Citation
Resource Management in Wireless Networks

Material Information

Title:
Resource Management in Wireless Networks
Creator:
CHEN, XIANG ( Author, Primary )
Copyright Date:
2008

Subjects

Subjects / Keywords:
Bandwidth ( jstor )
Local area networks ( jstor )
Motor vehicle traffic ( jstor )
Multimedia materials ( jstor )
Service time ( jstor )
Simulations ( jstor )
Timing devices ( jstor )
Traffic delay ( jstor )
Traffic estimation ( jstor )
Traffic loads ( jstor )

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Xiang Chen. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Embargo Date:
8/31/2008
Resource Identifier:
58564926 ( OCLC )

Downloads

This item is only available as the following downloads:


Full Text

PAGE 1

RESOURCEMANAGEMENTINWIRELESSNETWORKSByXIANGCHENADISSERTATIONPRESENTEDTOTHEGRADUATESCHOOLOFTHEUNIVERSITYOFFLORIDAINPARTIALFULFILLMENTOFTHEREQUIREMENTSFORTHEDEGREEOFDOCTOROFPHILOSOPHYUNIVERSITYOFFLORIDA2005

PAGE 2

Copyright2005byXiangChen

PAGE 3

TOMYPARENTSMYWIFE,YINMYSISTERANDBROTHER

PAGE 4

ACKNOWLEDGMENTSFirstandforemost,Iwouldliketoexpressmysinceregratitudetomyadvisor,Pro-fessorYuguangFang,forhisinvaluableguidance,encouragement,andsupportthroughoutmyPhDstudyattheUniversityofFlorida.Ihavebenetedalotfromhiswisdomwithregardtoworkandlife.IthankProfessorsJoseFortes,TanWong,Chien-LiangLiu,andShigangChenforservingonmydissertationcommitteeandfortheirvaluablesuggestionsandconstructivecriticism.ManythanksareduetomycolleaguesHongqiangZhai,WeiLiuandJianfengWangfortheircollaboration.IalsowanttothankDr.WenjingLou,Dr.Byung-SeoKim,Yan-chaoZhang,ShushanWen,XiaoxiaHuang,ChiZhangandothersintheWirelessNetworksLaboratoryfortheirhelpfuldiscussions.Immersedinsuchanexcitingandwarmenviron-ment,Ireallyhavehadagoodandrewardingexperience.Finally,Ioweaspecialdebtofgratitudetomyparents,mywife,andmysisterandbrother.Withouttheirselessloveandsupport,IwouldneverimaginewhatIhaveachieved.ThisdissertationisbasedupontheworksupportedinpartbytheNationalScienceFoundationFacultyEarlyCareerDevelopmentAwardundercontractANIR0093241.Theviews,opinions,and/orndingscontainedinthisdissertationarethoseoftheauthorsandshouldnotbeinterpretedasrepresentingtheofcialpolicies,eitherexpressedorimplied,oftheNationalScienceFoundation. iv

PAGE 5

TABLEOFCONTENTS page ACKNOWLEDGMENTS ................................ iv LISTOFTABLES ................................... viii LISTOFFIGURES ................................... ix ABSTRACT ....................................... xii CHAPTER 1INTRODUCTION ................................ 1 1.1Overview ................................. 1 1.2RelatedWork ............................... 4 1.2.1ResourceAllocationSupportingConnection-LevelQoS ..... 4 1.2.2SupportingDifferentiatedQoSinMANETs ........... 6 1.2.3SupportingQoSinWLANs .................... 8 1.3MainContributions ............................ 9 1.4OrganizationofThisDissertation ..................... 10 2DYNAMICRESERVATION-BASEDCALLADMISSIONCONTROLINCELLULARNETWORKS ............................. 12 2.1Introduction ................................ 12 2.2SystemModel ............................... 13 2.3ProposedScheme ............................. 15 2.3.1ConnectionAdmissionPolicy ................... 15 2.3.2Adaptationof G 1 , G 2 ,and G 3 ................... 15 2.3.3UpdatingFrequency ........................ 20 2.4PerformanceEvaluation .......................... 21 2.5Conclusion ................................. 26 3DYNAMICMULTIPLE-THRESHOLDBANDWIDTHRESERVATION(DMTBR) INCELLULARNETWORKS ......................... 28 3.1Introduction ................................ 28 3.2ProposedScheme ............................. 29 3.2.1QoSCriteria ............................ 29 3.2.2ConnectionAdmissionPolicy ................... 31 3.2.3CooperationamongCells ..................... 32 3.3ThresholdAdaptation ........................... 35 v

PAGE 6

3.3.1Adaptationof G 1 ......................... 35 3.3.2Adaptationof G 2 ......................... 36 3.3.3Adaptationof G 3 ......................... 37 3.3.4FurtherAdaptationofThresholds ................. 37 3.3.5ProbabilityofThrottling ...................... 38 3.3.6EstimationofArrivalRates rt and nrt ............. 40 3.3.7UpdatingFrequency ........................ 41 3.4PerformanceEvaluation .......................... 41 3.5Conclusion ................................. 51 4LOCATION-AWARERESOURCEMANAGEMENTINMOBILEADHOC NETWORKS .................................. 52 4.1Introduction ................................ 52 4.2SystemModel ............................... 55 4.3BandwidthPre-Reservation ........................ 57 4.3.1Pre-ReservationRequestPropagation ............... 58 4.3.2BandwidthPre-ReservationCriterion ............... 61 4.3.3UpperBoundofBandwidthtoBePre-reserved .......... 67 4.3.4IntervalofSendingPre-reservationRequests ........... 68 4.3.5ControlOverheadAnalysis .................... 68 4.4LocationAwareForwarding ........................ 69 4.4.1BandwidthCalculationandReservation ............. 69 4.4.2ImpactonRe-routing ....................... 71 4.5PerformanceEvaluation .......................... 71 4.6Discussions ................................ 78 4.6.1Mobility .............................. 78 4.6.2TradeoffbetweenOverheadandPerformance .......... 78 4.7Conclusion ................................. 83 4.8Appendix ................................. 83 4.8.1DerivationofConnectionBlockingProbability .......... 83 4.8.2FunctionsUsedinLAFA ..................... 84 5QOSSUPPORTINIEEE802.11EWIRELESSLANS ............. 88 5.1Introduction ................................ 88 5.2OperationsoftheIEEE802.11e ...................... 89 5.3QoSRequirementsforMultimediaServices ............... 90 5.4DelayEstimationoftheIEEE802.11e .................. 92 5.4.1SaturatedCasevs.UnsaturatedCase ............... 93 5.4.2MarkovChainModelfortheIEEE802.11e ........... 93 5.4.3G/M/1QueueModeltoEstimateMeanDelay .......... 95 5.4.4G/G/1QueueModeltoEstimateMeanDelay .......... 97 5.4.5ModelValidation ......................... 98 5.5DerivationoftheUpperBounds ..................... 100 5.5.1ProbabilityDistributionofMACServiceTime .......... 102 vi

PAGE 7

5.5.2UpperBoundsoftheAverageDelayandDelayVariation .... 105 5.5.3ModelValidation ......................... 106 5.6CallAdmissionandRateControlFramework .............. 107 5.6.1ChannelBusynessRatio ...................... 109 5.6.2CallAdmissionControlSchemeI ................. 110 5.6.3CallAdmissionControlSchemeII ................ 113 5.6.4RateControl ............................ 113 5.6.5Determinationof U max ...................... 115 5.7PerformanceEvaluation .......................... 116 5.7.1SimulationConguration ..................... 116 5.7.2SimulationResults ......................... 118 5.8CONCLUSION .............................. 125 6CONCLUSIONSANDFUTUREWORK .................... 126 6.1Contributions ............................... 126 6.2FutureWork ................................ 127 REFERENCES ..................................... 129 BIOGRAPHICALSKETCH .............................. 137 vii

PAGE 8

LISTOFTABLES Table page 3-1Simulationparametersfor DMTBR A and DMTBR G .......... 42 5-1Themean,standarddeviation(SD),and97'th,99'th,99.9'thpercentiledelays(s)forvoiceandvideowhenCACschemeIandRCareused. .... 121 5-2Themean,standarddeviation(SD),and97'th,99'th,99.9'thpercentiledelays(s)forvoiceandvideowhenCACschemeIIandRCareused. .... 122 viii

PAGE 9

LISTOFFIGURES Figure page 1-1Architectureofcellularnetworks ....................... 2 2-1Connectionadmissionpolicy ......................... 15 2-2Wrap-aroundsimulationmodel ........................ 21 2-3Connectionblockingprobabilityforreal-timetrafc ............. 23 2-4Handoffconnectiondroppingprobabilityforreal-timetrafc ........ 23 2-5Connectionblockingprobabilityfornon-real-timetrafc .......... 24 2-6Handoffconnectiondroppingprobabilityfornon-real-timetrafc ...... 24 2-7Throughput .................................. 25 2-8Performancevs.changingupdatefrequency ................. 26 2-9Performancevs.changing ......................... 27 3-1ThebandwidthreservationthresholdsinDMTBR .............. 33 3-2Connectionadmissionpolicy ......................... 34 3-3Reservationthresholdsadaptation ....................... 39 3-4Connectionblockingprobabilityinahomogeneousenvironment ...... 43 3-5Handoffconnectiondroppingprobabilityinahomogeneousenvironment .. 44 3-6Trafcthroughputinahomogeneousenvironment .............. 45 3-7Systemthroughputinahomogeneousenvironment ............. 46 3-8Throttlingprobability.(a) DMTBR A .and(b) DMTBR G . ....... 47 3-9Performancevs.QoS ............................. 48 3-10Performancevs.updatefrequency ...................... 49 3-11Connectionblockingprobabilityinaheterogeneousenvironment ...... 49 3-12Handoffconnectiondroppingprobabilityinaheterogeneousenvironment . 50 3-13Trafcthroughputinaheterogeneousenvironment .............. 50 ix

PAGE 10

3-14Systemthroughputinaheterogeneousenvironment ............. 51 4-1Illustrationofaquadrangle-shapedinuencearea .............. 60 4-2Exampleofpre-reservationfailure ...................... 61 4-3Locationandtimedivision .......................... 64 4-4Pre-reservationcriterion ............................ 66 4-5Anexampleoflocation-awarepre-reservation ................ 66 4-6Connectionblockingprobability.(a)i-connection.(b)ordinaryconnection. 73 4-7Re-routingsuccessprobability.(a)i-connection.(b)ordinaryconnection. .. 74 4-8Throughput .................................. 75 4-9Averagenumberofpre-reservedtimeslots .................. 75 4-10Pre-reservationrequestsperreceivedpacket ................. 76 4-11Connectionblockingprobability.(a)i-connection.(b)ordinaryconnection. 79 4-12Re-routingsuccessprobability.(a)i-connection.(b)ordinaryconnection. .. 80 4-13Throughput .................................. 81 4-14Averagenumberofpre-reservedtimeslots .................. 81 4-15Pre-reservationrequestsperreceivedpacket ................. 82 4-16Apathforderivation ............................. 84 5-1RTS/CTS/DATA/ACKfour-wayhandshake ................. 91 5-2802.11earchitecture .............................. 91 5-3Markovchainforthe802.11ebackoffprocedure ............... 96 5-4Averagedelay.(a) AIFS [2]=60 s , AIFS [3]=50 s , W 2 ; 0 =32 ,and W 3 ; 0 =16 .(b) AIFS [2]=75 s , AIFS [3]=50 s , W 2 ; 0 =64 ,and W 3 ; 0 =16 . ................................. 101 5-5PGFdiagramforthebackoff ......................... 103 5-6Delayperformancewhen AIFS [2]=60 s , AIFS [3]=50 s , W 2 ; 0 =32 and W 3 ; 0 =16 .(a)Mean.(b)Standarddeviation. ............. 108 5-7Overviewofthecalladmissionandratecontrolframework ......... 109 5-8Choiceof U max fordifferentpacketlengths .................. 117 x

PAGE 11

5-9Choiceof U max forprioritizedtrafcanddifferentnumbersofnodes .... 117 5-10Aggregatethroughput ............................. 121 5-11Channelbusynessratioandchannelutilization ................ 121 5-12Averagedelayofvoiceandvideotrafc ................... 122 5-13Delaydistributionofvoiceandvideotrafc ................. 122 5-14Aggregatethroughput ............................. 123 5-15Channelbusynessratioandchannelutilization ................ 123 5-16Averagedelayofvoiceandvideotrafc ................... 124 5-17Delaydistributionofvoiceandvideotrafc ................. 124 xi

PAGE 12

AbstractofDissertationPresentedtotheGraduateSchooloftheUniversityofFloridainPartialFulllmentoftheRequirementsfortheDegreeofDoctorofPhilosophyRESOURCEMANAGEMENTINWIRELESSNETWORKSByXiangChenAugust2005Chair:Yuguang“Michael”FangMajorDepartment:ElectricalandComputerEngineeringAsthenumberofusersisincreasinglygrowingandvarioustypesofservicessuchasaudio,video,anddataaresupported,resourcemanagementplaysanimportantroleinthedesignandanalysisofwirelesscommunicationnetworks.Essentially,thegoalofresourcemanagementistoefcientlyutilizenetworkresourcee.g.,bandwidthwhileprovidingqualityofserviceQoSforapplications.Inthisdissertation,weproposedifferentresourcemanagementschemestosupportdiverseQoSinthreetypesofwirelessnetworks.Incellularnetworks,wepresenttwodynamicreservation-basedcalladmissioncontrolCACschemestosupportconnection-levelQoSintermsofnewconnectionblockingprob-abilityandhandoffconnectiondroppingprobabilityformultimediatrafc.BothschemesadoptmultiplebandwidthreservationthresholdsthataredynamicallyupdatedaccordingtonetworktrafcsituationsandQoSstatus.Particularly,thedynamicmultiple-thresholdbandwidthreservationDMTBRschemecanproviderelativeprioritybetweenreal-timeandnon-real-timetrafc.Tosupportservicedifferentiationinmobileadhocnetworks,weproposeadistributedlocation-awareapproach.Inthisapproach,alocation-awarebandwidthpre-reservationmechanismutilizeseachnode'sgeographiclocationinformationtopre-reservebandwidth xii

PAGE 13

forhighpriorityconnectionsandhencereducespotentialschedulingconictsfortrans-missions.Inaddition,anend-to-endbandwidthcalculationandreservationalgorithmisproposedtomakeuseofthepre-reservedbandwidth.Asaresult,thisschemecanprovideservicedifferentiationandefcientlyusethenetworkbandwidth.Finally,westudyprovidingstrictQoSinsteadofdifferentiatedQoSintheemergingIEEE802.11eWLANs.Werstbuildananalyticalmodeltoanalyzethedelayperformancefortrafcofdifferentprioritiesintheunsaturatedcase.WeshowthattheQoSrequirementsofreal-timetrafccanbesatisediftheinputtrafcisproperlyregulated.Then,weproposeacalladmissionandratecontrolframeworkthatoperatesonthebasisofthedelayanalysisandthechannelbusynessratio,whichcanaccuratelyrepresentthenetworkstatus.TheframeworkhasbeenshowntosupportstrictQoSrequirementsofreal-timetrafcandatthesametimeachievehighchannelutilizationbyallowingbest-efforttrafctofullyusetheresidualbandwidth. xiii

PAGE 14

CHAPTER1INTRODUCTION1.1OverviewAsaresultoftheadvancementofwirelesstechnologyandtheproliferationofhand-heldwirelessterminals,recentyearshavewitnessedanever-increasingpopularityofwire-lessnetworks.Dependingontheservicecoverage,wirelessnetworkscanbedividedintowirelesslocalareanetworksWLANs[44]andwirelesswide-areanetworksWWANsorwirelesscellularnetworks[25,41,79,84,85].InWLANsorWWANs,mobilehostscommunicatewithanaccesspointorabasestationthatisconnectedtothewirednetworks.Obviously,onlyonehopwirelesslinkisneededforcommunicationsbetweenamobilehostandastationaryhostinwirednetworks.Unlikethosetypesofwirelessnetworks,mobileadhocnetworksMANETsdonotrelyonthesupportofxedinfrastructuresuchasbasestationsoraccesspoints.Rather,aMANETconsistsofself-organizingmobilehoststhatcommunicatewitheachotherthroughmultiplewirelesshops,withtheintermediatenodesactingasroutersbetweenasourceandadestination.Asthenumberofusersisincreasinglygrowingandtheofferedserviceportfoliosarebecomingdiverse,resourcemanagementplaysanever-increasingroleinthedesignandanalysisofwirelesscommunicationnetworks.Essentially,resourcemanagementisresponsibleforefcientutilizationofnetworkresourcee.g.,bandwidthwhileprovidingqualityofserviceQoSguaranteestoapplications[34,55,76,80].ThisdissertationdealswiththedesignandanalysisofresourcemanagementschemesthatareaimedtoprovideQoSindifferenttypesofwirelessnetworks.Specically,weconsidertwotypesofQoS,connection-levelQoSandpacket-levelQoS,whichwillbedescribedinthefollowing.AsshowninFig. 1-1 ,awirelesscellularnetworkistypicallyorganizedintogeograph-icalareascalledcellssothatthesameradiochannelsmaybereusedinothercellssome 1

PAGE 15

2 Figure1-1:Architectureofcellularnetworks distanceaway[79].Ineachcell,thereisabasestationBSwhichservesmobileuserswithinthecellcoverage.Further,amobileswitchingcenterMSCconnectstheuserstothepublicswitchedtelephonenetworkPSTN.Tocommunicatewithotherusers,amo-bileusermustissuetotheBSanewconnectionrequest,inwhichasetofdesirableQoSneedsareexplicitlyorimplicitlyspecied.TheBSthenwillmaketheadmissiondecisionbasedonthepremisethattheQoSofcurrentlyexistingconnectionswillnotbesacriced.GrantinganadmissionisequivalenttoacontractwherethenewlyconnectedserviceisguaranteedthesetofQoSrequirementsinthedurationoftheconnection.Whenamobileuserengagedinanongoingconnectionmovesfromonecelltoanother,ahandoffprocessoccurs.Toprovideuninterruptedservicetotheuser,thenetworkneedstomakesurethecelltheusermovesintohasenoughbandwidthtoaccommodatetheconnection.Anim-portantmeasureisthehandoffconnectiondroppingprobabilityCDP,whichisdenedastheprobabilitythatahandoffconnectionisdropped.Meanwhile,wearealsointerestedinthenewconnectionblockingprobabilityCBP,whichisdenedastheprobabilitythatanewconnectionrequestisblocked.Theconnection-levelQoSischaracterizedbybothCBPandCDP[19,42,78].

PAGE 16

3 Packet-levelQoSisusuallymeasuredinthroughput,delayanddelayjitter,andpacketlossrate.Thesemetricsbeardifferentlevelsofimportancefordifferentapplications.Forexample,real-timeserviceslikevoiceoverIPandteleconferencehaveverystrictdelayrequirement,butdataserviceslikeemailandletransferaredelay-tolerant.Ontheotherhand,real-timeservicescantolerateacertainlevelofpacketlosswhilethedataservicescannot.QoSsupportinwirednetworkshasbeenextensivelyresearched.IncombinationwiththeIntegratedServicesIntServarchitecture[10]andtheDifferentiatedServicesDiff-Servarchitecture[9],avarietyofmechanismsincludingscheduling,bandwidthreserva-tion,QoSrouting,andcalladmissioncontrolhavebeenproposed[16,24,31,76,100,101].However,theycannotbedirectlyappliedtothewirelessscenarioaswirelessnetworksfea-turesomeuniquecharacteristics. Itiswellknownthatwirelesschannelsarenotasreliableaswiredchannels.Asamatteroffact,theyaretime-varyingduetointerference,multipathfading,andshadowing.Asaresult,thechannelerrorrateisnotxedandneitheristhechannelbandwidth. Forsimplicityandlowcost,currentWLANsareoperatedonthecollision-basedanddistributedmediumaccesscontrolMACscheme[44].Duetothelackofcoordina-tion,guaranteeingQoSisverydifcult.Furthermore,thingsbecomemuchworseinmulti-hopwirelessnetworksasaresultofthehiddenterminalandexposedterminalproblems[6,52]. Inmobilewirelessnetworks,eachnodecanmovearound.ThismobilitycauseshandoffsincellularnetworksandWLANs.Thehandoffprocesstakesplacemorefrequentlywhenthecellsizehastobereducedinordertoaccommodatethein-creasingnumberofmobileusersandbandwidthrequirementsbydataorvideoap-plicationsthatdemandmorebandwidththantraditionalvoiceservices.InMANETs,mobilitymayresultinfrequentroutebreakagesandrenderend-to-endQoSprovisionextremelyhard. ComparedtowirednetworksthatenjoybandwidthashighasseveralhundredGbps,wirelessnetworkshavelimitedbandwidththatistypicallyatleastoneorderofmag-nitudelower,fallingintherangeofseveralhundredkbpstotensofMbps.TosupportQoSinwirelessnetworks,extensiveresearchhasbeencarriedouttoover-comethoseobstacles.Next,wewilldiscusstherelatedwork.

PAGE 17

4 1.2RelatedWork1.2.1ResourceAllocationSupportingConnection-LevelQoSThequestforefcientlyallocatingnetworkbandwidthamongdifferenttrafcclasseswhilegivinghandoffcallshigherprioritythannewcallrequestsoverwirelesscellularnetworkshasbeencourtingextensiveefforts.Consequently,agoodnumberofschemeshavebeenproposed.Thoseschemescanbemainlyclassiedintotwocategories:staticschemesanddynamicschemes.Therstandmostwell-knownstaticschemewastheGuardChannelorcutoffprior-ityschemeproposedbyHongandRappaport[42].Inthisscheme,handoffconnectionsaregivehigherprioritythannewconnectionsbypurposefullyreservinganumberofchan-nelsforhandoffconnections.Thus,thedroppingprobabilityofhandoffconnectionsisreduced.InChangetal.[13]andGuerin[36],somevariantsoftheschemewereproposed,whichbuffernewconnectionsorbothnewandhandoffconnectionsinthequeue.WhiletheguardchannelschemeisshowntobeoptimalforalinearobjectivefunctionoftheCBPandtheCDP,afractionalguardchannelscheme,duetoRamjeeetal.,isoptimalintermsofminimizingtheCBPsubjecttoahardconstraintontheCDP[78].Alltheseschemesarestaticinthesensethatthenumberofguardchannelsisdeterminedmainlybasedonaprioriknowledgeofthetrafcpatterns,therebybeingunabletocopewithnetworkdynam-ics.Moreover,onlyonetrafcclass,voicetrafc,isconsidered.Toallocatebandwidthforvoiceanddatatrafcinanintegratedmobilenetwork,Huangetal.[38]proposedamovableboundaryscheme,andYinetal.[94]proposedadual-thresholdreservationDTRscheme.Lietal.[61]analyzedandcomparedthehandoffperformanceofthesetwoschemes.Inanoptimalcalladmissioncontrolschemeusingsimulatedannealing,theoptimalsetofthresh-oldsarederivedtoguaranteethemaximumallowableCDPsforbothvoiceanddatatrafc,whileminimizingthemaximalCBPs[15].ChaoandChen[14]developedamultiple-classcallmodelwithusermobility,wherevariousguardchannelschemescanbeusedtosatisfyconnection-levelQoSrequirementsfordifferenttrafcclasses.UsingtheStochasticPetri

PAGE 18

5 NetSPNmodel,Lietal.[62]proposedandanalyzedahybridcutoffpriorityschemesupportingdifferentQoSrequirementsformulticlasstrafcbyadoptingmultiplethresh-olds.SimilartotheGuardChannelscheme,theseschemesarestaticastheboundariesorthresholdsarealwayspredeterminedandlackinginadaptability.However,adynamicnetworkwithtime-varyingtrafcconditionsmayrequiredy-namicbandwidthallocationschemes.NaghshinehandSchwartz[73]developedadis-tributedcalladmissioncontrolDCAscheme,inwhichthecalloccupancyandmobilityparametersintheoriginatingcellandintheadjacentcellsaretakenintoconsiderationforguaranteeingQoSforhandofftrafc.EpsteinandSchwartzimprovedthisschemetoaddresstheQoSofmulticlasstrafc[27].Byusingthepredenedcallblockingprobabil-ityprole,theirschemecannotonlyguaranteemaximumcalldroppingprobabilities,butalsomaintaintherelativeprioritiesamongdifferentclassesoftrafc.However,sincetheprolesarepredened,theirschememaynotadapttodynamictrafcmakeup.Besides,duetoitsunderlyingassumption,theperformanceoftheschemedependsonthedura-tionintheone-steppredictionalgorithm,especiallyintheheterogeneouscase.ComparedwithDCA,astabledynamiccalladmissioncontrolmechanismSDCA[90]isshowntoachievehigherchannelutilizationwhilesatisfyingapredeterminedboundonthecalldrop-pingprobability,sincethisschemeconsiderstheinuencesoflimitedchannelcapacityandtimedependenceaswellastheinuencesfromthenon-neighboringcells.Toavoidthefrequentinformationexchangesamongcells,Lietal.[63]proposedanon-linelocalestimationalgorithm,inwhichanexponentialsmoothingtechniqueisusedtoestimateparametersineachcontrolperiod.Asaresult,thisschemecanobtainalmostthesameperformanceasthatofSDCAwithoutincurringsignalingoverhead.However,thesetwoschemescannotbedirectlyappliedtomultimediawirelessnetworkswheremulticlasstraf-cexists.Ramanathanetal.proposedadynamicresourceallocationscheme[77].Byprobabilisticallyestimatingthepotentialnumberofhandoffconnectionsfromneighboring

PAGE 19

6 cellsandreservingtheexpectedmaximumnetbandwidthneededtodealwiththesehand-offconnections,thedroppingprobabilityofhandoffconnectionsisreduced.Thedrawbackofthisschemeisthatthepriorityamongdifferenttrafcclasses,eitherinhandofftrafcornewtrafc,isnotaddressed.Oliveretal.proposedanadaptivebandwidthreservationscheme[71].Here,real-timeandnon-real-timetrafcaredistinguishedbyexclusivelyre-servingsomebandwidthintheneighboringcellsforreal-timetrafc.However,thefairnessissueisnotconsidered.HouandFangusedtheconceptofinuencecurvetodeterminetheextentoftheinuenceonneighboringcellsbyeachmobileuserandreservedtheband-widthinproportiontoit[43].AresourcereservationalgorithmbasedontheshadowclusterconceptwasdevelopedbyLevineetal.[60],wherefutureresourcerequirementsareesti-matedbasedonthedetailedknowledgeofusers'movingpattern.Therefore,likeHouandFang[43],itsperformancedependsheavilyontheaccuracyoftheknowledgeofmobileusers'routesinthesystem.1.2.2SupportingDifferentiatedQoSinMANETsVariousschemeshavebeenproposedtoaddressQoSatbothroutinglayersandMAClayerforMANETs.OnMACprotocoldesign,muchworkhasbeendonetosupportQoS.ImprovementsbasedonrandomaccessMACprotocolsuchastheIEEE802.11werepro-posedbyseveralauthors[1,2,51,69,88].Adistributedapproachsupportingservicedifferentiationwasproposedforwirelesspacketnetworks[88].AdaandCastelluccia[1]proposedtoscalethecontentionwindow,usedifferentinterframespaceormaximumframelengthinordertosupportdifferentiatedservicesinwirelessLANs.BasedontheworkbyVeresetal.[88],Ahnetal.[2]proposedastatelessnetworkmodelwithdistributedcontrolalgorithmtodeliverdifferentiatedserviceinmobileadhocnetworks,whereservicesareregulatedtodealwithmobilityortrafcoverloading.Kanodiaetal.[51]combinedtwomechanisms,distributedpriorityschedulingandmulti-hopcoordination,toprovideQoSinrandomaccessmulti-hopwirelessnetworks.Sincemostcurrentcontention-basedMACschemescannotprovideuserswithguaranteedQoS,manyresearchershavelookedinto

PAGE 20

7 theTimeDivisionMultipleAccessTDMAMACandproposedvarioustechniquestoad-dressissuessuchasframelengthandslotallocationstrategiesinadhocnetworks[8,21,23].Chlamtacetal.[21]proposedatechniquecalledprotocolthreadingtoimprovetheperfor-manceoftime-spreadmultiple-accessTSMAprotocols.OptimizationofTimeDivisionMultipleAccessTDMAframelengthandslotallocationstrategieswerestudiedinadhocnetworks[8,23].NesargiandPrakashfocusedondynamicallyallocatingchannelstodifferentcellsincellularmobilenetworkssuchthatthereisnotransmissionconict[74].ConsiderableefforthasalsobeenspentonQoSroutingwiththeprimarygoaltondapathfromasourcetoadestinationthatsatisesthedesiredQoSrequirement.Lin[67][68]proposedaschemetocalculatetheend-to-endbandwidthofapathunderaCDMA-over-TDMAmechanism.ByrelaxingtheCDMA-over-TDMAMACscheme,Liaoetal.proposedaTDMA-basedbandwidthreservationschemeforQoSroutinginadhocnetworks[65].ZhuandCorson[102]proposedanefcientalgorithmcalledforwardalgorithmforcalculatingtheend-to-endbandwidthonapathandreservingrequiredband-widthsothataQoSroutecouldbeestablished.InsteadofresortingtoonesinglepathtofullltheneedsofQoS,Liaoetal.[66]proposedamulti-pathQoSroutingprotocol,inwhichmultiplepathsaresearchedtosupportQoStogether.ChenandNahrstedt[17]designedadistributedticket-basedQoSroutingschemethatcandealwithimprecisestateinformation.However,theunderlyingMACschemeisnotspeciedandprioritizationisnotsupported.Yeetal.[93]proposedaframeworkofreliableroutingbyplacingsomereliablenodesinsomespecialplacesinthenetwork.Locationinformationhasbeenleveragedtoimprovetheperformanceofroutingproto-col[4,5,18,53,58]andreferencestherein.Theoverheadassociatedwithroutediscoveryisgreatlyreducedsincethesearchforanewroutecanbelimitedtoacertainarea[4,58].Basagnietal.[5]usedlocationinformationtoestimatethenodemobilityandsetappropri-ateroutingupdatefrequency,leadingtothedecreaseofupdateoverhead.Further,locationinformationcanalsobeusedtoconstructlargeanddensewirelessadhocnetworks[53].

