(Slides) Video Ads Dissemination through WiFi-Cellular Hybrid Networks

Video Ads Dissemination through
WiFi-Cellular Hybrid Network
(revised)
Hiroshi Hanano1, Yoshihiro Murata2, Naoki Shibata3,
KeiichiYasumoto1 and Minoru Ito1
1NaraInstituteofScienceandTechnology
2HiroshimaCityUniversity
3ShigaUniversity

Organization
1. Background
2. RelatedWork
3. Problem Formulation
4. Proposed Algorithm
5. Evaluationthrough Simulations
6. Conclusion
2Video Ads Dissemination through WiFi-Cellular Hybrid Networks

Demand for Context-Aware Video Ad Dissemination
• Video deliveryto mobile terminals
has alreadybecome common
– Videoadvertisementdissemination
now attractingpublicattention
• Many studies forobtaining user contexts(situation, etc)
withsensors,userdata(profiles,schedules,etc),andsoon
• Context-aware video ad dissemination
– Muchhigheradvertisementeffectthanwithoutcontext-
awareness
3
Movie trailers
Video Ads Dissemination through WiFi-Cellular Hybrid Networks

Challenges
• Contexts are differentamong users, change overtime
– Eachvideoadshouldbedeliveredtousersatdifferenttime
4
→ larger data amount transferred via cellular network
→ need to consider influence to other cellular services
→ number of ads may be restricted
We need a dissemination method which
does not overload cellular network

Goals
• Distribute communication load forvideo ad dissemination
from cellularnetwork to WiFi network as much as possible
5
Cellular
Network
load
WiFi(Ad-hoc)

Organization
1. Background
2. RelatedWork
6. Conclusion

Complementing Cellular Network with MANET
– Somemobileterminalsincongestedcellareause
lightloadedcellviaMANET
7
Hongyi, W., Chunming, Q., Swades, D. and Ozan, T.:
”Integrated Cellular and Ad Hoc Relaying Systems:iCAR,”
Proc. of ICC 2001
It does not reduce overall data amount
transferred for video ads dissemination

Introducing BitTorrent into MANET
• Purpose:efficientfile exchangeamong mobile terminals
– RealizedBitTorrentmechanismonMANET
– Evaluationdoneintermsof mobilityandchunksizes
– Showrobustnessagainstnetworkpartition
8
S. Rajagopalan. and C.-C. Shen:
”A Crosslayer Decentralized BitTorrent for Mobile Ad hoc Networks,”
Proc. of MOBIQUITOUS 2006
It only uses MANET, and thus achieves
poor performance on data acquisition ratio/time

Organization
1. Background
2. RelatedWork
6. Conclusion

Assumptions
• Manyuserswithmobilephonesinurbanarea
– PhonescanusebothcellularandWi-Fi(ad-hocmode)networks
• Terminalautomaticallysets,accordingtouser’scontext:
– Videoadcontents(byfilenames)
– ReceptionDeadlineforeachcontent
10
ex) Context: Bob has a plan to go to a movie after work
Contents: Trailers of movies currently running
Deadline: Time when work finishes

Target Problem (see PerCom09 paper for more formal description)
• Input
– Videoadswithdeadlinesrequestedbyeachmobilenode
• Output
– Actionsofmobilenodesateachmoment
• Download, Broadcast, or Receive (part of) data
• Constraints
– Cellulartowercapacity(Nconnectionscanbeadmitted)
– Cellularnetworkbandwidthperconnection
– Wi-Firadiorangeofeachmobilenode
– Wi-Fibandwidthsharedbyallnodesineachradiorange
– Receptiondeadlineofeachcontent
• Objectivefunction
– Minimizedataamounttransferredthroughcellularnetwork

Organization
1. Background
2. RelatedWork
6. Conclusion

Approach
• File division (likeBitTorrent)
– Divideeachadcontentintochunkswithfixeddatasize
– Everynodecancontributeaddisseminationinfairmanner
• Utilizationof WiFi (ad-hoc mode)
– Reduceloadincellularnetwork
bylettingnodesexchangechunksthroughWi-Finetwork
• Fully distributed algorithm
– Eachterminaldecidesitsactionbasedonlyon informationof
neighbornodes
– Scalabletonumberofnodes,numberofcontents,etc

Possible Actions Taken by Mobile Terminal
• Downloada chunk through cellularnetwork
• Broadcast a chunk through WiFi (ad-hoc mode) network
• Waitto receive a chunk through WiFi network

Algorithm for Action Decision
• Eachnode takes an actionfor global benefit
– Foreachchunk,onlyoneterminalshoulddownloadit
– Ineachradiorange,weshouldavoidtwoormoreterminals
broadcastingthesamechunk
• Flow of proposed algorithm
15
Info Gathering Part
Gather neighboring info
Action Decision Part
Select a chunk
Take an action
Categorize chunks/compute
impact of chunks

