ITT REF: A0/1 – 5099/06/NL/NH Speaker: Dr. Cesare Rose

“TransportProtocolforDVB‐RCSInteroperablePEP”

ITTREF:A0/1–5099/06/NL/NH

WORKSHOPONSATELLITEPEPsNoordwijk–2ndDecember2008

Speaker:Dr.CesareRoseB

2

ProjectName•  TOP‐Transport

ProtocolforDVB‐RCSInteroperablePEP

DuraBon•  15months(from1st

November2006)Subdividedintwophases:• Phase1:Analysis,DesignandPreliminaryProtocolValidaBon

• Phase2:ImplementaBonandFinalProtocolValidaBon

ProjectOverview

ParBcipants•  ITS–InformaOonTechnologyServices(Coordinator)•  UniversityofRome“TorVergata”(Italy)•  IT‐InsOtutodeTelecomunicações(Portugal)•  ESATechnicalresponsible:Mr.KrisOansen

3

ProjectGoals

Objec3ve•  TodevelopanalternaBveconges3oncontrol

method[tostandardTCP],fullyinteroperablewiththeI‐PEPspecificaBonandspecificallyaimedatopBmizingwebbrowsingperformance(HTTPtransfer)intheI‐PEPenvironment,whilemaintaininggoodperformanceforotherTCPtrafficsuchaslargefilesdatatransfer(i.e.FTP).

Output•  FullspecificaBonofthenewprotocol•  SimulaBonmodule(NS‐2)•  PrototypeimplementaBon(testsoverarealsystem)

http ftp

http ftp

DAMA I-PEP

ReferenceScenario:TCPandDVB‐RCSNetworks

•  ProblemsofE2ETCPinheterogeneouspaths•  Linkasymmetry(bandwidth,delay,technology)

•  LossesoverterrestrialInternet

•  Overall,toomuchdelayonTCP’scontrolloop

•  TCPoverDVB‐RCSDAMAcontrolloop•  Extradelay(variableoverBme)perceivedatthetransportlayer

•  SatLabshasdevelopedandissuedthespecificaBonforInteroperablePEP(I‐PEP)inaDVB‐RCSframework

•  Splitarchitecture•  SCPS‐TPasatransportprotocol(congesBoncontrol:Vegas,standard

TCP+extensions)

TCP Terrestrial Internet

DVB HUB

DVB ST Terrestrial

LAN TCP

DVB-RCS

4

I-PEP

55

ProblemFormula3on

•  “SCPS‐TPwasnotdesignedspecificallyforthepoint‐to‐pointnatureofthePEP‐PEPsatellitelink…..noritisopOmizedspecificallyforwebtraffic,whichisthedominanttrafficprofile…”•  Start‐upsendingveryslow(SlowStartalgorithm).SeveralRTTs(~

halfsecond)areneededtotransmitfewkbytes(i.e.Webobject).

•  CongesBoncontrol(CongesBonAvoidance,FR‐FR)inefficientduetothehighRTT:toomuchBmeneededtoprobetheavailablebandwidth→underuBlizaBonoftheavailableresources.

•  DAMAalgorithmsmayintroducevariablecontribuBonstotheoverallRTT•  Furtherslowdownofthebandwidthprobing

•  MisleadingsignalsofnetworkcongesBon(i.e.forTCPVegas)

66

Analysis Requirements

•  CompaBbilitywithI‐PEPspec.

•  InteroperabilitywithI‐PEPsrunningonlySCPS‐TP.

•  OpBmizaBonofwebtrafficperformance

•  Goodperformanceforlargefiletransfers(i.e.FTPdownloads).

•  SatellitestararchitecturebasedonDVB‐RCSairinterface.

•  EfficientoperaBonoverallthesupportedDVB‐RCSDAMAschemes.

77

Analysis Assump3onsforthenewprotocoldesign

•  TCPdevelopedforquitelargefiletransfersovercongestedlinkswithrelaBvelylowBW‐delayproduct.

•  NewassumpBons:

•  Runoveracontrolledenvironment:point‐to‐pointlink,whosecharacterisBcsarewell‐known(i.e.buffersizes,MACalgorithms,overallBW,etc.)

•  Bandwidthmanagement/assignmentisinchargeofDAMAcontrolloop

•  Targettrafficprofileiswebbrowsing:relaBvelyshorttransfers,interspacedwithsilence.