PAGE 21

8 1.2.3SupportingQoSinWLANsTodatetherehavebeentwothreadsofpriorresearchontheIEEE802.11basedWLANs,namelyperformancemodelingandanalysis,andQoSprovisioning.Considerableeffortwasdevotedtotheoreticalanalysisoftheperformanceofthe802.11DistributedCoordinationFunctionDCF[7,12,20,32,40,59,89,91,99].Bianchi[7]proposedaMarkovchainmodelforthebinaryexponentialbackoffprocedure.Byassum-ingthecollisionprobabilityofeachnode'stransmissionisconstantandindependentofthenumberofretransmissions,hederivedthesaturatedthroughputfortheIEEE802.11DCF.BasedonthesaturatedthroughputderivedinBianchi'smodel,FohandZuckerman[32]usedaMarkovianstatedependentsingleserverqueuetoanalyzethethroughputandmeanpacketdelay.Calietal.[12]studiedthe802.11protocolcapacitybyusingap-persistentbackoffstrategytoapproximatetheoriginalbackoffintheprotocol.Theyalsoshowedhowtotunepinordertoobtainthemaximumsaturatedthroughput.Inadditiontocollisions,Hadzi-VelkovandSpasenovskitooktheeffectofframeerrorrateintoaccountintheiranal-ysisofsaturatedthroughputanddelay[37].Zhaietal.derivedanapproximateprobabilitydistributionoftheservicetime,andbasedonthedistribution,analyzedthethroughputandaverageservicetime[99].Zhai,ChenandFangalsocarefullystudiedtheperformanceof802.11DCFintermsofthroughput,delayanddelayjitter,andpacketlossrate.TheyfounditcansupportstrictQoSforreal-timetrafcaslongasthenetworkoperatesintheunsaturatedcase.Xiao[91]andKongetal.[59]studiedtheperformanceofthe802.11einthesaturatedcase.Thesecondthreadoftheresearchonthe802.11DCFwasfocusedonprovidingservicedifferentiation[1,51,81,83,88,92].Toprovideservicedifferentiation,AdaandCastelluccia[1]proposedtoscalethecontentionwindow,usedifferentinterframespacingormaximumframelengthforservicesofdifferentpriority.Thisideahasbeenincorpo-ratedintheemerging802.11estandard[45].DespiteprovidingprioritizedQoS,theEDCAstillcannotsupportstrictQoSforreal-timeapplicationslikevoiceandvideoasshown

PAGE 22

9 in[22,92].Veresetal.[88]proposedtwomechanisms,virtualMACandvirtualsource,toenableeachnodetoprovidedifferentiatedservicesforvoice,video,anddata.Bymodify-ingthe802.11MAC,adistributedpriorityschedulingschemewasdesignedtoapproximateanidealizedschedule,whichsupportsprioritizedservices[51].Bysplittingthetransmis-sionperiodintotwoparts,oneforreal-timetrafcandtheotherfornon-real-timetrafc,real-timetrafcissupportedwithQoSguarantee[81].However,theDCFmodewasdra-maticallychanged.TheBlackbust[83]providedhighpriorityforreal-timetrafc.Unfor-tunately,itimposesspecialrequirementsonhighprioritytrafcandisnotfullycompatiblewiththeexisting802.11standard.Motivatedbythediscovery[98],Zhai,ChenandFangproposedtousecalladmissioncontrolandratecontroltoregulatethenetworktrafcandhenceenablethenetworktosupportQoS[98].1.3MainContributionsInthisdissertation,wemakethefollowingcontributionstowardsprovidingQoSinwirelessnetworks: Wedesignedtwodynamicreservation-basedCACschemesforwirelesscellularnet-works.Intherstscheme,multiplebandwidthreservationthresholdsaredynami-callyupdatedaccordingtonetworktrafcsituations.Unlikeperviousapproaches,weconsidernotonlythemeanofthebandwidthneededtobereservedforhandofftrafc,butalsothevariationinthethresholdupdating.Inthisway,amoreaccu-rateestimatecanbeobtainedandhencebetterQoSrequirementssatised.Thesec-ondschemeguaranteestherequiredhandoffdroppingprobabilityfordifferenttrafcclassesandtherelativeprioritybetweendifferenttrafcclassesintermsofthenewconnectionblockingprobability.SinceallthethresholdsaredynamicallyadjustedaccordingtocurrenttrafcconditionsandQoSstatus,thisschemecanworkunderdynamicandasymmetrictrafcdistributionsandefcientlyutilizeresources.Wefurthergeneralizedtheconceptofrelativeprioritythatallowsnetworkoperatorstoadjustadmissioncontrolpoliciesexibly. Wedevelopedadistributedlocation-awareapproachtoprovidingserviceprioritiza-tioninabattleeldadhocnetworkoracluster-basedhierarchicalsensornetworkwithTDMAMAC.ItutilizesjointroutingandMACdesign.Inthisscheme,higherpriorityisgrantedtosomeimportantconnectionsthroughlocation-awarebandwidthpre-reservation.Then,aroutingschemethatisawareofMACcollisionsisusedtondandreservesufcientend-to-endbandwidthinordertosupportQoSandkeep

PAGE 23

10 theresidualresourcemoreefcientforfuturecommunications.Asaresult,thenet-workbandwidthisefcientlyusedduetothereductioninMACcollisionswhileservicedifferentiationissupported. FortheemergingIEEE802.11eWLANstandardthattargetedserviceprioritization,weanalyticallyderivedthedelayperformanceforthetrafcofdifferentpriorities.Then,weproposedacalladmissionandratecontrolframeworktoutilizetheanalyti-calresultsaswellasthechannelbusynessratio.Inthisway,the802.11eisenhancedtosupportstringentQoSinsteadofprioritizedservices,andatthesametime,achievehighchannelutilization.1.4OrganizationofThisDissertationInChapter 2 ,werstintroducethesystemmodelusedforcellularnetworks.Then,wedescribeadynamicreservation-basedCACschemewhichadoptsmultiple-thresholdreservationarchitecture.Themethodusedtodynamicallyupdatethebandwidthreservationthresholdsispresentedindetail.Attheendofthechapter,theperformanceoftheschemeiscomparedwiththatofageneralizeddual-thresholdreservationscheme.Adynamicmultiple-thresholdbandwidthreservationDMTBRschemeformobilemultimedianetworksisthefocusofChapter 3 .Inthischapter,westartwithintroducingtwomajorQoScriteria,inwhichfairnessisconsidered.Then,wepresenttheoverallarchi-tecture,followedbythedetaileddescriptionoftheadaptationofthebandwidthreservationthresholds.ExtensivesimulationsverifytheperformanceofDMTBR.Chapter 4 presentsalocation-awareresourcemanagementschemeformobileadhocnetworksthatsupportservicedifferentiation.Werstpresentthebandwidthpre-reservationscheme,followedbythedescriptionoflocation-awareforwarding.Then,weevaluatetheperformanceofthisschemeanddiscusssomedesignissues.InChapter 5 ,westartwithanalyzingthedelayperformancefordifferenttypesoftrafcina802.11eWLANthatoperatesintheunsaturatedcase.Basedontheanalysis,weproposetwocalladmissioncontrolschemesforacceptingreal-timetrafcintothenetworkandaratecontrolschemefordynamicallyadjustingthedataratefornon-real-timetrafc.Finally,weevaluatetheirperformancethroughputextensivesimulations.

PAGE 24

11 Atlast,wesummarizethecontributionsofthisdissertationanddiscusssometopicsthatweplantoexploreinthenearfutureinChapter 6 .

PAGE 25

CHAPTER2DYNAMICRESERVATION-BASEDCALLADMISSIONCONTROLINCELLULARNETWORKS2.1IntroductionToefcientlyutilizethenetworkresourcesandsatisfytheQoSrequirementsofeachacceptedconnection,aBStypicallyenforcesconnectionadmissioncontrolCACinthepresenceofnewconnectionrequests.WhendesigningaCACscheme,weshouldkeepseveralthingsinmind.First,weneedtogivehandoffconnectionrequestshigherprioritythannewconnectionrequests.Asweknow,ahandoffrequestoccurswhenauserengag-ingacallconnectionmovesfromonecelltoanother.TokeeptheQoScontractagreedduringtheconnectionsetupstage,thenetworkshouldprovideuninterruptedservicetothepreviouslyestablishedconnection.However,ifthedestinationcelldoesnothaveenoughresources,theongoingconnectionisforcedtoterminatebeforenormalcompletion.Sincemobileusersaremoresensitivetotheterminationofanongoingconnectionthantheblock-ingofanewcallconnection,handoffcallconnectionsareusuallygivenhigherpriorityoverthenewcallconnection.Second,sincethevariousservicesofferedbythenetworkhaveinherentlydifferenttrafccharacteristics,theirQoSrequirementsmaydifferintermsofbandwidth,delayandconnectiondroppingprobabilities.Itisthenetworks'responsibilitytoassigndifferentprioritiestotheseservicesinaccordancewiththeirQoSdemandsandtrafccharacteristicsaswell.Finally,whentherearemultipletypesofservicescoexistinginthenetwork,itiscriticalthatthenetworkcanprovidefairnessamongthoseserviceswhilesatisfyingtheirspecicQoSrequirement.Thus,thenetworkneedstofairlyallocatenetworkresourcesamongdifferentuserssuchthatdifferentiatedQoSrequirementscanbesatisedforeachtypeofserviceindependentoftheothers. 12

PAGE 26

13 Inthischapter,weproposeadynamicCACschemewhichemploysmultiple-thresholdreservationarchitectureinmultimediamobilewirelesssystems.Thisschemecangrantdifferentialprioritiestobothreal-timeandnon-real-timetrafcclassesandtonewandhandofftrafcbyusingthreebandwidthreservationthresholds,G1,G2,andG3,whicharedynamicallyadjustedaccordingtoinstantaneousnetworktrafcsituation.Moreover,unliketheschemeproposedbyRamanathanetal.[77],strictQoSrequirementscanbesatisedsincewetakeintoaccountthevarianceoftheestimatedbandwidthneededtobereservedwhenupdatingthebandwidththresholds.Notethattoperiodicallyupdatethesethresholdsrequiresabasestationtocommunicatewithitsneighborstoacquirethenecessarytrafcinformation.Consideringtheamountofinformationexchanged,however,thiswillnotposeaseriousproblemonbasestations,whichareinterconnectedwithhigh-speedcommunicationlinks.Therestofthischapterisorganizedasfollows.Section 2.2 describesthesystemmodelwewilluseforourstudy.Ourschemeispresentedindetailinsection 2.3 .Section 2.4 presentssimulationresultswithdiscussion.TheconcludingremarksaregiveninSection 5.8 .2.2SystemModelThesystemunderconsiderationisawirelessmultimedianetworkwithacellularin-frastructure,comprisingofanumberofcells.Eachcellisservedbyabasestation,andbasestationsareinterconnectedusinghigh-speedcommunicationlinks.WeassumethesystemusesxedchannelassignmentFCA[54],whichmeansthateachcellhasaxedamountofcapacity.Notenomatterwhichmultiple-accesstechnologyFDMA,TDMAorCDMA[79]isused,wecouldinterpretsystemcapacityintermsofbandwidth.Werec-ognizethatconnectionswithdifferenttrafcclassmaydifferintheirtrafccharacteristics,suchasconstantbitrate,variablebitrateandpeakbitrate,andthedesiredQoSguaranteesintermsofdelaybound,lossrateorthroughput.Forinstance,thebandwidthofdataser-vicesuchaswebbrowsingmayvarywithtime.Inthischapter,weassumeasinglenumber,

PAGE 27

14 “effectivebandwidth”[26],isadequateforguaranteeingthedesiredQoSforanyconnec-tionwithcertaintrafccharacteristics.Hereafter,wheneverwerefertothebandwidthofaconnection,wemeanitseffectivebandwidth.WeassumeeachcellhasCbandwidthunitsBU.Therearetwoclassesofincomingtrafc:ClassI–real-timetrafcandClassII–non-real-timetrafc.Typically,classItrafcincludesvoiceandvideoservicewhereasclassIItrafcconsistsofdataservicessuchasemail,letransferandwebbrows-ing.WeassumethatthearrivalsarePoissonprocesses,withrespectivearrivalratesrtandnrt.TheconnectiondurationtimesCDTsofbothconnectionsfollowexponentialdistributions,withmeans1/rtand1/nrt,respectively.WedenotebyDrttandDnrttdrttanddnrttthecumulativedistributionfunctionstheprobabilitydensityfunctionsoftheCDTsforthesetwokindsofconnections.Furthermore,weassumethatthecellres-idencetimeCRTdistributionsofthesetwokindsofconnectionsarealsoexponentiallydistributed,withmeans1/rtand1/nrt,respectively.WedenotebyRrttandRnrttrrttandrnrttthecumulativedistributionfunctionstheprobabilitydensityfunctionsoftheCRTsforthesetwokindsofconnections.ThenumberofBUsrequiredbyeachreal-timeandnon-real-timeconnectionisBWrtandBWnrt,respectively.Theseassumptionsareappropriateandcommonlyusedintheliterature. 1 Followingtheassumptions,wecaneasilyconcludethat,forthesetwotypesoftrafc,thechannelholdingtimeCHT,whichisdenedastheminimumofconnectiondurationtimeandcellresidenttime,isalsoexponentiallydistributed,withthemeanofdrt=1=rt+rtanddnrt=1=nrt+nrt,respectively[28,29]. 1Theseassumptionsareusedforpresentationpurposes.Infact,itcanbeeasilyseenthattheanalysisinthefollowingalsoholdsforgenerallydistributedCDTandCRT.

PAGE 28

15 //forincominghandoffconnectionsifareal-timeconnectionifavailableBUs>BWrtaccept;elsereject;else//anon-real-timeconnectionifavailableBUs>G2+BWnrtaccept;elsereject;//forincomingnewconnectionsifareal-timeconnectionifavailableBUs>G1+BWrtaccept;elsereject;else//anon-real-timeconnectionifavailableBUs>G3+BWnrtaccept;elsereject; Figure2-1:Connectionadmissionpolicy 2.3ProposedScheme2.3.1ConnectionAdmissionPolicyBasedonthethresholdsG1,G2,andG3,whicharecalculatedasdescribedinthenextsubsection,theadmissionpolicyisshowninFig. 2-1 .Ahandoffreal-timeconnectionisalwaysadmittedaslongasthereareBWrtBUsavailableandrejectedotherwise.Foranewreal-timeconnection,itwillbeadmittedifthereareBWrt+G1BUsavailableandrejectedotherwise.Similarly,ahandoffornewnon-real-timeconnectionwillbeadmittedonlyifthereareBWnrt+G2orBWnrt+G3BUsavailable.2.3.2AdaptationofG1,G2,andG3TheschemerequiresadjustingthevaluesofthesethreethresholdsG1,G2,andG3whenmakingadmissiondecisions.WeassumeG1isthenumberofBUsthatneedstobe

PAGE 29

16 reservedtodealwithhandoffreal-timeconnectionsthatwillarriveinaperiodt;t+drt],wheretdenotesthetimeinstantofthearrivalofanewreal-timeconnectionrequestinatypicalcelli,anddrtdenotestheexpectationoftheCHTforreal-timetrafc.Inthetimeintervalt;t+drt],someoftheexistingreal-timeconnectionsmayleavecellieitherduetocompletionorduetoahandofftotheadjacentcells.Wedeneanoutgoingeventtobesuchadeparture,whichwillresultinareleaseofBWrtBUs.Meanwhile,therearesomehandoffreal-timeconnectionscomingintocellifromitsneighboringcells.Wedeneanincomingeventtobeahandoffintocelli,whichwillconsumeBWrtchannels.Wewanttoobtainthenetbandwidthchangesofcelliint;t+drt].Letmdenotethenumberofreal-timeconnectionsincellithatwillleavecelli,andndenotethenumberofhandoffreal-timeconnectionsthatwillentercelliint;t+drt].Therefore,atotalofm+neventswilloccurincelliduringt;t+drt].Letsbeasequenceofthesem+neventsandSm;nbethesetofallpossiblesequencesthatmaytakeplaceint;t+drt];thenwecanobtainthecardinalityofSm;n:jSm;nj=m+n! m!n!.1IfwedeneXksasthenetchangeinthenumberofBUsallocatedtoreal-timeconnectionsfromtimettotheendofktheventins,itisobviousthatXks=8><>:Xk)]TJ/F24 7.97 Tf 6.587 0 Td[(1s)]TJ/F25 11.955 Tf 11.955 0 Td[(BWrt;ifktheventisoutgoingXk)]TJ/F24 7.97 Tf 6.586 0 Td[(1s+BWrt;ifktheventisincoming.2HereweassumeX0s=0and16k6m+n.LetYs=maxfXks:06k6m+ng,correspondingtothemaximumnetchangeinthenumberofBUsallocatedtoreal-timeconnectionsincelliint;t+drt].YscouldalsobethoughtofasthemaximumnumberofBUsthatneedstobereservedtodealwithhandoffreal-timeconnectionsthatarriveint;t+drt].Becausealltheseincomingandoutgoingeventsareindependent,eachpossiblesequencesisofequalprobability,1/jSm;nj,tooccurandmayhavedifferent

PAGE 30

17 Ys,weobtaintheexpectedvalueofYs: Ys=Xs2Sm;nYs jSm;nj.3wherejSm;njisthecardinalityofSm;n.Accordingtothetrafcmodeldescribedabove,similarto[77],wecanobtaintheparametersinEquation 2-3 .Thedetailsareshowninthefollowing.EstimationofdrtBydenition,theCHTistheminimumofconnectiondurationtimeandcellresidenttime,bothofwhichareexponentiallydistributed;thus,astheexpectationoftheCHTforreal-timetrafc,drtisequalto1/rt+rt.EstimationofmandnGivenareal-timeconnectionthatstartedattimep,enteredcelliattimeq,andisactiveattimet,theprobabilitythatthisconnectionwillincurahandoffintheintervalt;t+drt]isasfollows:P1p;q;t;drt=Rt+drttP[CRT=w)]TJ/F26 7.97 Tf 6.587 0 Td[(q] P[CRT>t)]TJ/F26 7.97 Tf 6.587 0 Td[(q]P[CDT>w)]TJ/F26 7.97 Tf 6.587 0 Td[(p] P[CDT>t)]TJ/F26 7.97 Tf 6.586 0 Td[(p]dw=Rt+drttrrtw)]TJ/F26 7.97 Tf 6.587 0 Td[(q 1)]TJ/F26 7.97 Tf 6.586 0 Td[(Rrtt)]TJ/F26 7.97 Tf 6.587 0 Td[(q1)]TJ/F26 7.97 Tf 6.586 0 Td[(Drtw)]TJ/F26 7.97 Tf 6.586 0 Td[(p 1)]TJ/F26 7.97 Tf 6.587 0 Td[(Drtt)]TJ/F26 7.97 Tf 6.586 0 Td[(pdw.4Likewise,theprobabilitythattheconnectionwillneitherincurahandoffnorcompleteintheintervalt;t+drt]isasfollows:P2p;q;t;drt=P[CRT>t+drt)]TJ/F26 7.97 Tf 6.586 0 Td[(q] P[CRT>t)]TJ/F26 7.97 Tf 6.586 0 Td[(q]P[CDT>t+drt)]TJ/F26 7.97 Tf 6.586 0 Td[(p] P[CDT>t)]TJ/F26 7.97 Tf 6.586 0 Td[(p]=1)]TJ/F26 7.97 Tf 6.586 0 Td[(Rrtt+drt)]TJ/F26 7.97 Tf 6.586 0 Td[(q 1)]TJ/F26 7.97 Tf 6.586 0 Td[(Rrtt)]TJ/F26 7.97 Tf 6.586 0 Td[(q1)]TJ/F26 7.97 Tf 6.586 0 Td[(Drtt+drt)]TJ/F26 7.97 Tf 6.587 0 Td[(p 1)]TJ/F26 7.97 Tf 6.586 0 Td[(Drtt)]TJ/F26 7.97 Tf 6.587 0 Td[(p.5Obviously,theprobabilitythattheconnectionwilleitherincurahandofforcompleteintheintervalt;t+drt]is1)]TJ/F25 11.955 Tf 11.956 0 Td[(P2p;q;t;drt.LetGirtdenotethesetofallreal-timeconnectionsincelliattimet.Foreachconnec-tionk2Girt,letpkdenotethetimeatwhichconnectionkstartedandqkdenotethetime

PAGE 31

18 connectionkenteredcelli.Then,mcanbeestimatedasm=6664Xk2Girt)]TJ/F25 11.955 Tf 11.955 0 Td[(P2pk;qk;t;drt7775.6Toestimaten,theBSincellineedsinformationfromitsneighboringcells.LetNidenotethesetofneighboringcellsofcelli.Then,ncanbeestimatedasn=&Xj2Ninjrt'.7wherenjrtdenotesthevaluereturnedfromcellj2Ni.Incellj,whenaconnectionkwillincurahandoff,itwillhandofftocelliwithprobabilityhj;ik.Thennjrtiscalculatedasnjrt=Xk2GjrtP1pk;qk;t;drthj;ik.8InthecasethattheCDTandCRTforreal-timetrafcbothareexponentiallydis-tributedwithmean1/rtand1/rt,weobtainP1p;q;t;drt=rt rt+rt)]TJ/F25 11.955 Tf 11.955 0 Td[(e)]TJ/F24 7.97 Tf 6.586 0 Td[(urt+rtdrtP2p;q;t;drt=e)]TJ/F24 7.97 Tf 6.587 0 Td[(urt+rtdrt.9However,asweknow,forarandomvariable,expectationisinadequatetofullychar-acterizethevariable.ToobtainabetterestimateofG1andprovidebetterQoSguarantees,wetakethevarianceofYsintoconsideration.Sincewedonotknowthespecicprob-abilitydistributionofthemaximumnetbandwidthneededtobereserved,thefollowingtechniqueisadoptedtoestimatethestandarddeviationofYs.SDEVrt=SDEVrt+)]TJ/F25 11.955 Tf 11.955 0 Td[(Rrt)]TJET1 0 0 1 420.611 198.455 cmq[]0 d0 J0.478 w0 0.239 m24.012 0.239 lSQ1 0 0 1 -420.611 -198.455 cmBT/F25 11.955 Tf 420.611 188.054 Td[(Ys.10whereRrtistheactualmaximumnetbandwidthneededtobereservedforhandlingreal-timehandofftrafc,whichwecanacquirebyrealtimemeasurement;and<<1isadesignparameter.Notethatasimilartechnique[48]hasbeenusedtosetTCPretransmit

PAGE 32

19 timer.Onceweobtainthestandarddeviation,wecansetthebandwidththresholdG1to:G1= Ys+KrtSDEVrt.11NotethatparameterKrt,whichisascalar,offersnetworkoperatorstheexibilitytofullldiverseQoSrequirements.Bydynamicallyadjustingitsvalueifnecessary,itisabletomeetdifferentQoSboundforhandofftrafc.SinceG2isthebandwidththresholdforbothnewandhandoffreal-timeconnections,itshouldincludesomebandwidthsreservedfornewreal-timeconnectionsinadditiontothoseforhandoffreal-timeconnections,whichhasalreadybeenreectedinG1.Lettbethetimeinstantofahandoffnon-real-timeconnectionrequestarrivalincelli.Letdnrtbetheexpectationofchannelholdingtimeofthisconnectionincelli. 2 Inthetimeperiodt;t+dnrt],giventhenewreal-timeconnectionarrivalratert,thereareapproximatelyrtdnrtarrivalsofnewreal-timeconnectionrequests.Ifnon-real-timeconnectiondurationislong,thisnumbermayexceedthetotalamountofbandwidthsavailableincelli.Thusweintroduceaparameter<<1toadaptthisnumberG2.WeestimateG2asG2=G1+rtdnrtBWrt.12Notethatalthoughreal-timetrafcalwayshashigherpriorityovernon-real-timetraf-c,thechoiceofprovidesustheabilitytodeterminehowhighthepriorityis.Smallergivesmoderatelyhigherprioritytoreal-timetrafc,causingtheincreaseofCBPforreal-timetrafcandthedecreaseofCDPfornon-real-timetrafc,whilelargerhasanoppositeeffect.BeforecalculatingG3,werstestimateG0.ItisthenumberofBUsthatneedstobereservedtodealwithhandoffnon-real-timeconnectionsthatwillarriveinaperiodofdnrt 2Accordingtooursystemmodel,theexpectationsofCHTforanewconnectionandahandoffconnectionareequal.

PAGE 33

20 fromnowtothefuture.Similarly,letm0denotethenumberofnon-real-timeconnectionsincellileavingcelliduetocompletionorhandoff,andletn0denotethenumberofhandoffnon-real-timeconnectionsenteringcelliinperioddnrt.SoweobtaintheexpectationY0s,themaximumbandwidthneededtobereservedforhandoffnon-real-timeconnections: Y0s=Xs02S0m0;n0Y0s jS0m0;n0j.13wheres0andS0m0;n0arethecounterpartsofsandSm;n.Thus,Y0scanbeobtainedusingthetechniquesintroducedearlier.UsingthesametechniqueshowninEquation 2-10 ,wecanobtainthestandarddevia-tionofthemaximumbandwidthneededtobereservedfornon-real-timetrafc,SDEVnrt.Therefore,wesetG0tothefollowingvalue:G01= Y0s+KnrtSDEVnrt.14whereKnrt,similartoKrt,isthedesignparametercorrespondingtonon-real-timetrafc.G3shouldbethesumofG2andG0,thusweobtaintheestimateofG3:G3=G2+G0.152.3.3UpdatingFrequencyItwouldbeidealthateachtimeaneworhandoffconnectionrequestarrivesincelli,allthethreebandwidththresholdsareupdatedsothatanadmissiondecisioncouldbemadefortherequest.However,basestationmaychoosetoupdatethethresholdseverytimeithasreceivedNconnectionrequests,consideringthateachupdatemayincursomecommunicationandcalculationoverheads.Ncouldbechosentoprovidethetrade-offbetweensystemperformanceandoverheads,whichisinvestigatedinsection 2.4 .

PAGE 34

21 Figure2-2:Wrap-aroundsimulationmodel 2.4PerformanceEvaluationInthissection,wecomparetheperformanceofourproposedschemewiththatofthegeneralizedDTR,whichistermedGDTRhereafter.Tofurtherillustratetheadvan-tageofusingvarianceinformation,wealsopresenttheperformanceofavariationofourschemewhichisidenticaltoourschemeexceptthatvarianceisnotconsidered.Here-after,wecallourproposedschemePSandthevariationwhichdoesnotconsidervariancePS NO VAR.OursimulationiscarriedoutusingOPNETModeler8.0[72].Oursim-ulationmodelisawrap-aroundmodel[96]showninFig. 2-2 .Eachcell,representedbyahexagon,hassixneighborssothathandoffdeparturefromtheedgecellswillnotbeignored.ThetotalnumberofBUsineachcellis40.ThenumberofBUseachreal-timeornon-real-timeconnectionneedsisBWrt=1orBWnrt=4.Thereal-timeconnectionmaybe

PAGE 35

22 voicecallsandthenon-real-timeconnectionmayrepresentletransferorwebbrowsing.Anewconnectionisrandomlydeterminedasareal-timeconnectionwithprobability0.8andasanon-real-timeconnectionwithprobability0.2,respectively.Forreal-timetrafc,themeanduration1/rt=300secondsandthemeancellresidencetime1/rt=150seconds.Fornon-real-timetrafc,1/nrt=1500secondsand1/nrt=750seconds.Soeachcon-nection,onaverage,willhandoffatleastonceduringitslifetime.Ahandoffrequestwillrandomlychooseadestinationfromthesixneighboringcells.Krtisequalto1andKnrtequalto3.Wechoose=0:05unlessotherwisenoted.ForGDTRscheme,K1isusedtoreservebandwidthsforreal-timehandoffconnections,whileK2isusedtoblocknon-real-timetrafcintothenetworkinordertoguaranteethereal-timetrafcperformance.WechooseK1=36andK2=30.Fig. 2-3 through 2-7 comparetheperformancesforthesethreeschemes,asafunctionofaveragenewconnectionarrivalrate.AsshowninFig. 2-3 and 2-4 ,forreal-timetrafc,theCBPandCDPforPS NO VARandPSaresignicantlyreducedcomparedtothoseforGDTR.Duetothefactthatittakesvarianceintoconsiderationwhenestimatingthebandwidththresholds,PScausessmallerCDPthanPS NO VARdoes.Fig. 2-5 and 2-6 presenttheperformancefornon-real-timetrafc.Again,bothPS NO VARandPSoutperformsGDTR.PSisalsobetterthanPS NO VARintermsofCDP.However,unlikethecaseforreal-timetrafc,thisisachievedatthecostthattheCBPforPSisabithigherthanthatforPS NO VAR.Thereasonisthat,G2increasesasthetrafcloadincreaseswhileG3doesnot.Wealsocomparethesystemthroughputforthesethreeschemes.Thesystemthrough-putisdenedasfollows:TP=PiBWiCHTofconnectioni CCELL NUMST.16whereC,asmentionedearlier,isthetotalnumberofBUsavailableineachcell,CELL NUMisthetotalnumberofcellsintheentirenetworkandSTisthetotalsimulationtime.Aswe

PAGE 36

23 Figure2-3:Connectionblockingprobabilityforreal-timetrafc Figure2-4:Handoffconnectiondroppingprobabilityforreal-timetrafc

PAGE 37

24 Figure2-5:Connectionblockingprobabilityfornon-real-timetrafc Figure2-6:Handoffconnectiondroppingprobabilityfornon-real-timetrafc

PAGE 38

25 Figure2-7:Throughput canseeinFig. 2-7 ,comparedtoPS NO VARorPS,thesystemutilizationforGDTRisverylow,onlyaboutahalfofthatcorrespondingtoPS NO VARorPS.SincemorebandwidthsarereservedforhandofftrafcinschemePS,itssystemthroughputisalittlelowerthanthatforPS NO VAR.Insum,theresultsareexpectedbecauseourproposedschemeiscapableofdynamicallyadjustitsbandwidththresholdsaccordingtotrafcsitua-tions,hencegivingamuchbetterperformancethanGDTRinalltheperformancemetricsdiscussedabove.ItisimportanttonotethattomeetstrictCDPrequirementsforreal-timeornon-real-timetrafc,KrtandKnrtcanbesettodifferentvalues.Next,weinvestigatehowdependentthesystemperformanceisonthevalueofN,theupdatingfrequency.Fig. 2-8 showsthattheperformanceintermsofCBP,CDPandsystemutilizationisnotverysensitivetothevalueofNinsomerangeforourproposedscheme,whichisadesirablefeature.Therefore,thesystemcanworkverywellwithoutincurringtoofrequentupdatesamongbasestations.Finally,Fig. 2-9 presentstheimpactofontheperformance.Generally,asin-creases,CBPandCDPforreal-timetrafcarereducedwhilenon-real-timetrafcblockingprobabilitiesarebecomingabitlarger.Therefore,couldbeusedtoadjusttherelativeprioritiesbetweenreal-timetrafcandnon-real-timetrafc.Consideringthedifferencein

PAGE 39

26 Figure2-8:Performancevs.changingupdatefrequency bandwidthrequirementsforbothtrafcclasses,thesystemthroughputdecreasesasisgettinglarger.2.5ConclusionInthischapter,weproposeaconnectionadmissioncontrolschemebasedondynamicmultiple-thresholdbandwidthreservationtoguaranteeQoSprovisioninginwirelessmul-timedianetworks.Accordingtonetworktrafcsituations,thebandwidththresholdsaredynamicallyupdatedbytakingintoaccountboththeexpectationandvarianceoftheband-widthsneededtobereserved.ThisschemeworkswelltoprovideQoSguaranteeandefcientlyusenetworkresource,asshowninthesimulation.Sincetheproposedschemeinvolvesslightchangestothearchitectureofthecurrentwirelessnetwork,itcanbeeasilyadoptedby3Gandbeyondwirelesssystems.

