White PapersSensor networks (sensor-nets) have emerged as one of the dominant technology trends of this decade (2000-2010). Sensors and actuators, wireless communications, and embedded computing are not new concepts but it is the recent low-cost large-scale integration of computation, communication, and sensing into “wireless sensor networks” that has captured the attention of many researchers. Sensor-nets enable observing the physical world at a granularity level which was unperceived before. Applications of sensor-nets encompass a variety of disciplines and domains, limited only by the imagination of the application developer. When thinking about future research directions for (wireless) networks it is important to consider which kind of devices would be connected to the network. In the coming years the most common type of devices on the network would be embedded processors, such as sensors and actuators (i.e. sensor-nets), and improving battery life will be more important than improving performance. Energy-consumption by computing chips is falling sharply per unit computation (Moore’s Law) whereas energy consumed by radios is determined by laws of physics. Thus, the wireless interface will be the primary consumer of energy in any device that combines computation and radios.
Medium Access Control Issues in Sensor NetworksMuneeb AliLUMS, CS DepartmentOpposite Sector U, DHALahore - 54792, Pakistanmuneeb@sics.seUmar SaifMIT, CSAILThe Stata Center, 32G-780,Cambridge, MA 02139umar@mit.eduAdam DunkelsSICSBox 1263SE-164 29 Kista, Swedenadam@sics.seThiemo VoigtSICSBox 1263SE-164 29 Kista, Swedenthiemo@sics.seKay R omerETH ZurichDepartment of CSCH-8092 Z urich, Switzerlandroemer@inf.ethz.chKoen LangendoenTU DelftMekelweg 4, 2628 CD DelftThe Netherlandsk.g.langendoen@ewi.tudelft.nlJoseph PolastreMoteiv Corporation55 Hawthorne St, Suite 550San Francisco, CA 94105joe@moteiv.comZartash Afzal UzmiLUMS, CS DepartmentOpposite Sector U, DHALahore - 54792, Pakistanzartash@lums.edu.pkABSTRACTMedium access control for wireless sensor networks has beena very active research area for the past couple of years.The sensor networks literature presents an alphabet soupof medium access control protocols with almost all of theworks focusing only on energy e ciency. There is muchmore innovative work to be done at the MAC layer, butcurrent eorts are not addressing the hard unsolved prob-lems. Majority of the works appearing in the literature are least publishable incremental improvements over the pop-ular S-MAC [1] protocol. In this paper we present researchdirections for future medium access research. We identifysome open issues and discuss possible solutions.Categories and Subject DescriptorsC.2.1 [Network Architecture and Design]: Wirelesscommunication; C.2.6 [Internetworking]: Standards; I.6.5[Model Development]: Modeling methodologiesGeneral TermsDesign, Experimentation, StandardizationKeywordsWireless Sensor Networks, Medium Access Control1. INTRODUCTIONSensor networks (sensor-nets) have emerged as one of thedominant technology trends of this decade (2000-2010). Sen-sors and actuators, wireless communications, and embeddedcomputing are not new concepts but it is the recent