88

Design TCPNoordwijk:thebasicidea

•  Burst‐basedtransmission:datasentinsuitablysized,relaBvelylargeburstsratherthanasasmoothflowofspacedpackets.•  AssumpBonofacontrolledenvironmentallowsburstsmanagement(i.e.

tailoredbuffers)

•  Burstsize(N)andBmebetweentwoburstsending(∆)regulatethetransmission

•  Twotransmissionphases•  IniBal“blind”phase(replacingSlowStart):N0and∆0valuesareused.

IniBalvalues:

•  N0≥commonwebobjectsize

•  ∆0suchtomatchatargetiniBalrate(N0/∆0)

•  A“ratecontrol”phase(replacingstandardcongesBoncontrol)whereNiand∆iarechangeddependingonbothlinkcharacterisBcsandloading

•  ACKBmingisusedtoesBmatetheopBmal“rate”.

99

Design TCPNoordwijk:senderbehaviour

IniBalseongs

Flow control

ACK-based estimation

Rate ControlEnd-restore init settings

1010

Design ACK‐basedes3ma3oncomponent

•  Dispersionmeasurements•  ACKpairdispersion(δ)spacebetweenconsecuBveACKs,indicatelink“servicecapacity”=Bme

neededtoprocessadatapacketandsendthecorrespondingACKs.

•  TCPNoordwijkcalculatesanewACKdispersionsample(δ)byaveragingitwithinandACKtrain.ThesampleisgivenasinputataEWMAfilter.

•  RTTmeasurements•  Foreachburst,TCPNoordwijkcalculatesRTTbyconsideringthefirstpacketofthe

burst(pilot).

•  TheminimumexperiencedRTTisstored(andconBnuouslyupdated)

•  ForeachACKtrain,ΔRTTiiscalculatedasthedifferencebetweenthecurrentRTTiandtheminimumRTT

•  Athreshold(β)isusedtodiscriminatebetweencongesBonandDAMAaccessdelay:•  βdependsontheselectedDAMAscheme

•  IfΔRTTi>β→congesBon

1111

Design FlowControl

•  FlowControlmanagesbothBmingofthebursttransmissionsandretransmissionofpossiblelostpackets

•  LossesarenoBfiedbytheACKbasedesBmaBoncomponent

•  SNACKsignaling(orduplicatedACKs)

•  RTOexpiraBon

•  Intheformercase,lostpacketsareimmediatelyretransmiued

•  Inordertonotaltertheoverallrate,thenextburstsize(Ni)isreducedbytheamountofretransmiuedpacket.

•  IncaseofRTOexpiraBon,retransmissionisperformedfromthelastacknowledgedpacketbyretrievingtheiniBaltransmissionseong:N0andΔ0

1212

Design RateControl(1/2)

•  RateControladaptstheburst‐basedtransmissiontoboththelinkcapacityanditsdynamicvariaBons.Thisisachievedintwoways:•  Increasing/decreasingtheburstsize(N),keepingTXintervalamongbursts

(Δ)constant.

•  Increasing/DecreasingTXinterval(Δ),keepingburstsize(N)constant.

•  Specifically,NandΔareupdatedateachACKtrainrecepBonbasedoninputsfromtheACK‐basedesBmaBoncomponent:δiand∆RTTi

•  Twoalgorithmsaresupported:•  RateTracking(inabsenceofcongesBon→∆RTTi<β)

•  RateAdjustement(inpresenceofcongesBon→∆RTTi>β)

1313

Design RateControl(2/2)

•  RateTrackingaimsatadapBngtransmissionratetothemaximumallowedratethroughthefollowingsteps:•  Burstsizeisgraduallyincreased(logarithmicgrowth)uptotheiniBalburst

value(N0).<N0isconsideredasareferencevalue>

•  ∆iissettotheopBmalvalueforN0‐bursts:∆i=δi•N0(maximumrateisachievedwhenN=N0)

•  RateAdjustmentreducesburstsizeproporBonallytotheexperiencedcongesBonlevel,while∆iiskeptunchanged•  SinceTCPNoordwijkisbasedonACKtrainrecepBons,burstsizecannot

bedecreasedunderagivenvalueNmin(e.g.3packets)

•  Incasesuchathresholdisreached,∆iisincreased(∆i+1=λ•∆i)

1414

Design Conceptualitera3onofTCPNoordwijksenderbehaviour

1515

PreliminaryanalysisthroughNs‐2(1/4)