PAGE 40

27 a b Figure2-9:Performancevs.changing

PAGE 41

CHAPTER3DYNAMICMULTIPLE-THRESHOLDBANDWIDTHRESERVATIONDMTBRINCELLULARNETWORKS3.1IntroductionInthischapter,weproposeadynamic,multiple-thresholdbandwidthreservationDMTBRschemeformobilemultimedianetworks.Theobjectiveoftheproposedschemeistwofold.TheschemerstprovidesQoSprovisioningbykeepingtheCDPbelowthepre-denedboundevenundernetworkcongestionsituation.Second,inafairmanner,itmaintainstherelativeprioritiesamongreal-timetrafcandnon-real-timetrafcintermsoftheCBPac-cordingtotheirtrafcprolesandinstantaneoustrafcsituations.Inthisway,thenetworkwillfairlyallocatebandwidthamongusersofdifferenttrafcclasseswhilemaintainingthehandoffdroppingprobabilityupperbounds.Tomeettheobjectivesstatedabove,threebandwidthreservationthresholds,G1,G2,andG3G1
PAGE 42

29 Itgeneralizestheconceptofrelativeprioritiesandfairnessamongtrafcclasses.Bydoingso,wecanfurtherreducetheCBPsofsomereal-timeserviceswhilenotseriouslydeterioratingtheQoSofnon-real-timeservices. Itusestheinformationinthecurrentcellandintheadjacentcellstoperiodicallyupdatethereservationthresholds,henceisabletorespondtothechangingnetworkconditionsquicklyandeffectively.Therestofthischapterisorganizedasfollows.Ourscheme,includingthetargetQoScriteria,ispresentedinSection 3.2 .Then,thethresholdadaptationintheschemeisaddressedindetailinSection 3.3 .InSection 3.4 ,theschemeisevaluatedthroughextensivesimulationsstudies.Finally,thischapterisconcludedinSection 3.5 .3.2ProposedSchemeTomeetdifferentQoSrequirementsforreal-timeandnon-real-timetrafc,wehavetotakeintoaccounttheirdistincttrafccharacteristics.Real-timeservicesaresensitivetodelay,whereasnon-real-timetrafccouldtoleratesomedelaywithoutdeterioratingser-vicequalityperceivedbythemobileusers.Moreover,handoffconnectionsshouldreceivehigherprioritythannewconnectionsforthesametype.Therefore,itisnecessarythattheyreceivedifferentialtreatmentsintermsofaccesstothenetworkresource.Thisistheratio-nalebehindourscheme,whichcompletelydifferentiatesconnectionrequestsbysettingupthreereservationthresholdsanddynamicallyadaptingtheirvaluestonetworktrafccon-ditions.Next,wedescribethetwoQoScriteriaweconsiderinthischapter,followedbyadetaileddescriptionofourproposedscheme.Itisnoteworthythatthetrafcmodelweuseforourstudyisnotgiveninthischapter,sinceitisthesameasthatpresentedinChapter 2 .3.2.1QoSCriteriaTherstcriterionistosetanupperboundforthehandoffdroppingprobability,theprobabilityofahandoffconnectionthatisbeingdroppedwhenitishandedofffromonecelltoanother.Thecriterionissatisedaslongasthefollowingtwoinequalitiesaresatised:Pd;rt6QoSrtPd;nrt6QoSnrt.1

PAGE 43

30 wherePd;rtQoSrtandPd;nrtQoSnrtareCDPsthemaximumallowabledroppingprob-abilitiesforreal-timeandnon-real-timetrafc,respectively.ThesecondcriterionistomaintaintherelativepriorityamongdifferenttypeoftrafcintermsofCBPs.Whenthereisnosuchcriterion,theblockingprobabilityfornewcon-nectionswillnotbethesame.ThetrafcclassesthatrequiresmallernetworkbandwidthwillhavealowerCBPascomparedtothosethatrequirelargerbandwidth.Obviously,thisisunfairtothetrafcclassesthatrequirelargerbandwidth.Toaddressthisproblem,weassumeintheproleforeachtrafcclass,thereexistsaparametercalledthetrafcpriorityweight,W,indicatingwhichprioritylevelthetrafcclasswillhavewhenthereareseveralclassesoftrafccoexistinginthenetwork.Thisparametercanbesetduringthenegotia-tionbetweentheuserandthenetworkoperatorbeforetheuserusesthenetworkservice,inwhichtrafccharacteristicsaretakenintoaccount.Asmallerweightmeansahigherpriority.ToachievefairnessintermsofCBPsamongalltrafcclasses,thenetworkmayrequiretheCBPstosatisfythefollowingequation:Pb;rt Pb;nrt=Wrt Wnrt.2wherePb;rtWrtandPb;nrtWnrtareCBPsthepredenedtrafcpriorityweightsforreal-timeandnon-real-timetrafc,respectively.Generallyspeaking,therearemanyfactorsthataffecttheCBP.TheactualCBPforeachtrafcclassdependsonthesystemcapacity,theofferedtrafcloadofeachtrafcclass,thepriorityofeachtrafcclass,theadmissionpolicyadoptedtofullltheQoScrite-riarelatedtothehandofftrafc,theactionthenetworkmaytakeintimesofcongestion,andsoon.Whilesomefactorssuchasthesystemcapacityorthepre-assignedtrafcprioritycouldbestatic,theofferedloadandtheactionstakentodealwithnetworkcongestionmaybedynamic.Inthissense,thecriterioninEquation 3-2 isstaticsinceitfailstoreecttherealtimenetworksituation.Therefore,wegeneralizetheconceptofrelativepriorityandproposeamoregeneralwaytomeetthesecondQoScriterion,thatis,tomaintain

PAGE 44

31 therelativepriorityamongdifferenttrafcclasses.Weusethefollowingequationforthispurpose:Pb;rt Pb;nrt=Wrt Wnrt.3ComparingtoEquation 3-2 ,weaddonefactor,,ontheright-handsideofEquation 3-2 .canbethoughtofasafunctionofsomeofthedynamicfactorsdescribedabove,representingthenetwork'srealtrafcconditionsorsomeproceduresrespondingtotrafcchangesorQoSstatus.Sincetheofferedloadisoneofthecommonlyusedmeasuresofnetworktrafcload,onewaytomakeconcreteistoletbeafunctionoftheofferedloadpercellforeachtrafcclass.Theofferedloadcanbedenedastheproductoftrafcarrivalrateforeachtrafcclass,theconnectionholdingtimeandthenormalbandwidth,OLrt=rtBWrt rtOLnrt=nrtBWnrt nrt.4Replacingwiththeratiooftheofferedloadofreal-timetotheofferedloadofnon-real-timetrafc,weobtainthefollowing:Pb;rt Pb;nrt=OLrt OLnrtWrt Wnrt.5Inthisway,wetaketheofferedloadofeachtrafcclassintoconsideration.Inotherwords,wemaintaintherelativeprioritybykeepingtheratiooftheCBPofeachtrafcclassequaltotheproductoftheircorrespondingratiooftrafcloadandthepre-denedweight.TheadvantageofthisapproachwillbediscussedinSection 3.4 .Hereafter,wecalltheschemesatisfying 3-1 and 3-2 DMTBR A,andtheschemesatisfying 3-1 and 3-5 DMTBR G.3.2.2ConnectionAdmissionPolicyIntheDMTBR,threereservationthresholds,namely,G1,G2,andG3,aremaintainedanddynamicallyadjusted.AsshowninFig. 3-1 ,G1isthebandwidthreservedforreal-time

PAGE 45

32 handoffconnectionsonly,andG2isthebandwidthreservedforhandofftrafcincludingreal-timehandoffandnon-real-timehandofftrafc.G3isthethresholdusedforadmittingthetwotypesofnewtrafc,andissetsuchthattherelativepriorityismaintained.Itsrolethusisnotxed,dependingontheinstantaneousstatusoftherelativepriorityforthetrafcclasses.Morespecically,in 3-2 or 3-5 ,iftheCBPforreal-timetrafcistoolargetomeettherequirement,G3shouldbeusedtofavorablyacceptmoreofnewreal-timeconnections;iftheCBPforreal-timetrafcistoosmall,G3shouldbeusedtofavorablyacceptmoreofnewnon-real-timeconnections.Bydenition,G1
PAGE 46

33 Figure3-1:ThebandwidthreservationthresholdsinDMTBR mannertopreventthisfromhappeningwhenexecutingtheadmissioncontrol.Weadoptthisideainthisschemetocopewiththesituationwherethenetworkisundergoingheavytrafcloadintimesofburstiness.Thedetailsareasfollows.Foratimeinterval,eachcellmeasuresCBPsandCDPs,Pb;rtandPd;rtorPb;nrtandPd;nrt.EachtimethecellndsthatthemeasuredCDPforcertaintrafcclass,eitherreal-timeornon-real-timetrafc,isequaltoorlargerthanacertainportionofQoSbounds,itwillincreasethecorrespondingbandwidthreservationthresholdforthetrafcclass.How-ever,incaseofheavynetworkcongestion,evenifthetotalavailableBUsinthecellareallreserved,thepredenedQoSrequirementstillmaynotbesatised.Toavoidthissit-uation,wewillactivelytakesomepreventiveactionsbeforehand.Wecountthenumberoftimesweincreasethereservationthresholdsforhandofftrafc.Onceonereservationthresholdisconsecutivelyincreasedforacertainnumberoftimes,saythreetimes,thecellisdeemedexperiencingheavyhandofftrafc.Inthiscase,toreducethepotentialincom-inghandofftrafcandkeeptheCDPbelowtheupperbound,thecellwillinformallofitsneighborstofurtherthrottletheadmissionsofnewconnectionsofthesametrafcclassasthehandofftrafcclassinthecurrentcell.Similartothepreviousapproaches[78][30],ourproposedschemeadmitsthenewconnectionrequestwithacertainprobabilitygener-atedonline,whichiscalledtheprobabilityofthrottlingnewconnections.DetailsonhowtogeneratetheprobabilityaregiveninSection 3.3 .

PAGE 47

34 iftheincominghandoffconnectionisreal-timeifavailableBUs>BWrtaccept;elsereject;else//theincominghandoffconnectionisnon-real-timeifavailableBUs>G1+BWnrtaccept;elsereject;ifswitch==TRUEiftheincomingnewconnectionisreal-timeifavailableBUs>G2+BWrt&&rand6probrtaccept;elsereject;else//theincomingnewconnectionisnon-real-timeifavailableBUs>G3+BWnrt&&rand6probnrtaccept;elsereject;ifswitch==FALSEiftheincomingnewconnectionisreal-timeifavailableBUs>G3+BWrt&&rand6probrtaccept;elsereject;else//theincomingnewconnectionisnon-real-timeifavailableBUs>G2+BWnrt&&rand6probnrtaccept;elsereject; Figure3-2:Connectionadmissionpolicy

PAGE 48

35 3.3ThresholdAdaptationThekeyideainthisschemeistoaccuratelyadjustthevaluesofthesethreethresholdsG1,G2andG3.First,weintroducethemethodtocalculateG1.Fromwhatisstatedintheadmissionpolicy,weknowG1isthebandwidthreservationthresholdforreal-timehandoffconnectionsonly;hereweadoptthetechniquedescribedinChapter 2 toestimateG1.Aswewillseeclearlylater,thetechniquesweusetoestimateG1orG02,thereservedbandwidthfornon-real-timehandoffconnectionsonlyservestoprovideagoodinitialvaluethatshouldbefurtheradapted.Therefore,ingeneral,anymethodthatcanprovideagoodinitialvaluecanbeusedinourscheme.Inthissense,ourschemeisindependentofanyspecictechniqueweusetoobtaintheinitialvalue.3.3.1AdaptationofG1Lettimetdenotethetimeinstantofthearrivalofanewreal-timeconnectionrequestinatypicalcelli.LetdbetheexpectationoftheCHTofthisnewconnectionincelli.Inthetimeintervalt;t+d],someoftheexistingreal-timeconnectionsmayleavecellieitherduetocompletionorduetoahandofftotheadjacentcells.Wedeneanoutgoingeventtobesuchadeparture,whichwillresultinareleaseofBWrtBUs.Meanwhile,therearesomehandoffreal-timeconnectionscomingintocellifromitsneighboringcells.Wedeneanincomingeventtobeahandoffintocelli,whichwillconsumeBWrtchannels.Letmdenotethenumberofreal-timeconnectionsincellithatwillleavecelli,andletndenotethenumberofhandoffreal-timeconnectionsthatwillentercelliint;t+d].Therefore,atotalofm+neventswilloccurincelliduringt;t+d].Letsbeasequenceofthesem+neventsandSm;nbethesetofallpossiblesequencesthatmaytakeplaceint;t+d];thenwecanobtainthecardinalityofSm;n:jSm;nj=m+n! m!n!.6IfwedeneYsasthemaximumnetchangeinthenumberofBUsallocatedtoreal-timeconnectionsincelliint;t+d].InasimilarwayasdescribedinChapter 2 ,wecan

PAGE 49

36 setG1totheexpectedvalueofYs:G1=Xs2Sm;nYs jSm;nj.73.3.2AdaptationofG2Again,lettimetdenotethetimeinstantofthearrivalofanewnon-real-timecon-nectionrequestincelli.Letd0betheexpectationoftheCHTofanewnon-real-timeconnectionincelli. 2 BeforecalculatingG2,wecalculatetheexpectedvalueofthemaximumnumberofchannels,G02,whichneedstobereservedforhandoffnon-real-timeconnectionsarrivingint;t+d0].Similarly,letm0denotethenumberofnon-real-timeconnectionsincellileavingcelliduetocompletionorhandoff,andletn0denotethenumberofhandoffnon-real-timeconnectionsenteringcelliint;t+d0].Inaverysimilarmanner,wecanobtainG02:G02=Xs02S0m0;n0Y0s jS0m0;n0j.8wheres0,S0m0;n0andY0sarethecounterpartsofs,YsandSm;ndenedearlier.Bydenition,G2shouldbethesumofG1andG02,thusweobtaintheestimateofG2:G2=G1+G02.9UsingthetechniquespresentedinChapter 2 ,giventhetrafcmodelmentionedearlier,wecaneasilyobtaintheparametersinvolved,d,d0,m,m0,n,n0,YsandY0s.AlsonotethatthecalculationorderofG1andG2impliesthathandoffreal-timetrafcisgrantedhigherprioritythanhandoffnon-real-timetrafc. 2Again,theexpectationsoftheCHTsforanewconnectionandahandoffconnectionareequalaccordingtoourassumptions.

PAGE 50

37 3.3.3AdaptationofG3BeforecalculatingG3,wecalculateG03,whichcouldbethoughtofasthebandwidththatisreservedfornewreal-timetrafcagainstnewnon-real-timetrafc,orthebandwidththatisreservedfornewnon-real-timetrafcagainstnewreal-timetrafc,dependingontheinstantaneousrelativeprioritystatusforthetrafcclasses.Forinstance,in 3-2 or 3-5 ,iftherelativeprioritybetweenreal-timetrafcandnon-real-timetrafcisviolatedbecausereal-timetrafcisunfairlyrejectedcomparedwithnon-real-timetrafc,G03shouldbethebandwidthreservedfornewreal-timeconnectionssothatmorereal-timeconnectionscanbeaccepted,andviceversa.TheinitialvalueforG03couldbesetasBWrtorBWnrt.Therefore,G3canbeestimatedasfollows:G03=8><>:BWrt;ifG03isfornewrtconnections:BWnrt;ifG03isfornewnrtconnections:G3=G2+G03.10NoticethatinEquation 3-10 ,weonlysettheinitialvalueofG03.Whentheschemeisrunning,wewilladaptthevalueofG03accordingtothesecondcriterionmentionedearlier.3.3.4FurtherAdaptationofThresholdsItisseenthatweusetheexpectedmaximumnetbandwidthornominalbandwidthtoestimatethereservationthresholds.Wemayexpectdeviationsfromtheminadynam-icallychangingnetworkenvironment,wheretheaccuracyoftheestimationmaydegrade.Therefore,itisinsufcienttodependonlyontheabovethreethresholdstofulllthetaskofQoSprovisioning.TomeettheQoScriteriamentionedinSection 3.2 ,furtheradaptationofthesethresholdsisneeded.ThedetailsareshowninFig. 3-3 .InFig. 3-3 ,up th1,down th1
PAGE 51

38 dothrottlingaswedescribedbefore.powup1;rt indexreferstothert indexthpowerofup1>1,inwhichrt indexisaninteger.Thepositiveornegativevalueofrt indexmeansweactuallyincreaseorreducethevalueoftheinitialvalueofG1,whichisobtainedasdescribedearlier.ItisworthnotingthatwhenthemeasuredCDPexceedsup th1QoSrt,weimmediatelyboostrt indextozeroifitwasnegativeinpreviousstep.Inthisway,thisschemeisalwaysabletoberesponsiveenoughtofullltheQoSboundcriterion.TheportionofhowtoadaptG02isomittedsinceitissimilartothatofadaptingG1.ToguaranteethesecondQoScriterion,G2andG3areusedtomake 3-2 or 3-5 hold.AswecanseeinthepseudocoderegardinghowtoadaptG3,therearethreeparameters,namely,switch,percentage,andadj index.switchisdenedasbefore.Percentagereferstothedeviationerrortheschememaytolerate,ortheerrorwithwhichthesecondcriterionisstillconsideredtobemet.Forinstance,ifpercentageissettobe0.1,thismeans,aslongastheratiooftheright-handsideandtheleft-handsideof 3-2 or 3-5 iswithintherange[0.9,1.1],theequationsholdandtheQoScriterionismet.Theroleofadj indexisverysimilartort index.However,intheadaptationhere,wechangeup3>1,accordingthevalueofadj indexinawaythat,thelargertheabsolutevalueofadj index,thefastertheadaptationspeed.ThisensurestheadaptationofG3canpromptlyrespondtothechangeoftheincomingtrafcand/orQoSstatus.Finally,whenadj indexislessthanathreshold,adj index th,whichmeansG03isnearlyzero,theschemewillreversetheparameterswitch,lettingG3reservedfortheothertrafcclassinsteadofthecurrenttrafcclassitisfor.3.3.5ProbabilityofThrottlingEachcellkeepsaJKnon-negativeintegerarrayAforeachtrafcclassforitsneighbors.JisthenumberoftrafcclassesandKisthenumberofneighboringcells.Sinceweconsiderreal-timeandnon-real-timetrafcinthischapter,Jisequalto2.Ifthecell'sithi=0;1;;K)]TJ/F15 11.955 Tf 12.222 0 Td[(1neighborsendsamessagetothecelltothrottleorde-throttleareal-timetrafcclass;thenA[0][i]isincreasedordecreasedby1.Itissimilarfor

PAGE 52

39 //AssumingtheinitialvaluesofG1,G02,G03areobtained.time1=0,time2=0;rt index=0,nrt index=0;switch=TRUE;ifPd;rt>up th1QoSrtfifrt index<0rt index=0;elsert index++;G1=G1powup1;rt index;time1++;iftime1%time th==0faskingneighboringcellstothrottle;time1=0;ggelseifPd;rtWrt=Wnrt[OLrt=OLnrt]+percentageadj index++;elseifPb;rt=Pd;nrt6Wrt=Wnrt[OLrt=OLnrt])]TJ/F25 11.955 Tf 11.956 0 Td[(percentageadj index--;gelsefifPb;rt=Pd;nrt>Wrt=Wnrt[OLrt=OLnrt]+percentageadj index--;elseifPb;rt=Pd;nrt6Wrt=Wnrt[OLrt=OLnrt])]TJ/F25 11.955 Tf 11.956 0 Td[(percentageadj index++;gG03=G03powup3;adj index;ifadj index
PAGE 53

40 non-real-timetrafc.Whenthereisanincomingnewconnectionrequest,thecellwillusethefollowingequationtogeneratetheprobabilityofthrottlingnewconnections:probrt=bmaxA[0][i]probnrt=bmaxA[1][i].11wherebisarealnumberlessthanandcloseto1,say0.9.Wecanseeprobrtorprobnrtisequalto1ifmaxA[0][i]ormaxA[1][i]iszero.Thismeanswedonotneedtothrottlethenewreal-timeornon-real-timeconnections.3.3.6EstimationofArrivalRatesrtandnrtInordertomakeschemeDMTBR Gworkproperlyaccordingtothecurrentnet-worktrafcsituation,weneedtoproviderealtimeinformationabouttheincomingtrafc.Thus,weneedtoestimatethecurrentarrivalrateofreal-timeconnectionnon-real-timeconnectionrtnrt.Assumethatwemeasurethearrivalrateataxedperiodp,andwedenotethemeasuredarrivalrateofreal-timeconnectionsandnon-real-timeconnectionsatthenthn=1;2;;measurementperiodasMrtnandMnrtn;thenwecanestimatethearrivalrateusingalow-passlterrtn+1=rtn+)]TJ/F25 11.955 Tf 11.955 0 Td[(Mrtnnrtn+1=nrtn+)]TJ/F25 11.955 Tf 11.955 0 Td[(Mnrtn.12whereMrtnandMnrtncanbeobtainedbyMrtn=#ofnewrtarrivalsinnthperiod pMnrtn=#ofnewnrtarrivalsinnthperiod p.13andisaweightingfactor,usually0:5<<1.Wecanseethatmoreweightisgiventothearrivalratesrecentlyobserved.

PAGE 54

41 3.3.7UpdatingFrequencyInresponsetothenetworktrafcconditionaspromptlyaspossible,itwouldbeidealthateachtimeaneworhandoffconnectionrequestarrivesincelli,allthethreereserva-tionthresholdsareupdatedsothatanadmissiondecisioncouldbemadefortherequest.However,basestationmayupdatethethresholdseachtimeithasreceivedNconnectionrequests,consideringthateachupdatemayincursomecommunicationandcomputationoverheads.Ncouldbechosentoprovidethetrade-offbetweensystemperformanceandoverheads.ItsimpactontheperformancewillbeinvestigatedinSection 3.4 .3.4PerformanceEvaluationInthissection,wepresenttheperformanceresultsforourproposedschemebasedonextensivesimulations.OursimulationiscarriedoutusingOPNETModeler.Thesimula-tionmodelisawrap-aroundmodelasshowninFig. 2-2 ,whichcomprises37cells.Eachcell,representedbyahexagon,hassixneighborssothathandoffdeparturefromtheedgecellswillnotbeignored.WeconsiderthesimulationparametersassetinTable 3-1 .ThetotalnumberofBUsineachcellis50.ThenumberofBUseachreal-timeornon-real-timeconnectionwillneedisBWrt=1orBWnrt=4.Thereal-timeconnectionmaybevoicecallsandthenon-real-timeconnectionmayrepresentletransferorwebbrowsing.Forreal-timetrafc,themeanduration1/rt=300secondsandthemeancellresidencetime1/rt=150seconds.Fornon-real-timetrafc,1/nrt=1500secondsand1/nrt=750seconds.Onaverage,eachconnectionwillhandoffonceduringitslifetime.Wheneverthereisahandoffrequest,itwillrandomlychooseadestinationfromthesixneighboringcells.Weassumethat25%oftrafcisreal-timetrafc,and75%ofthetrafcisnon-real-timetrafc.Thisisconsistentwiththefactthatdataserviceswilldominatenetworktrafcinthenearfuture.Newconnections,includingreal-timeandnon-real-timeconnections,arriveaccordingtoaPoissonprocess.Accordingtotheassumption,86.96%ofthenewconnectionarrivalsarereal-timeconnections,andtherestarenon-real-timeconnections.Notethatwhilethe

PAGE 55

42 Table3-1:SimulationparametersforDMTBR AandDMTBR G TotalnumberofBUsineachcell 50 BWrt 1 BWnrt 4 1/rt 300s 1/rt 150s 1/nrt 1500s 1/nrt 750s QoSrt 0.01 QoSnrt 0.05 Wrt=Wnrt 1 percentage 0.1 up th1 0.8 up th2 0.8 down th1 0.5 down th2 0.5 up1 1.1 up2 1.1 up3 1.1 time th 3 adj index th -30 N 20 defaultvalueofup3is1.1,itchangesto1.15or1.2whenjadj index thjislargerthan10or20.Theupdatefrequencyissetto20unlessotherwisespecied.Thesimulationtimeis3hours.Fig. 3-4 and 3-5 showinahomogeneousenvironment,thenewconnectionblock-ingprobabilitiesandhandoffdroppingprobabilitiesforbothtrafcclasses,asafunctionofaveragenewconnectionarrivalrateforbothDMTBR AandDMTBR G.Throughcalculation,weknowthatarrivalrate0.1connection/seccorrespondstoabout110Er-langs,whichare220%ofthefullload.InFig. 3-4 ,asexpected,wecanseethat,forDMTBR A,theblockingprobabilitiesforthetwoclassesarealmostequaltoeachother.ForDMTBR G,sincetheratioofofferedloadofeachtrafcclassistakenintoconsider-ation,whichisequalto1:3,theratioofblockingprobabilitiesforthetwotrafcclassesis

PAGE 56

43 Figure3-4:Connectionblockingprobabilityinahomogeneousenvironment alsoabout1:3.Throughdirectcalculation,wendoutthatonaverage,theCBPforreal-timetrafcisreduced58.23%inDMTBR GcomparedtothatinDMTBR A,whereastheCBPfornon-real-timetrafcisonlyincreased9.89%inDMTBR GcomparedtothatinDMTBR A.InFig. 3-5 ,weseebothschemessuccessfullykeepthehandoffdroppingprobabilitiesofbothtrafcclassesunderthepredenedQoSboundsasexpected,evenwhenthenetworkisexperiencingheavytrafc.Also,thereisnotbigdifferenceinthesetwoschemesintermsofhandoffdroppingprobabilities.InadditiontoCBPsandCDPs,wechecktheperformanceofthesetwoschemesintermsoftrafcthroughput.Thetrafcthroughputforreal-timeornon-real-timetrafcisdenedasfollows:TPrtnrt=BWrtnrtPCHTofrtnrtconnections: CCELL NUMST.14whereC,asmentionedbefore,isthetotalnumberofBUsavailableineachcell,CELL NUMisthetotalnumberofcellsintheentirenetworkandSTisthetotalsimulationtime.Ratherthanusingtheaveragetimespentbyeachconnectioninacell[27],wecounttheactualtimespentbyeachconnection.Obviously,thiswillgiveusamoreaccurateresult.Theentirenetworkthroughputisthesumofthethroughputforeachtrafcclassinthenetwork.

PAGE 57

44 Figure3-5:Handoffconnectiondroppingprobabilityinahomogeneousenvironment ThetrafcthroughputofeachtrafcclassforbothschemesisshowninFig. 3-6 .InDMTBR A,thetrafcthroughputisapproximatelyincreasingwiththeconnectionarrivalrate.ForDMTBR G,whentrafcarrivalrateislow,itperformsjustlikeDMTBR A.However,whentrafcarrivalrateisgettinghigher,orwhenthenetworkisoverloaded,itbehavesalittledifferently.Thereal-timetrafcthroughputisstillincreasingwiththear-rivalrate,whereasnon-real-timetrafcthroughputisremaininginacertainlevelandlow-eringabitattheend.Thisisbecause,inDMTBR G,thelowerCDPforreal-timetrafcisachievedatthecostofthehigherCBPfornon-real-timetrafc;accordingly,weobservethedecreaseinthethroughputofnon-real-timetrafcandtheincreaseinthethroughputofreal-timetrafc.Nevertheless,itcanbeobservedthatbothschemessuccessfullyachieveastablethroughputevenunderheavytrafcsituation.Fromsystem'spointofview,thesetwoschemesdifferverylittleintermsofnetworkthroughput,whichcanbeobservedinFig. 3-7 .Thenetworkthroughputkeepsincreasingastheofferedloadincreases,showingverylittledifferencefromeachother.CombiningtheobservationinFig. 3-4 and 3-5 ,wediscovertheadvantageofDMTBR GoverDMTBR A,whichisgainedbygeneralizingtheconceptofrelativepriority.Withoutcompromisingnetworkthroughput,DMTBR Gsignicantlyimprovestheusersatisfactionfornewreal-timetrafcintermsoftheCBP,

PAGE 58

45 Figure3-6:Trafcthroughputinahomogeneousenvironment whileonlyslightlyaffectingtheusersatisfactionfornon-real-timenewtrafc.Meanwhile,theusersatisfactionforhandofftrafcintermsoftheCDPiswellmaintainedforbothschemes.Next,weconsiderthedetailedthrottlingoperationsineachcell.Fig. 3-8a and 3-8b showineachscheme,thethrottlingprobabilitiesforbothtypesoftrafc,startingfromthebeginningofasimulationrunt=0forarrivalrate=0.1incell0andcell36.Inthesimulationmodel,cell0isinthecenterandcell36islocatedattheedge.Inbothgures,astimepasses,thethrottlingprobabilitiesforreal-timetrafcarealmost1,whichmeanstheneighboringcellsofcell0or36rarelythrottletheacceptanceofnewreal-timeconnections.ThisisconsistentwithFig. 3-5 ,wheretheCDPforreal-timetrafciswellbelowthepredenedQoSbounds,indicatingthereisnoneedtoreducethenewconnectionadmissionforfulllingtherstQoScriterion.Fornon-real-timetrafc,astimepasses,thethrottlingprobabilitiesrstdropandthenuctuatearoundacertainvalueafterthenetworkisinastablestate.Thus,weknowthatthecellskeeptherstQoScriterionfornon-real-timetrafcwiththehelpofcooperativeneighbors,whichreducetheadmissionprobabilityfornewconnectionsduetonon-real-timetrafcwhennecessary.

