PiCSE : a framework for simulation and emulation of pervasive computing applications

(1)

Pervasive Computing Appliations

Vinent Reynolds

A thesissubmitted to theUniversityof Dublin,Trinity College

infulllment of therequirementsfor thedegree of

Dotorof Philosophy(Computer Siene)

(2)

(3)

I, theundersigned, delare that this work hasnot previously been submitted to this or any

other University,and thatunlessotherwisestated, itis entirely myownwork.

VinentReynolds

(4)

(5)

I, the undersigned, agree that Trinity College Library maylend or opy this thesis upon

re-quest.

VinentReynolds

(6)

(7)

Oh odswallop! It's taken me eleven years, and it's perfet. Edmund: A

Butler's Tale a giant roller-oasterof anovel in fourhundred sizzlinghapters.

A searing inditment of domesti servitude in the eighteenth entury, with some

hot gypsies thrown in. My magnum opus, Baldrik. Everybody has one novel in

them and this one'smine,

EdmundBlakadder, 1793.

Ihavehadthepleasureofworkingwithmanyolleagueswhohavebeomesomeofmylosest

friends. Inpartiular,Anthony,Daire,Ray,AndronikosandMelaniehavesupportedmewhen

Ineededitandthatmeans alottome, morethanIan expressintheseshortwords. There's

fartoomanyotherstolistallthenames,butthanksforalltheoees,thefootball,theakes,

the bisuits, the Starraftand thebeers.

I owe a speial debt of gratitude to my supervisor, Vinny Cahill. His guidane and

patienethroughout the years havekeptthe shipsteady,and Iamindebted to him forall of

the opportunitieswhih have arisenout mytime intheDSG.

Finally, out in the real world, I wouldn't be where I am today without the support and

(8)

UniversityofDublin, Trinity College

(9)

The eld of pervasive omputing is hanging rapidly, withthe proliferation of and advanes

in hardware and ommuniation tehnologies resulting ina shift towards a large-sale,

het-erogeneous pervasive omputing world where people and devies move seamlessly between

previously isolated pervasive omputing appliations. Prototyping and evaluation of

per-vasive omputing senarios is a neessary yet inherently diult step between design and

deployment ofnew appliations, and simulation hastraditionally been both an eetive and

aepted modelling andevaluationmethodology thataddresses this step.

Previous eorts to provide generi simulation tools that fulll the unique requirements

of pervasive omputing have met only limited suess and these have typially foused on

extending or modifyingexistingtools thathave been designed for otherpurposes. There are

several existing simulators dediated to spei pervasive omputing domains suh as

wire-less sensornetworking, ontext-aware omputing, smart spaes, or intelligent transportation

systems. Additionally there are many emulators that provide platform-spei appliation

programming interfaes for the evaluation of pervasive appliations in these domains.

Per-vasive omputing nowenompasses so manysub-domains, tehnologies, and disiplines that

perhapsitisnosurprisethattrulygeneri pervasiveomputingsimulatorshaveyettobe

re-ated. Previous work generallysuers from one of two problems: simulators target a spei

appliationplatformorasub-domainofpervasiveomputing,withtheresultthatsubstantial

modiations are required to either simulate or integrate simulations of other sub-domains;

or, generisimulators thataretoo abstratrequire substantial eort in re-reatingreurring

elements whenspeializing the simulator toa partiular domain.

(10)

large-researh hallenge in this area, whih is to investigate whether a generi simulator an be

developed that an model a broad range of existing pervasive appliations as well as new

appliationsthatmayemergeinthefuture. Thehypothesisofthisthesisisthata

framework-basedapproahto buildinggeneriintegrated pervasive omputingsimulatorsandemulators

isa suiently exibleandextensibleapproah fortheevaluationandmodellingofpervasive

appliations.

Theuseofframeworksisawell-reognisedsoftwareengineeringparadigmthatis

appropri-atewhen exibility, extensibility and reusabilityare required from theoutset. A framework

omprisesa set of abstrat lasses that apture reurring patterns within a group of related

problems.

The ontribution of this thesisis to show thatusing a framework isa suiently exible

and extensible approah to reating simulators and simulations for a broad range of

perva-sive omputing domains. The PiCSE framework - the Pervasive Computing Simulation and

EmulationEnvironment -identies themostommon abstrationsfoundinthiseld,suhas

sensorsandthephysialenvironment,andprovidesanarhitetureforbuildingsimulatorsfor

themany dierent pervasive omputing sub-domains that exist. These simulators an then

be usedto reate spei simulationsof partiular senarios for these sub-domains. F

urther-more,theframework an beusedto provide an emulation environment for appliations, and

mediatetheinterationbetween thoseemulated appliationsandanysimulated omponents.

Thisthesis desribes thedesign and implementation of thePiCSE framework and shows

howthisapproahaddressestheresearhhallengeinthisarea. Theontributionisevaluated

by instantiating three senarios meeting dierent requirements suh as sale, use of diverse

hardware artefats, and omplex appliation exeution environments. These instantiations

validatetheframework'smodularity,exibilityandsuitabilityasanapproahforthereation

(11)

Aknowledgements vii

Abstrat viii

List of Tables xix

List of Figures xx

Chapter 1 Introdution 1

1.1 Pervasive Computing . . . 2

1.2 Motivation. . . 4

1.2.1 Current Approahes to Evaluating PervasiveComputing Appliations 5 1.2.2 ASetofChallengesfortheModellingandEvaluationofPervasive Com-puting Senarios . . . 7

1.3 A Framework Approah to Simulating and Emulating Pervasive Computing Appliations. . . 10

1.3.1 ThePiCSE Framework. . . 11

1.3.2 Using the PiCSE Framework . . . 12

1.4 TheThesis . . . 12

1.4.1 Implementation . . . 13

1.4.2 Evaluation . . . 13

1.4.3 Thesis Roadmap . . . 14

(12)

2.1 Introdution . . . 15

2.2 Model-Based Formalisms and Methodologies . . . 16

2.2.1 SimulationFormalisms . . . 16

The DisreteEvent SystemSpeiation . . . 16

The DisreteTime SystemSpeiation . . . 17

2.2.2 Emulation . . . 18

2.2.3 Alternative Software-Based Methodologies . . . 20

2.3 The Review Criteria . . . 21

2.4 An Overviewof Pervasive Computing Simulators . . . 25

2.5 The Key Simulators . . . 27

2.5.1 UbiWise . . . 28

RC1: FlexibleHeterogeneity. . . 29

RC2: Extensibility . . . 29

RC3: Arhiteture . . . 29

RC4: Appliation Emulation . . . 30

RC5: Interation ofEmulated and Simulated Components . . . 30

RC6: Network Simulation . . . 30

RC7: Experimental Support . . . 30

RC8: ExternalInteration . . . 31

2.5.2 Tatus . . . 32

(13)

RC1: Flexible Heterogeneity. . . 36

RC5: Interationof Emulated andSimulated Components . . . 37

RC8: External Interation . . . 37

2.5.4 UbiREAL . . . 38

2.5.5 SitCom . . . 42

2.5.6 P-VoT . . . 45

(14)

2.5.7 StarBED2 . . . 48

2.6 Perspetives . . . 52

2.6.1 RC1- FlexibilityinModellingPervasive Computing Components . . . 54

2.6.2 RC4- Appliation Emulation . . . 60

2.6.3 RC6- Network Simulator Integration . . . 61

2.7 Summary . . . 62

Chapter 3 The PiCSE Framework Arhiteture 65 3.1 Introdution . . . 65

3.2 A Framework-Driven Approahfor theTestingof Pervasive Computing Appli-ations. . . 66

3.3 A Frame ofReferene for a Pervasive Computing Simulator . . . 67

3.3.1 The Sopeof External-FaingFeatures . . . 68

3.4 Requirements . . . 70

(15)