low-costlarge-scale integration of computation, communication, andsensing into wireless sensor networks that has capturedthe attention of many researchers. Sensor-nets enable ob-serving the physical world at a granularity level which wasunperceived before [2]. Applications of sensor-nets encom-pass a variety of disciplines and domains, limited only bythe imagination of the application developer.When thinking about future research directions for (wire-less) networks it is important to consider which kind of de-vices would be connected to the network. In the comingyears the most common type of devices on the network wouldbe embedded processors, such as sensors and actuators (i.e.sensor-nets), and improving battery life will be more impor-tant than improving performance [3]. Energy-consumptionby computing chips is falling sharply per unit computation(Moore s Law) whereas energy consumed by radios is de-termined by laws of physics. Thus, the wireless interfacewill be the primary consumer of energy in any device thatcombines computation and radios [3].The Medium Access Control (MAC) layer sits directlyon top of the Physical layer and controls the radio. MACprotocols for sensor-nets focus on energy e ciency (singlemost important goal) instead of meeting traditional goals forwireless MAC design such as fairness, delay, and bandwidthutilization [4]. These protocols tradeo performance (fair-ness, delay, bandwidth utilization) for energy cost. Mainsources of energy wastage at the MAC layer are collisions,idle listening, overhearing, and control packet overhead [4].Unlike the 802.11 WLAN cards where the MAC is usuallyincluded as part of the chipset, in sensor-nets the MAC de-signer has absolute control on the MAC layer design. Thisabsolute control, and the fact that the wireless interfaceis the primary consumer of battery in energy constrainedsensor-nets, has made medium access for sensor networks a very active research area. However, recent studies on MACprotocols for sensor-nets observe that there is no clear trendindicating that medium access for sensor-nets is convergingtowards a unique best solution [5].ACM SIGCOMM Computer Communication Review33Volume 36, Number 2, April 2006Untitled DocumentThereisatendencyof re-inventingthewheel intheMACforsensor-netareawithmajorityoftheworksclaim-ingafewpercentbetterperformanceoverthepopularS-MAC[1]protocol.Insteadoffocusingonlyonenergye -ciency(whichalmostallrecentworksdo),it stimetoad-dressotherunsolvedproblemsattheMAClayer. Inthispaperwepresentresearchdirectionsfordesign,simulation,andexperimentationofMACprotocolsforsensor-nets.Wediscusstechnologytrends,identifysomeopenissuesanddis-cusspossiblesolutions. AdetaileddiscussionofdierentMACprotocolsforsensor-netsandtheirspeci ccharacter-isticsisoutofthescopeofthispaper.Readersnotfamiliarwithmediumaccessinwirelesssensor-netsareencouragedtoseeKoenandGertjan sbookchapteronMACforsensor-nets[5]beforeproceedingtoSection2.2. RESEARCHISSUES&DIRECTIONS2.1 TowardsaSensorNetworkArchitectureThecross-layerdesignsinsensor-netshaveleadtomono-lithic,verticallyintegratedsolutionswhichmightworkinde-pendentlybutarenotreallyusefulforotherresearchgroups.OneoftheearlyencouragingstepstowardsasensornetworkarchitectureisPolastreetal. sensornetprotocol(SP)[6]whichisa