PAGE 59

46 Figure3-7:Systemthroughputinahomogeneousenvironment Fig. 3-9 showshowvaryingQoSrequirementsaffecttheperformanceofthescheme.Sinceweassumethatalargeportionoftrafcisduetonon-real-timetrafc,wepresenttheresultsobtainedforDMTBR AwiththevariationofQoSnrtforpurposeofillustration.Asobserved,theCBPincreasesastheQoSbound,QoSnrt,isrelaxed.Thisisbecauseofthefactthat,alooseQoSboundmeanslessbandwidthneededtobereservedforthehandofftrafc.Asaresult,themoreavailablebandwidthcanbeusedtoacceptmorenewconnections.SinceaxedratiooftheCBPsforbothtrafcclassesaremaintained,theytogetherwillincreaseatthesamepace.CDPsforbothtrafcclassesaregettinglarger.FortheCDPofnon-real-timetrafc,itiseasytounderstandsinceweloosentheQoSbound.FortheCDPofreal-timetrafc,sincemorenewconnectionsareaccepted,itislikelythatitscorrespondingCDPwillincreasegiventhattheentirenetworkbandwidthisxed.ThethroughputsforbothtrafcclassesincreasebecausetheeffectoftherelaxedQoSnrtismagniedandthenreectedonthereductionofCBPs,whichresultinalargerthroughput.Accordingly,thenetworkthroughputisalsoincreasing.Wealsoinvestigatehowdependentthesystemperformanceisonthevalueofupdatefrequency,N.LargerNmeansslowerupdaterate.Fig. 3-10 showsthatalltheperfor-mancemetrics,includingtheCBP,theCDP,andthroughput,arenotverysensitivetothe

PAGE 60

47 a b Figure3-8:Throttlingprobability.aDMTBR A.andbDMTBR G.

PAGE 61

48 Figure3-9:Performancevs.QoS valueofN,evenwhenNisequalto80.Therefore,thesystemcanworkwellwithoutincurringtoofrequentupdates,hencecommunicationandcalculationoverheadsamongbasestations.Asaconsequence,thisschemecouldbeimplementedandusedincurrentsystems.Finally,wemeasuretheperformanceofthesetwoschemesinaheterogeneousenvi-ronment.Althoughthenetworktrafccompositionisfairlystableinalongrun,itmayuctuatesometimes.Inthesimulation,wemodelasystemwheretheportionofthenon-real-timetrafcintheentiretrafcischangingfrom70%,to75%,thento80%,whilealltheothertrafcparametersarethesame.Foreachspecicvalueofthetrafcportion,thenetworkstaysforonethirdofthetotalsimulationtime.TheresultsareshowninFig. 3-11 through 3-14 .Theresultsshowthattheschemesstillbehavethesameway,whichmeansourschemecouldworkwellinheterogeneouscase,too.Thisisexpectedsincetheschemeis,inessence,anadaptiveone.

PAGE 62

49 Figure3-10:Performancevs.updatefrequency Figure3-11:Connectionblockingprobabilityinaheterogeneousenvironment

PAGE 63

50 Figure3-12:Handoffconnectiondroppingprobabilityinaheterogeneousenvironment Figure3-13:Trafcthroughputinaheterogeneousenvironment

PAGE 64

51 Figure3-14:Systemthroughputinaheterogeneousenvironment 3.5ConclusionInthischapter,adynamicmultiple-thresholdbandwidthreservationDMTBRschemeisproposedtoguaranteeQoSprovisioninginwirelessmultimedianetworks.AccordingtonetworktrafcsituationsandQoSstatus,threebandwidththresholdsaredynamicallyadapted.Inaddition,whenthenetworkisunderheavytrafcload,cooperationamongneighboringcellsisexploited.Asaresult,thisschemeisabletoprovideQoSguaranteewhileefcientlyutilizingnetworkresourceundervarioustrafcloadsinbothhomoge-neousandheterogeneousenvironment.Also,wegeneralizetheconceptofrelativeprior-ityandshowtheacquiredperformancegain.Withthesedesirablefeatures,ourproposedschemeislikelytobeusefulinfuturewirelesssystems.

PAGE 65

CHAPTER4LOCATION-AWARERESOURCEMANAGEMENTINMOBILEADHOCNETWORKS4.1IntroductionAswirelessmobileadhocnetworksMANETsarendingmoreapplicationsinmanyelds,suchasbattleeldcommunications,disasterrescue,andinimicalenvironmentmon-itoring,thereisagrowingneedtoprovidequalityofserviceQoS.Duetothelackofxedwiredinfrastructureinadhocnetworks,allthecommunicationsbetweenwirelessmobilenodesareconductedoverbandwidthlimitedanderror-pronewirelesslinks.Theinforma-tionexchangesamongthesenodesmayinvolvemultiplehops,withintermediatenodesactingasroutersinbetweensourcesanddestinations.ThesalientfeaturesofMANETsandtheapplication-specicrequirementsposegreatchallengesonboththeroutinglayerandthemediaaccesscontrolMAClayer.Routingprotocolsmustbeabletohandleroutebreakageandre-routing,asanyfailureoftheintermediatenodesontherouteduetopoweroutageormobilitywilldisabletheentireroute.Meanwhile,MACprotocolsshouldeffec-tivelyreduceaccesscollisionsandachievehighchannelutilization.Moreover,awirelessmobileadhocnetworkshouldsupportserviceprioritizationintermsofnetworkaccess,givenitslimitedresources. 1 Inotherwords,someimportanttrafcneedstobeacceptedbythenetworkwithhigherprioritythanothers.Forinstance,inabattleeld,thecommunicationrequestsamongsomearmycommandersshouldbealwaysacceptedwithhigherprioritycomparedwiththecommunicationsbetweensoldierssincethecommandersmayneedtodiscussandcoordinatethestrategicplans.Another 1Unlessotherwisespecied,resourceandbandwidthareinterchangeableinthischap-ter. 52

PAGE 66

53 typicalscenarioisthatinclusterbasedhierarchicaladhocnetworks,connectionsamongclusterheadsshouldbespeciallytreated.Insuchcases,thecommunicationsbetweensuchimportantnodeslikecommanders,clusterheads,ornodesthattakeovercontroltasksinthenetwork,henceforthcalledi-nodesshouldbegrantedhigherpriorityintermsofnetworkaccess.Thus,thetasktosupporttheseimportantcommunicationsbetweentheseimportantnodeswithQoSguaranteeisimperative.However,incurrentliterature,theresearchonhowtosupportsuchimportantcommu-nicationswithQoSandhighacceptanceratioisstillthinonthegroundateitherroutinglayerorMAClayer.Attheroutinglayer,avarietyofQoSroutingschemeshavebeenproposed,withtheprimarygoaltondapathfromthesourcetothedestinationthatsat-isesthedesiredQoSrequirements[17,65,93,102].Intheseschemes,wheneverthereisaconnectionrequest,theroutingprotocolwillattempttondapathwithsufcientresourcesbandwidthtosupportQoSrequirements.Ifthereisnosuchapath,therequestwillberejected.Thisisundesirablefori-nodes,sinceinmostcases,theconnectionsbe-tweenthemcarryimportantandtime-criticalinformation.Moreover,theseschemestreatallconnectionsindiscriminatelyandthusdonotappearsatisfactorytoaccommodatetheaforementionedimportantcommunicationsbetweenthosehigh-prolednodes.Obviously,withoutaneffectiveandefcientresourcemanagementscheme,itisveryhardforQoSroutingschemestoovercomethisdeciency.Meanwhile,manycontention-basedMACprotocolswereproposedtoprovidediffer-entiatedservice[1,2,51,88].However,itisknownthatreservingresourcesincontention-basedMACisdifcultandhencemayfailtoprovidehighdegreeofserviceguarantee.Asaconsequence,someslot-basedMACschemesweredesigned[8,21,23].Thoughtheyndtheirnicheinwirelesslocalareanetworks,unfortunately,formulti-hopconnections,theyalonestillcannotsupportthoseimportantcommunicationswithQoS.Infact,theroutinglayerandtheMAClayershouldcollaborate.Insteadofran-domlyallocatingtheavailableresourcetothenewlyacceptedconnections,therouting

PAGE 67

54 layershouldintelligentlyallocatetheresourceforaconnectionalongitsentirepathsuchthattheresidualresourcecanaccommodatemoreconnections.Thismeansthattherout-inglayerneedstoknowtheallocationrestrictionsduetoMACcollisionsoftimeslots.Inaddition,adesirableroutingprotocolalsohelpstheMAClayermaintainhighresourceuti-lization.Further,itisobservedthatanode'slocationormovementisvaluableinformationthatcouldbeutilizedtoassistbothlayersinfulllingthetask.Alltheseobservationsthusmotivateustoresorttoapproachesthatarebasedonthecooperationofthesetwolayersandtheuseofnodes'locationinformation.Ontheotherhand,tosupportthecommunicationsbetweeni-nodes,theabovemen-tionedcollaborativeapproachisnotadequate.Inspiredbytheresourcereservationconceptincellularnetworks,webelieveitisadvantageoustopre-reservesomeresourcessothati-connections,whicharedenedasconnectionsbetweenanytwoi-nodes,canhavehigheracceptanceprobability.Althoughmanyreservationschemeshavebeenproposedforcellu-larnetworks, 2 duetothetremendousdifferencebetweenMANETsandcellularnetworks,theycannotbedirectlyappliedtoourcase.Ascanbeimagined,resourcereservationismuchmorecomplicatedinMANETs.Specically,incellularnetworks,bandwidthisre-servedatbasestation,whichisstaticandhasgoodknowledgeaboutthemobilityofallthemobileusersinthecellitserves.Thus,bandwidthcanbereservedeasilyandeffectively;andoncereserved,itisalwaysavailabletomobileusers.Bycontrast,inadhocnetworks,bandwidthisreservedateachmobilenodethatmaymove,andmostlikelyisnotawareofothernodes'mobility,whichgreatlyincreasesthedifcultyofmakingefcientresourcereservation.Therefore,designingeffectiveandefcientbandwidthpre-reservationinadhocnetworksisnottrivial. 2Forexample,theGuardChannelschemewasproposedincellularnetworkstogivehighprioritytohandoffcallsovernewlyincomingcalls[42].

PAGE 68

55 Inthischapter,byintroducingtheresourcepre-reservation 3 intoMANETsandtakingadvantageofnodes'locationinformation,weproposearesourcemanagementschemethatisdesignedonthebasisofcooperationbetweentheroutinglayerandtheMAClayerinaTDMAbasedadhocnetworks.Inthisscheme,wegranthigherprioritytoconnectionsbetweenanytwoi-nodesthroughlocation-awarebandwidthpre-reservation.Byutilizingeachnode'sgeographiclocationinformationandformingaquadrangle-shapedareabe-tweenanytwoi-nodes,thebandwidthpre-reservationmechanismpre-reservesbandwidthfori-nodesalongthepossiblepathsconnectinganytwoi-nodesinsuchawaythatpoten-tialcollisionsoftimeslotsaresignicantlyreduced.Subsequently,weuselocation-awareforwardalgorithmLAFA,abandwidthcalculationscheme,whichalsoexploitsthege-ographicinformationofeachnodeonthepath,tondandreserveenoughbandwidthtosupportQoSwhilekeepingtheresidualresourcemoreefcientforfuturecommunications.Asaresult,thenetworkbandwidthisusedmoreefcientlyduetothereductioninMACslotcollisions.Therefore,theconnectionblockingprobabilityfori-connectionsisgreatlydecreasedwithoutseriouslyhurtingadmissionsofotherconnections.Theremainderofthischapterisorganizedasfollows.ThesystemmodelisintroducedinSection 4.2 .Ourlocation-awarebandwidthpre-reservationschemeandtheLAFAalgo-rithmarepresentedinSection 4.3 andSection 4.4 ,respectively.InSection 4.5 ,perfor-manceevaluationisgivenwithsomeinsightfuldiscussions.WediscusssomeimportantdesignissuesinSection 4.6 .Finally,concludingremarksaregivenintheSection 4.7 .4.2SystemModelWeconsideramobileadhocnetworkconsistingofNnodes,amongwhichthereareMi-nodes.ThesystemadoptsTDMAasitschannelaccessmechanism.Thetransmissiontimescaleisdividedintoframes,eachofwhichcontainsaxednumberoftimeslots.Each 3TodistinguishfrombandwidthreservationinQoSrouting,wecallitbandwidthpre-reservation.

PAGE 69

56 frameiscomprisedoftwophases.Oneiscontrolphaseandtheotherisdataphase.Inthecontrolphase,allkindsofcontrolfunctionsareperformed,suchaspre-reservationrequestpropagation,bandwidthreservationandconnectionsetup,frameandslotsynchronization.Eachnodemayuseapredenedslottobroadcastcontrolmessagestoallofitsneighbors.Dataphaseisusedtotransmitandreceivedatapackets.Foreachtimeslotusedtotransmitorreceivedatacalleddataslot,itmaybeinoneofthefourstatesatoneparticulartime:FREE,PRE RESERVED,USED TXandUSED RX.Theyindicatethestateofthedataslotbeingfree,pre-reserved,usedtotransmitandusedtoreceive,respectively.Allconnectionsconsideredinthischapterareconnection-orientedandrequiresconstantbandwidth.InadhocnetworksusingTDMA,bandwidthismeasuredintermsoftimeslots.Hereafter,thetermsbandwidthandtimeslotareusedinterchangeably.AconnectionwillspecifyitsQoSrequirementasthenumberoftransmissiontimeslotsitneedspriortobeingadmittedintothenetwork.Inthenetwork,inadditiontoi-connectionswhosebothendpointsarei-nodes,alltheotherconnectionsareordinaryconnections.Weassumeinthisworkthateachnodeishalf-duplex.Inotherwords,itcannottrans-mitandreceivesimultaneously.Tosuccessfullytransmitapacket,boththetransmitterandreceivernodesneedtohaveoneormorecommontimeslots.Furtherrestrictionsapplytotheselectionoftimeslotsalongapathduetothehiddenterminalproblem.Forexample,inFig. 4-2 ,ifnodentransmitspacketsdestinedtonode0tonoden)]TJ/F15 11.955 Tf 9.96 0 Td[(1intimeslot1andnoden)]TJ/F15 11.955 Tf 12.091 0 Td[(1forwardsthepacketstonoden)]TJ/F15 11.955 Tf 12.091 0 Td[(2intimeslot2.Sincenoden)]TJ/F15 11.955 Tf 12.091 0 Td[(2,two-hopawayfromnoden,cannothearfromnoden,itmayalsousetimeslot1toforwardthepacketstonoden)]TJ/F15 11.955 Tf 10.288 0 Td[(3.Then,noden)]TJ/F15 11.955 Tf 10.289 0 Td[(1willdetectacollisionandcannotcorrectlydecodethepacketsfromnoden.Wealsoassumethateachnodeisawareofitsowngeographiclocationandtimeissynchronizedateachnode,ascurrentGlobalPositioningSystemGPS[75,87]canprovideaccuratelocationinformationandglobaltiming.Finally,itisassumedthatalli-nodesareawareofeachother'sgeographiclocation.Sincei-nodesneedtofrequently

PAGE 70

57 communicatewitheachother,theycanpiggybacktheirownlocationinformationandmo-bilityinformationsuchasvelocityanddirectionindatapackets.I-nodesotherthanthesourceandthedestinationofi-connectionscanalsoacquirethisinformationbyoverhear-ing.Amorepracticalsolutionmaybelocationregistrationandlookupservicethatmapsnodeidentitiestolocation[64].Furtherdiscussiononthisissueisbeyondthescopeofthischapter.Formally,anadhocnetworkismodeledasagraphG=N;L,whereLisasetofbi-directionallinks[102].AnodeihasasetofneighborsNBi=fj2N:i;j2Lg.AssumethesetofdataslotsinoneframeisS=fs1;s2;:::;sKg.ThesetofslotsTSiinwhichnodeitransmitsisdenedasitstransmissionschedule.ThesetofnodesRkj,Rkj2NBi,consistsofthereceiversinslotsk,sk2TSi.SetRSi=fsk2S:i2Rkj;j2NBigisthesetofslotsinwhichnodeiusestoreceivefromitsneighbors.Anothertwosetsaredenedfornodei:SRTi=fsk2S:sk=2TSi;sk=2RSi;sk=2[j2NBiRSjg,SRRi=fsk2S:sk=2TSi;sk=2RSi;sk=2[j2NBiTSjg.SRTiisthesetofslotsinwhichicantransmitwithoutcausinginterferencetoitscurrentreceivingneighborsandSRRiisthesetofslotsinwhichicanreceivewithoutsufferinginterferencefromitscurrenttransmittingneighbors,giventhecurrenttransmissionscheduleTS.4.3BandwidthPre-ReservationToreducetheblockingprobabilityofi-connections,weneedtopurposelypre-reservesomebandwidthbeforehand.Beforeweproceed,itisimportanttonotethedifferencesinbandwidthreservationorpre-reservationincellularnetworksandMANETs.Incellularnetworks,bandwidthisreservedatbasestations;however,inadhocnetworks,bandwidthisreservedineachmobilenode,whichdiffersfrombasestationsinseveralsignicantways.First,abasestationisaxedinfrastructure,whichmeansoncethebandwidthisreserved,itisalwaysavailabletomobileusers.Conversely,bandwidthpre-reservedinamobilenodemaybeunavailableasthemobilenodemoves.Second,abasestationmayhavegoodknowledgeaboutthemobilityofallthemobileusersinthecellitserves,andhence

PAGE 71

58 canmakeproperbandwidthreservation;inadhocnetworks,however,amobilenodedoesnotknowthemobilityinformationaboutothermobilenodes,whichmakesitdifculttoappropriatelypre-reservebandwidth.Finally,comparedwithabasestation,amobilenodehaslimitedpower,processingcapability,andbufferspace;therefore,unlikeabasestation,amobilenodecannotaffordanycomplicatedalgorithmforbandwidthpre-reservation.4.3.1Pre-ReservationRequestPropagationSincestaticallypre-reservingbandwidthisnotefcientduetonodemobilityortraf-cdynamics,thebandwidthneedstobedynamicallypre-reserved.Eachnodeneedstoknowwhentopre-reservebandwidth,howmuchbandwidthtopre-reserveforfuturein-comingi-connectionrequestsandwhentode-pre-reserve.Tothisend,eachi-nodeofani-nodecommunicationpairneedstoprovidetheintermediatenodeswithsuchinformation.Forthispurpose,apropagationandprocessingmechanismforpre-reservationrequestsisintroduced.AsshowninFig. 4-1 ,inordertopre-reservesomebandwidthalongthepathtoi-nodej,i-nodeibroadcastsamessagecalledpre-reservationrequesttoitsneighbors,suchasnoden.Therequestcontainsthefollowinginformation:athepreferrednumberoftimeslotstobepre-reserved,num reqitcanbesetbyi-nodesdependingontrafchistoryandapplicationtypes;btheduration,timer,forwhichthetimeslotsmaybepre-reserved;cthelocationofbothi-nodeiandi-nodej;di-nodeiandj'sidentitiesIDs;eauniquesequencenumber.Uponreceptionoftherequest,eachone-hopneighborofi-nodei,likenoden,willretrievethelocationinformationofthesourceandthedestinationi-node.Withthisinformation,aquadrangle-shapedinuenceareawillbebuiltasshowninFig. 4-1 .Theangleadjuststhesizeofinuenceareaandhencethecontroloverheadincurred.Iftheneighborhappenstofallintothisarea,itwilltrytopre-reservethenumberoftimeslotsasrequired.Fornoden,thisisthecase.Assumethepre-reservationupperboundatnodenisUB,andthecurrentnumberofpre-reservedtimeslotsisnum tot.Then,nodenwilltakeoneofthefollowingactions:

PAGE 72

59 1. num tot+num req6UB,whichmeansthenodecanpre-reservenum reqfreetimeslots.Thenthenodewillchooseeithernum reqtimeslotsorallcurrentlyfreetimeslotsdenotedbynum free,whicheverissmaller,andtagthemwithagPRE RESERVED.Thennum tot=num tot+minnum req;num free.Thecriteriongoverninghowtochoosewhichtimeslottotagwillbedescribedinthefollowingsubsection.Besides,eachchosentimeslotwillbeassociatedwithatimer,slot timerk,wherekistheIDofthetimeslot.Initially,slot timerkissetto0.slot timerk=timer.Whenslot timerkexpiresandthetimeslotwillbefreed,i.e.,markedwithagFREEagain. 2. num tot=UB,whichmeansnomorefreetimeslotscanbepre-reserved.Thenthenodewillrankallthepre-reservedtimeslotsinanincreasingorderofthevalueoftheirattachedtimer.Therstnum reqtimeslots'timerwillbeincreasedbyavalueoftimer,i.e.,slot timerk=slot timerk+timer,withtheagPRE RESERVEDunchanged.Bydoingthis,thoughthenumberofpre-reservedtimeslotsdoesnotin-crease,theirpre-reservationtimeisprolonged.Thus,thishasthesameeffectasincreasingthepre-reservedtimeslotnumber. 3. num tot>UB.NotethatthismayoccursincewedynamicallyadjustUB.Inthiscase,thenodewillchoosetherstUBtimeslotswiththelargesttimervalue.AmongtheUBtimeslots,num reqtimeslotswiththeleasttimervaluewillbeincreasedbyavalueoftimer.Itcanbeseenthatthenum tot)]TJ/F25 11.955 Tf 12.171 0 Td[(UBtimeslotsthathavesmalltimervaluemayexpiresoon. 4. num totUB.ThenthenodewillchooseUB)]TJ/F25 11.955 Tf 12.519 0 Td[(num totfreetimeslotsandpre-reservethemasdescribedin1.Theleftnum req)]TJ/F15 11.955 Tf 12.332 0 Td[(UB)]TJ/F25 11.955 Tf 12.331 0 Td[(num tottimeslotswillbeprocessedaccordingto2.Thennum tot=UB.Next,nodenwillbroadcastthepre-reservationrequestmessagetoitsneighbors.InFig. 4-1 ,theyarenoden+1,n+5.Then,noden+1andnoden+5willalsochecktheirposition.Iftheyarealsointhisquadrangle,suchasnoden+1,theywilltaketheactionsmentionedabove.Iftheyarenot,suchasnoden+5,theysimplyignoretherequest.Sonoden+5willnotpassonthepre-reservationrequesttonoden+6.Topreventanodefrompre-reservingmorethanonceforthesamepre-reservationrequest,repeatedreceptionofthesamerequestwillbedetectedbytheIDsofsourceanddestinationi-nodesandthesequencenumber.

PAGE 73

60 Figure4-1:Illustrationofaquadrangle-shapedinuencearea Thefrequencyofsendingoutpre-reservationrequestisadjustedbyatimer,TIMERupdate,maintainedateachi-node.Wheneverthistimerexpires,thei-nodewillsendanotherpre-reservationrequest.TheadaptationofTIMERupdatewillbeaddressedbelow.However,TIMERupdateisalwaysalittlesmallerthantimertoaccountforthepropagationdelayofpre-reservationrequests.Sothepre-reservedtimeslotatanodewillnotbefreedjustbecausethenodecannotreceiveanewpre-reservationrequesttoupdatethetimerduetodelay.Anotherimplicationis,onceTIMERupdateisdetermined,timerisalsodeter-mined.Sofar,itisassumedthatthepre-reservationrequestissuccessfullytransmittedincontrolphase.However,thismechanismstillworksevenwhenapre-reservationrequestmaybelostorcorruptedduringtransmission,sinceanewpre-reservationrequestwillbetransmittedsometimelater.Itisimportanttonotethatthepre-reservedtimeslotsarenotexplicitlytiedtotheuseofanytwospecici-nodes,thoughtheymaybepre-reservedaccordingtotherequestsbythem.Accordingly,everyi-connectioncanmakeuseofthemiftheyareavailable.

PAGE 74

61 Figure4-2:Exampleofpre-reservationfailure 4.3.2BandwidthPre-ReservationCriterionItisknownthatmulti-hoptopologyofadhocnetworksallowsspatialreuseofband-width.Differentnodescanusethesamebandwidthsimultaneouslyaslongastheyaresufcientlyseparated.However,withinone-hopandtwo-hopdistance,timeslotsbelong-ingtoneighboringwirelesslinksmaycollidewitheachother,duetothebroadcastingnatureofwirelesstransmission,half-duplexproperty,andthewell-knownhiddenterminalproblem.Asaresult,eventhoughpre-reservationcansetasidesomefreetimeslotsateachnodefori-connections,theseslots,ifcarelesslypre-reserved,maybeuselessbecauseofcollisions.Worseyet,inappropriatepre-reservationwillcausedramaticbandwidthun-derutilizationsincethoseimproperlypre-reservedtimeslotsmayotherwisebeavailabletootherordinaryconnections.WerefertoFig. 4-2 asaninstanceofpre-reservationfailure.Sourcenodenwantstoopenaconnectionwithdestinationnode0.Priortondingthepathwithsufcientbandwidth,supposethepre-reservationmessageisalreadysentfromnodentonode0.Aswesee,nodenpre-reservedslots0,1,7;noden)]TJ/F15 11.955 Tf 12.413 0 Td[(1pre-reservedslots0,5,6;noden)]TJ/F15 11.955 Tf 12.824 0 Td[(2pre-reservedslots0,1,5;noden)]TJ/F15 11.955 Tf 12.824 0 Td[(3pre-reservedslots2,8,9.Althoughallthesenodeshavepre-reservedthreetimeslots,duetotimeslotconict,wecannotsuccessfullyndapathgettingthroughfromnodentonoden)]TJ/F15 11.955 Tf 12.648 0 Td[(3withonebandwidthuniti.e.,oneslot,letaloneapathfromnodentonode0.

PAGE 75

62 Moreover,astheconnectionpathgetslonger,itislesslikelytondapathwithsuf-cientbandwidthduetothetransmissioncollisionwithinone-hopandtwo-hopneighbor-hood.Toillustratethis,weuseamathematicalmodeltoderivetherelationshipbetweentheconnectionblockingprobabilityandthehoplengthoftheconnection.Weassumeeachconnectionrequiresthebandwidthofonetimeslot.Foreachlink,thetotalbandwidthisS.ThelinkbandwidthLBofeachlinkisofi.i.dBernoullidistributionwithsuccessprobabilityPa,i.e.,theprobabilitymassfunctionPMFofLBisdenedasBS;i;pa=0B@Si1CApia)]TJ/F25 11.955 Tf 11.955 0 Td[(paS)]TJ/F26 7.97 Tf 6.587 0 Td[(i.1wherePaindicatescurrenttrafcloadinthenetwork.Wedeneqi=BS;i;pa,fori=0;1;:::;S.WealsodeneP1astheprobabilityofPrLB>1,whichisequalto1)]TJ/F25 11.955 Tf 12.606 0 Td[(BS;0;Pa.Basedonthis,wederivetheconnectionblockingprobabilitywhenthepathcomprisesone,twoandthreehops,asshowninAppendix 4.8.1 .Wecanseethat,asthenumberofhopsincreasesfrom1to3,theblockingprobabilityisalsoincreased.Forinstance,ifS=10,Pa=0:1,theblockingprobabilitiesforQoSpathwithone,two,andthreehopsare0.3487,0.5908,0.7695,respectively.Ifthenumberofhopsofapathisgreaterthan3,theblockingprobabilitybecomeslargerastimeslotconictincreases.Intuitively,ifawirelesslinkcanbeisolatedfromtheinterferenceduetoitsneighbor-inglinks,moretimeslotsatthelinkcanbeusedduetothereducedcollisions.Thus,moreconnectionscanbeadmittedintothenetworkandhencealowblockingprobability.Moti-vatedbythisintuition,weproposetopre-reservetimeslotsinsuchawaythat,withtheaidofeachnode'slocationinformation,thepre-reservedtimeslotsarespreadandhencetheyarenotwastedsimplybecauseofthetransmissioncollisionwithinone-hopandtwo-hopneighborhood.Next,wedetailthepre-reservationcriterioninthefollowing.AssumethenetworkgeographicallyoccupiesarectangularareawiththesizeofXYm2.Iftheareaisnotaregularrectangle,itcanbeapproximatedwiththesmallestrectangle

PAGE 76

63 whichcancoveralltheentirearea.ThetransmissionrangeofeachnodeisRm.Agridstructureisbuiltbydividingthenetworkareaintoagridofcellswithsize2R2R.Asaresult,alongthehorizontalaxis,thereareI=dX=2Recells.Andalongtheverticalaxis,thereareJ=dY=2Recells.Thecellsaredenotedbyi;j,asshowninFig. 4-3 .Wefurtherdenethedirectionofapre-reservationrequest.Assumewhenanoden,withlocationdenotedbyxn;yn,receivesapre-reservationrequestoriginatedbyi-nodes,withlocationdenotedbyxs;ys,itwillcompareitshorizontaldistancefromnodes,i.e.,jxn)]TJ/F25 11.955 Tf 11.005 0 Td[(xsj,withitsverticaldistancefromnodes,i.e.,jyn)]TJ/F25 11.955 Tf 11.004 0 Td[(ysj.Ifjxn)]TJ/F25 11.955 Tf 11.004 0 Td[(xsj>jyn)]TJ/F25 11.955 Tf 11.004 0 Td[(ysj,wedenethispre-reservationrequesttobehorizontalfornodenwithrespecttonodes;otherwise,wedeneittobevertical.Next,timeslotsinaframearealsodivided.AssumethereareSdataslotsinonetimeframe.Theyaredividedintotwoparts,denotedbyH1andH2,respectively.H1isthesetoftimeslotswhichwillbepre-reservedforhorizontalpre-reservationrequestsandH2isthecounterpartforverticalpre-reservationrequests.Tosimplifynotation,wealsouseH1andH2torefertotheirrespectivesize.Thepartitionisinproportiontothedimensionofthegeographicarea,i.e.,H1=I I+JSH2=J I+JS.2AsshowninFig. 4-3 ,thetimeslotsinH1arefurtherequallydividedintoKhKh>2sections,i.e.,section0;1;2;:::;Kh)]TJ/F15 11.955 Tf 12.135 0 Td[(1.Similarly,H2arefurtherequallydividedintoKvKv>2sections.Thegeometricalpositionofeachcellisthenmappedtothetimeaxis.Alongthehorizontaldirection,celli;jismappedtotimeslotsectionmodj;Kh.Alongtheverticaldirection,celli;jismappedtotimeslotsectionmodi;Kv.Functionmodp;qreturnsthemodulusobtainedbydividingpintoq.Morespecically,thepre-reservationalgorithmfornoden,locatedincelli;j,topre-reservebw numtimeslotsforapre-reservationrequestoriginatedati-nodes,isoutlinedinFig. 4-4 ,whereNsdenotestheIDofthetimeslotsection.

