Using Existing Network Simulators for Power-Aware Self-Organizing Wireless Sensor Network Protocols

HAL Id: inria-00113679

https://hal.inria.fr/inria-00113679v3

Submitted on 21 Nov 2006

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of

sci-entific research documents, whether they are

pub-lished or not. The documents may come from

teaching and research institutions in France or

abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est

destinée au dépôt et à la diffusion de documents

scientifiques de niveau recherche, publiés ou non,

émanant des établissements d’enseignement et de

recherche français ou étrangers, des laboratoires

publics ou privés.

Using Existing Network Simulators for Power-Aware

Self-Organizing Wireless Sensor Network Protocols

Thomas Watteyne

To cite this version:

Thomas Watteyne. Using Existing Network Simulators for Power-Aware Self-Organizing Wireless

Sensor Network Protocols. [Research Report] RR-6020, INRIA. 2006, pp.20. �inria-00113679v3�

inria-00113679, version 3 - 21 Nov 2006

a p p o r t

d e r e c h e r c h e

Thème COM

Using Existing Network Simulators for Power-Aware

Self-Organizing Wireless Sensor Network Protocols

Thomas Watteyne

N° 6020

Thomas Watteyne

∗†

ThèmeCOMSystèmes ommuni ants

ProjetARES

Rapportdere her he n°6020September200617pages

Abstra t: In this do ument, we ompare three existing simulation platforms (OPNET

Modeler,NetworkSimulator2,GeorgiaTe hSensorNetworkSimulator). Our omparative

study fo uses on ease of use, s alability, ease of implementing power onsumption model

and physi al layer modeling a ura y, mainly. Con lusions of this study are presented,

and will help us de ide whi h simulating environment to use for evaluating power-aware

self-organizingsensornetworksproto ols.

Key-words: WirelessSensorNetworks,Simulators.

∗

CITILaboratory,INSALyon,Fran e. thomas.watteyneinsa-lyon.fr

†

réseaux de apteurs sans ls, e a es en énergie et

auto-organisés

Résumé : Dans e do ument, nous omparons troisplateformes de simulationexistantes

(OPNETModeler,Network Simulator 2,GeorgiaTe h SensorNetworkSimulator). Notre

étude omparativeporte essentiellement sur la fa ilitéd'utilisation, le passageà l'é helle,

la fa ilitéd'implémenter unmodèlede onsommation énergétiqueet la nessedu modèle

de propagation radio utilisé. Les on lusions de ette étude nous aident à hoisir une

plateformedesimulationpourévaluerdesproto olesd'auto-organisatione a esenénergie

pourréseauxde apteurs.

1 Introdu tion and onstraints

We are interested in evaluating power-aware self-organizing proto ols for Wireless Sensor

Networks(WSNs). Althoughanalyti methodsareusefulin manysituations, the

omplex-ity of modern networks ombined with the inability to apply simplifying assumptions in

manyanalysisproblems(e.g.,itiswellknownthatMarkoviantra assumptionsareoften

inappropriateand anleadto misleading results)limit theappli abilityofpurely analyti

approa hes[1℄.

Nevertheless, ashas been shown in [2, 3, 4℄, simulatorsdo not alwaysree t the

per-forman es of the pro esses they model. Results are mainly orrupted by ina ura ies in

the physi al model (interferen es, error models, power onsumption, ...). This does not

meansimulationsshouldnotbeused,theyonlystressoutthat simulationresultsonlygive

a rough idea of what is really going on. What's more, as simulations are performed on

dis rete topologies with dis rete s enarios, simulation an never be used in a validation

approa h.

Severalsimulationplatformsexist. Theaim of thispaperis to omparethree of them

(OPNETModeler[5℄, Network Simulator 2[6℄and GTSNetS[7℄), in orderto pi koutthe

onethat best fulllsourneeds.

WSNarevery onstrained,soareWSNsimulators. AsWSN ommuni atewirelessly,it

isessentialthatthesimulatorembedsana uratepropagationmodel. What'smore,

de-pendingonthenalappli ation,WSNs anbe omposedoftensofthousandsnodes,whi h

means simulator s alability is important too. As WSNs are very appli ation spe i ,

nostandardsolution anbeused, oratleastit should bepossibleto tunethose solutions.

