<%BANNER%>

Mobile Encounters

Permanent Link: http://ufdc.ufl.edu/UFE0043622/00001

Material Information

Title: Mobile Encounters Pattern Analysis and Profile Embedding for Mobile Social Networking Testbeds
Physical Description: 1 online resource (111 p.)
Language: english
Creator: Moon, Sungwook
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2011

Subjects

Subjects / Keywords: contact -- encounter -- mobile -- networks -- periodicity -- profile -- robot -- social -- testbed
Computer and Information Science and Engineering -- Dissertations, Academic -- UF
Genre: Computer Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Study on human mobility is gaining increasing attention from the research community for use in mobile networks. To better understand the potential of mobile nodes as message relays, our study first investigates the encounter pattern of mobile devices. Specifically, we examine extensive network traces that reflect mobility of communication devices. We analyze the periodicity and consistency of encounter patterns by using power spectral analysis. Our result shows the presence of strong periodicity for rarely encountering mobile nodes and weak periodicity for frequently encountering nodes. In addition, our investigation on the encounter history shows that consistency depends on the encounter rate and length of history. With this understanding of human encounter patterns, we discuss profiling of mobile users based on their periodic properties in encounter pattern. In addition, we group mobile users based on encounter days and discover that the rank group size follows power-law distribution that we use in the assignments of communities for autonomous nodes. To enhance the mobile networks testing, we utilize our findings to effectively capture and embed personality of mobile users in simulation and testbed environment. We propose an encounter rule-based decision to mimic human encounter pattern, which is an important step toward efficient design of mobile social networking protocols and services. With the additions of group information and scheduler to the rule-based decision, we show that our approaches enable autonomous mobile nodes collectively mimic human encounter patterns. We experiment with various types of decision modes and compare the results to random mobility and real-world networking trace. The result shows that our proposed approach provides the range of knobs for adjusting parameters to capture power-law distribution of group sizes, encounter ratio with group members and periodical encounter patterns that are close to real-world networking trace while far outperforming random mobility. Finally, we propose a novel mobile networking testbed that blends the network of autonomous robots and participatory testing via personality profile. We implement a prototype mobile networking testbed with IRobot and PDAs.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Sungwook Moon.
Thesis: Thesis (Ph.D.)--University of Florida, 2011.
Local: Adviser: Helmy, Ahmed H.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2011
System ID: UFE0043622:00001

Permanent Link: http://ufdc.ufl.edu/UFE0043622/00001

Material Information

Title: Mobile Encounters Pattern Analysis and Profile Embedding for Mobile Social Networking Testbeds
Physical Description: 1 online resource (111 p.)
Language: english
Creator: Moon, Sungwook
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2011

Subjects

Subjects / Keywords: contact -- encounter -- mobile -- networks -- periodicity -- profile -- robot -- social -- testbed
Computer and Information Science and Engineering -- Dissertations, Academic -- UF
Genre: Computer Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Study on human mobility is gaining increasing attention from the research community for use in mobile networks. To better understand the potential of mobile nodes as message relays, our study first investigates the encounter pattern of mobile devices. Specifically, we examine extensive network traces that reflect mobility of communication devices. We analyze the periodicity and consistency of encounter patterns by using power spectral analysis. Our result shows the presence of strong periodicity for rarely encountering mobile nodes and weak periodicity for frequently encountering nodes. In addition, our investigation on the encounter history shows that consistency depends on the encounter rate and length of history. With this understanding of human encounter patterns, we discuss profiling of mobile users based on their periodic properties in encounter pattern. In addition, we group mobile users based on encounter days and discover that the rank group size follows power-law distribution that we use in the assignments of communities for autonomous nodes. To enhance the mobile networks testing, we utilize our findings to effectively capture and embed personality of mobile users in simulation and testbed environment. We propose an encounter rule-based decision to mimic human encounter pattern, which is an important step toward efficient design of mobile social networking protocols and services. With the additions of group information and scheduler to the rule-based decision, we show that our approaches enable autonomous mobile nodes collectively mimic human encounter patterns. We experiment with various types of decision modes and compare the results to random mobility and real-world networking trace. The result shows that our proposed approach provides the range of knobs for adjusting parameters to capture power-law distribution of group sizes, encounter ratio with group members and periodical encounter patterns that are close to real-world networking trace while far outperforming random mobility. Finally, we propose a novel mobile networking testbed that blends the network of autonomous robots and participatory testing via personality profile. We implement a prototype mobile networking testbed with IRobot and PDAs.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Sungwook Moon.
Thesis: Thesis (Ph.D.)--University of Florida, 2011.
Local: Adviser: Helmy, Ahmed H.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2011
System ID: UFE0043622:00001


This item has the following downloads:


Full Text

PAGE 1

MOBILEENCOUNTERS:PATTERNANALYSISANDPROFILEEMBEDDINGFORMOBILESOCIALNETWORKINGTESTBEDSBySUNGWOOKMOONADISSERTATIONPRESENTEDTOTHEGRADUATESCHOOLOFTHEUNIVERSITYOFFLORIDAINPARTIALFULFILLMENTOFTHEREQUIREMENTSFORTHEDEGREEOFDOCTOROFPHILOSOPHYUNIVERSITYOFFLORIDA2011

PAGE 2

c2011SungwookMoon 2

PAGE 3

ToJinYun,Dad,Mom,KaylinSaeyunandmygrandparents 3

PAGE 4

ACKNOWLEDGMENTS IsincerelythanktoDr.AhmedHelmyforhisexcellentguidanceformyPh.D.education.Heisindeedagreatresearchadviserandeducator.Igreatlyappreciateforhisguidance.Ialsothankmysupervisorycommitteefortheirvaluableandencouragingcomments.Iamtrulygratefultomyfamily.Mywife,whosharedallthegreatmomentsandencouragedmetocompletethiseducation.ItwasahappyjourneybecauseIsharedthiswholeprocesswithher.Myparentswhosupportedineverywaythattheycanpossiblydo.Justseeingmydadwasagreateducationandmotivationformetopushmyselfharder.Mymom'scountlesssupportingandcaringwordswerealwaysbigencouragement.Mygrandfatherwhowasarolemodelandallowedmetodreamofbecomingagreatperson.Mydaughterformakingmesmileandgivingamazinghappiness.Iappreciatemybrotherforgivingmehelpsinmanyofcomputingtechniquesandjoyofsharingknowledges.NOMADSgroupmembersunderDr.Helmyhelpedandsupportedmeingreatwayaswell.IparticularlythankUdayanKumarandWei-jenHsuforsharinggreattechnicalandnon-technicalideasthathelpedmetobroadenmyexperienceandknowledgeinsomanythings.IthankYibinWangonprogramminghelpforBluetoothscanningprogramandhisideaofeventbasedscenarios.IalsothankGautamS.ThakurformanydiscussionsandhelpsIreceivedforresearchanddefensepreparation.IthankShao-chengWangandSaponTenachaiwiwatfortheiradviceandencouragement,andJeeyoungKimforreviewcomments.ItwasagreatopportunityformetolearnsomuchfromgreatindividualsinNOMADswhonotonlyareintelligentbutalsohavegreatpersonalities.IalsothankallofmyfriendsinCISEandSouthKorea.Theirsupportwasveryencouraging.Thelastbutnottheleast,IthankGodforallowingmetoexperiencesuchwonderfulandvaluablemomentsinmylifeandtonishtheeducation. 4

PAGE 5

TABLEOFCONTENTS page ACKNOWLEDGMENTS .................................. 4 LISTOFTABLES ...................................... 8 LISTOFFIGURES ..................................... 9 ABSTRACT ......................................... 11 CHAPTER 1INTRODUCTION ................................... 13 1.1ResearchFramework ............................. 13 1.2ConceptualModel ............................... 14 1.3MobileNetworks ................................ 15 1.4ProblemStatements .............................. 16 1.4.1MotivationandChallenge ....................... 16 1.4.2ProblemStatements .......................... 17 1.4.3Approaches ............................... 17 1.5ResearchComponents ............................ 17 1.5.1AnalysisofEncounterPattern ..................... 18 1.5.2ProlesofMobileNodes ........................ 19 1.5.3ProleBasedMobileSocialNetworkingTestbed .......... 19 1.6Contribution ................................... 20 1.6.1EffortContribution ........................... 20 1.6.2IntellectualContribution ........................ 20 2RELATEDWORK .................................. 21 2.1AnalysisofEncounterPatterninMobileNetworks ............. 21 2.2MobileSocialNetworkingProtocols ..................... 22 2.3MobileNetworkingTestbeds .......................... 24 3MOBILITYTRACESDATA .............................. 26 3.1BasicDenitions ................................ 26 3.2NetworkTraces ................................. 26 3.3TransformingNetworkTracestoEncounterTraces ............. 27 3.3.1BluetoothEncounterTraces ...................... 28 3.3.2WLANTraces .............................. 28 3.3.3TransformedEncounterTrace ..................... 30 4UNDERSTANDINGENCOUNTERPATTERNSOFMOBILENODES ...... 31 4.1Introduction ................................... 31 4.2Relatedwork .................................. 32 5

PAGE 6

4.3UnderstandingPeriodicityandRegularityofMobileNodes ......... 34 4.3.1Methodology .............................. 34 4.3.2PeriodicitiesinNodalEncounters ................... 37 4.3.2.1WLANtrace ......................... 37 4.3.2.2Bluetoothtrace ........................ 43 4.3.3PeriodicityofIndividualEncounterPattern .............. 43 4.3.4RegularEncounter ........................... 47 4.3.4.1Topfrequency ........................ 47 4.3.4.2Normalizedregularencounter ............... 48 4.4TemporalStabilityofMobileEncounter .................... 50 4.4.1StabilityWindowofEncounterHistory ................ 51 4.4.2ConsistencyofEncounterPattern ................... 52 4.4.3ApplicationofEncounterStabilityAnalysis .............. 55 4.4.4ComparingRegularEncountertoNon-regularEncounter ...... 56 4.5Long-termBluetoothTraceAnalysis ..................... 57 4.6ConclusionsandFutureWork ......................... 59 5CAPTURINGANDEMBEDDINGPROFILESOFMOBILENODES ....... 61 5.1Introduction ................................... 61 5.2RelatedWork .................................. 62 5.3ModelingEncounterPeriodicitywithProle ................. 62 5.4DiscussionofRoutingBasedonEncounterProles ............. 64 5.5CommunityProlesofMobileUsers ..................... 65 5.5.1LocationVisitingPreference ...................... 66 5.5.2EncounterVector ............................ 67 5.6ConclusionsandFutureWork ......................... 68 6DESIGNOFAMOBILESOCIALNETWORKINGTESTBED .......... 71 6.1Introduction ................................... 71 6.2RelatedWork .................................. 74 6.2.1AnalysisofHumanMobility ...................... 74 6.2.2MobileNetworkingTestbeds ...................... 75 6.2.3ParticipatoryNetworks ......................... 77 6.3TestbedArchitecture .............................. 78 6.3.1AutonomousMobileNodes ...................... 78 6.3.2NetworkofAutonomousRobots .................... 79 6.3.3ParticipatoryTesting .......................... 80 6.3.4LimitationofControlledandUncontrolledEnvironment ....... 82 6.4ContactRule-BasedDecisionCriteria .................... 83 6.4.1Power-lawDistribution ......................... 86 6.4.2Memory ................................. 87 6.5Experiment ................................... 87 6.5.1SimulationSetup ............................ 87 6.5.1.1Communityprole ...................... 87 6

PAGE 7

6.5.1.2On-timescheduler ...................... 87 6.5.1.3Environmentsetup ...................... 88 6.5.2ExperimentModes ........................... 89 6.5.3ResultAnalysis ............................. 90 6.5.3.1Contactratiowithfriendstostrangers ........... 90 6.5.3.2Rankgroupsize ....................... 91 6.5.3.3Periodiccontactpattern ................... 92 6.5.4VisualizationofEncounterandMobility ................ 93 6.6ImplementationonAutonomousRobots ................... 95 6.6.1ControllingiRobot ............................ 95 6.6.2LabEnvironment ............................ 96 6.6.3EvaluationScenariosforAutonomousRobots ............ 97 6.7ConclusionandFutureWork ......................... 101 7CONCLUSIONS ................................... 102 7.1Conclusions ................................... 102 7.2FutureWork ................................... 103 REFERENCES ....................................... 105 BIOGRAPHICALSKETCH ................................ 111 7

PAGE 8

LISTOFTABLES Table page 3-1Statisticsofencountertraces. ............................ 27 3-2FormatofWLANtraces ............................... 29 3-3Formatofencountertrace .............................. 30 6-1Propertiesineachnode'sprole .......................... 87 6-2Schedulerforon-timeperiod ............................ 88 6-3Schedulerforon-timeduration ........................... 88 6-4Experimentenvironment ............................... 88 6-5Experimentmodes .................................. 89 8

PAGE 9

LISTOFFIGURES Figure page 1-1Frameworkofdissertation .............................. 14 4-1Exampleencountertraceforamobileuserwithoneoftheencounterednodes 35 4-2Averageautocorrelationcoefcient ......................... 36 4-3Dailyencounter:normalizedfrequencymagnitudeoffrequencycomponentsforencounteredpairs ................................ 38 4-4Encounterfrequency:normalizedfrequencymagnitudeoffrequencycomponentsforencounteredpairs ................................ 39 4-5Encounterduration:normalizedfrequencymagnitudeoffrequencycomponentsforencounteredpairs ................................ 40 4-6HourlyencounterforBluetoothpairs:frequencymagnitudeoffrequencycomponentsforhourlyencounteratUF08spring/fallBluetoothtrace ............. 42 4-7Dailyencounterforindividualnodes:normalizedfrequencymagnitudeoffrequencycomponentsforindividualnodes'encounterpattern ............... 44 4-8Encounterfrequencyforindividualnodes:normalizedfrequencymagnitudeoffrequencycomponentsforindividualnodes'encounterpattern ........ 45 4-9Encounterdurationforindividualnodes:normalizedfrequencymagnitudeoffrequencycomponentsforindividualnodes'encounterpattern ......... 46 4-10EmpiricalCDFofhighestfrequencyfortheencounteredpairsatUSC06springtraceaccordingtodailyencounterrate ....................... 48 4-11APsaccessingpreferenceatUSC06springtrace ................ 49 4-12Orderedlocationvisitingpreferencebyencounteredpairsaccordingtodailyencounterrate .................................... 49 4-13Windowofencounterhistory ............................ 52 4-14AveragedifferenceofthedailyencounterrateatUFandUSCcampusaccordingtothechangeofbreakpointfortherstwindow .................. 53 4-15AveragedifferenceofthedailyencounterrateatUFandUSCcampusaccordingtothechangeofbreakpointforthesecondwindow ................ 55 4-16AveragedifferenceofthedailyencounterrateatUSCcampusaccordingtothechangeofsizefortherstwindowofencounterhistory ........... 56 4-17Timeseriesdataforthemobileuser'sBluetoothencounter ........... 58 9

PAGE 10

4-18EmpiricalCDFforeachencountermetricwithencounterednodesinthelongtermBluetoothtrace ................................. 59 4-19Spectralanalysisoflong-termBluetoothencounter ................ 60 5-1CDFoftop7maximumfrequencies ........................ 63 5-2Overlaynetworksusingperiodicityasalinkweight ................ 65 5-3Exampleofembeddingprolesinvariousinterfaces ............... 66 5-4Locationbasedassociationmatrixforeachmobileuser(courtesybyHsu[ 1 ]) 67 5-5Rankedsizeofgroupsforcampustraces ..................... 69 6-1Bridgingthegapbetweencontrolledandnon-controlledenvironment ...... 73 6-2PictureofiRobot,itscontrollingNokiaN810PDAandhumancarryingNokiaN810PDA ....................................... 79 6-3Communicationstructureamongrobots,personalityinterface,andcommunicationprotocol ........................................ 81 6-4Statediagramoffriend-strangerdecisionmodel ................. 84 6-5Mobilityofsourcenode(Me)ineachdecision ................... 85 6-6Visualizationofsimulation .............................. 90 6-7Contactratiooffriendstostrangers ........................ 91 6-8Rankplotforgroupsizeinlog-logscaleinsimulation ............... 92 6-9Spectralanalysisforcontactwithfriendsover32daysinsimulation ....... 93 6-10Snapshotsofsimulationforeachmodewithvisualizationon. .......... 94 6-11Snapshotofaproofofconceptvideoformobilenetworkingtestbed ...... 100 10

PAGE 11

AbstractofDissertationPresentedtotheGraduateSchooloftheUniversityofFloridainPartialFulllmentoftheRequirementsfortheDegreeofDoctorofPhilosophyMOBILEENCOUNTERS:PATTERNANALYSISANDPROFILEEMBEDDINGFORMOBILESOCIALNETWORKINGTESTBEDSBySungwookMoonDecember2011Chair:AhmedHelmyMajor:ComputerEngineeringStudyonhumanmobilityisgainingincreasingattentionfromtheresearchcommunityforuseinmobilenetworks.Tobetterunderstandthepotentialofmobilenodesasmessagerelays,ourstudyrstinvestigatestheencounterpatternofmobiledevices.Specically,weexamineextensivenetworktracesthatreectmobilityofcommunicationdevices.Weanalyzetheperiodicityandconsistencyofencounterpatternsbyusingpowerspectralanalysis.Ourresultshowsthepresenceofstrongperiodicityforrarelyencounteringmobilenodesandweakperiodicityforfrequentlyencounteringnodes.Inaddition,ourinvestigationontheencounterhistoryshowsthatconsistencydependsontheencounterrateandlengthofhistory.Withthisunderstandingofhumanencounterpatterns,wediscussprolingofmobileusersbasedontheirperiodicpropertiesinencounterpattern.Inaddition,wegroupmobileusersbasedonencounterdaysanddiscoverthattherankgroupsizefollowspower-lawdistributionthatweuseintheassignmentsofcommunitiesforautonomousnodes.Toenhancethemobilenetworkstesting,weutilizeourndingstoeffectivelycaptureandembedpersonalityofmobileusersinsimulationandtestbedenvironment.Weproposeanencounterrule-baseddecisiontomimichumanencounterpattern,whichisanimportantsteptowardefcientdesignofmobilesocialnetworkingprotocolsandservices.Withtheadditionsofgroupinformationandschedulertotherule-baseddecision,weshowthatourapproachesenableautonomousmobilenodescollectively 11

PAGE 12

mimichumanencounterpatterns.Weexperimentwithvarioustypesofdecisionmodesandcomparetheresultstorandommobilityandreal-worldnetworkingtrace.Theresultshowsthatourproposedapproachprovidestherangeofknobsforadjustingparameterstocapturepower-lawdistributionofgroupsizes,encounterratiowithgroupmembersandperiodicalencounterpatternsthatareclosetoreal-worldnetworkingtracewhilefaroutperformingrandommobility.Finally,weproposeanovelmobilenetworkingtestbedthatblendsthenetworkofautonomousrobotsandparticipatorytestingviapersonalityprole.WeimplementaprototypemobilenetworkingtestbedwithIRobotandPDAs. 12

PAGE 13

CHAPTER1INTRODUCTIONResearchinthisdissertationhasthreemaincomponents.Thethreemaincomponentsare:analysisofencounterpatterns;capturingandembeddingprolesofmobilenodes;anddesigningoftestbedformobilenetworkswithautonomousmobilenodes.Werststudytheencounterpatternofmobilenodestounderstandtheirbehaviorinwirelessnetworking.Specically,weinvestigateperiodicityandregularityoftheencounterpatternsformobilenodes.Then,westudytheencounterhistoryofmobilenodes.Basedontheresultsoftheanalysis,weintroduceencounterbasedprolingofmobilenodesusingvectorsanditsuseinencounterpatternmodeling.Withtheanalysisandproleimplementationoftheencounterpatterns,wedevelopmobiletestbedsusingourproposedprolesforautonomousmobilenodesanddiscusstheexperimentresultsinsimulation.Furthermore,weintroduceaprototypeimplementationofthemobilenetworkingtestbedwithrobotsandPDAs.Weconcludewithourndingsandcontributionsofthisdissertation. 1.1ResearchFrameworkWeintroduceourresearchframeworkinthissection.Figure 1-1 showsanoverviewdiagramofourresearchworks.Thethreesquareboxesinthetoplevelshowsthebigpictureofeachresearchwork.Eachresearchcomponenthasitsowncomponents.Thedirectionofarrowindicatestheowofthestudy.Arrowtothetopindicatesthatthesmallcomponentsmakethebigcomponent.Whereas,downarrowindicatesthesmallcomponentsarethebranchesofthebigcomponent.Werstintroducethestudyofencounterpatternformobileusers.Weanalyzethenetworktracebyperiodicity,stabilityandregularity.Basedonourstudyontheunderstandingencounterpatternofmobilenodes,thesecondcomponentworkhasstarted.Specically,webeginthestudyofprolesinmobilenodestoutilizetheresultofanalysis.Webuildanencountervectorforeachmobileuser.Weclusterthemobilenodesbasedonencountervectoranddiscover 13

PAGE 14

Figure1-1. Frameworkofdissertation.Therearethreemaincomponents:analysisofencounterpatterns,prolingbasedoncommunityandembeddingprole.Theencounterpatternisanalyzedwithsubcomponentsofperiodicity,regularityandstability.Prolingbasedoncommunitystudyisperformedbysubcomponentofencountervectorandencountergroupsizeisanalyzedusingtheencountervectorforgroupsizesubcomponent.Prolesareembeddedinthetwosubcomponents:autonomousmobilenodesandmobilenetworkingtestbed.Autonomousmobilenodesaresimulatedandimplementedinitssubcomponentsofsimulationandnetworkofrobots.Mobilenetworkingtestbedcomponentconsistsoftwosubcomponentsthatarenetworkofrobotsandparticipatorytesting.Networkofrobotsisaduplicatecomponentasautonomousmobilenodesareused.Encountervectorandparticipatorytestingaremainlyusedtosupportothercomponents;thus,lledwithalightcolor. power-lawdistributioninthesizesofthegroups.Then,wedesignanddevelopmobiletestbedsbasedontheprolingresearch. 1.2ConceptualModelWearelivinginaworldwherewirelessLAN(WLAN)accessisavailableinmanyplaces,suchasschool,airportandcoffeeshop.Proliferationofmobiledevicessuchassmartphonesandtabletsdemandsmorewirelessnetworkingareas.AdventofpopularsmartphonessuchasiPhoneandAndroidenableseasyaccesstotheinternet 14

