• No results found

Multi-Timescale Modelling of Ethernet Traffic

N/A
N/A
Protected

Academic year: 2022

Share "Multi-Timescale Modelling of Ethernet Traffic"

Copied!
144
0
0

Loading.... (view fulltext now)

Full text

(1)

of Ethernet TraÆc

Patrik Carlsson

Karlskrona, December2003

Department of Telecommunicationsand Signal Processing

Blekinge Instituteof Technology

(2)

BlekingeInstituteofTechnology

LicentiateSeriesNo. 2003:09

ISSN1650-2140

ISBN 91-7295-031-5

Published2003

PrintedbyKaserntryckerietAB

Karlskrona2003

(3)

1932{2003

(4)
(5)

Abstract

Ethernetisoneofthemostcommonlinklayertechnologies,usedinlocalarea

networks,wirelessnetworksand wideareanetworks. There ishoweveralackof

traÆcmodelsforEthernetthatisusablein performanceanalysis.

In this thesis we describe an Ethernet traÆc model. The model aims at

matching multiple momentsofthe bitrate atseveral timescales. Tomatchthe

model parameters to measured traÆc, four methods havebeen developed and

tested on real traÆc traces. Once a model has been created, it can be used

directlyin a uid owperformanceanalysis.

Ourresultsshowthat, asthenumberofsourcespresentonanEthernetlink

grows,themodelbecomesbetterandlesscomplex.

Keywords

EthernetTraÆcModel,Multi-Timescale,Multifractal,BitRate,Moments,Fluid

FlowAnalysis,ModelMatching,Measurement.

(6)
(7)

Acknowledgments

First of all I would like to thank Professor Arne A. Nilsson, for accepting

me as aPhD student. I would also like to thank Dr.-Ing. Markus Fiedlerfor

plantingtheseedthat inspiredmeto startthisjourney. DocentA. Popescufor

the discussions that wehad, and my colleagues at the University during these

years. Withoutyouthejourneywouldnothavebeenthesame.

I wouldalsoliketothankmyfamily,withoutyoursupportIwouldnothave

madeit.

Patrik Carlsson

Karlskrona, November2003

(8)
(9)

Abstract vii

Acknowledgments ix

1 Introduction 1

1.1 IEEE802.3{Ethernet . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.1.1 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.1.2 RecentFeatures . . . . . . . . . . . . . . . . . . . . . . . . 5

1.1.3 ComputersandEthernet . . . . . . . . . . . . . . . . . . . 6

1.2 TraÆcMeasurement . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.2.1 ActiveMeasurements . . . . . . . . . . . . . . . . . . . . . 7

1.2.2 PassiveMeasurements . . . . . . . . . . . . . . . . . . . . . 9

1.2.3 TimestampAccuracy . . . . . . . . . . . . . . . . . . . . . 11

1.2.4 EthernetMeasurements . . . . . . . . . . . . . . . . . . . . 12

1.3 TraÆcModelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.4 Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.4.1 GeneticAlgorithms. . . . . . . . . . . . . . . . . . . . . . . 16

1.4.2 SimulatedAnnealing . . . . . . . . . . . . . . . . . . . . . . 19

1.4.3 ComparisonofGAandSA . . . . . . . . . . . . . . . . . . 20

1.5 TheNextStep{FluidFlowAnalysis. . . . . . . . . . . . . . . . . 20

2 State of the art 23 2.1 BellcoreTraces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.2 Network Measurements . . . . . . . . . . . . . . . . . . . . . . . . 25

2.3 TraÆcModelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

(10)

3 Measurements and MomentEstimation 29

3.1 Measurements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

MeasurementPoint. . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.2 BitrateEstimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

3.2.1 FractionalBits . . . . . . . . . . . . . . . . . . . . . . . . . 32

3.2.2 TimestampAccuracy . . . . . . . . . . . . . . . . . . . . . 34

3.3 MomentEstimation . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.4 Truncation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.5 Normalization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