exiblealternativetotherecentZigBeestan-dard.SPprovidesastandardizedinterfacetoMAC,withsomefeedbackinbothdirectionsandprovidesanimportantsteptobuildingalargersensornetworkarchitecture.UnlikeIPintheInternet,SPisnotatthenetworklayerbutin-steadsitsbetweenthenetworkanddata-linklayer(becausedata-processingpotentiallyoccursateachhop,notjustattheendpoints).Developingasensornetworkarchitecturewouldbeagrowingandorganicprocess.FutureMACde-signersshouldkeepthegeneralgoalofmovingtowardsasensornetworkarchitectureinmindandtrytomakeuseoftheservicesthatSP[6]hastooer.2.2 StandardizedRadioHardwareWhileSP[6]istheemergingstandardontopoftheMAClayerinthesensor-netnetworkstack,IEEE802.15.4istheemergingstandardforlowerlayers(physicalandmediumac-cess).IEEE802.15.4standardincludesbothphysical(PHY)andmediumaccesscontrol(MAC)speci cations. How-ever,weexpectthatresearcherswouldlargely override the802.15.4MAC(withMACprotocolstailoredforspeci cneedsofsensor-nets)andonlythe802.15.4PHYstandardwouldhaveimplicationsonfuturesensor-netMACdesigns.Newandupcomingsensor-nethardwareplatforms,likeTe-losMotes(Moteiv),SunSPOTSplatform(SunLabs),andMicaZMotes(CrossBow),alreadyuseIEEE802.15.4com-pliantradios.Byusingastandardizedradio,thenodescancommunicatewithanynumberofdevices(possiblyfromdif-ferentvendors)whilesharingthesamephysicallayer.OntheotherhandastheIEEE802.15.4radiointerfacesarepacket-based,thedevelopersloseconsiderable exibilityinsoftwareforcontrollingtheradios.2.3 PeacefulCoexistenceWiththeanticipatedlarge-scaleintroductionofsensor-netsintodailylives,asituationwillemergewheresensor-netsfromdierentvendors/operatorsneedtooperateatacommonfrequencybandinthesamephysicalenvironment(e.g.,2.4GHzinsidethehome).ThisputsarequirementontheMACprotocolstobehave nice toeachother.Simplyrunningatalowduty-cycleisnotenoughtowarrantcoex-istence.First,MACprotocolsneedtobeawareofotherstoguarantee awlessoperation;foreignmessagesneedtobe lteredout,congestiondetectionneedstobeenhancedtoavoidimproperbackos,schedulesneedtobeadjustedtoavoidoverlappingactiveparts,etc. Second,MACproto-colsmustconsidersecurityissuestoprotectagainsteaves-droppingandmaliciousbehavior.Althoughitwillbeclosetoimpossibletopreventdenial-of-serviceattacks,maybeasbrutalasjamming,observingthisandsignallingittotheroutinglayermaypreventatotalbreakdownoftheappli-cation.TinySec[7]isa rststepinsecuringMACproto-cols,butitsrelianceonasharedkeymakesitvulnerableand,thereforemoreadvancedschemesareneeded.Heterogeneity,nodefailuresandnetworkextensionsarecomplicatingfac-torsthathavereceivedlittleattentioninthesecurityarena,butareessentialtothesuccessfuloperationofsensor-netsinthefuture. Notethathardwaresolutions,likeprovid-ingmultiplechannelsattheradio,mayalleviatesomeoftheproblemsofcoexistence,butthesecurityissueswillstillneedtoberesolved.2.4 MobileSensorNetworksTheresearchcommunitygenerallyignoresmobilityattheMAC-layerbecausesensor-netswereoriginallyassumedtocompriseofstaticnodes.However,recentworkslikeRobo-Mote[8]andParasitic-Mobility[9]haveenabledmobilityinsensor-nets.Furthermore,recentlytherehasbeenanin-creasedinterestinmedicalcareanddisasterresponseappli-cationsofsensor-netsandtheseenvironmentsmakeuseofmobilesensornodese.g.sensorsattachedtopatients,doc-torsor