•  TCPNoordwijkhasbeenimplementedontheNetworkSimulatorNs‐2(“tcp‐burst.cc”)

•  ADVB‐RCSpoint‐to‐pointsatellitelink,includingPEPandterrestrialtails,hasbeenconfiguredinordertovalidatethenewprotocolalgorithms:•  Forwardlinkcapacity:10Mbit/s

•  Returnlinkcapacity:2Mbit/s(accessedbyDAMA–CRA,RBDC,VBDCormixedschemes)

•  FTPtransfersoverTCPNoordwijk

Preliminaryprotocolvalida3on

1616


  Performanceofasingle(long)TCPconnecBon

  FTPdownloadfromsatGWtoST

TCPsequencenumbervs.sim.3me

Throughput(bit/s)vs.sim.3me

TCPNoordwijk

TCPRenoTCPNoordwijk

TCPReno


1717


  MulBpleTCPNoordwijkconnecBons:fairness&uBlizaBon



α

α

α/2

α/2

α/2

α/2α/3

α/3

α/3


1818


  TCPNoordwijkfriendliness:1TCPNoordwijk+1TCPReno



TCPNoordwijk

TCPReno

Trendforastand‐aloneTCPReno


1919

Implementa3on Prototypedevelopment

20

Implementa3on GuidelinesforDevelopment

SDL(SystemDescripBonLanguage)hasbeenadoptedforaformaldescripBonofthemodifiedSCPS‐TPprotocol.

ThefollowingrequirementsareassumedforthedevelopmentacBviBes:

•  AgatewayapplicaBonsupporBngsplitandno‐splitarchitecture

•  AllowTCPNoordwijkconfiguraBon

•  NospecialHWrequirements

•  SCPS‐TP/TCPcompliance

SCPS‐TPMITREimplementaBonhasbeenusedasabaseline+PEPGWfuncBonaliBes

2121

Implementa3on GatewayApplica3on

TCP Noordwijk +PEP

I-PEP

2222

Implementa3on ConfiguringNoordwijkRun‐3meOp3ons

•  Therun‐BmeconfiguraBonofSCPSGatewayallowstomodifyallNoordwijkalgorithmparameters

•  iniBalbursttransmissionBme,minimumacceptableburstsize,iniBalburstsize,intervalBmebetweentwoconsecuBveACKS,congesBonthresholdonaverageRTT,etc..

•  TheconfiguraBonofSCPSGatewayismadeviathedirecBvesinthegatewayresourcefile(rfile).

2323

Implementa3on GatewayLogandGraphTraces

2424

Implementa3on GatewayNoordwijkGraph

2525

Implementa3on Socwaretes3ng:ModifiedWireshark

•  WiresharkNetworkprotocolanalyzerisused

•  SCPSCapabiliBesandOpBonsidenBficaBonhasimplemented

26

Valida3onPlaformIden3fica3ons

ThreevalidaBonpla}ormshavebeenidenBfiedinordertoperformthefollowingacBviBes:

•  So~waredevelopmentvalidaBon(ITSPla}orm)

•  PrototypevalidaBonwithDAMAemulaBon(UTORpla}orm)

•  ValidaBononRealDVB‐RCSlink(ESA‐ESRINpla}orm)

Finalprotocolvalida3on

27

ITSplaform

So~waredevelopmentvalidaBon(ITSPla}orm):


28

Func3onalitytes3ng–ITSplaform

•  Increasetheperformanceupto90%on1MBdatatransfer

Forwardlink:10Mbit;

Returnlink:512kb/s;

ruavg=500ms


29


NetworkconfiguraBon:

Forwardlink:10Mbit;

Returnlink:2Mbit;

Numberofsources:3

Source1:

Datatransferred=15Mbyte

StartBme=t0

Source2:


StartBme=t0+1sec

Source3:


StartBme=t0+5sec


30


burstbehaviors


CONNECTION1 CONNECTION2 CONNECTION3

3131

•  Emulatorhasbeenconceivedanddevelopedtoreproduceinreal‐BmeservicesandQoSofDVB‐RCSnetwork,whilesomeimpairmentsarereproducedinso~ware.

•  PhysicallylocatedintheUTORl@b(5PCs)andaccessibleviaaWebinterface.

•  ImplementaBonofaDAMAmodulerunningonST,compliantwith“ETSIEN301760”.

•  Interfacesforwirelessextensions(i.e.WiFi)fortesBngofhybridnet.

