<%BANNER%>

Enhanced Distributed Multimedia Services Using Advanced Network Technologies

Permanent Link: http://ufdc.ufl.edu/UFE0042024/00001

Material Information

Title: Enhanced Distributed Multimedia Services Using Advanced Network Technologies
Physical Description: 1 online resource (104 p.)
Language: english
Creator: Chung, Sung
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: architecture, content, design, distributed, fcal, high, loop, msg, mst, multimedia, multiple, p2p, pvr, quality, scalable, services, sharing, systems
Computer and Information Science and Engineering -- Dissertations, Academic -- UF
Genre: Computer Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: A variety of enhanced multimedia services, including multimedia live streaming and high-quality content sharing, have been enabled by the recent advances in network, storage, and compression technologies. Especially, the improvement of the network technologies has allowed the organization of a peer-to-peer (P2P) network where people can easily share their content with others. Integrated with the progress in multimedia technology, the advent of new electronic devices has accelerated to achieve those advanced multimedia services efficiently. In particular, a personal video recorder (PVR), one of those electronic devices, has emerged as an effective peer device in a P2P network, since it can store broadcast TV programs on its own embedded hard disk. In this dissertation, we develop an efficient network architecture such that all users can access high-quality multimedia content easily and the system can support various channels of access to the content. In order to achieve these goals, we adopt two leading technologies in our architecture in addition to a pool of disks as a backup storage; a fiber channel arbitration loop (FC-AL) as a reliable and broadband network connection and PVR as a peer in a P2P network. We have thus demonstrated the feasibility of a PVR-based network architecture by supporting the high-quality content sharing and distribution using the FC-AL. In fact, the promotion of the storage pool and the PVR capability on the FC-AL loop has enabled the system to become quite feasible for serving the emerging P2P streaming services. Nevertheless, the suggested PVR-based FC-AL network architecture has a critical limitation in terms of the number of attachable PVRs, since the FC-AL is intrinsically based on a single loop organization and the single loop only allows 127 attachable devices. We therefore develop a novel multiple-loop architecture providing a scalable network organization without using expensive switches. To connect multiple loops in our architecture, we then introduce shared disks that relay TV programs between loops like bridges. When extending into a multiple-loop architecture using the shared disks, it is realized that a topology design for the multiple-loop architecture has a major effect on the imposed total traffic in the whole system. We thus present and compare all possible topologies including linear, ring, edge-added, and complete graph (CG) topology. Finally, the analysis and experimental results reveal that the CG topology among all the possible topologies provides the best scalability. In addition, we recognize that, as the network size grows using the multiple-loop architecture, popular programs tend to be stored redundantly in many PVRs due to users' viewing skewness, thereby wasting the storage space that could otherwise be used to store additional TV programs. Thus, we also propose to extend program storage hours in terms of the whole system by presenting efficient storage saving algorithms for both PVRs and a backup storage. Through extensive simulations, we finally show that our proposed schemes significantly extend program storage hours by an average of 69.7%. Lastly, we present a practically constructible architecture solution, named MSG, which maintains both advantages of CG and MST. Thus, the MSG can be expected to be employed in a real constructible network infrastructure by its balance between system performance and cost.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Sung Chung.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Liu, Chien-Lian.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0042024:00001

Permanent Link: http://ufdc.ufl.edu/UFE0042024/00001

Material Information

Title: Enhanced Distributed Multimedia Services Using Advanced Network Technologies
Physical Description: 1 online resource (104 p.)
Language: english
Creator: Chung, Sung
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: architecture, content, design, distributed, fcal, high, loop, msg, mst, multimedia, multiple, p2p, pvr, quality, scalable, services, sharing, systems
Computer and Information Science and Engineering -- Dissertations, Academic -- UF
Genre: Computer Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: A variety of enhanced multimedia services, including multimedia live streaming and high-quality content sharing, have been enabled by the recent advances in network, storage, and compression technologies. Especially, the improvement of the network technologies has allowed the organization of a peer-to-peer (P2P) network where people can easily share their content with others. Integrated with the progress in multimedia technology, the advent of new electronic devices has accelerated to achieve those advanced multimedia services efficiently. In particular, a personal video recorder (PVR), one of those electronic devices, has emerged as an effective peer device in a P2P network, since it can store broadcast TV programs on its own embedded hard disk. In this dissertation, we develop an efficient network architecture such that all users can access high-quality multimedia content easily and the system can support various channels of access to the content. In order to achieve these goals, we adopt two leading technologies in our architecture in addition to a pool of disks as a backup storage; a fiber channel arbitration loop (FC-AL) as a reliable and broadband network connection and PVR as a peer in a P2P network. We have thus demonstrated the feasibility of a PVR-based network architecture by supporting the high-quality content sharing and distribution using the FC-AL. In fact, the promotion of the storage pool and the PVR capability on the FC-AL loop has enabled the system to become quite feasible for serving the emerging P2P streaming services. Nevertheless, the suggested PVR-based FC-AL network architecture has a critical limitation in terms of the number of attachable PVRs, since the FC-AL is intrinsically based on a single loop organization and the single loop only allows 127 attachable devices. We therefore develop a novel multiple-loop architecture providing a scalable network organization without using expensive switches. To connect multiple loops in our architecture, we then introduce shared disks that relay TV programs between loops like bridges. When extending into a multiple-loop architecture using the shared disks, it is realized that a topology design for the multiple-loop architecture has a major effect on the imposed total traffic in the whole system. We thus present and compare all possible topologies including linear, ring, edge-added, and complete graph (CG) topology. Finally, the analysis and experimental results reveal that the CG topology among all the possible topologies provides the best scalability. In addition, we recognize that, as the network size grows using the multiple-loop architecture, popular programs tend to be stored redundantly in many PVRs due to users' viewing skewness, thereby wasting the storage space that could otherwise be used to store additional TV programs. Thus, we also propose to extend program storage hours in terms of the whole system by presenting efficient storage saving algorithms for both PVRs and a backup storage. Through extensive simulations, we finally show that our proposed schemes significantly extend program storage hours by an average of 69.7%. Lastly, we present a practically constructible architecture solution, named MSG, which maintains both advantages of CG and MST. Thus, the MSG can be expected to be employed in a real constructible network infrastructure by its balance between system performance and cost.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Sung Chung.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Liu, Chien-Lian.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0042024:00001


This item has the following downloads:


Full Text

PAGE 2

2

PAGE 3

3

PAGE 4

IwouldliketothankallpeoplewhoprovidedmewiththeirhelpduringmyPh.Dyears.Firstofall,Iwouldliketothankmyadvisor,Dr.JonathanC.L.Liuforallhissupportandencouragement.Withouthisconstantinspiration,thisdissertationwouldnothavebeenpossible.Thediscussionwithhimonanytopichasalsomademepleasantandrelaxed.Iamalsogratefultomysupervisorycommitteemembers,Dr.ShigangChen,Dr.DouglasD.DankelII,Dr.PaulFishwick,andDr.PaulW.Chunfortheirinvaluablesuggestionsandcommentsformyresearch.Inaddition,IwouldthankallmyKoreanfriendsandfamilieswhosharedhappymemoriesinGainesville.Lastbutnotleast,Itrulyappreciatemyparents,SangkabChungandMalnamSeo,whohavebeensupportingmeandhavealwaysstoodbehindmeformywholelife,heartfeltlybelievinginmewithoutanydoubtevenatamoment.Ialsoappreciatemysister,KyungaeChung,whohaspleasantlyhelpedmeoutwithherwarmheartallthetime.IwouldalsoliketothankallmyfamilymembersandfriendsinKorea. 4

PAGE 5

page ACKNOWLEDGMENTS .................................. 4 LISTOFTABLES ...................................... 7 LISTOFFIGURES ..................................... 8 LISTOFALGORITHMS .................................. 10 ABSTRACT ......................................... 11 CHAPTER 1INTRODUCTION ................................... 13 1.1Overview .................................... 13 1.2ProblemNature ................................. 18 1.2.1ProblemDenitionsandRequirements ................ 18 1.2.2ProposedSchemes ........................... 18 1.2.2.1DesignofMultiple-LoopArchitecture ............ 19 1.2.2.2AnEfcientStorageScheme ................ 20 1.2.2.3APracticallyConstructibleMultiple-LoopArchitecture .. 21 1.3TechnicalBackground ............................. 22 1.3.1Multimedia-enabledSmallAreaNetwork ............... 22 1.3.2FiberChannelArbitrationLoop .................... 23 1.4OutlineofDissertation ............................. 24 2ASCALABLEPVR-BASEDCONTENTSHARINGARCHITECTURE ...... 26 2.1Motivation .................................... 26 2.2RelatedWork .................................. 28 2.3AScalableTVContentSharingArchitecture ................. 30 2.3.1SingleLoopArchitecture ........................ 30 2.3.2MultipleLoopArchitecture ....................... 31 2.4DesignofaScalableLoopTopology ..................... 32 2.4.1LinearTopology ............................. 34 2.4.2RingTopology .............................. 36 2.4.3CompleteTopology ........................... 37 2.4.4Edge-AddedTopology ......................... 38 2.4.5Multiple-InterfacedSharedDisks ................... 38 2.5ExperimentalEvaluation ............................ 42 2.5.1ScalabilityComparisonamongDifferentTopologies ......... 42 2.5.2ImpactoftheNumberofInterfacesperDisk ............. 45 2.6Summary .................................... 46 5

PAGE 6

..... 48 3.1Motivation .................................... 48 3.2RelatedWork .................................. 49 3.3TVContentDistributionArchitectureforCommunityNetworks ....... 50 3.4ExtendingContentStorageHours ...................... 53 3.4.1DesignIssuesforStorageEfciency ................. 54 3.4.2ImpactofDuplicatedStorageofPrograms .............. 54 3.4.3StorageSavingSchemes ....................... 57 3.4.4ReplacementSchemes ........................ 59 3.5ExperimentalEvaluation ............................ 61 3.5.1EffectivenessofProposedSchemesinaSingleLoop ........ 62 3.5.2EffectivenessofProposedSchemesinMultipleLoops ....... 63 3.5.3EffectivenessofOurProposedArchitecture ............. 67 3.5.4ImpactofPVRs'StoragePortionforTime-Shifting .......... 67 3.5.5ImpactofStorageCapacity ...................... 68 3.6Summary .................................... 69 4ANMST-BASEDNETWORKARCHITECTUREFORSHARINGBROADCASTTVPROGRAMS ................................... 71 4.1Motivation .................................... 71 4.2RelatedWork .................................. 73 4.3TVContent-SharingArchitecture ....................... 73 4.3.1MultipleLoopArchitecture ....................... 73 4.4EnhancedMultiple-loopNetworkArchitecture ................ 78 4.4.1Overview ................................ 78 4.5ProblemDenitionandFormulation ..................... 80 4.5.1ProblemStatement ........................... 80 4.5.2MinimumSpanningTree ........................ 83 4.5.3MinimumSpanningTree-basedGraph ................ 84 4.6PerformanceEvaluations ........................... 87 4.6.1AverageLoopSize ........................... 89 4.6.2TotalTrafcandAverageRejectRatio ................ 90 4.7Summary .................................... 92 5CONCLUSIONSANDFUTUREWORK ...................... 94 REFERENCES ....................................... 97 BIOGRAPHICALSKETCH ................................ 104 6

PAGE 7

Table page 2-1Trafconeachedgeinasixlooplineartopology ................. 35 2-2Trafconeachedgeinasixloopringtopology .................. 36 2-3mvaluesinthecompletetopology ......................... 39 2-4Simulationparameters ................................ 41 3-1PVRs'ContributableStorageinthePVR-basedFC-ALsystem ......... 56 3-2Symbolsforreplacementalgorithms ........................ 58 3-3Simulationparameters ................................ 62 3-4Theratioofpvrsandnetworkdisksinacommunitynetwork ........... 66 4-1Simulationparameters ................................ 88 7

PAGE 8

Figure page 1-1OverviewofTVcontentbroadcastingandsharingsystem ............ 15 2-1OverallTVcontentdistributionarchitecture .................... 29 2-2Examplesofmultiplelooparchitectures ...................... 31 2-3Examplesoftopologywithsixloops ........................ 33 2-4Examplesofcompletetopologiesusingmultiple-interfacedshareddisks .... 40 2-5Totaltrafcofdifferenttopologiesaccordingtothenumberofloops ....... 43 2-6Numberofshareddisksrequiredbydifferenttopologiesaccordingtothenumberofloops ........................................ 44 2-7Numberofattacheddevicesofdifferenttopologiesaccordingtothenumberofloops ........................................ 44 2-8Numberofattacheddevicesinthecompletetopologyaccordingtothenumberofinterfacesperdisk ................................. 46 2-9Averageblockoverheadtimeaccordingtothenumberofinterfacesperdisk 47 3-1Examplesofmultiplelooparchitectures ...................... 52 3-2Probabilityofstoringeachprogramaccordingtoitspopularity .......... 55 3-3EffectsofPVRsandnetworkdisks ......................... 64 3-4EffectsofPaliveandthreshold ............................ 65 3-5Effectivenessofourproposedarchitecture ..................... 66 3-6ImpactofPVRs'storageportionfortime-shifting ................. 68 3-7Impactofstoragecapacity .............................. 69 4-1OverallTVcontentdistributionarchitecture .................... 74 4-2HowtorelayaTVprogramusingshareddisksinatriple-loop .......... 76 4-3ExamplesofCG,MST,andMSGarchitectures .................. 79 4-4MSTformulations .................................. 82 4-5MSGformulations .................................. 84 4-6Flowchartforload-balancedMSG ......................... 86 8

PAGE 9

.................. 89 4-8Averageloopsize .................................. 90 4-9Totaltrafcandaveragerejectratio ......................... 91 9

PAGE 10

Algorithm page 3-1Programplacementalgorithm(input:new program) 59 3-2ReplacementalgorithmforPVRs(input:new program) 60 3-3Replacementalgorithmfornetworkdisks(input:new program) 61 10

PAGE 11

Avarietyofenhancedmultimediaservices,includingmultimedialivestreamingandhigh-qualitycontentsharing,havebeenenabledbytherecentadvancesinnetwork,storage,andcompressiontechnologies.Especially,theimprovementofthenetworktechnologieshasallowedtheorganizationofapeer-to-peer(P2P)networkwherepeoplecaneasilysharetheircontentwithothers. Integratedwiththeprogressinmultimediatechnology,theadventofnewelectronicdeviceshasacceleratedtoachievethoseadvancedmultimediaservicesefciently.Inparticular,apersonalvideorecorder(PVR),oneofthoseelectronicdevices,hasemergedasaneffectivepeerdeviceinaP2Pnetwork,sinceitcanstorebroadcastTVprogramsonitsownembeddedharddisk. Inthisdissertation,wedevelopanefcientnetworkarchitecturesuchthatalluserscanaccesshigh-qualitymultimediacontenteasilyandthesystemcansupportvariouschannelsofaccesstothecontent.Inordertoachievethesegoals,weadopttwoleadingtechnologiesinourarchitectureinadditiontoapoolofdisksasabackupstorage;aberchannelarbitrationloop(FC-AL)asareliableandbroadbandnetworkconnectionandPVRasapeerinaP2Pnetwork.WehavethusdemonstratedthefeasibilityofaPVR-basednetworkarchitecturebysupportingthehigh-qualitycontentsharinganddistributionusingtheFC-AL.Infact,thepromotionofthestoragepoolandthePVR 11