rstresponders[10]. Theonlywork,thatweareawareof,whichexplicitlyconsidersmobilityattheMAClayerisMMAC[11]andthereismuchroomforresearchinthisarea.Mobilityevaluationsshouldconsidermorere-alisticmobilitymodelse.g. CodeBlue[10]onlyconsidersthetypicalmovementofadoctorinhospitalhallways,andMMAC[11]usesafairlysimple random mobilitymodelforsimulations.Furthermore,comprehensiverealmobilitytracescouldproveusefulinsuchmobilityevaluations.2.5 NewOptimizationCriteriaTodate,theprimarydesigngoalforsensornetworksingeneralandMACinparticularhasbeenenergye ciency.However,asnewapplicationsofsensornetworksemerge,otheroptimizationcriteria(orQuality-of-Serviceparame-ters)suchaslatencyandcompliancewithreal-timecon-straints(e.g.,monitoringandcontrolinindustrialenviron-ments),orreliabledatadelivery(e.g.,medicalapplications)maygainimportance.Sofar,littleattentionhasbeenpaidtotheminthecontextofsensornetworks. Oneparticu-larissueisthatmanyapplicationsneedtobeoptimizedformultiple,con ictingcriteria(e.g.,energy-e cientandrelia-bility).Hence,applicationsneedawaytoimplementpartic-ulartrade-osbetweenthesecon ictinggoals.Here,MACprotocolswouldberequiredthatprovide turningknobs toprovideapplication-speci ctrade-os.2.6 UnderstandingandExploitingTraf cPatternsMostexistingMACprotocolsforsensornetworksareratherACM SIGCOMM Computer Communication Review 34Volume 36, Number 2, April 2006Untitled Documentgeneralpurposeinthattheysupportarbitrarycommunica-tionpatterns.Whilethisallowstheimplementationofarbi-trarydistributedprotocols,thepricefor exibilityisoftenlimitede ciency(e.g.,intermsofenergyconsumption).However,manyapplicationsofsensornetworksexhibitafewratherspeci ctra cpatternssuchasbroadcastfromthesinktoallsensornodes(tasking,querydistribution,etc.)andconvergecastfromallnodestowardsthesink(datacollection).Also,tra cgenerationshowsspeci cpatternsinsensornetworks. Whilesensornetworksforcontinuousmonitoringexhibit(aggregated)datastreamsfromnodestothesink,event-basedapplicationsgeneratetra conlyupontheoccurrenceofcertaininterestingevents.Sincereal-worldeventsareoftenconcurrentlyobservedbymanynodesintheneighborhoodoftheevent,suchevent-basedapplica-tionsoftenresultinhighlycorrelated(bothtemporallyandspatially)generationoftra cinthenetwork.Inaddition,certaintra cpatternsoftenappearinphasessuchasoc-casionalquerydistributionsfollowedbylongphasesofdatacollectionoraburstofeventreportsfollowedbylongidlephases.Withrespecttothee ciencyofMACprotocols,muchcouldbegainedbybetterunderstandingandexploitingthesetra cpatternsintheMACdesign.Optimalsolutionsforspeci cpatternscouldthenbeintegratedintotra c-adaptiveMACprotocolsthatlearnthecurrentpatternsandadapttheirbehavioraccordingly.2.7 SimulationsConsideredHarmfulAssumptionsmadeinmostsimulationenvironments(aradio stransmissionareaiscircular,allradioshaveequalrangeetc.)donotnecessarilyre ectthereal-worldcondi-tions.Inordertofullyunderstandthecomplexityofdesign-ingaMACprotocolandtodevelopsolutionswhichworkinreallife,itisnecessarytonotonlymodelorsimulatebutalsotoimplementandtestonrealworldsystems.Axiom0:Theworldis at.Axiom1:Aradio