Information Gathering Part
• Informationrequired to decide action
– DemandInfo.：whichcontentseachnoderequires
– ChunkInfo.：whichchunkseachnoderetains
• CreateNeighbor Table by exchanginghello messages
– Eachnodeperiodically(ex.30s)broadcasthellomessage
includingdemandinformationandchunkinformation
16
a1 a2 a3 …A, B
Demand
Info.
Chunk Info.
:retain :no
Hello Message
a1 a2 a3 …A, C
b1 b2 b3 …B, D
a1 a2 a3 …A, D
n1
n2
n3
↑Terminal ID
Neighbor Table

Action Decision Part
1. Select a chunk foraction
– Categorizechunks
• Chunks ALREADY obtained broadcast via WiFi
• Chunks NOT obtained download from cellular
– Impactcomputation
• Prioritize chunks by putting impact values
– Rouletteselection
• Select a chunk by roulette
2. Takean actionfor the selectedchunk
– Forbroadcast,weleteachnodetakeactionat probability
b toavoidcollision

Action Decision Phase –Categorizing Chunks–
• C1: Chunks to be downloadedfrom cellular
– Nodewantsa chunk,butnoneighbornodesretainit
• C2: Chunks to be broadcast through WiFi
– Noderetainsa chunk,andsomeofneighbornodeswantit
• C3: Other chunks
– Nodedoesnottakeanyactiontothosechunks
18
Node under
consideration
Node under
consideration
Neighbor nodes
Neighbor nodes

Action Decision Phase –Impact Computation(1)–
• C1:Chunknotretainedbyanyneighbornodes
• Num.ofneighbornodesthatwantchunks:
19
Impact for chunk s:
Demand low High
Impact high Low
3
1
3
1 3
1
Avoid redundant
download

Action Decision Phase –Impact Computation(2)–
• C2: Chunk which some nodes retain,and some of
neighbor nodes want the chunk
• Num of neighbor nodes broadcasting chunk s:
20
Impact for s:
Number of
retaining nodes
small large
Impact high low
2
1
2
1
Avoid redundant
broadcast

Action Decision Phase –Roulette Selection–
• RouletteSelection
– Preparetworoulettesfor cellularandWiFi
– Decideportionofeachchunkinproportiontoimpact
– Selectachunkbyrouletteselection
– Probabilityofselectingachunkwithimpactwi
21
Impact low High
Probability low High
High impact chunk
Low impact chunk

Problem in Basic Algorithm
• Initially, allnodes intensivelyuse cellularnetwork while
neighboring informationis not known
• As time elapses,cellularis not used (neighbornodeshavechunks)
22
Time
100
Time
Not adaptive to remaining time for deadline
Short increase cellular usage to meet the deadline
Long reduce cellular usage so that nodes exchange chunks via WiFi
Download
Rate (%)
Cellular
Usage

Adaptation Mechanism (Extended from PerCom09 WiP)
• Consider target downloadyield at elapsedtime to deadline
– Currentdownloadyieldishigherthanthis,donotusecellular
– Currentdownloadyieldis lowerthanthis,usecellular
23
100
Elapsed Time deadline
：target DL yield
Not use cellular
Use cellular
Download
Yield (%)

Organization
1. Background
2. RelatedWork
6. Conclusion

Evaluation Experiments
Preliminary Experiment
– Measureperformance,changinghellomessageperiodhandcollision
avoidancerateb
– Purpose: obtaingoodvaluesforhandb
Experiment1
– Measurecellularnetworkusageanddownloadtimeforvariousdeadlines
– Purpose: showadaptabilityforvariousdeadlinelengths
Experiment2
– Measurecellularnetworkusageanddownloadtime,changingnumber
ofcontents
– Purpose:showadaptabilityforvariousnumberofadcontents
 Metric:cellularnetworkusageratio(cellularusage)
25
Extension
from PerCom
version
Cellular usage (%) =
amount actually downloaded via cellular
amount when all nodes download all chunks via cellular

Common Simulation Parameters
• We developed our own simulatorfor simulations in 1ms
granularity
26
Common simulation parameters
Size of each video ad 1.5MB (divided to 1000 chunks)
Size of each chunk 1.5KB (sent in a packet)
Num of ads that
each node requests
2 (selected by Zipf distribution)
Mobility Random Way Point (4km/hour)
Field size 500m×500m

Network Model
27
Cellular network Rate/connection: 1.2Mbps
Maximum connections: 64
WiFi
(ad-hoc mode)
Bandwidth/range: 1.7Mbps
Radio range: 200m
Fading model: Nakagami fading
MAC: CSMA/CA
0
0.25
0.5
0.75
1
0 50 100 150 200
Distance[m]
M. Killat, F. Schmidt-Eisenlohr, H. Hartenstein,
C. Rössel, P. Vortish, S. Assenmacher and F. Busch:
”Enabling Efficient and Accurate Large-Scale Simulations
of VANETs for Vehicular Traffic Management,”
Proc. of VANET2007
Packet
Reception
Rate (%)