PAGE 12

Nevertheless,thesuggestedPVR-basedFC-ALnetworkarchitecturehasacriticallimitationintermsofthenumberofattachablePVRs,sincetheFC-ALisintrinsicallybasedonasinglelooporganizationandthesinglelooponlyallows127attachabledevices. Wethereforedevelopanovelmultiple-looparchitectureprovidingascalablenetworkorganizationwithoutusingexpensiveswitches.Toconnectmultipleloopsinourarchitecture,wethenintroduceshareddisksthatrelayTVprogramsbetweenloopslikebridges.Whenextendingintoamultiple-looparchitectureusingtheshareddisks,itisrealizedthatatopologydesignforthemultiple-looparchitecturehasamajoreffectontheimposedtotaltrafcinthewholesystem.Wethuspresentandcompareallpossibletopologiesincludinglinear,ring,edge-added,andcompletegraph(CG)topology.Finally,theanalysisandexperimentalresultsrevealthattheCGtopologyamongallthepossibletopologiesprovidesthebestscalability. Inaddition,werecognizethat,asthenetworksizegrowsusingthemultiple-looparchitecture,popularprogramstendtobestoredredundantlyinmanyPVRsduetousers'viewingskewness,therebywastingthestoragespacethatcouldotherwisebeusedtostoreadditionalTVprograms.Thus,wealsoproposetoextendprogramstoragehoursintermsofthewholesystembypresentingefcientstoragesavingalgorithmsforbothPVRsandabackupstorage.Throughextensivesimulations,wenallyshowthatourproposedschemessignicantlyextendprogramstoragehoursbyanaverageof69.7%. Lastly,wepresentapracticallyconstructiblearchitecturesolution,namedMSG,whichmaintainsbothadvantagesofCGandMST.Thus,theMSGcanbeexpectedtobeemployedinarealconstructiblenetworkinfrastructurebyitsbalancebetweensystemperformanceandcost. 12

PAGE 13

75 80 81 ],peoplecaneasilyuseandutilizeconsumerelectronicdevicestoreceiveandstorehigh-qualityTVprogramsorvideotitles.Typicalexamplesoftheelectronicdevicesarevideocassetterecorders(VCRs),DVDplayer/recorderanddigitalvideorecorders(DVRs)withaharddisksuchasTiVo. Atthesametime,theadvancesinnetworkingtechnologieshaveenabledpeopletoexperiencevariouschannelsforenjoyinglivestreamingservices,suchasbroadcastingTVprograms,andtosharetheircontentmuchmoreefcientlyinawaythatausercanaccessotherusers'content,andviceversa.Inparticular,theprogressofthenetworktechnologieshaveacceleratedtodistributehigh-qualitymultimediacontentsandtosharetheircontentsasapeer-to-peer(P2P)network. Fundamentally,thedistributedmultimediaservicestrytodeliverlivemultimediacontents,i.e.,streamingvideo/audio,vianetworksinrealtime.Thus,therapidadvancesincomputingtechnology,compressiontechnology,largebandwidthstoragedevices,andhighspeednetworktechnologyhavemadeitfeasibletoproviderealtimemultimediaservicesoverthenetworks.Inordertoachievethisgoal,variousissueshavetobeconsideredsothatanendusercaneasilyenjoythelivemultimediaservices.Hence,themaindistributedmultimediaresearchissuesareasfollows:multimediacontent 13

PAGE 14

24 58 66 ]. 20 69 77 82 ]. 2 27 32 ]. 23 26 29 40 72 ]. WeconsiderthesefundamentalfactstodevelopaneffectiveTVcontentsharinganddistributionnetworkarchitecture.Thus,inordertohandlethebroadcastTVprograms,oursystemsupportstheMPEG-2compressioninadditiontothereliableandbroadbandnetworktechnologybywhichournetworkarchitecturecandeliverHD-qualityTVprograms.Furthermore,ournetworkarchitecturemaintainsminimizedserver-functionalsystemcomponentsforatargetprogramroutingandabackupstorage.Thatis,inordertosupportanappropriateQoScontrol,oursystemprovidesthenecessaryinformationtondandretrievearequestedprogram.Also,adistributed/networkedsystemstoragecanofferthebackupstoragefunctioninoursystem.Thus,oursystemcanprovidethementionedfourkeyfunctionssothatitcanserveeffectivelyasadistributedmultimediasystem. 14

PAGE 15

BP2Pnetworkinasinglebuilding CSmallareaP2Pnetwork OverviewofTVcontentbroadcastingandsharingsystem Specically,Fig. 1-1A illustratespossibleTVcontentdistributionsusingtheadvancednetworkarchitecturedemonstratedinFig. 1-1B ,whereuserscaneasilyobtainvariousTVprogramsandtheycansharetheircontentwithothersusinghomecontent-storingdevices.Particularly,peopleenjoytheTVprogramsastheyarebroadcastorfromtheTV-contentprovidingserverviaaglobalnetwork,i.e.,theInternet.Inaddition,smallareanetworksinthecommunitycanbeformedasP2P 15

PAGE 16

1-1C .TheP2Pnetworkprovidercanserveseveralwaysforcontent-providingwiththepoolofstoragedevicesforlargecapacitiesandreliablebackupstorageswhilealluserswithintheP2Pnetworkscandocontent-sharingbetweenthepeers. HowtobuildtheseP2Pnetworkswitheffectiveschemesisthemajorthemeofthisstudy.Itisindeedanimportanttopicsincetherehavebeenseveralresearchefforts[ 31 59 ]thathavetriedtofacilitateallofthosestate-of-arttechnologiestogether,suchasadvancednetworktechnologiesandelectricappliances,forenhancedmultimediaservices.However,thesesystemsdonotnecessarilyguaranteetheoverallsystemperformanceandefciencysincetheyarebasedonInternetinfrastructure.Viaapilotstudy[ 38 ]basedontheFiberChannelArbitrationLoop(FC-AL),wehavedemonstratedthefeasibilityofaPVR-basednetworkarchitecturebysupportinghigh-qualitycontentsharinganddistribution,wherePVRscanworkaspeersviatheFC-ALloop.TheFC-ALisoneofleadingnetworktechnologies,offeringsuchadvantagesaslargebandwidth(e.g.,upto1Gbps),longdistancecoverage(e.g.,10km),andafairnessarbitrationalgorithm,whichisatypicalstructuretoorganizestorageareanetworks(SANs)[ 56 70 73 74 ]. OneuniquedesignconsiderationthatwehavebuiltintothesystemisthepromotionofDigitalVideoRecorder(DVR)totheconceptofPersonalVideoRecorder(PVR).Traditionally,VCRuserswouldrecordthebroadcastprogramsonvideohomesystem(VHS)tapes.Similarly,today'suserscansimplystorethedigitalvideo,includingbroadcastmoviesandTVprograms,ontheembeddedharddiskintheDVRs.JustlikeVHStapeswithrecordedvideocontentcanbesharedwithfriends,theuser-storeddigitalvideoontheharddiskscanalsobesharedamongfriends.InsteadofphysicallysendingthetapesorDVDs,digitalvideocontentcanbesharedsimplybyon-the-ytransmission.Atthismoment,DVRsonthemarkethaveonlylimitedtransmissioncapability(mainlyfordownloadingtheprogramdetails).Itisexpectedthateventually 16

PAGE 17

DuetothepromotionofthestoragediskpoolandthePVRcapabilityonFC-ALloops,thesystembecomesquiteexibleonsupportingtheemergingP2Pstreamingservices.Usually,amultimediastreamingservicecanbeclassiedinoneofthefollowing:purestreaming,implicitduplication,andexplicitduplication.Inpurestreaming,avideoleisplayedinrealtimewithoutstoringontoalocalharddisk.Theimplicitduplicationspeciesthatacertainportionofthevideoisbuffered,inordertosupportaneffectivemediaplayingwithoutenduringadelayorajitter.Lastly,intheexplicitduplication,theentirevideoleshouldbedownloadedforplayingthele.OursystemcansupporttheimplicitandexplicitduplicationaswaysofP2Pcontent-sharing,sinceourarchitecturehasthecapabilitythatstoresaprogramusingtheembeddedlocaldiskinPVRsandtransmitsareceivedcontentinrealtimewiththehigh-speednetwork,i.e.,FC-AL. Inotherwords,aprogramcanbesharedinawaythataPVRisreceivingandtransmittingittotheotherPVRsatthesametime.Infact,withalladvancedfunctionscombined,PVRshavethepotentialstoemergeasoneoftheadvancedmultimediaelectricappliancesthatsupportvariousadvancedservices,suchasVCR-likeoperations,electronicprogramguides(EPGs),andtime-shiftingTVprograms.Inparticular,weareinterestedinanintensiveinvestigationonthedegreeoftime-shift.Fundamentally,PVRsareabletostorerealtimebroadcastTVprogramsontheirownembeddedharddisks.Thus,theadventofPVRshasledtothechangeofpeople'sTVviewingpatterns[ 52 ]becausepeoplecanbecomeindependentofthespecicTVbroadcastingtimewiththePVR'stime-shiftingfunctionality.Signicantly,thePVRshave 17

PAGE 18

1-1 1.2.1ProblemDenitionsandRequirements Moreover,tobuildupamultimedia-service-enablednetworkarchitecture,weshouldadoptanappropriatebroadbandnetworkasanetworkconnection.Inaddition,eachusercanhaveamultimedia-compatibleappliancetoenjoyandsharethemultimediacontentseffectively. Furthermore,severalsystemcomponentsaswellasPVRsarealsonecessaryinordertocoordinateeachsystemcomponentandtosupportvariousaccesschannelsformultimediacontents,suchasbackupstorages.SinceeachPVRisunderanindividualuser'scontrol,wealsoneedtodevelopaschemetoutilizethewholesystemeffectively,regardlessofeachuser'sdiscretion. Lastbutnotleast,anetworkarchitecturehastobescalablewithoutalimitationonthenumberofaccommodableusers,sothatthenetworkarchitecturecanbeapplicableforvariouspurposes,whilestillsupportingitsoriginatedadvantages. 38 ]. 18

PAGE 19

34 ],whichmeansthatallattachablePVRsarelimitedatmostto127PVRs.Hence,thisindicatesthatthesingle-looparchitectureisnotsuitablewheremanyusershavetobeaccommodatedbythenetworkarchitecture,suchasadenselypopulatedbuildingoraregionalnetworkarealikeacommunityoracampus. Inthisdissertation,werstdevelopascalablePVR-basednetworkarchitecturebasedontheFC-AL-basednetworkwithmultipleloops.Therecanbeseveralconsiderationstodetermineabetterdesign.Fundamentally,atopologydesignaffectsthecongurationofamultiple-looparchitectureandtheorganizedmultiple-looparchitecturehasitsowncharacteristicsaccordingtotheselectedtopologydesign.Forexample,ifamultiple-looparchitectureisconguredbyalineartopology,themiddleloopsandshareddiskshavetobeinvolvedmorethanbothendloopsbuthavetheleastconnectivitybetweenloops.Ontheotherhand,aringtopologyimposeslesstrafcloadinthemiddleloopsandshareddisksbutrequiresmoreconnectivitycomparedtothelineartopology.Noticeably,thetopologydesignforamultiple-looparchitecturehasaneffectonthetrafcloadandtheconnectivity,whichiscloselyrelatedtothenumberofnecessaryshareddisks. Anadditionalobservationisthattheobtainablescalability,i.e.,thenumberofattachabledevicesincludingPVRs,networkdisks,andshareddisks,isaffectedbythenumberofrequiredshareddisks.Thatis,ifamultiple-looparchitectureneedsmoreshareddisks,themultiple-looparchitectureprovideslessscalabilityatthecostofdeployedshareddiskswithrespecttothegivennumberofloops.Inotherwords,ifa 19

PAGE 20

Therefore,weanalyzeandcomparepossiblemultiple-looptopologies:alineartopology,aringtopology,anedge-addedtopology,andacompletetopology.Withthegoalofndingthetopologytoprovidethemostscalability,wedeterminewhichtopologydesignisthemostsuitableforthemultiple-loopnetworkarchitecture,enablingmanymorePVRstoshareanddistributehigh-qualitycontentsbetweenthemwithupto10kmcoverage. Wethenintroduceashareddisktoconnectdifferentloopsandrelayarequestedprogramfromonetotheother,whichisakeysystemcomponentwithmultipleinterfacesinordertoorganizeamultiple-looparchitecture,workingasabridge.Weproposethemultiple-looparchitectureusingtheshareddisksbycomparingpossibletopologydesignsviathoroughanalysesandextensivesimulations.Finally,wedeterminewhichtopologydesignisthemostdesirableintermsofscalability,supportinghigh-qualitycontentsharinganddistributioneffectivelyatalowcost. 20

PAGE 21

Withourproposedschemes,wedemonstratethatourarchitecturecanextendprogramstoragehourssignicantlybyoursimulationresults.Infact,wehavefoundtheprogramstorageredundancytendstorapidlyincreasewhenmoreloopsandmorePVRsareattached.Therefore,agoodschemeshouldutilizethewholesystemstorageefcientlywhilereducingtostorepopularprogramsfromthewholesystempointofviews.Inthisstudy,wedemonstratethatanefcientstorage-savingschemecanbedesignedtostorefewerduplicatedprogramsinthewholesystem.Viaourthoroughsimulation,theexperimentalresultsrevealthatourstorage-savingschemecanutilizethesystemstoragecapacitiesefciently,whileextendingtheprogramstoragehoursbytheaverageof67.7%. 21

PAGE 22

Nevertheless,thesuggestedtopologyrequirestoomanyshareddisks,justforthetopologyconnectivity.Specically,iftherearenloopsinthearchitecture,thesystemdemandsatleastnC2shareddisksjustforthesystemorganization,therebymakingitlesslikelytobeabletoconstructarealnetworkarchitecture. Therefore,wepresentapracticallyconstructiblemultiple-looptopologywhichconsidersboththesystemperformanceandcost.Werstlydescribetheminimumspanningtree(MST)topologyforthesystemcost.However,itshowsunacceptablesystemperformancesinceitisatree-basedtopologywherethetrafcloadisconsecratedontheselectedxed-patheverytime.Consequently,weaddgraphcharacteristicstotheMSTtopologysothattheloaddistributioncanbeachievedforsatisfactorysystemperformance.Wenallydescribehowtocongureourproposedtopology,whatcharacteristicsithas,andhoweffectivelytoexposeitssuperiorityintermsofbothperformanceandcost. 1.3.1Multimedia-enabledSmallAreaNetwork 40 62 71 76 ]. 22

PAGE 23

47 57 84 ].However,theproxyserverfunctionsmainlydiscusshowtoreducethenumberofoutsiderequestsandhowtoabsorbtherequestinitsnetworkbycache/prefetchschemes. OurproposedsystemfurtheraddsP2Pfunctionswithinthenetworkarchitectureinadditiontothebasicfunctionsofasmallareanetwork.Thatis,theresidentsinthenetworkcannotonlyservetheircontentstotheirneighborsascontentprovidersbuttheycanalsoretrievetheirtargetprogramsfromothers,therebyprovidingmoresystemloaddistributionandreliablecontentsharing. 7 15 55 61 ]. OneoftheFC-ALfabrictopologiesistheFiberChannelArbitrationLoop(FC-AL)thatisemergingtobeemployedintheStorageAreaNetwork(SAN)architecture[ 4 21 44 74 ].IntheFC-ALtopology,theoutgoingandincomingberstoaportaresplitofftoattachtodifferentremoteports,suchthattheaggregationofbersformsaunidirectionalloopwhichpassesthrougheveryportexactlyonce.Thegeneralpictureof 23

PAGE 24

OnenotableadvantageofFC-AListhatitprovidesfairnesstoallattacheddevices.EachdeviceparticipatesinthearbitrationinordertoaccesstotheFiberChannel,andtherebynodevicesuffersstarvationduetothearbitrationprotocol.Inaddition,FiberChannelcancoverupto10kmdistancewhichishighlyappropriatetoorganizeasmallareanetworksuchasacommunityoracampus. Withtheseadvantages,therehasbeenresearchtoorganizeamultimedia-enablednetworkarchitectureusingFC-AL[ 12 13 22 ].However,thesesystemsaremainlyfocusedonthefeasibilityandperformanceattainability,notaimingforarealresidentnetworkarchitecture.Wehaveinvestigatedafurtherconsiderationfortheactualnetworkinfrastructure.Inotherwords,wearefocusinghowtomakeitfeasibletobescalablesothatmanymoreusercanbesupported,byovercomingtheintrinsicFC-ALaccommodatablelimitation.Thus,weconsideraneffectivesystemdesignintermsoftopology,andfurthermorewedevelopapracticallyconstructibletopologydesignwhichmakesabalancebetweensystemperformanceandcost. Chapter2describesourproposedFC-ALmultiple-looparchitecturesforsmallareanetworks.Weanalyzevarioustopologytypes,(e.g.,complete,ring,edge-added,andcompletetopology)todeterminethebetterarchitecturedesignintermsofscalability,i.e.,thenumberofattachabledevices.Withthoroughinvestigationandextensivesimulations,werevealthatthecompletetopologydesignprovidesthemostscalability,andtherebyismostappropriateforthemultiple-loopnetworkarchitectures. 24