stransmissionareaiscircular.Axiom2:Allradioshaveequalrange.Axiom3:IfIcanhearyou,youcanhearme.Axiom4:IfIcanhearyouatall,Icanhearyouperfectly.Axiom5:Signalstrengthisasimplefunctionofdistance.Table1: Mistakenaxiomsofwireless-networkre-searchItisimportanttorevisitKotz smistakenaxiomsofwireless-networkresearch[12](seeTable1)tounderstandwhyMACprotocolsthatyieldextremelyaccurateresultsinsimula-tionfailinreallifedeployments(see[13]forexperiencesfromarealsensor-netdeployment).Kotzetal.surveyedMobiComproceedingsfrom1995to2002andclassi edthesimulationradiomodelsusedintheworksas:FlatEarth,Simple,andGood[12]. Flatearth modelsassumethattwonodescould perfectly communicateiftheyarewithinsomedistance,sayd,ofeachother. Simplemodels arens-2modelswhicharemorerealisticthanthe FlatEarth modelsbutarestillfairlylimitedinemulatingrealradiopropagation(ns-2simulationsarepopularinsensor-netre-searchaswell). Goodmodels aremainlyusedbythecel-lulartelephonycommunityandconcentrateonexactmech-anismsofRFpropagation(takesintoaccountfactorsliketerrain,treedensity,3-Dantennalocation,foliagetypes,wavelength,etc.). Kurkowskietal.[14]surveyedMobi-Hocproceedingsfrom2000-2005andshowedthatpublishedMANETsimulationresultslackbelievability.Sensor-netre-searchersneedtoconductstudiessimilartoKotzetal.[12]andKurkowskietal.[14]onthesensor-netliteratureandidentifythe mistakenaxioms intheradiomodelsusedbypopularsensor-netsimulationenvironments.InSection2.8,wediscussSoftwareDe nedRadiosaspossiblemeansofbridgingthegapbetweensimulationandrealworldperfor-manceofMACprotocols.2.7.1 CurrentSimulationEnvironmentsChoiceofthesimulatormakesanimpactonthevalidityoftheresults.Theradiomodelsprovidedinthestandarddis-tributionofns-2(free-spacemodelandtwo-raygroundre ec-tionmodel)donotre ectthechannelpropagationtypicalofsensor-netenvironments.Shadowingmodelisthelatestad-dition(2000)tons-2radiomodelsandittakesintoaccounttheeectsofindoorobstructionsandoutdoorshadowing.However,thereisstillaneedforbetterRFpropagationtreatmentinns-2.IfusingOMNeT++discreteeventsimulatorforsensor-netMACsimulationsitisbettertouseMACSimulator0.2.2[15]thathascodeforcomparisonprotocolsalreadyavailable. TheOMNeT++MACSimulator0.2.2imple-mentsradiocharacteristicsofEYESnodesradio.However,itsuersfromsomeofthemistakenaxioms(Table1)e.g.itassumesthattheradiosignalsarecircular.TOSSIM[16]enablestheMACdeveloperstochoosetheaccuracyandcomplexityoftheradiomodelasnecessaryfortheirsimulationsasthemodelsareexternaltothesimulator.However,likens-2andMACSimulator0.2.2,theavailableradiomodelsinTOSSIMarefairlysimple.Emstar[17]helpsthesensor-netdeveloperstoeasilymovefromsimulationtoprototypetodeployment.TheimportantpointtonoteaboutEmstaristhatitprovidesaninterfacetoreallow-powerradiosinsteadofasimulatedchannel.Em-starprovidessupportformultipleunderlyingradiotypesanddriversforradiolinkhardware(IEEE802.11andseveral avorsofMicaMotes)arealreadyimplementedinEmstar.2.8 EmbraceSoftwareRadiosSoftwarede nedradiosmightbeasrevolutionaryandtransformingforwirelesscommunicationaspacketswitch-ingwastocircuits[3]. Insoftwareradioscharacteristicslikemediumaccesscontrol,dataencoding,frequencyusageetc.areprogrammableinsteadofhard-wiredandintroduc-ingnewmediaaccessrulesbecomesamatterofasimplesoftwarechange.WearguedinSection2.7thattheperformanceresultsofMACprotocolsinsimulationsshouldbetakenwithagrainofsaltbecauseofpossibleunrealisticassumptionsmadeinradiomodelsofthesimulationenvironment.Apossibleso-lutionaroundthisproblemcouldbetostartdesigningforsoftwarede