Extensibility of the model is therefore essential. We are interesting in the power

on-sumptionofself-organizingproto ols,soaproperpower onsumption modelshould be

implementedin the simulator. Finally, thesimulatorshould be easy to use and to learn,

andits ostshould bekeptaslowaspossible.

2 OPNET Modeler

OPNET Modeler [5℄ is a simulation environment developed and maintained by OPNET

Te hnologies,In . (NASDAQ:OPNT).Itsaimistoenableuserstoevaluatehownetworking

equipment, ommuni ationste hnologies, systems,and proto olsperformunder simulated

network onditions. ItisthereforemainlyusedbyServi eproviders(AT&T,BT,Deuts he

Telekom, Fran eTele om, NTTDoCoMo...), Entreprises(DaimlerChrysler,IBM Global

Servi es,Mi rosoft,Ora le,S hneiderEle tri ...),Network-EquipmentManufa turers

(Al- atel,Cis o Systems,IntelCorporation,Motorola,Nokia ...) and Government&Defense

(DARPA,FederalAviation Administration,FederalBureauofInvestigation,NASA,NATO

...). Itisa ommer ialprodu t,andali enseisfarfrom free. Useofaedu ationalli ense

in ooperationwithanindustrial partnerisprohibited.

TheOPNETModelersimulationenvironment anbedividedinthreeparts: themodeling

Figure1: Modelingin OPNETisdoneusinglayers.

2.1 The modeling environment

Modeling isgreatlysimplied bybuildingthemodelhierar hi ally. As ommuni ationhas

been utinto several OSI layers[8℄, amodel built using OPNET Modeler is omposed of

severalinter hangeablelayers. Itisforexampleveryeasytomodifythepropagationmodel

beingusedbymodifyingthe losuremodel parameterusedin theemittinglayer.

Thehighestlayeristhes enariomodel. As enariois omposedofanodetopology,and

globalsimulationattributes.

Thenode level model isdes ribedbyagraphi alrepresentationofitspro esses. Those

pro esses ommuni ate using interrupts. Node level models also have attributes, su h as

geographi nodeposition. Aninterfa eispresenttoletthenodelevelmodel ommuni ate

withthe highermodel (the s enario). This way, anode model attribute anbeinitialized

bythes enariowhenenteringrun-time,andthisdependingonwhi h s enarioisused.

Asinthenodemodel,aninterfa eandattributesarepresentinthepro esslevel model.

Thepro ess is des ribed graphi allyby astatesinter onne ted by onditional transitions.

Time anonlyowin states. Atransitionisred onlywhenitsgiven onditionsare

satis-ed. What'smore,itis possibleto perform omputationonagiven variablewhilering a

transition,whenenteringagivenstate,orwhenleavingit. Threemajor odeblo ksas

de-ned: SV (StateVariable)whi hdes ribedthepro essvariables,HB(HeaderBlo k)whi h

des ribesthe onditionstorethetransitions,andFB(Fun tionBlo k)whi h ontainsall

thefun tions' bodies. FB is written in C programminglanguage,with anadded OPNET

Modelerfun tionlibrary. ChoosingCasaprogramminglanguagegivesa esstoall lassi al

Cfun tions (formanipulating les, generating pseudo-randomnumbers ...). Thelayered

modelstru tureisdepi tedin Figure1.

2.2 The simulation engine

OPNET Modeleris adis rete event simulator. The omplete system'sbehavior is

Duringexe ution,alistofup omingeventsisthusatta hed toea hentity. Whenthis

ex-e ution timeis rea hed, thepro essing atta hedto theeventsis performed, andthe event

isremovedfrom thelist. A newevent anbes heduled(i.e. atta hedto thelist) dire tly

byusingOPNETfun tions,orindire tlyforexamplewhenapro essre eivesaninterrupt.

Evenifthenotionof dis reteeventsimulationis internal tothe simulator,it isimportant

fortheuserassheorhe an dire tlyintera t usingtheappropriatefun tionsonthelistof

events(adding,removing,lteringagivenlistofevents).

