evolutionary power‐aware routing in vanets using...
TRANSCRIPT
Madrid, Spain 2-6 July 2012
J.Toutouh,S.Nesmachnow,andE.Alba Evolu&onaryPower‐AwareRou&nginVANETsusingMonte‐CarloSimula&on
OPTIM 2012 HPCS 2012
EvolutionaryPower‐AwareRoutinginVANETsusingMonte‐CarloSimulation
OptimizationIssuesinEnergyEfficientDistributedSystems(HPCS2012)
JamalToutouh,SergioNesmachnow*,andEnriqueAlbaUniversityofMalaga,Spain UniversidaddelaRepública,Uruguay*
Madrid, Spain 2-6 July 2012
J.Toutouh,S.Nesmachnow,andE.Alba Evolu&onaryPower‐AwareRou&nginVANETsusingMonte‐CarloSimula&on
OPTIM 2012 HPCS 2012
3
Outline
Introduction
Power‐AwareAODV
ExperimentalResults
ConclusionsandFutureWork
1
2
4
IntroductionPower‐AwareAODV
ExperimentalAnalusisConclusionsandFutureWork
1/13
Madrid, Spain 2-6 July 2012
J.Toutouh,S.Nesmachnow,andE.Alba Evolu&onaryPower‐AwareRou&nginVANETsusingMonte‐CarloSimula&on
OPTIM 2012 HPCS 2012
1.Introduction.VANETsandEnergyIssues Vehicularadhocnetworks(VANETs)areself‐configuringwirelessadhocnetworks inwhich thenodesarevehicles, elementsof roadside,sensors,andpedestrianpersonaldevices(smartphones,PDAs,etc.)
2/13
• VANETandEnergyIssues• Motivation
TransportEfficiency Safety
Deployedtoenableup‐to‐minuteroadtrafficinformationexchange
IntroductionPower‐AwareAODV
ExperimentalAnalysisConclusionsandFutureWork
VANETs involve devices fed with limited powersources. Therefore, power‐aware networkarchitecturesandprotocolsarehighlydesirable
Theroutingprotocolaffectsthenodespowerconsumptionintwoways: Theprotocoloperationamountofenergyusedtocomputetheroutingpaths The computed routing paths the terminals power consumption whenforwardingpackets
Madrid, Spain 2-6 July 2012
J.Toutouh,S.Nesmachnow,andE.Alba Evolu&onaryPower‐AwareRou&nginVANETsusingMonte‐CarloSimula&on
OPTIM 2012 HPCS 2012
1.Introduction.Motivation
3/13
WestudytheapplicationofDifferentialEvolution(DE)tocomputepower‐awareroutingprotocolconfigurations
The optimization process is guided by evaluating tentative solutions(protocolparameterizations)bymeansofVANETsimulations
StochasticpropagationmodelsthatproducedifferentresultswhenevaluatingthesameVANETscenario(upto57%forsomemetricsandscenarios)
A number of simulations of the same scenario is performed applyingMonte‐Carlomethodtoevaluateasolution
AstheVANETsimulationsrequirehighcomputationalcoststhesimulationsareperformedinparallelusingdifferentprocessunitsofacluster
• VANETandEnergyIssues• Motivation
IntroductionPower‐AwareAODV
ExperimentalAnalysisConclusionsandFutureWork
Madrid, Spain 2-6 July 2012
J.Toutouh,S.Nesmachnow,andE.Alba Evolu&onaryPower‐AwareRou&nginVANETsusingMonte‐CarloSimula&on
OPTIM 2012 HPCS 2012
2.Power‐AwareAODV.ProblemDefinition
4/13
Finding the best AODV configurations to enable green communications inVANETsisthemainsubjectofthiswork