PAGE 15

frommobiledevicesthatpeoplecarryallthetime.However,thosemobiledevicesaredependentontheinfrastructurenetworks(i.e.wirelessaccesspoint(AP)andcelltower).Mobileadhocnetworksallowthedevicestocommunicateviawirelesscommunicationwhentherearenosuchnetworkinginfrastructuresavailable.Yet,theycanalsobelimitedinthedistanceofcommunicationwheremobiledevicesaresparselydeployed.DelayTolerantNetwork(DTN)[ 2 ]isanetworkingconceptthatallowsadelayincommunicationbetweennodes.DTNcanbemanytypesofforms.Forexistingnetworkinginfrastructure,therecanbesomelinksthatareoffbutcanbeonagainaftersometime.Formobilenetworkswheremobilenodesplayaroleofrouters,theendtoendpathsmayexistwhenopportunityforcommunicationcomesinwithmobilityofmobilenodesastheybecomerelayingnodes.Consequently,withdelays,therangeofwirelesscommunicationcanbewiderwiththehelpsofmobilenodesforwardingdataormessagestothetargetnodes.Weassumethatmobilenodescanberelaynodesbystoreandforwardfashion.Theybecomeroutersinasensethattheydecideanextrouteforthedatapacketandforwardittothenextnodeorendnode.Inthisdissertation,westudythemobilenetworkswheresuchconceptscanbeapplied. 1.3MobileNetworksMobilenetworksarenetworksthatconsistofmobilenodesthatmayroamaroundandhavewirelessconnectivity.Directend-to-endpathmaynotexistinsuchanetworkbutpermissionofdelaymakesthedeliverypossibleasthenewpathsmaybecomeavailableaftercertaindelay.Forinstance,thenewlinkscanbeopenedwhenmobilenodesgetclosetoothernodes.Bluetoothcommunication(e.g.,PDA,smartphone)isanexampleofsuchnewcommunicationopportunity.Bluetoothcommunication,however,islimitedtodirectcommunication.Unlikesinglehopnetworks,weassumeanetworkwithmulti-hopdeliverycapabilityasinexistinginfrastructurenetworks,onlythedifferencebeinganodecanbebothendnodeandrelaynode.Thisassumptionmakesrobustnetworkagainstthemalfunctionsofseveralnodesandeliminatesthe 15

PAGE 16

needofinfrastructure;therefore,itisparticularlyusefulinasituationlikedisasterrelieformilitaryoperation.Anotherupsideofsuchnetworksisimplicitmulticast,whichisspreadingamessagetocertainnodesofinterestorgroupsbasedontheirlocationvisitingpreference.[ 3 ]Oneofthemainchallengesinthisnetworkisunpredictableconnectivitywithanexceptionofscheduledmove,whichisnotourfocusinthiswork.Ourinterestliesinanenvironmentwherethemostofcarriersofthemobiledevicesarehuman.Ashumanmobilityisunpredictable,socanbemessagedeliverytothedestinationnode.Toincreasethepredictionrateofconnectivityamongnodesincertainsituations,signicantamountofrecentresearcheshavefocusedonsocialaspectsofusermobility.Manyofthesestudiesproposedaroutingprotocolthatusessocialcharacteristicsofhumangroupsormobilitymodelstobeusedinsimulation[ 3 5 ].However,littleisknownabouttheencounterpatternofmobilenodes,whichisimportantinpredictingnextconnectivity. 1.4ProblemStatementsWerstdiscussthemotivationofourworksalongwithchallengesassociatedwiththem.Then,wedescribetheproblemstatements.Explanationofourapproachesinabigpicturearefollowed. 1.4.1MotivationandChallengeMotivationsofourworkarethreefolds:1)understandingmobileusersencounterpattern,2)discoveringgroupencounterpatternofmobileusers,and3)designingamobilesocialnetworkingtestbed.Fortherstmotivationofunderstandingmobileusers'encounterpattern,itiscrucialidentifyingspecicdimensiontoexplore.Inaddition,domainknowledgeobtainedfromanalysisshouldbehelpfulinapplicationofunderstanding.Thesecondmotivationbringsthechallengesofdeneanappropriategroupingvectoranddiscoveringsimilaritymetrictoclustermobileusers.Thethirdmotivationhasgreatchallengesofdesigningandimplementationofthetestbed.Tocreatearealisticmobiletestbedusinghumanmobility,itiscriticaltouseautonomous 16

PAGE 17

mobilenodethatcanmakeindependentmobilitydecisionbythemselveslikehumandoes.Deployingautonomousmobilenodesbringsanotherchallengeofemulatinghumanmobilitypatternwithoutglobalknowledge. 1.4.2ProblemStatementsTothemotivationsandchallenges,wedeneourproblemstatementsinfourcategories:1)identifyperiodicencounterpatternofmobileusers;2)discoverencounterpatternsofmobileusersusingencounterbasedcommunity;3)designautonomousmobilenodescollectivelymimickingmobileusersencounterpattern;and4)proposeanovelideatodesignamobilesocialnetworkingtestbedandshowaprototypeimplementation. 1.4.3ApproachesOurapproachestoeachproblemstatementsareasfollows.1)Periodicencounter:performspectralanalysisforWLANandBluetoothencountertracesthatreectreal-worldhumanmobility.2)Communityencounterpattern:clustermobileusersbasedonthesimilaritiesofencountervectorandembedasacommunityprole.3)Autonomousmobilenodes:embedencounterrule-baseddecisiontotheautonomousmobilenodesalongwithschedulerandcommunityprole.4)Designamobilesocialnetworkingtestbed:implementaproofofconceptontherobotsandmobiledevicesanddevelopasimulationtoolforlarge-scaleexperimentwhileshowingthemobilityofthenodesingraphicsandvalidatetheresultbycomparingtothereal-worldencounterpatternappearfromWLANtraces. 1.5ResearchComponentsInthissection,wedescribethemainresearchcomponentsindetail.Eachcomponentistightlylinkedwithstrongrelationtoothercomponents.Weanalyzetheencounterpatternsrst.Then,weproposethemethodstocapturemobileusers'behaviorandcreateprolesbasedonthem.Weproposeprolebasedmobilesocial 17

PAGE 18

networkingtestbedutilizinganalyzedencounterpatternofmobileusersandprolingmethods. 1.5.1AnalysisofEncounterPatternWeinvestigatetheencounterpatternsofmobilenodesandanalyzetheirpatterntoapplytothehumanencountermodelandmessagebundleprotocolthatusestheencounterpattern.Weusealowerboundmetrictocalculatethenumberofrequiredrelaynodesforgivensuccessratiocondition.Advantageofthismetricisthatthesourcenodecandeterminethenumberofmessagecopiesaccordingtotheimportanceofmessagewithoutcausingoverloadtothenetwork.Then,weanalyzethenetworktracesinvariousenvironmentsandshowthepresenceofstrongperiodicpatternsinnodalencounter(a.k.a.periodiclinkconnection),usingspectralanalysis.Withthisanalysis,wendtheperiodicpatternsthatarerepetitiveanddiscussitsapplication.Toachievethisgoalofstudy,weanalyzevarioustypesofWirelessLAN(WLAN)andBluetoothencountertraces.First,wegenerateencountertracewithanadequateassumptionfromWLANtrace.BluetoothtraceisnaturallyanencountertracewithoutanytransformationasitrecordedtheBluetoothdevicesidenticationeachdeviceactuallyencountered.However,thescaleofBluetoothtraceislimited;thus,wealsousetheprocessedWLANtraceforencounterpatternanalysis.Inaddition,westudytheencounterhistoryofmobilenodes.Theirencounterpatternscanbeeitherconsistentorinconsistentdependingonthebreakpointofthehistory.Wedividetheencounterhistorybytwodifferentwindowsandanalyzetheconsistencyofencounterinbothwindows.Ouranalysisresultshowsthataswehavemoreencounterhistorydataintherstwindow,theoverallconsistencyofencounterpatternimproves.However,morehistorydatadidnotaffecttotheconsistencyofencounterpatternsignicantly.Further,weinvestigatetheencounterconsistencybycontrollingthesizeofthesecondwindow.Theresultshowsthatbiggersecondwindowsizelowerstheoverallencounterconsistency.Withtheseresultofanalysis,wedesignaencounterpatternmodelofmobilenodesthat 18

PAGE 19

reectsthemetricswefound.Wealsoprovideasnapshotoftheoverlaynetworkswiththecharacteristicsouranalysisresultshows. 1.5.2ProlesofMobileNodesWiththeanalysisofmobilenodes,weaimtobuildprolesofmobilenodesbytheirencounterpattern.Twotypesofprolescanbeconstructed.Firstly,theprolecanbebuiltforaggregateencounterpatternforeachnode.Secondly,pairwiseprolescanbebuiltforencounterwitheachnode.Thelattertypeofprolescanbeheavierinsize.Wefocusonthesecondtypeofproleasitaccuratelydepictstheencounterpatternforeachuser.Weclusterthemobileusersbasedontheencountervectorwedeneandanalyzethetrend.ThisencountervectoristhesecondtypeofprolethatrequireNnumberofcolumns,whereNisthenumberofexistingnodes.Basedontheanalyzedtrace,werevealthedistributionofclustersizes,whichisabasisinassignmentofcommunitiesinthemobilesocialnetworkingtestbedwediscussinthefollowingsection. 1.5.3ProleBasedMobileSocialNetworkingTestbedSomeofthemobilenetworkingtestbedshavebeenproposedinpreviousliteratures.Thebehaviorsofmobilenodes,however,arelimitedtorandommobility.Tocreatearealisticmobilenetworkingtestbed,itisimperativetouseabehavioralmodelthatcloselyresembleshumanbehavior.Inordertoachievethisgoal,weuseaconceptofencounterprolewediscussintheprevioussection.Weembedthecreatedprolesontherobottobuildatestbedthatemulatesthebehaviorofhuman.Byimplementingtheprolesontherobots,wehavefreedomofcontrollingtheproles,yet,wealsohaveadvantageofemulatehumanmobilityonthetestbeds.However,scalabilityofthenetworksarestilllimitedtothenumberofmobilenodescreatedinthelab.Forinstance,usingmobilerobotscostssignicantlyhighifpurchasedmorethanhundreds.Forsuccessfuldeploymentofmobilenetworking,itisessentialtohavelargenumberofnodesfornetworkingpurpose.Weadoptaconceptofparticipatorytestingtoimprovethescalability.Withthisapproach,voluntaryhumanparticipantscarryingsmartphones 19

PAGE 20

becometestsubjects.Thiscreatesahugetestbedasthesizeofthetestbedcanbeaslargeasthenumberofparticipants.Efcientrecruitingstrategyisakeyindrawingmoreparticipants.However,itisoutofscopeofthisdissertation;thus,ourfocusisonproposingtherealistictestbeds,evaluatingthetestbedviasimulationandshowingtheprototypeofthetestbeds.Ourtestbedsprovidethecontrollableprolesthatreecthumanbehaviorwithscalabilitybycrowdsourcing. 1.6ContributionOurcontributionsincludetwoareas:effortcontributionandintellectualcontribution. 1.6.1EffortContribution1)Webuildatracelibraryandprocessthetracesforthepurposeofencounterpatternanalysis.2)Wedevelopasimulationtooltovisualizetheencounterpatternofmobilenodesandgenerateencounterstatisticsforencountervectorandtime-seriesdata.3)Webuildaprototypemobiletestbedforproofofconcept,whereweembedanemulationproleonrobotnodestomimichumanencounterpattern.4)Weevaluatetheautonomousmobilenodesforcommunityencounterpattern,groupdistributionandperiodicencounterpatternwithvariousencountermetricsofencounterdays,frequencyanddurationviasimulation. 1.6.2IntellectualContribution1)Ourndingfromperiodicityanalysisprovidesinsightintotheperiodicalnatureofencounterpatternformobileusers.2)Wedeneencountervectoranddiscoverpower-lawdistributionofmobileencounterclusters.3)Weproposeaencounterrule-baseddecisionforautonomousmobilenodeswhichcollectivelyemulatecommunityencounterpatternfoundfromthenetworkingtraces. 20

PAGE 21

CHAPTER2RELATEDWORKWediscussrelatedresearchworksintheliterature.Werstintroducetheotherresearchworksthatusesrelateddenitions,assumptionsandnetworkenvironments.Inaddition,wediscusstheprojectsandliteraturesthatincludethedatasetsweuse.Wefurtherdiscusstheresearchworksthatstudythemobilenetworkingbydescribingtherelevantstudiesofanalysisofmobilenodes,modelingandprotocolsinmobilenetworking.Weintroducetherelatedtestbedsprojectsandliteraturesofmobilenetworksaswell. 2.1AnalysisofEncounterPatterninMobileNetworksTheadventofDelayTolerantNetworks(DTN)makesthemobileadhocnetworkstohavebroaderconceptofnetworks.Withitsmobilitycharacteristicsinmobileadhocnetworks,mobilenetworksbecomeavailablewithoutexistinginfrastructure.Allowanceofdelayencompassesthemobileadhocnetworksnotonlytolargerspatialcommunicationspacebutalsotolongertemporalcommunicationspace.WithsimilardenitionofDTN,intermittentconnectivityandopportunisticnetworksalsoplaythesamerole;thus,thesearethebasisofthenetworksenvironmentwestudy.PocketSwitchedNetwork[ 6 ]isanetworkwherecommunicationisperformedamongonlythemobiledevices.Thisisalsoasimilarenvironmentwithourfocusofstudy;however,PocketSwitchedNetworksdiffersinthatitisforonlyamongthemobiledevicescarriedbyhuman.Whereas,thenetworksofstudyinthisdissertationincludesthestatisticnetworkssuchasnetworkinfrastructures.[ 7 ]analyzethenetworktrafcandrevealedthepresenceofcombinationofperiodicityinthecaseofdenialofserviceattacksontheinternet.TheyapplypowerspectralanalysisthatappliesACFtothetimeseriesdatabeforetransformingtofrequencydomain.Thisremovessynctermsthatmightappearforaniteset,therefore,weadaptedasimilarapproachratherthanapplyingDFTtothedatasetsdirectly.Kim 21

PAGE 22

etal.studytheperiodicpropertiesofWLANusersassociationwithaccesspoints[ 8 ].TheymeasurethediameterofvisitedAPsforhighlymobileuserswhosemaximumdiameterwithinanhouris100meterormorefromtheDartmouthcampusWLANdata.Theresultshowsthestrongpresenceofperiodicityofdiameter,particularly24hourand1week,fortheselected360users.Periodicpropertiesoftraveldistanceforhighlymobileusersareinterestingndings.ThisistheclosetworktoourfrequencyanalysisinthatitusesDFT(DiscreteFourierTransform)toanalyzetheperiodicityofusermobility.However,weinvestigatethedifferentaspectofusermobilitypattern,namely,theirencounterpattern.Further,ACF(AutoCorrelationFunction)isusedbeforeapplyingDFTtoremovesynctermsthatmightappearforaniteset;additionally,weanalyzetherichdataset,includingover10,000usersperWLANtraceandBluetoothtrace. 2.2MobileSocialNetworkingProtocolsManyofthestudiesinmobilenetworksweredevotedinrouting.Specically,largeportionofmobilenetworksroutingstudyfocusedonusinghumanmobilityforroutingmessagesanddata.Themaincharacteristicsusedinmobilenetworksroutingfromsocialnetworksarecommunity,mobilityandencounterofhuman.Tousesuchcharacteristicsinrouting,deepunderstandingofhumanbehaviorisessential.Gonzalezetal.[ 9 ]hasshownthatindividualhumantendstofollowsimplereproduciblepatternsbasedonthecellphoneusertraces.Hsuetal.proposedtime-variantcommunitymodel[ 10 ]thatreectstheperiodicencounters.Communitystructureofsocialnetworksfromnetworkingtracesarediscussedin[ 11 ].SocialrelationshipbetweenmobilenodesforDTNroutingisdiscussedin[ 4 ][ 5 ].Miklasetal.[ 12 ]dividedhumanencounterstofriendsandstrangersaccordingtothelengthofencounter.Theseworksstudyhumanencounterpattern;however,wearethersttoanalyzetheperiodicityofhumanencounterextensivelybyspectralanalysis.Prophet[ 13 ]isoneoftherstroutingalgorithmsusingencounterhistoryinDTN[ 2 ].Itusesencounterfrequencytodeterminearelaynode.Thechosennodewillforwardthemessagebundletotheencountered 22

PAGE 23

nodeanddelegatetheresponsibilityofdeliverytothenodeifthenodehashigherencounterfrequencytothedestinationnode.However,itisunclearhowtodeterminetheprobabilityofencounterusingfrequency.Touseencounterfrequencyforprobability,totalnumberofpossibleencountersshouldbeknown,whichotherwisecouldbeinnite.Further,itispossiblethatfrequentencountersinshorttimecanmisleadpredictionoffutureencounter.Timely-countprobability[ 14 ]isanideathatregardsencountersbelongingtothesameintervalasoneencounter;thus,itprovidesstandardproceduretocalculatetheencounterprobability.Inourwork,weuseditforanideaofdailyencounterwhoseintervalisaday.OuranalysispartofthestudyinthisdissertationcanbethebasistoimprovetheperformanceofprotocolsinDTN.Prole-cast[ 3 ]isaforwardingprotocoltothegroupofnodes,sharingthesameinterest.Ithastwowaysinmessagedistribution:1)sourcenodepropagatesthemessagetothenodewithsimilarproletothedestinationgroup,whichmayalsoforwardtoothernodesthataremoresimilar(gradientascending);2)thesourcenodecantrytodistributethemessagetothenodesthathavedifferentproles,suchthatitsdifferentmobilitycangivemorechancetoreachthedestinationgroup.Bothprole-castandprophetprotocolscanenhancetheirstabilityofdeliverybyincorporatingperiodicpropertiesofencounterforconsistentpredictability.Studiesforpredictabilityofhumanmobilityandencounter[ 15 ][ 16 ]canalsobeextendedbyconsideringdifferentperiodicpropertiesaswellaswormpropagationpatternviahumanencounter[ 17 ][ 18 ].Recentstudiesshowtheapplicationsofusinghumanmobilityandencounterformessagepropagation[ 19 ][ 13 ][ 20 ][ 21 ][ 22 ].MobiTrade[ 22 ]isimplementedonAndroidphonesforsharingthedatawithincentive.Humanmobilityinthemeparkisimplementedonmessagedeliveryin[ 19 ].Popularityoflocationsisshowntoinuenceroutingdecisionforefcientdeliveryofmessageusinghumanmobility[ 21 ]. 23

PAGE 24

2.3MobileNetworkingTestbedsThereareresearchprojectstodesignanddevelopmobilenetworkingtestbeds.Weexplainthetestbedsthatarerelatedtoourworks.MiNT[ 23 ]isaminiaturizednetworktestbedthatsolelyusesiRobotsinacontrolledspace.AservercomputercontrolsthemovementsandcommunicationamongstiRobotsthatareequippedwithWLAN.AlthoughthistestbedcanexpandwithmultiplenumbersofiRobotsandbeeffectiveinexperimentsforsmall-scalemobileadhocnetworks,itstillsuffersfromscalabilityandadiversityofnodes.RoombaMADNeT[ 24 ]showedthecapabilityofusingiRobotforDTN.TheresearchersmountedawirelessrouterthatrunsonLinuxbyconnectingthroughmodiedserialcable.Thisprocessmighttakeadvantageofacostumedlightweightprogrammingboardtoutilizethewirelesscommunicationfeaturespecicallyforthetestingpurpose.However,thisprocesscanbetediousandcumbersometomanyresearcherswhoarenotskilledinthisarea.Ourmethodissimpleandusestheexistingdevice.Connectionrequiresaminimumofstepsandeffort:eitherBluetoothpairingorconnectingaserialcabledirectlywithadistantorattachedcomputer.In[ 25 ],theauthorsproposedaDTNtestbed.TheyusedenclosurestocontainthelaptopcomputersandmeasurethesignalattenuationforimplementedDTNprotocol.Thedesigniscentralizedanduserscanviewthewirelessnodesmovingaroundfromtheservercomputer.Mobilityislimitedtoacontrolledenvironment,astheparticipantsaretofollowthegivenpathsandrequiredtobeintheexperimentrange.MobileEmulab[ 26 ]iswirelesssensornetworkstestbedthatmanagesitsMiCa2motebasedrobotnodesfromacentralcomputerwithvideocamera.GUIinterfaceprovideslocationprecisionof1cmformovingrobotnodes.Inourpresentation,nodescanhavecompletecontroloftheirmovements,includingmessagepropagationdecisions.Moreover,measurementsaredecentralizedinourexperiment,aseachmeasurementrecordiskeptinsidethemobilenodes.SCORPION[ 27 ]isaheterogeneousnetworkingtestbed,whichusesiRobots,Buses,Aircraftsand 24

PAGE 25

humanswithbriefcasenodes.Itprovidesatestbedtoexperimentcommunicationbetweendiversemovements;however,mobilityofallthemobilenodesarelimitedtocontrolledmovements;thus,socialaspectsarenotreected.Ouruniquecontributioncomesfromtheautonomousrobotswithbehavioralprolesandparticipatoryhumansthatprovidesuncontrolled,thus,realistictestingenvironment. 25

PAGE 26

CHAPTER3MOBILITYTRACESDATAItiscrucialtohaveextensiveandrealisticdatasetsfortheanalysisofmobility.Weintroducethedatasetsweusedandcollectedforouranalysisalongwithdenitionsandassumptions.Inaddition,wedescribethedetailsoftransformingthenetworkdatasetsintotheencountertracesforouranalysis. 3.1BasicDenitionsBeforeweexplainthenetworktraces,werstdescribethebasicterminologiesthatweuseinourwork.Mobilenodeisanentitythatcanmovearoundwithdifferentspeedsindifferenttimeandspace.Mobilenodeiscapableofwirelesscommunication.Mobilenodescanencountereachotherwhentheyarecloseenoughtodiscovereachother.Specically,thetermencounterinthisworkindicatestheeventthattwoormoremobilenodespresentwithinthewirelesscommunicationrange.Theterms,encounterandcontact,areusedinterchangeablyinliteratures[ 1 ][ 28 ]andweusethetermencounterthroughoutthepaperforconsistency.Mobilenodeshaveon-lineandoff-linebehaviors.Inoff-linemode,othernodesmaynotdiscoverthemobilenodeseveniftheyareintheproximityofwirelesscommunicationrange.ThesemobilenodescanbethesmartphonesorPDAshumancarryaroundorwirelesscommunicationdevicesthatmovearoundorstayinstaticlocations. 3.2NetworkTracesForaccurateanalysisofencounterpatternsformobilenodes,itisessentialtohavelargedatasets.Therearetwotypesofacontacttrace.Firstapproachisobtainingatracebythesynthetictrace.Thesynthetictraceisaarticiallymadetracebasedonmobilitymodel.Theadvantageofusingsuchtraceisfreedomofmanipulatingthedatatotthepurposeofanalysis.However,thesynthetictraceislimitedinthatitisbasedontheassumptionsandobservedparametersfrompreviousanalysis;thus,itsclosenesstorealitydependsonthequalityandquantityofsamplesusedinanalysis 26