4 Process Description 39 4.1 Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

4.2 Sub-processAnalysis . . . . . . . . . . . . . . . . . . . . . . . . . . 41

4.2.1 MomentAnalysis . . . . . . . . . . . . . . . . . . . . . . . . 41

4.2.2 NumericalProblems . . . . . . . . . . . . . . . . . . . . . . 43

4.2.3 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

4.3 TheProcess . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.3.1 Construction . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4.3.2 Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

4.3.3 Gradient. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

5 Process MatchingMethods 53 5.1 ManualMatching . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

5.2 GeneticAlgorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

5.2.1 FitnessFunction . . . . . . . . . . . . . . . . . . . . . . . . 55

5.2.2 BoundaryValues . . . . . . . . . . . . . . . . . . . . . . . . 58

5.2.3 Example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

5.3 SimulatedAnnealing . . . . . . . . . . . . . . . . . . . . . . . . . . 62

5.3.1 CostFunction. . . . . . . . . . . . . . . . . . . . . . . . . . 62

5.3.2 BoundaryValues . . . . . . . . . . . . . . . . . . . . . . . . 62

5.3.3 Example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

5.4 HeuristicMethod1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

5.5 HeuristicMethod2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

5.6 Matchingin Practice . . . . . . . . . . . . . . . . . . . . . . . . . . 72

5.6.1 NumberofTimescales . . . . . . . . . . . . . . . . . . . . . 72

5.6.2 NumberofSub-Processes . . . . . . . . . . . . . . . . . . . 72

(11)

5.6.4 MatchingProblems. . . . . . . . . . . . . . . . . . . . . . . 74

6 Resultsand Discussions 75 6.1 BellcoreTraces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

6.1.1 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . 75

6.1.2 Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

6.2 InternetAccessLink . . . . . . . . . . . . . . . . . . . . . . . . . . 82

6.2.1 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . 82

6.2.2 Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

6.3 ADSLLink . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

6.3.1 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . 90

6.3.2 Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

6.4 DiscussionofMeasurementEquipment . . . . . . . . . . . . . . . . 97

6.5 ObservationonMeasuredtraces. . . . . . . . . . . . . . . . . . . . 97

6.6 DiscussiononMatchingMethods . . . . . . . . . . . . . . . . . . . 99

7 Conclusions and Outlook 105 A KroneckerAlgebra and MatrixConstruction 107 B Derivation of Momentsand their limits 109 B.1 FirstMomentAnalysis . . . . . . . . . . . . . . . . . . . . . . . . .110

Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .111

B.2 SecondMomentAnalysis . . . . . . . . . . . . . . . . . . . . . . .112

Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .116

B.3 ThirdMomentAnalysis . . . . . . . . . . . . . . . . . . . . . . . .117

Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .122

Bibliography 123

(12)
(13)

1.1 LANcomponents . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.2 Exampleofanactivemeasurement . . . . . . . . . . . . . . . . . . 7

1.3 Threewaysto attachapacketmonitor . . . . . . . . . . . . . . . . 9

1.4 10/100Base-Twiretap . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.5 Comparisionoftimestampaccuracy . . . . . . . . . . . . . . . . . 13

1.6 Constraintsonamodel . . . . . . . . . . . . . . . . . . . . . . . . 15

1.7 GATerminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1.8 Fluid owfunnel . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.1 CDFframesizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.2 CDFinterarrivaltimes . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.1 Filestructure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

3.2 Captureheaderstructure . . . . . . . . . . . . . . . . . . . . . . . 30

3.3 Measuementpoint . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.4 Bitrateestimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.5 Fractionalbit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.6 Timestampinaccuracy . . . . . . . . . . . . . . . . . . . . . . . . . 34

3.7 Exampleofmeasurementerror . . . . . . . . . . . . . . . . . . . . 35

3.8 Estimationofbitrates atdi erenttimescales. . . . . . . . . . . . . 36

4.1 Statediagramfora2-stateMMRP . . . . . . . . . . . . . . . . . . 40

4.2 Exampleofoutputfromasub-process. . . . . . . . . . . . . . . . . 40

4.3 Sub-processwithsub-process . . . . . . . . . . . . . . . . . . . . . 45

4.4 Sub-processwithoutnumericalproblems . . . . . . . . . . . . . . . 45

4.5 Sub-processwith xedoutputratesandvaryingtransitionrates. . 46

4.6 Sub-processwith xedtransitionratesandvaryingoutputrates. . 47

(14)