AODVisaroutingprotocolformobileadhocnetworksusedinVANETsAODVRFC3561
Anexcessiveenergyconsumptionreductioncanleadtoprotocolmalfunction
QoSrestrictionMaximumallowedPDRdegradation15%(overthestandard)
parameter RFCvalue type rangeHELLO_INTERVAL 1.0s real [1.0,20.0]ACTIVE_ROUTE_TIMEOUT 3.0s real [1.0,20.0]MY_ROUTE_TIMEOUT 6.0s real [1.0,40.0]NODE_TRAVERSAL_TIME 0.04s real [0.01,15.0]MAX_RREQ_TIMEOUT 10.0s real [1.0,100.0]NET_DIAMETER 35 integer [3,100]ALLOWED_HELLO_LOSS 2 integer [0,20]REQ_RETRIES 2 integer [0,20]TTL_START 1 integer [1,40]TTL_INCREMENT 2 integer [1,20]TTL_THRESHOLD 7 integer [1,60]
• ProblemDefinition• Methodology• OptimizationMethodDetails• FitnessEvaluation
IntroductionPower‐AwareAODV
ExperimentalAnalysisConclusionsandFutureWork
Madrid, Spain 2-6 July 2012
J.Toutouh,S.Nesmachnow,andE.Alba Evolu&onaryPower‐AwareRou&nginVANETsusingMonte‐CarloSimula&on
OPTIM 2012 HPCS 2012
Automatic optimization tool coupling Differential Evolution (DE) andMonte‐CarloVANETsimulation
Methodology
(energy, PDR)
Differen4al
Evolu4on
AODV configuration
Energy-efficient AODV configuration
Solution evaluation new solution (AODV configuration)
fitness value
x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11
AODV protocol ns‐2VANETsimula4onn
AODV protocol ns‐2VANETsimula4onn
AODV protocol AODV protocol ns‐2VANETsimula4onn
AODV
x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11
Fitnessfunc4on
Monte Carlo
method
nprocessingunits
2.Power‐AwareAODV.Methodology
• ProblemDefinition• Methodology• OptimizationMethodDetails• FitnessEvaluation
5/13
IntroductionPower‐AwareAODV
ExperimentalAnalysisConclusionsandFutureWork
Madrid, Spain 2-6 July 2012
J.Toutouh,S.Nesmachnow,andE.Alba Evolu&onaryPower‐AwareRou&nginVANETsusingMonte‐CarloSimula&on
OPTIM 2012 HPCS 2012
ProblemEncoding:
FitnessFunction:Penaliza4onmodel(PDRdegrada&on>15%)
2.Power‐AwareAODV.OptimizationMethodDetails
Realvalued Integers
Δ=0.1,ω1=0.9,andω2=‐0.1
parameter type rangeHELLO_INTERVAL real [1.0,20.0]ACTIVE_ROUTE_TIMEOUT real [1.0,20.0]MY_ROUTE_TIMEOUT real [1.0,40.0]NODE_TRAVERSAL_TIME real [0.01,15.0]MAX_RREQ_TIMEOUT real [1.0,100.0]NET_DIAMETER integer [3,100]ALLOWED_HELLO_LOSS integer [0,20]REQ_RETRIES integer [0,20]TTL_START integer [1,40]TTL_INCREMENT integer [1,20]TTL_THRESHOLD integer [1,60]
• ProblemDefinition• Methodology• OptimizationMethodDetails• FitnessEvaluation
6/13
IntroductionPower‐AwareAODV
ExperimentalAnalysisConclusionsandFutureWork
Madrid, Spain 2-6 July 2012
J.Toutouh,S.Nesmachnow,andE.Alba Evolu&onaryPower‐AwareRou&nginVANETsusingMonte‐CarloSimula&on
OPTIM 2012 HPCS 2012
Initialization:Uniforminitializationtoassurethattheinitialpopulationcontainsindividualsfromdifferent areas of the search space. It splits the search space into pop_size(numberofindividuals)diagonalsubspaces.