PAGE 27

Table3-1. Statisticsofencountertraces. TraceTraceAnalyzedUniqueusersEncounterpairssourcedurationduration USC2006Jan-May128days28,17325,359,4542007Jan-May35,27419,057,0892008Jan-May42,58731,289,100UF2007Aug-Dec128days46,11512,493,4032008Jan-May50,54916,807,427Monteral2004Aug-Dec128days4552,512Bluetooth2/25-3/7/2008256hours101,27711/17-27/2008271,655Long-termBluetooth2010/9-2011/4180days127,870 forthemodel.Secondapproachisusingarealworldtrace.Therealworldtracecanbedividedintothetracebythemobilerobots,humanmobilityandhybridofboth.Byusingtherobots,researcherscancontrolthemobilityofeachrobotforthepurposeoftheirexperiments.Thisissimilartothesynthetictracebutitdiffersinthatitcanbeexperimentedinrealworldenvironmentasopposedtoarticialenvironmentinsynthetictrace.Collectinghumancontactpatternislimitedinsizetotheparticipantswillingtofollowtheinstructions.Table 3-1 showsthetracesweuseinouranalysis.Privacyissuesofthetracesareoutofscopeofthisdissertation.Interestedreadersregardingtheanonymityoftracesandprivacyissuesareencouragedtoread[ 29 ][ 30 ][ 31 ][ 32 ][ 33 ].TheWLANtracesarepubliclyavailablefordownloadat[ 34 ][ 35 ].Bluetoothtracesareobtainedbyclassstudentsandobtainedwithconsentbyparticipantsforresearchpurpose. 3.3TransformingNetworkTracestoEncounterTracesWeusethetwotypesofdatasetsfornodalencounter:BluetoothtracesandWLANtraces.BluetoothtracesreecttheencountersofuserscarryingmobiledeviceswithBluetoothcommunicationcapabilities.ThelimitationofBluetoothtracesisthescalability,becausethedatasetislimitedtothenumberofparticipantswillingtoruntheBluetoothdiscoveryprogram.Whereas,WLANtracescanbeverylargeasthecentralizedsystemcancontinuouslycollectthedataviaaccesspointsbelongingtotheparticular 27

PAGE 28

organization(e.g.,collegecampus).Asdiscussedearlier,weuseanassumptionthattheuserswhoaccessedtothesameaccesspoints(APs)haveencounterevents.Thisassumptionmaynotreecttheexactencounter;however,itisclosetorealencounterconsideringtheusersaccessedthesameAPswereatthecloseproximityofeachotherandcouldhavecommunicatedeachotherthroughtheAP. 3.3.1BluetoothEncounterTracesScaleofBluetoothencounterdataisconsiderablysmall,comparedtoWLANtracesduetothedifcultyofndingsubjectstoparticipate.SomeoftheavailableBluetoothencounterdataincludetheconferenceencounter[ 36 ]andbusencounter[ 37 ][ 38 ].Whilethesedatasetsmaybeusefulforparticularscenarios,weconductedourownexperimenttoobservegeneralBluetoothencounter,whichmatchestotheWLANtracewealsocollect.EachofgraduatestudentstakingtheComputerNetworkingcoursein2008wasassignedaPDA(HPiPAQorNokiaN800/810)andwasstronglyencouragedtocarrythemobiledeviceasoftenaspossiblewiththeBluetoothencountercollectionprogramrunning.Thisprogrambroadcaststhebeaconsignalevery60secondsandlogstheBluetoothdeviceinformationthatacknowledgesthebeaconsignal,includingthetimestamp.Thisexperimentwasperformedfortwosemesters(2008springandfall)[ 35 ],eachwithdifferentgroupsofstudents.Duetotheshortlengthofexperiment,weobservedhourlyencounterinsteadofdailyencounter.AsTable 3-1 shows,thereare10and27subjectsinspringsemesterandfallsemesterrespectively.ThesecollectedBluetoothtracescontaintheinformationoftheencounterednodes,namelytheirMACaddressesandtimestampsforacknowledgements. 3.3.2WLANTracesTherearemanyformsofnetworktracesavailableinpublic,whichcanbeobtainedfrom[ 34 ],includingthecityofMontrealtrace[ 34 ]thatweuseinthispaper.Toobtainlargescalenetworktracesthatcovertheentirecampusoveralengthofmorethanoneacademicsemester,wecollectedcampus-wideWLANtracesattheUniversityof 28

PAGE 29

Table3-2. FormatofWLANtraces MACAccessPointTimestampDurationofAssociation ab-cd-34-4d-23-12lbw343-win-ap1200-1120188930034003a-3d-4f-fe-ca-b8lbw343-win-ap1200-112018894741200 Florida(UF,2007fall-2008springsemester)[ 35 ].WealsousedtheWLANtracesoftheUniversityofSouthernCalifornia(USC,2006-2009springsemesters)[ 35 ]asinFig. 3-1 .TheseWLANtraceshavethefollowinginformation-MACaddress,associatedAPandtimestampforstartandendtimeofassociation.Basedontheassumptiondescribedearlier,theWLANtracesareprocessedtoencountertracesthatlogtheMACaddressesoftheencounteredWLANdevices,timestamps,durationsandlocationsofencounter.Conversiontotheencountertraceiscomputationandstorageconsumingprocess.Givenminputsinaveragefornnumberofnodes,thecomputationneededisn(n)]TJ /F6 7.97 Tf 6.58 0 Td[(1) 2m2asitrequiresthecomparisonforeachinputbetweentwonodestodeterminetheoccurrenceofencounteranditsduration.Therefore,weobtainO(n2m2)foroverallcomputationtimeingeneratingencountertraceandO(n2m)forthesizeofencountertracedata.Iftheorignaltraceissortedintimesequence,thecomputationreducestoO(n2m)becausethecomparisonfortheinputsoftwonodescanbeperformedinsequenceofproceedingtime.Forthesizeofverylargedata,whichhasatleastover28,000nodeswiththeinputsranginguptoseveralmegabytesforamonthlydata,itisrealistictobreakdowntheentiretracebythecertainperiodsforanalysispurpose.Weanalyzethe128daysofdatafromeachtracefortheabovereasonsandconsistencyincomparison.Further,thisspecictimespanroughlycoverstheentiresemesterforbothcampuses.OriginalformatofWLANtracecontainsmuchinformation.TogenerateencountertraceweonlyneedtouseIDofanuser,hisassociationtimewithaccesspointsandtheIDofaccesspoints.Therefore,weprocessthetracetosimpleformatthatcontainsonlythenecessaryinformationasshowningure 3-2 .UserIDisidentiedbytheMACaddresstheirmobiledeviceshas.Timestampisthetimeauserstartstoaccessthe 29

PAGE 30

Table3-3. Formatofencountertrace SourceMACEncounteredMACTimestampEncounterduration ab-cd-34-4d-23-123a-3d-4f-fe-ca-b812018894741200 WLANaccesspoint.Eachaccesspointhasuniqueidentierthatsometimeschangeafterasemesterisover.Thus,weinvestigateeachsemesterseparately.Numberofaccesspointsalsoincreasesovertimeaswell.USCtracecontainsaround200APsandUFtracecontainsover500APs.DurationofassociationisthetimeauserspendsatcertainAPs. 3.3.3TransformedEncounterTraceTable 3-3 showsthetransformedencountertrace.ThisformatappliestotheBluetoothtraceaswell,whichdoesnotrequiretransformation.BasedonthetimestampandAPthatduplicatebetweentwomobilenodes,webuildanencountertracewithsourceMACandencounteredMAC.Thistransformedtraceisthetraceweuseinthisdissertation. 30

PAGE 31

CHAPTER4UNDERSTANDINGENCOUNTERPATTERNSOFMOBILENODESItisessentialtounderstandtheencounterpatternsofmobilesnodesforanyoftheprotocolsusingmobilenodesasforwardingnodes.Weanalyzetheperiodicityofmobilenodesandproposeheuristicapproachestondregularlyencounteringpairs.Furthermore,westudytheencounterhistoryofmobilenodes.Thisanalysisisthecriticalbasisfordesigninganencounterpattenmodelformobilenetworkingprotocolsandservices. 4.1IntroductionMobilityandnodalencountersareutilizedtodelivermessagesinintermittentlyconnecteddelaytolerantnetworks(DTNs)[ 2 ].MuchofDTNresearchsofarhasbeendevotedtothestudyofmessagedeliveryprotocolsandthedesignofmobilitymodels.WhilethesestudiesareessentialforeventualimplementationofthemobileadhocnetworksusingDTNconcept,understandingofnodalencounterpatternisacriticalbasisforthesuccessofprotocoldeploymentasdeliverymechanismdependsonnodalencounter.Inthispresentation,weexploretheperiodicitypresencesinencounterpatternsandanalyzethem.Bythespectralanalysisofencounterpattern,wendtheperiodicpatternsthatarerepetitiveanddiscusstheirapplicationslater.Toachievethisgoalofstudy,weanalyzedvarioustypesoftheWirelessLAN(WLAN)andBluetoothencountertraces.First,wegeneratetheencountertraceswithareasonableassumptionfromWLANtraces.BluetoothtracesarenaturallyencountertraceswithouttheneedofanytransformationastheylogtheidenticationoftheBluetoothdevicethatthesubjectBluetoothdeviceshavediscovered.However,thescaleofBluetoothtracesarelimitedtothenumberofsubjectscarryingthedevcieswiththediscoverprogramon.Hence,weusetheWLANtracesforscalableanalysisofencounterpattern.InordertousetheWLANtraceasencountertrace,weusecommonassumptionthathadbeenusedbyotherpublications[ 1 ][ 39 ],whichdenestheencounteroccurrenceintheWLAN 31

PAGE 32

environmentasthenodesthatareassociatedwiththesameaccesspoints(APs)inthesameperiodoftime.Aftertransformationtotheencountertrace,thenextstepisgeneratingatimeseriesdataforthenumberofmetrics,namely,daily/hourlyencounter,encounterfrequencyandencounterduration.WeapplytheAutoCorrelationFunction(ACF)toidentifytherepetitivepatternsandperformpowerspectralanalysistondthedistinctperiodicitiesinencounterpatternsforeachmetric.FastFourierTransform(FFT)wasperformedinconversiontofrequencydomainforcomputationefciencyandanalyzesthefrequencymagnitudeinthespectrum.Wehighlighttheimportantperiodicitiesbydifferentgroupsanddiscusstheutilizationoftheresultinthemobilenetworks.Afteranalyzingtheperiodicity,weshowsomeofapproachestoextracttheperiodicallyencounteringnodepairsandconcludewiththesummaryandapplications.Inthefollowingsection,weintroducethemethodologytoanalyzetheperiodicityalongwiththeencountertracesinsectionII.AnalysisofperiodicityfortheencounteredpairsandindividualencounterpatternsinWLANandBluetoothtracesfollowinsectionIII.Then,sectionIVdescribestheapproachestoextractregularlyencounteringnodepairsanddiscusstheresults.WeexplainaboutrelatedworksinsectionV,andwrapupwithconclusionsandsummaryinsectionVI. 4.2RelatedworkManyofstudiesonDTN/opportunistic/intermittentconnectivityroutingweredevotedinusingthesocialaspectsofnetworks,suchascommunity,mobilityandencounter.Gonzalezetal.[ 9 ]hasshownthatindividualhumantendstofollowsimplereproduciblepatternsbasedonthecellphoneusertraces.Hsuetal.proposedtime-variantcommunitymodel[ 10 ]thatreectstheperiodicencounters.SocialrelationshipbetweenmobilenodeswerediscussedforDTNroutingin[ 4 ][ 5 ].Miklasetal.[ 12 ]dividedhumanencounterstofriendsandstrangersaccordingtothelengthofencounter.Theseworksstudyhumanencounterpattern;however,wearethersttoanalyzetheperiodicityofhumanencounterextensivelybyspectralanalysis.Prophet[ 13 ]isoneoftherst 32

PAGE 33

routingalgorithmsusingencounterhistoryinDTN[ 2 ].Itusesencounterfrequencytodeterminearelaynode.Thechosennodewillforwardthemessagebundletotheencounterednodeanddelegatetheresponsibilityofdeliverytothenodeifthenodehashigherencounterfrequencytothedestinationnode.However,itisunclearhowtodeterminetheprobabilityofencounterusingfrequency.Touseencounterfrequencyforprobability,totalnumberofpossibleencountersshouldbeknown,whichotherwisecouldbeinnite.Further,itispossiblethatfrequentencountersinshorttimecanmisleadpredictionoffutureencounter.Timely-countprobability[ 14 ]isanideathatregardsencountersbelongingtothesameintervalasoneencounter;thus,itprovidesstandardproceduretocalculatetheencounterprobability.Inourwork,weuseditforanideaofdailyencounterwhoseintervalisaday.OurperiodicityandregularityanalysiscanbethebasistoimprovetheperformanceofprotocolsinDTN.Prole-cast[ 3 ]isaforwardingprotocoltothegroupofnodes,sharingthesameinterest.Bothprole-castandprophetprotocolscanenhancetheirstabilityofdeliverybyincorporatingperiodicpropertiesofencounterforconsistentpredictability.Studiesforpredictabilityofhumanmobilityandencounter[ 15 ][ 16 ]canalsobeextendedbyconsideringdifferentperiodicpropertiesaswellaswormpropagationpatternviahumanencounter[ 17 ][ 18 ].[ 8 ]studiedtheperiodicpropertiesofWLANusersassociationwithaccesspoints.TheymeasuredthediameterofvisitedAPsforhighlymobileuserswhosemaximumdiameterwithinanhouris100metersormorefromtheDartmouthcampusWLANdata.Theresultshowedstrongpresenceofperiodicityofdiameter,particularly24hoursfortheselected360users.ThisistheclosestworktoourfrequencyanalysisinthatitusesDFTonthetimeseriesdatatoanalyzetheperiodicityofusermobility.[ 7 ]analyzedthenetworktrafcandrevealedthepresenceofcombinationofperiodicityinthecaseofdenialofserviceattacksontheinternet.TheyappliedpowerspectralanalysisthatappliesACFtothetimeseriesdatabeforetransformingtofrequencydomain.Thisremovessynctermsthatmightappear 33

PAGE 34

foraniteset,therefore,weadaptedasimilarapproachratherthanapplyingDFTtothedatasetsdirectly. 4.3UnderstandingPeriodicityandRegularityofMobileNodesWediscussthemethodologyandanalysisresultsinthissection. 4.3.1MethodologyForspectralanalysisoftheencountertraces,multiplestepsarerequired.Inourwork,rawnetworktracesareprocessedtotheencountertracesintheformoftimeseriesdata.Weapplytheautocorrelationfunction(ACF),andthentransformthemtothefrequencydomainbyperformingthediscrete-timeFouriertransform(DFT).Tocapturethemultipleaspectsofencounterbehavior,welookintothefollowingvariables:encounterfrequencyFd(i,j),dailyencounterEd(i,j),houlyencounterEh(i,j)anddurationofencounterLd(i,j),wheregivennnumberofnodes,(i,j)istheencounterednodeiandj(0i
PAGE 35

Figure4-1. Exampleencountertraceforamobileuserwithoneoftheencounterednodes hours,respectively.Asnotedearlier,128daysarethetimespansforeachofthetraceinthisanalysis.ThistimespansarealsobenecialtotheuseofFastFourierTransform(FFT)infrequencyanalysisbecauseitrequiresthelengthofdatatobethepoweroftwo.ApplyingFFTforeachsemesterdataenablesfastprocessingofmassiveencounterdataandhelpsobservingdistinctcharacteristicsinanergranularitybypreventingtheseasonaleffect(repeatedbehaviorbyeachsemester)fromaffectingtheresult.Figure 4-1 showsanexampleencountertraceofamobileuseranditsencounterhistorywithoneoftheencounterednodesfordailyencountermetric,Ed(i,j).Inthegure,boxedperiodisthedaysthetwonodesencounteredeachother,thus,settingavalueofoneforEd(i,j).Eachuserhasanumberofsuchtracesdependingonthenumberofencounterednodes.ACFisameasureofcorrelationbetweenobservationsatdifferentlags(distances)apart[ 40 ],thusprovidinginsightintothestreamofdata.WeuseACFtondtherepetitiveperiodicalpatternsfromtheprocessedtime-domainrepresentationofencountertraces.Whenlagk=0,itcomparesthedatastreamtoitself,andautocorrelationismaximum,whichresultsinVariance().Giventhemeanofapair(i,j),wecalculateautocorrelationcoefcients(autocoefcients)forthepair,rk(i,j)foreachlagk(1k
PAGE 36

Ararelyencounteringpairs,0.1Drate<0.2 Bfrequentlyencounteringpairs,0.5Drate<0.6Figure4-2. Averageautocorrelationcoefcient Afterproducingtheautocorrelationcoefcients,weobtainaresultshowingperiodictrendsforcertainlagsasinFigure 4-2 .Yet,severaldistinctperiodicitiesarehiddenandrequireafurtherprocessing.ApplyingDFTisthenecessaryprocessinordertotransformtheautocoefcientsofthetimeseriesdatatothefrequencydomain.Itproducesthepowerspectrumycofthepair(i,j)foreachfrequencycomponent 36

PAGE 37

c(1c
PAGE 38

ADailyencounterforrarelyencounteringpairs BDailyencounterforfrequentlyencounteringpairsFigure4-3. Dailyencounter:normalizedfrequencymagnitudeoffrequencycomponentsforencounteredpairs(i,j)(0i
PAGE 39

AEncounterfrequencyforrarelyencounteringpairs BEncounterfrequencyforfrequentlyencounteringpairsFigure4-4. Encounterfrequency:normalizedfrequencymagnitudeoffrequencycomponentsforencounteredpairs(i,j)(0i
PAGE 40

AEncounterdurationforrarelyencounteringpairs BEncounterdurationforfrequentlyencounteringpairsFigure4-5. Encounterduration:normalizedfrequencymagnitudeoffrequencycomponentsforencounteredpairs(i,j)(0i
PAGE 41

Montrealtrace.Figure 4-3B showsthatthehighestspikeappearsatthefrequencycomponentof2forthefrequentlyencounteringpairs.Thisimpliesthattheencountereventsmayhavetwobigwavesbutitstillshowsthepresenceof7daycycle.Thepresenceofstrongweeklypatterninencounteredpairsisaninterestingresultas[ 8 ]showedweeklymobilitypatternwasnotamongthedominanttrendsofmobileusers'mobilitydiameter.Considerthelogarithmicnatureofencounterratethatthelargenumberofpairsencounteredlessthan20percentofdaysover128days.Existenceofweeklypatternforthepairswithlowencounterrateisparticularlyimportantinmessagedelivery.Choosingarelaynodeisaharddecisioninacasewheremajorityofnearbynodesencounteredinfrequentlywiththedeliverytarget.However,thenodesthatshowtheconsistentencounterpatternsuchasweeklyencounterwiththetargetofinterest,wouldlikelyprovidemoreaccurateestimationfordeliveryprobability.Withthelowererrormargin,thesourcenodeorintermediatenodecanfurthercalculatetherequirednumberofrelaynodestosatisfythegivendeliverysuccessrate.Moreover,thisimpliesthatthethresholdcriteriacanmeasuredaccordingtotheimportanceofmessage.Althoughmanyothermessageforwardingschemescanbedevelopedbasedontheperiodicencounterpattern,ourfocusisontheanalysisthatcanprovidemorebasisforsuchapplications.InMontrealtrace,outstandingspikesarehardlyshownexceptintherstfrequency,whichcouldsuggesttheburstencounterpatternbutthemainreasonistheveryfewnumberofpairshaveencounteredrepeatedly.Notethattherewerenopairs,0.1
PAGE 42

AHourlyencounter BHourlyencounterfrequencyFigure4-6. HourlyencounterforBluetoothpairs:frequencymagnitudeoffrequencycomponentsforhourlyencounteratUF08spring/fallBluetoothtrace 42

PAGE 43

4.3.2.2BluetoothtraceWestudythehourlyencounterforBluetoothexperimentduetotheshortlengthoftheBluetoothexperiment.Thedifferencefromobservingdailyencounterisgranularityofobservationisner.Inthisexperiment,welookatthe256hours,whichisapproximately10days.Figure 4-6A showsthehourlyencounterpatternsofencounteredpairswith0.2Drate<0.3.IntheFigure,X-axisindicatesthefrequencyofcyclesfortheexperimentperiodinhours.GivenDratewasselectedbecause0.1Drate<0.2wasunavailableduetoexperimentlengthlimitation.Accordingtothegure,24-hourperiodicityisstrongestinbothoftheBluetoothencountertraces.HourlyencounterfrequencyinFig 4-6B displaysstrongerperiodicpatternbutitisstillsimilartohourlyencounter.Thegraphsindicateperiodicencounteroccursaroundevery24hourinaverage;thus,suggestthatmostoftheencountereventsmayoccurduringthesimilartimespanoftheday.This24hourperiodicityinencounterpatterncorrespondstotheresultinmobilitydiameterstudyin[ 8 ]. 4.3.3PeriodicityofIndividualEncounterPatternTheperiodicityintheencounterpatternofindividualnodeisevenstrongerthanintheencounterpatternofpairs.LetDrate(i)beadailyencounterrateforanodei,suchthatDrate(i)=PNd=1Ed(i) N,whereEd(i)is1ifatleastoneencountereventoccurredonthedayiand0otherwise.NotethatpeakforweeklyencounterinencounterpatternofeachnodeinFigure 4-7A ismoredistinctthanthepeakineachencounterpairinFigure. 4-7B .Thisimpliesthataggregateencounterbehaviorofeachnodeismoreperiodicalthantheencounterbehaviorofpairs;thus,consistentlymorepredictable.However,forthepurposeofmessagedelivery,understandingtheencounterbehaviorofpairsismoreusefulthanstudyingtheencounterbehaviorofindividualnode.Thisisbecausethesourcenodewillmakeadecisionofselectingarelaynodebasedontheinformationaboutencounterbehaviorofthecandidatenodetothedestinationnoderatherthanitsoverallencounterbehaviorwithallnodes.Inthegure,periodicencounterpatternis 43