PAGE 25

Chapter4explainsapracticallyconstructiblearchitecturesolutionwhichtriestondabalancebetweensystemperformanceandcost.Infact,theCGtopologycanbeimpracticalwhenthenumberofloopsbecomeslarge.Thus,anminimumspanningtree-basedgraphtopology,calledMSG,isdevisedwhichisderivedfromanMSTtopologybutalsoreectstheCGtopology'sadvantages.TheMSGthenrevealsitssuperioritybyshowingaconstantaverageloopsizeaswellasanacceptabletotaltrafcandrejectratiocomparedtobothGCandMST. Chapter5concludesthisdissertationregardingallthetopicsandpresentsourresearchdirectionsforfuturework. 25

PAGE 26

Inaddition,apersonalvideorecorder(PVR)isanelectronichomeappliancethatcanrecordbroadcastTVprogramsontoitsembeddedharddiskinadigitalformat[ 52 ].Unliketheconventionalvideocassetterecorder(VCR),itusesaharddiskasastoragemediumratherthanatape.However,itstillsupportsadvancedfeaturesincludingelectronicprogramguide(EPG)andtime-shifting(e.g.topauseliveTV),inadditiontoconventionalVCR-likeoperations,suchasfastforward,rewind,pause,andplay.Remarkably,thePVR'stime-shiftingfunctionalityhasamajoreffectonpeople'sTVwatchingpatterns.Thatis,thetime-shiftingfeatureallowspeopletobecomeindependentofaTVbroadcastingtime.Moreover,theycanevenpausealiveTVprogramorskipsomepartsofaliveTVshowduetothetime-shiftingfunctionality.Furthermore,aPVRhasthecapabilitytobeconnectedbyabroadbandnetwork,therebyservingasanetworkcomponenttosharestoredTVprogramswithotherPVRs.ThisfeasibilityofaPVR-basedP2PnetworkhasbeeninvestigatedandtestedintheShare-itproject[ 49 ]andtheShareTVbyNDS[ 35 ]. Therefore,weadoptthisPVRasamultimediadeviceineachhome,sothateachusercannotonlyenjoyvariousmultimediaservicesbutoursystemcanalsobeorganizedforaP2Pnetworkusingeachuser'sPVR.Thatis,oursystemassumesthateachuserhasaPVRsothathecanstorehisTVwatchingTVcontentsthatcanberetrievedbyotherusersintheproposednetworkarchitecturewhichservesasaP2Pnetwork. 26

PAGE 27

Toconnectmultipleloopsinourarchitecture,weintroduceshareddisksthatareresponsibleforrelayingTVprogramsbetweenloopslikebridges.Withshareddisks,multipleloopscanbeconguredbyvarioustypesoftopologiesinordertoconstructacommunitynetwork.Toseewhichtopologycanprovidethebestscalabilityintermsofthetotalnumberofdevicesthatcanbeattachedtoallloops,weanalyzefourpossibletypesoftopologies:linear,ring,edge-added,andcompletetopologywhereanodeandanedgerepresentaloopandashareddisk,respectively.Thefourtopologiesareclassiedaccordingtohowmanyadjacentedgeseachnodehas. Asthenumberoftheshareddisksincreases,thetotalnumberofdevicesattachedtoallloopsdecreasesbecausetheyareattachedtomorethanoneloopatthesametime.Thus,weneedtoreducethetotalnumberofshareddiskstoprovidebetterscalability.Wehaveobservedthatthesystem'stotalloadedtrafcaffectstherequirednumberofshareddisks.Inaddition,thetotaltrafcdependsonthesystem'stopologydesign,i.e.,whichtopologyisemployedforthewholesystem'scongurationdesign.Ultimately,thesystem'stopologydesigndeterminesthenecessarynumberofshareddiskstoorganizeaneffectivemultiple-looparchitecture.Thisisbecauseeachtopologygeneratesadifferentnumberofhopstoreachadestinationloopfromasourceloop.Inordertoseehoweachtopologyaffectsthetotaltrafc,weanalyzefourdifferenttopologies.Theanalysisrevealsthatthecompletetopologyhastheleasttotaltrafcamongfourtopologies.Thereasonforthisisthatanyloopcanalwaysreachanotherloopwithinaminimumnumberofhops,i.e.,onehop.Wealsoexaminetheimpactofthenumberofinterfacespershareddiskonthesystemscalability. 27

PAGE 28

Theremainderofthischapterisorganizedasfollows:Section 2.2 describesrelatedwork.Section 2.3 proposesascalableTVcontent-sharingarchitectureincludingshareddisksanddirectoryservers.Section 2.4 analyzestheimpactoftopologytypesofacommunitynetworkandthenumberofinterfacesperdiskonthesystemscalability.Section 2.5 presentsextensivesimulationresults.Finally,Section 2.6 offerssummarizations. 6 68 ],buttheyhavefocusedonhowtoorganizeacommunitynetworkratherthanhowtoconstructascalablearchitectureforacommunitynetworkincludingvarioustypesoftopologies. Inaddition,PVRs'impactonTVviewingpatternshasbeendescribedin[ 45 ]andvariousfunctionalitiesofPVRshavebeenstudiedin[ 43 83 ].However,theyconcentratedonPVR-specicapplicationssuchastime-shiftingandvideo-rateadaptation. Inthemeantime,theberchannelhasbeenproposedasoneofthestandardstoorganizethestorageareanetwork[ 34 ].Thus,itsfeasibilityandcapabilityhavebeenstudiedasahigh-speednetworkin[ 18 ].Additionally,theFC-AL-basedstoragesystemshavebeenstudiedformultimediaserverarchitecturesin[ 13 22 ].Moreover,the 28

PAGE 29

10 49 ],buttheydonotaddressthescalabilityissueofthesystem. Whenorganizingnetworktopologies,designissueshavebeenintroducedin[ 11 33 48 ].However,theystudiedautonomoussystemsorclustersastargetsystems. BAcommunitynetworkforalargenumberofdevices OverallTVcontentdistributionarchitecture 29

PAGE 30

2-1A illustratesastructureofanFC-ALsingleloopthatcanbeconstructedinasmallarea.Inordertoprovidehigh-qualitycontentsharinginrealtimeamongPVRs,itisessentialtoadoptahighspeednetworkasaninterconnectinginfrastructure.Theberchannel(FC)hasemergedasaleadingtechnologyduetoitsseveraladvantages.Firstofall,FCcanprovidehighbandwidth,i.e.,currently800Mbps,andlargecoverage,i.e.10km,whichareverysuitabletoorganizeacommunitynetworkwherealargenumberofPVRsareinvolved.Inaddition,theFC-ALemploysafairnessarbitrationalgorithmthatcanguaranteethereisnostarvationamongallattacheddevices.Therefore,weemployFC-ALtechnologytoconnectallsystemcomponentswithinacommunitynetwork[ 38 ]. OncePVRsbecomeequippedwithnetworkdevices,theycansharecontentsstoredontheirharddiskswithothers.Inotherwords,eachPVRcanworkasapeerasinaP2Pnetwork,especiallywithinacommunitynetwork.Thecommunitymemberscansharethehigh-qualityTVcontentsviathePVRsthatareconnectedtotheFC-AL.PVRsrecordliveTVprogramsthatuserswant.TheyalsocontinuetorecordbroadcastTVprogramswhichusershavewatchedforagivendurationsothatuserscantime-shifttheprograms. Asmentionedabove,however,thesingleloopcanaccommodateonly127devicesbecausetheFC-ALstandardallowsa7-bitaddressspace[ 34 ].Inotherwords,itisnotsuitableasacommunitynetworkifthecommunityhasmorethan127consumerdevices.Therefore,itisessentialtoextendthecommunitynetworkinaneconomical 30

PAGE 31

BFourandFiveloopconguration Examplesofmultiplelooparchitectures 2-1B .Toextendthesinglelooparchitecturetothemultiplelooparchitecture,weemploytwoadditionalsystemcomponents:oneisashareddiskandtheotherisadirectoryserver. 31

PAGE 32

34 ].SincedisksattachedtotheFC-ALareatleastdual-interfaced,wecanconnecteachinterfacetooneofmultipleloops. ThedirectoryservercoordinatesthecontentsharingbetweenPVRswithinaloopandamongloops.Locatedineveryloop,itkeepscollectingandexchangingtheinformationonwhichPVRsarestoringwhichprograms.Basedonthisinformation,itcanperformtheschedulingofrequestsforcontentsharing. Fig. 2-2 illustratesourproposedmultiplelooparchitecturewithtwo,three,four,andveloopcongurations.Notethateachlooprepresentsasinglelooparchitectureshownin 2-1A .Itcanbeseenthattheshareddisksworkasbridgesbetweenloops.Inordertoeffectivelyextendthearchitecturewithagivennumberofloops,itisimportanttodeterminehowmanyshareddisksshouldbeemployedbetweenloops.Themoredisksaredesignatedasshareddisks,theworsescalabilitywemayachieve.Moreover,itcanbeaperformancebottleneckwhentheyareoverloadedduetoimproperdesign. 32

PAGE 33

BEdge-addedandCompletetopology Examplesoftopologywithsixloops FromEq. 2 ,itisclearthatweshouldreducethetotalnumberofshareddiskstoprovidebetterscalability.Asmentionedabove,wehaveobservedthatthetotaltrafcloadedonallshareddisksdependsonthetypeoftopologyconstructingamultiplelooparchitecture.Thisisbecausetherequirednumberofhopstotransferdatabetweenaspecicpairofloopsdiffersaccordingtoeachtopology. Inordertoseehoweachtopologyaffectsthetotaltrafcloadedonallshareddisks,weillustratefourdifferenttopologycongurationswithsixloopsasshowninFig. 2-3 .Sincethecongurationswithalargernumberofloopsaremuchhardertoillustrateingures,weusethesix-loopcongurationsfordetailedanalysisofeachtopologyinthissection. 33

PAGE 34

2-3A and 2-3B showalinear,ring,edge-added,andcompletetopology,respectively.Anodeandanedgerepresentasingleloopandashareddisk,respectively.Wecomparethesetopologiesintermsofthetotaltrafcwithagivennumberofloops.Weusedual-interfacedshareddisksfortheconvenienceofcomparison.Wethenexaminetheimpactofthenumberofinterfacespershareddiskonthesystemscalability. 2-3A .Inthelineartopology,itcanbeseenthatthenumberofedgesislessbyonethanthenumberofnodes. Inourmultiple-looparchitecture,whenaPVRrequestsaprogram,thesystemrsttriestondtherequestedprogramwithinthelocalloopwheretherequestingPVRislocated.Iftheprogramisnotfoundinthelocalloop,i.e.,theprogramisstoredonlyinotherloops,therequestingPVRmustreceivetheprogramfromoneofotherloopsviaallshareddiskslocatedbetweenthetwoloops.Theroutingisdeterminedbasedonthetotalnumberofhopsbetweenasourceandadestinationloop,i.e.,theshortestpathischosenamongpossiblepaths.ItisassumedthattheprogramsarestoredevenlyamongPVRsandthepopularityofprogramsisuniformsothattheamountofdatatrafcbetweenallpairsofloopscanbefairlyevaluated. Inthelineartopology,itisobviousthattheedgescloselylocatedtoacenternodehandlemoretrafcbecausetheytendtobemoreinvolvedinrelayingprogramsbetweenloops.Forexample,inFig. 2-3A ,thecenteredge(i.e.,edgec)handlesthemosttrafcamongallveedgesduetothetaskofrelayingprograms.Thesecondandfourthedge(i.e.,edgebandedged)handlethesecondmosttrafcforthesamereason.Therstandthefthedge(i.e.,edgeaandedgee)havetheleasttrafc.Specically,whendenotingmarelativeunitoftrafc,theveryleftnode(i.e.node1)generatestrafcasmuchas(5/5)m,(4/5)m,(3/5)m,(2/5)m,and(1/5)mtoedgea,edgeb,edgec,edged, 34

PAGE 35

Trafconeachedgeinasixlooplineartopology 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m andedgee,respectively.Similarly,node2generatestrafcasmuchas(1/5)m,(4/5)m,(3/5)m,(2/5)m,and(1/5)mtoedgea,edgeb,edgec,edged,andedgee,respectively.Table 2-1 showsthetotaltrafconeachedgeinthelineartopologywithsixloops.Thetotaltrafconedgea,edgeb,edgec,edged,andedgeeare2m,(16/5)m,(18/5)m,(16/5)m,and2m,respectively.Thetotaltrafconalledgesiscomputedas14msimplybysummingthetrafconveedges. Ingeneral,wecanderivethetotaltrafcinthelineartopologyaccordingtothenumberofloopsasfollows: (n1)n1 2Xk=2(k2nk)35m,n:odd&n52444 (n1)n2 2Xk=2(k2nk)+n2 wherenisthenumberofloopsandmisarelativeunitoftrafc. 35

PAGE 36

2-3A ,allthenodeshavetwoedgesandwecannotdesignateanyedgeasacenteredge.Anodecanreachanyothernodewithinasmanyhopsashalfthetotalnodes. Table2-2. Trafconeachedgeinasixloopringtopology 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m Node4 0 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 5m 2-3B ,node1generatestrafcasmuchas(3/5)m,(2/5)m,(1/5)m,(1/5)m,and2/5)mtoedgea,edgeb,edgec,edgee,andedgef,respectively.Notethatwhenthereisatieindeterminingtheroutingpathintermsofthenumberofhops,wechoosetheloopintheclockwisedirection.Thus,edgeddoesnothaveanytrafcgeneratedbynode1duetothistie-breakingpolicybasedontheshortestpathrouting.Similarly,wecancalculatethetrafconotheredgesasshownin 36

PAGE 37

2-2 .Wecanthusgenerallyformulatethetrafcofeachnodeasfollows: (n1)n1 2Xk=1k=n+1 435m,n:odd242 (n1)n2 2Xk=1k+n=2 2(n1)35m=n2 WithEq. 2 ,wecanobtainthetotaltrafcintheringtopologysimplybymultiplyingthetrafcofeachedgebythenumberofedgesasfollows: 4nm=n(n+1) 4m,n:oddn2 First,wederivethetrafcofeachedge.InFig. 2-3B ,eachnodegenerates(1=5)mtrafctoallotheradjacentedges.Sinceeachedgereceivesthesametrafcfromtwoconnectednodes,thetrafcofeachedgeis(2=5)m. Thus,wecangenerallyderivethetrafcofeachedgeinthecompletetopologyasfollows: (n1)m(2) Sincethetotalnumberofedgesinthecompletetopologyisn(n1)=2,wecomputethetotaltrafcforalledgesbymultiplyingthetrafcofeachedgebythetotalnumberof 37

PAGE 38

(n1)mn(n1) 2=nm(2) ThetotaltrafcofthecompletetopologywithsixloopsinFig. 2-3B is6m.ComparedtothetotaltrafcofthelinearandringtopologywithsixloopsinFig. 2-3A ,amountingto14mand10.8m,respectively,itcanbeseenthatthecompletetopologyhastheleasttrafc. 2-3B However,itishardtoderivegeneralformulaeforthetotaltrafcintheedge-addedtopologiesduetotheirirregularities.Thus,wehaveperformedextensivesimulationstoinvestigatethetotaltrafcofallthetopologieswhereeachnodehas2to(n1)adjacentedgesincludingringandcompletetopology.ThesimulationresultswillbepresentedinSection 2.5 38