PAGE 77

64 Figure4-3:Locationandtimedivision

PAGE 78

65 Undersuchamappingrule,pre-reservedtimeslotsarespreaddependingonthenode'slocationandthedirectionofthepre-reservationrequest.Sincethepotentiali-connectionwilllikelyfollowthesameshortestpathexperiencedbypre-reservationrequests,inmostcases,thepre-reservedtimeslotsfordifferenti-connectionswillnotcollidewitheachotheriftheydonotcomeinthesamedirectionwithrespecttothenodedoingpre-reservation.Foraconnectionwiththreeormorehops,itislikelythatanythreeofitsconsecutivehopsspantwoneighboringcellsandthuscanhaveaccesstodifferenttimeslotsBysettingthecelldimensionto2R,ifthereisaconnectionacrossthecell,thereareatmosttwohopsinthecellfortheconnection,asthepre-reservedtimeslotschosenfromdifferentsectionswillnotoverlapinmostcases.Therefore,transmissioncollisionamongpre-reservedslotsisreducedamongdifferentconnectionsanddifferenthopswithinoneconnection.Asaresult,whentheQoSroutingprotocolismakingbandwidthreservationalongthepath,itwillndapathwithsufcientbandwidthwithalargerprobabilitysincethepre-reservedslotsareavailable.Twoimportantpointsarenoted.First,thereissomeoverlap,So,betweeneverytwoadjacentsections.Thereasonwhythereisoverlapintimeslotsectionistoallowforthecaseswheresomewirelesslinksmayspantheboundaryoftwoadjacentcells.Inotherwords,oneendofthelinkislocatedinonecellwhiletheotherendislocatedinanadjacentcell.Withtheoverlapping,bothnodeofthewirelesslinkwillbeabletoreservethesametimeslotfortransmissionandreception.Second,ascanbeseen,thisschemeismoreeffectivewhenthenetworksizeisrelativelylarge,e.g.,forthenetworkwhereQoSconnectionscouldtraverseafewhops>4hops.Fig. 4-5 showsthebandwidthpre-reservationfollowingtheabovecriterion,inwhichnodenandn)]TJ/F15 11.955 Tf 12.091 0 Td[(1belongtoacell,andnoden)]TJ/F15 11.955 Tf 12.076 0 Td[(2andn)]TJ/F15 11.955 Tf 12.077 0 Td[(3belongtoaneighboringcell.SimilartothecaseshowninFig. 4-2 ,threetimeslotsarepre-reserved.However,withourlocation-awarepre-reservation,itcanbeseenthattheslotsarepre-reservedsuchthatwecanndapathgettingthroughfromnodentonoden)]TJ/F15 11.955 Tf 11.955 0 Td[(3,andtonode0withonebandwidthunit.

PAGE 79

66 BandwidthPre-reservationAlgorithm//assumenodenneedstopre-reservebw numtimeslotsforsourcei-nodes//locationofs:Ls=xs;ys;locationofnincelli;j:Ln=xn;yn;whilebw numfifjxn)]TJ/F25 11.955 Tf 11.955 0 Td[(xsj>jyn)]TJ/F25 11.955 Tf 11.955 0 Td[(ysj//pre-reservetimeslotsalonghorizontaldirectionfNs=modj;Kh;ifbxn=Rc==0//inthelefthalfofthecellrandomlychooseonefreeslotfromthersthalfofNs;else//intherighthalfofthecellrandomlychooseonefreeslotfromthesecondhalfofNs;gelse//pre-reservetimeslotsalongverticaldirectionfNs=modi;Kv;ifbyn=Rc==0;//intheupperhalfofthecellrandomlychooseonefreeslotfromthersthalfofNs;else//inthelowerhalfofthecellrandomlychooseonefreeslotfromthesecondhalfofNs;gbw num–;g Figure4-4:Pre-reservationcriterion Figure4-5:Anexampleoflocation-awarepre-reservation

PAGE 80

67 4.3.3UpperBoundofBandwidthtoBePre-reservedAnupperbound,UB,isusedtoensurethatbandwidthisnotexcessivelypre-reserved,causingbandwidthunderutilization.Toadapttoinstantaneousnetworkdynamics,suchastrafcarrivalinformation,node'sgeographiclocation,andnodedegree,itmustbedynam-icallyadjusted.Meanwhile,theupperboundprovidesthetradeoffbetweentheperfor-mancesofthesetwotypesofconnections.Iftheupperboundislarge,thei-connectionblockingprobabilitywillbelowwhiletheblockingprobabilityforordinaryconnectionswillbehigh.Smallupperboundimpliesjusttheopposite.Eachnodeinthenetworkwillmonitortheincomingpre-reservationrequestsandthebandwidthrequirementofeachrequest.LetTIMERavgbetheaverageofthetimersasso-ciatedwiththepre-reservedtimeslots,andletpbethearrivalrateofthepre-reservationrequestsobservedatanode.Besides,denotetheaveragebandwidthrequirementofeachpre-reservationrequestbyBW.Then,TIMERavgpBWrepresentstheaveragetotalbandwidthpre-reservationrequirementsforonenodeduringtheaveragepre-reservationperiod.Therefore,wesettheupperboundtothefollowing:UB=TIMERavgpBW.3where<<1isadesignparameterweintroducetoadapttheupperboundUBsothattheupperboundwillnotbecometoolarge.Thus,eachnodeneedstoestimatethecurrentarrivalrateofp.AssumethateachnodemeasuresthearrivalrateataxedtimeperiodT,andthemeasuredarrivalrateatthelthl=1;2;:::measurementperiodisMTl.Thenthearrivalratecanbeestimatedusingexponentialmovingaverage:pl+1=pl+)]TJ/F25 11.955 Tf 11.956 0 Td[(MTl.4whereMTlcanbeobtainedbyMTn=#ofnewarrivalsinlthperiod T.5

PAGE 81

68 andisaweightingfactor,usually0:5<<1.Wecanseethatmoreweightisgiventothearrivalratesrecentlyobserved.4.3.4IntervalofSendingPre-reservationRequestsThecontroloverheadincurredbysendingpre-reservationrequestsisdeterminedbythesendinginterval,TIMERupdate.Atrstglance,theoverheadcanbereducedbyin-creasingTIMERupdate.However,withlargeTIMERupdate,bandwidthpre-reservationcannotquicklyrespondtothenetworktopologychange,whichwillcauseperformancedegradation.Thus,TIMERupdateshouldbesetsoastostrikeabalancebetweenoverheadandperformance.Duetothefactthatonlyi-nodessendoutpre-reservationrequest,andtheyarenotawareofintermediatenodes'mobility,ourapproachwilladjustTIMERupdatebasedonthemobilityofthesendingi-node.TIMERupdate=MINR 2Vavg;TIMERmax.6whereTIMERmaxisthemaximumvaluethatTIMERupdatecantakeon,Vavgistheaveragespeedofthesendingi-node.Itcanalsobeobtainedusingexponentialmovingaverage.TIMERmaxisintroducedfortworeasons.First,itcanpreventTIMERupdatefrombeinginnite,whenthespeedisverysmall.Second,evenifthesendingi-nodeisstationary,theintermediatenodesmaymoveinormoveout.Soitensuresthatthei-nodesendspre-reservationrequeststopre-reservesometimeslotsattheintermediatenodes.4.3.5ControlOverheadAnalysisWeassumethatthenetworkconsistsofNnodesuniformlydistributedinanareaofA.DuringanintervalofTIMERupdate,num datapacketsaretransmittedfromi-nodeitoi-nodej.Iftheaveragedistancebetweentwoi-nodesisD,wecancomputetheareaofthequadrangleasfollows:Aquadrangle=D2 2tan.7

PAGE 82

69 Ifwecountthecontroloverheadasthenumberofnodeswhichsendsorforwardsthepre-reservationrequest,weobtainthecontroloverheadperpacket:Overheadperpacket=NAquadrangle Anum data=ND2tan 2Anum data.8Wecanobservethattheoverheadperdatapacketincreaseswiththeincreaseofnodedensity,theareaofthequadrangleanddecreaseswithincreaseofdatapacketssentduringTIMERupdate.ThisisveriedbysimulationinSection 4.5 .4.4LocationAwareForwarding4.4.1BandwidthCalculationandReservationToguaranteeQoS,QoSroutingprotocolsneedtoestablishapathorroutewithsufcientbandwidthforadmittingnewlyarrivingconnections.ForanyQoSconnectionoriginatedatanode,theadmissioncontroltestisasfollows.Thenoderstattemptstondapathbyroutediscoveryorcheckitsownroutingtable,dependingwhetheranon-demandroutingprotocoloratabledrivenroutingprotocolisused.Ifapathfromthesourcetothedestinationisfound,thenthesourcewillcheckifthepathhassufcientbandwidthtoaccommodatetheconnectionforon-demandroutingprotocol,thisisdonewithroutediscoverysimultaneously.Ifyes,thebandwidthisreservedandtheconnectionadmitted;ifno,theconnectionisrejected.Thedifferentialtreatmentofi-connectionliesinthesearchforpathbandwidth.Forordinaryconnections,theycannotreservethetimeslotswhichhavebeenpre-reservedfori-connectionswhilei-connectionscanreserveeitherfreeorpre-reservedtimeslots.Thebandwidthpre-reservationschemedescribedearlierensuresthateachnode'spre-reservedtimeslotsaredispersedintimeaxisaccordingtoitslocation.Therefore,thechancethatthediscoveredpathhasadequatebandwidthfori-connectionsisgreatlyincreasedastimeslotcollisionsarereduced.i-connectionsthuswillbeadmittedwithhigherprioritythanordinaryconnections.ItcanbeseenfromtheabovedescriptionthatQoSroutingprotocoldependsonthepathbandwidthcalculationalgorithmtomakeadmissiondecision.Unfortunately,

PAGE 83

70 inTDMAbasedlinkaccessschemes,thepathbandwidthcalculationproblemisNP-complete[102].Wethushavetoresorttoheuristicapproaches.ForwardAlgorithmFAisproposedasanefcientanddistributedalgorithmforcalculatingandreservingtheend-to-endbandwidthalongapath.Sinceeverynodeisawareofitslocation,basedonFA,weproposeaLocation-AwareForwardAlgorithmLAFA.Thekeyideais,whendoingband-widthreservation,LAFAwillreservethebandwidthwithrespecttothenode'slocationaccordingtothelocation-to-time-slotmappingcriterion.Inthisway,thegainofreducedtimeslotcollisionduetothecarefullydesignedlocation-dependentpre-reservationcanbeaccrued.Thealgorithmisgivenindetailasfollows.AssumethereisapathP=fnm;nm)]TJ/F24 7.97 Tf 6.586 0 Td[(1;:::;n0g,whereni2N,ni;ni)]TJ/F24 7.97 Tf 6.586 0 Td[(12L,i=m;m)]TJ/F15 11.955 Tf 11.999 0 Td[(1;:::;1.nmisthesourceandn0isthedestination.Thesetoftimeslotsusedonlinkni;ni)]TJ/F24 7.97 Tf 6.587 0 Td[(1tosupportpathfnm;nm)]TJ/F24 7.97 Tf 6.586 0 Td[(1;:::;nkgisdenedasPBki. 1. Ifm=1,PB01=LABW1node;IN;n.9 2. Ifm=2,PB02;PB01=LABW2LB2;LB1;.10 3. Ifm>3,PBm)]TJ/F24 7.97 Tf 6.587 0 Td[(2m;PBm)]TJ/F24 7.97 Tf 6.587 0 Td[(2m)]TJ/F24 7.97 Tf 6.587 0 Td[(1=LABW2LBm;LBm)]TJ/F24 7.97 Tf 6.587 0 Td[(1;.11fork=m)]TJ/F15 11.955 Tf 11.955 0 Td[(3to0doPBkk+3;PBkk+2;PBkk+1=LABW3PBk+1k+3;PBk+1k+2;LBk+1;.12whereLBi=SRTiSRRi)]TJ/F24 7.97 Tf 6.587 0 Td[(1,denotingthelinkbandwidthoflinkni;ni)]TJ/F24 7.97 Tf 6.587 0 Td[(1.Theend-to-endbandwidthofpathPisjPB01j.Noteintheabovealgorithm,functionsLABW1,LABW2,andLABW3aregiveninAppendix 4.8.2 .Sinceeachnodehaspre-reservations,weneedtomodifythetwosets,namely,SRTiandSRRi,whichareusedinFA.Fori-connections,SRTiisdenedasfsk2S:sk=2TSi;sk=2RSi;sk=2[j2NBiRSjgandSRRiisdenedasfsk2S:sk=2TSi;sk=2RSi;sk=2[j2NBiTSjg;forordinaryconnections,SRTiisdenedasfsk2S:sk=2TSi;sk=2RSi;sk=2[j2NBiRSj;sk=2[j2NBiPRVjgandSRRiisdenedasfsk2

PAGE 84

71 S:sk=2TSi;sk=2RSi;sk=2[j2NBiTSj;sk=2[j2NBiPRVjg,wherePRVi,thepre-reservationsetatnodei,isdenedasfsk2S:skistaggedPRE RESERVEDg.4.4.2ImpactonRe-routingInQoSroutingalgorithms,analternatepathneedstobesearchedandusedwhentheprimarypathisbrokenduetomobility.Sinceeachi-nodepairhasaskedtheintermediatenodeintheinuenceareatopre-reservebandwidth,thisquadranglemaycontainseveralalternatepaths.Therefore,whentheprimarypathisnotavailable,theroutingalgorithmcaneasilyndanalternatepathwithsufcientbandwidth.Therefore,there-routingsuccessfulprobabilitywillbeincreasedaswellastheconnectionacceptanceratio,whichwillbeshowninSection 4.5 .4.5PerformanceEvaluationInthissection,theperformanceofourproposedschemeisevaluatedthroughexten-sivesimulations.OursimulationstudyiscarriedoutusingOPNETModeler8.0.Anadhocnetworkconsistingof80mobilenodesissimulatedina1600400m2area.Thetransmissionrangeofanodeis200m,soalongQoSconnectionmaytraversemorethan4hops.Theinitialpositionofeachmobilenodeisuniformlydistributedintheentirenet-work.Theirmobilityfollowsthewaypointmobilitymodel[95].Thatis,afterremainingstationaryforaperiodofpausetime,anoderandomlychoosesadestinationandstartstomovetowardit.Whenmoving,anodewillrandomlychooseaspeedfrom110m/s.Itcanbeseenthattheshorterthepausetime,thehigherthemobilityofeachnode.Thereare8i-nodes,whichare10%ofallnodes.ConnectionrequestsarrivefollowingPoissonprocess,eachrequiringonedataslotineachframe.Thedurationofeachconnectionisexponentiallydistributedwithmeanof60seconds.Ineachtimeframe,thedataslotis5msandthecontrolslotis0.1ms.Weassumethereare30controlslotsincontrolphaseand72dataslotsindataphase.Sotheframelengthis30*0.1+72*5=363ms.Wealsoassumeadatapacketcanbetransmittedinonedataslot.Theangleis45degreeunlessotherwisestated.Thesimulationdurationis900seconds.

PAGE 85

72 Forcomparisonpurpose,weconsiderthreeschemesinoursimulation.Therstoneistheschemewithoutanypre-reservation.Thesecondisourproposedscheme,whichmakesbandwidthpre-reservationinaquadrangleareafori-connections.Wealsocreatedthelastonethatadoptsbandwidthpre-reservation.However,itonlypre-reservestimeslotsalongtheshortestpathbetweenanytwoi-nodes.Notethatdependingonwhichroutingprotocolisused,theshortestpathinformationmayormaynotbeavailablewhenani-nodesendsoutthepre-reservationmessage.Inthesimulation,weassumethatthisschemeknowsthisinformation.Inthegurespresentedbelow,thesethreearedesignatedasNopre-reservation,QuadrangleandSP,respectively.Throughsimulationresults,weobservethat,whenthenetworktrafcloadislight,theperformancesofallthethreeschemesareveryclose.Thisisbecausethat,underlighttrafcconditions,thenetworkalwayshasenoughbandwidthtoaccommodatenewconnec-tions.Therefore,theconnectionblockingprobabilitiesareallverylow.However,itisevenmorecriticalfortheseschemestoperformwellundermediumtoheavytrafcload.Forthisreason,wemainlyfocusontheseschemes'performanceinarelativelyheavytrafcsituation,wheretheaverageconnectionarrivalrateateachnodeis0.2connection/second.Allresultspresentedhereareaveragedover10simulationruns.Fig. 4-6 showstheperformancesofthethreeschemesintermsofconnectionblockingprobability.Fori-connection,ourschemeprovidesthelowestblockingprobabilityamongallthreeschemes.Thisisexpectedaseachi-nodeissendingpre-reservationrequeststopre-reservesometimeslotsattheintermediatenodeswithintheinuencearea.SchemeSPisworsesinceitonlypre-reservesslotsatnodesontheshortestpath,whichmaybeun-availableduetomobility.Wealsoobservethatasthemobilityisincreased,theconnectionblockingprobabilityisslightlyreduced.Thisisnotunexpectedduetothefollowingfact.Whenmobilityincreases,theongoingconnectionmaybedropped,releasingthebandwidthoccupiedfornewlyincomingconnections.Therefore,theconnectionblockingprobability

PAGE 86

73 a b Figure4-6:Connectionblockingprobability.ai-connection.bordinaryconnection.

PAGE 87

74 a b Figure4-7:Re-routingsuccessprobability.ai-connection.bordinaryconnection.

PAGE 88

75 Figure4-8:Throughput Figure4-9:Averagenumberofpre-reservedtimeslots

PAGE 89

76 Figure4-10:Pre-reservationrequestsperreceivedpacket dropsinsteadofincreasing.Forordinaryconnections,theperformanceofschemeNopre-reservationisthebest.Thisisreasonablesincetheentirenetworkbandwidthisxed,asmorebandwidthisgiventoi-connection,lessbandwidthisleftforaccommodatingordi-naryconnections.However,thedifferencesareverysmall.Next,weinvestigatetheperformancesofthesethreeschemeswhenre-routingiscon-sidered.Fig. 4-7 presentsthere-routingsuccessfulprobabilitiesforbothtypesofcon-nections.Again,fori-connections,ourscheme,Quadrangleshowsthebestperformance,sinceitpre-reservesbandwidthinthequadrangle.Whenoneconnectionisdroppedduetomobility,itwillhavehighprobabilitytondanotherpathwithsufcientbandwidthinthequadrangle.SPisworseandNopre-reservationistheworst.Forordinaryconnections,theirperformancesareinanoppositeorderforthesamereasondescribedearlier.Fig. 4-8 illustratesthethroughputperformanceintermsofsuccessfullyreceivedpack-ets.Forallthreeschemes,throughputincreasesasmobilitydecreases.Thisshowsthenegativeeffectofmobilityonthethroughput:highermobilitycausesmoreongoingcon-nectionstobedroppedandinturnreducethethroughput.Ontheotherhand,weobservethatthroughputisclosetooneanotheramongthethreeschemes,whichmeans,although

PAGE 90

77 ourschemeadoptsbandwidthpre-reservationfori-connection,thenetworkthroughputisnotseriouslyaffected.Thenumberofpre-reservedtimeslotsinthenetworkispresentedinFig. 4-9 .Asexpected,ourschemepre-reservesmoretimeslotscomparedwiththeothertwo.However,consideringtheentirenetworkhasanumberoftimeslot72*80=5760,weonlyneedtopre-reserveabout2.4%.Fig. 4-9 alsoillustratesadesirablefeatureofthescheme:thenumberofpre-reservedtimeslotsisinsensitivetothechangesinmobility,whichensurestheschemedoesnotpre-reservetoomanytimeslotsinhighmobilityscenarios.Fig. 4-10 showsthecontroloverheadincurredbysendingpre-reservationrequests.Wecounttheoverheadasthenumberoftimesforwhicheachnode,eitheri-nodesorordi-narynodes,broadcaststhepre-reservationrequest.Weseethattheoverheadissmallanddecreaseswithmobility,showingtheadaptabilityofourschemetomobility.Withthelow-estblockingprobabilityandhighestre-routingsuccessfulprobabilityfori-connectionsandthesmallcontroloverheadincurred,ourscheme,i.e.,schemeQuadrangleisabletoprovidehighprioritycommunicationservicetoi-nodeswithQoSguaranteeatalowcost.SchemeSPseemstoperformfairlywell,however,itcannotbeusedwithsomeon-demandroutingprotocolsastheshortestpathinformationisnotavailableforbandwidthpre-reservation.Finally,westudytheimpactofangleontheperformanceofschemeQuadrangle,asshowninFig. 4-11 4-15 .InFig. 4-11 and 4-12 ,fori-connection,asincreases,theconnectionblockingprobabilityisdecreasedandre-routingsuccessfulprobabilityisincreased.Thisisexpected,sincelargerimplieslargerquadrangle,whichinturnwillhavemoretimeslotspre-reservedfori-connections.Forordinaryconnections,largehasanoppositeeffect.Itcanalsobeseenthatthesuperiorityofourproposedschemeisstillmaintained.Fig. 4-13 showsthethroughputisslightlyaffectedbytheincreasein.Also,moreoverheadisincurredtopropagatethepre-reservationrequestsandmoretimeslotsneedto

PAGE 91

78 bepre-reservedfori-connections,asincreases,asshowninFig. 4-14 and 4-15 .There-fore,canbeusedasadesignparameterdeterminingtheextenttowhichi-connectionshouldbegrantedhighpriority.Fig. 4-15 alsoveriestheinsensitivityoftheschemetothechangesinmobility,whichishighlydesirableasover-pre-reservationinthecaseofhighmobilityisavoided.4.6DiscussionsInthissection,wediscussseveraldesignissues.4.6.1MobilityThroughthewaythataquadrangle-shapedinuenceareaisformedbetweenanytwoi-nodes,ani-nodepaircanalwayspre-reservesometimeslotsintheintermediatenodesinbetweenit.Forintermediatenodes,whentheyarelocatedintheinuencearea,theywillalwayspre-reservesometimeslotsforpossiblecommunicationbetweenthetwoi-nodes;whentheymoveout,thepre-reservedtimeslotsforthei-nodepairwillbefreed,sincefuturepre-reservationrequestsfromthetwoi-nodesmaynotbereceivedforupdatingthetimersassociatedwiththetimeslots.Whenoneofthetwoi-nodesmoves,sowillthequadrangle,whichwillbegeneratedaccordingtothenewlocationofthetwoi-nodes.Therefore,thismechanismsuitswellwiththemobilenatureofadhocnetworks.Ofcourse,QoSisonlyfeasibleinthenetworkthatisstaticorwithlow-to-mediummobility;whenthenetworktopologychangestoofast,thereisnowaytosupportQoS[17,102].4.6.2TradeoffbetweenOverheadandPerformanceThecontroloverheadinbandwidthpre-reservationismainlythepropagationofpre-reservationrequests,whichconsumesnetworkbandwidth.SincethereareMi-nodesinthenetwork,theymayformuptoMM)]TJ/F15 11.955 Tf 12.551 0 Td[(1half-duplexcommunicationpairs,eachofwhichsendspre-reservingrequestaccordingitsownsendinginterval.Notethatweneedtodistinguishthedirectionofcommunicationforeachi-nodepair,sinceeachdirectionwilldemandaQoSconnectiontobesetupbeforesendingdata.Thus,theoverheaddependsonthenumberofi-nodesinnetwork,thegeographicallocationofeachi-node,aswellasthe

PAGE 92

79 a b Figure4-11:Connectionblockingprobability.ai-connection.bordinaryconnection.

PAGE 93

80 a b Figure4-12:Re-routingsuccessprobability.ai-connection.bordinaryconnection.

PAGE 94

81 Figure4-13:Throughput Figure4-14:Averagenumberofpre-reservedtimeslots

PAGE 95

82 Figure4-15:Pre-reservationrequestsperreceivedpacket nodedensityandthetrafcloadbetweeni-nodesduringTIMERupdate,whichareshownintheoverheadanalysisinSection 4.3 .Itisimportanttonotethatwhentwoi-nodesarefarawayfromeachotherandhavealargeamountofdatatotransmit,thequadrangleband-widthpre-reservationmayinvolvemanynodesifthenodedensityishigh.Consequently,thecontroloverheadmaybehigh.Inthiscase,wemayadoptthestrategyofhoplimit[50].Thatis,eachpre-reservationrequestcontainsaeldof“hoplimit”.Everytimetherequestisforwardedbyanode,thislimitisdecremented;ifitbecomeszero,therequestissimplydiscarded.Inthisway,welimitthenumberofintermediatenodesallowedtoforwardtherequestandhencereducetheoverhead.Moreover,wecanalsoadapttheangletothedistancebetweentwoi-nodesandthenodedensityinthenetwork.Nevertheless,theoverheadisrelativelysmallforthefollowingreasons.First,itisanticipatedonlyaverysmallpercentofnodesarei-nodesinadhocorsensornetworks.Second,aswecanseefromtheestimationofTIMERupdate,itadaptstonodemobility.Ifnodesarestaticormoveatalowtomediumspeed,theoverheadcanbefurtherreducedasTIMERupdatewillbeincreased.ItisworthnotingwhenestimatingTIMERupdate,wehavenotconsideredthetrafctypeortrafchistoryofeachi-node.Inreality,forthose

PAGE 96

83 i-nodesthattransmitdatalessfrequently,TIMERupdatecanbefurtheradaptedtoreduceoverhead.Weintendtofurtherexplorethisaspectinthefuture.4.7ConclusionSinceneitherroutingnorMACprotocolscaneffectivelyprovidehighprioritycom-municationswithQoSguaranteeforsomeimportantnodesinmobileadhocnetworks,weproposedanovelcross-layerapproachformobileadhocnetworksinthischapter.Werstproposealocation-awarebandwidthpre-reservationmechanismtopre-reservebandwidthforconnectionsbetweenimportantnodesbyutilizingeachnode'sgeographiclocationin-formation,therebyreducingthepotentialtransmissioncollisions.Then,alocation-awareforwardalgorithmLAFAisproposedtocalculateandreserveend-to-endbandwidthforsuchhighpriorityconnections.Inthisway,ourschemecannotonlyreducethetransmis-sioncollisionandhenceincreasingresourceutilization,butalsoadapttonetworktopologychangesduetomobility.Therefore,highprioritycommunicationservicesbetweenim-portantnodescanbeprovidedbythenetworkwithhighprobabilityandQoSguaranteeswithoutincurringtoomuchoverheadandseverelyblockingotherconnections,agoalthatcannotbeachievedwithoutthecollaborationbetweentheroutinglayerandtheMAClayerandtheuseoflocationinformation.Finally,extensivesimulationveriestheperformanceofourproposedscheme.4.8Appendix4.8.1DerivationofConnectionBlockingProbability1Forapathwithonlyonehop,itisobviousthattheconnectionblockingprobabilityPB1isequaltotheprobabilitythatLB=0,i.e.,PB1=q.2Forapathwithtwohops,saylinkAandB,rstweconsidertheconnectionaccep-tanceprobabilityPA2.WeknowthataconnectionrequestwillbeacceptedifbothlinkshaveLB>1,theprobabilityofwhichisP21.However,inthiscase,thereisonesce-nariothatwecannotbuildtheconnection,whichisequaltoq2=S.SotheprobabilityofacceptanceisP21)]TJ/F25 11.955 Tf 10.461 0 Td[(q2=S.Therefore,theblockingprobabilityisPB2=1)]TJ/F15 11.955 Tf 10.461 0 Td[(P21)]TJ/F26 7.97 Tf 11.657 5.699 Td[(q2 S.