PAGE 44

ADailyencounterforrarelyencounteringpairs BDailyencounterforfrequentlyencounteringpairsFigure4-7. Dailyencounterforindividualnodes:normalizedfrequencymagnitudeoffrequencycomponentsforindividualnodes'encounterpattern.Rareencounter:0.1Drate<0.2Frequentencounter:0.5Drate<0.6(FrequentencounterofMontrealtrace:0.2Drate) 44

PAGE 45

AEncounterfrequencyforrarelyencounteringpairs BEncounterfrequencyforfrequentlyencounteringpairsFigure4-8. Encounterfrequencyforindividualnodes:normalizedfrequencymagnitudeoffrequencycomponentsforindividualnodes'encounterpattern.Rareencounter:0.1Drate<0.2Frequentencounter:0.5Drate<0.6(FrequentencounterofMontrealtrace:0.2Drate) 45

PAGE 46

AEncounterdurationforrarelyencounteringpairs BEncounterdurationforfrequentlyencounteringpairsFigure4-9. Encounterdurationforindividualnodes:normalizedfrequencymagnitudeoffrequencycomponentsforindividualnodes'encounterpattern.Rareencounter:0.1Drate<0.2Frequentencounter:0.5Drate<0.6(FrequentencounterofMontrealtrace:0.2Drate) 46

PAGE 47

mostprevalentandconsistentinthesequenceofdailyencounter,encounterfrequencyandencounterdurationatbothencounterpatternsinthepairsandindividualnodes.Wealsoobservedperiodicencounteroccurrenceatevery24hourfromhourlyencounterpatterninbothoftheWLANandBluetoothtracesforindividualencounter.Naturally,wecaninferthatthenodesarepenchanttoencounterinsimilarhoursofthedayeachday.Notethatinthegure,astheshapeoftheconcaveiswider,periodicnatureislessaccurateorhaswidermarginoferror.Thenarrowerandhigherthebellshapeofpoweris,thestrongertheperiodicpropertyis. 4.3.4RegularEncounterToutilizetheperiodicpropertiesofencounterpairs,itisessentialtodevelopaschemetodiscoversuchpairsinsuccessofforminganetworkamongthosenodes.Withthetransformeddatainfrequencydomain,itissimpletoextractthepairsthatencounterconsistentlyinaperiodicfashion,whichwedeneasregularlyencounteringpairs.Thechallengeinregularityisthatitcanhavemultiplevariablesinit.Severaltrendscanbehiddenandnoisemayinterferefromobservingsomeoftheperiodicities.WediscusstheseveralapproachesthatextractregularityfromthetracesandpresenttheresultforUSC'06trace. 4.3.4.1TopfrequencyThelargefrequencymagnitudeattheleftmostfrequencyinaxis-Xsuggeststhattheencountereventsarelikelyoccurringataconcentratedtime,formingonebigwave.Conversely,thelargefrequencymagnitudesathighfrequenciesinX-axisindicatethemorewavesexistedinthetimedomainrepresentation,closertouniformdistribution.Therefore,thepairswithstrongtrendinhighfrequenciesaresearchofinterests-regularlyencounteringpairs.Althoughthenotableweeklypatternwasshowinginaverage,periodicityappearsdifferentlybyeachpair.Figure 4-10 showscdfofthehighestfrequencyoftheencounteredpairsinorderoffrequencymagnitude.Rightkneeintheupperrightsideofthegraphindicatesthatperiodicbehaviorisstrongerfor 47

PAGE 48

Figure4-10. EmpiricalCDFofhighestfrequencyfortheencounteredpairsatUSC06springtraceaccordingtodailyencounterrate(x=Drate) someofthepairs,suggestingregularencounteractivity.Basedonthisobservation,wegroupedthepairsthatencounterregularlybytakingthepairswhosetopfrequenciesareoverthekneepoint(top20percent).WeplottheAPswheretheregularlyencounteringpairshaveaccessedandoverallpairshaveaccessedinFigure 4-11 .Itisclearthatthelocationvisitingpatternsatthetimeofencountersarenotablydifferentinmanyofthelocations.Notethatthelowestfrequencywasnotconsideredingroupingtheregularlyencounteringnodesforthereasonthatitdoesnotindicatetheregularencounter,ratherburstencounter. 4.3.4.2NormalizedregularencounterTakingonlythetopfrequencymagnitudemaynotcapturetheregularlyencounteringpairsaccurately,becausethenumberofspikesinthefrequencygraphcanbeoftwoormore.Thisisnormalforregularpatternassomeofthetrendcanconsistofseveralminorsincludingartifactsoffrequencyanalysis.Forexample,cyclesatevery8timeunitcancausetohavecyclesatevery16timeunit.Tobettercapturetheregularlyencounteringpatternwhileconsideringthepossiblenoisefactors,weobservednormalizedtopfrequenciesforeachpair.Toachievethenormalization,the 48

PAGE 49

Figure4-11. APsaccessingpreferenceatUSC06springtracefor40Drate<60 Figure4-12. Orderedlocationvisitingpreferencebyencounteredpairsaccordingtodailyencounterrate summationofseveraltopfrequencymagnitudeswascomparedtothesummationofallthefrequencymagnitudesineachpair.Weobservedthatthetop3frequenciesweretakingupmorethanonethirdofallforthemostofthepairsconsistentlyacrossthedifferentdailyencounterrate.Hence,weusethetopthreefrequenciesascriteriatoextracttheregularlyencounteringpairs.Afterapplyingthisrule,itappearsthattheratioofregularlyencounteringpairstotheoverallpairsdifferbythedailyencounterratewith 49

PAGE 50

therangebeingfrom0.15to0.4.Rareencountercanhardlyyieldregularityduetoitssmallnumberofsamples.Fortheoppositereason,frequentencounteroftenleadstotheuniformdistributionofencounter,thus,anypatternsarehardlynoticeableespeciallyindailyencounter.Werstdiscussthegeneraltrendoflocationvisitingpatternatthetimeofencounterevents.Then,wecomparetheresulttotheregularlyencounteringpairs.InFigure 4-12 ,thedataissortedaccordingtothenumberofencountereventsinthelocation.Itisclearfromthegurethatencountereventsareskewedtoafewlocations,thus,thegraphshowingexponentialcurvetowardhighlyvisitedlocations.Anotherobservationisthatfrequentlyencounteringpairsareexhibitingdifferentlocationvisitingpatternsfromlessfrequentpairsatthetimeofencounterevents.Thisimpliesthatthelocationswithrareencountereventsaredisadvantagedinmessagedelivery.Nowweturnourattentiontothetrendfortheregularlyencounteringpairs.IntheFigure 4-11 ,theregularlyencounteringpairsareshowingdifferentvisitingpreference.Thisdifferentlocationvisitingpatternsupportsthestrongvalueofusingregularencounterpattern.Deliveryattempttothenodesthatmainlyappearinthelocationofscarceencountereventsmaysufferfromndingtheforwardingnodesthathavehistoryofencounterevent.Therefore,overloadonthenetworkcanbeexpectedduetotoomanynumberofmessagecopies.Insuchasituation,1)usingtheregularlyencounteringnodesiftheencounterrateissimilar,wouldmakemoreaccurateestimationforthenumberofrelaynodestoreachthedeliveryprobabilitygoaland/or2)thesourcenodehasamorechancetodiscoverthenodesthatregularlyencounterwiththetargetnodeinsomelocations.Moreapplicationscanbedevelopedthatusesuchcharacteristics;thus,ouranalysisopensupformoreapplicationsandisoneofourcontributions. 4.4TemporalStabilityofMobileEncounterInthissection,weinvestigatetheencounterhistoryofmobilenodesanditschangesovertheperiodoftimefromWLANtraces.Morespecically,weconducta 50

PAGE 51

preliminary,yetsystematicevaluationofhistory-basedanticipation,byvaryingthesizeofthetwowindowsofencounterhistory-oneformerandonelatter.Practically,wecompareencounterpatternsintherstpartoftracetothelaterpartoftrace;andobservehowwellencounteranticipationworksasthesizeoftheencounterhistorychanges.Ourndingsindicatethatencounterpatternsofrarelymeetingnodesaresurprisinglyquiteconsistentingeneral,whereasfrequentlyencounteringnodestendtobeinconsistent.Further,moreencounterhistoryappearstohelptoobtainbetterconsistencyinencounteranticipation.Basedondailyencounterpatterns,theencounterratedifferenceisgenerallylessthan20%betweentwowindows.Encounterratedifferenceisevensmaller(lessthan10%)forregularlyencounteringpairs. 4.4.1StabilityWindowofEncounterHistoryKnowledgeaboutthepastencounterpatternbetweentwomobilenodescanbebenecialinEncounterpredictioncanbebenecialinmessagebundledeliveryespeciallyinasituationwheremobilityisnotrandom.Properuseofencounterhistorycanbehelpfulinpredictionofencounterincasethefutureencountermaybaseonthepastencounter.Forthisreason,therearemultipleliteraturesproposingroutingprotocolsusingencounterhistory(orcontacthistoryassimilarterminology)[ 4 5 13 14 ].In-depthanalysisonencounterpatterncanhelptheroutingprotocolstooperatemoreefcientlybysophisticatedlyselectingtherelaynodes.Tohelpinrouting,weinvestigatetheencounterstability,whichshowsthetemporalchangeofencounterpattern.Webelieveunderstandingofencounterstabilityanditsrelationshiptoencounterpredictionisakeyindevelopinganappropriatecontactpatternmodelingofmobilenodesandroutingschemethatusesthemobilenodesasforwardingnodes.Toanalyzetheencounterstabilityinlargescale,weinvestigatewirelessLANusageof10,000samplemobilenodesfromtheUniversityofFlorida(1Jan2008-3Mar2008)andUniversityofSouthernCalifornia(9Jan2006-11Mar2006).Thesesamplenodeswerechosenrandomlyamongaround20,000nodesatUSCand50,000nodesatUF. 51

PAGE 52

Figure4-13. Windowofencounterhistory Toobservethedifferencebetweentherstpartoftheencounterhistoryandthelatterpartoftheencounterhistory,wedividetheencounterhistoryofeachpairtoW1andW2fromtheoriginalencounterhistorywindowasinFigure 4-13 ,wherebothparametersindicatethesizeofthewindowineachofrstandsecondperiodrespectively.Wecomparethedifferenceofdailyencounterrateforencounterpair(i,j)betweentwoperiodsW1andW2,byprovidingDi(i,j)=jDrate1(i,j))]TJ /F4 11.955 Tf 10.15 0 Td[(Drate2(i,j)j,where0Drate1(i,j)1,0Drate2(i,j)1.Toseethevarioustrendsbytheencounterrate,werstcategorizetheencounterpairstovegroupsaccordingtotheirdailyencounterrateandlookintotheiraverageDi(i,j),where0in,0jnandi6=jforeachgroup. 4.4.2ConsistencyofEncounterPatternFigure 4-14 showstheaveragedifferenceofthedailyencounterrateineachofUFandUSCtraceaccordingtothedifferentdatesofthebreakpoint,td,fortherstwindow.EachgraphshowsthechangeofDiaccordingtothelengthofencounterhistory,W1.Specically,theX-axisindicatesthesizeoftherstwindowW1indays,whiletheY-axisistheaveragedifferenceofencounterratebetweenthetwowindowsW1andW2.ThesizeofthesecondwindowW2wassetto21daystocomparetheencounterrateofthreeweeksperiod.BiggersizeofthesecondwindowW2wasdiscouragedasitoverlapswiththespringbreakperiod,whichbytheobservationaffectstheresultmeaningfully.Thus,weobservedtheresultsbeforethespringbreaktimeforbothoftheUFandUSCcampustoobservethetrendwithouttheinterferenceofsignicantsocialcontext.TheguresshowthatDibecomessmalleraswehavemoreencounterhistory.Thisindicatesthatmorehistoryintherstwindowwouldyieldthebetter 52

PAGE 53

AUFcampuswiththebreakpointat28thday BUFcampuswiththebreakpointat35thday CUFcampuswiththebreakpointat42ndday DUSCcampuswiththebreakpointat28thday EUSCcampuswiththebreakpointat35thday FUSCcampuswiththebreakpointat42nddayFigure4-14. AveragedifferenceofthedailyencounterrateatUFandUSCcampusaccordingtothechangeofbreakpointfortherstwindow 53

PAGE 54

consistencyinencounterpattern.Wecaninferfromthisresultthatmoreknowledgeontheencounterhistoryingeneralleadstohigherconsistencyinanticipatingtheencounterrateinthesecondwindow.AnotherinterestingtrendisDiisgreaterforthegroupswithmoredailyencounterrate.InFigure 4-14 ,differentlinesrefertodifferentdailyencounterrates.ForbothUFandUSCtrace,itshowsthattheencounterrateofthefrequentlyencounteringpairsarelikelyinconsistentintermsofdailyencounterrate.NotealsothatDiwassmallestingeneralwhentd=28forthesamesizeofencounterhistoryW1atbothoftheUFandUSCtraces.Thisimpliesthatconsistencyoftheencounterratecanchangedependingonthebreakpointofthetwowindows.Nowwemoveontocontrollingthesizeofthesecondwindow.Figure 4-15 showstheaveragedifferenceofthedailyencounterrateaccordingtothechangeinthesizeofthesecondwindow.Inallofthegures,thesizeoftherstwindowcorrespondstothetd.Forinstance,W1=14whentd=14asweobservetheencounterhistoryfromtherstdayofthetrace.Eachgraphshowstheconsistencyofencounterratebetweentwowindowsaccordingtothechangeofthesizeinthesecondwindowforgivensizeoftherstwindow.Wecanseefromtheguresthatencounterconsistencydropsslightly,particularlyforthegroupsofpairswithhigherencounterrate.Thistrendisstrongerwithmorehistoryintherstwindowwheretd=21thanlesshistorywheretd=14.IncaseofUFtracewithtd=14,itappearsthatconsistencyofencounterpatternbecomesbetterasthesizeofthesecondwindowgrows.Itisaninterestingtrendasingeneraltheconsistencyisexpectedtodropduetoshortlengthofprevioushistorytocompare.WethinkthestartingdateofthetracemayplayaroleinthisuniquetrendasthestartingdateoftheUFtrace(Jan1)doesnotmatchtothestartingdateoftheclassdate.Itisuncertainatthispoint,whatplaysaroleingeneratingsuchadifferenttrendfromothertraces.Welearnfromthisstudyaboutthevaryingsizeofthesecondwindowthatwithshortencounterhistoryintherstwindow,theconsistencyinencounterpatterncanbe 54

PAGE 55

AUFcampuswiththebreakpointat28thday BUFcampuswiththebreakpointat35thday CUSCcampuswiththebreakpointat35thday DUSCcampuswiththebreakpointat42nddayFigure4-15. AveragedifferenceofthedailyencounterrateatUFandUSCcampusaccordingtothechangeofbreakpointforthesecondwindow unpredictable.Withlargerencounterhistoryandproperbreakpoint(td=21),encounterconsistencydecreasesasthesizeofthesecondwindowgrows. 4.4.3ApplicationofEncounterStabilityAnalysisKnowledgeonconsistencyofencounterbehaviorcanbebenecialfordevelopinganintelligentmessagebundledeliverysystem,particularlywhereencounterpatternsamongstmobilenodesarenotrandom.Properuseofencounterhistorycanbehelpfulinmeasuringconsistencyofencounterpattern.Forthisreason,therearemultiple 55

PAGE 56

Figure4-16. AveragedifferenceofthedailyencounterrateatUSCcampusaccordingtothechangeofsizefortherstwindowofencounterhistory literaturesproposingroutingprotocolsusingencounterhistory;however,in-depthanalysisontheencounterpatternsaccordingtothedifferencesinencounterhistoryhasbeeninneedtodeviseasophisticatedprotocolthatreectstherealencounterpattern.Ouranalysisprovidesabasisfordevelopingprotocolsthatutilizeencounterhistoryfordeterminingrelaynodesaswellasefcientbufferingandcachingintheforwardingmobilenodes. 4.4.4ComparingRegularEncountertoNon-regularEncounterWeshallinvestigatetheencounteredpairsregardingtheirconsistencyofencounterpattern.Weusetheapproachesweproposeinabovesectiontondtheregularlyencounteringpairsfromtherstpartoftheencounterhistory.Wealsouseanewapproachthatdividestherstwindowofencounterhistorytoseveralpiecestoseetheconsistencyinthem.Inthisnewapproach,encounterconsistencyiscomparedwithintherstwindowandweselecttheregularlyencounteringpairsfromthepairsshowingmoreconsistencyintherstwindow. 56

PAGE 57

Wecomparetheregularlyencounteredpairswiththenon-regularlyencounteringpairs.Bydenition,regularlyencounteringpairsaremuchmorelikelytomaintainconsistentencountereachotherthanirregularpairsonaregularbasis.Weusedthenewapproachofselectingtheregularlyencounteringpairsbytheevaluationofconsistencyforthetwodividedwindowsintherstwindow.Encounteredpairswithlessthan0.5ofdifferenceinthesedividedwindowswerechosenforregularlyencounteringpairs.Therestoftheencounterpairswereconsideredirregularlyencounteringpairs.InFigure 4-16 ,weobservetheoverallconsistencybetweentwowindowsandcomparebetweenregularlyencounteringpairsandnon-regularlyencounteringpairs.Welookatthepairswithatleastoneencountereventforbothoftheregularandnon-regularcases.Itshowsthattheregularlyencounteringpairshavesignicantlybetterconsistencyovertheirregularlyencounteringpairs.Regularlyencounteringpairsshowtheaveragedifferenceofencounterratebetweenthetwowindowsisaround0.1,whichissurprisinglyconsistent.Whereas,thenon-regularlyencounteringpairsshowrelativelylargeinconsistencywithanaverageof0.3atthedifferentofencounterratebetweentwowindows.Thisimpliesthattheregularlyencounteringpairsintherstperiodofthetracetendtomaintainsimilarencounterrateinthesecondperiod.Notethatthepairswithverylowencounterratecanskewthisoveralltrend.Furtheranalysisinthefuturetothisresultisobservingtheencounterconsistencybythegroupsofdifferentencounterrate. 4.5Long-termBluetoothTraceAnalysisInthissection,weanalyzethelong-termBluetoothcontacttraceofamobileuser.AnanonymousmobileusercarriedourBluetoothcontact-measuringdeviceover6monthsperiodandweanalyzetheindividualencounterdata.NotethatthisdataincludestheBluetoothcontactsoccurredduringthemobileuser'stravel.Therefore,whilethisdatadoesnotreecttheaccuratedescriptionofuser'scontactbehaviorinonelocation,itcapturesoverallcontactpatternregardlessoflocation. 57

PAGE 58

AEncounterdays BEncounterfrequency(count) CEncounterdurationFigure4-17. Timeseriesdataforthemobileuser'sBluetoothencounter Figure 4-17 showsthetimeseriesdataforthemobileuser'sBluetoothencounter.Allthemetrics(encounterdays,frequencyandduration)showsimilartrendinpeaks.Figure 4-18 showstheCDFforthesizeofencounterdays,frequencyandduration.Itshowsaroundhalfofencountereventsbelongtoverylowmagnitude.About5%oftheencounterednodesthroughoutallthemetricsshowhighdegreeofencounters.Inthegure 4-19 ,spectralanalysisshowsthatperiodicalpatternisnotasstrongasshowninWLANtraceanalysisandshort-termBluetoothanalysis.Particularly,encounterdurationshowsveryweakperiodicalpattern.Possiblereasonscouldbeabreakperiodsuchasvacation,summerbreakandinternationaltravel.Welearnfrom 58

PAGE 59

AEncounterdays BEncounterfrequency(count) CEncounterdurationFigure4-18. EmpiricalCDFforeachencountermetricwithencounterednodesinthelongtermBluetoothtrace thisanalysisthatperiodicitymaychangeovertimeandmayyielddifferentresultthanshort-termanalysis. 4.6ConclusionsandFutureWorkWeinvestigateaWLANtraceandBluetoothtracetodiscoverperiodicityofmobileencounterpattern.Ouranalysisshowsthatspectralanalysiscanapplytoanalyzetheperiodicityandtheresultshowsthatweeklyencounterpatternisstrongovermostofthetraces.Wealsoshowthestabilityofencounterpatternoverasemesterperiodandshowthedifferenceofencounterpatternovertime.Encounterratedifferencebetweentherstperiodandsecondperiodincreasesasthesizeofthewindowincreases.Inourlong-termBluetoothcontactanalysis,weanalyzedover180daysoftraceforasinglemobileuser.ItappearsthatperiodicityisweakerthanWLANtraceandshort-termBluetoothtrace.Thisshowsthatanobservationforashortperiodcandifferfroma 59

PAGE 60

AEncounterdays BEncounterfrequency(count) CEncounterdurationFigure4-19. Spectralanalysisoflong-termBluetoothencounterforeachmetric long-termperiod.Futureworkincludestheapplicationofusingtheseperiodicpropertiesinmobilityprediction[ 41 ][ 42 ][ 43 ].Overlaynetworksamongsttheperiodicallycontactingpairsisonedirectiontostudyasanapplication.Anotherapplicationisanticipatingtheuser'scontactwithcertainnodesbasedonregularityandstabilityofencounterpatternforthepair. 60