UTOREmula3onPlaform


Finalprotocolvalida3on 3232

Configura3onforTOPdemo

3333

DEMO

•  hup://www.tlcsat.uniroma2.it/archives/category/laboratorio/en/•  Tohaveanaccess,pleasesendamailto:

•  [email protected]

•  Test1:IperftransferoverTCPNoordwijkwithCRA–(testBme~30s)

•  Test2:IperftransferoverTCPNoordwijkwithRBDC–(testBme~30s)

•  Test3:IperftransferoverTCPNoordwijkwithVBDC–(testBme~30s)


3434

PerformanceoftheI‐PEPtransportprotocols•  2Mbytesfiletransfer(FTP)performance

comparison

Average throughput-transfer time TCP Vegas TCP Reno TCP

Noordwijk

CRA 2,57 Mbit/s 6,21 s

2,46 Mbit/s 6,51 s

4,2 Mbit/s 3,78 s

RBDC 192,15 kbit/s 143,27 s

1,9 Mbit/s 8,64 s

3,4 Mbit/s 4,7 s

VBDC 1,01 Mbit/s 15,8 s

1,03 Mbit/s 15,45 s

2,61 Mbit/s 6,13 s


3535

PerformanceoftheI‐PEPtransportprotocols•  30Mbytesfiletransfer(FTP)performance

comparison

Average throughput-transfer time TCP Vegas TCP Reno TCP

Noordwijk

CRA 963,2 kbit/s 249,18 s

8,13 Mbit/s 29,50 s

8,17 Mbit/s 29,35 s

RBDC > 45 min. 7,24 Mbit/s 33,15 s

7,83 Mbit/s 30,66 s

VBDC > 45 min. 5,96 Mbit/s 40,23 s

7,42 Mbit/s 32,33 s


36

RealDVB‐RCSlink

ValidaBononRealDVB‐RCSlink(ESA‐ESRINpla}orm):


37

TrialsatESA‐ESRIN

•  ESRINtestbedmadeavailablebyESA•  C‐bandsatellite

•  DVB‐RCSThales/AlcatelA9780R1.0pla}orm•  Forwardlink:4Mbit/s

•  Returnlink:512kbit/s

•  DAMAalgorithms•  CRA

•  RBDC

•  VBDC


38

TrialsatESA‐ESRIN

•  Short‐objectthroughputtests•  Objectsizes:1MByte,10MByte

•  AllowassessmentofimpactofTCPstart‐up

•  Protocols•  TCPVanJacobson(non‐SCPS)•  TCPVanJacobson(SCPSopBons‐SNACK)•  SCPSVegas

•  SCPSNoordwijk(fourconfiguraBons)


TrialsatESA‐ESRIN

39Finalprotocolvalida3on

TrialsatESA‐ESRIN


TrialsatESA‐ESRIN

•  Summary•  TrialsusefultoobserveoperaBonofprotocolinreal

environment•  Shouldbeseenaspreliminary

•  Last‐minutelimitaBonsoftestbedavailabilityreducedscopeoftestsplanned

•  WithinagivenallocaBonmethod,resultsareconsistentwithwhatisexpectedfrompreviousstudy•  UnexpectedlowthroughputresultsforRBDC

•  MoredetailsneededonalgorithmusedinHub•  TCP‐Noordwijkpresentsoverallgoodperformance


TOPExtension

•  TOPextension(currentlyinprogress)aimstocarryouttestacBviBes,whichcanbegroupedasfollows:1.  InteroperabilitytestsbetweentheTOPprototyped

protocolandacommercialI‐PEPprovidedbyThalesAleniaSpace(TAS),

2.  FinalizaBonofunreliableTOPperformancetests,reviewofthetestbedandperformtests,whichwerecharacterizedbyunexpectedandunexplainableoutcomes.

42

CONCLUSIONS

•  TheNEW“TCPNoordwijk”hasbeendesigned,implementedandvalidatedthroughsimulaBon,SDL,emulaBonandrealtrials.

•  ResultsfulfilliniBalrequirementsintermsofbothcompliancewithI‐PEPandtargetperformance(handlingofWWWtraffic,steady‐statebehaviour)

•  Aprototypeiscurrentlyavailableforfurthertrials

43

44

Q&A

ITT REF: A0/1 – 5099/06/NL/NH Speaker: Dr. Cesare Rose

Documents

Transcript of ITT REF: A0/1 – 5099/06/NL/NH Speaker: Dr. Cesare Rose