3.6.1 MinimumRequirementsfor aPiCSE Instantiation . . . 78

3.6.2 PiCSE Instantiations asSub-Frameworks . . . 80

3.7 Supported Abstrations . . . 81

3.7.1 Supporting the Modelling of the Physial Pervasive Computing Com-ponents . . . 82

TheObjetlass . . . 83

Physial Domains. . . 85

3.7.2 Supporting theModelling oftheSoft Pervasive Computing Abstrations 86 3.8 Emulationsupportwithin PiCSE . . . 87

3.8.1 Enabling Emulation . . . 87

3.8.2 ThePhysial Arhiteture for Emulated Appliations . . . 89

3.8.3 Supporting MultipleAppliations . . . 91

Disretisingan Appliation . . . 92

Exeuting MultipleAppliations . . . 93

3.9 Experimental Support . . . 93

3.10 Summary . . . 94

Chapter 4 Modelling the Key Pervasive Computing Components 95 4.1 ThePC_Abstration lassategory . . . 95

4.2 ExeutionEnvironments . . . 97

4.3 Sensors . . . 100

4.3.1 Push and Pullmodels . . . 101

4.3.2 Measured Phenomenon. . . 102

4.3.3 Querying . . . 102

4.3.4 Charateristisof Individual SensorReadings . . . 103

4.3.5 External Interfaes . . . 104

4.4 Atuators . . . 104

4.4.1 ExeutionEnvironment Interation . . . 105

(16)

4.5.1 A Single ModelledPhenomenon . . . 106

4.5.2 A Loation-BasedAPI . . . 109

4.5.3 EnvironmentLayer Complexity . . . 111

4.6 EntityLayers . . . 111

4.6.1 Loation basedInteration. . . 113

4.7 PiCSE asa Combinatorial Framework . . . 115

4.7.1 Physial Dependenies . . . 116

4.7.2 Logial Dependenies. . . 116

4.7.3 Limiting theeetsof updates . . . 117

4.8 Requirements Analysis . . . 119

4.9 Conlusion. . . 121

Chapter 5 Evaluation 123 5.1 Introdution . . . 123

5.2 ExperimentalMethodology . . . 123

5.2.1 The Senarios . . . 126

5.3 The STEAMEmulationSenario . . . 126

5.3.1 Senario Modelling requirements . . . 127

5.3.2 Building theSenario Model. . . 129

ModellingtheSenario's Software Components . . . 129

Modellingthe Senario's Hardware Components. . . 138

Interlinking ofSimulatedHardware andSoftware Components . . . 139

5.3.3 Exeuting theSenario Simulation . . . 141

Initialising theExperimentalSenario . . . 141

Exeution of theSimulation . . . 143

5.3.4 Senario Analysis . . . 146

Insights . . . 147

Evaluatingthe PiCSErequirements. . . 153

(17)

5.4.2 Building theSenario . . . 156

Modelling the senario'ssoftwareomponents . . . 156

Modelling the senario'shardwareomponents . . . 159

Interlinking of simulatedhardwareand softwareomponents. . . 160

5.4.3 Exeution ofthemodelled domain . . . 162

Initialising the experimentalsenario . . . 162

Exeution ofthe simulation . . . 162

Insights . . . 165

EvaluatingthePiCSE requirements. . . 165

5.5 TheIntelligent Transportation SystemsSenario . . . 167

5.5.1 Senario ModellingRequirements . . . 168

5.5.2 Building theModel . . . 169

Modelling the Senario's Software Components . . . 169

Modelling thesenario'sHardware Components . . . 169

5.5.3 Exeution ofthe modelled domain . . . 173

Initialising theExperimental Senario . . . 173

Exeution ofthesimulation . . . 174

Insights . . . 175

Evaluatingthe PiCSE requirements. . . 178

5.6 RequirementsValidation and Conlusion . . . 180

5.6.1 Requirements R1,R2,andR3 . . . 180

5.6.2 Requirement R4 . . . 182

(18)

6.1 The Challenge Restated . . . 187

6.2 The Contribution . . . 188

6.3 Lessons Learned . . . 189

6.4 FutureWork . . . 190

Appendix A Appendix 193 A.1 PiCSE Arhiteture Header Files . . . 193

A.2 EvaluationSenarios . . . 230

(19)

2.1 Simulator ReviewCriteria . . . 22

2.2 Comparing the reviewriteria for the keypervasive omputingsimulators . . 53

2.3 Pervasive Computing Components Modelled . . . 54

5.1 Requirement evaluationmetris . . . 125

5.2 TheSTEAM API . . . 129

5.3 System allsmadebySTEAMto the underlying operatingsystem. . . 130

5.4 TheDUMMY-SEAR API . . . 132

5.5 Lines ofode requiredto implement theSTEAM emulationsenario . . . 149

5.6 Requirementsanalysis for theSTEAM emulation senario . . . 153

5.7 Mappingfrom systemallsto theiremulated equivalent . . . 158

5.8 Requirementsanalysis for theCar Hardwaresenario . . . 166

5.9 Requirementsanalysis for theITS senario. . . 179

5.10 Overallrequirements analysisfor R1,R2and R3 . . . 181

5.11 Thehardwareand softwaremodels requiredbythethreeevaluation senarios 182 5.12 Overallrequirements analysisfor R4 . . . 183

5.13 Overallrequirements analysisfor Requirement 5. . . 183

(20)

2.1 Overlapping Domainsof Pervasive Computing Simulators . . . 26

3.1 The PiCSE_Core arhiteture. . . 74

3.2 Coneptual arhiteture showing layers within a generi instantiation of the PiCSE framework. . . 76

3.3 Coneptual diagram ofa PiCSE simulation inAlgorithmMode . . . 79

3.4 A PiCSE instantiationinAppliation Mode . . . 80

3.5 UML Sequene diagramdepiting olloatedobjets senario . . . 84

3.6 A splitlevelimplementation ofthePiCSE supportfor emulatingappliations 88 4.1 The loation ofthe PC_Abstration lasses withinthelogial arhiteture. . 96

4.2 Objets instantiated from the PC_Abstration lass ategory are pre-dened to interat using ertainmethods. . . 97

4.3 The proess bywhih a reading passesthrough a pipeline formed of a ombi-nation of bloking and modifyinglters . . . 103

4.4 Temp_environment lass. . . 107

4.5 A PreipitationLayerobjet models rainfallwithin thesimulatedenvironment. 108 4.6 EnvironmentLayerinstantiations an interat diretlywith otherLayers . . . 109

4.7 Objetinstaneswithinloseproximity,denedbythegrid'sdisretisedspae, are bundledinto asingle list . . . 112

(21)

5.1 CommandSensor announesevent types periodially . . . 127

5.2 TheCommandSensor appliation broadasts eventsto hissubsribers. . . 128

5.3 The appropriate point at whih STEAM should be emulated must be

deter-mined bytheuser andis not onstrained byPiCSE. . . 130

5.4 STEAMis omprised ofan RTE andSUMMY_SEAR library.. . . 131

5.5 Extrats from the emulated STEAM library, as dened in steamExeEnv.h.

Thefull denitionof thisprogram maybe found intheAppendix. . . 133

5.6 OriginalSTEAM libraryand emulated STEAM library. . . 134

5.7 ComparingoriginalbehaviouroftheDummy-SEARandtheemulatedDUMMY 2

-SEAR omponents. . . 135

5.8 Extrats from the CommandSensor program, whih is dened in

Command-Sensor.pp . . . 136

5.9 Extrats fromthe TestDevie program, whihis dened inTestDevie.pp . . 137

5.10 TheTestDevieevent allbak funtion,whihis dened inTestDevie.pp . 137

5.11 Modelling of a stati objet an be ahieve by passing a loation parameter

duringthe instantiation of anExeutionEnvironment objet.. . . 138

5.12 Physial objets an be parameterised with a mobility pattern whih denes

their movement duringthesenario experiment. . . 139

5.13 Code fragment demonstratingthe interlinkingof thesenario'sTestDevie

ap-pliation and emulated STEAMmiddlewareinstane. . . 140

5.14 Code fragmentdemonstrating the interlinkingof thesenario's

CommandSen-sorappliation, emulated STEAM middleware instaneand themodelled

mo-bile devies . . . 141

5.15 Code fragment showing the initialisation of the PiCSE_Core aspets of the

STEAMsenario environment. . . 142

5.16 Code fragment showing theinitialisation oftheSTEAM senarioenvironment. 143

(22)