PAGE 61

CHAPTER5CAPTURINGANDEMBEDDINGPROFILESOFMOBILENODESInthepreviouschapter,wehaveanalyzedtheencounterpatternsofmobilenodes.Resultofsuchanalysisinformsofgroupbehavioranalysisofencounterpatterns.Wediscusstheideaofcapturingperiodicpropertiesfrommobileusers'contactpattern.Wethendiscusstheencountercommunityprolesofmobilenodesandtheircommunitypattern.Ourgoalofencounterprolecreationformobilenodesistoemulatehumanbehaviorintermsofcommunitybasedencounterpattern. 5.1IntroductionMobilityproleofthemobilenodescanbeanusefulinmessagedeliveryfordecidingtargetsofthemessagewhentherearemultiplerecipientsinprolebasedprotocolssuchasprole-cast[ 3 ],bubble-rap[ 5 ]andsocialawarenetworking[ 4 ].Thoseprotocolsusetheprolesofsimilarityinlocationvisitingpreferenceandcommunitypropertyconstructedbasedonthesimilaritiesinlocationvisitingpreference.Ourworkfocusesoncapturingperiodichumanencounterpatternandtheircommunitydistribution.Location-basedprolinghasbeenstudiedbutencounterbasedprolinghasrarelybeenstudied,particularlyfocusingonperiodicproperties.Encounterpatternisimportantinformationforencounterpatternmodelingandmessagepropagationsystem,wherelocationinformationisnon-existent.Forinstance,BluetoothencounterhistorydoesnothavelocationinformationwithoutthehelpofGPSorWLANaccessrecordwithAPs.BothofGPSandWLANquicklyconsumethebatteryofmobiledevices;thus,havingtheBluetoothorshort-rangecommunicationdeviceforthisspeciccommunicationwillhaveanadvantageinbatteryconsumption.Besides,therearefrequentoccasionsormanyplaceswherelocationinformationisunavailable(i.e.insideabuildingwhereGPSdoesnotworkandAPsarenotinstalled).Moreover,messagedeliveryoractualencounterbetweenmobilenodesshouldoccurwhentheyareinproximityofwirelesscommunication.Encountereventmaynotoccureveniftwonodesareaccessingthe 61

PAGE 62

samelocationunlesstheyareincloseproximity.Therefore,studyontheencounterpatternexcludinglocationinformationhasseveraladvantagesoverlocation-basedanalysis. 5.2RelatedWorkProlingofmobileuserswasstudiedinmanyliteraturesbasedonusermobility[ 44 ][ 45 ];however,littleefforthasbeenmadeforprolingbasedperiodicityofcontacts.Weusespectralanalysis[ 46 ]torevealtheperiodicityandgroupmobileusersbasedonperiodicalcontactpattern.Withanunderstandingofhumancontactpatternandtheirperiodicproperties,wecapturemostimportantcharacteristicsandshowthatitcanreplicatethereal-worldcontactpatternintermsofperiodicity.Communitystructureinmobileusershasbeencriticalsubjectofstudy[ 47 ][ 48 ][ 49 ];yetlittleisknown.Revealingandcapturingcommunitystructureareimportantworksascommunityinformationcanbeusedininterest-basedmessageforwarding[ 3 ][ 50 ][ 51 ]orndingcriticalnodesfordatapropagation[ 5 ][ 52 ].Researchersshowedthatmobileusers'mobilitycouldbegroupedintocommunities.Thedistributionofgroupsizesbyhierarchicalclusteringappearedfollowingapower-lawdistributionin[ 1 ].Wedeneencountervectorandanalyzeencountertracetondcommunitiesinencounterpattern.Wealsoapplythehierarchicalclusteringtothemobileusers'contactpatternanddiscovertheexistenceofpower-lawdistribution. 5.3ModelingEncounterPeriodicitywithProleWiththepropertiesofencounterpatterndiscussedabove,weproposetheencountermodelformobilenodes.TVCmodelprovidestheinputofperiodicityamongmobilenodes.However,TVCmodeldoesnotgeneratetheperiodicityofencounterpatternthatemulatestheobservedencounterpattern.Instead,ittakesthecommunitystructuresfoundfromthemobilitytraces.Withabasisontheunderstandingofindividualencounterperiodicities,synthetictracethatemulatesthereal-worldhumanencounterpatterncanbeconstructed(i.e.reectingthenotableweeklyencounterpattern).There 62

PAGE 63

Figure5-1. CDFoftop7maximumfrequenciesfromthe2000samplepairswithdailyencounterratebetween0.2and0.4for128daysofUF07trace aretwotypesofimplementationofperiodicitygenerator:1)Periodicityofindividualencounterpatternthatreectsoverallstatisticsofindividualencounter2)Periodicityofencounteredpairs.Thereisatrade-offineachtypeoftracegenerator.Thersttypegeneratesatracedatawithoverallindividualencounterpattern.Thisreectstheperiodicityofencounter;yet,itislimitedtoindividualencounterpattern.Thesecondtypegeneratestheperiodicityforeachencounteredpair,whichemulatesthecloselyreal-worldencounterevents.However,implementingperiodicityforeachencounterpairincurscumbersomeimplementation.Figure 5-1 showstheCDFoftop7maximumfrequencyfromthe2000sampleencounterpairsfor128days.Inthegure,thekneealsoappearsfor9%ofallfrequencieswhichindicatesroughly7frequenciesoutof63aredominatingtrend.Thisimpliesthattop7frequenciescanbeusedasasignaturefrequencyforperiodicityvector.Thissavesthesizeofvectorfrom63to7.Astop5locationswerechoseninminingbasedonlocationvisitingpreferencework[ 1 ],periodicityvectorcanbebuilt 63

PAGE 64

accordinglytorepresenttheperiodicityofpairs.Weleavetheactualconstructionofvectorstofuturework. 5.4DiscussionofRoutingBasedonEncounterProlesInthissection,wediscussthefutureapplicationofperiodicproperties.Actualimplementationusingperiodicpropertiesisoutofscope.Wedescribethedetailsofhowimplementationcanbedoneonly.Encounterconsistencymaydifferbytheencounteredpairs.Ourresultsshowthattheencounteredpairswithconsistentencounterrateintherstwindowarelikelytokeepthesimilarencounterrateinthesubsequentwindow.Assumingtheencounterhistoryintherstwindowisthepasteventsandthebreakpointisthecurrenttimewhereroutingdecisionismade,thesecondwindowoftheencounterhistoryistranslatedintothefutureencounter.Inthisscenario,consistentencounterpatterninencounterhistoryinfersthatthefutureencounterpatternisalsolikelyconsistenttotheencounterhistory.Withthisidea,routingdecisionofselectingforwardingnodecanbemadebylookingattheencounterhistoryanditsconsistencyforndingconsistentencounterpattern.Inadditiontotheconsistencyofencounterpattern,encounterprobabilityiscalculatedbasedontheencounterhistory.Prophetusestheencountercountfornextencounterprobabilitybasedonencounterhistoryforndingrelaynodesofamessagebundle.However,howtocalculatetheprobabilityofencounterisunclear.Forinstance,encountercountcanhaveburstpatternthatyieldsbiasedresultsinprobability.Weovercomethisproblembyusingbinarytimeunit.Wedividethetimeseriesdataaccordingtothedenedtimeunitsdependingonthescenarios.Forinstance,hourlyencounterisatimeseriesdatawithtimeunitofanhourwithavalueof1inthepresenceofencountereventwithinanhour.Differenttimeunitscanbeappropriatedependingonthescenarios.Selectingtheregularlyencounteringnodesisacriticaldecisioninthisroutingscheme.Theirregularlyencounteringpairsmaynotbeappropriateforprobabilitycalculationbecausetheirencounterhistoryisinconsistent;thus,theirfutureencounterisunpredictable.Furthermore,therandomly 64

PAGE 65

Figure5-2. Overlaynetworksusingperiodicityasalinkweight encounteringpairsarealsonotgoodcandidatesformessageforwardingastheirbehaviorcannotbeestimated.Therefore,interestofcalculatingprobabilitybasedontheencounterhistoryliesintheregularlyencounteringnodes.Howmuchofnon-regularlyencounteringpairsshouldbefactoredindecisiondiffersbyscenarios.Figure 5-2 showsanexamplescenarioofanetworkwithlinksweightdecidedbythestrengthofregularity.Applicationofthisnetworkcanbeperiodicalnews,informationorelectroniccoupontothesubscribersviatheregularitybasedoverlaynetworks.UsingMobiTrade[ 22 ],thedatacanbesharedwithincentivesforsharingmoredataviathisoverlaynetworks.ThisoverlaynetworkscanalsobeusedinestablishingthelevelofencounterbasedtrustsuchasPROTECT[ 29 ][ 30 ][ 31 ]byaugmentingtheencounterbasedtrustmetricwiththeadditionofperiodicitymetric. 5.5CommunityProlesofMobileUsersInthissection,wediscusstheclusteredbehaviorofmobilenodesfromreal-worldtraces.Prolingmobileusersbasedonlocation-visitingpreferencewasintroducedin[ 1 ].Ouranalysisisperformedforcontactbehaviorofmobileusers.Werstdescribethepreviousndingfromthelocationbasedprolework,andexplaintheapproachto 65

PAGE 66

Figure5-3. Exampleofembeddingprolesinvariousinterfaces discovercontactbasedprole.Then,wecomparetheresultstorevealthepower-lawdistributioningroup(community)sizesfrombothtypesofproles.Theseprolescanbeembeddedonautonomousnodes.ExamplesareshowninFig. 5-3 .Inourimplementationforautonomousmobilenodes,power-lawdistributionthatwediscoverfrombothtypesofprolesisusedtoassigncommunityIDs.Notethatonlytheanalyzedresult(i.e.power-lawdistribution)isusedinthisworkinsteadofembeddingeithertypeofprolesdirectlythoughtheycanbeembeddedinourtestbed.Complicatedcommunitystructuressuchasoverlappingcommunitymodelisstudiedin[ 53 ][ 54 ],however,communitymodelingisnotoutofscopeinthisdissertation. 5.5.1LocationVisitingPreference[ 1 ]showsthatmobileuserscanbeclusteredbytheirlocation-visitingpattern.Behavioralsignatureoflocation-visitingpreferenceforamobileusercanbebuiltbyanassociationmatrixasinFig. 5-4 (courtesyofHsu[ 1 ]).Foreachdayi,eachrowsavesthedurationoftimeateachlocationj.Percentageoftimespentineachlocationjisputineachcolumnofthematrix.Eachmatrixiscondensedtoavectorrepresentation 66

PAGE 67

Figure5-4. Locationbasedassociationmatrixforeachmobileuser(courtesybyHsu[ 1 ]) ofmostsignicantbehaviorsbyperformingSingularVectorDecomposition(SVD).Similarityscoreisproducedfortheencounterednodes.Eachnodecanmakeadecisionbasedonthissimilarityscorewhethertheencounterednodesshowsimilarbehaviortoatargetgroup.Prole-cast[ 3 ]usesthislocationsignaturefordeliveringmessagestotargetgroupofnodes.Ifimplementedonrobots,theycanbuildthesameassociationmatrixbycollectingitsownlocation-visitinginformation.Prole-castwasimplementedonNokiaPDA;therefore,itisdirectlyusableinourtestbedwithoutfurthermodication. 5.5.2EncounterVectorToanalyzeencounterpatternofmobileusers,wedeneanencountervector.Anencountervector,Vi,foruseri,representsanencounterbehaviorofthemobileuseriwithallothernodes.Specically,eachelementinaencountervectorindicatesaratioofencounterindayswithamobileuser.Notethatdifferentmetricssuchasencounterrateordurationcanbeusedbutweuseencounterdays[ 46 ]inthiswork.Formobileuser,i,totaldaysofencounter,Tiiscalculatedasinthefollowing( 5 ),whereMisanumberoftotalnodes,NisanumberoftotalobserveddaysintraceandEd(i,j)indicatesabinaryprocessofencounterbetweenanodeiandjinadayd. 67

PAGE 68

Ti=MXj=0N)]TJ /F6 7.97 Tf 6.58 0 Td[(1Xd=0Ed(i,j)(5)Usingthesumoftotalencounter,Encountervector,Vi,isformedinthefollowingwaytoreecttheratioofencounterwitheachnode: Vi=fN)]TJ /F6 7.97 Tf 6.59 0 Td[(1Xd=0Ed(i,0)=Ti,......,N)]TJ /F6 7.97 Tf 6.58 0 Td[(1Xd=0Ed(i,M)=Tig(5)Nowweanalyzethereal-worldtraceusingthisvectorrepresentationofencounter.USC06andUF07traces[ 35 ]areusedforanalysis.In[ 1 ],themobileusersshowpower-lawdistributioninsizeofgroupswithhierarchicalclusteringaccordingtotheirlocationvisitingvectorasshownatlog-logscalerepresentationinFig. 5-5 .Wealsoapplythesamehierarchicalclusteringtomobileusers'contactpatternbasedonproposedencountervector.Theresultshowsthesizeofrankedgroupsforcontactpatternalsofollowspower-lawdistributionasshowninFig. 5-5 .Speccally,slopeof-1.48reectsthedistributionofcontactbasedgroupsizes.Inthefollowingsection,weexplainhowtoimplementself-decisionmakingmobilenodeswhilefollowingthiscommunitydistributionpattern.Withthisunderstandingofcommunitydistribution,weintroducetheprocessofembeddingcommunityinformationaccordingtopower-lawdistribution. 5.6ConclusionsandFutureWorkWediscusstheprolingprocessofmobileusersbasedonperiodicityofcontact.Wedescribetheprocessofgeneratingacontactperiodicitythatmatchestoreal-worldstatistics.Inaddition,wedeneencountervectorandshowthatreal-worldcontactpatterningroupsfollowspower-lawdistributionintermsofrankgroupsize,whichisthepatternobservedfromgroupingbylocationvisitingpreference.Futureworkincludesexploringmorepropertiesofcommunityexistinsocialcontactpatternsuchascommunitiesbasedoncontactdurationorfrequency.Selectedgroups(i.e.highly 68

PAGE 69

AClusteringbyencounterpattern BClusteringbylocation-visitingpattern(CourtesyofHsu[ 1 ])Figure5-5. Rankedsizeofgroupsaftergroupingbyhierarchicalclusteringforcampustraces.Bothgraphstapower-lawpatternforthesizeofgroups. 69

PAGE 70

activeusers)canbegroupedseparately.Studyingcommunitygroupsbasedonlocationvisitingpreferenceandencounterpatterntoseethesimilarityanddifferencewillprovidevaluableknowledgeinunderstandingcommunitystructureofmobileusers'society. 70

PAGE 71

CHAPTER6DESIGNOFAMOBILESOCIALNETWORKINGTESTBED 6.1IntroductionRecentadvancesinmobilenetworksbroughtmarriageofsocialandmobilenetworking,whichisgoingtoconstitutethefuturefrontierwiththeproliferationofsmartphones.Growthofmobilesocialenvironmentsrequiresappropriatenetworkingprotocolsandservicesusinghumansocialpattern.Toevaluatesuchprotocolsandservices,itisessentialtohavealarge-scaleandrealistictestingenvironment,includingsimulationsandtestbeds,whichincorporatevariousaspectsofmobileuserbehavior.Towardtacklingthesechallenges,weintroduceanovelmobilenetworkingtestbed,whereautonomousmobilenodescollectivelymimichumancommunitycontactpattern.Specically,wediscoverpower-lawdistributioningroupsizesfromcontactpatternanduseitforcontactproles.Furthermore,wedevelopseveralcontactrule-baseddecisioncriteriaforautonomousmobilenodes.Oursimulationfor1000mobilenodesshowsthatmobilenodescanmatchhumancommunitycontactpattern.Inaddition,weimplementaprototypetestbedbyembeddingthesamedecisioncriteriaonmobilerobots.Tothebestofourknowledge,ourworkisthersttoembedcommunityinformationfromreal-worlddataanalysisonautonomousmobilenodes,ineithersimulationortestbed.Ourndingsprovideopportunitiesforbuildingrealistictestbedswithautonomousmobilenodesforvarioustypesofmobilesocialnetworksandservices.Withproliferationofmobiledevicesandsocialnetworkservices,itisimportanttohaveanenvironmenttotestnewnetworkingprotocolsandservicesusingmobileusers'mobility.RealisticmobilenetworkingtestbedisimportantforsuchamobilenetworkingenvironmentasbreakdownoflinksduetomobilitycreateschallengingissuesbothforMobileAdhocNetworksandwirelesssensornetworks[ 55 ].Towardsrealistictestbedformobilenetworks,weaddressthefollowingthreekeychallenges:1)Testingenvironmentshouldsupportrealisticmovementsofmobilenodesbasedonreal-worldmobilitydata.Particularly, 71

PAGE 72

weintendtocapturemobileusers'community-basedcontactpattern.2)Autonomousmobilenodesmaketheirowndecisionswhileemulatinghumanmobility.Existingtestbedsarelimitedtorandommobilityandpredenedpaths.Othersimulationandmobilitymodelscancapturerealistichumanmobility;yet,theyrequirehavingglobalknowledge,includinglocationofothernodes.Thisknowledgemaynotbeavailabletoeverynodeinarealmobilenetworkingenvironment.Hence,mobilenodesneedtobeabletodecidetheirownmovementsbasedonavailableinformationinrealtime.3)Bothofsimulationandprototypeimplementationonrealmobilenodesshouldbeavailable.Simulationprovideseasyadjustmentsforexperimentparametersandprolesofmobilenodesalongwithvisualization.Ontheotherhand,prototypeimplementationshowsthefeasibilityofdeployment.Notethatweinterchangeablyusertheterm,contact,encounterandmeeting,allinthesamecontextofphysicalclosenessthatissufcientfordirectwirelesscommunication(i.e.Bluetoothdiscovery)asin[ 46 ][ 16 ][ 56 ][ 57 ].Existingmobilenetworkingtestbedsuserobots,emulationormobilitymodelingtodeploymobilenodes'mobility[ 24 ][ 23 ][ 26 ][ 27 ].Robotsaredeployedinrandommobilityorpredenedpaths.Emulationofrobotmovementsfollowsthesamecriteriaandrequirestohavelocationinformation.Modelingneedsglobalknowledgeofnodelocationtogeneratemobilityandcontactstatistics.Todeployontheautonomousmobilenodesthatcollectivelymimicmobileencountersinrealworld,itisessentialtohaveamobilityprolesandself-decisionmakingcriteriaforeachnode.Thisapproachenablestoembedthealgorithmbothonsimulatedmobilenodesforlarge-scaleexperimentandonphysicalmobilenodes(i.e.robots).Inordertoachievethisgoal,wedevelopcontactrule-baseddecisioncriteriathatareembeddedoneachmobilenodetomakeamobilitydecisionforthemselvesuponcontactingothermobilenodes.Furthermore,weinvestigatereal-worldnetworktracetorevealthecommunitydistributionbasedoncontactpattern.Wediscoverpower-lawdistributionforthesizeofgroupsfromcommunitycontactpattern,and 72

PAGE 73

Figure6-1. Bridgingthegapbetweentheemulatedcontrolledenvironment(networkofrobots)andthenon-controlledchaoticrealmobileworld(participatornetworks).Bothenvironmentsareconnectedviamobiledevicesthatenablecommunicationbetweentheseenvironments.Robotsmovewithmimickedhumanmobility,thus,emulatingrealchaoticworldwithoutneedingtorecruithumanparticipants assigncommunityinformationtothemobilenodesaccordingly.Inaddition,weembedaschedulertoemulateperiodicalcontactwithcommunitymembers.Finally,weevaluatewithmultiplemetricsincludingcontactdays,frequencyanddurationcomparedtoreal-worldcontactpatternthatisprocessedfromcampusnetworktrace.Anotherpartofourworkisaproposalofnovelideaforarealisticmobilenetworkingtestbedthatcanblendanetworkofrobotsandparticipatorynetworks.Theconceptofbridgingtwodifferentenvironmentofcontrolledanduncontrolledwasdiscussedinourearlierwork[ 58 ]andshowninFig. 6-1 :1)Anetworkofrobotsisaswarmofrobotsthatroamaroundwithwirelesscommunicationdevicesforopportunisticcommunication.2)Participatorynetworkisanetworkspacewhereparticipantsintheexperimentcarrythemobiledevicesashumancohorts. 73

PAGE 74

Bothoftheseenvironmentshavelimitations:Inanetworkofrobots,robotmobilityisoftenlimitedtothelabenvironmentspace.Moreover,currenttestbedssupportrandommobilityandpredenedmobilitypaths,whichareunrealistictoemulatehumanmobility.Participatorynetworkisanidealformoftestingenvironmentifsufcientsizeoftargetpopulationiswillingtocooperateintheexperiment.However,itbringsissuesofrecruitingappropriateparticipantsandcontrolofhumancohorts,particularlyifthesamplesizeisbig.Tobridgethegapbetweenthesetwotypesofthetestbedsandtakethestrengthsofboth,weproposetoembedapersonalityproleontherobots.Specically,itcancontaintherule-baseddecisioncriteriaandcommunityproleinformationontherobotmobilenodes.WeprovideimplementationdetailsofthisbridgingeffortusingiRobotCreateandNokiaPDA,whichisaprototypeofmobilesocialnetworkingtestbed.Ourmaincontributionsinthisworkare1)designingacontactrule-baseddecisioncriteriaforautonomousmobilenodesinbothofsimulationandrobots;2)applyingareal-worldcommunitydataonthemobilenodesthatemulatecollectivegroupbehaviors;3)proposinganovelarchitectureofrealisticmobilenetworkingtestbedthatbridgesthegapbetweencontrolledtestbed(networksofrobotswithxedmobility)anduncontrolledtestbed(participatorytesting);and4)implementingaprototypeforthenetworkofautonomousrobots. 6.2RelatedWorkTherearethreepartsofrelatedworks:1)analysisofhumanbehavior,2)participatorynetworks,and3)mobilenetworkingtestbeds. 6.2.1AnalysisofHumanMobilityRelatedsubjectofhumanbehaviorresearchinrelationtonetworkingissocialbehaviorofhumanmobility.SeveralnetworkingprotocolsusingcommunitiesandsocialbehaviorhavebeenproposedincludingProle-cast[ 3 ],SimBet[ 4 ],BubbleRap[ 5 ],PRO[ 50 ],SIMPS[ 59 ]andSANE[ 60 ].Theresultsintheseworkswereevaluatedvia 74