PAGE 39

Table2-3. 3 4 5 6 7 8 9 10 3 1 N/A N/A N/A N/A N/A N/A 6 3 1 N/A N/A N/A N/A N/A N/A 5loops 10 4 3 1 N/A N/A N/A N/A N/A 6loops 15 6 3 3 1 N/A N/A N/A N/A 7loops 21 7 4 3 3 1 N/A N/A N/A 8loops 28 11 6 4 3 3 1 N/A N/A 9loops 36 12 7 4 3 3 3 1 N/A 10loops 45 17 8 6 4 3 3 3 1 Werstderivetheminimumnumberofmultiple-interfacedshareddisksinthecompletetopologywhereanynodecanreachanyothernodewithonlyonehopasfollows: (k1) wheren,k,andMrepresentthenumberofloops,thenumberofinterfacesthatoneshareddisksupports,andtheminimumrequirednumberofshareddisks,respectively. Thecompletetopologycomprisingdual-interfacedshareddisksisaspecialcasewherek=2inEq. 2 .Table 2-3 illustratesMvalueswhenvaryingthenumberofloopsandinterfacesperdisk. Fig. 2-4 showsthreeexamplesofcompletetopologiesusingmultiple-interfacedshareddisks.Fig. 2-4A showsthecompletetopologywithfourloopsandthreethree-interfacedshareddisks.Inthistopology,shareddiska,b,andcconnectloop1/2/4,1/2/3,and2/3/4,respectively.Fig. 2-4A showsthecompletetopologywitheightloopsandfourve-interfacedshareddisks.Similarly,fourdisksconnecteightloops 39

PAGE 40

Examplesofcompletetopologiesusingmultiple-interfacedshareddisks usingveinterfacessothateverypairofloopscanbereachedwithonlyonehop.Fig. 2-4B showsthecompletetopologythatcanbeconstructedwithonlyoneshareddisk.Thisispossiblebecausetheshareddiskhasteninterfacesthatareabletoconnecttenloopsatatime.Wecanthusderivethetrafcofeachshareddiskwithkinterfacesasfollows: (n1)km(2) 40

PAGE 41

2 and 2 asfollows: (n1)km266666n(n1) (k1) ItcanbeseenfromEq. 2 and 2 thatthetotaltrafcofthecompletetopologywithmultiple-interfacedshareddisksissameasthatofdual-interfacedshareddisks.Thismayimplythatthetotaltrafcisnotaffectedbywhichtypeofshareddiskisemployed.However,notethatwemayneedmoreshareddisksasthetrafcofeachedgebecomesheaviereventhoughonlyoneshareddiskcanconnectallloopsasshownFig. 2-4B .Sinceeachshareddiskhasalimitationonhandlingtrafc,weshouldalsotakeprocessingcapabilityofeachshareddiskintoaccounttoestimatetherealscalability.Inthenextsection,wewillshowtheactualrequirednumberofshareddisksconsideringboththeprocessingcapabilityofeachshareddiskandthetrafcloadedonit. Table2-4. Simulationparameters datatransferrate 400MB/s propagationdelay 3.5ns/meter pernodedelay 240ns Disk capacity 300GB cache 32MB datatransferrate 2965MB/s seektime 4.7ms rotationallatency 3.0ms 41

PAGE 42

ItisassumedthateachTVprogramhasHDqualityof19.4Mbpsplaybackrateandis60minuteslong.Weemploydual-interfacedshareddisksbetweentwodistinctloopsinordertoorganizeamultiplelooparchitectureunlessotherwiseindicated.Wealsovarythenumberofinterfacespershareddisktoshowtheimpactontheoverallscalability.Theprogramsaredistributedevenlyamongloopsandtheprobabilitythateachprogramisrequestedisuniformsothattheamountofdatatrafcbetweeneachpairofloopscanbefairlycompared.TherequestrateofprogramplaybackfollowsaPoissondistribution.EachavailablePVRissuesrequestsfordifferentprogramsevery60minutes.Itisalsoassumedthat126PVRsareconnectedtoeachloopandhalfofthemonaverageareavailableataspecictime.Notethattheroutingbetweentwoloopsisdeterminedbythenumberofhopsandthetie-breakingpolicyasmentionedabove.ThedetailsofparametersusedfordisksandFC-ALarealsoillustratedinTable 2-4 Thesimulationresultsdemonstratethatthecompletetopologyprovidesthebestscalability,whichsupportsouranalysispresentedintheprevioussection. 42

PAGE 43

Figure2-5. Totaltrafcofdifferenttopologiesaccordingtothenumberofloops Fig. 2-5 showsthetotaltrafcaccordingtotopologytypewhilevaryingthenumberofnodes(i.e.,loops)fromthreetoten.Thetopologytypeisdeterminedbythenumberofaverageedgesconnectedtoeachnode.Forinstance,inthecaseofn=10,thetopologieswithtwoandnineedgespernodedenotetheringandcompletetopology,respectively.Thesixtopologieswiththreetoeightedgespernoderepresentallpossibleedge-addedtopologies.Itcanbeseenthat,asthenumberofedgespernodeincreases,thetotaltrafconeachtopologydecreases.Forexample,inthecaseofn=10,wheneachnodehastwoedges,i.e.,theringtopology,thetotaltrafcisapproximately1,579programsonaverage.Ontheotherhand,wheneachnodehasnineedges,i.e.,thecompletetopology,thetotaltrafcisonaverageonly577programs.Thistrendisconsistentwithallothertopologieswithdifferentnumbersofnodes.Thereasonforthis 43

PAGE 44

Numberofshareddisksrequiredbydifferenttopologiesaccordingtothenumberofloops Figure2-7. Numberofattacheddevicesofdifferenttopologiesaccordingtothenumberofloops 44

PAGE 45

Itisclearthatthetopologieswithlesstrafcrequirefewershareddisks.Thisimpliesthatwecanachievebetterscalabilitybecauseshareddisksareattachedtomorethanoneloopatthesametime.Fig. 2-6 illustratestherequirednumberofshareddisksforeachtopologytypewhilevaryingthenumberofnodes.Inthecaseofn=10,thecompletetopologyrequiresonly135shareddiskswhiletheringtopologyrequires320shareddisks.Notethatthenumberofshareddisksperedgeintheringtopologyismuchlargerthanthatinthecompletetopologyeventhoughthenumberofedgesintheringtopologyissmaller. Fig. 2-7 showsthetotalnumberofattacheddevicesforeachtopology.Asexpected,thecompletetopologycansupportthegreatestnumberofattacheddevicesamongallthetopologiestoconstructamultiplelooparchitecture.Forexample,inthecaseofn=10,thecompletetopologycanattach1,125deviceswhiletheringtopologysupports940devices,whichindicates19.68%improvedscalability.Itcanbealsoseenthatthecompletetopologyoutperformsalltheothertopologiesforallthecaseswithdifferentnumbersofnodes. 2-8 showstheimpactofthenumberofinterfacespershareddiskinthecompletetopologywhilevaryingthenumberofloops.Itcanbeseenthatthedifferenceinthenumberofattacheddevicesisnegligibleforallthenumbersofloopsfromtwototen.Forexample,inthecaseofn=10,thenumberofattacheddevicesis1,125whenemployingthedual-interfacedshareddiskswhilethesystemwithteninterfacedshared 45

PAGE 46

Numberofattacheddevicesinthecompletetopologyaccordingtothenumberofinterfacesperdisk diskssupports1,144.Thisindicatesonly1.69%differenceinscalabilityeventhoughtheaverageprocessingutilizationofshareddisksisquiteabitlowerwithdual-interfaceddisksthanwithteninterfaceddisks.Thisisbecause,asthenumberofinterfacesincreases,theoverheadincludinglooparbitrationincreasesrapidly.Fig. 2-9 illustratestheincreasedaverageone-block(512KB)processingtimeastheimposedoverhead,accordingtothenumberofinterfacespershareddisksinthe10-looparchitecture.Thus,wecanseethatmoreinterfaceddisksdonotcontributesignicantlytothesystemscalability. 46

PAGE 47

Averageblockoverheadtimeaccordingtothenumberofinterfacesperdisk expensiveswitchesforanetworkarea,wehaveproposedascalablemultiple-looparchitectureusingshareddisksthatrelaytherequestedprogramsfromonelooptotheother. Inordertodeterminewhichtopologyhasthebestscalability,weanalyzedthetotaltrafcforfourpossibletopologiessuchaslinear,ring,edge-added,andcompletetopology.Throughanalysisandextensivesimulations,weshowedthatthecompletetopologyhasthebestscalabilitybecauseitrequiresthefewestnumberofshareddisksbygeneratingtheleasttrafcamongthefourtopologies.Wealsoshowedthatthesystemscalabilityisnotaffectedbythenumberofinterfacesperdisksignicantly. Weshowedthattenloopsconnectedviashareddiskscanservemorethan1,120deviceswithoutemployingswitchdevices.OurproposedarchitectureisthereforeexpectedtobeaveryscalablecommunitynetworkcapableofsupportingmanymorePVRswiththeincreasednumberofloopssincetheberchannelstandardtechnicallyallowsupto10kmcoverage. 47

PAGE 48

14 ].Withtheshareddisksinthesystem,communicationbetweenthehosts/disksindifferentloopsbecomesfeasible.Therefore,ourdesignfocusonthesystemisshiftedtomakesurethegivenstoragespaceiseffectivelyutilized. UnliketheconventionalSCSI-basedserversthatmanagethevideoleallocationissuesinthedominatingrole,FC-AL-basedP2PnetworksprovidethesystemmoreexibilityinthateveryPVRcanequallyaccessallthedisksonthenetworks.Thuscontent-sharingamongthelocaldisksinPVRsisquitefeasible,whichallowsthesystemtobemoredynamicforusingthewholesystemstoragecomparedtotheserver-onlysystem.Thus,thePVR-basedFC-ALarchitectureingeneralshouldnotrequireasmanycopiesoftheprogramsastheconventionalSCSI-basedservers/systems.Therefore,PVR-basedFC-ALsystemsmayreducetherequirednumberofcopiesofprogramssignicantlyandeffectively. WebelievethattheoperationsofeachPVRshouldbeunderthecontrolofitsuser.UserscanthereforearbitrarilystoreorremoveanyprogramsontheirPVRsindependentlyofhowotherPVRsarestoringordeletingthesameprogram.AsthecommunitynetworkcanbeextendedwithaddedloopsinordertoaccommodatemorePVRs,however,thestorageofpopularprogramstendstobeduplicatedinmanyPVRs.ThisimpliesthatthecommunitynetworkoverallwouldwastethestoragespacethatcouldotherwisebeusedtostoreadditionalTVprograms.Inaddition,PVRscansupportthefeatureofauto-recordwiththeirpre-assignedlocalstorage,bywhichthePVRscanfunctionfortime-shiftingservicelikeTiVo[ 5 ].Consequently,thisimpliesthatwecan 48

PAGE 49

Wethereforeanalyzetheproblem,andproposenovelschemestoextendprogramstoragehoursinacommunitynetwork.Theseschemesincludethealgorithmtodeterminetheminimalnumberofcopiesoftheassociatedprograms,andstorage-savingalgorithmsforbothPVRsandnetworkdisks.TheseschemesjointlyconsiderthestoragesystemdesignissuesincludingaccessskewamongprogramsandPVRs'limitedavailability.Wethenperformextensivesimulationswhileconsideringthevarioussystemparametersinordertodemonstratetheeffectivenessofourproposedarchitectureandalgorithms.Thesimulationresultssuccessfullydemonstratethatourarchitecturecanextendprogramstoragehourssignicantly. Theremainderofthischapterisorganizedasfollows:Section 3.2 describespreviousworkrelatedtoourchapter.Section 3.3 explainstheoverallarchitectureforourproposedcommunitynetworks,especiallyFC-ALmultiplelooparchitecture,todeliverhigh-qualitymultimediadata.Section 3.4 explainsourstoragesavingschemesforPVRsandnetworkdisks.Section 3.5 illustratesextensivesimulationresults.Finally,Section 3.6 offersasummarization. 34 ].Thus,itsfeasibilityandcapabilityhavebeenstudiedasahigh-speednetwork[ 18 19 ]anditsattainableperformanceshavebeenanalyzed[ 30 60 ].TherehasbeenresearchonFC-AL-basedmultimediaserverarchitecturesusingFC-ALstoragesystems[ 12 13 22 ]. TherehasalsobeenliteratureonPVRsthatcanofferentertainmentservicesathomesuchthattheend-userscansharewhattheyarestoringintheirownPVRsinP2Pnetworks[ 3 36 49 79 ].Inaddition,high-qualitystreamingservicesbasedoncontentdistributionnetworkshavebeenexploited[ 9 16 53 ]althoughthoseservices 49

PAGE 50

39 ].However,thissystemonlyfocusedonthefeasibilityofthesystemarchitecturewiththesingle-loopFC-ALconguration.Thus,ithaslimitationsasacommunitynetworkinfrastructurebecauseitcanaccommodateonlyalimitednumberofcommunitymembers. Inaddition,asthenetworkandmultimediatechnologieshaveadvanced,themultimediastreaminghasbecomemuchmorefeasibleinavarietyofservicessuchasP2Psharing.MultimediastreamingoverP2Pnetworkcanbecategorizedinthreetypes:purestreaming,implicitduplication,andexplicitduplicationasdescribedinSection 1.1 Todate,therehavebeenintroducedseveralP2PmultimediaapplicationssuchasJoost[ 51 ],PPLive[?],TvAnts[ 63 ],Cabos[ 65 ],KiwiAlpha[ 54 ],Shareaza[ 50 ],BitTorrent[ 41 ],BitComet[ 46 ],Azureus[ 67 ],andsoon.ThepurestreamingtypesareJoost[ 51 ],PPLive[?],andTvAnts[ 63 ].Next,theimplicitduplicationtypesareCabos[ 65 ],KiwiAlpha[ 54 ],andShareaza[ 50 ].Lastly,theexplicitduplicationtypesincludeBitTorrent[ 41 ],BitComet[ 46 ],andAzureus[ 67 ]. Inoursystem,weadoptimplicitduplicationasaP2Pcontent-sharing,sothattheprogramstoredineachPVRcanbeeffectivelysharedwithotherPVRuserswithoutsufferingadelayorajitter.Inaddition,oursystemcanalsoworkasbyexplicitduplicationmethodonceawholeprogramisrecordedinaPVRuntilwhentheprogramisdeleted. 50

PAGE 51

Inthissection,weintroduceanovelTVcontentdistributionarchitectureforacommunitynetworktomeettheserequirements.WehaveexplainedtheFC-ALsingle-loopcommunitynetworkinsection 2.3.1 .Then,wedescribeandreviewascalablemultiple-looparchitectureusingshareddisksastheinfrastructureofthecommunitynetwork.Theproposedarchitecturecanadaptivelyincreasethecommunitynetworksimplybyaddingloopseventhoughcommunitysizecontinuestogrow[ 14 ]. Wethereforedevelopthescalablemultiple-looparchitectureforacommunitynetworkthathasnolimitationonthenumberofattacheddevicesasshowninFig. 2-1B .Whenextendingthesingleloopsystemtothemultipleloopsystem,however,weneedtodeviseabridgingmechanismtorelaytherequestedprogrambetweendifferentloops.Thus,weadoptshareddisksthatconnecttwodistinctloopsandrelaytherequestedprogramfromonelooptotheother.SincetheFCstandardallowsmultipleinterfacedevicesforsuchreasonsasdurability,wecanimplementtheshareddiskswithoutmodifyingFCprotocols. Fig. 3-1 illustratesproposedmultiple-looparchitecturesinordertosupportaninfrastructureofacommunitynetwork,usingtheshareddisksandthedirectoryserverwithpossible2-loop,3-loop,and4-loopcongurations.Notethateachloopindicatesasingle-looparchitecture,asinFig. 3-1A Fig. 3-1A showsthefundamentalarchitecturetoextendfromsinglelooptodoubleloopusingshareddisks.Inthisarchitecture,wecanseethattheshareddisksworklikebridges,whichenablethemultiple-looparchitectureandcanbeapathbetweentheloops.Fig. 3-1A illustratesa3-looparchitecture.Withmorethan3loops,wenoticethattherewillbesomedesignissuesintermsofhowtoorganizeamultiple-looparchitecturewiththeshareddisks.However,inordertoorganizeeffectivelyintermsofaccessibilitysothateachloopcanreachalltheotherloopswithoutinvolvingmoreshareddisksand 51

PAGE 52

BFourloopconguration Examplesofmultiplelooparchitectures moreloops,itisdesirabletodeployashareddiskbetweeneverytwoloopsasinFig. 3-1A .Forexample,wecanseefromFig. 3-1A thattheloop-1canreachtheloop-3withonlyoneshareddiskbetweenthem,ratherthanpassingthroughtheloop-2withtwoshareddisksinvolved.Similarly,wecanorganizea4-looparchitectureasinFig. 3-1B .WecanalsoseethattheeveryloopcanreachanyotherloopsviaonlyoneshareddiskinFig. 3-1B 52