evaluation.. . . 145

5.19 Original STEAMlibrary and emulated STEAMlibrary . . . 147

5.20 Tableshowingthenumberofmodiedlinesofodewiththeemulated

DUMMY-SEAR library. . . 150

5.21 Photograph of the hardware being emulated. . . 154

5.22 The hardware setup of the Car Hardware Senario. The program running on

theIPAQ,onsumes sensordatavia the PICboard. . . 155

5.23 System allsmade bythe IPAQ appliation . . . 157

5.24 System all denitionles for theinvoked systemalls . . . 159

5.25 Co-loation ofExeutionEnvironment andtheIPAQappliation . . . 161

5.26 Co-loation ofExeutionEnvironment andSensor Objets . . . 161

5.27 A sreenshotshowing theloggedoutputfromtheCarHardwareSenario

eval-uation. . . 163

5.28 The Dublin Cityroadtra network . . . 167

5.29 A snippet from the ityentre.xml le desribing the road network topology

and onstraints. . . 170

5.30 A sampleentry fromthepath denition le . . . 172

5.31 A setionof odedemonstrating howthearpath leis parsed. . . 174

5.32 Parsing the CityCentre.xml maple . . . 174

5.33 Two sreenshotsfroma visualisation oftheDublin tra senario. . . 175

5.34 A vehileinterats withits environment to inform its nextmovement . . . 177

A.1 ArhitetureDiagram for typialinstantiation . . . 193

(23)

Introdution

Simulationanplayanimportantroleinthelife-yleofsoftwareengineeringprojets(Zeigler00 ).

Thisispartiularlytrueintheaseofpervasiveomputingwherethesaleofappliationsan

belarge and they areoftendesigned to be deployed in lessthan ideal physial environments

(Razandralambo 10 ). Designing a simulator that meetsthe modelling requirements of

per-vasiveomputingisadiulthallengearisingfromthreefators: thereisabroaddiversityof

appliation areaswithinthedomain;theseappliation areasarenowinterseting,andnally,

new appliation areas are emerging quikly. At present, new simulationtools are being

de-velopedto keep pae witheahnew appliation area(Mangharam 06 ;Eugster 06;Jouve 09)

within the evolving pervasive omputing domain. This thesis desribes the Pervasive

Com-putingSimulationandEmulation(PiCSE) 1

framework,itsarhiteture anddesign,the

om-binationof whihaddresses these hallenges.

PiCSE is a framework (Johnson88 ) that supports both the reation of pervasive

om-puting simulators and individual simulations through the modelling of reurring pervasive

omputing abstrations and their relationships. The reation of domain-spei pervasive

omputing simulators suh as intelligent transportation system (ITS) simulators or

smart-spae simulators is enabled by speialising the PiCSE framework's abstrations to reate

domain-spei abstrations.

Theinstantiation ofa groupofPiCSE abstrations,or of anyspeialiseddomain-spei

1

(24)

abstrations provided by a domain-spei simulator results in the reation of an instane

simulation: a simulated model of a spei pervasive omputing senario. Moreover, the

PiCSE framework supportsthe emulationof appliationsthat forman integral part ofthese

senarios, and these emulated appliations an be integrated into instane simulations. The

PiCSE simulation engine, whih underpins the framework, mediates the interation of

em-ulated appliations with simulated artefats. PiCSE's exible approah enables a person to

instantiate theframework to reate domain-spei simulatorsand simulations to evaluate a

diverse rangeof pervasiveomputing appliation areas.

The remainder of this hapter is strutured as follows. Setion 1.1 provides a broad

introdution to the area ofpervasive omputing. Themotivation and urrent methodologies

forevaluatingpervasiveomputingappliationsarepresentedinsetion1.2whihthendraws

onadisussionoftheseapproahesintheeldtoderivethekeyhallengesinthisarea. Setion

1.3 introdues the onept of frameworks and suggests why a framework-driven approah

might be suessful inaddressing these hallenges. A desription of the ontribution of the

thesisandthe evaluationmethodology used isprovided insetion1.4and aroad map ofthe

remainder ofthe thesisonludes thishapter.

1.1 Pervasive Computing

Pervasive omputing (Weiser91 ; Weiser 93) as an area of researh is one that enompasses

more foused areas suh as ambient intelligene (Bresiani04 ), sensor networking (He04 ),

ontext-awareomputing(Benereetti 01),andsmartspaes(Dearle 03 ). Considerthe

follow-ingpervasiveomputingappliations. Smart-spaes(Tapia04)areoftenpassiveappliations,

atinginthebakgroundanddrivenbyloalsensors,whihanthenadaptsomeaspetofthe

spae suh as aess ontrol (Selvarajah 10 ) or telephone all re-routing to a user's

require-ments. An environmental-monitoring wireless sensornetwork (Ji 04) ouldomprise sensors,

perhaps monitoring preipitation, noise, or temperature, attahed to wireless sensor motes

that propagate that sensor information to other sensor motes through an ad-ho

ommuni-ationnetwork. Intelligent transportationsystems(Klein 01 )typiallyuseembedded sensors

(25)

the throughput of thevehiles in the system. Despite thediversityof these appliations, all

of theman beonsidered to liewithin the broad domain ofpervasive omputing.

Morereently, there hasbeen a shift towards the integration of heterogeneous pervasive

omputing appliations. As adoption and deployment of pervasive omputing tehnologies

takeshold,the physial and logialboundariesbetween these appliationsarebeingredued

andnowoverlapinmanyases (Handte09 ). Thetraditionalnotionofapervasiveomputing

appliationasalosed,isolatedsystemisbeingerodedwiththeintegrationofservie-oriented

arhitetures into pervasive omputing appliations (Guinard10), enabling users to exploit

the serviesand funtionalityof dierent pervasive omputingappliations astheymove

be-tween them. Iftheintegrationofthese pervasiveappliationsis tobesuessful, information

desribingusersoftheseappliationsandtheappliationsthemselvesmustbefreelyavailable

and understandable by all so that it an be onsumed by all pervasive omputing

applia-tions when and where they need it. As a result, there has been attempts reently towards

entralisingtheseservies: tworeentexamplesareGoogleLatitude 2

whihprovidesservies

formanaging auser'sloation andConserv(Hynes09 )whihmanages awiderspetrumofa

user's ontextual datasuh astheiralendar events andpersonaldevies thatmight identify

them.

In addition, a new lass of ad-ho pervasive omputing is now emerging with the user

at its entre. In itizen sensing (Campbell 08) or partiipatory sensing as it is sometimes

known, userswith powerful sensor-enabledmobiledeviesareaugmenting,and insome ases

replaing, statiembedded sensors inthe physial environment. The wisdom ofrowds, or

their olletive sensed information in this ase, is leading to innovative rowd-soured

ap-proahestoenvironmentalmonitoring(Lu 09)andevensupportingthereationofinterative

artinstallations (Vyas10 ).

These examples represent a small sub set of all the domains that pervasive omputing

address,yetdespitethediversenatureofthesepervasiveomputingsub-domains,therearea

setof reurringharaterististhatareommon withinthem. Theyaretypiallydeployed in

sensor-equippedspaes,whetherthosesensorsarestatiormobile,whereinformation

regard-2

(26)

ingtheenvironment and loalontextual information isleveragedbyappliationsto provide

more tailored and ontextualised funtionality and servieswithin thatenvironment. In

ad-dressing this area, pervasive omputing appliations have to ontend with many non-trivial

problems inluding dependability, large sale, physial distribution, seurity and timeliness.

Moreover, typial environments in whih these appliations are deployed inlude buildings

forsmart-spaesenarios(Tapia04),ruggedterrainforenvironmentalmonitoring (Ji04)and

roadnetworksforintelligent transportationsystems(Klein 01 ). Theseenvironmentsareoften

physiallyinaremoteloationmakingdeployment andmaintenanebothdiultandostly.