PAGE 75

randommobilitysimulationandreal-worldtrace.Experimentsonsimulationortestbedusingautonomousmobilenodescanprovideextendedtestingenvironmentforsuchprotocols.Tocreatesuchanenvironment,itiscriticaltounderstandhumanmobilityandtheirsocialinteractionpattern.Usingreal-worldnetworkingtraces,manystudiesfocusedonndinginter-contacttimeandlocationvisitingpatternofmobileuserstouseinmodelinghumanmobilityanditssocialinteractionpattern[ 28 ][ 9 ][ 16 ][ 12 ][ 15 ][ 61 ].Researchndingsfromthesestudiesarefoundationsofcreatingatestbedthattakeshuman-likeinteractionbetweenmobilenodes.Yet,itisunclearhowtoadaptthemobilitymodelsontheautonomousmobilenodes,whichdonothaveknowledgeoflocationofothernodes.Ourprole-basedapproachandself-decisionapproachisuniqueinthatitcanbeimplementedinadistributedfashionwithoutglobalknowledge.Thisenablesimplementationofthealgorithmonmobilerobotnodeswithoutsignicantmodication.Theotherpatternweapplyinourworkisperiodicalpattern.PeriodicalpatternwasobservedataggregatelocationaccesspatternforWLANaccesspointsandcapturedformodelinginvariousworks[ 47 ][ 8 ].Itwasstudiedthatindividualandpair-wisecontactpatternshowsstrongperiodicaltrendviaWLANandBluetoothtraceanalysis[ 46 ].Followingthistrend,weuseaschedulerthatassignson-timeofmobilenodesbasedonthisperiodicalcontactpattern. 6.2.2MobileNetworkingTestbedsOurimplementationpartofnetworkofrobotslieswithintheareaofmobilenetworkingtestbed.Prolebasedapproachcanbeimplementedandaugmentedtotheexistingmobiletestbedbyembeddingthemobilityproles.iRobotisprogrammablerobotforresearchpurposethatisoriginatedfromRoombavacuumrobot.Ithasbeenusedinseveralothertestbedsandwealsousethisrobotintheimplementationofthetestbedasamobilenode.Severalmobilenetworkingtestbedsuserobotsasmobilenodes.MiNT[ 23 ]isaminiaturizednetworktest-bedsolelyusingiRobotsinacontrolledspace.AservercomputercontrolsthemovementsandcommunicationamongstiRobots 75

PAGE 76

thatareequippedwithWLAN.Althoughthistest-bedcanexpandwithmultiplenumbersofiRobotsandbeeffectiveinexperimentforsmall-scalemobileadhocnetworks,itstillsuffersfromscalabilityanddiversityofnodes.RoombaMADNeT[ 24 ]showedcapabilityofusingiRobotforDTN.TheymountedawirelessrouterthatrunsonLinuxbyconnectingviamodiedserialcable.Thisprocessmaytakeadvantageofcostumedlightweightprogrammingboardtoutilizewirelesscommunicationfeature,specicallyforthetestingpurpose.However,thisprocesscanbetediousandcumbersometomanyofresearcherswhoarenotskilledinthisarea.Ourmethodissimpleandusestheexistingdevice.Connectionrequirestheminimumstepandeffort:eitherBluetoothpairingorconnectingaserialcabledirectlywithadistantorattachedcomputer.MeshTestWirelessTestbed[ 25 ]isanotherDTNtest-bed.TheyuseenclosurestocontainthelaptopcomputersandmeasurethesignalattenuationforimplementedDTNprotocol.Thedesigniscentralizedanduserscanviewthewirelessnodesmovingaroundfromtheservercomputer.Mobilityislimitedtocontrolledenvironment,astheparticipantsaretofollowthegivenpathsandrequiredtobeintheexperimentrange.Inourpresentation,nodescanhavecompletecontroloftheirmovementsincludingmessagepropagationdecisions.Moreover,measurementsaredecentralizedinourexperiment,aseachmeasurementrecordiskeptinsidethemobilenodes.SCORPION[ 27 ]incorporatesvariousmobiledevices.Theyuse20iRobotsalongwithatoyairplaneandseveralvehiclestotestvarioustypesofmobility.TheycarrythespecicallydesignedDTNdevicestestthecommunicationprotocolsviaWLANandBluetooth.Randommobilityisappliedforrobotsandtoys.Othercarriersmoveinadesigneddirection.Thus,itcantesttheperformanceofDTNprotocolswithcontrolledmobility.Ourdesignphilosophycanalsobeincorporatedtothistestbeds.SCORPIONdiffersfromourtestbedsinthatweusecommondevicesthatarecarriedbynormalhumanforcommunicationandcontrollingrobots.Theotheruniquefeatureofourtestbedsisthatweuseprolestoemulaterealhumanmobilitywithfreedomsofprolemodication.Inaddition,our 76

PAGE 77

participatorytestingenvironmentallowsanonymoushumansubjectstoparticipateinthetesting.Ourcommunitycontactrule-baseddecisioncriteriaformobilenodestomakemobilitydecisionwasinuencedbyPlausiblemobility[ 62 ],whichinfersmobilitytracefromencountertrace.Pursuit-evasiongameinrobotics[ 63 ]hastwogroupsofmobilenodes-pursuersandevaders.Ourrule-baseddecisioncriteriaalsohavetwogroupofmobilenodes-friendsandstrangers. 6.2.3ParticipatoryNetworksWedeneparticipatoryNetworksasanetworkwherevoluntaryhumanparticipantswhousemobiledevicescommunicateviamobiledevices.Originalconceptisinspiredbytheparticipatorysensingproject[ 64 ].Participatorysensinghasbeenwidelydeployed,includingCenceMe[ 65 ],Micro-Blog[ 66 ]andPEIR[ 67 ].WeuseasimilarideainourBluetoothdiscoveryprogramasitcollectstheinformationforencountereddevicesasasensor.Similarexperimentswereperformedin[ 38 ][ 68 ].Ifconductinganexperimentwithparticipants,itisanidealparticipatorytestingenvironmentfortheprotocolsandservicesthatusehumanmobilityformessagepropagation.ParticipatorytestingwaspartiallyinvestigatedviaourownBluetoothtracecollectionprogramaswellasCrowdLab[ 69 ].InCrowdLab,virtualmachinesrunguestprocessesdevelopedbyresearchersonthevolunteer'sphone.Thisenablesanideaofparticipatorytestingtosomeextent.OtherattemptalongthesamelineisPhoneLab[ 70 ]whereauthorsattempttobuildanenvironmentwithathousandphonestoprovideusablespacefortestingkernellevelcommunicationprotocols.Themainstrengthsofprolebasedmobiletestbedascomparedtoothermobiletestbedsareautonomyofparticipatingsubjectsandreplicationofpersonality.Criticalfeatureofourtestbedisthathumanmobilenodescanbeemulatedbyembeddinghumanmobilityprolesontoanetworkofrobots;hence,weprovideacontrollableenvironmentinourtestbed.Thisisparticularlyusefulwhenresearcherswanttoexperimentforcertaintargethumanproles.Further,prolesandmobilitydecision 77

PAGE 78

criteriaisinterchangeablebetweensimulationandimplementation.Hence,large-scaleexperimentthatisdifculttoperformwithmobiledevicescanbesimulated.Additionaladvantageofourtestbedisthatitcanoperateontopoftheothertestbeds;thus,researcherscantakeadvantageofexistingtestbedswhileenhancingitsperformancewiththefunctionalityofourtestbeds. 6.3TestbedArchitectureWeaimtobuildamobilenetworkingtestbedthatcanintegratecontrollableexperimentalenvironmentwithuncontrollablecrowdsourcingenvironment.Inthissection,wediscussthedesignofprototypetestbedusingrobots.Werstdescribeautonomousmobilenodesthatareusedforbothinsimulationandprototypeimplementation.Then,wediscusstheimplementationdetails. 6.3.1AutonomousMobileNodesAutonomousmobilenodesaremobilenodesthatmakemobilitydecisionsindependentlyuponcontactevents.Themobilenodesdonothaveglobalknowledgeofothernodes.Theyscannearbyareaviawirelesssignal.Upondiscoveringothernearbynodes,themobilenodeeithermakesanothermobilitydecisionorestablishescommunicationwithencounterednodes.Tocollectivelymimichumancontactpatterns,itiscriticaltounderstandgroupcontactbehaviorofhuman.Weextractcontactpatternsfromreal-worldnetworktraceandanalyzethegroup(community)behaviorsbyclusteringmobilenodesbasedoncontactpattern.Communitydistributionpatternisappliedtomobilenodesastheirproles,namely,communityidentityinformation.Mobilenodes,uponencounteringothernodes,lookupthecommunityinformationandmakeadecisionbasedonpre-conguredrules.Thisresembleshumanmobilityanditsdecisionmakingofmobility.Wedescribethisinvestigationofreal-worldcommunitycontactpatternandrule-baseddecisioncriteriainthelatersectionindetail.Notethatthisrule-baseddecisioncriteriaandcommunityinformationareembeddedforbothoftherobotsandmobilenodesinsimulation. 78

PAGE 79

Figure6-2. PictureofiRobot,itscontrollingNokiaN810PDAandhumancarryingNokiaN810PDA 6.3.2NetworkofAutonomousRobotsInthenetworkofautonomousrobots,robotsfreelymovearoundwhileemulatingthemovementofhumans.Itsmobilityisdecidedbyembeddedrule-baseddecisioncriteriaandcommunityinformation(ID)ofcontactednodes.ThisrulesandcommunityIDareprogrammedintothemobiledevices,whichwecallasapersonalityinterfacethatmakesthedecisionandcontrolsthemobilityofrobots.WechooseiRobotCreateforimplementation.ThecontrollingmobiledeviceinthisimplementationisNokiaN810PDAbutanyotherdeviceswithBluetoothcanbeused(i.e.Android,WindowsMobile).ThepersonalityinterfaceinthemobiledevicecansendcommandstoiRobotviaBluetooth.WemountthecontrollingdevicesontopoftherobotsbysimplyputtingontheiRobot.Inthissetting,wetakeadvantageofmassivelyproducedandpopularmobiledevicesthatarecosteffectivefortheircapabilitiescomparedtospecicallydesignedhardwareforresearchpurposes.Moreover,becausebothoftherobotsand 79

PAGE 80

humanscarrythesametypeofmobiledevices(i.e.NokiaPDA),itisfeasibletousethesamecommunicationprotocolsbetweenthem.OurprototypeimplementationisshowninFig. 6-2 .Demonstrationvideoisavailableonlineat[ 35 ]toshowthemobilityaccordingtocontactrule-baseddecision.Notethatalthoughweimplementaprototypeofanetworkofautonomousrobots,ourfocusisonanovelideaofrealisticmobilenetworkingtestbed.Thisincludesunderstandingandgroupingofhumancontactpatternandaheuristicrule-baseddecisionalgorithmthatcollectivelyemulatescommunitybehaviorwithoutlocationinformationofothernodes.WediscussmoreimplementationdetailsonrobotsinSection7.Theonlydifferenceofmobiledeviceatbetweenrobotsandhumansisthatrobot-controllingdevicerunsanpersonalityinterfaceprocessinbackground.BluetoothdiscoveryprogramrunsonmobiledevicestosearchforothernearbyBluetoothdevicesandndouttheircommunityIDs.ThecommunicationstructurebetweenmobiledevicesonrobotsandhumanisillustratedintheFig. 6-3 6.3.3ParticipatoryTestingTheotherpartofthedesignisparticipatorytesting.NetworkofRobotsislimitedinscalabilityatboththenumberofnodesandsizeofthespacebecauseofcostsanddifcultyofdeploymentinthelargespace.Therefore,robotsmayshowlimitedandunrealisticmobility.Byusingcrowdsourcing,wecanenhancethetestingenvironmentsignicantlyatbothofthecriteria.Participatorytestingisanovelideathatuseshumansocietyastestbed.Humancohortswhoarevoluntaryparticipantsinexperimentofnetworkingprotocolsarecrowdsourceinthistestbed.Participantscandownloadandinstallthenetworkingprotocolstotheirdevicesandcarrythemaround.Withproliferationofsmartphonesthesedays(e.g.,iPhone,Android),itiseasiertoseekforhumancohorts.Weconductanexperimentwithsmallsetofhumansubjectsandseektoexplorethepotentialofparticipatorytestinginmobilenetworks.Specically,werecruitedgraduate 80

PAGE 81

Figure6-3. Communicationstructureamongrobots,personalityinterface,andcommunicationprotocol studentstakingacomputernetworkingcourseincomputerscienceovertwosemesters.EachhumansubjectcarriedeitherHPiPAQ(fallsemester)orNokiaN800/810(springandfallsemesters)oncampus.BluetoothdiscoveryprogramranineachofmobiledeviceandrecordedBluetoothencounterevery90seconds.Eventhoughmorehumanparticipatesinfallsemester,morerecordsarecollectedinspringsemesterbecauseoflongerlengthofexperiment.Whereas,moreuniquenumberofdevicesareencounteredinfallsemester.Itisworthnotingthatafewparticipants(lessthanthreeintotal)failtorecordproperlyduetomisuseofdevicesortheirscheduletogooutoftown.We 81

PAGE 82

learnedfromourexperiencethatevensmall-scaleexperimentrequiresextensiveefforttoeducatetheparticipantsandwemaystillexpectafewexceptions.EncountermeasurementshavebeenperformedbymultipleresearchprojectssuchasMITsRealityMining[ 61 ]andHaggleprojects[ 38 ].Yet,thetotalnumbersofparticipantswaslimitedtolessthanahundred.ResearchersusedWLANtraceasanindirectwayofmeasuringencounters[ 1 39 ]becauseofitslargesamplesize.Weovercomethisscalabilityissuebypubliclyrecruitinghumansubjects.Specically,wecreateapubliccommunityforthetestbedsouserscandownloadtheprogramstoparticipateinthetest.Theseparticipantsformalargetestbed.Toachievethisgoal,itisessentialtodevelopaprogrammableinterfacetomakethetestingeasier.WeimplementedcommunicationprogramsontheWindowsMobile-based(HPiPAQ)andLinux-basedMAEMO(NokiaN810)smartphones.Manyscenariosandprotocolscanbetestedthroughthisparticipatorytesting,includingtheDTNroutingprotocolsbasedonsocialmodels.TheProle-cast[ 3 ]isoneofsuchprotocolsthatprovidecommunicationbasedonthebehavioralprolesofhumansviasocialsensing.Amessageisroutedbasedonuserprolesasafunctionoftheusersmobilitypreferences.TheProle-castwastestedwithstudentsincomputernetworkingclassin2010spring.However,thesystemhasyettobetestedinapublicforum,whichisnotourfocusofthisdemonstration. 6.3.4LimitationofControlledandUncontrolledEnvironmentParticipatorysensingisacrowdsourcingofsensordatafromvoluntaryparticipants[ 64 ][ 67 ].Participatorytestingissimilartoparticipatorysensinginthatitleansonhumanparticipants;however,thetaskisevenharderasparticipantsshouldbeengagedactivelyintestingovercertaindurationoftime.Thisinvolvessensingmobilecontactsandrunningnetworkingprotocolsorservices.Itbecomesparticularlydifcultwhentherearetoomanyhumanparticipantstomanage.Forinstance,participantsmaynotfollowtheexperimentinstruction.Further,maliciousparticipantsmaydamageor 82

PAGE 83

misleadtheexperimentresults.Therefore,participatorytestingisachaoticenvironmentwhereconsiderableefforttocontrolthetestingisrequired,yet,maystillhavedifcultyingatheringcorrectexperimentdata.Hence,thereismuchdemandforatestingenvironmentwhererealisticsubjectscanbeusedinanexperimentwhilecontrollabilityissupported.Testingenvironmentwithautonomousmobilenodesprovideexibilityincontrolwhilesimulatinghumanmobility.Capturingcommunitybehaviorofhumanmobileusersandembeddingthecommunityinformationastheirprolescanbeastepforward.Inthefollowingsection,weprovidedetailsofprolingmobileusersbasedonreal-worlddataanalysisanduseofresultsincreationofrule-baseddecisioncriteriaforindividualmobilenodes.Weshowfromsimulationofautonomousmobilenodesthatproposedapproachescancreatearealistictestingenvironmentwhilecollectivelymatchingreal-worldcontactpatterns. 6.4ContactRule-BasedDecisionCriteriaItisachallengingtasktoreplicatethecontactpatterninadistributedimplementationwithoutlocationinformation.Heuristicmobilitydecisionisacriticalfactorinthisproblemasavailableinformationformobilityislimited.Plausiblemobility[ 62 ]infersmobilitytracefromencountertraceusingdraw,repulsionandattractionproperties.Attractionisanintentiontomovetowardsothernodes;repulsionisaforcetomoveawayfromparticularnodes;dragisapropertytostayandnottobeaffectedbyothernodes.Itrequiresglobalknowledgeofallnodes'encounterinformationandlocationinformation.Implementationofdrawandrepulsionareviablebyallowingnodestostayawayfromeachotherwhentheyaresupposedtoencounterinfrequentlyandmakingthemstopwhentheyaresupposedtoencounterfrequently,however,attractionpropertycannotbeimplementedaseachnodedoesnothaveinformationofothernodes.Hence,itisnotappropriatetoimplementonautonomousmobilenodesthatworkinadistributedfashion.Adopting 83

PAGE 84

Figure6-4. Statediagramoffriend-strangerdecisionmodel thesimilaridea,however,adecisionmechanismbasedoncommunityinformationofencounterednodescanachievethegoal.Oneassumptioninformingacommunityisthatgroupmembersaredecidedbasedontheircontactdurationandperiodicity;thus,theobjectiveofgroupmemberistondtheirowngroupmembersandmaintainperiodicalcontactwiththem.Tothisgoal,weusethesimilarideasinplausiblemobilityformakingdecisionsofnextmovement.Repulsionisimplementedinawayofgettingawayfromothergroupmembersuponencounter.Drawistriggeredwhenencounteringmultiplegroupmembers.Attractionisimplementedbyslowingdownsearchspeedofgroupmembersandthoroughlysearchingthenearbyareauntilndingmoregroupmembers.Thisalgorithmdoesneitherperfectlyndallofthegroupmembersnorpreventcontactswithothergroupmembers.However,assmallworldinwirelessnetworksanalysis[ 57 ]shows,afewmoreaddedlinkscanbridgemostofthedisconnectednetworks;hence,unavoidablecontactswithothergroupmembersarestillrealistic.Friend-Strangerisadecisioncriteriausedwhenencounteringgroup/non-groupmembers.Adecisionismadeaccordingtoacommunityidentity(friendorstranger)of 84

PAGE 85

Figure6-5. Mobilityofsourcenode(Me)ineachdecision encounterednodes.Afriendisanodebelongingtothesamecommunity(i.e.groupmembers)andallothernodesareconsideredasstrangers(i.e.non-groupmembers).Asdiscussedintheprevioussection,anodemeetsfriendsformorethancertaintimeduration,whilemeetsstrangersforashorttime.ThestatediagraminFigure 6-4 showsfourstatesthatareincludedinthedecisioncriteria.Meisthenodethatismakingadecisioninthiscase;however,notethatotherencounterednodesalsomaketheirowndecisionsasMenodedoes.Fourscenariosareintroduced:1)Whentherearenonodesnearby,Meisinasearchmode,whichmakestheMenodetomovewithfastspeedinsquareshapedirection.2)Uponencounteringafriend,Megoesintoaslowsearchmodewhereitassumestheremightbeotherfriendnodesaroundandslowsdownthespeedofsearch.Further,thenodemovesinoctagonshapedirectiontosearchtheareathoroughly.3)WhenMediscoversmultiple 85

PAGE 86

friendnodes,MestopsandholdsthepositionbecausethisindicatesthatMeprobablyislocatedinsideafriendcommunity.4)Whenthereisfarmorenumberofstrangersthanfriends,Meescapesfromthecurrentpositionbygoingintoarun-awaymode.Itmovesbackwardtomoveoutoftheareafullofstrangers.Additionally,friendnodesthatcontactedmorethandurationthresholdareconsideredashavingafullcontact.Thismeansthereisnoneedofmaintainingcontactswiththesamenode;thus,theyareconsideredasstrangersthoughtheircommunityinformationstillindicatesthattheybelongtothesamecommunity.Thisensuresthenodestomovearoundwithoutbeingstuckinoneplacetocontactthesamenodesrepeatedly.Inthispresentation,thethresholdfordurationis240minutesperday.MobilityofsourcenodeineachdecisionisdescribedinFigure 6-5 .Proposeddecisionmodel,however,haslimitationasstayinginacommunitycanleadtomissingoutopportunitiestocontactwithothercommunitymembers.Weproposedthismodelinourposterpresentation[ 71 ]withuniformdistribution.Themodelwasabletomimiccontactdurationandperiodicity,however,didnotcapturethecontactdaysandfrequency.Weintroduceseveralmethodstoimproveinthisareasothatthegroupcontactbehaviorsshowclosercontactpatterntoreal-worldtraceinmultiplemetrics. 6.4.1Power-lawDistributionOurnewdiscoveryofpower-lawdistributionforthesizesofthecommunitiescanleadtoadifferentresultthanpreviousimplementationofuniformdistribution.Unlikeequalnumberofcommunitymembersforeachcommunity,communityIDscanbeassignedtofollowthepower-lawdistribution.Specically,distributionofcommunitysizescanfollowtheslopeof-1.48inlog-logscalethatmatchesgroupsizesfromcontactpattern.Anotherimprovementtogeneratecloseresultstoreal-worldcontactpatterncomesfrommodiedrule-basedalgorithm.Insteadofstayinginonelocationuponencounteringmultiplecommunitymembers,anodecanchoosetoexploretheareatomovetowardalocationwheremorememberscanbefound.Thesamelogicapplies 86

PAGE 87

Table6-1. Propertiesineachnode'sprole ProlePropertyDescription ContactfrequencyForcontactednodesContactdurationForcontactednodeson-offDayindaysofaweekon-offTimeinaday toaslowsearchmode,whereitcansupportofmovementtoalocationwheremorecommunitymembersarefoundafterthoroughsearchinsteadofstoppingimmediatelyupondiscoveringmultiplenodes.Thisimprovementbringsmorecontactswithcommunitymembersthatthenodeisalreadycontactingaswellastheopportunitiestocontactwithothernearbymembers. 6.4.2MemoryAddingamemorytorememberthelocationwhereanodehascontactedthemostnumberofnodesrecentlycanhelpthenodetoavoidunnecessarycontactswithotherstrangers.Thus,havingaMemorytorememberallowmobilenodestoshowmorecontactswithfriendscomparedtocontactswithstrangers. 6.5ExperimentInhissection,wereporttheexperimentresultsofsimulationcomparedwithreal-worldcontacttrace(UFtrace)[ 35 ]. 6.5.1SimulationSetupParametersforsimulationismadeclosetoarealworldmobilenetworktrace. 6.5.1.1CommunityproleEachmobilenodemaintainsitsowncontactprole.Table 6-1 showsthepropertiesinaprole.Eachnoderecordscontactdays,frequencyanddurationwithothernodesinitsprole,whichisusedtodetermineafullcontact.inaddition,on-timeinformationiscarriedbyeachmobilenode,whichweexplaininthefollowingsection. 6.5.1.2On-timeschedulerStudyshowsthatcontactpairsexhibitperiodicalencounterfromanalysisofWLANusersandBluetoothcontactsbymobileusers[ 46 ].Toemulatethisbehavior,weadopt 87