nedradiosystemsandmovefromsimulationtoprototypes.Softwareradiosallownetworkresearcherstouseinexpensiveo-the-shelfnetworkingcardstoexperimentwithnewMACprotocols.Neufeldet.al.havedevelopedasoftwarearchitecture(softwarede nedradio)SoftMAC,ontopoftheradiosubsystem(IEEE802.11familynetworkACM SIGCOMM Computer Communication Review 35Volume 36, Number 2, April 2006Untitled Documentcards)thatpermitsnetworkresearcherstoeasilyconstructanddeployexperimentaldynamicMAClayers[18].Devel-opingasoftwarearchitecturelikeSoftMAC[18]ontopoftheIEEE802.15.4compliantradiosiscrucialforfuturemediumaccessresearchinsensornetworks.2.9 Real-WorldExperimentsWhilesoftwarede nedradiosareonedirectiontowardsobtainingmorerealisticinsightintoMAClayerperformance,afurtherstepisreal-worldexperiments.Long-termexper-imentssuchastheoneatGreatDuckIsland[19]arethemostrealisticones,butrequirealotofengineeringeortbeforeresearchquestionscanbeanswered[13].Controlledindoorexperimentsarenotasexpressiveaslarge-scalede-ploymentsbutsincerealradiohardwareisused,theydonotsuerfromthemistakenaxiomslistedinTable1.Ritteretal.[20]haveproposedtoreplacethebatterieswithhigh-capacitycapacitors,so-calledGoldCaps,forthepurposeofexperimentalvalidationoflifetimeboundsforwirelesssen-sornetworks.Thisapproachcanalsobeusedtocomparetheenergy-e ciencyofMACprotocolswithoutwaitingforthebatteriesofthesensornodestodrain.Whilethisap-proachhasinherentdrawbacks,e.g. itdoesnottakethebatteryrelaxationeectintoaccount,resultsobtainedus-ingthisapproacharemorerealisticthansimulations.3. CONCLUSIONSWhenreviewingtheopenissuesintheMACforsensor-netareawe ndthatthereisnoclearsingledirectioninwhichfutureeortsshouldbedirected.Somegeneralrec-ommendations,however,couldbemade:"Energye ciencymightbetheprimarydesigngoalforMACprotocolsinsensor-netsbutitshouldnotbetheonlydesigngoal.Optimizationcriteriasuchaslatency,reliabledatadelivery,andcompliancewithreal-timeconstraintsmaygainimportanceinthefuture."Applicationsofsensor-netsexhibitafewspeci ctra cpatternsandmuchcouldbegainedbybetterunder-standingandexploitingthesetra cpatternsintheMACdesign(insteadofsupportingarbitrarycommu-nicationpatterns)."MACprotocolsshouldconsidersecurityissuestopro-tectagainsteavesdroppingandmaliciousbehavior."DierentMACprotocolsforsensor-nets(possiblyfromdierentvendors/operators)operatingatacommonfrequencybandinthesamephysicalenvironmentshouldbeabletopeacefullycoexistwitheachother."ExistingMACprotocolsfor(static)sensor-netsfailtoprovideacceptableperformancewhenappliedtosensor-netswithmobilesensornodes.MACprotocolsforsensor-netsshouldexplicitlyaddresstheeectsofmobilesensornodesintheprotocoldesign."InordertoobtainmorerealisticinsightintoMAClayerperformance,sensor-netresearchersshouldmovefromsimulationtoprototypes(softwarede nedradios)orreal-worldexperiments.4. REFERENCES[1]W.Ye,J.Heidemann,andD.Estrin, Anenergy-e cientmacprotocolforwirelesssensornetworks, inProceedingsoftheIEEEInfocom. NewYork,NY,USA:IEEE,June2002,pp.1567 1576.