PAGE 97

84 Figure4-16:Apathforderivation 3Forapathwiththreehops,wealsorstcalculatetheconnectionacceptanceprob-abilityPA3.TheprobabilitythatallthreehopshaveatleastonetimeslotavailableisP31.However,weneedtosubtractfromthisprobabilityseveralpossibilities.aAnytwohavethesametimeslotonly.Theprobabilityis3q2P1S)]TJ/F24 7.97 Tf 6.587 0 Td[(2q3 S2.bOnelinkhastwotimeslotsandtheothertwolinkseachhaveoneofthetwotimeslots.Theprobabilityis6qq2 S2.cTwolinkshavethesametwotimeslotsandtherestonelinkhasoneofthetwotimeslots.Theprobabilityis12q2q S2S)]TJ/F24 7.97 Tf 6.587 0 Td[(1.dAllthethreelinkshavethesametwotimeslots.Theprobabilityis4q3 S2S)]TJ/F24 7.97 Tf 6.586 0 Td[(12.Itcanbeshownthattherearenoothercaseswhereaconnectioncannotbeadmitted.Therefore,PA3istheprobabilitybysubtractingalltheseprobabilitiesfromP31.There-fore,theblockingprobabilityisPB3=1)]TJ/F15 11.955 Tf 11.774 0 Td[(P31)]TJ/F24 7.97 Tf 12.969 5.699 Td[(3q2P1S)]TJ/F24 7.97 Tf 6.587 0 Td[(2q3+6qq2 S2)]TJ/F24 7.97 Tf 12.969 5.699 Td[(12q2q S2S)]TJ/F24 7.97 Tf 6.586 0 Td[(1)]TJ/F24 7.97 Tf -414.074 -18.209 Td[(4q3 S2S)]TJ/F24 7.97 Tf 6.586 0 Td[(12.4.8.2FunctionsUsedinLAFAfunctionOUT=BWlocationnodegetnode'slocationxn;ynincelli,j;getsource'slocationxs;ys;ifjxn)]TJ/F25 11.955 Tf 11.955 0 Td[(xsj>jyn)]TJ/F25 11.955 Tf 11.955 0 Td[(ysjNtimesection=modj;Kh;OUT=halfofallthepre-reservedorfreetimeslotsintimesectionNtimesection;else

PAGE 98

85 Ntimesection=modi;Kv;OUT=halfofallthepre-reservedorfreetimeslotsintimesectionNtimesection;return;functionOUT=LABW1node;IN;nassertn6jINj;S=BWlocationnode;E1=SIN;E2=SIN;ifjE1j>nrandomlychoosenelementsfromINasOUT;return;elserandomlychoosen)-222(jE1jelementsfromE2asE3;OUT=E1E3;return;functionOUT2;OUT1=LABW2node2;node1;IN2;IN1C=IN1IN2;E1=IN1 IN2;E2=IN2 IN1;ifjE2j>jIN1jOUT1=IN1;OUT2=LABW1node2;E2;jIN1j;return;

PAGE 99

86 elseifjE1j>jIN2jOUT1=LABW1node1;E1;jIN2j;OUT2=IN2;return;elseT=floorjIN1[IN2j/2;C2=LABW1node2;C;T)-222(jE2j;C1=C C2;OUT1=LABW1node1;C1[E1;T;OUT2=LABW1node2;C2[E2;T;return;functionOUT3;OUT2;OUT1=LABW3node3;node2;node1;IN3;IN2;IN1assertjIN3j=jIN2j&&IN2IN3=;C21=IN2IN1;C31=IN3IN1;E1=IN1 C21 C31;E2=IN2 C21;E3=IN3 C31;ifjE1j>jIN2jOUT1=LABW1node1;E1;jIN2j;OUT2=IN2;OUT3=IN3;return;elseifjE3j>jLABW2node2;node1;IN2;IN1jOUT2;OUT1=LABW2node2;node1;IN2;IN1;

PAGE 100

87 OUT3=LABW1node3;E3;jOUT1j;return;elseifjE2j>jLABW2node3;node1;IN3;IN1jOUT3;OUT1=LABW2node3;node1;IN3;IN1;OUT2=LABW1node2;E2;jOUT1j;return;elseT=floorjIN1[IN2[IN3j/3;C331=LABW1node3;C31;T)-222(jE3j;C131=C31 C331;C221=LABW1node2;C21;T)-222(jE2j;C121=C21 C221;OUT1=LABW1node1;E1[C121[C131;T;OUT2=E2[C221;OUT3=E3[C331;return;

PAGE 101

CHAPTER5QOSSUPPORTINIEEE802.11EWIRELESSLANS5.1IntroductionWirelessLANsbasedontheIEEE802.11DistributedCoordinationFunctionDCF[44]havebeenwidelyusedinrecentyearsduetotheirsimpledeploymentandlowcost.SincethecurrentDCFcanonlysupportbestefforttrafc,theIEEE802.11TaskGroupErecentlyproposedanewcontention-basedchannelaccessmethodcalledEnhancedDis-tributedChannelAccessEDCAintheIEEE802.11estandard[22,45,70].Despitepro-vidingprioritizedqualityofserviceQoS,theEDCAstillcannotsupportstrictQoSforreal-timeapplicationslikevoiceandvideo[22].ThischapterstudieshowtheEDCAcanbeenhancedtomeetthischallenge.Inourpreviouswork[97],wehavefoundthatitisintheunsaturatedcasethatthe802.11achievesthemaximumthroughputandsmalldelaybecauseofthelowcollisionprobability;bycontrast,whenworkinginthesaturatedcase,itsuffersfromalargecolli-sionprobability,leadingtolowthroughputandexcessivelylongdelay.Motivatedbythisdiscovery,weaimtotunethenetworktoworkintheunsaturatedcaseinordertosupportstrictdelayrequirementsofreal-timeservices.However,effectivetuningisnoteasytoachievegiventhatthe802.11EDCAisinnaturecontention-basedanddistributed,therebymakingithardtocharacterizeactualtrafcconditionsinthenetwork.Toovercomethesedifculties,weproposetwocalladmissioncontrolschemesandaratecontrolschemethatfunctionbasedonthenoveluseofthechannelbusynessratio.ItisimportanttonotethatwhiletheIEEE802.11erecommendstheuseofcalladmissioncontrol,noalgorithmisspecied.Inaddition,theIEEE802.11ehasnotaddressedanyissueonratecontrol. 88

PAGE 102

89 Inthischapter,wemakethefollowingcontributions.First,webuildananalyticalmodeltoderiveanaveragedelayestimateforthetrafcofdifferentprioritiesintheunsat-urated802.11ewirelessLAN.Wealsoderivetheupperboundsforboththedelaymeananddelayvariationintheunsaturatedcase.Theanalyticalresultsshowthatifthetrafcisproperlyregulated,the802.11eWLANiscapableofsupportingQoSrequirementsforthereal-timetrafc.Thedelayestimateisthenusedinthecalladmissioncontrol.Second,sincethechannelbusynessratioiseasytoobtainandcanaccuratelyrepresentthenetworkstatus,itprovidesaverysuitablecontrolvariableforboththecalladmissioncontrolandtheratecontrol.Asaresult,thecalladmissionandratecontrolschemesaresimpleandeffective.Theadmissioncontroloverthereal-timetrafcguaranteesitsQoSrequirementscanbesatised,andtheratecontrolallowsthebestefforttrafctomakefulluseoftheresidualchannelcapacitywhilenotaffectingQoSofthereal-timetrafc.Theremainderofthischapterisorganizedasfollows.InSection 5.2 ,wegiveabriefintroductionoftheIEEE802.11eEDCA.WerevisittheQoSrequirementsofdifferenttypesofservicesinSection 5.3 .ThedelayestimateisanalyzedandveriedinSection 5.4 .ThedelayupperboundsarederivedandveriedinSection 5.5 .WethenpresentourproposedschemesinSection 5.6 .InSection 5.7 ,theperformanceisevaluatedthroughcomprehensivesimulationstudies.Finally,Section 5.8 concludesthischapter.5.2OperationsoftheIEEE802.11eThelegacyIEEE802.11DCFDistributedCoordinationFunctionisbasedoncarriersensemultipleaccesswithcollisionavoidanceCSMA/CA[52].Beforestartingatrans-mission,eachnodeperformsabackoffprocedure,withthebackofftimeruniformlychosenfrom[0,CW-1]intermsoftimeslots,whereCWisthecurrentcontentionwindow.Ifthechannelisdeterminedtobeidleforabackoffslot,thebackofftimerisdecreasedbyone.Otherwise,itissuspended.Whenthebackofftimerreacheszero,thenodetrans-mitsaDATApacket.Ifthereceiversuccessfullyreceivesthepacket,itacknowledgesthepacketbysendinganacknowledgmentACKafteranintervalcalledshortinter-frame

PAGE 103

90 spaceSIFS.Sothisisatwo-wayDATA/ACKhandshake.Ifnoacknowledgmentisre-ceivedwithinaspeciedperiod,thepacketisconsideredlost;sothetransmitterwilldoublethesizeofCWandchooseanewbackofftimer,andstarttheaboveprocessagain.Whenthetransmissionofapacketfailsforamaximumnumberoftimes,thepacketisdropped.Toreducecollisionscausedbyhiddenterminalsandimprovechannelefciencyforlongdatatransmissions[6],theRTS/CTSrequesttosend/cleartosendmechanismisemployed.Therefore,afour-wayRTS/CTS/DATA/ACKhandshakeisusedforapackettransmission,asshowninFig. 5-1 .BasedontheDCF,theEDCAismeanttoprovideprioritizedservices.IntheEDCA,trafcofdifferentprioritiesisassignedtooneoffourtransmitqueues,whichrespectivelycorrespondtofouraccesscategoriesACs,asshowninFig. 5-2 .Eachqueuetransmitspacketswithanindependentchannelaccessfunction,whichimplementstheprioritizedchannelcontentionalgorithm.Inotherwords,differentchannelaccessfunctionsusedif-ferentcontentionwindowstheminimumandmaximumcontentionwindowsandbackofftimers.Specically,forACii=0;1;2;3,theinitialbackoffwindowsizeisCWmin[i],themaximumbackoffwindowsizeisCWmax[i],andthearbitrationinter-framespaceisAIFS[i].For06iCWmin[j],CWmax[i]>CWmax[j],andAIFS[i]>AIFS[j].Notethatintheaboveinequalities,atleastonemustbestrictly“notequalto”.Thus,weseethattheACwithahigherlevelhasahigherpriority,sinceithasahigherprobabilitytogainchannelaccess.Whenanapplicationisadmitted,itwillbeattachedwithaspecicpriorityandassignedtothecorrespondingqueue,whichperformslikeasinglenodeintheDCF.5.3QoSRequirementsforMultimediaServicesAstheInternetexpandsitssupportedtrafcfrombesteffortdatatoavarietyofmul-timediaservices,includingvideoconferencing,voiceoverIPVoIP,streamingaudioandvideo,WWW,e-mail,andletransfer,etc.,QoSprovisioninghasbecomeanimportantissue.ThecommonlyacceptedQoSmetricsmainlyincludebandwidth,delay,delayjitter

PAGE 104

91 Figure5-1:RTS/CTS/DATA/ACKfour-wayhandshake Figure5-2:802.11earchitecture

PAGE 105

92 delayvariation,packetlossrateorbiterrorrate.AccordingtotheirQoSrequirements,currentmultimediaservicescanbegroupedintothreeclasses:real-time,streaming,andnon-real-timeorbesteffort.Real-time:Real-timetrafchasstringentrequirementsindelayanddelayjitter,whichisnecessaryforinteractivecommunicationslikeVoIPandvideoconferencing.Theonewaytransmissiondelayshouldbepreferablylessthan150ms,andmustbelessthan400ms[46,47].However,itisnotverysensitivetopacketlossrate.Typically,alossrateof1%isacceptableforreal-timevideowithrate16384Kbpsandalossrateof3%forreal-timeaudiowithrate464Kbps.Becausedelayedpacketsarenottolerable,retransmissionoflostpacketsisnotuseful.Thus,UDPisusedtotransmitreal-timetrafc.Streaming:Streamingaudioorvideobelongstothisclass.Comparedwithreal-timetrafc,itislesssensitivetodelayordelayjitter.Attheexpenseofincreaseddelay,playoutbufferorjitterbuffercanbeusedtocompensatefordelayjitterintherangeof2050ms.Acceptabledelaymaybeupto10seconds[47],whilethepacketlossrateisabout1%.StreamingtrafcisnormallytransportedviaUDP,althougharetransmissionstrategycanbeaddedintheapplicationlayer.Non-real-time:Non-real-timeservicescomprisee-mail,letransfer,andwebbrows-ing.Mostnon-real-timeservicesaretoleranttodelayrangingfromsecondstominutesorevenhours.However,thedatatobetransferredhastobereceivederror-free,whichmeansreliabletransmissionisrequired.Sonon-real-timetrafcistransportedwithTCP.5.4DelayEstimationoftheIEEE802.11eThissectionfocusesonthedelayanalysisoftheIEEE802.11eEDCAintheunsat-uratedcase,wherethecollisionprobabilityissmallandpacketsdonotaccumulateinthetransmitqueue.WeconsiderthecasewheretheRTS/CTSmechanismisused,althoughouranalysiscanbeextendedtothebasicaccessmechanism.Thechannelisassumedtobeperfect,i.e.,nopacketislostduetochannelfading.InaccordancewiththeIEEE802.11eprotocol,thereareatmostfourtransmitqueuesineachactivenodes.Leti=0,1,2,3

PAGE 106

93 denotethepriorityofthefourqueues,withi=3beingthehighestpriority.Also,letnide-notethenumberofqueuesofpriorityiinthenetwork.Eachpriorityqueueistreatedasanindependentnode.Next,werstdistinguishbetweenthesaturatedcaseandtheunsaturatedcase.5.4.1SaturatedCasevs.UnsaturatedCaseBysaturation,wemeanthenetworkisoverloadedandeachnodealwayshaspacketstotransmit.Inotherwords,thetransmitqueueateachnodeisalwaysnotempty.Asamatteroffact,allthenodeswillkeepcontendingforthechannel,leadingtoahighlevelofpacketcollisionsespeciallyinthepresenceofalargenumberofnodes.Asaresult,thepacketcannotgetthroughandthetransmitqueuewillbuildupandcausepacketlossesduetobufferoverow.Onthecontrary,ifthenetworkworksintheunsaturatedcase,notallthenodesarecontendingforthechannelatthesametime.Therefore,thepacketcollisionislowandpacketsgettransmittedquickly.Also,thequeueisnotalwaysnonempty.Inthiscase,weneedtoexplicitlyconsiderthispossibilitywhenbuildingtheanalyticalmodel.Whilethesaturatedthroughputwasshownbestablewhenthenetworkisoverloaded[7],wehavedemonstratedthatthemaximumthroughputisachievedintheunsaturatedcaseandthedifferencebecomesmorevisiblewhenthenumberofactivenodesisfairlylarge[97].Furthermore,inthesaturatedcase,thepacketcollisionprobabilitygiventhenumberofnodesinthenetworkisthehighest,leadingtolongMACservicetime.Also,thequeuebuild-upresultsinlongqueueingdelay.Clearly,thesaturatedcaseisundesirabletosupportreal-timetrafcthathasstrictdelayrequirement.5.4.2MarkovChainModelfortheIEEE802.11eConsiderapriorityiqueue.Wedenebi;tasastochasticprocessrepresentingthevalueofthebackoffcounterattimet,andsi;tasastochasticprocessrepresentingthebackoffstagej,where06j6.Hereisthemaximumnumberofretransmissionsandisequalto7accordingtothestandard.LetCWi;minandCWi;maxbetheminimumandmaximumcontentionwindowforpriorityi,thenCWi;max=2mCWi;min,wheremisthe

PAGE 107

94 maximumnumberofthestagesallowedintheexponentialbackoffprocedureandisequalto5accordingtothestandard.Forconvenience,wedeneWi;0=CWi;min.Therefore,atdifferentbackoffstagej2;,thecontentionwindowsizeWi;j=8><>:2jWi;0if06j6m2mWi;0ifm>>>>>><>>>>>>>:Pfj;kjj;k+1g=1k2;Wi;j)]TJ/F15 11.955 Tf 11.955 0 Td[(2j2;Pf0;kjj;0g=)]TJ/F25 11.955 Tf 11.955 0 Td[(pi=Wi;0k2;Wi;0)]TJ/F15 11.955 Tf 11.955 0 Td[(1j2;)]TJ/F15 11.955 Tf 11.955 0 Td[(1Pfj;kjj)]TJ/F15 11.955 Tf 11.956 0 Td[(1;0g=pi=Wi;jk2;Wi;j)]TJ/F15 11.955 Tf 11.955 0 Td[(1j2;Pf0;kj;0g=1=Wi;0k2;Wi;0)]TJ/F15 11.955 Tf 11.955 0 Td[(1;.2wherePfj1;k1jj0;k0g=Pfsi;t+1=j1;bi;t+1=k1jsi;t=j0;bi;t=k0g.Lettingbj;k=limt!1Pfsi;t=j;bi;t=kgbethestationarydistributionofthechain,wehavebj)]TJ/F24 7.97 Tf 6.587 0 Td[(1;0pi=bj;00<>:)]TJ/F25 11.955 Tf 11.955 0 Td[(pi)]TJ/F24 7.97 Tf 6.587 0 Td[(1Pl=0bl;0+b;0j=0pibj)]TJ/F24 7.97 Tf 6.586 0 Td[(1;00
PAGE 108

95 GivenEquation 5-4 ,Equation 5-5 canbesimpliedasbj;k=Wi;j)]TJ/F25 11.955 Tf 11.956 0 Td[(k Wi;jbj;006j6:.6Furthermore,byusingthenormalizationcondition1=Xj=0Wi;j)]TJ/F24 7.97 Tf 6.587 0 Td[(1Xk=0bj;k=Xj=0bj;0Wi;j)]TJ/F24 7.97 Tf 6.586 0 Td[(1Xk=0Wi;j)]TJ/F25 11.955 Tf 11.955 0 Td[(k Wi;j=Xj=0bj;0Wi;j)]TJ/F15 11.955 Tf 11.956 0 Td[(1 2;.7wehaveb0;0=8><>:21)]TJ/F24 7.97 Tf 6.586 0 Td[(2pi)]TJ/F26 7.97 Tf 6.586 0 Td[(pi Wi;0)]TJ/F24 7.97 Tf 6.587 0 Td[(pi+1)]TJ/F26 7.97 Tf 6.586 0 Td[(pi+)]TJ/F24 7.97 Tf 6.586 0 Td[(2pi)]TJ/F26 7.97 Tf 6.586 0 Td[(p+1i6m21)]TJ/F24 7.97 Tf 6.587 0 Td[(2pi)]TJ/F26 7.97 Tf 6.587 0 Td[(pi Wi;0)]TJ/F24 7.97 Tf 6.587 0 Td[(pim+1)]TJ/F26 7.97 Tf 6.586 0 Td[(pi+)]TJ/F24 7.97 Tf 6.586 0 Td[(2pi)]TJ/F26 7.97 Tf 6.586 0 Td[(p+1i+Wi;02mpm+1i)]TJ/F24 7.97 Tf 6.587 0 Td[(2pi)]TJ/F26 7.97 Tf 6.586 0 Td[(p)]TJ/F27 5.978 Tf 5.756 0 Td[(mi>m:.8Therefore,theprobabilitythatanodeofpriorityitransmitsinarandomslot,giventhatthequeueisnotempty,denotedbyi,isi=Xj=0bj;0=8><>:2)]TJ/F24 7.97 Tf 6.586 0 Td[(2pi)]TJ/F26 7.97 Tf 6.586 0 Td[(p+1i Wi;0)]TJ/F24 7.97 Tf 6.587 0 Td[(pi+1)]TJ/F26 7.97 Tf 6.586 0 Td[(pi+)]TJ/F24 7.97 Tf 6.586 0 Td[(2pi)]TJ/F26 7.97 Tf 6.586 0 Td[(p+1i6m21)]TJ/F24 7.97 Tf 6.586 0 Td[(2pi)]TJ/F26 7.97 Tf 6.586 0 Td[(p+1i Wi;0)]TJ/F24 7.97 Tf 6.587 0 Td[(pim+1)]TJ/F26 7.97 Tf 6.586 0 Td[(pi+)]TJ/F24 7.97 Tf 6.586 0 Td[(2pi)]TJ/F26 7.97 Tf 6.586 0 Td[(p+1i+Wi;02mpm+1i)]TJ/F24 7.97 Tf 6.587 0 Td[(2pi)]TJ/F26 7.97 Tf 6.587 0 Td[(p)]TJ/F27 5.978 Tf 5.757 0 Td[(mi>m:.9Onceiisknown,picanbeobtainedbypi=1)]TJ/F26 7.97 Tf 12.741 14.944 Td[(i)]TJ/F24 7.97 Tf 6.587 0 Td[(1Yl=0)]TJ/F15 11.955 Tf 11.955 0 Td[()]TJ/F25 11.955 Tf 11.955 0 Td[(Pl;0lnl)]TJ/F15 11.955 Tf 11.955 0 Td[()]TJ/F25 11.955 Tf 11.955 0 Td[(Pi;0ini)]TJ/F24 7.97 Tf 6.586 0 Td[(13Yl=i+1)]TJ/F15 11.955 Tf 11.955 0 Td[()]TJ/F25 11.955 Tf 11.955 0 Td[(Pl;0lnl;.10wherePi;0istheprobabilitythatapriorityiqueueisempty.5.4.3G/M/1QueueModeltoEstimateMeanDelayWemodelapriorityiqueueasaqueueingsysteminordertoderivetheprobabilityPi;0.Inthequeueingsystem,thepacketarrivalprocessisdeterminedbythetrafcchar-acteristicsofapriorityiapplicationthatemitspacketstotheMAClayer.Withoutlossofgenerality,weassumethepacketinterarrivaltimeisgenerallydistributed.Theservicetimeofthequeueingsystem,whichisalsocalledtheMAClayerservicetime,isthetimeperiodfromtheinstantthatapacketmovestotheheadofthequeueandbeginstobeservicedbytheMAClayertotheinstantthatitissuccessfullytransmittedordroppedafterallthe

PAGE 109

96 Figure5-3:Markovchainforthe802.11ebackoffprocedure timesofretransmissionsfail.Asshowninourpriorwork[99],wehavederivedtheprobabilitygeneratingfunctionPGFoftheMACservicetimeandhenceitsprobabilitydistribution.WealsodemonstratedthattheMAClayerservicetimecanbewellapprox-imatedwiththeexponentialdistribution.Thus,forapriorityiqueue,theservicetimeisexponentiallydistributedwithmean1=i,where1=icanbeobtainedfromthePGFandexpressedasafunctionofthecollisionprobabilitypi,i,andPi;0.ThequeueingsystemnowcanbecharacterizedbyaG/M/1queueingmodel.Meanwhile,theprobabilityPi;0canbeobtainedasPi;0=1)]TJ/F25 11.955 Tf 13.258 8.088 Td[(i i;.11whereiistheaveragepacketarrivalrateforpriorityitrafcandisknowninthetrafcspecication.Thus,givennii=0,1,2,3isknown,wecanusenumericalmethodstosolvethenonlinearsystemrepresentedbyEquations 5-9 5-10 5-30 andobtaintheunknownparameterspi,i,andPi;0.Notethatalltheseparameterslieintheinterval;1.Oncetheseparametersbecomeknown,ii=0,1,2,3isalsosolved.Nowwecanobtaintheaveragedelayexperiencedbyapacketofpriorityi.IntheG/M/1system,iftheprobabilitydistributionfunctionPDFofthepacketinterarrivaltimeisdenotedbyAt

PAGE 110

97 andthecorrespondingLaplacetransformisdenotedbyAs,theaveragesystemtime,i.e.,theaveragepacketdelayTi,thatapacketexperiencescanbeexpressedas[56]Ti=1 i)]TJ/F25 11.955 Tf 11.955 0 Td[(i;.12whereiistheuniquerootofi=Ai)]TJ/F25 11.955 Tf 11.955 0 Td[(ii.13intherangeof0
PAGE 111

98 whereWiistheaveragewaitingtime,i=i=iisthetrafcintensity,and2Aiand2Biare,respectively,thevariancesoftheinterarrivaltimeandservicetime.Thisboundgetsbetterasi!1;however,thisisnotthecasefortheunsaturatedcasewhereiisrelativelysmallandnoqueuebuildsup.Hence,weuseaweightingfactor,2i2Ai+2Bi 2Ai+2Bi,toscaledowntheboundtoachieveagoodestimate[35].Then,theaveragewaitingtimeforapacketinthequeuecanbeapproximatedascWi=i2Ai+2Bi 21)]TJ/F25 11.955 Tf 11.956 0 Td[(i2i2Ai+2Bi 2Ai+2Bi=i2i2Ai+2i2Bi 2i)]TJ/F25 11.955 Tf 11.955 0 Td[(i:.15Then,theaveragepacketdelaythatapacketexperiencesisthesumoftheaveragewaitingtimeinthequeueandtheaverageMACservicetime1=i:Ti=i2i2Ai+2Bi 21)]TJ/F25 11.955 Tf 11.955 0 Td[(i+1=i:.165.4.5ModelValidationInthissection,wevalidateouranalyticalresultsthroughsimulations.Weconsidertwokindsofreal-timetrafc,namelyVBRvoicetrafcandCBRvideotrafc.ItisknownthattheinterarrivaltimeforCBRtrafcisdeterministic.However,itismorecomplicatedforVBRtrafcthatcanbemodeledwithanon/offtrafcmodel[24,39].Inthismodel,activeperiods,whichareexponentiallydistributedwithmeanTon,alter-natewithidleperiods,whichareexponentiallydistributedwithmeanToff,accordingtoacontinuous-timeMarkovchain.Duringanactiveperiod,packetsaregeneratedatregularperiodsofTp.WeassumethatanactiveperiodtypicallyconsistsofmultipleconsecutivepacketsTp=Ton<1,whichisthecaseofpracticalinterest.Theon-offsourcecanthusbeconvenientlyviewedasarenewalprocesswithinterarrivaltimedistributiongivenbyAt=[)]TJ/F25 11.955 Tf 15.635 8.087 Td[(Tp Ton+Tp Ton)]TJ/F25 11.955 Tf 11.955 0 Td[(e)]TJ/F18 5.978 Tf 7.782 4.689 Td[(t)]TJ/F27 5.978 Tf 5.757 0 Td[(Tp Toff]Ut)]TJ/F25 11.955 Tf 11.955 0 Td[(Tp;.17

PAGE 112

99 whereUtistheunitstepfunction,andtheLaplacetransformAs=Z10e)]TJ/F26 7.97 Tf 6.586 0 Td[(stdAt=[1)]TJ/F25 11.955 Tf 15.635 8.088 Td[(Tp Ton+Tp Ton+sToff]e)]TJ/F26 7.97 Tf 6.587 0 Td[(sTp.18withthepeakpacketarrivalrate1=Tpandthemeanpacketarrivalrate1=Tp+TpToff=Ton.ForCBRtrafc,wesimplyletTon!1andToff!0inEquations 5-17 5-18 andobtainAt=Ut)]TJ/F25 11.955 Tf 11.955 0 Td[(Tp.19As=e)]TJ/F26 7.97 Tf 6.587 0 Td[(sTp;.20whereTpnowbecomestheconstantpacketinterarrivaltime.Wesimulatean802.11ebasedwirelessLANwith100mobilenodes.Allnodesarewithinthetransmissionrangeofoneanother.Thechannelrateis2Mb/s.Thetrafcparametersarelistedasfollows.VoiceTrafcVBR:ThevoicetrafcismodeledasVBRusinganon/offsourcewithexponentiallydistributedonandoffperiodsof300msaverageeach.Trafcisgeneratedduringtheonperiodsatarateof32kb/swithapacketsizeof160bytes,thustheinter-packettimeis40ms.VideoTrafcCBR:ThevideotrafcismodeledasCBRtrafcwitharateof64kb/swithapacketsizeof1000bytes,thustheinter-packettimeis125ms.Accordingtothe802.11e[45],weassignthevideotrafctoAC2andthevoicetrafctoAC3.Twosetofparametersareusedtoverifytheanalysis.Insettinga,AIFS[2]=60s,AIFS[3]=50s,W2;0=32,andW3;0=16;insettingb,AIFS[2]=75s,AIFS[3]=50s,W2;0=64,andW3;0=16.Itcanbeseenthatitbecomesharderforthevideotrafctogainchannelaccessinsettingbthaninsettinga.Inbothsettings,thenumberofqueuesforeachtrafcclassisequal,i.e.,n2=n3.Notethatthenetworkworksintheunsaturatedcase.Fig. 5-4a and 5-4b respectivelyillustratetheaveragedelayasafunctionofthetotalnumberofows,i.e.,n2+n3forbothsettings.Ineachgure,boththeanalytical

PAGE 113

100 andsimulationresultsarepresented.Severalobservationsaremadehere.First,asthenumberofowsincreases,foreithertheanalyticalorsimulationresults,thedelaysforbothtrafcclassesincrease.Thereasonisasfollows.Intheunsaturatedcase,thecollisionisnotseverewiththecollisionprobabilitylessthan0.1,andthequeuedoesnotbuildup.Asaresult,thequeueingdelayissmallandtheMAClayerservicetimedominatesthedelay.Whenthenumberofcompetingowsincrease,thecollisionincreasesandsodoestheMAClayerservicetime.Second,thedelayforthevoicetrafcismuchsmallerthanthatforthevideotrafc,whichisconsistentwiththefactthatthevoicetrafchasahigherprioritythanthevideotrafcintermsofchannelaccess.Especially,asexpected,thedelayforthevoicetrafcinsettingbissmallerthanthatinsettingaandthedelayforthevideotrafcinsettingbgreaterthanthatinsettinga.Third,theG/M/1andtheG/G/1modelsdeliververyclosedelayresults,bothgreaterthanthesimulationdelays,indicatinginpracticetheycanprovidetheupperboundsfortheaveragedelay.Asshownlater,weusethemintheproposedcalladmissioncontrolscheme.Wealsoobservethatasthenumberofowsincreases,thegapsbetweenthesimulationandanalyticalresultsbecomelarger.Nevertheless,ouranalyticalresultscanserveasupperboundsoftheaveragedelay.Finally,itisimportanttopointoutthatwhenwekeepthenetworkworkingintheunsaturatedcase,thedelaysforbothtrafcclassesaresufcientlysmalltosatisfytheirQoSrequirementsasspeciedin[46,47],wheretheonewaytransmissiondelayforinteractivecommunicationslikeVoIPandvideoconferencingshouldbepreferablylessthan150ms,andmustbelessthan400ms.5.5DerivationoftheUpperBoundsInthissection,wederivetheupperboundsfordelaymeansandvariationsofdifferenttypesoftrafcfortheIEEE802.11eEDCAintheunsaturatedcase.Weusethesameassumptionsandnotationsasintheprevioussection.

PAGE 114

101 a bFigure5-4:Averagedelay.aAIFS[2]=60s,AIFS[3]=50s,W2;0=32,andW3;0=16.bAIFS[2]=75s,AIFS[3]=50s,W2;0=64,andW3;0=16.