PAGE 88

Table6-2. Schedulerforon-timeperiod DailyscheduleProbability Daytime(9am-5pm)0.98Eveningtime(5pm-1am)0.017Dawntime(1am-9am)0.003 Table6-3. Schedulerforon-timeduration On-timedurationProbability 0-8hours0.78-16hours0.216-24hours0.1 aschedulertoeachmobilenode.AsshowninTable 6-2 ,anodehason-offdayinaweekandon-offtimeinaday.Eachnodeisassignedrandomon-offtimeinadayandweek.Inaddition,toreectarealisticschedule,differentweightsareputforstartingtimeofaday,on-timedurationofadayandbetweenweekdaysandweekends.Parametersforon-timewithaweightindailyscheduleisshowninTable 6-2 .Forweekdays,anodeisgiven99%chancetobeactiveforeachdayand1%chancetobeactiveduringweekends.Schedulerforon-timedurationshowsinTable 6-3 .Nodesappearbetween0-8hourswithprobabilityof0.7,8-16hourswith0.2and16-24with0.1.Thesevaluesarefromhumanobservationandnotobtainedfromthetraces;thus,theydonotaccuratelyreectthetrendinthenetworktraces. 6.5.1.3EnvironmentsetupTable 6-4 showstheenvironmentsetupforsimulationandUFtrace.Numberofcommunitiesissettothetotalnumberofnodes20=50.Speedofslowsearchis10meters/minand40meters/minforfastsearchandrunawaymode.Eachmobilenodescans40metersofradiustodiscoverothernodes.InUFtrace,nodesthatshowedfor Table6-4. Experimentenvironment(UFdataisanapproximatevalue.) DomainSimulationUFtrace Space1000meters5000metersNodes10001000Period64days64daysCommunities5050Granularity10seconds1second 88

PAGE 89

Table6-5. Experimentmodes NameMobilityCommunity RandommobilityrandomwalkuniformUniformdistributioncontactrule-baseddecisionuniformPowerlawdistributioncontactrule-baseddecision+schedulerpower-lawMemorycontactrule-baseddecision+memory+schedulerpower-lawNoSchedulecontactrule-baseddecision+memorypower-lawUFtracerealWLANuserspower-law atleast20%ofthedayswereselected.BothofsimulationandUFtraceareexaminedfor64daysforcontactratioandgroupranksizeexperiments,and32daysforperiodicity(spectral)analysis. 6.5.2ExperimentModesWedescribetheexperimentmodesusedforcomparisonsandmetrics.1)Randommobility:Thisismobilitymodethatiscurrentlyembeddedinthemostoftheexistingtestbeds.Thismodeldoesnotembedrule-baseddecisioncriteria.2)Uniformdistribution:Thismodeappliesbasicrule-baseddecisioncriteriawithfourrulesinadditiontothescheduler.Communitiesareassigneduniformly.3)Power-lawdistribution:TotheUniformdistributionmode,communitiesareassignedinapower-lawdistributionfashion.Additionalmobilityinstayandsearchmodeisalsoapplied.4)Memory:TothePower-lawdistribution,locationmemorywherethenodecontactedmostnumberoffriendsisadded.5)Noscheduler:TotheMemory,theschedulerisremoved.6)UFtrace:Thisisaprocessedcontacttrace[ 46 ].ContactstatisticsisanalyzedfromtheUFWLANtracewithacommonassumptionthatWLANusersaccessingthesameaccesspointsarewithinsignalrangeofwirelesscontact[ 46 ][ 1 ].Summaryofthemodesareshownin 6-5 .Theexperimentsareperformedforcontactdays,frequencyandduration.Themetric,contactdays,indicatesthenumberofdaysacontacteventoccurredwithanothernode.Contactfrequencyiscountingthenumbercontactsbetweennodes.Contactdurationisdurationoftimeinminutesbetweencontactednodes. 89

PAGE 90

Figure6-6. Simulationtoolshowsthemobilityandencounterofautonomousmobilenodes.Eachcircleindicatesasignalrangetodetectothernearbymobilenodes.Eachcommunityisidentiedbydifferentcolors.Extraemptyinnercircleisforthenodesinslowsearchmode.Nodesthatareholdinginthecurrentpositionsaredisplayedwithextrainnercirclesthatarelledwithacolor. 6.5.3ResultAnalysisOursimulationtoolwedevelopedvisualizesthemobilityofautonomousnodesaswellastheirsignalrangetoshowifencounteredasinFig. 6-6 .Weturnoffthevisualizationfeaturewhengeneratingstatisticstoproduceresultsquickly.Thissimulationtool,codeandvideoofmobilenodesareavailableon-linein[ 35 ]. 6.5.3.1ContactratiowithfriendstostrangersFigure 6-7 showscontactratiooffriendstostrangers.Higherratioindicatesgreaterdegreeofcontactwithfriendsthanstrangers.Resultsshowninthisgureindicatethatdifferentmethodscancapturedifferentmetrics.Contactdurationandfrequencyarecapturedbymostofthemethods.ContactdaysarecapturedinMemorymethodasinUFtrace.Randommobilityistheonlymethodthatdoesnotshowcommunitycontactbehaviorinanyofmetrics.Thisshowsourproposedmethodsworkforcapturingcommunitycontactpatternfordifferenttypesofscenarios. 90

PAGE 91

ARatioforcontactdays BRatioforcontactfrequency CRatioforcontactdurationFigure6-7. Contactratiooffriendstostrangersaccordingtodifferentmetrics.Eachmethodsshowsdifferentresultscomparedtoreal-worldpattern(UFtrace). 6.5.3.2RankgroupsizeFigure 6-8 showstherankplotsinlog-logscaleforthesizeofgroups.Hierarchicalclusteringwasappliedafterencountervectorinformationwasgenerated.Theplotsshowthatdifferentmodesfollowpower-lawdistributionmorecloselyfordifferentmetrics.Randommobilityanduniformdistributiondonotexhibitanyclosenesstothepower-lawdistributionforanymetrics.Thisshowstheimportanceofcapturingcommunitiesbyunderstandingthereal-worldpatternrst.Theresultalsoshowsthatourproposedrule-basedmodescanshowtheclosepatterntoreal-worldcommunitydistributiondependingonthemetrics. 91

PAGE 92

ARankgroupsizeforcontactdays BRankgroupsizeforcontactfrequency CRankgroupsizeforcontactdurationFigure6-8. Rankplotinlog-logscaleaccordingtothesizeofgroupsafterhierarchicalclustering.Randommobilityshowsleastclosetopower-lawdistributionofgroupsizes.Differentmodesshowdifferentclosenesstopower-lawdistribution. 6.5.3.3PeriodiccontactpatternPeriodiccontactpatternwithfriendsisanalyzedinFig. 6-9 .Thegureshowsspectralanalysisofcontactpatterninfrequencydomain.Interestedreaderinthisanalysisisencouragedtoread[ 46 ].Inthegure,X-axisindicatesthefrequencyoftherepetitivepatternovertheperiodof32days,whileY-axisindicatesthemagnitudeforthefrequencyofrepetitivepatterns.Fromthegraph,weeklyencounterpattern(peakat 92

PAGE 93

APeriodicityanalysisforcontactdays BPeriodicityanalysisforcontactfrequency CPeriodicityanalysisforcontactdurationFigure6-9. Spectralanalysisforcontactwithfriendsnodesover32days.Peakat4and5indicatesstrongweeklypattern.Randommobilityandnoscheduleraretheonlymodesthatdonothavescheduler,thus,displayingnoperiodicpattern. frequency4and5)appearsparticularlystrongformostofthemodeswiththescheduler.Thisweeklypatternwasalsoobservedfromvariousreal-worldtraces[ 46 ].RandommobilityandNoschedulerexhibitnodistinctrepetitivepattern. 6.5.4VisualizationofEncounterandMobilityVisualizationhelpstodiscovercertainpatternsthatarenotwellobservedinstatisticalanalysis.Italsogivesaroughideaofwheretostartanalysis.Oursimulationtoolcanshowthecurrentrule-baseddecisionstatusofallnodes.Ingraphicalrepresentationofmobileencounters,mobilenodesshowsimilartrendtostatisticalresults.SnapshotsofvisualsareshowninFigure 6-10 .Animationvideoandexecutablelesofeachmodeisavailableon-lineat[ 35 ].Randommodebarely 93

PAGE 94

ARandommode BPower-lawmode CMemorymode DNoschedulemodeFigure6-10. Snapshotsofsimulationforeachmodewithvisualizationon.Randommodesshowtheleastnumberofcommunityconvergence.Randommodeandnoschedulemodeshowssignicantlymorenumberofnodesbecausetheschedulerdoesnotexistforbothofthemodes.Notethatthissnapshotonlyshowsthestatusofcertaintime;thus,thestatuscandifferovertime.However,thissnapshotshowsthecommonstatusthroughoutthesimulation. showsthecommunityencountersaslledinnercirclesarerarelyobserved.Power-lawmodeshowsmorecommunityconvergence.Italsoshowslessnumberofnodesthanrandommodebecauseaschedulecontrolsthenumberofnodesthatappearsduringeachperiod.Communityconvergenceisobservedmoreinmemorymode.Finally,noschedulemodeshowsmorenumberofnodesalongwithrandommodesasallofthenodesappearatthesametime.Comparedtorandommode,however,noschedulemodeshowssignicantlymoreconvergenceofcommunity.Notethatthissnapshotonly 94

PAGE 95

showsthestatusofcertaintime;thus,thestatuscandifferovertime.However,thissnapshotshowsthecommonstatusthroughoutthesimulation. 6.6ImplementationonAutonomousRobotsInthistestbedimplementation,communityinformationispreconguredforeachmobiledevice.TherearethreeconcurrentprocessesrunninginaiRobotcontroller:1)BluetoothconnectionsetupwithaniRobot;2)Bluetoothscanning;and3)Personalityinterface.BluetoothsetupisnecessaryafterpairingofBluetoothbetweenthecontrollingmobiledeviceandtheiRobot.ItcreatesavirtualtunneltosenddatafrommobiledevicestotheiRobots.AnotherprocessisBluetoothscanningprogram,whichprobesthenearbyBluetoothdevicesbysendingbeaconsignalforeveryfewseconds.ItlogstheMACaddressandtimestampinale.Theotherprocessispersonalityinterfacethatmakesmobilitydecision.Itreadsthelogleandndtheencounteredmobilenodes'identitiesfromtheirMACaddressandcomparewithcommunityinformationithas.NextbehavioroftheiRobotisdecidedbyrule-baseddecisioncriteria.PersonalityinterfacesendscommandstotheiRobotaftermakingamovementdecision.WedemonstratedthisprototypetestbedwithtwoiRobotsandtheircontrollerstogowith12otherNokiaPDAsintheconferences[ 72 ][ 73 ][ 74 ][ 75 ].12NokiaPDAsweretaggedwithblueandredcolortoindicatetheircommunityIDs.RandomattendantsoftheconferencesparticipatedinthedemonstrationbycarryingthePDAsandemulatingtheappearance/disappearanceofthemselvesbyturningon/offthevisibilityofBluetooth.Programcodesandvideosofthetestbedareavailablein[ 35 ]. 6.6.1ControllingiRobotForeasyprogrammingandwidedeploymentsofthedevice,wechoosetheiRobottoformthenetworksofrobots.Theconnectionbetweenamobiledevice(e.g,.HPiPAQ)andtheiRobotisperformedviaBluetoothcommunication.Inthisway,theiRobotcanbecontrolledremotely.Thepersonalityinterfacecansendcommandsandreceivesensorreadingsto/fromiRobot.Forcompatibilityandportability,weusemobiledevices 95

PAGE 96

insteadoflaptops.AsiRobotdoesnotcontrolitselfbutisrathercontrolledbythepersonalityinaseparatecomputer(mobiledevice),itisthepersonalityinterfacethatdecidesthenextbehavioroftheiRobotuponreceivingsignalreadingsfromtheiRobotoracommunicationonthemobiledevices.Theproposedstructurecanincorporateanycommunicationmethods/protocolsbyusingthemountediRobotcontrollerasacommunicationdevice.ThecontrollercancommunicateviaBluetoothorWLANformessagedeliveryandproleexchange.IntheProle-castimplementation,weuseWLANforcollectinglocationinformationtocreateprolesandBluetoothformessageexchanges.WecreateapersonalityontheiRobotbysettingupafewbehavioralrules.Variouspersonalityprolescanbecreatedasmentionedabovedependingonthescenarios,protocols,andexperimentalenvironments.Viableexamplesare1)abehavioralsignatureoflocationvisitingpreferences;2)regular/irregular/randomcontactpatternswithothermobilenodes;3)attractiontofriendlycommunityandrepulsiontounfriendlycommunity.Theirmovementsareachievedbythethreerulesmentionedearlier:attraction,repulsionanddraw.Uponencounteringthedesiredtarget,iRobotrecognizesitandstops.Ifthetargetnodemovesaway,thepersonalityinterfacetriggerstheattractionsothatitcangetclosertothetargetnode.RepulsionistriggeredtogetawayfromthecurrentlyencounteringnodewhentheiRobothasspentenoughtimeaccordingtothegivenpersonalityprole. 6.6.2LabEnvironmentOurmobiletestbeddoesnotrequiretohaveanyunusualfacilityorenvironmentforexperiments.SpacetodeployiRobotisnecessarybutnotrequiredastherobotscanmovebetweenattendees.iRobotandPDAsmayrequirebatterycharge;therefore,outletpowerstripswillbeneeded.Posterboardtoexplainthedesignandadesktoputthedevicesarealsonecessary.Bythedemonstration,notonlywepresentanoveltestbeddesignbutalsoshowcaseourresearchresultssuchasanalysisofperiodicityin 96

PAGE 97

encounterandprole-caseprotocol.Inaddition,thisdemocanleveragenewresearchideasconcerningparticipatorytestingandpromotepotentialcollaborationonthelarge-scaledeploymentoftheexperiments.WecreateanenvironmentwheretheiRobotmovearoundwithgivensetofinstructionstoreectthebehavioralprole.Forcommunicationprotocol,weimplementProle-castthatshowtwotypesofdeliverymodes:1)todelivermessagebundlestotargetnodeswithmatchingbehavioralproles,or2)todisseminatemessagestotheinterestednodeswhensuchprolesareunavailable.DemonstratingaprotocolsuchaProle-castonalarge-scaleisourgoalforthetestbed;however,itrequiresabehavioralprolecollectionstep,whichassociatesthenodeswithlocation-visitingpreferences.WethereforemodifytheProle-castprotocoltotintheconferenceenvironment,yetstillrevealthefullconceptofProle-cast.Ourdemonstrationscenariointhelabenvironmentisasfollows:1)Selecteduserscarrythemobiledevices.Thesedevicescollecttheencountertraceforthemobilecarrier.2)Basedonthecollectedtraces,weplantapersonalityproletotheiRobotsmimickingeachuser.3)MobiledevicescontainProle-castimplementation,andmessageexchangeistested.4)Evaluatetheresultsaftertheexperimentisover. 6.6.3EvaluationScenariosforAutonomousRobotsThereareseveralfactorsneedtobeconsideredinpresentingthesetestbeds.First,thescenarioshouldbeabletostartandendinagiventimewhileshowingthefullaspectsofthemaincomponentsinthetestbeds.Anotherimportantfactoristhetimelimit.Topresentademonstrationofthetestbedsinarelativelysmallspace,itisimperativetolimitthescopeofthetestbeds.However,itstillneedstoshowthecoreideaofthedesignandstrengthsoftheimplementation.Theotherfactortoconsideriseffectivelyshowingthemainunderlyingconceptsofthetestbeds.Inaconferenceenvironment,notonlythepresentationtimeandspaceofthetestbedsislimited,butalsotimeandbasicknowledgeforlectureislimited.Yet,itisavaluableopportunitytootherresearchersseekingfortheideaoftestbedstotesttheirdevelopedconceptsand 97

PAGE 98

results.Inordertoeffectivelydelivertheideaanddesignofthetestbeds,itisnon-trivialtohaveconcisescenarios.Wepresentthreedemonstrationscenariosintheconferenceenvironment,whichwehaveimplementedandshowedintheconference[ 72 74 ].Thesescenariosstartfromthebasicscenariothatisthebasisofthesubsequentscenarios.Scenariosareincrementalinthedifcultyofunderstandinganddisplayingtheimplementationdetails.1)BasicscenarioTherstscenarioweintroduceisthemostbasicscenariothatshowstheconceptoffriendshipanditsincorporationintothetestbeds.Inthisscenario,asinglerobotisused.ThisrobotwithanattachmentofNokiaN810asitsbrainmovesforwardandbackwarduntilitndsafriend.ItdiscoversafriendbysearchingaBluetoothdevicewhosesignature(MACaddress)belongstoagroupofitsfriends.Upondiscovering,therobotstopsthemovementtoshowthatitiscurrentlystayingwithfriends;thus,thereisnoneedofmovementanylongerasfriendsarenearby.Thisshowsthebasicscenarioofcommunitybasedencounter.Thisfriendshipinformationisstoredinthememoryspaceofthemobiledevice(i.e.NokiaN810)controllingtherobot.Thefriendshipinformationisbasedonsimilarityofthelocationvisitingpreference;therefore,itiseasytouseprole-castincaseofmessagepropagation.Asprole-castisalreadyimplementedinthemobiledevice,thereisnoneedoffurtherdevelopmentoftheprole-castimplementationorportingissue.ThesameprogramforcontrollingtherobotandcommunicationprotocolcanbeimplementedinanyofLinuxbasedsystem(i.e.laptoprunningontopofLinux).Notethatfrienddevicescanbemultipledevicesorsingledevice.Toshowtheappearanceoffriendmobiledevice(s),eitherthefriendmobiledevicecomeclosertotheBluetoothsignalrangeofthedevicecontrollingtherobotorwecanemulateappearance/disappearanceoffriendbehaviorbyturningon/offtheBluetoothdeviceofthefriendmobiledevice.2)FriendvsenemyWithsimilarimplementation,morecomplexscenarios,yet,describingthecapabilityofthetestbedsispossible.Thesecondscenario,thus, 98

PAGE 99

introducesadditionalelementtofriendconcept-enemy.Welaterexplainthevariousapplicationsofthefriends-enemiesconceptinmobilitymodelingandcommunicationprotocol.Toachievethisgoalofshowingcommunitybasedbehaviors,wesetupabehavioralruleandaddaspeedelementtoshowthebehavioralrule.WeuseasinglerobotandmultipleBluetooth-enabledmobiledevices(i.e.smartphonewithBluetooth)inthisscenario.Thedesigngoalofthisscenarioistoshowthecommunitypropertyofourtestbeds.Toachievethisgoal,wedeneagroupoffriendsandagroupofenemies.Thesegroupsareobtainedbyrunningthesimilaritybasedcommunitymodel.Hence,theobjectiveoftherobotistostaytogetherwithafriendcommunityandtostayawayfromanenemycommunity.Thisscenariohasdifferentbehaviorsaccordingtotheidentityofthediscoverednodes-1)Nofriendsandenemies,2)Onefriend,3)Multiplefriendsand4)Numberofenemies>numberoffriends.Therobotreactstotherespectivecircumstanceby1)Fastsearch,2)Slowsearch,3)Stopand4)Panic.Specically,whentherearenofriendsandenemies,therobotwillsearchfastforitsfriends.Weimplementthisbyshowingthebehaviorofmovementinasquaredirection.Specically,therobotwillgoforwardfast,thenturnby90degreeandrepeatthesamebehaviors.Thisisasearchphraseandweemulatetheintensesearchoffriendsbytherobot.Whenitdiscoversafriend,itattemptstogetclosertothefriend,yet,itstilltriestondtheotherfriendswithanassumptionthatfriendstendtoocktogether.Thus,itslowsdownitssearchspeedandmovesforwardslowlyforgettingclosetothefriendcommunity.Whenitnallydiscoversmultiplefriends,itconcludesthatitreachestothecoreofthefriendcommunityandstopsthemovement.However,whenitdiscoversanenemy,therobotactsbasedonthebehavioralrule.Accordingtothebehavioralrule,therobotgoesintoapanicmodewhenthenumberofenemiesnearbyisgreaterthanthenumberoffriendsnearby.Regardlessofpreviousstates,therobottriestomoveawayfromtheenemygroupwhilesearchingforitsfriendsgroupinpanicmode.This 99

PAGE 100

Figure6-11. Snapshotofaproofofconceptvideoformobilenetworkingtestbed panickingsituationisemulatedbymovingbackandforthfast.Thisbehaviorsimulatesabehaviorofgettingawayfromenemies.3)TeamscenarioLastscenario,wedemonstrateisteam-basedbehavior.Furtherextendingfriendsandenemycommunity,welabeleachteamasredandblueteam.Thistime,weusetwoormorerobotsandeachteamhasatleastonerobot.Thebasicbehavioralruleisthesameasthesecondscenario.Robotstrytostaywiththeirownteamswhilesearchingforitsteammembers,stayingwiththeteamwhentherearemultiplemembersdiscovered,andgettingawayfromtheotherteammemberswhenoutnumbered.Thisbehaviorsimulatestheteam-basedstrategy.Multiplemobiledevicesareassignedtoeachteamarbitrarily.Figure 6-11 isasnapshotofaproofofconceptvideoformobilenetworkingtestbed.Thevideoisavailableat[ 35 ]anditshowstheteamscenario. 100