2.3 The result olle tor

TheOPNETModelerenvironmentalsoin ludesaresult olle tor. This an al ulateoutput

statisti s,andplotgraphswhilethesimulationisrunning. Using spe i fun tions,probes

arepla ed atthe appropriatelo ationsin the odesoasto measure agiven value. These

probes feed the result olle tor using ve torial or s alar result les. Afterward, OPNET

Modeler anplotthoseles. Despitethisintegratedtool,wehavenotuseditforitsla kof

exibility. Indeed,youarebound tothefun tions OPNETModelerproposes,and itisfor

exampleveryunnaturaltoperform omplexpost-simulation al ulations. Wehavetherefore

de idedto olle toutputdatausingstandardCfun tions,andwritingdatainoutputASCII

les. Post-pro essingandplotting anthenbeeasilyperformedbysoftwaresu hasgnuplot

[9℄.

2.4 The physi al layer model

Aswehaveseenwith[2℄,thereisagreatimpa tofthephysi al modelonoverallresultsin

WSNs. Soagoodmodeloftheradiointerfa eisessential. Radiointerfa eisrepresentedin

OPNETModelerbyapipeline: thephysi al models(antennagain,delays...) areapplied

sequentiallyonthe11stagelongpipeline. Ea hofthesestagesis ongurable,andthisfor

emittingandre eivingradio entities.

The ongurablepipeline-basedradiomodel(seeFigure2)makesitpossibletosuppress

uninteresting phenomena (or phenomena whi h are not takeninto a ount yet in agiven

ommuni ationproto ol). For simple Unit Disk Graph transmission model, it is possible

to turno power ontrol (tagain,ragain, ragain model, power model),errors(error model,

e model),andnoise(bkgnoisemodel,snrmodel,bermodel). The losuremodel parameter

denes the a tual propagation model, with mil_ losure_dist being the unit disk graph

model.

2.5 Con lusions on OPNET Modeler

ThankstoourCITI'sedu ationalli ense,wehaveworkedwithOPNETModeler. Evenifit

hasbeenprimarilydesigned fornetwork planning,it anbeusedeasilyforwireless ad-ho

proto ols providing its wirelessextension is installed. Themain advantageof this

omprehensible graphi al modeling interfa e. Even with a mixture between pure

graphi- al(s enarioand nodemodel)and textual ode-basedmodeling(pro essmodel),modeling

staysstraightforward. Nevertheless,there are ountless parametersandsometimes theline

between model interfa es and global/lo al parameters gets fuzzy, yielding to moments of

great onfusion.

Asfors alabilitya ordingto [1℄OPNETModelerisableto modelnetworksofseveral

hundredstoseveralthousandnodes. Of ourses alabilitydependsonnumerousparameters

su h asthemodel renement,availablememory,pro essingpower,availabletime et . We

willmorespe i allyaddresss alabilityinSe tion 4.

Despite the fa t the propagation model seems orre t, the la k of s alability and the

pri eofaexploitationli ensemakeusreje tOPNETModelerasasimulationenvironment.

3 Network Simulator 2

Ns[6℄isadis reteeventsimulatortargetedatnetworkingresear h. Nsprovidessubstantial

support for simulation of TCP, routing, and multi ast proto ols over wired and wireless

(lo alandsatellite)networks.

NsbeganasavariantoftheREALnetworksimulatorin1989andhasevolved

substan-tiallyoverthepastfewyears. In1995nsdevelopmentwassupportedbyDARPAthroughthe

VINTproje tatLBL,XeroxPARC,UCB,andUSC/ISI.Currentlynsdevelopmentis

sup-portthroughDARPAwithSAMANandthroughNSFwithCONSER,bothin ollaboration

with other resear hersin luding ACIRI.Ns hasalwaysin luded substantial ontributions

fromotherresear hers,in ludingwireless odefromtheUCBDaedelusandCMUMonar h

proje tsandSunMi rosystems. Fordo umentation,see[10℄.

Ns-2 is free and open-sour e software. It is extensively used in edu ational/resear h

institutions. Contributorsareverynumerous,soin ludedpa kages omefromverydierent

people. Themainadvantageisthatns-2in ludes utting-edgemodeledproto ols. Themain

drawba k is the omplexityof resulting groupof pa kages and the existen eof untra ked

bugs. Thesamehappensfor do umentation: despiteaneortfor entralizinginformation

(forexampleinthenewns-2manual[10℄),information/do umentationisstilldisseminated

on numerous Web sites. It is generally advies to seek information on the ns-2 dedi ated

forums,ratherthanwritten downmanuals.