PAGE 53

Therefore,withapplyingthismultiple-looporganization,wecanconstructthefollowingformulainordertoobtainthenumberofattachabledeviceswithrespecttothegivenloopnumber: wherenisthenumberofloops. Now,wecanorganizethemultiple-looparchitecturebyusingtheshareddisksforthecommunitynetworkinfrastructure.Withtheproposedmultiple-looparchitectures,wecandelivertheHD-qualitycontentsharinganddistributionamongcommunitymembers,i.e.,PVRs,withoutlimitingthenumberofcommunitymembers[ 14 ]. 53

PAGE 54

1 42 ].Thus,theprobabilitythateachPVRrecordstheithmostpopularprogramamongthe$n$programscanbecomputedasfollows: whererepresentstheaccessskewnessdegree(i.e.,asincreases,theskewnessdegreedecreases). Toshowtheimpactofstorageduplicationofprograms,Fig. 3-2 illustratesanexamplewherethereare50TVchannels,i.e.,50differentprogramsarealwaysbeingbroadcast.Inaddition,wesettheto0.271,whichisatypicalparameterforpopularitydistributionofvideorentals. 54

PAGE 55

Probabilityofstoringeachprogramaccordingtoitspopularity WecanseefromFig. 3-2 thattherstranked(themostpopular)programneedsbestoredin13.3%ofallPVRs,whilethe50thrankedprogramneedstobestoredinabout1%ofPVRs.Inotherwords,ifwedeploy1,000PVRs,133copiesoftherstrankedprogramwillberecordedandonly10copiesofthe50thrankedprogramwillberecorded.Thisindicatesthat,iftherearetoomanyduplicatedcopiesofpopularprogramsamongPVRs,thecommunitynetworkwillwastestoragespacesignicantly. Inaddition,theconventionalleserversystemsbasedonSCSIdonotsharetheinterfacebandwidthamongconnecteddiskssincealldisksarefullycontrolledbytheservers[ 78 ].TheSCSI-baseddisksarepassiveinnature,andonlyaservercanaccessthembytheSCSIcommands.Moreover,theserversarelesssuccessfulingettingtheprogramcontentsfromtheclientspromptlyandadaptivelycomparedtotheP2Papproach.However,theFC-ALwithPVRsenablesthesystemtosharethediskbandwidth.Theyalsohavemoreadvantagessuchaspeer-matchingandpeer-switching[ 39 ],whichallowthesystemtopiggy-backontheexistingstreams. 55

PAGE 56

PVRs'ContributableStorageinthePVR-basedFC-ALsystem ProgramNumberstoStoreAdditionally 182 364 547 729 912 1094 1276 1459 1641 1824 Furthermore,content-sharingamongthelocaldisksinPVRsallowsthesystemtobemoreexiblecomparedtotheserver-onlysystemforusingthewholesystemstorage.Thus,thePVR-basedFC-ALarchitectureingeneraldoesnotrequireasmanycopiesoftheprogramsastheconventionalSCSI-basedservers/systems.Table 3-1 explainsthatPVRs'contributablestoragespaceintermsofthenumberofprogramsinthePVR-basedFC-ALarchitecturewhenassuminghalfofPVRsaredeployed,thathalfofthemarealive,andeachPVRprovidestheindicatedsharingportiontothesystem,whichiscomparabletothepassiveSCSI-basedsystem.Ideally,PVR-basedFC-ALsystemscanreducetherequirednumberofcopiesofprogramssignicantlyandeffectively. Inourarchitecture,aPVRcanreceiveanyprogramstoredonotherPVRsornetworkdisks.WhenaPVRrequestsaprogramthatmorethanonealivePVRorthenetworkdisksarestoring,itcanreceivetheprogramfromoneofthealivePVRsornetworkdisksviaP2Pstreamingwithinthecommunitynetwork.Thus,aslongaswecanguaranteethatsufcientcopiesofeachprogramareavailableatanygiventime,we 56

PAGE 57

5 ].Inoursystem,thePVRs'auto-recordcanworkautomaticallyfortime-shiftingservices,oncethestorageissetbyitsuser.Forexample,ifaPVRhas500GBstorageanditsusersets10%ofstorageforauto-record,then50GBisassignedforstoringtime-shiftingprograms.Specically,onceaPVRwatchesaprogram,theprogramisautomaticallystoredintheauto-recordstorageofthePVR. Ourschemehasbeendevisedbasedonthesefacts.Thatis,ourschemeproposestostoreonlyasufcientnumberoftheprograms,initiallybroadcastwithagivenexistenceprobability.Inotherwords,ifwedeneathresholdvalueforaprogramexistence,wecandeterminehowmanyPVRsshouldbealiveatleast,inordertoobtaintherequirednumberofcopiesforthatprogramwithrespecttothatthresholdvalue.GiventhosealivePVRshavingthetargetprogram,wecandeterminetheminimumcopynumbersoftheprogramasfollows: wheremistheminimumrequirednumberofcopiesforaprogramandPaliveistheprobabilityofeachPVR'sbeingalive. Forexample,iftheexistencethresholdvalueisgivenas0.9andthePaliveis0.5,thenwecangettheminimumcopynumberforPVRs,m,as4fromEq. 3 .Thatis,if4alivePVRseachhavethetargetprogram,oursystemdeterminesthatthatprogramcanberegardedasminimal-copywithrespecttothegiventhresholdvalueandPalive. 57

PAGE 58

Table3-2. Symbolsforreplacementalgorithms SND Table 3-2 showsthesymbolsusedinourstoragesavingschemessuchasprogramplacementandreplacementalgorithms.TheprogramplacementalgorithminAlgorithm 3-1 triestoholdatleastthesufcientnumberofprogramsonbehalfofthewholesystem.IfthePVRstrytostoremthcopiesofaprogramwhenthenetworkdiskscurrentlyholdthatprogram,thesystemallowsthePVRstostoremthcopybutit 58

PAGE 59

program) ThedesigngoalisthereforetomaintainatleastsufcientcopiesofeachprogramonthealivePVRsorthenetworkdisksforlateraccess,whilealsoreducingtheexcessiveredundantcopies.Wethereforeproposereplacementschemesforboth 59

PAGE 60

program) program; 3-2 .Thealgorithmrstcheckswhetherornotthereexistanyprogramsthatarealsostoredonnetworkdisks:i.e.,ifSi\SND6=.Aslongasnetworkdisksstoretheprograms,PVRsmaynotneedtostorethemanylongerbecausethenetworkdisksarealwaysavailable,providingsufcientdiskbandwidth.Ifso,wechoosetheoldestprogramamongthem,i.e.,theprogramthathadbeenstoredearliest,becausetheoldestoneislesslikelytobeaccessedinthefuture.ThesystemwillsendanautomaticmessagetorecommendthePVRownerremovetheprogramcopywiththeassurancethatthecopiesonthenetworkdiskswillremainavailable.Ifthereisnoprogramsatisfyingtheabovecondition,wecheckwhetherthereexistanyprogramsthatarealsostoredinotheralivePVRs:i.e.,iffpijpi2Si\Ni>1g6=.Iffound,wechoosetheprogramthathasthemostnumber 60

PAGE 61

program) program; 3-3 .WerstcheckwhetherthereexistanyprogramsthatarestoredinanyalivePVRs:i.e.,iffpijNi>0g6=.Ifso,wechoosetheprogramthathasthemostnumberofcopiesinAlgorithm 3-2 .Ifnotfound,itindicatesthatalltheprogramsstoredinnetworkdiskshaveonlyonecopywithinacommunitynetwork.Thus,wechoosetheoldestprogramamongthemandreplaceitwiththeincomingprogram. 61

PAGE 62

3-3 .Theyareusedthroughoutoursimulationsunlessotherwiseindicated.ItisassumedthateachTVprogramhasHDqualityof19.4Mbpsplaybackrate.ThedegreeofaccessskewnessamongTVprogramsfollowstheZipfdistributionwith=0.271,whichistypicallyusedforvideorentaldistribution[ 17 79 ]. Table3-3. Simulationparameters datatransferrate 400MB/s propagationdelay 3.5ns/meter pernodedelay 240ns Disk capacity 500GB cache 32MB datatransferrate 2965MB/s seektime 4.7ms rotationallatency 3.0ms Storagesavingscheme probabilityofeachPVRbeingalive 70% threshold 0.9 storageportionfortime-shifting 30% Zipfdistribution(value) 0.271 Multiple-loop shareddiskinterfaceconguration dual-loopinterface numberofnetworkdisks 10%ofalldevices Ourproposedstorage-savingschemesarerstevaluatedinthesingle-looparchitecturewithvariousparametersettingssuchastheratioofPVRsandnetworkdisks,Palivevaluesandthresholdvalues.Fig. 3-3A showstheimprovedstoragehoursaccordingtothedifferentnumberofnetworkdisksdeployedintermsofpercentagewithourproposedschemes.Wecanseethattheimprovementreachesalmost133%betterwithsevennetworkdisks,(i.e.,119PVRs).Theimprovementstendtodecreasewith 62

PAGE 63

3-3B showstheeffectoftheratioofPVRstonetworkdisks.Whentheratioisincreasing,theimprovementtendstogetbetterupto16%,whichimpliesthattheimprovementcannotreachhigherwithoutaddingmanynetworkdisks. Lastly,Fig. 3-4A demonstrateshowthevariousPalivecanaffectthesystemperformance.Asweexpected,thelargerthevalueofPalive,thebettersystemperformance.Thatis,theperformanceisimprovedwhenalargevalueofPaliveissetbytheusers.ThismeansthatwhentheprobabilitythateachPVRcanbealiveishigh,thesystemshowsbetterperformance.Fig. 3-4B illustratestheeffectsofvariousthresholdvalues.Asthevaluegetslarger,thesystemperformanceisdecreased.ThisisbecausethesystemneedsmorePVRsinordertomeetthegiventhresholdvalues.Thus,thelargerthresh-oldshowsthesmallernumberofstoragehoursintermsofthesystemperformance. Table 3-4 showshowmanyPVRsandnetworkdisksthereareinacommunitynetworkdependingonthenumberofloops.Thoughwearestillsearchingfortheoptimalratio,wehavechosentoassign10%ofallpossibleattacheddevicestonetworkdisksinthischapter.Itcanbeseenthat,whenemployingtenloops,ourarchitecturecansupportmorethan1100devicesinacommunitynetwork.Thesimulationresultsdemonstratethatourproposedarchitecturesignicantlyincreasesprogramstoragehoursbyreducingtheduplicatedstorageofpopularprogramsinacommunitynetwork.Thefollowing 63

PAGE 64

BEffectsofPVRtoNDiskratio EffectsofPVRsandnetworkdisks 64

PAGE 65

EffectsofPaliveandthreshold 65

PAGE 66

Theratioofpvrsandnetworkdisksinacommunitynetwork 115 12 2 228 25 3 341 37 4 452 50 5 563 62 6 673 74 7 782 86 8 890 98 9 997 110 10 1103 122 subsectionsdescribetheachievedefciencyofthesystemperformancebasedonvarioussystemaspects. Figure3-5. Effectivenessofourproposedarchitecture 66

PAGE 67

Fig. 3-5 showsthatschemeoncaseoutperformsschemeoffcasebyanaverageof69.7%.Thisisbecausetheschemeoncasecanreducetheduplicatedstorageofprogramseffectively,consideringthecharacteristicsofTVrecordingpattern,i.e.,accessskewnesstopopularTVprograms.AsthenumberofPVRsincreases,thenumberofcopiesofeachprogramalsoincreases.Thisimpliesthat,asmorePVRsareinvolved,ourschemecanbefurtherimprovedbyreducingthenumberofredundantcopiesofeachprogram.Thisalsoenablesnetworkdiskstosavestoragespacethatwouldotherwisestoreexcessivecopiesofpopularprograms.NotethatthenumberofPVRsinFig. 3-5 isproportionaltothenumberofloops,aslistedinTable 3-4 3-6 showsthat,asahigherstorageportionisassignedfortimeshifting,performanceimprovementalsoincreases.Forexample,letusconsiderthecasewhereweemploy900PVRsinacommunitynetwork.Comparedtowhenonly10%ofPVRs'storageisreservedfortime-shifting,theprogramstoragehoursbecome17.1%,32.9%,44.7%,and56.1%longerwhen20%,30%,40%,and50%arereserved,respectively.Therearelikelytobemoreduplicatedprogramsaswehavemorestoragespacefortime-shifting. 67

PAGE 68

ImpactofPVRs'storageportionfortime-shifting Itisalsointerestingthatthedegreeofimprovementpercentagedecreasesastheportionsincrease.Forexample,whenemploying900PVRs,extendedhoursbetweeneachoftwoconsecutiveportionsfrom10to50%are24,21,18,and16hours,respectively.Thus,itcanbeseenthat,asmoreprogramsareaccommodated,ourarchitecturecontinuestoimprovetheperformancebuttheimprovementpercentagedecreases. 3-7 showsthatschemeoncaseonaverageachievedtheperformanceimprovementby56.3%,65.3%,and69.7%for100GB,300GB,and500GB,respectively,comparedtotheschemeoffcase.Thatis,asthestoragecapacityofeachPVRincreases,theprogramstoragehoursgetlonger.ThisisbecausePVRs'storageportionfortimeshiftingincreaseswiththeincreasedstorage 68

PAGE 69

Impactofstoragecapacity capacityastheirstoragecapacityislarger.Itcanbealsoseenthattheperformanceat300GBintheschemeoncasecanworksimilarlyas500GBintheschemeoffcase.Thisimpliesthatourproposedarchitecturecanachieveacceptableperformanceevenwithrelativelylittlestoragecapacity. Basedonthemultiple-loop-enabledcommunitynetworkinfrastructures,wehaveproposedstoragesavingschemesforbothPVRsandnetworkdisks.ThegoalofstoragesavingschemesistoreducethenumberofduplicatedcopiesofpopularprogramstoextendprogramstoragehoursofTVprograms.Theextensivesimulations 69

PAGE 70

70

PAGE 71

Additionally,inmultiple-looporganization,theinitialshapeofaloopshouldbechangedinordertoplaceashareddiskbetweenloops,whichresultsinanincreaseintheloopsizeasshareddisksareadded.Forexample,inordertoconnecttwoloopsthatarephysicallysomewhatapart,itisclearthattheinsertionoftheshareddiskrequiresthemodicationofthosetwoloops.Asaresult,thesizeofeachloopincreaseswhenashareddiskislocatedbetweenanytwoloops.Itiswellknownthat,toprovidebetterperformance,thesizeofanFC-ALloopissignicantinordertobesuitableforhigh-qualitycontentdeliveryinrealtime[ 60 ].Thisindicatesthattheaverageloopsizeshouldbelargerwheneachloophasmoreneighborloopsconnectedbyshareddisksandwoulddeliverworseperformanceduetoincreasedpropagationdelaysandarbitrationtimes. Therefore,inthischapter,werstintroduceanovelmultiple-looparchitectureusingaminimumspanningtree(MST)basedtopologywithlessconnectivityandlessaverageloopsize.Infact,theCGtopologyisanextremecase,havingthemostnumberofconnectivitiesby1-hopreachabilityandthustheleasttotaltrafcwithdistributedtrafc 71

PAGE 72

WethendescribehowwecanorganizetheMST-basedtopology,sothataverageloopsizeinthetopologycanbeminimizedandeveryloopcanreachanyotherloopswithina2-hopor3-hop,whichaffectsthetopology-causedtotaltrafcsincetheloadedtrafcisadditivewhenmorehopsareinvolvedinrouting.Nonetheless,oncethetopologyisdetermined,theMST-basedtopologyonlypermitsaxedroutingpath,whichstillcausesatrafcloadconcentrationonthexedroutingpath. Consequently,weproposeaninnovativeMST-basedgraphtopology,namedMSGtopology,whichisbasedonthe2-hopand3-hopMSTtopology,butallowsmultiplepathsandgraphorganizations,usingenablingtrafcloaddistributionviaaless-loadedpathamongthosepossiblepaths.WedescribehowtheMSGtopologyisachievedandcomparetheMSGtopologywiththeMSTtopologyandtheCGtopology.Oursimulationresults,then,revealthattheMSGtopologymultiple-looparchitecturecanprovideaverycloseaverageloopsizecomparedtoMSTtopologycaseswithremainingconstantandcloseenoughintermsoftotaltrafcandaveragerejectratiocomparedtotheCGcasesevenwithmuchlesstopologyconnectivity. Theremainderofthischapterisorganizedasfollows:Section 4.2 describespreviousworkrelatedtothischapter.Section 4.3 explainstheoverallarchitecturefortheFC-ALmultiple-looparchitecturewithshareddisks.Section 4.4.1 speciesthemotivationbywhichtheMSG-basedtopologyisdevised.Section 4.5 presentsthedetailedalgorithmsonhowtoorganize2-hop/3-hopMSTsandMSGs,respectively.Section 4.6 illustratesextensivesimulationresultswithCGs,MSTs,andMSGs.Finally,Section 4.7 offersconclusions. 72