[2]J.ElsonandD.Estrin,WirelessSensorNetworks. KluwerAcademicPublishers,2004,ch.WirelessSensorNetworks:AbridgetothePhysicalWorld.[3]D.Clark,C.Partridge,R.Braden,B.Davie,S.Floyd,V.Jacobson,D.Katabi,G.Minshall,K.Ramakrishnan,T.Roscoe,I.Stoica,J.Wroclawski,andL.Zhang, Makingtheworld(ofcommunications)adierenceplace, ACMSIGCOMMComputerCommunicationReview,July2005.[4]W.YeandJ.Heidemann,WirelessSensorNetworks. KluwerAcademicPublishers,2004,ch.MediumAccessControlinWirelessSensorNetworks,AlsoasimilarversionisavailableasISI-TR-580.[5]K.LangendoenandG.Halkes,EmbeddedSystemsHandbook.CRCPress,Aug.2005,ch.Energy-E cientMediumAccessControl,Availableonlineat:http://pds.twi.tudelft.nl/[6]J.Polastre,J.Hui,P.Levis,J.Zhao,D.Culler,S.Shenker,andI.Stoica, Aunifyinglinkabstractionforwirelesssensornetworks, inProc.ACMSenSys 05,Nov.2005.[7]C.Karlof,N.Sastry,andD.Wagner, TinySec:alinklayersecurityarchitectureforwirelesssensornetworks, inProc.SenSys 04,Nov.2004,pp.162 175.[8]K.Dantu,M.Rahimi,H.Shah,S.Babel,A.Dhariwal,andG.S.Sukhatme, Robomote:Enablingmobilityinsensornetworks, inIEEE/ACMFourthInternationalConferenceonInformationProcessinginSensorNetworks(IPSN/SPOTS),Apr.2005.[9]M.LaibowitzandJ.A.Paradiso, Parasiticmobilityforpervasivesensornetworks, inThirdInternationalConferenceonPervasiveComputing(PERVASIVE2005),Munich,Germany,May2005.[10]V.Shnayder,B.rongChen,K.Lorincz,T.R.F.Fulford-Jones,andM.Welsh, Sensornetworksformedicalcare, HarvardUniversity,Tech.Rep.TechnicalReportTR-08-05,April2005.[11]M.Ali,T.Suleman,andZ.A.Uzmi, MMAC:Amobility-adaptive,collision-freemacprotocolforwirelesssensornetworks, inProc.24thIEEEIPCCC 05,Phoenix,Arizona,USA,April2005.[12]D.Kotz,C.Newport,R.S.Gray,J.Liu,Y.Yuan,andC.Elliott, Experimentalevaluationofwirelesssimulationassumptions, inProceedingsoftheACM/IEEEInternationalSymposiumonModeling,AnalysisandSimulationofWirelessandMobileSystems(MSWiM),October2004,pp.78 82.[13]K.Langendoen,A.Baggio,andO.Visser, Murphylovespotatoes:Experiencesfromapilotsensornetworkdeploymentinprecisionagriculture, in14thInt.WorkshoponParallelandDistributedReal-TimeSystems(WPDRTS),Rhodes,Greece,April2006.[14]S.Kurkowski,T.Camp,andM.Colagrosso, MANETsimulationstudies:theincredibles, SIGMOBILEMob.Comput.Commun.Rev.,vol.9,no.4,pp.50 61,2005.[15] Macsimulator0.2.2. [Online].Available:http://www.consensus.tudelft.nl/software.html[16]P.Levis,N.Lee,M.Welsh,andD.Culler, TOSSIM:accurateandscalablesimulationofentiretinyosapplications, inACMSensys 03. NewYork,NY,USA:ACMPress,2003,pp.126 137.[17]L.Girod,T.Stathopoulos,N.Ramanathan,J.Elson,D.Estrin,E.Osterweil,andT.Schoellhammer, Asystemforsimulation,emulation,anddeploymentofheterogeneoussensornetworks, inACMSensys 04. NewYork,NY,USA:ACMPress,2004,pp.201 213.[18]M.Neufeld,J.Fi eld,C.Doerr,A.Sheth,andD.Grunwald, SoftMAC exiblewirelessresearchplatform, inProc.HotNets-IV,CollegePark,Maryland,USA,Nov.2005.[19]R.Szewczyk,J.Polastre,A.Mainwaring,andD.Culler, Lessonsfromasensornetworkexpedition, inProc.EWSN2004,Berlin,Germany,Jan.2004.[20]H.Ritter,J.Schiller,T.Voigt,A.Dunkels,andJ.Alonso, Experimentalevaluationoflifetimeboundsforwirelesssensornetworks, inProc.EWSN2005,Istanbul,Turkey,Jan.2005.ACM SIGCOMM Computer Communication Review 36Volume 36, Number 2, April 2006