1.2 Motivation

Modelling and evaluation plays a ritial role in the life-yle of any software engineering

projet but presents some partiular hallenges in pervasive omputing appliation

senar-ios. The appliation typially extends beyond the omputer into the physial environment

through the use of sensors, atuators, or loation-based servies. The impliations for the

testingofthesesenarios isthattheomplete pervasiveomputingeo-system, inludingthe

environment, devies, their users and other fators should be onsidered. However, this

in-trodues diulties thatmust be overome. Testing inphysial environments suh asoean

oors (Jiang 09), oes and homes (Selvarajah 10 ) to road networks (Salim08) and sports

environments (Kranz 07) an be invasive, disruptive and often hampered bybureauray, if

aessis even available. Manyof these appliations areprediated upon hardware

tehnolo-giessuhasheap,reliablesensors,forexample,radiofrequenyidentiationtags(Xu10)or

wirelesssensors(Medagliani 10 ),whihareyettobewidelydeployed orevenavailableinthe

salesthatareenvisioned,and asaresult, aurate real-world deploymentsinsome domains

may not yet be feasible. Pervasive omputing appliations in the areas of intelligent

trans-portation systems or urban-sensing for example, an involve theinteration of thousands or

tensofthousands of elements. Finally,theevolvingnatureof theeldmeans thatnewareas

areemerging and the pervasive omputing ommunity urrently laks the appropriate tools

to evaluate some of these areas suh as partiipatory-sensing (Campbell 08). In these ases,

(27)

suhastheseimpatuponthetimeandeortrequiredtoevaluateanddebugapervasive

om-putingprototype,partiularly whenonsidering thatanexhaustivesetofontrolledsenarios

mayhave to be deployed inorder to evaluatea prototyperigorously.

1.2.1 Current Approahes to Evaluating Pervasive Computing

Applia-tions

(Nakata07 )liststestbeds,small-salelaboratorytestsandsimulation asthreemethodologies

for evaluating pervasive omputing appliations. Whih of these methodologies is hosen to

evaluate any partiular pervasive omputing senario is inuened by experimental fators

suh asthe appliation's urrent stage of development and on what the desired outome of

the evaluation is. The methodology hosen an additionally be inuened by loal fators

suh asthe tester'saessto hardware deviesand testingfailities,osts, andthetimethat

is available to test theappliation. If the purpose ofthe evaluation isto validatewhether a

pervasive omputing appliation is robust enough to be ready for a ommerial release and

subsequent real-world deployment, then anappropriate evaluation methodology might be to

usea large-sale testbed,whereas inthe reationof an appliation prototypeinan emerging

large-sale eld, asimulation-based evaluationmaybeused.

PhysialtestbedssuhastheTwist(Handziski06 )andMoteLab(Werner-Allen 05)testbeds

provide an experimental platform in whih appliations are exeuted on hardware devies

in a stable environment, where there is usually a supporting infrastruture in the form of

bak-hannels and monitors for the management, ontrolled exeution and debugging of

ex-periments. Using nativehardwaredeviesastheplatformforan experimentpotentiallygives

greaterredibilitytoresultsgatheredfromanyexperimentwhenompared,forexample,with

simulatedresultshowever,themethodologyandsupportinginfrastrutureusuallyrestritsthe

testbedto thatpartiularplatformand thus to aspeidomain ofpervasiveomputing. To

date, most testbeds are very domain spei, have not been designed with integration of

otherdomainsinmindandthus arenotsuitableasageneripervasiveomputingevaluation

platform.

(28)

experiments areexeutedon realhardwaredevies. In ontrast to testbedshowever, there is

rarelya permanent infrastruture supporting theexperiments, resulting in experiments that

areoftennot verysalable and thatan be diult toreprodue inotherlaboratories.

Finally, there are several existing pervasive omputing simulators but few thatreognise

or reet that pervasive omputing is beoming a more multi-disiplinary domain with new

emerging areas of researh and their rossover with existing areas of researh. As an

eval-uation methodology, simulation's popularity is demonstrated by its wide aeptane within

the pervasive omputing ommunity as a valid methodology, and many simulation libraries

are open-soure and are available for little or no ost (Fall 01 ; Martin 06a) so it has a low

barrierof entry in this respet. However, in general,these simulators arerarely designed for

inter-operation(Kuhl 99 )andarenotapableofmodellingthediverserangeofpervasive

om-putingsenarios that exist,i.e., a network simulator isnot readily adapted to themodelling

ofaontext-awareomputingsenariobeauseitwasnot intendedtomodeloneptssuhas

userbehaviourandativities. Therefore,anysimulatedsenariothatinludesbothaspetsof

ontext-aware omputing and wireless networking requireseither the reation of a new

sim-ulationtoolfor thatpartiular domain,or the modiation and integrationof existingtools.

Inadditiontosimulation,emulationprovidesalowostalternativetotheevaluationof

appli-ationsbyremovingthephysialrequirement fortheintendeddevieplatformtobeavailable

for theevaluation ofa partiular appliation. Salableand ost-eetive prototyping an be

ahievedasmultipleinstanesofanemulatoranberuninparallel,withouttheexperimental

overhead of managing and deploying an appliation on thehardware devies. Many existing

emulators within the pervasive omputing domain are written for appliations targeted for

a single spei platform. For example, the TOSSIM emulator (Levis03) supports TinyOS

(Hill00 )appliationsonly,althoughtheseanrunonseveralrelatedhardwareplatformssuh

astheTMoteSky 3

andtheMoteIV.Byonstraininganemulatortoasingleappliation

plat-form,anemulatorisimpliitlyonstrainingitssopetoaspeidomainastheseappliation

platformsare generallydomain-spei, inthisase, wireless sensornetworking.

3

(29)

1.2.2 A Set of Challenges for the Modelling and Evaluation of Pervasive

Computing Senarios

This thesis is onerned with methodologies that are both suitable for the evaluation of a

wide rangeof diversepervasiveomputingsenarios andthatan be extendedtoevaluate new

senarios thatmayemergein thefuture.

Itanbeseenfromtheapproahesthathavebeendesribedinthepreedingsetionsthat

testbeds,small-salelaboratorytesting,simulatorsandemulatorsmeettheserequirementsto

varying degrees of suess. Testbeds are very eetive tools and an provide exellent

ex-perimental support, however they are expensive to reate and maintain, and are usually

onstrained toasingleappliation domain andhardware platform. Inmanydomainssuh as

intelligenttransportationsystems(ITS)andurban sensing,large-saletestbedsmaynoteven

befeasibletoanyoneexeptthelargestindustrialresearhlabs. Whilstitispossibletoreate

bespoke small-sale laboratory tests for almost any domain, these are inherently limited in

sale, areoftenbuiltwithout onsiderationfor extensibilityandrequire physial

reongura-tionfor eahnewsenario. Intheaseofbothexistingsimulatorsand emulators, therearea

widerangeofdomain-speitoolsandlibrariesinexistenewhihareoftenextensiblewithin

their domain but annot be easily integrated witheah otherdespite their software-only

ap-proah. They additionally address omplimentary aspetsof pervasive omputing senarios.

Simulatorsan modelandbeusedtoevaluatephysialelementssuhassensordevies,users,

awirelessnetwork,andtheenvironment,whilstanemulator anmodeltherun-time

environ-ment of appliations thatexeute within those senarios and potentially interat with those

elements. Inonlusion, there is no one size ts all evaluation strategy that an be applied

to pervasive omputingappliations however,a exible andombinedapproah of both

sim-ulation and emulation ould potentially support the evaluation of large-sale heterogeneous

appliations.

Based onthe issuesidentied inthe overviewof pervasive omputingdomains insetion

1.1 andin the rangeof modelling and testing methodologies usedto evaluate those domains

in setion 1.2.1, the following researh problem has been identied. As pervasive

(30)

onsiderationwhen evaluatingthose appliations isalso inreasing. However, all ofthe

eval-uationmethodologies suerto aertaindegree fromthe sameweakness,inthattheyaretied

to a spei pervasive omputing domain and are generally not designed to be extended to

evaluatenew appliationswhih oulddraw frommore than oneindependent domain.

One of the main arguments for maintaining the status quo of testing individual aspets

of partiular pervasive omputing domains in isolation is the need for rigorousness in the

experimentalmethod. Byisolatingasmanyexperimentalfatorsaspossible,itistheneasier

to ahieve quantiably independent, reproduible and objetive results. Certainly, using a

singleindependent simulator isatleasttheoretially moresuitable forthis task. However, as

hasbeennoted,newmulti-disiplinary areasofpervasiveomputingsuhasvehiularad-ho

networks(VANETs)andsmartspaesmakesitmorediulttotreattheseareasinisolation.