PAGE 73

45 52 ].Moreover,thecapabilitiesofTVcontent-sharingbetweenPVRshavebeenexaminedin[ 35 49 ]. Meanwhile,theberchannelhasbeenpresentedasoneofstandardstocongurethestorageareanetworks[ 34 ]andtheperformanceandcharacteristicsofFC-ALhavebeenanalyzedin[ 19 37 60 ].Additionally,S.Chenetal.haveinvestigatedthemultimediaserverorganizationbasedonFC-ALbyseveraladvantagessuchasadirectattachmenttothestoragesandthedesirableperformanceatalowcost[ 12 13 ]. IntegratingallthosecomponentsincludingPVRsandtheFC-AL,thefeasibilityofamultimediacontent-sharingarchitecturehasbeenproposedin[ 14 38 ].However,E.Kimetal.havenotconsideredthescalabilityissuestoconstructtheFC-AL-basedPVRnetworkarchitecture[ 38 ]andthecompletegraphtopologyhasbeenmainlydiscussedindesigningascalablemultiple-looparchitecture[ 14 ]. 14 ]. 14 ].Whenextendingthesingle-loopsystemtothemultiple-loopsystem,however,weneedtodeviseabridgingmechanismtorelaytherequestedprogrambetweenloops.Wethusadoptshareddisksthatconnecttwodistinctloopsandrelayrequestedprogramsfromonelooptotheother. 73

PAGE 74

BThestructureofanFC-ALsingleloop CSamplethree-looparchitecturewithshareddisks OverallTVcontentdistributionarchitecture 74

PAGE 75

34 ],wecanimplementtheshareddiskswithoutmodifyingFCprotocols. 4-1C isone,itcanbegreaterthanoneifongoingtrafcismorethancanbecoveredbyasingleshareddiskasshowninFig. 4-2 Tomakeitfeasibletorelayprogramsbetweenapairofsystemcomponentswithoutinvolvingservers,weadoptFCdisksthatprovideadirectdatatransferfunctionamongthemsuchasextendedcopy(X-copyorE-copy)anddisk-to-disk(D2D)transfer.Withthisfunction,aserveronlydesignatesinvolveddisksfortheirroles,i.e.,oneforsourceandtheotherfordestination.Infact,thisfunctionhasalreadybeendevelopedbyseveralcompaniesforserverlessdatabackuptechnologiesusingFCdiskswithlimitedprocessingcapabilities[ 8 ].Inourarchitecture,eachloophasaDSthatmaintainsalistofTVprogramsstoredinallcomponents.TheDSisresponsiblefortriggeringX-copymoduleembeddedineachshareddiskasaninitiatorwhenperformingthedirectdatatransfer.Thisfunctioncanreduceadata-relayinglatencyandavoidthesituationwhereserversbecomeaperformancebottleneckwhentheymustrelayallthedataamongloopsthroughtheirmainmemory. Theprogramrelayingprocessinoursystemisperformedintwophases:locationandroutingdeterminationusingFC-ALprotocolandactualblockread/writebetweentwodevicesusingFCprotocol.Intherstphase,thePVRrstasksaDSwheretherequestedprogramislocatedandthentheDSinformsthePVRoftheentireroutingpath 75

PAGE 76

HowtorelayaTVprogramusingshareddisksinatriple-loop fromtargetPVRfromwhichitthePVRcanreceivetheprogram.Inthesecondphase,theactualdatablocktransferfromthetargetPVRviashareddisksisperformed. Fig. 4-2 illustrateshowaprogramcanbetransferredfromarequestingPVRtoatargetPVRviashareddisksinatripleloopusingFC-ALandFCprotocols.Basically,eachDSprovidesnecessaryinformationtoarequestingPVRandalsoworksasaninitiatortotriggertheX-copyfunctionbetweenanypairoftwodevices.Intherstphase,whenPVR(a)hasarequestforaTVprogram,itasksDS0whichdeviceitcanbeservedfrom().TheDS0thenndsoutfromitsmaintainedinformationthatitislocatedinPVR(d)anddeterminesaroutingpath,includingproperloops,i.e.,loop-1()andloop-2()inthisscenario.Oncetheroutingpathincludingmultipleloopsisdetermined(),theDS0informsPVR(a)ofarelayingdevice,Sdisk(b).ByrepeatingthisprocedureuntilreachingthetargetPVR,PVR(a)obtainsaconnectionpathtoPVR(d)throughshareddisks,i.e.,Sdisk(b)andSdisk(c).InordertoestablishanactualconnectiontothetargetPVR,PVR(a)sendsARB(a)toloop-0foranarbitrationandthensendsOPN(b)toSdisk(b)toopenaconnectiontoSdisk(b)afterwinningthearbitration.Sdisk(b)thensendsR RDYbacktoPVR(a)tosettleacommunicationwithPVR(a)().Atthesametime,DS1issuesadirectdatatransfercommandtoSdisk(b)sothatSdisk(b)should 76

PAGE 77

RDYisalsosentintheloop-1forthecommunicationbetweenSdisk(b)andSdisk(c)().Intheloop-2,DS2coordinatesthecommunicationbetweenSdisk(c)andPVR(d)asaninitiatorwiththesamecommandsasonesusedinpreviousloops(). AfterallroutingconnectionsareestablishedfromPVR(a)toPVR(d)viaSdisk(b)andSdisk(c),inthesecondphase,datablocksaretransmittedalongwiththedeterminedpath.DS2issuesanX-copycommandtoPVR(d)sothatPVR(d)shouldwritetherequestedprogramontoSdisk(c).PVR(d)thussendsFCP CMNDIUtoSdisk(c)asawriteoperation.Sdisk(c)thensendsFCP XFER RDYIUtoPVR(d)toreceivedatablocksandPVR(d)sendsFCP DATAIUbacktoSdisk(c)totransmitanadditionaldatablock.ThispairofFCP XFER RDYIUandFCP DATAIUbetweenSdisk(c)andPVR(d)continuestobetransferredwithinaspecicperiodoftime,i.e.,cycletime,untilalltheblock-writingtoSdisk(c)isnished().Infact,eachDScoordinatestheblocktransmissionssothateachrelayingshareddiskcancontinuouslytransfertheincomingdatablockstothenextdevicewithonlystoringthemintoitsbuffertemporarily,notstoringthemontoitsharddisk.Similarto,thedatablocktransmissionbetweenSdisk(c)andSdisk(b)()andbetweenSdisk(b)andPVR(a)()isperformedwiththesamecommandssuchasFCP CMNDIU,FCP XFER RDYIU,andFCP DATAIU,bythecoordinationofeachloop'sDS,i.e.,DS1andDS0,respectively.Notethat,ifarequestingPVRcanreceivetherequestedprogramfromatargetPVRinthesameloop,otherDSsandshareddisksdonotneedtobeinvolved. Basedontheprogramrelayingmechanismviashareddisks,Fig. 4-1C illustratesthefundamentalarchitecturecongurationwiththreeloopsandshareddisks.Specically,everyloopisdirectlyconnectedbyalltheotherloopsviaashareddiskperloop.Forexample,loop-0isdirectlylinkedtoloop-1andloop-2byeachshareddisk.Thisorganizationallowseverylooptoreachalltheotherloopswithonlyonehop,i.e.,one 77

PAGE 78

Thus,wecanmakeoneimportantobservation:theCGtopologyrequiresatleastasmanyshareddisksasthedirectlyconnectedloopnumberonlyforthepurposeofmaintainingtopologyconnectivity.Wethereforeanalyzethisproblematictopologyconnectivityissueandproposeanovelmultiple-looptopologyinSection 4.4 ,inordertoorganizeanefcientmultiple-looparchitecture. 4.4.1Overview Whenplacingashareddiskbetweentwoloops,wecanregardthemodiedshapeofeachloop,asinFig. 4-3A ,withoutlossofgenerality,whilemaintainingeachloopstructureinordertoorganizethemultiple-looparchitecture.Specically,iftwoloopsaredkilometersapart,wethenputashareddiskinthemiddleofthedistance,i.e.,d=2kilometers,byattachingeachlooptoitsshareddisk.Fig. 4-3A illustrateshow,asabasiccase,eachloopcanbemodiedwhenashareddiskisinsertedbetweentwodifferentloops.Thekeyobservationhereisthateachloopsizeshouldincreaseinordertoreachacontactpoint,i.e.,theshareddiskbetweentwoloops.Inaddition,Fig. 4-3B showshowweorganizeaCGtopologywithfourloops,demonstratinghoweachshareddiskcanbedirectlyconnectedinordertomaintaineachloopstructure.Infact,wecanlinktheinsertedshareddisktoanydeviceontheloopaslongastheextendedloopsize 78

PAGE 79

B4-loopCG C4-loopMST D4-loopMSG ExamplesofCG,MST,andMSGarchitectures canbereducedandtheloopstructurecanbemaintained.Nevertheless,theloopsizetendstoincreasewheneverashareddiskislocatedbetweentwodifferentloops.ThismeansthattheCGtopologymultiple-looparchitecturewouldhavethelargestloopsizecomparedtoanyothertopologyarchitecturessinceithasthemostconnectedloops,asmanyas(n1)neighbors,wherenisthenumberofdeployedloops. 79

PAGE 80

60 ],althoughitisbelievedthattheFC-ALcancoverupto10km.Furthermore,practicalissuesarisewhenamultiple-looparchitectureisreallyappliedasanetworkinfrastructureinrealtime.Thatis,itisalsoanimportantfactorinhowmanyloopsarenecessaryorhowlongthephysicalberchannelhastobeintermsoftheactualcostandtheeffectivenessofthesystem. Moreover,whenashareddiskisinsertedinthemiddleoftwoloops,wecanfurtherreducethemodiedloopsizeifwechoosethedesirable,i.e.,closest,loopswhereashareddiskwouldbeplaced.Whenthisapproachisapplied,thepossible4-loopMSTarchitectureisshowninFig. 4-3C .Nonetheless,theMSTarchitectureonlyallowsonexedroutingpathbetweenanypairofloops,therebycausingload-concentrationonthatpath. Consequently,weintroduceaninnovativeMSGtopologymultiple-looparchitectureinthischapterthatprovidesacceptableperformanceintermsofaverageloopsizeandaveragerejectratiowhilesupportingmultiple-pathroutingwithin2-hopor3-hopreachability.Anexampleof4-loopMSGtopologyisshowninFig. 4-3D ,implyingtherecanbemultiplepathsbetweentwoloopsforloaddistribution.Forinstance,loop-0canreachloop-2viasdisk-a,loop-1,andsdisk-borviasdisk-d,loop-3,andsdisk-cinFig. 4-3D .WedescribetheMSGtopologyingreaterdetailinthenextsectionandexamineitsperformancebyvarioussimulationsinSection 4.6 4.5.1ProblemStatement 4.4.1 ,thepreviousCGapproachproposedin[ 14 ]hasnottakenintoaccountpracticalaspectsoftheproblem,i.e.,loopsize,andhasbeenchosenasthebestselectionbycomparingseveralpossiblenetworktopologies.Theonlyfactorconsideredin[ 14 ]isthetotalnetworktrafctraversingnetworktopologies 80

PAGE 81

However,thetotalnetworktrafc,whichisproportionaltothenumberofshareddisks,isnottheonlydesignfactorwehavetoconsiderwhendesigningnetworktopologiesforhigh-qualitycontent-sharinginaPVR-attachednetwork.Theprincipaldesignfactorsthatshouldbeconsideredare:Networkthroughput,Hopnumberbetweenpairsofloops,Accept/rejectratioofrequests,Numberofusersthatcanbeaccommodated,Numberofshareddisks,Loopcapacity,andLoopsize.Networkthroughputisawell-knownperformancemetrictoevaluateacertainnetworksituation.Hopnumberbetweenpairsofloopsshouldbeboundedsowecanguaranteethatarequestedprogramcanbeaccessiblewithinsomedelayandwithoutqualitydegradation.ThehardwareattributesofFCshowsthatthequalityofcommunicationbetweentwodevicesinaFCgetspoordrasticallyasthedistanceisgettingfarther.Accept/rejectratioofrequestsandthenumberofaccommodableusersaremajordesigncriteriabeforedeployingnetworktopologies.Loopsizeandthenumberofshareddisksreectbudgetissuesaswellasfeasibilityoftopologies.LoopcapacityisoneofinherentlimitationsofFCloopsaccordingtoFCcapacities. However,itisnottruethateachofthesedesignfactorsisorthogonaltotheotherfactors.Ingeneral,networkthroughputcanbeapproximatedbytotalnetworktrafc averagehopnumber.Thus,networkthroughputandhopnumberbetweenpairsofloopsarecloselycorrelatedsincenetworkthroughputdecreasesasaveragehopnumberofpathsonthatrequestshouldtraverse.Numberofshareddisksisalsodependentonnetworkthroughput,whichmeanspoornetworkthroughputleadstoagreaternumberofshareddiskstohandleincreasednetworktrafc.Whenlookingatthenumberofshareddisksandloopsize,theyalsohaveoninverselyproportionalrelationship.AsshowninFig. 4-1 ,theloopsizehasanincreasingtendencyasthenumberofshareddisksbetweenloopsincreases.Inthischapter,welookforasolutioninwhichloopsizeisminimized 81

PAGE 82

4.4.1 ,wecantranslatetheproblemofminimizingloopsizeintotheproblemofminimizingtotalinter-distanceofloopsconnectedbyshareddisks.ThecomplicationsthattheproblemhasasobjectivesorasconstraintsdeterusfromdeployingsimplecombinatorialalgorithmssuchasKruskal'sMinimalSpanningTree(MST)algorithmandsoon. Therefore,weformulatethisdesignproblemasanoptimizationproblemwithoneoftwopossibleobjectivefunctionsandseveralconstraints.OurmathematicalformulationismotivatedbytheDiameter-constrainedMinimumSpanningTreeProblem(DMST)basedonnetworkowmodels[ 25 ],whichismixed-integerlinearprogramming(MILP).Onceaproblemisformulatedinalinearprogrammingmodel,thelinearprogrammingmodelallowsusmoreexibilitycomparedtoothersimplecombinatorialalgorithms.Inthefollowingsubsections,wedescribeoptimizationcriteriaandassociatedconstrainingconditionstosolveourproblem. MSTformulations 82

PAGE 83

25 ]imposesaboundonthediameterofatree,whichisthemaximumnumberofedges,i.e.,themaximumhopnumber,inanypathsbetweenanypairsofitsnodes.WhenD4whereadiameterisdenotedbyD,theDMSTproblemisprovedtobeNP-Hard[ 25 ]whereaswhenD=2or3,theproblemissolvableinpolynomialtime. TheDMSTbasedonnetworkowmodelstswellintoourproblemsincethehopnumberforanypathisconstrainedbyagivenconstant,sosubsequentlywecanboundthetotalnetworktrafc.Asforthediameterconstraint,wedonotneedD4becauseinthatcase,obtainingatargetprogramfrompeersbeyond3hopsishighlylikelytobemorecostlythangettingonefromalocalserverfromtheperspectiveofdelayandnetworkresourceconsumption.Thus,asarsttry,wedirectlyappliedtheDMSTbasedonnetworkowmodelswithdiameter2or3toouroptimizationproblem.Intermsoftimecomplexity,itdoesnottakeverylongtosolvetheMILPformulationwhenthenumberofloopsdoesnotexceedaround20basedonanassumptionthatanadoptednetworkhasroughly1,0002,000subscribers.Thenotationsandequationsareborrowedfrom[ 28 ]wheneverpossible. TheobjectiveofthisMILPformulationistominimizetotaldistancebetweenloopsconnectedbyshareddisksinEq. 4 wheneverynodesendsaunitamountofow,1,toallothernodes.Thedecisionvariablexerepresentswhetheracertainedgeisincludedinatreeornot.Anotherdecisionvariableypqijdenotestheamountofowonedge(i,j)betweennodespandq.Theconstantceisanedgedistance,determinedoncethelocationsofloopsareknowninadvance.TheresultinggraphisconstrainedtobeatreebyEq. 4 .Theconstraints 4 and 4 comefromthoseofwell-knownmulti-commoditynetworkowproblem,makingeveryowobeyowconservationandlinkcapacitylimitrule.TheEq. 4 limitsthehopnumberofallthepathsbyD,whichanodeshouldtraversetoreachtheothernodes. 83