PAGE 115

102 5.5.1ProbabilityDistributionofMACServiceTimeFollowingthesameproceduredescribedinsubsection 5.4.2 ,wecansolvetheMarkovchainforanACiasshowninFig. 5-3 .Next,wederivetheprobabilitydistributionoftheMACservicetime,denotedbyTsi,forapriorityiACusingthemethod[99].Sincethereexistsaone-to-onecorrespondencebetweentheprobabilitygeneratingfunctionPGF,i.e.,theZ-transformoftheprobabilitydistributionfunctiondenotedbyGiZ,andtheprobabilitydistributionoftheMACservicetime,wechoosetocalculatethePGFrst.AsdescribedinSection 5.4.2 ,giventhecollisionprobabilitypi,wecanmodelthebackoffprocesswithaMarkovchain.Inthischainmodel,ifthePGFofthestatetransfertimebetweentwostatesisknown,thenwecanobtainthePGFoftheMACservicetime.Specically,ifwedenotethePGFsofacollisionperiod,asuccessfultransmissionperiod,andthedecrementofthebackofftimerasCiZ,SiZ,andDiZ,respectively,wecantransformthechainmodelinFig. 5-3 intothePGFdiagramshowninFig. 5-5 .NowwedescribehowtocalculateCiZ,SiZ,andDiZ.SincethecollisionperiodassociatedwithapriorityiACisRTS+SIFS+CTS+AIFS[i],wecanobtainCiZasCiZ=ZRTS+SIFS+CTS+AIFS[i].21Similarly,wecanobtainSiZasSiZ=ZRTS+CTS+3SIFS+DATAi+ACK+AIFS[i].22whereDATAiistheaveragepackettransmissiontimeinasuccessfultransmissionperiodforACi.ps=3Pj=0;j6=i264DATAjnj)]TJ/F25 11.955 Tf 11.955 0 Td[(Pj;0j)]TJ/F15 11.955 Tf 11.956 0 Td[()]TJ/F25 11.955 Tf 11.955 0 Td[(Pj;0jnj)]TJ/F24 7.97 Tf 6.586 0 Td[(1)]TJ/F15 11.955 Tf 11.956 0 Td[()]TJ/F25 11.955 Tf 11.955 0 Td[(Pi;0ini)]TJ/F24 7.97 Tf 6.587 0 Td[(13Qk=0;k6=i;j)]TJ/F15 11.955 Tf 11.955 0 Td[()]TJ/F25 11.955 Tf 11.955 0 Td[(Pk;0knk375+ni)]TJ/F15 11.955 Tf 11.955 0 Td[(1)]TJ/F25 11.955 Tf 11.955 0 Td[(Pi;0i)]TJ/F15 11.955 Tf 11.955 0 Td[()]TJ/F25 11.955 Tf 11.955 0 Td[(Pi;0ini)]TJ/F24 7.97 Tf 6.586 0 Td[(23Qk=0;k6=i)]TJ/F15 11.955 Tf 11.955 0 Td[()]TJ/F25 11.955 Tf 11.955 0 Td[(Pk;0knk.23

PAGE 116

103 Figure5-5:PGFdiagramforthebackoff

PAGE 117

104 DATA=1 ps3Pj=0;j6=i264DATAjnj)]TJ/F25 11.955 Tf 11.955 0 Td[(Pj;0j)]TJ/F15 11.955 Tf 11.955 0 Td[()]TJ/F25 11.955 Tf 11.955 0 Td[(Pj;0jnj)]TJ/F24 7.97 Tf 6.586 0 Td[(1)]TJ/F15 11.955 Tf 11.955 0 Td[()]TJ/F25 11.955 Tf 11.955 0 Td[(Pi;0ini)]TJ/F24 7.97 Tf 6.586 0 Td[(13Qk=0;k6=i;j)]TJ/F15 11.955 Tf 11.955 0 Td[()]TJ/F25 11.955 Tf 11.955 0 Td[(Pk;0knk375+1 psDATAini)]TJ/F15 11.955 Tf 11.955 0 Td[(1)]TJ/F25 11.955 Tf 11.955 0 Td[(Pi;0i)]TJ/F15 11.955 Tf 11.955 0 Td[()]TJ/F25 11.955 Tf 11.955 0 Td[(Pi;0ini)]TJ/F24 7.97 Tf 6.586 0 Td[(23Qk=0;k6=i)]TJ/F15 11.955 Tf 11.955 0 Td[()]TJ/F25 11.955 Tf 11.956 0 Td[(Pk;0knk.24TocalculateDiZ,weneedtoexaminehowthebackofftimervaries.Afteranidletimeslot,denotedby,itwilldecreaseby1;afteracollisionperiodorsuccessfultrans-missionperiod,itwillstayunchanged.Thissuccessfultransmissionperiodcanbecharac-terizedbyS0iZ:S0iZ=ZRTS+CTS+3SIFS+ DATA+ACK+AIFS[i].25where DATAistheaveragepackettransmissiontimeinasuccessfultransmissionperiodforalltheACsexcepttheACiunderconsideration.DenotebypstheprobabilitythatamongthoseACs,thereisoneACthattransmitssuccessfully.Clearly,itcanbeobtainedasshowninEquation 5-23 .Then, DATAcanbeobtainedasshowninEquation 5-24 .Then,DiZcanbeobtainedasDiZ=)]TJ/F25 11.955 Tf 11.955 0 Td[(piZ 1)]TJ/F25 11.955 Tf 11.955 0 Td[(psS0iZ)]TJ/F15 11.955 Tf 11.955 0 Td[(pi)]TJ/F25 11.955 Tf 11.955 0 Td[(psCiZ.26WethencanusetheMasonformulatosolveforthetransferfunctionfromthe“start”pointtothe“end”point,i.e.,thePGFoftheMACservicetimeGiZasfollows.GiZ=)]TJ/F11 10.909 Tf 10.909 0 Td[(piSiZPj=0"[piCiZ]jjQk=0HkZ#+[piCiZ]+1Qk=0HkZ.27whereHkZ=8>>><>>>:2kWi;0)]TJ/F24 7.97 Tf 6.587 0 Td[(1Pl=0DliZ 2kWi;0;6k6m2mWi;0)]TJ/F24 7.97 Tf 6.587 0 Td[(1Pl=0DliZ 2mWi;0;m6k6.28

PAGE 118

105 NotethatGiZisafunctionofthecollisionprobabilitypi.OnceweobtainGiZ,boththemeanandvariationoftheMACservicetimecanbederivedbytakingthederivativewithrespecttoZ:8><>:1 i=G0iZjZ=12i=G00iZjZ=1+G0iZjZ=1)]TJ/F15 11.955 Tf 11.955 0 Td[(G0iZjZ=12.29Meanwhile,theprobabilityPi;0canbeobtainedasPi;0=1)]TJ/F25 11.955 Tf 13.258 8.088 Td[(i i.30whereiistheaveragepacketarrivalrateforpriorityitrafcandisknowninthetrafcspecication.Thus,givennii=0,1,2,3isknown,wecanusenumericalmethodstosolvethenonlinearsystemrepresentedbyEquations 5-9 , 5-10 and 5-30 ,andobtaintheunknownparameterspi,i,andPi;0.Notethatalltheseparameterslieintheinterval;1.Oncetheseparametersbecomeknown,GiZisalsocompletelydetermined.5.5.2UpperBoundsoftheAverageDelayandDelayVariationThedelaythatapacketbelongingtoACiexperiences,denotedbyTi,canbeexpressedasfollows:TiTsi+Ri.31whereRiistheresidualMACservicetimeseenbythepacketunderconsideration.Notethatintheaboveequation,whenthepacketarrives,thequeueisassumedtobeemptyexceptthepacketcurrentlybeingserved.Thisisthecaseintheunsaturatedcase[97].SincetheprobabilitydistributionoftheMACservicetimeisknown,usingtheResidualLifeTheorem[56],wecaneasilyobtainboththemeanandvariationofRi:E[Ri]=PbE[Ts2i] 2E[Tsi]VAR[Ri]=PbE[Ts3i] 3E[Tsi])]TJ/F15 11.955 Tf 11.955 0 Td[(PbE[Ts2i] 2E[Tsi]2.32wherePbistheprobabilitythattheserverisbusywhenthepacketarrives.

PAGE 119

106 Then,theupperboundsofthemeanandvariationoftheMACservicetimecanbeobtainedbyusingthefactthatPbbelongsto[0;1],E[Ti]E[Tsi]+E[Ri]6E[Tsi]+E[Ts2i] 2E[Tsi]VAR[Ti]VAR[Tsi]+VAR[Ri]6VAR[Tsi]+5E[Ts3i] 12E[Ts3i])]TJ/F15 11.955 Tf 11.955 0 Td[(E[Ts2i] 2E[Tsi]2.335.5.3ModelValidationInthissection,wevalidateouranalyticalresultsthroughsimulations.Wesimulatean802.11ebasedwirelessLAN.Allnodesarewithinthetransmissionrangeofoneanother.Thechannelrateis2Mb/s.Again,weconsidertwokindsofreal-timetrafc,VBRvoicetrafcandCBRvideotrafc.Thetrafcparametersarelistedasfollows.VBRVoiceTrafc:anon/offsourcewithexponentiallydistributedonandoffperiodsof300msaverageeach.Trafcisgeneratedduringtheonperiodsatarateof32kb/swithapacketsizeof160bytes,thustheinter-packettimeis40ms.CBRVideoTrafc:aconstantrateof64kb/swithapacketsizeof1000bytes.Theinter-packettimeis125ms.Accordingtothe802.11e[45],weassignthevideotrafctoAC2andthevoicetrafctoAC3.AIFS[2]=60s,AIFS[3]=50s,W2;0=32,andW3;0=16.Thenumberofowsforeachtrafcclassisequal,i.e.,n2=n3.Notethatthenetworkworksintheunsaturatedcase.Fig. 5-6a and 5-6b respectivelyillustratethedelaymeanandstandarddeviationasafunctionofthetotalnumberofows,i.e.,n2+n3.Ineachgure,boththeanalyticalandsimulationresultsarepresented.Severalobservationsaremade.First,asthenumberofowsincreases,foreithertheanalyticalorsimulationresults,thedelaysandstandarddeviationsforbothtrafcclassesincreaseasaresultoftheincreasingcollisionlevel.Sec-ond,thedelayforthevoicetrafcismuchsmallerthanthatforthevideotrafc,whichisconsistentwiththefactthatthevoicetrafchasashortpacketsizeandahigherpriority

PAGE 120

107 thanthevideotrafcintermsofchannelaccess.Third,theupperboundsforboththemeanandvariationhold,indicatingthattheycanbeusedintheproposedcalladmissioncontrolschemepresentedbelow.Finally,ourobservationinsubsection 5.4.5 isagaincon-rmed:whenthenetworkworksintheunsaturatedcase,thedelaysforbothtrafcclassesaresufcientlysmalltosatisfytheirQoSrequirements.belessthan400ms.5.6CallAdmissionandRateControlFrameworkTokeepthenetworkoperatingintheunsaturatedcase,wherethecollisionprobabilityissmall,thethroughputishigh,andthedelayisshort[97],itiscrucialtoregulatetotalinputtrafc.Sincethereal-timetrafcisnotgreedyintermsofbandwidthusage,andmoreimportantly,hasstrictdelayrequirements,calladmissioncontrolCACisasuitabletrafccontrolmechanismforit.Ontheotherhand,forthebestefforttrafclikeTCP,whichcantoleratedelayrangingfromsecondstominutesbutaregreedyintermsofbandwidthusage,ratecontrolRCisappropriate.Therefore,weproposeacalladmissionandratecontrolframeworktocontrolthenetworktrafcsoastokeepthenetworkoperatingintheunsaturatedcase.Fig. 5-7 givesanoverviewofthisframework.Whentheupperlayertrafcisadmittedintothenetwork,itrstgoesthroughaclassierandisbufferedinatransmitqueuecorrespondingtoitspriority.Then,real-timetrafcwillbedirectlyservedbytheMACwhilenon-real-timewillbesubjecttotheratecontrolmechanism.InFig. 5-7 ,real-timepacketsareinredandnon-real-timepacketsareinblue.FortheIEEE802.11ethatreliesoncontention-basedchannelaccess,itishardtocharacterizethecurrenttrafcconditions.Therefore,weneedtondanappropriatecontrolvariableforboththeadmissioncontrolandratecontrol.WecanseeinFig. 5-7 thatboththecalladmissioncontrolandtheratecontrolcountonanindexcalledchannelbusynessratiofedbackbytheMAC.Inthefollowing,werstbrieydiscusstheconceptofchannelbusynessratio.

PAGE 121

108 a bFigure5-6:DelayperformancewhenAIFS[2]=60s,AIFS[3]=50s,W2;0=32andW3;0=16.aMean.bStandarddeviation.

PAGE 122

109 Figure5-7:Overviewofthecalladmissionandratecontrolframework 5.6.1ChannelBusynessRatioThechannelbusynessratio,denotedbyrb2[0;1],isdenedastheportionofthetimethatthechannelisbusyinanobservationperiod,whichcanbedirectlymeasuredateachnodesincetheIEEE802.11eMACisbasedoncarriersensing.Meanwhile,thechannelutilization,denotedbyu,isdenedastheportionofthetimethatthechannelisusedforsuccessfultransmissionsinanobservationperiod.Clearly,u6rb,andtheequalityholdsonlywhenallthechannelbusytimeisusedforsuccessfultransmissions.Inotherwords,thereisnocollisionatall.Becauseofspacelimit,weonlygiveseveraladvantagesofthechannelbusynessratioreferto[97]formoredetails.First,asweshowedin[97],intheunsaturatedcase,sincethecollisionprobabilityisverysmalltypicallybelow0.1andthetimewastedduetochannelcollisioncanbeignored,channelutilizationisalmostequaltothechannelbusynessratio.Asamatteroffact,thisisalsoconrmedinSection 5.7 .Forthisreason,wemayuse

PAGE 123

110 thesetwotermsinterchangeably.Moreover,intheunsaturatedcase,thereexistsanoptimalpointwherethenetworkachievesthemaximumthroughputandshortdelay.Thechannelbusynessratioatthispointchangesverylittlewhenthenumberofnodesorthepacketlengthschange.MoredetailsonhowtodeterminethecorrespondingchannelutilizationisgiveninSection 5.6.5 .Second,sincetheEDCAisbasedoncarriersensing,itiseasytomeasurethechannelbusytime.Ontheotherhand,thechannelutilizationisnotreadilymeasurableasanodecannotdistinguishchannelcollisionfromchannelfading.Next,wepresenttheCACandRCschemesinorder.ForCAC,wepresenttwoschemes,namelyacomprehensiveoneandasimpliedone.5.6.2CallAdmissionControlSchemeIAsspeciedintheIEEE802.11eEDCA,theadmissioncontrolisconductedattheQoSaccesspointQAPwhentheinfrastructuremodeisused.Ifthenetworkisworkingintheadhocmode,amobilenodecanbeelectedtocoordinatetheadmissioncontrolusingoneofmanyalgorithmsintheliterature[33,82].Furtherdiscussionsontheelectionalgorithmisbeyondthescopeofthischapter.Hereafter,weusethecoordinatortodenotetheQAPorthecoordinatingnodewithoutdifferentiation.Itshouldbenotedthatsuchacoordinatorisnecessarytoavoidtheso-calledover-admissionproblem,whichwilloccurwhenseveralindividualnodes,ifnotcoordinated,admitnewreal-timeowsatthesametimeandcausetheadmittedtrafctoexceedthenetworkcapacity.Intheadmissioncontrol,weshouldsetaquotaontheamountofthereal-timetrafcthatthenetworkcanadmit[24].Namely,ifwemeasurethetrafcamountintermsofitscontributiontothechannelutilizationorthechannelbusynessratio,weshouldsetaquotaonthechannelutilizationthatisduetothereal-timetrafc.Wesetsuchaquota,denotedbyUrt,to80% 1 ofthemaximumchannelutilization,denotedbyUmaxfortworeasons.It 1Thisnumberistunableandcouldbechangeddependingonthetrafccompositioninrealnetworks.Wechoose80%forourstudyonly.

PAGE 124

111 rstensuresthatthebestefforttrafcisoperationalallthetime,sincethebestefforttrafcisatleastentitledto20%ofthechannelutilization.Inaddition,the20%ofthechannelutilizationforthebestefforttrafccanbeusedtoaccommodatesizableuctuationscausedbytheVBRreal-timetrafc.IntheCACscheme,threeparameters,Rmean,Rpeak,L,areusedtocharacterizethebandwidthrequirementofareal-timeow,whereRmeanistheaveragedatarateandRpeakthepeakdatarateinbit=s,andListheaveragepacketlengthinbits.ForCBRtrafc,Rmean=Rpeak.ForVBRtrafc,Rmean
PAGE 125

112 UponreceivingtheADDTS,thecoordinatorassociatestheowwiththeappropriateACiandobtainsui;meanandui;peakaccordingtoEquation 5-35 .Then,itdeterminesiftheowcanbeadmittedusingthefollowingtests: First,theremainderofthequotaUrtandUmaxshouldbeabletoaccommodatethenewreal-timeow,i.e.,uA;mean+ui;mean
PAGE 126

113 5.6.3CallAdmissionControlSchemeIIItcanbeseenthatwhenmakingadmissiondecisions,CACschemeItakesintoac-countboththepeakrateandmeanrateforthereal-timetrafc.Whilethisensuresthatthenetworkwillnotbecongestedintheworst-casescenario,inwhichalltheVBRreal-timetrafctransmitsatitspeakrate,itmayunnecessarilyrejectmanyreal-timeowswhentheratioRpeak=Rmeanislargeforsomereal-timeapplications.Toresolvethisproblem,weonlyconsiderthemeanrateintheadmissioncontrolscheme.Meanwhile,recognizingthatitmaynotbepracticalfornon-QAPnodestocalculatetheaveragedelaybeforehandifthenetworkworksintheadhocmode,Wefurtherremovethedelaytestfromtheadmissionscheme.NotethatthismightnotbeatoobadideaifweconsiderthatassuggestedinSection 5.4.5 ,aslongasthenetworkiskeptworkingintheunsaturatedcaseandthebestefforttrafciswellcontrolledtoisolateitseffectonthereal-timetrafc,thedelayforthereal-timetrafcshouldbesmallenoughtomeettheQoSrequirements.AftermakingthesetwochangestoCACschemeI,wegetthesimpliedCACschemeIIasfollows.WhenanodesendsarequestwiththecorrespondingTSPECtothecoordinator,thecoordinatorgrantsadmissionifthefollowingtestispassed: TheremainderofthequotaUrtshouldbeabletoaccommodatethenewreal-timeow,i.e.,uA;mean+ui;mean
PAGE 127

114 andCWaresetmuchlargerthanthoseforthereal-timetrafc.However,thisapproachisproblematicinthatitwillunnecessarilyimpedethebestefforttrafcfromaccessingthechannelevenwhenthereisnoheavyreal-timetrafcinthenetwork,leadingtochannelunderutilizationandunreasonablylargedelayforthebestefforttrafc.Second,thebestefforttrafcshouldbeabletopromptlyaccesstheresidualbandwidthleftbythereal-timetrafcinordertoefcientlyutilizethechannel.Clearly,tomeetthesecriteria,eachnodeneedstoaccuratelyestimatethetotalinstan-taneousrateoftheongoingreal-timetrafc.However,thisisnotaneasytaskifthenetworkworksintheadhocmode,wherenodescancommunicatewithoneanotherdirectlywithoutinvolvingQAP.Meanwhile,evenifthenetworkworksintheinfrastructuremode,sincetheIEEE802.11eallowsdirectlinksbetweentwonon-QAPnodes,allcommunicationsmaynotnecessarilygothroughtheQAP.Itcanthusbeconcludedthatineithermode,thereisnonodethatcanaccuratelymonitorallthetrafcintheairandcontrolthetrafcrateofalltheothernodes.Therefore,aneffectivedistributedratecontrolschemeisdesired.Intheratecontrolscheme,eachnodeneedstomonitorthechannelbusynessratiorbduringaperiodofTrb.Letusdenotebyrbrthecontributionfromthereal-timetrafctorb,anddenotebyRbethedatarateofthebestefforttrafcatthenodeunderconsideration,withtheinitialvalueofRbebeingconservativelyset,sayonepacketpersecond.ThenodethusadjustsRbeaftereachTrbaccordingtothefollowing:Rbenew=RbeoldUmax)]TJ/F25 11.955 Tf 11.955 0 Td[(rbr rb)]TJ/F25 11.955 Tf 11.955 0 Td[(rbr;.39whereRbenewandRbeoldarethevalueofRbeafterandbeforetheadjustment.TwopointsarenotedonEquation 5-39 .First,weseethatthenodeincreasestherateofthebestefforttrafcifrb
PAGE 128

115 wherePRbeoldU)]TJ/F24 7.97 Tf 6.587 0 Td[(1rb)]TJ/F25 11.955 Tf 12.096 0 Td[(rbrisduetothefactthatthechannelbusynessratioisequaltothechannelutilizationandrb)]TJ/F25 11.955 Tf 11.837 0 Td[(rbristhecontributionfromthetotalbestefforttrafctorb.ThusafteronecontrolintervalTrb,thechannelutilizationwillbeapproximatetoUmax.Toestimationofrbr,eachmobilenodeneedstomonitorallthetrafcintheair.How-ever,tobeconsistentwiththeoriginal802.11eprotocol,ourschemeonlyrequiresmobilenodestodecodetheMACheaderpart,astheoriginal802.11edoesintheNAVprocedure.Todistinguishreal-timepacketsfrombesteffortpackets,weonlyneedtocheckthemostsignicantbitofthesubtypeeld,whichisdenedintheIEEE802.11eastheQoSsubeldindatapackets.Therefore,theobservedchannelbusynessratiocomprisesthreepiecesofcontribution:thecontributionfromthebestefforttrafcwithadecodableMACheaderrb1,thatfromthereal-timetrafcwithadecodableMACheaderrb2,andthatofallthetrafcwithanundecodableMACheaderrb3duetocollision.Sowegiveanupperboundandalowerboundforrbrasfollows:rb26rbr6rb2+rb3:.41Toenforceaconservativelyincreasingandaggressivelydecreasinglaw,wethussetrbrasfollows:rbr=8><>:rb2;ifrbUmax:.42WealsonotethatthecontrolintervalTrbshouldbesetsuchthattheschemecanberesponsivetothechangeofthechannelbusynessratioobservedintheairandcansmoothouttheinstantaneousdisturbance.5.6.5DeterminationofUmaxItisclearthatchoosinganappropriatemaximumchannelutilization,Umax,iscriticalinmakingboththecalladmissioncontrolandratecontrolwork.Next,weshowhowtodetermineanappropriatevalueforUmax.

PAGE 129

116 First,weconsiderhowthepacketlengthaffectsthechannelutilization.Weconsideranetworkof40nodesandeachnodegeneratesCBRtrafc.Thedefault802.11DCFsystemparametersareused:SIFS=10s,DIFS=50s,andtheinitialCW=32.TheRTS/CTSmechanismisused.Fig. 5-9 showsboththethroughputanddelayasafunctionoftheinputnetworktrafc.Hereweusethenominalchannelutilizationtodenotethetrafcload,thatis,thechannelutilizationthatwouldresultfromthetrafcloadasiftherewerenocollisionsatall.Wecanseethatregardlessofthepacketlength,asthetrafcloadincreases,thethroughputrstincreasesandthendecreases;meanwhile,thedelayrstincreasesveryslowlyandthenincreasesdramatically.Inotherwords,thenetworkentersfromunsaturationtosaturation.Thechannelutilizationvaluescorrespondingtotheturningpointsortheboundariesbetweenunsaturationandsaturation,isUmax.Itcanalsobeobservedthatwhenthepacketlengthincreases,sodoesUmax.Obviously,toachievethemaximumthroughputandshortdelay,Umaxshouldbesetintherangeof0.9to0.95.Second,weconsiderhowrobustsuchachoiceofUmaxisintheprioritizedscenarios.Weconsidertwotypesoftrafc,namelythehighprioritytrafcandlowprioritytrafc.Forhighprioritytrafc,AIFS=50sandtheinitialCW=16;thepacketlengthis500Bytes.Forlowprioritytrafc,AIFS=60sandtheinitialCW=32;thepacketlengthis1000Bytes.Eachnodegenerateseitherahighpriorityorlowprioritytrafcow.Fig. 5-9 showsthethroughputanddelaywhenthetrafcloadincreases.Again,wecanseethatthechoiceofUmaxwithintherangeof[0.9,0.95]leadstogoodperformance.NotetheseobservationsarealsotrueofthecasewhereRTS/CTSisnotused.5.7PerformanceEvaluation5.7.1SimulationCongurationToevaluatetheperformance,weconductsimulationsinOPNETModeler10.0[72].An802.11ebasedwirelessLANwith100mobilenodesissimulated.Allnodesarewithinthetransmissionrangeofoneanother.Inallsimulations,channelrateis2Mb/sandtheRTS/CTSmechanismisused.Inadditiontothetwotypesofreal-timetrafcmentioned

PAGE 130

117 Figure5-8:ChoiceofUmaxfordifferentpacketlengths Figure5-9:ChoiceofUmaxforprioritizedtrafcanddifferentnumbersofnodes

PAGE 131

118 insection 5.4.5 ,wealsoconsiderthegreedybest-effortTCPtrafcAC0,whichisofapacketsizeof1000bytes.TCP-Renoisused.Sovoice,video,anddatacorrespondtoAC3,AC2,andAC0respectively.TheAIFSandCWparametersaresetasfollows.AIFS[0]=80s,AIFS[2]=60s,AIFS[3]=50s;W0;0=128,W2;0=32,andW3;0=16.Insuchasetting,itisclearthatthevoicetrafchasthehighestpriorityandtheTCPtrafchasthelowestpriorityintermsofchannelaccess.Umax=0:93andUrt=Umax80%=0:744.TheperiodofmeasuringthechannelbusynessratioTrb=2s.InCACschemeI,D2=200msandD3=100ms.Thesimulationtimeis120seconds.Inthesimulation,thetrafcloadisgraduallyincreased.Specically,anewvoice,videoorTCPowisperiodicallyaddedinaninterleavedway,inordertoobservehowtheschemeworksandhowanewlyadmittedowimpactstheperformanceofpreviouslyadmittedows.Until94seconds,anewvoiceowisaddedatthetimeinstantof6isecond6i615.Likewise,avideoowisaddedtwosecondslaterandaTCPowisadded4secondslater.Furthermore,tosimulatetherealscenariowherethestartofreal-timeowsarerandomlyspreadovertime,thestartofavoiceowisdelayedarandomperioduniformlydistributedin[0ms,40ms],andthatofavideoowdelayedarandomperioduniformlydistributedin[0ms,125ms].Notethatinthesimulationperiodbetweens,120s],wepurposelystopinjectingmoreowsintothenetworkinordertoobservehowwelltheschemeperformsinasteadystate.5.7.2SimulationResultsCACschemeIandRCFromthesimulationresults,wendthereareatotalof10voiceowsand10videoowsadmittedby56seconds;andnomorevoiceorvideoowsareadmittedthereafter.ThenumberofTCPowsincreasesbyoneevery6secondsuntil94seconds.After94seconds,asexpected,thereisnochangeinthenumberofows.Thisisexpected.Accord-ingtoEquation 5-35 ,weknowthattheu3;meanandu3;peakforavoiceoware0.0248and0.0496,respectively;andtheu2;mean=u2;peakforavideoowis0.04283.Following

PAGE 132

119 theadmissioncriteriainCACschemeI,afterthenetworkadmits10voiceowsand10videoows,uA;mean=0:6763anduA;peak=0:9243.Obviously,nomorereal-timeowscanbeacceptedduetotheconstraintofUmax=0:93.Weshouldmentionthatupto56seconds,noreal-timeowsarerejectedbecausethedelaycriterionspeciedinEquation 5-37 cannotbemet.Duringthesimulation,neitherreal-timeorbesteffortpacketsarelost.Fig. 5-10 showsthethroughputforthethreetrafcclassesthroughoutthesimulation.Atthebeginning,theTCPtrafchashighthroughput;thenasmorereal-timeowsareadmitted,itgraduallydropsasaresultoftheratecontrol.BecausewesetanupperboundUrtforthereal-timetrafc,itcanbeobservedthatevenwhenthetrafcloadbecomesheavy,TCPtrafc,asdesired,isnotcompletelystarved.BecauseTCPtrafcisallowedtouseanyavailablechannelcapacityleftbythereal-timetrafc,thetotalchannelutilization,namelythesumofthechannelutilizationduetodifferenttypesoftrafc,stabilizesatashighas0.9,asshowninFig. 5-11 .Fig. 5-11 alsoshowsthatintheunsaturatedcase,asaresultoftheverysmallcollisionprobability,thechannelutilizationcurvecoincideswiththechannelbusynessratiocurve.ThepacketdelayisillustratedinFig. 5-12 ,inwhicheverypointisaveragedover2seconds.Asexpected,itcanbeobservedthatthedelayforthereal-timetrafciskeptbelow20ms;moreover,thedelayforthevoicetrafcismuchsmallerthanthatforthevideotrafc.Initially,asthenumberofadmittedreal-timeowsincreases,thedelayincreases.NotethattheincreaseofdelayisnotduetotheTCPtrafc,butmainlyduetotheincreasingnumberofcompetingreal-timeows.Then,thedelayoscillatesaroundastablevalue.Fig. 5-13 presentsthedelaydistributionforthevoiceandvideotrafcwithoutanyaveraging.MoredetailedstatisticsofdelayanddelayvariationaregiveninTable 5-1 .Again,noaveragingistaken.AsshowninTable 5-1 ,the97percentiledelayvaluesforvoiceandvideoare18.5msand29.2msrespectively,andthe99percentiledelayvaluesforvoiceandvideoare24.6msand37.1msrespectively.Itisknownthatforthereal-timetrafc,

PAGE 133

120 packetsthatfailtoarriveintimearesimplydiscarded.Giventheallowable1%3%packetlossrate,thesedelaysarewellwithintheboundsgivenin[46,47].ThegooddelayperformanceindicatesthatCACschemeIandRCtogethercaneffectivelyguaranteethedelayanddelayjitterrequirementsofthereal-timetrafc,eveninthepresenceofhighlydynamicTCPtrafc.CACschemeIIandRCUnlikethepreviouscase,whenCACschemeIIandRCareused,weobservethatthereareatotalof11voiceowsand11videoowsadmittedby62seconds;andnomorevoiceorvideoowsareadmittedthereafter.Again,thenumberofTCPowsincreasesbyoneevery6secondsuntil94seconds.After94seconds,thereisnochangeinthenumberofows.Thereasonthatmorereal-timeowsareadmittedinthiscaseisthefollowing.Inthepreviouscase,after10voiceowsand10videoowsareadmitted,uA;peakisclosetoUmaxandthusnomorereal-timeowscanbeaccepted.SinceCACschemeIIeliminatesthatconstraint,nowonlytheconstraintUrtworks.After11voiceowsand11videoowsgetintothenetwork,uA;meanisequalto0.7439andclosetoUrt.Thus,nomorereal-timeowscanbeadmitted.Duringthesimulation,neitherreal-timeorbesteffortpacketsarelost.InFig. 5-14 ,weseeasonemorevoiceowandonemorevideoowareacceptedcomparedtothepreviouscase,theTCPthroughputinthesteadystatedropsbyacor-respondingamount.Thechannelutilizationalsoremainssteadilyhighexceptthatsomeslightuctuationsareobservedasopposedtothatinpreviouscase,sincemoreVBRvoiceowsinthenetwork.Fig. 5-16 and 5-17 demonstratethatthedelayrequirementsofthereal-timetrafccanbeadequatelymet.However,asexpected,theresultsareabitworsethanthoseinthepreviouscase.ThiscanalsobeseeninTable 5-2 ,wherethe97percentile,99percentile,and99.9percentiledelayvaluesforvoiceandvideoslightlyincrease.Asawhole,however,thegoodperformanceintermsofboththroughputanddelayindicatesthatthissimpliedCACschemeIIincombinationwithRCstillworkswell.

PAGE 134

121 Figure5-10:Aggregatethroughput Figure5-11:Channelbusynessratioandchannelutilization Table5-1:Themean,standarddeviationSD,and97'th,99'th,99.9'thpercentiledelayssforvoiceandvideowhenCACschemeIandRCareused. mean SD 97%ile 99%ile 99.9%ile VBRVoice 0.0065 0.0051 0.0185 0.0246 0.0411 CBRVideo 0.0123 0.0074 0.0292 0.0371 0.0708

PAGE 135

122 Figure5-12:Averagedelayofvoiceandvideotrafc Figure5-13:Delaydistributionofvoiceandvideotrafc Table5-2:Themean,standarddeviationSD,and97'th,99'th,99.9'thpercentiledelayssforvoiceandvideowhenCACschemeIIandRCareused. mean SD 97%ile 99%ile 99.9%ile VBRVoice 0.0069 0.0066 0.0209 0.0306 0.0684 CBRVideo 0.0130 0.0089 0.0338 0.0421 0.0738

PAGE 136

123 Figure5-14:Aggregatethroughput Figure5-15:Channelbusynessratioandchannelutilization

PAGE 137

124 Figure5-16:Averagedelayofvoiceandvideotrafc Figure5-17:Delaydistributionofvoiceandvideotrafc

PAGE 138

125 5.8CONCLUSIONWhiletheemergingIEEE802.11ewirelessLANsupportsprioritizedservices,itcan-notprovidestrictQoSforthereal-timetrafc.Inthischapter,weenhancethe802.11ebyproposingtwocalladmissionschemesandaratecontrolscheme.Werstbuildananalyt-icalmodeltoanalyzetheaveragedelayforthetrafcwithdifferentprioritiesandderiveanestimate,whichisthenusedinthecalladmissioncontrolmechanism.Wealsoderivetheupperboundsforboththedelaymeananddelayvariationintheunsaturatedcase.Theanalyticalresultsshowthe802.11eWLANcansatisfythedelayrequirementsofthereal-timetrafcaslongasthenetworkistunedtooperateintheunsaturatedcase.Then,relyingonthenovelusechannelbusynessratio,wedemonstratethatthetwocalladmissioncon-trolschemesensureQoSguaranteesforthereal-timetrafcandtheratecontrolschemeallowsthebestefforttrafctousetheresidualchannelcapacityleftbythereal-timetraf-c.Finally,thesimulationresultsshowthattheproposedschemessuccessfullyguaranteestringentQoSrequirementsofreal-timeservices,whileachievinghighchannelutilization.