In the ase of VANETs(Klein01 ), the behaviour model responsible for the movement of a

vehileisafator inthe eetivenessofarouting protoolthatmight be runningwithinthat

VANET. Of ourse, the behavioural model of the vehile an be redued to a plug-in or a

mobilitypatternthatmightbeintegratedintoanetworksimulator,butthatmodelstillhasto

bebuilt byanexpertinthevehiular ommunity. Furthermore,theremaybeappliationsin

thefuturewherebytheoutomeofanappliationthatisusingtheroutingprotoolinuenes

thebehaviourofthe vehile. In thatase, thelimitations ofasimple plug-inbasedapproah

willbe exposedand theintegrationof both networkand vehilesimulators will thenhave to

ahieved.

Thefollowinghallengeshavebeenidentiedaskeyissuesinaddressingthisopenresearh

problem:

1. FlexibleHeterogeneity: Anysimulatororemulatorshouldbeabletosupport

hetero-geneouselements, both hardwareand softwarebased, and thefatthat these elements

aninterat inamannerthatisunrestritedbyanyadoptedapproah. Theseelements

mayomefrom dierent pervasive omputingdomains.

2. Extensibility: Anysimulatoror emulatorshouldbeableto aommodatenew

emerg-ingareasofpervasiveomputingnotyetidentied,wherethesenewareasanomprise

(31)

3. Reusability: Anysimulator oremulator should simplifythetaskof buildingand

inte-grating elements thatareto be evaluated.

Simulation andemulationaretwo evaluationmethodologies thatansupportlarge-sale

se-narios and that are ommonly implemented using extensible arhitetures, albeit generally

within their domain only. The hypothesis of this thesis is that a new framework-based

ap-proah to building generi integrated simulators and emulators for pervasive omputing

ap-pliations is a suiently exible, extensible, and reusable approah for the evaluation and

modelling of these appliations. These three hallenges are not exhaustive by themselves

however. There areadditional hallenges inthis areathat inlude:

1. Modelling Large-Sale Senarios: Manypervasive omputing appliations require

theinteration of manythousands of interatingfeatures suh asappliations, sensors,

andusers,andinsomeasesevenmore. Thismotivatestheneedforasalableapproah

inaddressing this hallenge.

2. Timeliness: Therealisti modellingof appliation behaviour inmany pervasive

om-putingdomainsneessitates the aurate modellingofthetimetakenfor anevent suh

asa sensorreadingor an appliation behaviour.

Whilsttheseareimportant hallengesinthe simulation of pervasiveomputingappliations,

they are not the key hallenges when addressing the issue of the modelling and simulation

of overlapping of newand emerging pervasive omputingappliation domains. Nevertheless,

these two hallenges present seondary hallenges and their onsideration is addressed in

the thesis where appropriate. Finally, appliation-spei requirements suh asseurity and

dependability provide additional hallenges but these are pertinent to the appliations that

may run within the senarios and are not spei features thatare addressedindependently

(32)

1.3 A Framework Approah to Simulating and Emulating

Per-vasive Computing Appliations.

Johnsonetal(Johnson 88 )dene a framework as

a set of lasses that embodies an abstrat design for solutions to a family of

relatedproblems.

Thisdenitionenapsulatestheabstratnatureofframeworks,whileapturingtheir

suitabil-ityto addressing families of related problems. Other denitions, notably those of Campbell

(Campbell 91 ) and Booh (Booh93) inorporate the onept that the olletion of lasses

onstituting a framework, interat in a dened manner or pattern. The benets of using

frameworks are well known, and Shmidt (Shmidt 97 ) lists these primarily as modularity,

reusability and extensibility. He states, thatframeworks

enhanemodularitybyenapsulating volatileimplementation detailsbehind

sta-ble interfaes. This loalisation redues the eort required to understand and

maintain existingsoftware.

Thisdenitionhighlightstheuseofstableinterfaesdened byaframeworkto abstratfrom

theimplementation ofthe framework. Hegoeson to state thatstableinterfaes

providedbyframeworksenhanereusabilitybydeninggeneriomponentsthat

anbe reappliedto reate new appliations. Framework reusabilityleverages the

domainknowledgeand prioreort ofexperieneddevelopers inorderto avoid

re-reatingandre-validatingommonsolutionstoreurringappliationrequirements

andsoftwaredesign hallenges.

Thisdenitionofreusabilityaknowledges thatifreurringrequirementsandsoftwaredesign

hallenges an be identied amongst a set of related problems, then these an be exploited

within a framework to redue the eort in reatingnew appliations. Finally,Shmidt et al

(33)

by providing expliit hook methods that allow appliations to extend its stable

interfaes.

and that

framework extensibility isessential to ensure timely ustomisationof new

appli-ation serviesand features.

Thebenetsofextensibilityarelearwithinframeworks,asitmeans thattheframeworkan

be extended to support new appliation features identied inaddition to theoriginal set of

relatedappliation features.

1.3.1 The PiCSE Framework

The PiCSE framework is implemented as a set of abstrat lasses that an be grouped into

two lass ategories (Kruhten 95), the PC_Abstrations and the Abstrat_Interfaes lass

ategories,thataddressthethreekeyhallengesidentied;exibleheterogeneity,extensibility,

and reusability. A third lass ategory, theCore_Components lassategory,provides a set

of ore omponents, for example, the simulation engine, that underpin and are ommon, to

all simulationsreated using the PiCSE framework.

PiCSE's PC_Abstrations lass ategory provides the lass denitions of reurring

ele-ments of pervasive omputing appliations suh as sensors, environments, and appliation

platforms. These abstrations are designed to be suiently generi that they may be

in-stantiated to reate a diverse range of onrete instanes with variable harateristis, thus

meetingtherequirementofsupportingheterogeneity. Inadditiontothis,theabstrationstake

into onsiderationthelikelyinterations betweenthese instantiations, andtheframework

ap-proah allows onreteinstanes to be modularly ombined, allowing theexible reation of

more omplex instantiations.

Theframework'sAbstrat_Interfaes lassategory omprisesasetofabstratinterfaes

that also underpin the PC_Abstrations lass ategory, allowing for the extension of the

framework to reate new abstrations. These set of abstrations apture onepts suh as

(34)

arereurringarosspervasiveomputingdomains. Thislassategoryaddressesthehallenge

of extensibility within the framework's arhiteture that an aount for emerging domains

inthe future.

The framework-driven approah addresses thehallenge of reusability by providing

om-monsolutionsto reurringappliationrequirementsandsoftwaredesignhallenges. Allthree

lassategories, the PC_Abstrations,Abstrat_Interfaes andCore_Components lass

at-egoriesontribute herebyproviding the base of an extensibleframework,that also aptures

how these omponents an potentially interat. The apture of these reurring omponents

and their interations in a generi way within the framework redues theeort required to

reate newsimulatorsand simulations.

1.3.2 Using the PiCSE Framework

ThePiCSEframeworkanbespeialisedbydomainexpertstoarangeofpervasiveomputing

sub-domains. Two experts in separate domains an speialise the PiCSE framework using

PiCSE's reurringinteration patternsto reate simulators for their respetivesub-domains,

andthatthese simulatorsan themselves be provided intheformofframeworksthatan be

furtherspeialised where appropriate. Furthermore, simulated and emulated elementsof the

respetive sub-domainsan exist andinterat within the same simulated spaeifdesired.

1.4 The Thesis

Domain-spei simulators address partiular appliation domains suh as wireless sensor

networking or ontext-aware omputing and these play an important role in the evaluation

of individual appliation areas. However, their onstrained and inexible arhitetures do

not meet the modelling and evaluation requirements of large-sale, heterogeneous pervasive

omputing appliations so there remains a signiant open researh hallenge in this area,

whihisto investigate whetheragenerisimulator an be developed thatanmodelabroad

range of existing pervasive appliations aswell as new appliations that may emerge in the

(35)

This thesis addresses this open researh hallenge. PiCSE's framework-driven approah

and lass ategory implementations will be shown to be suiently exible and extensible

that users an reate simulators for a broad range of pervasive omputing appliations that

have diverse harateristis, and isanovel ontribution to thestate oftheart inthiseld.