PAGE 84

MSGformulations 4.5.1 ,loopcapacityisnotreectedintoconstraintsoftheMSTformulation. ToovercomethesedrawbacksoftheMSTapproach,wehaveadjustedpreviousconstraintssuchthattrafcgoingthroughaloopisboundedandtheresultinggraphmaynotnecessarilybeatree. Asobjectivefunctions,wecantakeoneoftwopossiblefunctionsasbelowdependingonwhichdesignfactorhasapriority.Iftheloopsizeisthetoppriority,settingminimizingtotaldistancetotheobjectivefunctionisreasonable.Ontheotherhand,if 84

PAGE 85

Formally,theobjectivefunctionscanbesetasinEq. 4 .TheconstraintenforcingtheresultantgraphtobeatreeinthepreviousMILPformulationisremovedinthisformulation.Instead,theconstraintisloosenedbysettingtheupperlimitofthenumberofedgesto2(n1)asinEq. 4 .Moreover,theloadbalancingconstraintinEq. 4 isadded. TheinequalityofEq. 4 canbeunderstoodintuitivelyasboundingthetotalamountofowsinwhichanodeisinvolvedbyL(n1).Ifacertainnodeisnotusedasaintermediatenodewhereowsbetweenothernodestraverse,thetotalamountofowsinwhichanodeisinvolvedis(n1).IfwesetLto2,itcanbetranslatedthatonenodedoesnotsacriceitsbandwidthforothernodes,morethanforitselfsincehereonenoderepresentaloop.Itisareasonableconstraintthatnoparticipantdoesnotwanttogiveitsresourcemorethanitneeds.Hence,inthischapter,weconductfollowingexperimentswhenLisgivenas2.Theotherconstraints,Eq. 4 throughEq. 4 ,aresameasconstraintsintheMSTformulation. Thereasonloadbalancingamongloopsispossibleisthatourformulationisbasedonmulti-commoditynetworkowmodels.TheDMSTbasedonmulti-commoditynetworkowmodelsassumesthateachnodegeneratesthesameamountoftrafctotheothernodes;thiscanberegardedasthemostcongestednetworkstatus.Forthisreason,theloadconstraintsoneachnodecaneasilyreectthenetworksituationwhererequestsarerejectedduetoashortageofloopcapacity. InSection 4.6 ,wewillseeloadbalancingasanobjectivefunctionresultsinthebesttopologyintermsofbothloopsizeandrejectratio.Alternatively,toachievebothobjectives,minimizingloopsizeandloadbalancing,wecantakeeitheroftwoapproaches.Oneistocombinetwoobjectivefunctionsadditivelywithscaling 85

PAGE 86

Flowchartforload-balancedMSG factors,whichiscalledmultivariateobjectivefunction.Theotherapproachistouseoneobjectivefunctionwhilesearchingtheallsolutionspaceoftheotherobjectivefunction.Supposethatweneedasolutionthatsatisesthelimitofloopsizewhilemaximizingloadbalancing.WecangetthesolutionbyperformingabinarysearchonL,themaximumtrafclimitvariable,tominimizeLaslongasthesolutionisfeasible,i.e.,everyloopsizeislessthanorequaltothelimitofloopsize.Thefeasibilityisveried 86

PAGE 87

4-6 .Inthisalgorithm,thelowerbound(lbound)ofLissetto1,whereastheupperbound(ubound)ofLissetton(n1)becauseeachloopsendsow1toeveryotherloopsinDMSTnetworkowmodel,thus,totaltrafcinDMSTnetworkowmodelcannotexceedn(n1).IttheniteratesabinarysearchuntilitreachestheminimumvalueofL.Therefore,thesolutionisfeasibleaswellasitisoptimizedforthemostloadbalancinggraphtopology.Thestep-by-stepoperationsofthealgorithminFig. 4-6 aregivenbelow: Inthesimulations,whenTVprogramsarebroadcast,theprogramsaredistributedevenlyamongloopsandtheprobabilitythateachprogramisrequestedisuniformsothattheamountofdatatrafcbetweeneachpairofloopscanbefairlycompared.TherequestrateofprogramplaybackfollowsaPoissondistributionandeachPVRissues 87

PAGE 88

Simulationparameters datatransferrate 1Gb/s propagationdelay 3.5ns/meter pernodedelay 240ns Disk capacity 300GB cache 32MB datatransferrate 2965MB/s seektime 4.7ms rotationallatency 3.0ms TVprograms HD-quality 19.4Mbps SD-quality 5Mbps length 60minutes requestsfordifferentprogramsevery60minutes.ItisalsoassumedthathalfofattacheddevicesarePVRs,andtherestarefornetworkdisksandshareddisksineachloop.Thedetailsofparametersusedfordisks,FC-AL,andTVprogramsareillustratedinTable 4-1 Fig. 4-7 illustratesthataverageconnectionsetuptimeinadouble-looparchitectureasabasiccaseofmultiple-loops,basedonthedescriptioninSection 4.3.1 .Wehaveevaluatedtheaveragesetuptimebasedonthedouble-loopcongurationwhileaddingmorePVRs,i.e.,from10upto70,ineachloop.AsshowninFig. 4-7 ,thesetuptimetendstoincreasewithmorePVRs.Nevertheless,wecanseethatthisconnectionsetuptimeisreasonablesinceoursystemperformslessthan16msecwithhigh-qualitymultimediadatainthe70-PVRcase,whereas[ 64 ]showsitssetuptimeasaround6.2msecusingtheFC-ALswitchdevicewithonly50%ofpossiblegeneratedtrafc. Wehavealsoevaluatedtheworstcaseofstartuplatencywhenthreeloopsareinvolvedforrelayingaprogram.Thelatencyofoursystemismainlycausedbytherstphaseoftherelayingprocesssincethestartuplatencyisgreatestwhenstreaming 88

PAGE 89

Impactonaveragesetuptimeinadouble-loop aprogrambetweentwoPVRs.WehaveobservedwithparametervaluesinTable 4-1 thatstartuplatencyisaround29.17msecintheworstcasewhenallthedevicesareparticipatingineverylooparbitration,involveddeviceshavelowestpriorities,andrequireddatablocksmustbereadfromdisks.Thisisareasonabledelayconsideringonecycletimeisusuallyover10timeslongerthanthisdelay. Moreover,inordertoevaluateaverageloopsize,ina30km30kmarea,werandomlydistributeagivennumberofloopsaslongastheCGcongurationisachieved.OncetheCGtopologyisorganizedwiththegivennumberofloops,wethencongurebothMSTandMSGtopologiesfromthedeployedloopsfortheCG.WealsovalidatetheproposedMSGalgorithmdescribedinSection 4.5.3 intermsoftheeffectivenessofthetheload-balancingamongloopsandshareddisks. 4-8 illustratesthat,whenloopsareadded,averageloopsizeintheCGcaseincreaseslinearly,whiletheMSTandMSGtopologycasesstayconstant.Essentially,theloopsizeiscloselyrelatedtotheaveragedistancebetweentwoloopssincethe 89

PAGE 90

Averageloopsize shareddiskisplacedinthemiddlebetweenthosetwoloops.Inotherwords,ifthedistancebetweentwoloopsbecomeslarger,themodiedloopsizewiththeshareddiskwouldbecomequitesizable.Inaddition,alargeFC-ALislessdesirableduetoperformancedegradationfromtherapidincreaseofpropagationdelay[ 60 ].Thus,wecanseethattheMSTtopologyperformsbestandtheCGtopologyperformsworstintermsofaverageloopsize.ThisisbecauseCGrequireseverylooptobeconnectedtoallotherloops,whereastheMSTtopologyonlyneedsoneortwoconnectedloopsviashareddisks,andtheMSGtopologyshows2(n1)connectivityonaveragebythealgorithminFig. 4-6 4-9A ,itisclearthattheCGtopologypresentstheleasttotaltrafcinagivennumberofloopssinceitguaranteesonly1-hoproutingbetweenanypairofloops,requiringatleastnC2connect-disksforthe1-hop 90

PAGE 91

BAveragerejectratio Totaltrafcandaveragerejectratio 91

PAGE 92

InFig. 4-9B ,MSTshowstheworstaveragerejectratiocomparedtoCGandMSG.CGhasthebestrejectratio,evenlessthan0.5%,sinceitguaranteesthebestloaddistributionwithitsowndedicatedroutingpathtoallloops.Nevertheless,withoutusingtherequiredhard-connectedroutingpathstoeveryloop,MSGcanperformwithlessthana5%rejectratioevenintheworstcasewithfewerconnect-disksthanCG.Surprisingly,duetomoreexibleroutingpathsthanMST,theMSGalsomaintainsaconstantrejectratioevenwhenmoreloopsaredeployed. Consequently,theMSGrevealsthesuperiorityreectedbytheextremeCGandMSTtopologies,intermsofaverageloopsize,totaltrafc,andaveragerejectratio.SincetheMSGisderivedfromtheMST,italsohasaconstantaverageloopsize.Moreover,sinceMSGfurtherallowsthegraphcharacteristic,itcanprovidemoreexibleroutingpathssupportingloaddistributions,whichcanaffecttheamountoftotaltrafcandtheaveragerejectratio.Therefore,wecanconcludethattheMSG-basedmultiple-looparchitecturecandistributehigh-qualitycontent-sharingeffectivelyamongPVRusers,deliveringoutstandingperformancetothewholesystematlowertotalcost. 92

PAGE 93

93

PAGE 94

Thisdissertationmadesomecontributionsfororganizingamultimedia-enablednetworkarchitecturewherehigh-qualityP2Pcontentsharingissupported.Inordertocongureaneffectivenetworkarchitecture,wehaveintroducedtheFC-ALtechnologyasahigh-speedandbroadbandnetworkconnectionandthePVRasamultimediahomedevice.Moreover,wehavepresentedtheshareddiskinordertoextendthemultiple-looparchitectureforascalablenetworkarchitecturedesign. Intherstpartofthisdissertation,wehaveproposedascalablemultiple-looparchitectureusingshareddisks,whichsupportsHD-qualitycontentdistributionefcientlyandeffectively.Inourproposedarchitectures,wehaveshowedthatthetopologydesignaffectsthetotalloadedtrafcamountwhichinuencesthescalabilityintermsofthetotalattachabledevicenumber,therebygivingusawaytodeterminewhicharchitecturedesignismoredesirable.Inaddition,wehaverealizedthatwehavetoreducethenumberofshareddisksinordertoobtainabetterscalabilitywhichisalsoakeyfactortodesignthemultiple-looparchitectures.Basically,wehavetriedtoreducethetopology-causedtotaltrafcbyadoptingtheCompleteGraph(CG)topologydesigninordertousethefewestnumberofshareddisks,whichistheinitialapproachtodeterminethedesirablemultiple-looparchitecture. Inthesecondpart,wehavedevisedthestoragesavingschemewhichtriestoutilizethewholesystemstorageincludingbothnetworkdisksandPVRstorage.ThestoragesavingschemeoperatesbasedonthegiventhresholdvaluewhichdeterminewhetherornotanewincomingprogramcanbestoredeitherinthenetworkdiskorPVRsothatthesystemtriestoreducetheredundancyofstoringsameprograminthesystem.Particularly,thisschemecanworkeffectivelywhenthesizeofscalablearchitecturebecomeslargesinceitwillbemuchlikelytostoreduplicatedprogramsredundantlyduetoaccommodatingmanymorePVRusers. 94

PAGE 95

Moreover,theinsertedshareddisksnecessarilycausethemodicationofthedeployedloopshape,resultinginanincreaseinloopsize.Thisimpliesthatalongerberchannelisneededwhenextendingintoamultiple-loopnetworkarchitectureusingshareddisks.Itisexpectedthattheperformanceoftheberchannel-arbitrationloop(FC-AL)decreaseswhentheloopsizeincreases,althoughitisbelievedthattheFC-ALcancoverupto10km. Inaddition,therealcostisanimportantfactorwhenconstructinganactualnetworkinfrastructure,i.e,howmanyshareddisksareneededandhowlongthephysicalberchannelisrequired.Inotherwords,theactualcosthastobetakenintoaccountaswellasthescalability. Therefore,thepracticallyconstructiblenetworkarchitecture,namedMSGarchitecture,hasbeenexploredastheorganizationofamultiple-looparchitecture,whichcanmakebalancebetweentheactualarchitecturecostandthesystemperformance.Wehasshownthattheproposedarchitecturerevealsitssuperioritybythetrade-offbetweentheCGtopologyarchitectureandtheMSTtopologyarchitecture.Infact,theMSGarchitectureexposestheconstantaverageloopsizesimilartotheMSTandtheclosetotaltrafcinadditiontoacceptablerejectratiocomparedtotheCG. Ibelievethattherearestillmoreissuesthatweneedtoaddresstofurtherimprovetheperformanceofmultimedia-enablednetworkarchitecture. 95

PAGE 96

Inaddition,thecurrentrealtimemultimediaservicescanalsoapplytooursystem.Particularly,theIPTVservicesarenowspreadingintomanymorehomes,especiallyintegratedwithhigh-speedInternetservicesandIPtelephoneservices,knownasTriple-playservices.Thisispossiblebecausetheberlinesaredeployingtosupportvariousbroadbandnetworkservices.Nevertheless,thecurrentIPTVservicesaremainlyconsideringthedeliveryofhigh-qualityTVcontentsviatheberlines.However,wecanfurthertakeintoaccountthatoursystemenvironmentcanbeapplicableintheIPTVservicessothateveryIPTVusercansharetheircontentswithothersintheIPTVnetworkarchitecture.Furthermore,variousIPTVservicesincludingbidirectionalInternetserviceswhilewatchingaTVprogramcanbealsoapplicabletooursystem.Thatis,theintegrationofIPTVservicesandoursystemenvironmentcanbethebasisofourfuturework. 96

PAGE 97

[1] Adamic,L.A.andHuberman,B.A.Zipfslawandtheinternet.Glottometrics3(2002):143. [2] Ahmed,T.,Mehaoua,A.,andBuridant,G.ImplementingMPEG-4videoon-demandoverIPdifferentiatedservices.ProceedingsofIEEEGLOBECOM.NewYork,2001,2489. [3] Androutsellis-Theotokis,S.andSpinellis,D.Asurveyofpeer-to-peercontentdistributiontechnologies.ACMComputingServey36(2004).4:335. [4] Barker,RichardandMassiglia,Paul.StorageAreaNetworkEssentials:ACompleteGuidetoUnderstandingandImplementingSANs.JohnWiley&Sons,NewYork,2001. [5] Barton,Zim.TiVo-lution.Queue4(2006).3:28. [6] Beck,M.andMoore,T.TheInternet2distributedstorageinfrastructureproject:AnarchitectureforInternetcontentchannels.ComputerNetworkingandISDNSystems30(1998).22:2141. [7] Benner,AlanF.FibreChannel:gigabitcommunicationsandI/Oforcomputernetworks.McGraw-Hill,Massachusetts,1995. [8] Bolosky,W.J.,Douceur,J.R.,Ely,D.,andTheimer,M.FeasibilityofaserverlessdistributedlesystemdeployedonanexistingsetofdesktopPCs.ProceedingsoftheACMSIGMETRICSinternationalconferenceonmeasurementandmodelingofcomputersystem.SantaClara,CA,2000,34. [9] Cahill,A.J.andSreenan,C.J.VCDN:Acontentdistributionnetworkforhigh-qualityvideodistribution.ProceedingsofInformationTechnologyandTelecommunicationsConference.Letterkenny,Ireland,2003. [10] Callaly,F.andCorcoran,P.ArchitectureofaPVRAppliancewithlong-tailInternet-Vcapabilities.IEEETransactionsonConsumerElectronics52(2006).2:454. [11] Calvert,K.I.,Doar,M.B.,andZegura,E.W.ModelingInternettopology.IEEECommunicationsMagazine35(1997).6:160. [12] Chen,S.andThapar,M.Abrechannel-basedarchitectureforInternetmultimediaserverclusters.ProceedingsofInternationalAlgorithmsandArchitecturesforParallel.Taipei,Taiwan,1997,437. [13] Chou,C.,Golubchik,L.,andLui,J.C.S.Designofscalablecontinuousmediaservers.MultimediaToolsandApplications17(2002).2:181. 97

