing “efficient, scalable, and non-interfering protocols” for VANETs

Query Dissemination Query dissemination exploiting vehicle

mobility Query “originator” periodically advertises

its query to 1-hop neighbors Vehicles “carry” received queries w/o further

relaying

Q1

Q2

Q1

Q2

Yellow Car w/ Q1

Red Car w/ Q2

Distributed Query Processing Received query is processed to find a match of

interests Eg. Q1 – buy iPod / QM – sell iPod / Q2 – buy Car

QM

QM

Q2

Q2

(1) Find a matching query

for Q2

No match found

QM

LocalMatchQMQ1

(2) Send a match notification msg to the originator of query QM

Red car w/ Q2

& carries Q1

Cyan car w/ QM

Q1

(1) Find a matching query

for QM

Found query Q1

Transaction Notification After seeing a match, use Last Encounter

Routing (LER) to notify seller/buyer Forward a packet to the node with more

“recent” encounter

QM

LocalMatchQMQ1

Q1

Q1

Q1

Q1 T-1s

T-5s

T-10s

T

Encounter timestamp

Current Time: T

Originator of Q1

Cyan car

Red car

Blue car

Green car Yellow

carTRXRESP

TRXREQ

FleaNet Latency Restricted mobility patterns are harmful to opportunistic

data dissemination However, latency can be greatly improved by the popularit

y of queries Popularity distribution of 16,862 posting (make+model) i

n the vehicle ad section of Craigslist (Mar. 2006)

Freq

uenc

y (l

og)

Items (log)

FleaNet Scalability Assume that only the query originator can “period

ically” advertise a query to its neighbors We are interested in link load Load depends only on average number of neighb

ors and advertisement period (not on network size)

Example: Parameter setting : R=250m, 1500B packet size, BW

=11Mbps N=1,000 nodes in 2,400m x 2,400m (i.e., 90 nodes

within one’s communication range) Advertisement period: 2 seconds Worst case link utilization: < 4%

Simulations Ns-2 network simulator 802.11b - 2Mbps, 250M

radio range Two-ray ground reflection

model “Track” mobility model

Vehicles move in the 2400mx2400m Westwood area in the vicinity of the UCLA campus

Metric Average latency: time to

find a matched query of interest

Westwood area, 2400mx2400m

Simulation Results Impact of density and speed

0

50

100

150

200

250

300

350

400

450

5 10 15 20 25

Average Speed (m/s)

Late

ncy

(S

eco

nd

s) N=100N=200N=300

Simulation Results Impact of query popularity

Popularity: the fraction of users with the same interest For a single buyer, increase the number of sellers (e.g.,

N=200/0.1 = 20 sellers)

0

10

20

30

40

50

60

70

0.05 0.1 0.15 0.2 0.25

Popularity

Late

ncy

(S

eco

nds)

N=100/V=5

N=100/V=25

N=300/V=5

N=300/V=25

Simulation Results Impact of ad-station location

Given N=100, fix each node in its initial location, and set it as a “stationary” ad-station (as a buyer)

measure the average latency to the remaining 99 mobile nodes (run 99 times, by taking turns as a seller: 1 buyer 1 seller)

0

50

100

150

200

250

300

350

400

450

500

1 11 21 31 41 51 61 71 81 91

Rank

Late

ncy

(S

eco

nds)

N=100/V=25m/s

avg. stationaryavg. mobile

Latency rank

Conclusions Proposed a virtual market concept in VANETs:

A mix of mobile and stationary users carry out buy/sell transactions (or any other matching of common interests) using vehicular networks

Mobility-assisted query dissemination and resolution (scalable and non-interfering) Node density/speed are closely related to the performance Popularity of a query greatly improves the performance Location of an ad-station is important to the performance

Future work Query aggregation to improve the performance

Unpopular queries/queries from ad-stations How to enforce cooperativeness of users? Security: false query injection and spamming?