1.4.1 Implementation

Thisframeworkapproah is implemented asaset ofC++ (Stroustrup98 )libraries, that are

grouped into three separatelassategories. Thesimulation engine underlying thesystemis

builtuponanexistingevent-basedsimulator,Simpak(Fishwik95),whihhasbeenextended

usingamulti-threadapproahtoseamlesslyandexiblymanagemultipleheterogeneous

emu-latedappliationsandsimulatedhardwareomponents. Furthermore,theenginesupportsthe

interationofboththesesimulatedhardwareomponentsandemulatedappliations, enabling

realistiinputand outputhannelsto be reatedfor theemulatedappliations. Appliations

that interat diretly with an underlying operating system via system alls or alternatively

thatarebuilt upon middleware an be modelled.

1.4.2 Evaluation

Three instantiations of the frameworkwerereated to validate thehypothesisof thethesis's

framework-drivenapproah. IntherstSoial-SensingSenario,anevent-basedmiddlewareis

emulated. Appliations that arebuilt upon that middleware thenexhange messages whilst

moving around a onstrained physial spae. In the seond Car-Hardware Emulation

Se-nario, Linux system alls are emulated, and appliation instanes use those systemalls to

interat withsimulated sensors reading data from theenvironment. The third and nal

In-telligent TransportationSystems Senario models vehiular tra in partof Dublin, Ireland

anddemonstratesthesimulator'ssupportforlarge-salesimulations, aomplexenvironment,

(36)

1.4.3 Thesis Roadmap

Theremaininghaptersof thisthesisprovideadetailed desriptionofthePiCSE framework,

its ontribution to the state of the art and an evaluation of the work. Chapter 2 provides

an examination of the state of the art in simulators and emulators for pervasive omputing

andrelated domains. Chapter 3desribes theframework,its arhiteture and design and an

additionalsetionwithinthathapteralsodesribestheframework'ssupportfortheemulation

ofappliations. Chapter4 examinestheimplementation and featuresofthekey omponents

supportedinmore detail. An evaluationof thethesisand itsontributions is doumentedin

Chapter5 andthe onlusionsarepresentedinChapter 6.

1.5 Summary

Thishapteroutlinesthemotivationfor,approah,andontributionofthethesis-thedesign

of a framework that an be speialised to reate simulations of pervasive omputing

senar-ios. An introdution to the area of pervasive omputing is presented to dene the sope of

thethesis. The work isthen motivatedbyexamining theurrent methodologies for

evaluat-ing pervasive omputing senarios and identifying a set of open hallenges. An overview of

PiCSE's framework-based approah is thenprovided before the ontribution of the thesis is

(37)

Related Work

2.1 Introdution

The use of simulation as a valid methodology for themodelling and evaluation of pervasive

omputingappliationshasbeen established, andthreeorehallenges,exibility,

extensibil-ity,andreusability have been identied thatmustbeaddressedbygenerisimulatorsfor the

broad pervasive omputing domain. This thesis desribes a modular generi and extensible

approah that enables the simulation of a broad range of existing pervasive omputing

ap-pliations as well as new appliations that may emerge in the future. The objetive of this

hapteristo examineestablishedandpriorworksinthisareathathaveattemptedto address

these researh hallenges. In doing so, the strengths and weakness of these approahes are

identied and these will be used to derive requirements for the proposed framework-driven

approah thatforms theontribution to this thesis.

In this hapter, an overview is provided of the many simulators in this eld; examining

both generi pervasive omputing simulators that attempt to address the entire domain, as

well as some of the simulators for the many sub-domains of pervasive omputing. What is

established is that there are few truly generi simulators that take a exible and extensible

approah to modelling the pervasiveomputing domain asawhole.

This hapter begins by providing a brief overview of some of the lassial simulation

(38)

aset of evaluation riteria are identied by whih thesimulators established inthis state of

the art will be examined. A broad overview of the representative work in this eld is then

provided, before seven of the most well known and ited simulators representing the state

of the art inthis spae arethen examined. The salient pointsand other notable featuresof

eah simulator arenoted, and these simulators aremeasured againstthe evaluation riteria.

Finally, Setion 2.6 takes a step bak and oers a wider view on the issues raised by the

examinationof the simulators. Additional simulatorsareonsidered atthis stage to broaden

theoverallperspetive,andassess alternative approahesandfeaturesthatarenotemployed

bythose examined indetail.

2.2 Model-Based Formalisms and Methodologies

2.2.1 Simulation Formalisms

Simulationisasoftware-orientedmodellingandevaluationmethodologyandisoneofthemost

popular and widely usedwithin the pervasive omputing ommunity today. A simulation is

adesriptionofa system,or atleastofsome ofits omponentsexpressedintermsofasetof

statesand events (Zomaya 96 ). Several simulation formalisms exist (Law00), most notably

thedisreteeventsystemspeiationbutalsothedisretetimesystemspeiation,andthe

dierential equationsystemspeiation formalisms.

The Disrete Event System Speiation

The most popular formalism, the Disrete Event System Speiation, known as DEVS

(Zomaya 96), speies a system in terms of a time base, a set of inputs, outputs, urrent

state, and a set of state transition funtions. An implementation of a DEVSsimulator

typ-ially ontains a time-ordered list of events, often in the form of funtion pointers, a lok,

some state information, a set of funtions for handling events and a loop that advanes the

simulation. As the lok progresses to the time of the next sheduled event, that event is

removed from the event list and any orresponding funtionality to that event is invoked.

(39)

andseondly,theeventlistmaybeupdated,wherebyeithermore eventsaresheduledinthe

future or existing future events are anelled. The hallenge in building a suessful model

or simulation of any system is in the implementation of funtionality assoiated with any

proessedevents.

DEVS, as a formalism, an also be extended to parallel and distributed arhitetures

(Fujimoto00 ; Tropper02). Simulations of omplex systems an take a long time on single

sequential mahines, and opportunities to parallelise these simulations an greatly speed up

the exeution of these simulations. The amount of atual gain that an be ahieved in suh

systemsisdeterminedbyanumberoffatorsinludingthespeiationofthelevelofdelity

required and theinter-dependene of anyparallelised datasetsand proesses.

Oneweakness of theDEVSformalism is thatthe approah isnot easily adaptedto

real-time simulations, i.e., simulations thatappearto proeedat thesame speed asthemodelled

behaviourwouldappearintherealworld. IntheDEVSformalism,thesimulation stepsfrom

event to event, andthere isno behaviour modelled or exeutedinbetween those events. As,

suh,thereisnomodellingofthepassingoftime inbetweenthoseevents, andthesimulation

usually ompletes as quikly as possible. Simulators built using the Disrete Time System

Speiation formalismtake some steps towards addressingthis deieny.

TheDEVSmodelisarguablythemostpopularmodelamongstthesimulationommunity

at the present timeand several onferenes 1

and workshops 2

,both aademi and industrial,

aredediated to solely advaning themodel. There are manypopularDEVSsimulators and

tools thatare pervasiveomputing orientedsuh asns-2 (Fall 01 )and Diasim(Jouve 09). In

addition to that, there are several generi simulation tools and libraries, suh as C-Sim 3

,

JavaSim 4

,and SimPywhihare alsoavailable underopen sourelienses.

The Disrete Time System Speiation

IntheDisreteTime SystemSpeiation (Zomaya 96), asimulation advanes atxed

time-steps,andateahtime-steptheorrespondingmodelledfuntionalityorbehaviourisinvoked.

1

http://www.ss.org/onfern/springsim/springsim09/f p/springsi m09.htm

(40)

Animplementationofthis formalismtypiallyontainsalok, somestateinformation,a set

of funtions that model that state information, and a loop that invokes those funtions at

eahxedtime-stepthusadvaningthesimulation. IntheDisreteTimeSystemSpeiation

(DTSS)formalism, allevents ourat thesame logial time and areexeuted inthe order

inwhih theyweresheduled, whereasintheDEVSformalism theevents aretotally ordered

globally.