PAGE 139

CHAPTER6CONCLUSIONSANDFUTUREWORKResourcemanagementisimportantinwirelessnetworksasitdirectlydealswithef-cientlyutilizingnetworkresourcesandprovidingdiverseQoSforavarietyofservicessuchasvoice,videoanddata.Amongvarioustypesofwirelessnetworks,mobilecellularnet-works,mobileadhocnetworks,andwirelessLANshaveeachtakentheirplacesinmeetinguserdemandsindifferentapplicationscenarios.Clearly,thereisnoone-size-t-allsolu-tionsforresourcemanagementinthesedifferingnetworks.Inthisdissertation,weproposedifferentresourcemanagementandQoSsupportschemesforeachtypeofnetworksinlightoftheirspeciccharacteristics.6.1ContributionsInsummary,wemakethefollowingcontributionsinthisdissertation. Formobilecellularnetworks,weproposedtwodynamicmultiplethresholdband-widthreservationschemestosupportconnection-levelQoS.Intheseschemes,thethresholdsaredynamicallyadjustedbasedonnetworktrafcconditionsandQoSre-quirementsofserviceswithdifferentpriorities.Especially,inthesecondscheme,twoQoSrequirementsareconsidered:1keepingthehandoffdroppingprobabilitylessthanapredenedQoSthreshold2maintainingrelativeprioritiesfordifferenttrafcclassesbasedonconnectionblockingprobability.Besides,intimesofnetworkcongestion,apreventivemeasureistakentothrottletheacceptanceofnewconnec-tions.Moreover,theconceptofrelativepriorityisgeneralizedtogivethenetworkoperatormoreexibilitytoadjustadmissioncontrolpolicybytakingintoaccountsomedynamicfactorssuchasofferedload. Formobileadhocnetworks,Weproposedaresourcemanagementscheme.Sincecommunicationsbetweensomeimportantnodesinthenetworkaremorecriticalthanothers,theyshouldbeacceptedbythenetworkwithhighpriorityintermsofnetworkresourceusageandqualityofserviceQoSsupport.Inthisscheme,wedesignedalocation-awarebandwidthpre-reservationmechanism,whichtakesadvantageofeachmobilenode'sgeographiclocationinformationtopre-reservebandwidthforsuchhighpriorityconnectionsandthusgreatlyreducespotentialschedulingconictsfortransmissions.Inaddition,anend-to-endbandwidthcalculationandreservationalgorithmwasproposedtomakeuseofthepre-reservedbandwidth.Inthisway, 126

PAGE 140

127 timeslotcollisionsamongdifferentconnectionsandinadjacentwirelesslinksalongaconnectioncanbereducedsothatmorehighpriorityconnectionscanbeacceptedintothenetworkwithoutseriouslyhurtingadmissionsofotherconnections.Thankstothecross-layerdesignthattakesadvantageofthecollaborationbetweentheroutingandMAClayer,thisschemecansuccessfullyprovidebettercommunicationqualitytoimportantnodesatarelativelylowprice. FortheemergingIEEE802.11ewirelessLANs,westudiedprovidingstrictQoSin-steadofdifferentiatedQoSforreal-timetrafc.Werstbuiltananalyticalmodeltoderivetheapproximationsofthedelaymeansforthetrafcofdifferentpriori-tiesintheunsaturated802.11eWLAN.Wealsoderivedtheupperboundsofdelaymeansandvariations.WeshowedthattheQoSrequirementsofthereal-timetraf-ccanbesatisediftheinputtrafcisproperlyregulated.Then,weproposedtwocalladmissioncontrolschemesandaratecontrolschemethatrelyontheaveragedelayapproximationsandthechannelbusynessratio,anindexthatcanaccuratelyrepresentthenetworkstatus.Thekeyideais,whenacceptinganewreal-timeow,theadmissioncontrolalgorithmconsidersitseffectonthechannelutilizationandthedelayexperiencedbyexistingreal-timeows,ensuringthatthechannelisnotoverloadedandthedelayrequirementsarenotviolated.Atthesametime,theratecontrolalgorithmallowsthebestefforttrafctofullyusetheresidualbandwidthleftbythereal-timetrafc,therebyachievinghighchannelutilization.6.2FutureWorkWirelesstechnologieshavebeenexperiencingcontinuousevolutionandposingnewchallenges.Nextwediscusssomefutureresearchdirections.Inter-systemresourcemanagementWemainlyresearcheddifferentresourcemanagementschemesthataresuitedfordifferenttypesofwirelessnetworksinthisdissertation.However,suchintra-systemre-sourcemanagementschemesarenotadequateforsupportingglobalandseamlessmobilitythroughwhichmobileuserscanroambetweensystemswithheterogeneouswirelesstech-nologies.Therefore,weneedtocarefullystudyinter-systemresourcemanagement.Thismayinvolveaseriesofproblemsincludingadaptiveresourcereservation,verticalhandoffchoices,andQoSmapping.DynamicSpectrumAllocationDuetotheincreasingdensityofwirelessdevicesandthedeploymentofvariouswire-lessservices,currentstaticspectrumallocationisinefcientandcanhindertheseamless

PAGE 141

128 integrationbetweennetworksofdifferentphysical-layertechnologies.Duetotheadvance-mentsalongthelineofsoftware-denedradioSDR[86]thatcanadjustmodulationschemesandpowercontrolonline,dynamicspectrumallocationhasrecentlybeenpro-posedtodealwiththespectrumshortage.Itssuccesshingesonresolvingissuessuchasfreespectrumscanninganddetection,centralizedordistributedallocationcoordination,andsoon.WirelessMeshNetworksUnliketraditionalmultihopadhocnetworks,wirelessmeshnetworksWMNsconsistofmeshroutersandmeshclients[3].Meshroutersarenon-mobileandformthebackboneofmeshnetworks,relayingtrafctoandfromtheInternet.WMNshaverecentlyattractedenormousinterestsduetosuchadvantagesaslowcost,easymaintenance,androbustness.InorderforWMNstoevolveasaviablewirelessaccessmethod,itisimportanttosupportQoSfortrafcbetweenthemeshclientsandthewirednetworksthroughmultiplewire-lesshops.Also,itisexpectedmultiplechannelsandradiosshouldbeusedinordertoboostthroughput.Thus,thejointstudyofroutingandchanneland/orradioallocationbetweenmeshrouterswouldbeveryinterestingandmayleadtosignicantperformanceimprovement.

PAGE 142

REFERENCES [1] I.AdaandC.Castelluccia,“DifferentiationmechanismsforIEEE802.11,”inPro-ceedingsofIEEEINFOCOM,pp.209-218,Anchorage,Alaska,Apr.2001. [2] G-SAhn,A.T.Campbell,A.Veres,andL-HSun,“SWAN:servicedifferentiationinstatelesswirelessadhocnetworks,”inProceedingsofIEEEINFOCOM,pp.457-466,NewYork,Jun.2002. [3] I.F.Akyildiz,X.Wang,andW.Wang,“Wirelessmeshnetworks:asurvey,”Com-puterNetworksJournalElsevier,vol.47,no.4,pp.445-487,Mar.2005. [4] R.BankaandG.Xue,“Angleroutingprotocol:locationaidedroutingformobilead-hocnetworksusingdynamicangleselection,”inProceedingsofIEEEMILCOM,pp.501-506,Anaheim,California,Oct.2002. [5] S.Basagni,I.Chlamtac,V.Syrotiuk,andB.Woodward,“AdistanceroutingeffectalgorithmformobilityDREAM,”inProceedingsofACMMobiCom,pp.76-84,Dallas,Oct.1998. [6] V.Bharghavan,A.Demers,S.Shenker,andL.Zhang,“MACAW:amediaaccessprotocolforwirelessLAN's,”inProceedingsofACMSIGCOMM,pp.212-225,London,UnitedKingdom,Oct.1994. [7] G.Bianchi,“PerformanceanalysisoftheIEEE802.11distributedcoordinationfunc-tion,”IEEEJournalonSelectedAreasinCommunications,vol.18,no.3,pp.535-547,Mar.2000. [8] P.Bjorklund,P.Varbrand,andD.Yuan,“ResourceoptimizationofspatialTDMAinadhocradionetworks:acolumngenerationapproach,”inProceedingsofIEEEINFOCOM,pp.818-824,SanFrancisco,Mar.-Apr.2003. [9] S.Blake,D.Black,M.Carlson,E.Davies,Z.Wang,andW.Weiss,“Anarchitecturefordifferentiatedservices,”RFC2475,1998. [10] R.Braden,D.Clark,andS.Shenker,“IntegratedservicesintheInternetarchitecture:anoverview,”RFC1633,1994. [11] F.Cali,M.Conti,andE.Gregori,“IEEE802.11protocol:designandperformanceevaluationofanadaptivebackoffmechanism,”IEEEJournalonSelectedAreasinCommunications,vol.18,no.9,pp.1774-1786,Sep.2000. 129

PAGE 143

130 [12] F.Cali,M.Conti,andE.Gregori,“TuningoftheIEEE802.11protocoltoachieveatheoreticalthroughputlimit,”IEEE/ACMTransactionsonNetworking,vol.8,no.6,pp.785-799,Dec.2000. [13] C.-J.Chang,T.-T.Su,andY.-Y.Chiang,“Analysisofacutoffprioritycellularradiosystemwithnitequeueingandreneging/dropping,”IEEE/ACMTransactionsonNetworking,vol.2,no.2,pp.166-175,Apr.1994. [14] C.ChaoandW.Chen,“Connectionadmissioncontrolformobilemultiple-classpersonalcommunicationsnetworks,”IEEEJournalonSelectedAreasinCommuni-cations,vol.15,no.8,pp.1618-1626,Oct.1997. [15] T.-C.Chau,K.Y.MichaelWong,andB.Li,“OptimizingcalladmissioncontrolwithQoSguaranteeinavoice/dataintegratedcellularnetworkusingsimulatedanneal-ing,”inProceedingsofIEEEGlobecom,pp.809-813,Taipei,Nov.2002. [16] S.ChenandK.Nahrstedt,“Anoverviewofquality-of-serviceroutingforthenextgenerationhigh-speednetworks:problemsandsolutions,”IEEENetwork,vol.12,no.6,pp.64-79,Nov./Dec.1998. [17] S.ChenandK.Nahrstedt,“Distributedquality-of-serviceroutinginadhocnet-works,”IEEEJournalonSelectedAreasinCommunications,vol.17,no.8,pp.1488-1505,Aug.1999. [18] X.Chen,W.Liu,H.Zhai,andY.Fang,“Location-awareresourcemanagementinmobileadhocnetworks,”toappearinACMWirelessNetworks. [19] X.Chen,B.Li,andY.Fang,“Adynamicmultiple-thresholdbandwidthreservationDMTBRschemeforQoSprovisioninginmultimediawirelessnetworks,”IEEETransactionsonWirelessCommunications,vol.4,no.2,pp.583-592,Mar.2005. [20] X.Chen,H.Zhai,andY.Fang,“EnhancingtheIEEE802.11einQoSsupport:analysisandmechanisms,”inProceedingsoftheSecondInternationalConferenceonQualityofServiceinHeterogeneousWired/WirelessNetworksQShine'05,Or-lando,Aug.2005. [21] I.Chlamtac,A.Farago,andH.Zhang,“Time-spreadmultiple-accessTSMAproto-colsformultihopmobileradionetworks,”IEEE/ACMTransactionsonNetworking,vol.5,no.6,pp.804-812,Oct.1997. [22] S.Choi,J.Prado,S.Mangold,andS.Shankar,“IEEE802.11econtention-basedchannelaccessEDCFperformanceevaluation,”inProceedingsofIEEEICC'03,pp.1151-1156,Anchorage,Alaska,May2003. [23] A-M.ChouandV.O.K.Li,“SlotallocationstrategiesforTDMAprotocolsinmul-tihoppacketradionetworks,”inProceedingsofIEEEINFOCOM,pp.710-716,Flo-rence,Italy,May1992.

PAGE 144

131 [24] D.Clark,S.Shenker,andL.Zhang,“Supportingreal-timeapplicationinaninte-gratedservicespacketnetwork:architectureandmechanism,”inProceedingsofACMSIGCOMM,pp.14-26,Baltimore,Aug.1992. [25] D.C.Cox,“Wirelesspersonalcommunications:whatisit?,”IEEEPersonalCom-munication,vol.2,no.2,pp.20-35,Apr.1995. [26] A.I.ElwalidandD.Mitra,“EffectivebandwidthofgeneralMarkoviantrafcsourcesandadmissioncontrolofhigh-speednetworks,”IEEE/ACMTransactionsonNetworking,vol.2,no.2,pp.166-175,Apr.1994. [27] B.M.EpsteinandM.Schwartz,“PredictiveQoS-basedadmissioncontrolformulti-classtrafcincellularwirelessnetworks,”IEEEJournalonSelectedAreasinCom-munications,vol.18,no.3,pp.523-534,Mar.2000. [28] Y.FangandI.Chlamtac,“TeletrafcanalysisandmobilitymodelingforPCSnet-works,”IEEETransactionsonCommunications,vol.47,no.7,pp.1062-1072,Jul.1999. [29] Y.Fang,I.Chlamtac,andY.B.Lin,“ChanneloccupancytimesandhandoffrateformobilecomputingandPCSnetworks,”IEEETransactionsonComputers,vol.47,no.6,pp.679-692,Jun.1998. [30] Y.FangandY.Zhang,“Calladmissioncontrolschemesandperformanceanalysisinwirelessmobilenetworks,”IEEETransactionsonVehicularTechnology,vol.51,no.2,pp.371-382,Mar.2002. [31] V.Firoiu,J.-Y.LeBoudec,D.Towsley,andZ.-L.Zhang,“Theoriesandmodelsforinternetqualityofservice,”ProceedingsoftheIEEE,vol.90,no.9,pp.1565-1591,Sep.2002. [32] C.H.FohandM.Zukerman,“PerformanceanalysisoftheIEEE802.11MACpro-tocol,”inProceedingsofEuropeanWireless2002,pp.184-190,Florence,Italy,Feb.2002. [33] H.Garcia-Molina,“Electionsinadistributedcomputingsystem,”IEEETransac-tionsonComputers,vol.31,no.1,pp.48-59,Jan.1982. [34] E.Gelenbe,X.Mang,andR.Onvural,“Bandwidthallocationandcalladmissioncontrolinhigh-speednetworks,”IEEECommunicationMagazine,vol.35,no.5,pp.122-129,May1997. [35] D.GrossandC.M.Harris,FundamentalsofQueueingTheory.3rdEdition,NewYork:Wiley,1998. [36] R.Guerin,“Queueing-blockingsystemwithtwoarrivalstreamsandguardchan-nels,”IEEETransactionsonCommunications,vol.36,no.2,pp.153-163,Feb.1988.

PAGE 145

132 [37] Z.Hadzi-VelkovandB.Spasenovski,“SaturationthroughputdelayanalysisofIEEE802.11DCFinfadingchannel,”inProceedingsofIEEEICC,pp.121-126,Anchor-age,Alaska,May2003. [38] Y.-R.Haung,Y.-B.Lin,andJ.-M.Ho,“Performanceanalysisforvoice/dataintegra-tiononanite-buffermobilesystem,”IEEETransactionsonVehicularTechnology,vol.49,no.2,pp.367-378,Feb.2000. [39] H.HeffesandD.M.Lucantoni,“AMarkovmodulatedcharacterizationofpack-etizedvoiceanddatatrafcandrelatedstatisticalmultiplexerperformance,”IEEEJournalonSelectedAreasinCommunications,vol.4,no.6,pp.856-868,Sep.1986. [40] T.S.HoandK.C.Chen,“PerformanceanalysisofIEEE802.11CSMA/CAmediumaccesscontrolprotocol,”inProceedingsofIEEEPIMRC,pp.407-411,Taipei,Oct.1996. [41] H.HolmaandA.Toskala,WCDMAforUMTS:radioaccessforthirdgenerationmobilecommunications.2ndEdition,NewYork:Wiley,2002. [42] D.HongandS.S.Rappaport,“Trafcmodelandperformanceanalysisforcellularmobileradiotelephonesystemswithprioritizedandnonprioritizedhandoffproce-dures,”IEEETransactionsonVehicularTechnology,vol.35,no.3,pp.77-92,Aug.1986. [43] J.HouandY.Fang,“Mobility-basedcalladmissioncontrolschemesforwirelessmobilenetworks,”WirelessCommunicationsandMobileComputing,vol.1,no.3,pp.269-282,Jul.-Sep.2001. [44] IEEE802.11WG,IEEEstandardforWirelessLANMediumAccessControlMACandPhysicalLayerPHYspecications,ISO/IEC8802-11:1999E,1999. [45] IEEE802.11WG,DraftSupplementtoPart11:WirelessMediumAccessControlMACandPhysicalLayerPHYspecications:MediumAccessControlMACEnhancementsforQualityofServiceQoS,IEEEStd802.11e/D8.0,Feb.2004. [46] ITU-TG.114,One-waytransmissiontime,1996. [47] ITU-TG.1010,End-usermultimediaQoScategories,2001. [48] V.Jackson,“Congestionavoidanceandcontrol,”inProceedingsofACMSIG-COMM,pp.314-329,Stanford,California,Aug.1988. [49] S.Jamin,P.B.Danzig,S.Shenker,andL.Zhang,“Ameasurement-basedadmissioncontrolalgorithmforintegratedservicepacketnetworks,”IEEE/ACMTransactionsonNetworking,vol.5,no.1,pp.56-70,Feb.1997. [50] D.Johnson,D.Maltz,andJoshBroch,“DSR:TheDynamicSourceRoutingPro-tocolforMulti-HopWirelessAdHocNetworks,”inAdHocNetworking,editedbyCharlesE.Perkins,Chapter5,pp.139-172,Addison-Wesley,2001.

PAGE 146

133 [51] V.Kanodia,C.Li,A.Sabharwal,B.Sadeghi,andE.Knightly,“Distributedmulti-hopschedulingandmediumaccesswithdelayandthroughputconstraints,”inPro-ceedingsofACMMOBICOM,pp.200-209,Rome,Italy,Jul.2001. [52] P.Karn,“MACA-anewchannelaccessmethodforpacketradio,”inProceedingsofARRL/CRRLAmateurRadio9thComputerNetworkingConference,pp.134-140,London,ON,Canada,Sep.1990. [53] B.KarpandH.Kung,“GPSR:Greedyperimeterstatelessroutingforwirelessnet-works,”inProceedingsofACMMobiCom,pp.243-254,Boston,Aug.2000. [54] I.KatzelaandM.Naghshineh,“Channelassignmentschemeforcellularmobiletelecommunicationsystems:acomprehensivesurvey,”IEEEPersonalCommunica-tion,vol.3,no.3,pp.10-31,Jun.1996. [55] J.KimandA.Jamalipour,“TrafcmanagementandQoSprovisioninginfuturewirelessIPnetworks,”IEEEPersonalCommunication,vol.8,no.5,pp.46-55,Oct.2001. [56] L.Kleinrock,QueueingSystems,volumeI.NewYork:JohnWiley&Sons,1975 [57] L.Kleinrock,QueueingSystems,volumeII.NewYork:JohnWiley&Sons,1975 [58] Y.KoandN.Vaidya,“Location-aidedroutingLARinmobileadhocnetworks,”inProceedingsofACMMobiCom,pp.66-75,Dallas,Oct.1998. [59] Z.Kong,D.H.K.Tsang,andB.Bensaou,“PerformanceanalysisofIEEE802.11econtention-basedchannelaccess,”IEEEJournalonSelectedAreasinCommunica-tions,vol.22,no.10,pp.2095-2106,Dec.2004. [60] D.Levine,I.Akyildiz,andM.Naghshineh,“Aresourceestimationandcallad-missionalgorithmforwirelessmultimedianetworksusingshadowclusterconcept,”IEEE/ACMTransactionsonNetworking,vol.5,no.1,pp.1-12,Feb.1997. [61] B.Li,L.Li,B,Li,andX.Cao,“Onhandoffperformanceforanintegratedvoice/datacellularsystem,”ACMWirelessNetworks,vol.9,no.4,pp.393-402,Jul.2003. [62] B.Li,C.Lin,andS.T.Chanson,“Analysisofahybridcutoffpriorityschemeformultipleclassesoftrafcinmultimediawirelessnetworks,”ACMWirelessNet-works,vol.4,no.4,pp.279-290,Jul.1998. [63] B.Li,L.Yin,K.Y.Wong,andS.Wu,“Anefcientandadaptivebandwidthallo-cationschemeformobilewirelessnetworksusinganon-linelocalestimationtech-nique,”ACMWirelessNetworks,vol.7,no.2,pp.107-116,Mar./Apr.2001. [64] J.Li,J.Jannotti,D.DeCouto,D.Karger,andR.Morris,“Ascalablelocationserviceforgeographicadhocrouting,”inProceedingsofACMMobiCom,pp.120-130,Boston,Aug.2000.

PAGE 147

134 [65] W-HLiao,Y-CTseng,andK-PShih,“ATDMA-basedbandwidthreservationpro-tocolforQoSroutinginawirelessmobileadhocnetwork,”inProceedingsofIEEEICC,pp.3186-3190,NewYork,Apr.-May2002. [66] W.-H.Liao,Y.-C.Tseng,S.-L.Wang,andJ.-P.Sheu,“AMulti-PathQoSRoutingProtocolinaWirelessMobileAdHocNetwork,”TelecommunicationSystems,vol.19,no.3-4,pp.329-347,2002. [67] C.R.Lin,“Admissioncontrolintime-slottedmulithopmobilenetworks,”IEEEJournalonSelectedAreasinCommunicationsvol.19,no.10,pp.1974-1983,Oct.2001. [68] C.R.LinandJ.-S.Liu,“QoSroutinginadhocwirelessnetworks,”IEEEJournalonSelectedAreasinCommunications,vol.17,no.8,pp.1426-1438,Aug.1999. [69] W.Liu,X.Chen,Y.Fang,andJ.Shea,“Courtesypiggybacking:supportingdiffer-entiatedservicesinmultihopmobileadhocnetworks,”IEEEtransactionsonMobileComputing,vol.3,no.4,Oct.-Dec.2004. [70] S.Mangold,S.Choi,P.May,O.Klein,G.Hietz,andL.Stibor,“IEEE802.11ewirelessLANforqualityofservice,”inProceedingsofEuropeanWireless'02,pp.32-39,Florence,Italy,Feb.2002. [71] C.Oliver,J.B.Kim,andT.Suda,“Anadaptivebandwidthreservationschemeforhigh-speedmultimediawirelessnetworks,”IEEEJournalonSelectedAreasinCom-munications,vol.16,no.6,pp.858-874,Aug.1998. [72] OPNETTechnologies,OPNETModeler,http://www.opnet.com/products/modeler/home.html,Jun.2005. [73] M.NaghshinehandM.Schwartz,“Distributedcalladmissioncontrolinmo-bile/wirelessnetworks,”IEEEJournalonSelectedAreasinCommunications,vol.16,no.4,pp.711-717,May.1996. [74] S.NesargiandR.Prakash,“Distributedwirelesschannelallocationinnetworkswithmobilebasestations,”IEEETransactionsonVehicularTechnology,vol.51,no.6,pp.1407-1421,Nov.2002. [75] B.W.ParkinsonandS.W.Gilbert,“NAVSTAR:globalpositioningsystem-tenyearslater,”ProceedingsoftheIEEE,vol.71,no.10,pp.1177-1186,Oct.1983. [76] H.G.PerrosandK.M.Elsayed,“Calladmissioncontrolschemes:areview,”IEEECommunicationMagazine,vol.34,no.11,pp.82-91,Nov.1996. [77] P.Ramanathan,K.M.Sivalingam,P.Agrawal,andS.Kishore,“Dynamicresourceallocationschemesduringhandoffformobilemultimediawirelessnetworks,”IEEEJournalonSelectedAreasinCommunications,vol.17,no.7,pp.1270-1283,Jul.1999.

PAGE 148

135 [78] R.Ramjee,D.Towsley,andR.Nagarajan,“Onoptimalcalladmissioncontrolincellularnetworks,”ACMWirelessNetworks,vol.3,no.1,pp.29-41,Mar.1997. [79] T.S.Rappaport,WirelessCommunications:principlesandpractice.2ndEdition,UpperSaddleRiver:Prentice-Hall,Inc.2002. [80] M.Schwartz,“Networkmanagementandcontrolissuesinmultimediawirelessnet-works,”IEEEPersonalCommunication,vol.2,no.3,pp.8-16,Jun.1995. [81] S.-T.SheuandT.-F.Sheu,“Abandwidthallocation/sharing/extensionprotocolformultimediaoverIEEE802.11adhocwirelessLANs,”IEEEJournalonSelectedAreainCommunications,vol.19,no.10,pp.2065-2080,Oct.2001. [82] S.SinghandJ.Kurose,“Electing'good'leaders,”J.Par.Distr.Comput.,vol.21,pp.184-201,May1994. [83] J.L.SobrinhoandA.S.Krishnakumar,“Real-timetrafcovertheIEEE802.11mediumaccesscontrollayer,”BellLabsTechnicalJournal,pp.172-187,Autumn1996. [84] 3GPartnershipProject,http://www.3gpp.org/,June2005. [85] 3GPartnershipProject2,http://www.3gpp2.org/,June2005. [86] W.Tuttlebee,“Software-denedradio:facetsofadevelopingtechnology,”IEEEPersonalCommunications,vol.6,no.2,pp.38-44,Apr.1999. [87] U.S.CoastGuardNavigationCenter,GPSGeneralInformation,http://www.navcen.uscg.gov/gps/default.htm,May2003. [88] A.Veres,A.T.Campbell,M.Barry,andL-H.Sun,“Supportingservicedifferentia-tioninwirelesspacketnetworksusingdistributedcontrol,”IEEEJournalonSelectedAreasinCommunications,vol.19,no.10,pp.2094-2104,Oct.2001. [89] H.Wu,Y.Peng,K.Long,S.Cheng,andJ.Ma,“PerformanceofreliabletransportprotocoloverIEEE802.11wirelessLAN:analysisandenhancement,”inProceed-ingsofIEEEINFOCOM,pp.599-607,NewYork,Jun.2002. [90] S.Wu,K.Y.Wong,andB.Li,“AdynamiccalladmissionpolicywithprecisionQoSguaranteeusingstochasticcontrolformobilewirelessnetworks,”EEE/ACMTransactionsonNetworking,vol.10,no.2,pp.257-271,Apr.2002. [91] Y.Xiao,“EnhancedDCFofIEEE802.11etosupportQoS,”inProceedingsofIEEEWCNC,pp.1291-1296,NewOrleans,Louisiana,Mar.2003. [92] Y.Xiao,H.Li,andS.Choi,“ProtectionandguaranteeforvoiceandvideotrafcinIEEE802.11eWirelessLANs,”inProceedingsofIEEEINFOCOM,pp.2152-2162,HongKong,China,Mar.2004.

PAGE 149

136 [93] Z.Ye,S.V.Krishnamurthy,andS.K.Tripathi,“Aframeworkforreliableroutinginmobileadhocnetworks,”inProceedingsofIEEEINFOCOM,pp.270-280,SanFrancisco,Mar.-Apr.2003. [94] L.Yin,B.Li,Z.Zhang,andY.-B.Lin,“Performanceanalysisofadual-thresholdreservationDTRschemeforvoice/dataintegratedmobilewirelessnetworks,”inProceedingsofIEEEWCNC,pp.258-262,Chicago,Sep.2000. [95] J.Yoon,M.Liu,andB.Noble,“Soundmobilitymodels,”inProceedingsofACMMobiCom,pp.205-216,SanDiego,Sep.2003. [96] H.Zeng,Y.Fang,andI.Chlamtac,“CallblockingperformancestudyforPCSnetworksundermorerealisticmobilityassumptions,”TelecommunicationsSystems,vol.19,no.2,pp.125-146,Feb.2002. [97] H.Zhai,X.Chen,andY.Fang,“HowwellcantheIEEE802.11wirelessLANsupportqualityofservice,”toappearinIEEETransactionsonWirelessCommuni-cations. [98] H.Zhai,X.Chen,andY.Fang,“Acalladmissionandratecontrolschemeformul-timediasupportoverIEEE802.11wirelessLANs,”toappearinACMWirelessNet-works. [99] H.Zhai,Y.Kwon,andY.Fang,”PerformanceanalysisofIEEE802.11MACproto-colsinwirelessLANs,”JournalofWirelessCommunicationsandMobileComput-ing,vol.4,pp.917-931,Dec.2004. [100] H.Zhang,“Servicedisciplinesforguaranteedperformanceserviceinpacket-switchingnetworks,”ProceedingsoftheIEEE,vol.83,no.10,pp.1374-1396,Oct.1995. [101] L.Zhang,S.Deering,D.Estrin,S.Shenker,andD.Zappala,“RSVP:anewresourceReSerVationProtocol,”IEEENetwork,vol.7,no.5,pp.8-18,Sep.1993. [102] C.ZhuandM.S.Corson,“QoSroutingformobileadhocnetworks,”inProceedingsofIEEEINFOCOM,pp.958-967,NewYork,Jun.2002.

PAGE 150

BIOGRAPHICALSKETCHXiangChenreceivedhisB.E.andM.E.degreesinelectricalengineeringfromShang-haiJiaoTongUniversity,Shanghai,China,in1997and2000,respectively.FromFeb2000toJuly2000,heworkedasanMTSmemberoftechnicalstaffinBellLaboratories,Beijing,China.HeiscurrentlyworkingtowardthePh.D.degreeintheDepartmentofElectricalandComputerEngineeringattheUniversityofFlorida.Hisresearchinterestsincluderesourcemanagement,mediumaccesscontrol,andQoSinwirelessnetworks.HeisastudentmemberofIEEEandamemberofTauBetaPi. 137