PAGE 101

6.7ConclusionandFutureWorkWeproposeanoveltestbedconceptofblendinganetworkofrobotsandparticipatorytesting.Wealsoshowhowtoembedacommunityproleandproposeacommunitycontactrule-baseddecisioncriteriaalongwithafewmodicationstooperateonautonomousmobilenodes.Thesimulationresultsmatchhuman-likecontactpatternintermsofcontactdays,frequencyanddurationalongwithperiodiccontactpatternwithproposedapproaches.Ourproposedapproachesmaketheautonomousmobilenodetoformcommunitiesthatshowpower-lawdistributionforthesizeofthegroupsasitshowsinreal-worlddata.Thisisastepforwardtoanimplementationofrealistictestbedaseachnodewithpersonalityprolecanreplicatehumancontactpatternwithoutneitherofaglobalknowledgenorlocationinformationofanyothernodes.Finally,weprovideimplementationofpersonalityprolesonrobotsthatreactsbasedontheircontactinformation.Ourworkisthersttoinvestigatethisdirectionandopensopportunitiestowardsbuildingrealisticmobiletestbedusingpersonalityprolestoevaluatemobilesocialnetworks.Futureworkincludesdevelopmentofrulesandmobilityroutesofautonomousnodestoemulatedifferentscenarios.Targetprolestomimiccanbemixedamongnodestoshowdifferentmobilitypatterns.Implementationofmobilesocialnetworkingprotocolsorembeddingprolesondifferentphysicalnodes(i.e.legorobots)isanotherfutureworks.VisualizationofmobilenodescanalsobeextendedbydisplayingonGoogleEarth.Usingtheframeworkwedevelopedinourpreviousworks[ 76 ][ 77 ][ 78 ],decisionstatusofeachmobilenodecanbevisualizedonGoogleEarth. 101

PAGE 102

CHAPTER7CONCLUSIONSWediscusstheconclusionsandfutureworkofthisdissertation. 7.1ConclusionsInordertounderstandhumanencounterbehavior,weanalyzetheperiodicityinencounteredpairsandindividualnodesundervariousconditions[ 46 79 80 ].Wecategorizethemaccordingtotheirdailyencounterrateandshowedtheperiodicitywiththefollowingmetrics:dailyandhourlyencounter,encounterfrequencyandencounterduration.Forthemajorityoftheencounteredpairs,aweeklyencounterpatternisprevalent,whichmobilitydiameterpatternstudydidnotobserve.Wealsoobservethatperiodicityappearedstrongerfortherarelyencounteringpairsthanthefrequentlyencounteringpairs.Incaseofrareencounterevents,theregularencounterpatternisparticularlyusefulasitcanprovidetheestimationforthenumberofrequiredrelaynodestosatisfythegivendeliveryprobability.Wealsoproposeviableapproachestodiscovertheregularlyencounteringpairs.Ouranalysisshowstheutilityofspectralanalysisforcharacterizingencounterregularity,whichisvitalforfuturemobilenetworks.Additionally,weshowedthatregularlyencounteringpairsmayhavedifferentlocationvisitingpatterns,whichfurtherreinforcestheimportanceofregularencounterpattern.Weanalyzethereal-worldencounterdatasets(Bluetoothencounter)andtheWLANtraceswithadequateassumptionforlarge-scaleencounterdata,whereperiodicityinvariousset-upwascommonlyobservedatbothtypesoftraces.Tosumup,ouranalysisshowstheutilityofspectralanalysisforcharacterizingencounterregularity,whichisvitalforthestudyoffuturemobilenetworks.Ourperiodicityanalysisisuniqueinthatwe1)investigatetheperiodicityandregularityofnodalencounterpatternbyusingpowerspectralanalysis,2)proposenewapproachestodiscovertheregularpatternandanalyzetheregularlyencounteringpairs,and3)analyzearichsetofrealworlddataforlongperiodsoftime,includingupto50,000userspersemester 102

PAGE 103

periodoverthreeyearsandBluetoothtracesintwosemesters.Inadditionwestudytheconsistencyofencounterpatternformobileusersaccordingtothechangesinsizeofpastwindow.Theresultsobtainedsofar,indicatesthatmoreknowledgeislikelyleadingtobetterprediction.Thistrendismoreapparentwithfrequentlyencounterednodes.Furthermore,weshowthatregularlyencounteringpairsaresignicantlymoreconsistentthanirregularlyencounteringpairs.Ourprolingstudyofmobileusersshowedthatgroupingbasedonencounterdaysresultsinpower-lawdistributionofsizesingroups.Wecomparedthedistributiontolocationbasedgroupingandobservedsimilardistribution.Basedonthisunderstanding,weapplythegroupinginassigningcommunitiesonautonomousmobilenodes.Hence,communitycontactproleembeddedoneachmobilenodeisbuilttofollowthispower-lawdistributionofcommunity.Weproposeanovelmobilesocialnetworkingtestbed.First,weblendanetworkofrobotsandparticipatorytestbedbyembeddingapersonalityproleonautonomousmobilenodes.Byemulatinghumanmobility,focusingoncommunitycontactpattern,wetakethestrengthsofbothtestbedcomponents,thus,becomesabridgebetweenthem.Toemulatehumancontactpattern,weembedcommunitycontactrule-baseddecisioncriteriaonautonomousmobilenodesalongwithaschedulerandamemory.Simulationresultsshowthatdifferentmodescanemulatedifferentmetricsofcontactpattern,includingcontactdays,frequencyandduration.Groupsize,encounterratiowithfriendsandspectralanalysisareprovidedforexperimentswiththesemetrics.Furthermore,weprovideaprototypeimplementationofthetestbedusingiRobotCreateandNokiaPDAs,whichwedemonstratedinmanyconferencesandavailableon-lineat[ 35 ]. 7.2FutureWorkFutureWorkincludesthemobilenetworkingprotocolsandservicesusingtheobtainedunderstandingofhumancontactpattern.Ourcontactproleembeddingonautonomousmobilenodesshowspromisingdirectionofcreatingatestbedwith 103

PAGE 104

variousscenarios.Actualtestingofmobilesocialnetworkingprotocolsonthetestbedisanotherdirectionofresearch.Ourunderstandingonperiodicityofcontactpatterncanalsobeimplementedformobileapplicationsthatusehumancontactpatternsuchasadvertisementandinterest-basedmessagepropagation. 104

PAGE 105

REFERENCES [1] W.Hsu,D.Dutta,andA.Helmy,Miningbehavioralgroupsinlargewirelesslans,inProc.ACMMobiCom,September2007. [2] K.Fall,Adelaytolerantnetworkarchitectureforchallengedinternets,inProc.ACMSigComm,2003,http://dtnrg.org. [3] W.Hsu,D.Dutta,andA.Helmy,Prole-cast:Behavior-awaremobilenetworking,inProc.IEEEWCNC,March2008. [4] E.M.DalyandM.Haahr,Socialnetowrkanalysisforinformationowindisconnecteddelay-tolerantmanets,IEEETransactionsonMobileComputing,vol.8,no.5,May2009. [5] P.Hui,J.Crowcroft,andE.Yoneki,Bubblerap:social-basedforwardingindelaytolerantnetworks,inProc.ACMMobiHoc,May2008. [6] P.Hui,A.Chaintreau,J.Scott,R.Gass,J.Crowcroft,andC.Diot,Pocketswitchednetworksandhumanmobilityinconferenceenvironments,inProc.ACMSIG-COMMworkshoponDelay-tolerantnetworking,August2005. [7] A.Hussain,J.Heidemann,andC.Papadopoulos,Identicationofrepeateddenialofserviceattacks,inProc.IEEEINFOCOM,April2006. [8] M.KimandD.F.Kotz,Periodicpropertiesofusermobilityandaccess-pointpopularity,PersonalandUbiquitousComputing,Springer-Verlag,vol.11,August2007. [9] M.C.Gonzalez,C.A.Hidalgo,andA.-L.Barabasi,Understandingindividualhumanmobilitypatterns,Nature,,no.453,pp.779,June2008. [10] W.Hsu,T.Spyropoulos,K.Psounis,andA.Helmy,Modelingspatialandtemporaldependenciesofusermobilityinwirelessmobilenetworks,IEEE/ACMTransac-tionsonNetworking,vol.17,no.5,October2009. [11] G.S.Thakur,U.Kumar,A.Helmy,andW.jenHsu,Ontheefcacyofmobilitymodelingfordtnevaluation:Analysisofencounterstatisticsandspatio-temporalpreferences,inProc.IWCMC,June2011. [12] A.Miklas,K.Gollu,K.Chan,S.Saroiu,K.Gummadi,andE.Lara,Exploitingsocialinteractionsinmobilesystems,inProc.UbiComp,September2007. [13] A.Lindgren,A.Doria,andO.Schelen,Probabilisticroutinginintermittentlyconnectednetworks,LectureNotesinComputerScience,vol.3126,pp.239,September2004. 105

PAGE 106

[14] L.SongandD.F.Kotz,Evaluatingopportunisticroutingprotocolswithlargerealisticcontacttraces,inACMMobiComworkshoponChallengedNetworks(CHANTS),September2007. [15] C.Song,Z.Qu,N.Blumm,andA.-L.Barabasi,Limitsofpredictabilityinhumanmobility,Science,vol.327,no.5968,February2010. [16] P.HuiandJ.Crowcroft,Humanmobilitymodelsandopportunisticcommunicationsystemdesign,RoyalSocietyPhilosophicalTransactions,,no.366,pp.1872,June2008. [17] S.TanachaiwiwatandA.Helmy,Encounter-basedworms:Analysisanddefense,Elsevier,AdHocNetworks,vol.7,September2009. [18] P.Wang,M.C.Gonzalez,C.A.Hidalgo,andA.-L.Barabasi,Understandingthespreadingpatternsofmobilephoneviruses,Science,vol.324,no.5930,May2009. [19] V.VukadinovicandS.Mangold,Opportunisticwirelesscommunicationinthemeparks:astudyofvisitorsmobility,Sep. [20] S.Grasic,E.Davies,A.Lindgren,andA.Doria,Theevolutionofadtnroutingprotocol-prophetv2,Sep. [21] A.Aljarhi,H.Arafa,S.G.Aly,andK.A.Harras,Rethinkingopportunisticroutingusingspacesyntax,inCHANTSProceedingsofthe6thACMworkshoponChallengednetworks,2011. [22] A.Krifa,C.Barakat,andT.Spyropoulos,Mobitrade:Tradingcontentindisruptiontolerantnetworks,inCHANTSProceedingsofthe6thACMworkshoponChal-lengednetworks,2011. [23] P.De,A.Raniwala,S.Sharma,andT.Chiueh,Mint:Aminiaturizednetworktestbedformobilewirelessnetwork,inProc.IEEEINFOCOM,2005. [24] J.Reich,V.Mishra,andD.Rubenstein,Roombamadnet:Amobilead-hocdelaytolerantnetworktestbed,ACMMobileComputingandCommuncationReview(MC2R),January2008. [25] B.Walker,I.Vo,M.Beecher,andM.Seligman,Ademonstrationofthemeshtestwirelesstestbedfordtnresearch,inProc.ACMMobiComworkshoponChallengedNetworks(CHANTS),2008. [26] D.Johnson,T.Stack,R.Fish,D.M.Flickinger,L.Stoller,R.Ricci,andJ.Lepreau,Mobileemulab;aroboticwirelessandsensornetworktestbed,inProc.IEEEINFOCOM,2006. [27] S.Bromage,Scorpion:Aheteroheneouswirelessnetworkingtestbed,ACMMobileComputingandCommuncationReview(MC2R),2009. 106

PAGE 107

[28] T.Karagiannis,J.-Y.L.Boudec,andM.Vojnovic,Powerlawandexponentialdecayofintercontacttimesbetweenmobiledevices,inProc.ACMMobiCom,September2007. [29] U.Kumar,G.S.Thakur,andA.Helmy,Protect:proximity-basedtrust-advisorusingencountersformobilesocieties,inProc.IWCMC,June2010. [30] U.KumarandA.Helmy,Proximity-basedtrust-advisorusingencounters,MobileComputingandCommunicationsReview,vol.14,October2010. [31] U.Kumar,G.S.Thakur,andA.Helmy,Proximitybasedtrust-advisorusingencountersformobilesocieties:Analysisoffourlters,December. [32] U.KumarandA.Helmy,Humanbehaviorandchallengesofanonymizingwlantraces,inProc.GlobeCom,December2009. [33] H.ZangandJ.C.Bolot,Anonymizationoflocationdatadoesnotwork:Alarge-scalemeasurementstudy,inProc.ACMMobiCom,September2011. [34] D.F.Kotz,Crawdad:Communityresourceforarchivingwirelessdataatdartmouth,2008,http://crawdad.cs.dartmouth.edu. [35] A.Helmy,Archieveatnomadsgroupinciseattheuniversityoforida,2008,http://nile.cise.u.edu. [36] P.Hui,A.Chaintreau,J.Scott,R.Gass,J.Crowcroft,andC.Diot,Pocketswitchednetworksandhumanmobilityinconferenceenvironment,inProc.ACMSigcommworkshoponDTN,August2005. [37] J.LeBrunandC.-N.Chuah,Bluetoothcontentdistributionstationsonpublictransit,inProc.MobiShare,September2006. [38] J.Crowcroft,P.Hui,C.Diot,J.Scott,A.Chaintreau,andR.Gass,Haggleproject,2006,http://www.haggleproject.org. [39] A.Chaintreau,P.Hui,J.Crowcroft,C.Diot,R.Gass,andJ.Scott,Impactofhumanmobilityonthedesignofopportunisticforwardingalgorithms,inProc.IEEEINFOCOM,April2006. [40] C.Chateld,AnalysisofTimeSeries,pp.18,105,ChapmanandHall,1989. [41] S.-J.Lee,W.Su,andM.Gerla,Wirelessadhocmulticastroutingwithmobilityprediction,August. [42] W.Su,S.-J.Lee,andM.Gerla,Mobilitypredictionandroutinginadhocwirelessnetworks,January. [43] M.H.SunandD.M.Blough,Mobilitypredictionusingfutureknowledge,inProc.MSWiM,2007. 107

PAGE 108

[44] I.F.AkyildizandW.Wang,Thepredictiveusermobilityproleframeworkforwirelessmultimedianetworks,IEEE/ACMTransactionsonNetworking,vol.12,no.6,pp.1021,2004. [45] M.A.Bayir,M.Demirbas,andN.Eagle,Discoveringspatiotemporalmobilityprolesofcellphoneusers,inWoWMoM,2009. [46] S.MoonandA.Helmy,Understandingperiodicityandregularityofnodalencountersinmobilenetworks,inIEEEGlobeCom,December2010. [47] K.P.Wei-jenHsu,ThrasyvoulosSpyropoulosandA.Helmy,Modelingtime-variantusermobilityinwirelessmobilenetworks,inProc.IEEEINFOCOM,May2007. [48] G.S.Thakur,A.Helmy,andW.jenHsu,Similarityanalysisandmodelinginmobilesocieties:themissinglink,inACMMobiEval1stinternationalworkshoponSystemevaluationformobileplatforms,2011. [49] F.Tan,Y.Borghol,andS.Ardon,Emo:Astatisticalencounter-basedmobilitymodelforsimulatingdelaytolerantnetworks,inWoWMoM,2008. [50] M.A.BayirandM.Demirbas,Pro:Aprole-basedroutingprotocolforpocketswitchednetworks,inProc.GlobeCom,2010. [51] A.MoghadamandH.Schulzrinne,Interest-awarecontentdistributionprotocolformobiledisription-tolerantnetworks,inProc.WoWMoM,June2009. [52] E.Yoneki,P.Hui,andJ.Crowcroft,Distincttypesofhubsinhumandynamicnetworks,inEuroSysWorkshoponSocialNetworkSystems,April2008. [53] M.Goldberg,S.Kelley,M.Magdon-ismail,K.Mertsalov,andA.Wallace,Findingoverlappingcommunitiesinsocialnetworks,inProc.SocialCom,2010. [54] N.P.Nguyen,T.N.Dinh,S.Tokala,andM.T.Thai,Overlappingcommunitiesindynamicnetworks:Theirdetectionandhowtheycanhelpmobileapplications,inProc.ACMMobiCom,Sep2011. [55] E.M.DalyandM.Haahr,Thechallengesofdisconnecteddelay-tolerantmanets,AdHocNetworks,vol.8,pp.241,2010. [56] W.jenHsuandA.Helmy,Onnodalencounterpatternsinwirelesslantraces,IEEETransactionsonMobileComputing,vol.9,no.11,November2010. [57] A.Helmy,Smallworldsinwirelessnetworks,CommunicationsLetters,vol.7,pp.490,2003. [58] S.MoonandA.Helmy,Designandimplementationofmobilenetworkingtestbedswithpersonalityproles,inProc.IWCMC,July2011. [59] V.BorrelandM.D.Amorim,Simps:Usingsociologyforpersonalmobility,IEEE/ACMTransactionsonNetworking,vol.17,no.3,2009. 108

PAGE 109

[60] A.Mei,G.Morabito,P.Santi,andJ.Stefa,Social-awarestatelessforwardinginpocketswitchednetworks,inProc.(Mini-Conference)INFOCOM,2011. [61] N.Eagle,A.Pentland,andD.Lazer,Inferringsocialnetworkstructureusingmobilephonedata,Proc.oftheNationalAcademyofSciences(PNAS),vol.36,pp.15274,September2009. [62] J.Whitebeck,M.Amorim,V.Conan,M.Ammar,andE.Zegura,Fromencounterstoplausiblemobility,PervasiveandMobileComputing,vol.7,2011. [63] M.A.M.Vieira,R.Govindan,andG.S.Sukhatme,Scalableandpracticalpursuit-evasionwithnetworkedrobots,vol.2,no.4,pp.247,2009. [64] J.Burke,D.Estrin,M.Hansen,A.Parker,N.Ramanathan,S.Reddy,andM.Srivastava,Participatorysensing,inWorldSensorWebWorkshop,2006. [65] E.Miluzzo,N.D.Lane,K.Fodor,R.Peterson,H.Lu,M.Musolesi,S.B.Eisenman,X.Zheng,andA.Campbell,Sensingmeetsmobilesocialnetworks:thedesign,implementationandevaluationofthecencemeapplication,inProc.SenSys,2008. [66] S.Gaonkar,J.Li,R.R.Choudhury,L.Cox,andA.Schmidt,Micro-blog:sharingandqueryingcontentthroughmobilephonesandsocialparticipation,inProc.MobiSys,2008. [67] M.Mun,S.Reddy,K.Shilton,N.Yau,J.Burke,D.Estrin,M.Hansen,E.Howard,R.West,andP.Boda,Peir,thepersonalenvironmentalimpactreport,asaplatformforparticipatorysensingsystemsresearch,inProc.MobiSys,2009. [68] E.Nordstrom,C.Diot,andR.Gass,Experiencesfrommeasuringhumanmobilityusingbluetoothinquiringdevices,inACMMobiEval1stinternationalworkshoponSystemevaluationformobileplatforms,2007. [69] E.Cuervo,P.Gilbert,B.Wu,andL.P.Cox,Crowdlab:Anarchitectureforvolunteermobiletestbeds,inProc.COMSNETS,2011. [70] G.W.Challen,M.Demirbas,S.Y.Ko,andT.Kosar,Phonelab:Alarge-scaleparticipatorysmartphonetestbed,inProc.NSFWorkshoponPervasiveComput-ingatScale(PeCS),Jan.2011. [71] S.MoonandA.Helmy,Contactrule-baseddecisionformobilesocialnetworkingtestbed,inACMMobiCom(Poster),2011. [72] S.Moon,G.S.Thakur,J.Kim,Y.Wang,S.Mogahaddam,andA.Helmy,Integrationofrobotsandhumanbehaviorformobiletestbeds,inWINTECHworkshop,co-locatedwithACMMobiCom(Demo),September2010. [73] S.Moon,G.S.Thakur,J.Kim,Y.Wang,S.Mogahaddam,andA.Helmy,Mobiletestbedswithanattitude,inACMMobiCom(Demo),September2010. 109

PAGE 110

[74] S.MoonandA.Helmy,Mobiletestbedswithanattitude,inACMGlobeCom(Demo),December2010. [75] S.MoonandA.Helmy,Mobiletestbedswithanattitude,Tech.Rep.,arXiv.org,Sep2010. [76] S.Moon,U.Kumar,J.Kim,W.jenHsu,andA.Helmy,Visualizationandrepresentationofmobilenetworkusers,inProc.IEEESECON,June2008. [77] U.Kumar,S.Moon,W.jenHsu,andA.Helmy,Visualizationandrepresentationofmobilenetworkusers,inIEEEINFOCOMdemosession,April2008. [78] J.Kim,W.jenHsu,S.Moon,U.Kumar,andA.Helmy,Visualizationandrepresentationofmobilenetworkusers,inIEEEICNPposter,October2008. [79] S.MoonandA.Helmy,Spectralanalysisofperiodicityandregularityformobileencountersindelaytolerantnetworks,ACMSIGMOBILEMobileComputingandCommunicationsReview,vol.14,October2010. [80] S.MoonandA.Helmy,Understandingperiodicityandregularityofnodalencountersinmobilenetworks,Tech.Rep.,arXiv.org,Aug2010. 110

PAGE 111

BIOGRAPHICALSKETCH SungwookMoonwasborninSeoul,SouthKoreain1978.Hisinterestsintechnologyandsciencegrewnaturallybylookinguptohisfarther,YoungkeyMoon,whostudiedappliedphysicsincollege.SungwookgraduatedfromSangmoonHighSchoolinSeoulandthenattendedSogangUniversity,Seoulin1997.HeservedmandatorymilitaryservicefromOct1999toDec2001,wherehewasattachedtotheU.S.8thArmyandreceivedARCOMandAAMmedalsforhisleadership.Aftercomingbacktocollege,HereceivedB.S.inComputerSciencefromSogangUniversityandM.S.inComputerEngineeringattheUniversityofFloridain2004and2006respectively.Sungwookmarriedhiswife,JinYunin2004andtheirdaughterKaylinSaeyunMoonwasbornin2009.Hebeganhisdoctoralstudyin2006andreceivedhisPh.D.fromtheUniversityofFloridainthefallof2011.Hisresearchinterestliesintheeldofmobileandsocialnetworking,prolingofmobileusersandnetworkingtestbed.Asidefromresearch,Sungwookwasinuencedbyhismother,SeungjaBae,whostudiedclassicalmusiccompositionincollege.Sungwooklovesmusicandhadseveralchancestoperforminfrontofthousandsofaudience,includingchorusperformancewithSogangChorusasabassinaprestigiousartcenterinSeoul.HewasalsoamemberofclassicalguitarteamandperformedintwoofcialregularperformancesinmarryhallatSogangUniversity.DuringhistimeatUniversityofFlorida,heenjoyedwatchingandtalkingaboutGatorsfootballandbasketballwithfellowGators.Morethananything,heloveshisfamilythemost. 111