Real-timesimulations areapartiular variationof the DTSSsimulation formalismwhere

the time steps are synhronised with the time of an entity observing the simulation. In

this ase, the simulation appears to run at what is alled the wall-time. This partiular

variation has many benets, most notably that it appears real froma timing perspetive to

an external observer and is thus suitable for approahes where the observer plays a role in

the simulation. This is often the ase where there is a graphial interfae for the observer

in, for example, a training simulator, or a game. This synhronisation with the real world

introduesa timelinessrequirement of ourse,whihis thatallevents thatmustbeobserved

externallymustbe modelledwithinthattime-step. Thisanintrodue anartiial bound on

thenumber of events that an be simulated given that a simulation is being exeuted on a

deviewitha xednumberof omputingresoures.

2.2.2 Emulation

Analysisof appliation behaviour isan important part ofthe evaluationof apervasive

om-putingsenarioandisaddressedbynumerousemulators(Girod04a ;Sobeih05 ). Anemulator

an be dened asa systemthatimplements thebehaviour of anoriginal system, i.e., athird

systemannot dierentiate between the behaviour of the original and the emulated system.

Howeverduetosystemresouresandfatorssuhastherun-timeinterpretationofinstrution

sets, there may be dierenes in the speed at whih the emulated appliation runs, usually

slower. Unlikeasimulationwhihattemptstomodelthestate ofasystem,anemulator's

em-phasisison realistiallymodellingtheinteration between thesystemandan external party,

anappliation inthis ase. Emulators allowappliationsto be exeutedasiftheywere

(41)

meet within the overall modelling and evaluation of pervasive omputing senarios. This is

important as it allows developers to bypass some of the problems inurred whilst bridging

the gap between a simulatedalgorithm and areal appliation. Theatof implementing and

evaluatingode twie(Nishikawa 06),onefor the simulation and onefor theatual

imple-mentation,istimeonsumingandthetranslationfromalgorithmtoappliationanintrodue

dierenes inappliation behaviour. Emulation addresses this problem byproviding a

simu-lationplatformfor theevaluationofappliation odebut requiresthatthetarget platformis

realistially modelled.

Therearevarious approahestoemulatingomputerprogramsbut theseanbesplitinto

either hardware- or software-level emulation and the approah that is hosen is usually

de-pendenton the interfaesuponwhih theemulated program isbuilt. For example,when the

emulated appliation is programmedsothat itinterats diretlywiththehardware, perhaps

addressing hardware suh asmemory, then a hardware emulation tehnique suh as binary

translation an be used. This involves the interpretation and translation of mahine

in-strutionsfrom theiroriginal forminto instrutions thatthehost platforman interpretand

exeute. The hardware-level approah is also used when the aspet of the program under

evaluation is dependent on the hardware being used. For example, within wireless sensor

networks, evaluatingtheperformaneofasensormotean be dependent onthebehaviourof

the CPU or theradio.

Forprogramsthatdonotaddresshardwarebut perhapsareimplementedatahigher

logi-allayer,forexample,appliationsthatarebuiltuponsystemalls,thensoftwareemulationis

moretypiallyused. Inthisase,aompatibilitylayermustbedesribedthatprovidesa

map-pingfromtheoriginalsystemallstosystemallsonthehostplatform. Forexample,

transla-tionsoftwareimplementationssuhasWine(Win07 )andWABI(Sun Mirosystems 96)both

employthistehnique,e.g.,allowing Windowsprogramstorununmodiedonx86UNIXand

Solaris SPARC mahines respetively. The TOSSIM (Levis 03 ) and EMTOS (Girod04b )

emulators, bothemulatingTinyOS (Hill00)appliations,implementthetraditional

software-level emulation by interepting and redireting funtion alls when the program is being

(42)

systemthatprovidesallof therequired funtionalityoftheatualoperatingsystembut that

an interat with the emulator. Using this approah, there is no need to interept funtions

duringtheexeution ofthe program.

It should be notedthat some oftherelated worksexamined inthis hapter usetheterm

simulation when the term emulation would stritly be more aurate. TOSSIM, a TinyOS

emulator is an example of suh an instane, and these instanes are noted throughout the

thesiswhenthey our.

2.2.3 Alternative Software-Based Methodologies

There are several software-based approahes that ould be adopted here inaddressing these

hallenges but these areinadequate in meeting the broad hallenges identied in hapter 1.

Forexample,oneapproahwouldbetobuildasuper-simulator,anall-in-onesimulatorofthe

entire pervasive omputingarea. However, inorder to build this type of simulator, it would

requirealevelof domainexpertiseinallof thesub-domainsinorderto implement modelsof

alltherequiredelementsandwouldnotneessarilysupportareasofpervasiveomputingthat

areyet to emerge. Simulators thatlaim to be general pervasive omputing simulatorssuh

asUbiWise(JohnJ. Barton 03)ommonlyonlymodelasubsetoftheompletesetofaspets

ofthe domain. Anyapproah thatdoesnot expresslyinorporateexibilityandextensibility

into its initial design and arhiteture isnot well positioned to be adapted to the hallenges

ofmodelling overlapping andemerging new domainswithinpervasive omputing.

A seond alternative approah is the modiation and integration of existingsimulators.

For example, researhers in aademia and the motor industry have long speulated on the

impatthat theemerging area of vehiular ad-ho networks, might have on vehiular

appli-ationsinludingdriver safetyand vehile oordination (Salim 08 ). There aremany existing

vehileandnetworksimulatorssoanintegrationofoneofeahofthesesimulatortypesshould

be ableto leveragethe independent domain expertise ofboth thevehilesimulation

ommu-nity and the network simulation ommunity, and ould address the modelling needs of the

vehiular ad-ho networking (VANET) ommunity. Depending on the implementation and

(43)

of ommon properties suh as the simulated time and the loation of ommon objets. In

a simulated VANET senario (Mangharam 06), the ommon objets, the vehiles and their

network interfaes have thesame loation whih mustbe synhronised within therespetive

vehileand network simulators.

Clearly this approah an address the hallenge of modelling overlapping domains, but

until standardised API's are agreed upon for all simulators, this an only be ahieved on a

bespoke basis. At present there are no agreed interfaes, and few simulators thatan

inter-operate. The integration of two or more simulators does not address the hallenge of being

able to model new and emerging areas of pervasive omputing, and this approah is only as

extensible asits onstituent simulatorsare. Finally, an integrated approah neessitates the

testerto have the required expertiseineah ofthe simulators thatareto beintegrated.

Inonlusion,building asuper-simulator isnot a viableapproah asitisinherently

on-strained withrespet to futureextensibilitywhilst eorts to integrate independently reated

simulators arelimited by a lakof standardisation and bythe exibility and extensibility of

the onstituent simulators. So what are the alternatives? One approah is to inorporate

exibility,extensibilityandre-usabilityfromthe startandmakethesepriniplesorefeatures

of theadoptedapproah rather than add-onsat the end.

2.3 The Review Criteria

The simulators in this hapter are examined using eight review riteria, of whih abridged

working denitions are provided in table 2.1 and are subsequently expanded. These

rite-ria an be ategorised as follows. Review riteria one to three speially map to the ore

hallengesidentied inhapter1: namelyexibleheterogeneity,extensibilityand reusability.

Review riteria four through eight address spei features that are ommon to pervasive

omputing simulatorsthat merit examination andare inluded inorder to draw out spei

noteworthy features. These riteria an help to identify features of simulators that might

not neessarily addressthe identied ore hallenges but may inuene its arhiteture and

implementation. As will be seen in the following denitions, there is some small overlap in

(44)

# ReviewCriteria Denition

RC1 Flexible

Heterogeneity

Examines whether thesimulator supportstheexible

modelling ofmultiple instanes ofpervasive omputing's

ommon omponents.

RC2 Extensibility Examines whether thesimulator isextensible andan be

modied witheaseto supportdierent pervasive omputing

domains.

RC3 Arhiteture

(inl.

Reusability)

An overviewofthe arhiteture from asoftwareengineering

perspetive. Issuesovered heremayoverlap withother

evaluated riteria.