Ns-2hasnographi alinterfa e,norhasitalayermodelingapproa hasOPNETModeler

forexample. Thisla kofeaseofuse ans areawaypotentialusers. ns-2usestwo

omple-mentaryobje t-orientedprogramminglanguages,aideaknown assplit-language

program-ming: C++a ompiledlanguagesis usedforlongand omplexrun-timephases,while

3.1 Modeling the radio interfa e

3.1.1 Radio propagationmodels

Ns wasrst dedi ated to simulating wirednetworks. Wirelesssupport wasmade possible

byportingtheCMU'sMonar hgroup'smobilityextensiontons. Anodewillonlyre eivea

signalifitisre eivedabovea ertainpowerthreshold. Re eptionpoweris al ulatedthanks

to apropagationmodel. Three standardpropagationmodelsare provided withns-2. The

rstoneistheFrissfreespa e model(Eq. 1), whi his notreally realisti ineveryday-use.

The two-way ground ee t model onsiders the radio wave travels using two paths: the

dire t line of sight path and a path whi h boun es o the ground (Equation 2). A more

realisti modelwouldin ludetherandomfadingee t,duetomultipathpropagationee ts.

Wewillnotexpli itthoseequationshere.

P

r

(d) =

P

t

G

t

G

r

λ

2

(4π)

2

d

2

L

P

r

(d) =

P

t

G

t

G

r

h

2

t

h

2

r

d

4

L

Basi energy models are verysimple in ns-2. The node has a bu ket of energy. When a

message is sent/re eived, some energy is taken out of the bu ket. When it empties, no

more messages an be sent nor re eived. This energy model is of ourse mu h to s ar e

for Wireless Sensor Network appli ations for example, as it only takes into a ount the

pa ket-basedmodel,andnotthetime-basedmodel,northephysi almodel(seeSe tion 1).

3.2 Add-ons for ns-2

Be auseof theopen-sour enature ofns-2 ontributionsareadded daily. Weheregivethe

exampleofvisualizationtools.

The"o ial"visualization toolforns-2is alledNam [11℄. Whereasis hasbeen

exten-sivelyusedforresear honwirednetworks,ithasnotbeenextendedforwirelessvisualization.

Kurkowskiet al. hasre entlyproposed avisualizationtool alled iNSpe t [12℄ (seeFigure

3). Theauthorsarguethatthisvisualizationtool anbeusedfor(1)validatingthea ura y

ofamobilitymodel'soutputand/orthenodetopologylesusedtodrivethesimulation;(2)

validate thenew versions of theNS-2 simulatoritself; and (3)analyze theresultsof NS-2

simulations.

3.3 Con lusions on ns-2

As for s alability, a ording to [13℄, on single pro essor ma hine, ns-2 an omfortably

simulatenetworksofabout1000nodeswithpopularroutingproto ols[13℄. Wewilldes ribe

Ns-2lookslikeaverygood andidateforevaluatingpower-awareself-organizingwireless

sensornetworksproto ols. Itis freeandopen-sour e,whi hmeansthe ommunityof users

is ontinuously updating the simulator with the latest proto ols. Nevertheless, it is hard

to use. What's more, for Wireless Sensor Network evaluation, ns-2 la ks s alability and

satisfa toryenergymodels.

4 Georgia Te h Sensor Network Simulator

The Georgia Te h Network Simulator (GTNetS) is urrently developped by Pr. George

Rileyandhis students. It isfreelyavailable online[7℄andwasrst presentedin2003 [14℄.

Beforedeveloppingthisnewsimulationenvironment,Pr. Rileywasworkingons alability

issuesinsimulatingnetworks.S alability anbeamainissuewhentryingtosimulatewired

peer-to-peersystemsorwirelesssensornetworks omposedoftensof thousandnodes.

S alability is limitedmainly byavailable memoryand omputationtime. Another

im-portantfa toris howne-grainedsimulationmodelsarebuilt. Assinglepro essorPC-like

omputerareverylimited,itwasinterestingtostudythefeasibilityofparallelsimulation. A

rstattemptwastheParallel/DistributedNetworkSimulatorinwhi haruntime

infrastru -turewasdeveloped in order to make severalns-2 simulators runningon separatema hine

worktogetheronthesamesimulationtask (seeFig. 4). Buttheresultingar hite turewas

not"natural",asns-2is notintendedto beused in adistributed fashion, soGTNetS was

developed. Itsmaingoalsare:

ˆ distributed simulationand s alability

ˆ easeofuseandextensibility

ˆ support forpopularproto ols

4.1 distributed simulation and s alability

InGTNetS,asimulationis utintoapossiblelargenumberofsmallersimulations,ea hone

simulatingasubsetofthenodes. Fornodesofonesubsetto beableto ommuni atewith

nodeof anothersubset,relatedlinks needtobehandled bytwosub-simulators. Theseare

alledremote-link,theatta hednodeare alledremote-nodes. Routingofteninvolvehaving

routingtablesatea hnode,butthisleadstomemoryspa egrowingquadrati allywiththe

numberofnodes. Therefore,routingtablesarebuiltinside thesimulatoronanon-demand

basis,usinganoptimization alledNix-routing. Finally,nottoover-utilizediskspa e,logged

events an beltered,inorderfortheresultinglog-le tobeassmall aspossible.

S alability of theGTNetS platformhas been demonstrated in [15℄. Here GTNetS was

used to evaluate performan eof the AODV wireless network routingproto ol. Studywas

performedonaupto50,000nodenetwork.Apartfromthispaper,simulationwererunusing

Figure4: Buildingaar hite turetolinkns-2simulatorstogetherisnotaneasytask[drawn

Figure5: Apa ketisrepresentede ientlyinGTNetS [drawnfrom [13℄℄.

ona17ma hine-wideLinux-based luster platform(136 CPUs),and ona750server-wide

super omputeratthePittsburghSuper omputingCenter(3000CPUs).

Althoughitisnotthisdo ument'sprimarys ope,wewillqui klydes ribehowGTNetS

a hieves good s alability [13℄. Simulator s alability is limited mainly by the amount of

memoryavailable(in ludingvirtualmemory). Other limitationsin lude limiteddiskspa e

forlogle,andsimulationrun-timedurationlimitations. GTNetShasbeendesignedwith3

goalsin mind: (1) redu ingtheeventlistsize,(2)oeringoptimizedmemorymanagement

and(3)redu ingthesizeofthelogle. Byredu ingthenumberofevents,simulationtime

isredu ed. Inorderto do so,GTNetS usesFIFOre eivequeues, abstra t pa ket queuing

te hniques, and timers bu kets. The overall memory footprint is redu ed by using

NIx-ve torroutingandoptimizingthepa ketrepresentation. Finally,theloglesizeisredu ed

byallowingtheusertodeneexa tlywhathastobelogged. What'smore,built-instatisti s

olle tionavoidshavingtobuildstatisti safterrun-timeonahugelogle.

4.2 ease of use and extensibility

4.2.1 using GTNetS

GTNetS is written entirelyin obje t-orientedC++. Theperson in harge of runningthe

simulationstartswritingthemainC++program. Afterin ludingtheappropriatepa kages

(espe ially thesimulator.h header le), she orhe buildsthe topologyby instantiating the

appropriatenode,interfa eandlink obje ts. Additionalproto ols,randomowgenerators

andstart/stoptimesandrenedandthesimulationisstarted. Aftersu essfully ompiling

the main program  using any C++ ompliant ompiler , it is linked with the GTNetS

obje tlibraries. Theresultingexe utablebinaryissimplyexe utedasanyotherappli ation.

The omplete GTNetSmanual anbefoundonline[16℄.

4.2.2 Implementingnode mobilityin GTNetS

Newappli ation emergeforwirelessad-ho andsensornetworks. Inapervasive omputing

s enario,PDAs, mobilephonesorevensmartwat hespeoplewear ould ommuni ateand

ex hange information (su h as propagating interesting lo ation-spe i information). In

order to gather onsistent performan e results, simulation an beused. [2℄ hasshown us

theimportan e ofthe physi al layermodel. Here wefa e another hallenge: themobility

Figure6: MobiREAL's ute interfa eanimatesthemodelof shoppingstreetsindowntown

Osaka[drawnfrom[18℄℄.

openspa esu hasanairporthall. This isneverthelessnotrealisti in mosts enarios,and

amoregeneralmobilitymodel isneeded.

Konishietal. haveproposedapedestrianmobilitymodel[18℄. Thismodelisbestsuited

for urban pedestrian shopping street. Modeling starts with inputting information on the

simulation eld. Building, walls, and roads are drawn, and hot spots are dened (very

attra tiveshops for example). Rules will then model the movement of pedestrian.