PAGE 98

Chung,S.,Kim,E.,andLiu,J.C.L.AscalablePVR-basedcontentsharingarchitectureforcommunitynetworks.IEEETransactionsonConsumerElectronics54(2008).3:1192. [15] Clark,Tom.Designingstorageareanetworks:apracticalreferenceforimplement-ingFibreChannelSANs.Addison-Wesley,NewJersey,1999. [16] Cranor,C.D.,Green,M.,Kalmanek,C.,Shur,D.,Sibal,S.,derMerwe,J.E.Van,andSreenan,C.J.Peer-to-PeerBasedMultimediaDistributionService.IEEETransactionsonMultimedia6(2004).2:343355. [17] Dan,A.andSitaram,D.Multimediacachingstrategiesforheterogeneousapplicationandserverenvironments.MultimediaToolsandApplications4(1997):279. [18] Du,D.H.C.,Chang,T.,Hsieh,J.,Shim,S.,andWang,Y.TwoEmergingSerialStorageInterfacesforSupportingDigitalLibraries:SerialStorageArchitecture(SSA)andFibreChannel-ArbitratedLoop(FC-AL).MultimediaToolsandApplica-tion10(2000).2:179. [19] Du,D.H.C.,Hsieh,J.,Chang,T.,Wang,Y.,andShim,S.Interfacecomparisons:SSAversusFC-AL.IEEEConcurrency6(1998).2:55. [20] Eleftheriadis,A.andAnastassiou,D.MeetingarbitraryQoSconstraintsusingdynamicrateshapingofcodeddigitalvideo.LectureNotesinComputerScience1018(1995).1:89. [21] Farley,Marc.StorageNetworkingFundamentals:AnIntroductiontoStorageDevices,Subsystems,Applications,Management,andFileSystems.CiscoPress.,Indiana,2004. [22] Fumagalli,A.,Krishnamoorthy,P.,andCai,J.Thesub-channeledLightRingforbandwidthefcientstorageareanetworks.ProceedingsofIEEEGlobalTelecom-municationsConference(GLOBECOM).Taipeicity,Taiwan,2002,2832. [23] Gafsi,J.andBiersack,E.W.Performanceandreliabilitystudyfordistributedvideoservers:Mirroringorparity.ProceedingsofIEEEInternationalConferenceonMultimediaComputingandSystems.Florence,Italy,1999,628. [24] Gall,D.L.MPEG:avideocompressionstandardformultimediaapplications.CommunicationsoftheACM34(1991).1:46. [25] Garey,M.andJohnson,D.Computerandintractability:aguidetothetheoryofNP-completeness.Freeman,SanFrancisco,1979. [26] Gemmell,D.J.,Vin,H.M.,Kandlur,D.D.,andRangan,P.V.Multimediastorageservers:atutorial.IEEEComputerMagazine28(1992).5:40. 98

PAGE 99

Goossens,K.,Wielage,P.,Peeters,A.,andMeerbergen,J.V.NetworksonSilicon:CombiningBest-EffortandGuaranteedServices.ProceedingsoftheConferenceonDesign,AutomationandTestinEurope.Paris,France,2002,423. [28] Gouveia,L.andMagnanti,T.L.Networkowmodelsfordesigningdiameter-constrantedminimum-spanningandsteinertrees.Networks41(2003).3:159. [29] Guha,A.Theevolutiontonetworkstoragearchitecturesformultimediaapplications.ProceedingsofIEEEInternationalConferenceonMultimediaComputingandSystems.Florence,Italy,1999,68. [30] Heath,J.R.andP.J.Yakutis.High-speedstorageareanetworksusingabrechannelarbitratedloopinterconnect.IEEENetwork14(2000).2:51. [31] Heffeeda,M.,Habib,A.,Botzv,B.,Xu,U.,andBhargava,B.PROMISE:peer-to-peermediastreamingusingCollectCast.ProceedingsofACMMulti-media.Berkeley,CA,2003,45. [32] Hemy,M.,Hengartner,U.,Steenkiste,P.,andGross,T.MPEGsystemstreamsinbest-effortnetworks.ProceedingsofPacketVideoWorkshop.NewYork,1999. [33] Huang,M.L.andNguyen,Q.V.Afastalgorithmforbalancedgraphclustering.Pro-ceedingsofInternationalConferenceInformationVisualization.Zurich,Switzerland,2007,46. [34] ISO,IEC,JTC,andSC.ISO,IEC14165-122:2005,Informationtechnology-FibreChannel-Part122:ArbitratedLoop-2,FC-AL-2.AmericanNationalStandardsInstitute,DC,2007. [35] Jensen,JensF.Interactivetelevision:newgenres,newformat,newcontent.Proceedingsofthe2ndAustralasianconferenceoninteractiveentertainment.Sydney,Australia,2005,89. [36] Jung,S.,Kim,E.,andLee,D.DesignandImplementationofanEnhancedPersonalVideoRecorderforDTV.IEEETransactionsonConsumerElectronics47(2001).4:915. [37] Kim,E.andLiu,J.C.L.Performanceofascalablemultimediaserverwithshared-storageclusters.MultimediaToolsandApplications18(2002).1:31. [38] .Anintegratednetwork/storagearchitectureforsharinghigh-qualitybroadcastTVcontents.IEEECommunicationsMagazine43(2005).9:86. [39] .DesignofHD-qualitystreamingnetworksforreal-timecontentdistribution.IEEETransactionsonConsumerElectronics52(2006).2:392. [40] Kunii,T.L.,Shinagawa,Y.,Paul,R.M.,Khan,M.F.,andKhokhar,A.A.Issuesinstorageandretrievalofmultimediadata.MultimediaSystems3(1995).5:298. 99

PAGE 100

Legout,A.,Liogkas,N.,Kohler,E.,andZhang,L.ClusteringandsharingincentivesinBitTorrentsystems.ProceedingsoftheACMSIGMETRICSinter-nationalconferenceonMeasurementandmodelingofcomputersystems.SanDiego,CA,2007,301. [42] Li,W.Zipfslaweverywhere.Glottometrics5(2002):14. [43] Lim,S.Y.,Choi,J.H.,Seok,J.M.,andLee,H.K.AdvancedPVRarchitecturewithsegment-basedtime-shift.ProceedingsofInternationalConferenceonComsumerElectronics.LasVegas,Nevada,2007,1. [44] Long,James.StorageNetworkingProtocolFundamentals.CiscoPress.,Indiana,2006. [45] Louise,B.andBarkhuus,L.Thetelevisionwillberevolutionized:effectsofPVRsandlesharingontelevisionwatching.ProceedingsofSIGCHIconferenceonHumanFactorsinComputingSystems.Montreal,Canada,2006,663. [46] Lv,J.,Cheng,X.,Jiang,Q.,Ye,J.,Zhang,T.,Lin,S.,andWang,L.LiveBT:providingvideo-on-demandstreamingserviceoverBitTorrentsystems.Proceed-ingsoftheeighthinternationalconferenceonparallelanddistributedcomputing,applicationsandtechnologies.Adelaide,Australia,2007,501. [47] Ma,W.andDu,D.H.C.Reducingbandwidthrequirementfordeliveringvideooverwideareanetworkswithproxyserver.IEEETransactionsonMultimedia4(2002).4:539. [48] Magoni,D.andPansiot,J.J.Analysisoftheautonomoussystemnetworktopology.ACMSIGCOMMComputerCommunicationReview31(2001).3:26. [49] Marusic,B.,Finkst,T.,Jere,A.,Meza,M.,Dobravec,S.,andTasic,J.Shareit!-contenttransferinhome-to-homenetworks.ProceedingsofIEEEMediterraneanElectrotechnicalConference.Dubrovnik,Croatia,2004,669. [50] Meshkova,E.,Riihijarvi,J.,Petrova,M.,andMahonen,P.Asurveyonresourcediscoverymechanisms,peer-to-peerandservicediscoveryframeworks.ComputerNetworks:theinternationaljournalofcomputerandtelecommunicationsnetworking52(2008).11:2097. [51] Nafaa,M.andAgoulmine,N.AnalysingJoostpeertopeerIPTVprotocol.Proceedingsofthe11thIFIP/IEEEinternationalsymposiumonintegratednetworkmanagement.LongIsland,NY,2009,291. [52] Oliphant,A.andScudamore,B.TVgetsPersonal(PersonalVideoRecorders).IEEReview47(2001).5:9. [53] Pallis,G.andVakali,A.Insightandperspectivesforcontentdeliverynetworks.CommunicationsoftheACM49(2006).1:101. 100

PAGE 101

Park,H.,Yang,J.,Park,J.,S.G.Kang,S.G.,andChoi,J.K.Asurveyonpeer-to-peeroverlaynetworkschemes.Proceedingsofthe10thinternationalconferenceonadvancedcommunicationtechnology.PhoenixPark,Korea,2008,986. [55] Parnell,T.Guidetobuildinghigh-speednetworks.McGraw-Hill,NewYork,1996. [56] Phillips,B.Havestorageareanetworkscomeofage.Computer31(1998).7:10. [57] Rejaie,R.,Yu,H.,Handley,M.,andEstrin,D.MultimediaproxycachingmechanismforqualityadaptivestreamingapplicationsintheInternet.ProceedingsofIEEEINFOCOM.Tel-Aviv,Israel,2000,980. [58] Richardson,I.E.H264andMPEG-4videocompression:videocodingfornextgeneration.Wiley,NewJersey,2003. [59] Ripeanu,M.Peer-to-PeerArchitectureCaseStudy:GnutellaNetwork.TechnicalReport,2001.DepartmentofComputerScience,UniversityofChicago. [60] Ruwart,T.M.Performancecharacterizationoflargeandlongbrechannelarbitratedloops.ProceedingsofIEEEMassStorageSystemsSymposium.SanDiego,CA,1999,11. [61] Saunders,S.,Ranade,J.,andAdvisor,S.Thehigh-speedLANshandbooks.McGraw-Hill,NewYork,1996. [62] Sharda,N.K.Multimediainformationnetworking.PrenticeHall,NewJersey,1998. [63] Silverston,T.andFourmaux,O.P2PIPTVmeasurement:acasestudyofTVants.Proceedingsofinternationalconferenceonemergingnetworkingexperimentsandtechnologies.Lisboa,Portugal,2006. [64] Sinha,V.andDu,DavidH.C.SwitchedFC-AL:anarbitratedloopattachmentforbrechannelswitches.ProceedingsofIEEESymposiumonMassStorageSystems.CollegePark,MD,2000,225. [65] Springett,M.V.andGrifths,R.N.Innovationforinclusivedesign:anapproachtoexploringtheiDTVdesignspace.Proceedingsofthe1stinternationalconferenceondesigninginteractiveuserexperiencesforTVandvideo.SiliconValley,CA,2008,49. [66] Srinivasan,U.,Pfeiffer,S.,Nepal,S.,Lee,M.,Gu,L.,andBarrass,S.AsurveyofMPEG-1audio,videoandsemanticanalysistechniques.MultimediaToolsandApplications27(2005).1:105. [67] Steiner,M.andBiersack,E.W.Whereismypeer:evaluationoftheVivaldinetworkcoordinatesysteminAzureus.Proceedingsofthe8thInternationalIFIP-TC6NetworkingConference.Aachen,Germany,2009,145. 101

PAGE 102

Szabo,C.A.,Chlamtac,I.,andBedo,E.Designconsiderationsforbroadbandcommunityareanetworks.ProceedingsofHawaiiInternationalConferenceonSystemSciences.BigIsland,Hawaii,2006,50123.2. [69] Tan,W.T.andZakhor,A.Real-timeInternetvideousingerrorresilientscalablecompressionandTCP-friendlytransportprotocol.IEEETransactionsonMultime-dia1(1999).1:172. [70] Telikepalli,R.,Drwiega,T.,andYan,J.Storageareanetworkextensionsolutionsandtheirperformanceassessment.IEEECommunicationsMagazine42(2004).4:56. [71] Tennenhouse,D.L.,Smith,J.M.,Sincoskie,W.D.,Wetherall,D.J.,andMinden,G.J.Asurveyofactivenetworkresearch.IEEECommunicationsMagazine35(1997).1:80. [72] Tewari,R.,Mukherjee,R.,Dias,D.M.,andVin,H.M.Designandperformancetradeoffsinclusteredvideoservers.ProceedingsofIEEEInternationalConferenceonMultimediaComputingandSystems.Hiroshima,Japan,1996,144. [73] Thornburgh,R.H.andSchoenborn,B.J.Storageareanetworks:designingandimplementingamassstoragesystem.PrenticeHall,NewJersey,2000. [74] Vacca,JohnR.TheEssentialGuidetoStorageAreaNetworks.PrenticeHall,NewJersey,2001. [75] Vandalore,B.,Feng,W.,Jain,R.,andFahmy,S.Asurveyofapplicationlayertechniquesforadaptivestreamingofmultimedia.RealTimeImaging7(2001).3:221. [76] Vogel,A.,Kerherve,B.,Bochmann,G.V.,andGecsei,J.DistributedmultimediaandQOS:asurvey.IEEEMultimedia2(1995).2:10. [77] Wang,X.andSchulzrinne,H.ComparisonofadaptiveInternetmultimediaapplications.IEICETransactionsonCommunications82(1999).6:806. [78] Wang,Y.,Liu,J.C.L.,Du,D.,andHsieh,J.Efcientvideoleallocationschemesforvideo-on-demandServices.MultimediaSystems5(1997):283. [79] Xiang,Z.,Zhu,Q.ZhangW.,Zhang,Z.,andZhang,Y.Peer-to-PeerBasedMultimediaDistributionService.IEEETransactionsonMultimedia6(2004).2:343355. [80] Yan,B.andNg,W.K.AsurveyonthetechniquesforthetransportofMPEG-4videooverwirelessnetworks.IEEETransactionsonConsumerElectronics48(2002).4:863. 102

PAGE 103

Yang,Z.andKuo,C.C.J.Surveyonimagecontentanalysis,indexing,andretrievaltechniquesandstatusreportofMPEG-7.TamkangJournalofScienceandEngineering2(1999).3:101. [82] Zhang,Q.,Wang,G.,Zhu,W.,andZhang,Y.Q.RobustscalablevideostreamingoverInternetwithnetwork-adaptivecongestioncontrolandunequallossprotection.ProceedingsofPacketVideoWorkshop.Kyongju,Korea,2001. [83] Zhang,Y.,Huangfu,W.,andQiao,Y.Amulti-channelHDvideostreamingindigitalcommunitynetwork:videorateadaptationandplaybackadjustment.IEEETransactionsonConsumerElectronics53(2007).4:1449. [84] Zhang,Z.L.,Wang,Y.,Du,D.H.C.,andShu,D.Videostaging:aproxy-server-basedapproachtoend-to-endvideodeliveryoverwide-areanetworks.IEEE-ACMTransactionsonNetworking8(2000).4:429. 103

PAGE 104

SungwookChungwasborninPusan,RepublicofKorea,in1976.HereceivedaB.S.incomputersciencefromSogangUniversity,Korea,in2002andanM.S.fromtheComputerandInformationScienceandEngineering(CISE)DepartmentoftheUniversityofFloridain2005.Since2005,hehasbeenconductingresearchwithDr.JonathanC.L.LiuintheCISEdepartmentattheUniversityofFlorida.Hisresearchinterestsincludedistributedmultimediasystems,homeandcommunityareanetworksandarchitectures,storageareanetworks,high-qualitymultimediacontentdistribution,multimediaconsumerdevices,andIPTVservices. 104