Parameters for preliminary experiment
• Adaptation mechanism is not used
28
Parameters
Number of ad contents 4
Number of nodes 500
Deadlines 30 min from starting time
Collision avoidance rate b 0.0005～0.005 (by 0.0005 step)
Hello message period h 5s, 10s, 30s, 60s
Mobility Fixed and RWP

Preliminary Experimental Result (cellular usage)
• Asbdecreases,cellularusageimprovesreducecollision
• Ashdecreases,cellularusageimprovesreduceredundantDL
• Whenh=5andb=0.0005,cellularusagewas7.3%
29
7.3%
Cellular
Usage (%)

Preliminary Experimental Result (download time)
• b between0.001and0.002smallerdownloadtime
– β<0.001:lowWiFiusage
– β>0.002:manycollisions
30
Download
time(s)
- Small h suppresses cellular usage and
prolongs download time

Experiment1: Adaptability to Deadline Lengths
• Evaluateadaptabilityto variousdeadline lengths
• Suppose cold start case
– Allnodessimultaneouslystartacquiringvideoadsafter
eachnodedecidestwoadcontentsandtheirdeadlines
– Hot startcaseisexplainedlater
31
Parameters used in Experiment1
Number of ad contents 10
Number of nodes 500
Deadlines 15, 30, 60, 90 minutes
from starting time (fixed)
h, b 5s, 0.0015

Result of Experiment1 (DL Yield over Time)
• Withoutdeadlineadaptationmechanism
– DLyieldincreasesslowlyafter90%.Manynodesmissdeadlines
• Withdeadlineadaptationmechanism
– Downloadyieldincreasesasexpected.Allnodesmeettheirdeadlines
32
Time
DL
Yield
(%)
Without mechanism
With mechanism
(deadline=90min)
60min
30min
15min

Result of Experiment1 (Cellular Usage over Time)
• Withoutdeadlineadaptationmechanism
– Usescellularnetworkintensivelyonlyininitialphase
• Withdeadlineadaptationmechanism
– Constantlyusescellularnetwork,butincreasesusageinfinishing phase
• In end phase, it takes long time to receive blank chunks via WiFi
33
40.5%
16.9%
4.3%
1.1%
Time
Without mechanism
15min
30min
60min 90min
Number of chunks
downloaded via
cellular

Experimental Result with “Hot Start” Configuration
• Mobilenodesjoinandleavethesystem
• Newnodesstartsobtainingvideoadsjustafterjoining
• Measuredfor267nodesappearingafter200minuteselapses
– Cellularusagewasreducedfurtherthan“coldstart”case
– Minimumcellularusagewasreducedtoabout0.3%
34
Deadlines
Cellular
Usage (%) Cold start
Hot start
Configuration
Initially 500nodes
Poisson arrivals of 15s
120min lifetime
10 ad contents
Deadlines: 15, 30, 60
(random)

Experiment2: Scalability for Number of Contents
• In actualsituation, many video ads may exist
– Investigateifourmethodadaptivelyworksfordifferent
numberof adcontents
35
Parameters for Experiment2
Number of Ad Contents 10, 20, 30, 40
Mobile Terminals 500
Time to Deadline 60 minutes

Result of Experiment2
• Whenincreasingnumberofadcontents
– Smallernumberofterminalswantthesamead
– Ourmethodadaptivelyregulatedcellularusage
• All terminals completed downloading all required contents
– Evenfor40contents,cellularusagereducedtoabout34.9%
36
34.9%
Number of ad contents
Time
Cellular usage (%)
# of chunks
via cellular
40 contents
30
20
10

Conclusion
• Cooperativedownloadmethodformobilephonesforcontext-aware
videoaddissemination
– DoesnotsuppresscellularnetworkbyusingWi-Finetwork
• Evaluationresultssupposingtypicalurbanscenarios
– Cellularnetworkusagewasreducedbyupto98.9%
– Inhotstartscenario,theusagewasreducedfurther
– Adaptivetovariouslengthsofdeadlinesandnumberofvideoad
contents
FutureWork
– Experimentwithrealisticsituation(realtraces,realmaps,etc)
– Incentiveformobileterminals
– multi-hoppathsinadhocnetwork

Hanano, H., Murata, Y., Shibata, N., Yasumoto, K.,
and Ito, M.: Video Ads Dissemination through
WiFi-Cellular Hybrid Networks, Proc. of the 7th
IEEE Int'l. Conf. on Pervasive Computing and
Communications (PerCom 2009), pp. 322-327.
DOI:10.1109/PERCOM.2009.4912787
[ PDF ]
Video Ads Dissemination through WiFi-Cellular Hybrid Networks 38

(Slides) Video Ads Dissemination through WiFi-Cellular Hybrid Networks

Recommandé

Recommandé

Contenu connexe

Plus de Naoki Shibata

Plus de Naoki Shibata (20)

Dernier

Dernier (20)

(Slides) Video Ads Dissemination through WiFi-Cellular Hybrid Networks