Anti- ollisionalgorithmsmodelthewayhumans walkin rowded streetswithoutbumping into

ea h other. An interestingfa t is that the appli ation ae ts the pedestrian's movement:

whenauserre eivesinformationaboutaspe iallyattra tivehotspot,itwillwalkthatway.

Arulealsomakesthesimulatedpedestriansprefernon- rowdedstreetsiftheyknowwhere

theyaregoing.

This mobility model is inje ted in theMobiREAL simulationenvironment,whi h also

oers a very ni e looking interfa e (see Figure 6). The underlying simulation engine is

GTNetS,whi hprovidesthephysi alandMAClayermodels.

4.2.3 Modeling wirelesssensornetworks with GTSNetS

As wireless sensor networks an potentially omprise several tens of thousands of nodes,

s alabilityinthesimulationenvironmentisanimportantissues. What'smore,asWSNsare

Figure7: InGTSNetS,asensoris omposedofsensing, ommuni ation, omputationunits,

andabattery[drawnfrom [19℄℄.

to anexisting simulator. GTNetS oersthe desireds alability andextensibility, soit has

beenextendedtosupportWSNs. TheGeorgiaTe hSensorNetworkSimulator(GTSNetS)

[19℄oersvarious GTNetS C++wireless sensormodels. Apartfrom s alabilitywhi h is

inheritedfromGTNetS GTSNetSfo uses onenergy onsumption.

Figure7illustratesthewirelesssensormodelusedbyOuld-Ahmed-Valletal. Logi ally,

ea h oneof those omponentwill haveatunableC++ obje tasso iated,whi h allin lude

several energy models. Details of those models an be found in [19℄, and the GTSNetS

extensiontoGTNetS anbefoundonlinein [7℄.

5 Con luding Remarks

Table 1 summarizes this do ument. We are interested in simulating power-aware

self-organizingproto ols for Wireless SensorNetworks. As WSN anbe omposed of alarge

numberof nodes, s alabilityis amajor fa tor in hoosing asimulation platform. What's

more,astheresultsweareinterestedingatheringarerelatedtopower onsumption,having

Simulator Cost Ease of use Extensibility

OPNETModeler expensive graphi al hard

ns-2 free atmodel possible,open-sour e

GTSNetS free obje t-oriented easy,open-sour e

Simulator S alability physi al layer power onsump.

OPNETModeler 1,000nodes 11stage nemodel none

ns-2 1,000nodes 3models none

GTSNetS 100,000+nodes omplete802.11 WirelessSensors

Table1: Comparingthedierentsimulationenvironments.

This leadsus tothe on lusionthat, foroutneeds,thebest suitedsimulationplatform

is GTNetS. Future work in lude mastering this platform. It would be interesting to see

whatenergy onsumptionmodelshavebeenimplemented,andtoseeiftheyare losetothe

modelswehavedened. What'smore,asfors alability,itwouldbeinterestingto seehow

bigaWSNwe ansimulateonasinglepro essorpersonal omputer,andwhattheoverhead

wouldbetouseaLinux-based luster.

Referen es

[1℄ R. M. Fujimoto, K. S. Perumalla, A. Park, H. Wu, M. H. Ammar, and G. F. Riley,

Large-s alenetworksimulation: Howbig? howfast? in11thIEEEInternational

Sym-posium on Modeling, Analysis, and Simulation of Computer and Tele ommuni ations

Systems (MASCOTS). Orlando,FL, USA:IEEE,O tober2003,pp.116124.

[2℄ D. Cavin,Y.Sasson,andA.S hiper,Onthea ura yofmanetsimulators, in

Work-shop on Prin iples Mobile Computing (POMC). Toulouse, Fran e: ACM, O tober

2002, pp.3843.

[3℄ F. Haq and T. Kunz, Simulation vs. emulation: Evaluating mobile ad ho network

routingproto ols,inInternationalWorkshoponWirelessAd-Ho Networks(IWWAN),

London, UnitedKingdom,May2005.

[4℄ V. Shnayder, M. Hempstead, B. Chen, G. W. Allen, and M. Welsh, Simulatingthe

power onsumptionoflarges alesensor networkappli ations, in Conferen e on

Em-bedded Networked Systems (SenSy ). Baltimore, Maryland, USA: ACM,November

2004.