RC4 Appliation

emulation

Examines whether thesimulator supportstheemulation of

appliations thatexist withinthemodelled appliation

domain. RC5 Interation between emulated and simulated omponents

Examines whether theinteration between instantiations of

abstrations andemulatedappliations issupported.

RC6 Network

simulation

Examines thesupportthe simulator provides for network

simulationof bothwired andwireless networks.

RC7 Experimental

support

Examines thesupportthe simulator provides for evaluating

simulations. Thisrelates also to usability.

RC8 External

Interation

Examines whether thesimulator isself-ontained, or whether

itis possible to interat withexternal simulators, serviesor

proesses.

(45)

Flexible Heterogeneity

This riteria examines whether the simulator supports the modelling of heterogeneous

fea-tures of pervasiveomputing appliations. Theseheterogeneous featuresmayinlude a

vari-etyofsensorsandotherhardwaredevies,multipleappliationspossiblyrunningondierent

platforms,users,thephysial environment,and otherfeatures, both mobileandstati.

Addi-tionally,the simulatorsarereviewed to determine theirsupportfor theexibleinteration of

these heterogeneous features, afeature that isrequired ifa simulator isto support emergent

and ross-over pervasive omputerappliation domains.

Extensibility

This review riteria examines whether the simulator is extensible by design. Systems are

examined to determine what support exists for the modelling of new emerging appliation

areas of pervasive omputing not yet identied, where these appliation areas an omprise

of both new hardwareand softwareelements.

Arhiteture inluding Reusability

Thethirdreviewexaminesthesystem'sarhiteturefromasoftwareengineeringperspetive.

Issuesoveredhere mayoverlap withotherevaluated riteria. Thereusabilityaspets of the

arhiteture arealso examined in ases where thearhiteture an be used to model several

pervasive omputingdomains.

Appliation Emulation

Thisreviewriteria examineswhatsupportthesystemprovidesfortheemulation of

applia-tions thatwould typiallyexist withinthe modelled appliation domain.

Interation between Simulated and Emulated Components

Thisreviewriteriaexamineswhatsupportisprovided formodellinginterationbetween any

simulated elementsandanyemulated appliationswithintheone appliationdomain. There

(46)

Network Simulation

Thisreviewriteria examineswhatsupportthe systemprovidesfor thenetworksimulationof

anynetworked elementswithinthemodelleddomain. Thismayinludebothwiredand

wire-lessnetworks. Thisreview riteria also examines what interfaes areavailable to algorithms

andanyemulated appliationswithin thesimulator.

Experimental support

Thisreviewriteriaexamineswhatsupportthesystemprovidesfortherunning,exeutionand

subsequentevaluationofanymodelledexperiments. Thisouldinludefor examplesripting

support,debugging features, or the abilityto program user behaviour.

External Interation

Thisreviewriteriaexaminesthesystemsabilitytointeratwithexternalsimulators, servies

orproesses. OftensystemsoeranAPItoonnetexternalserviesfordebuggingorsystem

extension. AnyAPI'sthatareprovided to support thisfuntionality areexamined.

Omitted Criteria

Thesimulators underreview ouldalso have evaluated againstriteria suhasperformane,

easeofuse,andauray,however,thenumberofsimulatorsattemptingtoaddressthegeneri

modellingofpervasiveomputingissofew, thattobenhmarktheseagainsteahotherwould

notbeto omparelikewithlike. Theperformaneorease ofuse ofStarBED, adistributed

platformforsimulationannotbeomparedinameaningfulwayagainstUbiWise,asimulator

basedupon a PC-basedgames engine.

Theeldofpervasiveomputingsimulatorsisnotasstronglydevelopedasthatofnetwork

simulation for example, where there are maybe twenty aepted simulators that an all be

runonaloalmahine. Inmaturesimulation eldssuhasthese,theresearhquestionshave

moved on from whether or not itis possible to model some aspet of network behaviour to

(47)

Finally,theaurayanddelityofanysimulatorisimportant,andthisisapturedwithin

RC1 for simulated hardware devies, and inRC4 for emulated appliations, and is a theme

whih isarriedthroughout theframeworks implementation and itsevaluation.

2.4 An Overview of Pervasive Computing Simulators

Setion 1.1 in hapter 1 introdued themany overlapping areas of pervasive omputing, all

of whihinvolve dierent tehnologies and havedierent modellingrequirementsyetontain

some ommonalities. For example, smart spaes (Tapia04) are often reative appliations,

driven bysensors suh asdoor ontatswithes embedded within thespae, whih an then

adaptsomeaspetofthespaetoauser'srequirements. Anenvironment-monitoring wireless

sensornetwork (Ji 04)ouldomprise of sensors,perhaps monitoring preipitation,attahed

to wireless nodes that propagate that sensor information to other nodes through an ad-ho

network. ITSsystems(Klein01)typiallyuseembeddedsensorssuhasindutiveloopsinthe

roadnetwork to improve the throughput of thevehilesinthe system. Allof these domains,

and others onsidered, formthebroad overlapping spaethatis pervasiveomputing.

This review of related work intersets several disiplines, as shown in gure 2.1 , some

of whih are subsets of pervasive omputing and some that are independent researh areas

that have some ommonalities with pervasiveomputing. There aremany interpretations of

pervasive omputing, and oasionally simulators that are targeted at pervasive omputing

atuallyonlyaddressasubsetofthearea. Comparedwithanareaofresearhsuhaswireless

sensor networking, there are surprisingly few well-known simulators targeted speially at

theoverallpervasiveomputingarea. Thisismainlyduetoitsbroad denition,andthatfew

researh groupsaddresspervasiveomputing asa ompletespae, insteadpreferringto fous

on speipervasive appliation areas,suh asontext-aware or loation-aware omputing.

Simulatorsfor pervasiveomputingan bebroadlyategorised into twogroups. Therst

groupinludes simulators suh asUbiWise(John J.Barton 03 ), Tatus (O'Neill 05), and the

Lanastersimulator(Morla 04),whiharepervasiveomputingsimulatorsnotaddressingany

partiular sub-domain. UbiWise and Tatus support reating omplex 3-D simulations of a

(48)

(49)

suhaswirelesssensorsthatmayexistwithinthoseenvironments. Theseondgroupof

simula-torsarethosethataddressspeialisedsubsetsofthepervasiveomputingeld. Forexample,

Siafu (Martin 06a ) is a simulator for ontext-aware omputing appliations. Additionally,

there aremanyinstanes of simulators suh asnetwork simulators (Sundresh 04 ; Sobeih 05)

and agent simulators (North 06; Narendra 05 ) being modied and usedto model subsets of

pervasive omputing.

Infat,ofthemanysimulatorsidentiedingure2.1,onlythoseinsidetheredovallaim

to speially address pervasive omputing, in the broad multi-disiplinary sense that this

thesisonsiders. Thesesevenareperhapsthemostimportantofthesimulatorsonsidered, as

they attemptto addressthe hallengesthat have been identiedfor a exibleand extensible

simulator for pervasive omputing. An additional tensimulators, those insidethe blue oval,

and outsidetheredoval,eitheraddresssubsetsofpervasive omputing, orareintheoverlap

of appliation domainssuh asintelligent transportation systemswithpervasive omputing.

Todate,therehasonlybeenonesimulator developedusing aframeworkforthepervasive

omputing domain. Shibuya applied a framework approah (Shibuya 04 ) to modelling the

spatial aspetsofhumanbehaviourina pervasiveomputing environment. Thefousof this

work was on human interation within these environments, rather than the environments

themselves, however,andthephysial aspetsofthedevies,and appliationsthatmightuse

those devieswithin a pervasive omputing environment were negleted.

2.5 The Key Simulators

..This thesis is onerned with methodologies that are both suitable for the evaluationof a

wide rangeof diversepervasiveomputingsenarios andthatan be extendedtoevaluate new

senarios thatmayemergeinthefuture.... Asseenintheprevioussetion, there have been

numerousworks addressing a subset of the requirements of pervasive omputing simulation,

but few taking a more omplete approah to modelling the domain as a whole. Of these

simulators, sevenarehosen thatoerthemostinteresting andnote-worthyapproahes, and

that are losest in spirit and exeution to a framework-based approah to building exible