4.7 Sub-processwithasymmetricaltransitionrates. . . . . . . . . . . . 48

4.8 Processbasedonthreesub-processes . . . . . . . . . . . . . . . . . 49

4.9 Exampleofathree slopeprocess. . . . . . . . . . . . . . . . . . . . 51

5.1 Arti cialprocess . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

5.2 Fitnessfunction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

5.3 GAmatchingexample,arti caldata, defaultbounds . . . . . . . . 59

5.4 GAmatchingexample,arti caldata, tightbounds . . . . . . . . . 61

5.5 SAmatching example,arti caldata,defaultbounds . . . . . . . . 63

5.6 SAmatching example,arti caldata,tightbounds . . . . . . . . . 65

5.7 HM1matchingexample . . . . . . . . . . . . . . . . . . . . . . . . 69

5.8 HM2matchingexample. . . . . . . . . . . . . . . . . . . . . . . . . 71

6.1 BellcorepOCt89 rstthreemoments. . . . . . . . . . . . . . . . . 76

6.2 BellcoreOCt89Ext4 rstthreemoments . . . . . . . . . . . . . . 77

6.3 BCpOct89model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

6.4 BCOct89Ext4model . . . . . . . . . . . . . . . . . . . . . . . . . . 81

6.5 MeasurementsetupfortheIALscenario . . . . . . . . . . . . . . . 82

6.6 IAL-1-IN rstthreemoments . . . . . . . . . . . . . . . . . . . . . 84

6.7 IAL-1-NI rstthreemoments . . . . . . . . . . . . . . . . . . . . . 85

6.8 IAL-1-INmodel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

6.9 IAL-1-NImodel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

6.10 MeasurementsetupfortheADSLscenario. . . . . . . . . . . . . . 90

6.11 ADSL8RM rstthreemoments . . . . . . . . . . . . . . . . . . . . 91

6.12 ADSL8MR rstthreemoments . . . . . . . . . . . . . . . . . . . . 92

6.13 ADSL8RMmodel. . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

6.14 ADSL8MRmodel. . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

6.15 Comparisionbetweentimestampaccuracies . . . . . . . . . . . . . 98

6.16 BetterBCpOct89model . . . . . . . . . . . . . . . . . . . . . . . .102

6.17 BetterIAL-1-NImodel. . . . . . . . . . . . . . . . . . . . . . . . .103

B.1 Statediagramfora2-stateMarkovModulatedRateProcess. . . .109

References

Related documents

Clusteranalyserna Visar att Region Skåne och Region Sydöst har en högre Väg- markeringsstandard och retroreflexion än övriga regioner, medan Region Stockholm och Region Mitt har

Two of the screened individuals appeared to be carriers of both the familial variants in this family, four of six screened individuals carried the TNNT2-variant with or

Det går också att använda information från det föregående och det efterkommande paketet för att beräkna fram en ersättning till den förlorande informationen.[2] Denna

Secondly, it is expected to ensure QoS (Quality of Service) when it comes to latency, worst-case latency and buffer fill level so that the network can perform

The IEEE Power Profile [23] defines “specific or allowed values for PTP networks used in power substations”. Some of these values that are tampered with from the default profile

En frågeställning som går att testa med prak- tiska experiment blir alltså: är samtliga spjutspet- sar från Uppåkra som uppvisar skador avsiktligt förstörda, eller kan vissa

Från Tabell 4 går det att utläsa att arbetslöshetsnivån Granger-orsakar inflationen i Australien, Nya Zeeland, Sverige och USA, vilket ger stöd för Phillipskurvans

Vad en erfaren djurskötare tittar efter vid tillsyn av djuren är viktig kunskap för ny personal, men är samtidigt något som kan vara svårt att