[5℄ Opnet te hnologies, in . last visited on 16/09/2006. [Online℄. Available:

http://www.opnet. om/

[7℄ Gtnets home, last visited on 18/09/2006. [Online℄. Available:

http://www.e e.gate h.edu/resear h/labs/MANIACS/GTNetS/

[8℄ A.S.Tanenbaum,ComputerNetworks,FourthEdition,A.S.Tanenbaum,Ed. Prenti e

Hall, August2002.

[9℄ gnuplot homepage, last visited on 18/09/2006. [Online℄. Available:

http://www.gnuplot.info/

[10℄ K. Fall and K. Varadhan, The ns manual, January 2006. [Online℄. Available:

http://www.isi.edu/nsnam/ns/do /ns_do .pdf

[11℄ Nam: Network animator, last visited on 18/09/2006. [Online℄. Available:

http://www.isi.edu/nsnam/nam/

[12℄ S. Kurkowski,T.Camp, N.Mushell,andM.Colagrosso,Avisualizationandanalysis

tool forns-2wireless simulations: inspe t, in International Symposiumon Modeling,

Analysis, andSimulation of Computer andTele ommuni ation Systems (MASCOTS).

Atlanta,GA,USA:IEEE,September2005,pp.503506.

[13℄ G.Riley,Large-s alenetworksimulationswithgtnets, inWinter Simulation

Confer-en e,vol.1. NewOrleans,Louisiana,USA:IEEE,De ember7-102003,pp.676684.

[14℄ G.F.Riley,Thegeorgiate hnetworksimulator, inworkshop onModels,methodsand

tools forreprodu ible networkresear h. Karlsruhe,Germany: ACMSIGCOMM,2003,

pp.512.

[15℄ X.ZhangandG.F.Riley,Performan eofroutinginverylarge-s alemobilewirelessad

ho networks, in International Symposiumon Modeling, Analysis, and Simulation of

Computer and Tele ommuni ation Systems (MASCOTS). Atlanta, GA,USA:IEEE,

September2005, pp.115122.

[16℄ G. F. Riley, Using the Georgia Te h Network Simulator,

Geor-gia Institute of Te hnology, Atlanta, GA, USA. [Online℄. Available:

www.e e.gate h.edu/resear h/labs/MANIACS/GTNetS/do /gtnets.pdf

[17℄ B.KarpandH.Kung,Gpsr: Greedyperimeterstatelessroutingforwirelessnetworks,

in AnnualInternationalConferen eonMobile ComputingandNetworking(Mobi om).

ACM,August2000,pp.243254.

[18℄ K.Konishi,K.Maeda,K.Sato,A.Yamasaki,H.Yamagu hi,K.Yasumoto,andT.

Hi-gashino, Mobireal simulator - evaluating manet appli ations in real environments,

in International Symposium on Modeling, Analysis, andSimulation of Computer and

Tele ommuni ationSystems(MASCOTS). Atlanta,GA,USA:IEEE,September2005,

[19℄ E. Ould-Ahmed-Vall,G.F.Riley,B.S.He k,andD.Reddy,Simulationoflarge-s ale

sensornetworksusinggtsnets,inInternationalSymposiumonModeling,Analysis,and

Simulation ofComputerandTele ommuni ationSystems(MASCOTS). Atlanta,GA,

Unité de recherche INRIA Rhône-Alpes

655, avenue de l’Europe - 38334 Montbonnot Saint-Ismier (France)

Unité de recherche INRIA Futurs : Parc Club Orsay Université - ZAC des Vignes

4, rue Jacques Monod - 91893 ORSAY Cedex (France)

Unité de recherche INRIA Lorraine : LORIA, Technopôle de Nancy-Brabois - Campus scientifique

615, rue du Jardin Botanique - BP 101 - 54602 Villers-lès-Nancy Cedex (France)

Unité de recherche INRIA Rennes : IRISA, Campus universitaire de Beaulieu - 35042 Rennes Cedex (France)

Unité de recherche INRIA Rocquencourt : Domaine de Voluceau - Rocquencourt - BP 105 - 78153 Le Chesnay Cedex (France)

Unité de recherche INRIA Sophia Antipolis : 2004, route des Lucioles - BP 93 - 06902 Sophia Antipolis Cedex (France)

Éditeur

INRIA - Domaine de Voluceau - Rocquencourt, BP 105 - 78153 Le Chesnay Cedex (France)

http://www.inria.fr