Crossover:WeusedBlendCrossoverOperator(BLX‐α),awell‐knownrecombinationoperatorforrealcodedEAs
2.Power‐AwareAODV.OptimizationMethodDetails
x
y
• ProblemDefinition• Methodology• OptimizationMethodDetails• FitnessEvaluation
7/13
IntroductionPower‐AwareAODV
ExperimentalAnalusisConclusionsandFutureWork
Madrid, Spain 2-6 July 2012
J.Toutouh,S.Nesmachnow,andE.Alba Evolu&onaryPower‐AwareRou&nginVANETsusingMonte‐CarloSimula&on
OPTIM 2012 HPCS 2012
RealisticVANETsimulationsshouldreflectrealworldinteractions.Inthissense,stochastic signal propagationmodels are used. In ourwork,we have used theNakagamipropagationmodel.Thesimulationresultsarenon‐deterministic(deviationsupto57%)
Monte‐Carlosimulation isusedtoperformmoreaccuratefitnessevaluations.EnergyandPDRarecomputedrepeatingntimesthesimulation
2.Power‐AwareAODV.FitnessEvaluation
nprocessingunits
Perform VANET simulations
Computes the fitness by using Monte-Carlo method
MASTER
SLAVE SLAVE SLAVE
F( )
Asns‐2 VANET simulation require large execution times (minutes)we appliedtheparallelmultithreadingmaster‐slaveparadigmtorunthensimulationsovernprocessingunits
Reducingdrasticallytherequiredtimetoevaluatethesolutionfitness
Inourwork,n=24
• ProblemDefinition• Methodology• OptimizationMethodDetails• FitnessEvaluation
8/13
IntroductionPower‐AwareAODV
ExperimentalAnalysisConclusionsandFutureWork
Madrid, Spain 2-6 July 2012
J.Toutouh,S.Nesmachnow,andE.Alba Evolu&onaryPower‐AwareRou&nginVANETsusingMonte‐CarloSimula&on
OPTIM 2012 HPCS 2012
VANETscenariosdefinition: Geographicalareas:
G1:240,000m2
G2:360,000m2
Numberofnodes:G1:20G2:30,45,and60
Generateddatatrafficrate: 128,256,512,and1024kbps
OptimizationscenarioG1,20,512kbps
DE(C++,MALLBA,andstandardpthreadlibrary): Populationsize=8individuals Numberofgenerations=50 Crossover:probability=0.9,α=0.2
9/13
• ExperimentsDefinition• DEOptimizationResults• ValidationResults
G2
G1
N
MÁLAGA
3.ExperimentalAnalysis.ExperimentsDefinition
AberperformingDEconfigura&onanalysisexperimentsusingG1,20,128kbpsscenariotofindthebestDEparameteriza&on
IntroductionPower‐AwareAODV
ExperimentalAnalysisConclusionsandFutureWork
Madrid, Spain 2-6 July 2012
J.Toutouh,S.Nesmachnow,andE.Alba Evolu&onaryPower‐AwareRou&nginVANETsusingMonte‐CarloSimula&on
OPTIM 2012 HPCS 2012
3.ExperimentalAnalysis.DEOptimizationResults
Bestconfigurationfound: HELLO_INTERVAL=11.994,ACTIVE_ROUTE_TIMEOUT=12.439, MY_ROUTE_TIMEOUT=15.965,NODE_TRAVERSAL_TIME=8.106, MAX_RREQ_TIMEOUT=42.466,NET_DIAMETER=66, ALLOWED_HELLO_LOSS=6,REQ_RETRIES=9,TTL_START=12, TTL_INCREMENT=19,andTTL_THRESHOLD=54
Mainproperties: itgenerateslowercontroltraffic eachnodespendslesstimeinthetransmittingandthereceivingstates itmanageslongerpathssincethenetworkdiameterandthetimetolivearelarger itismoretoleranttodisconnectionsandpacketlossbecauseitallowsgreaterhellopacketlossanditresendRREQpacketsmoretimes
The average power consumption obtained with the optimized AODVconfiguration reduced 33.55% the RFC energy requirements, while thePDRdegradationwasonly5.17%
• ExperimentsDefinition• DEOptimizationResults• ValidationResults
10/13
IntroductionPower‐AwareAODV
ExperimentalAnalysisConclusionsandFutureWork
Madrid, Spain 2-6 July 2012
J.Toutouh,S.Nesmachnow,andE.Alba Evolu&onaryPower‐AwareRou&nginVANETsusingMonte‐CarloSimula&on
OPTIM 2012 HPCS 2012
3.ExperimentalAnalysis.ValidationResults
Bestenergy‐awareAODVconfigurationvsRFC3561over9scenariosinG2area Results:
Averagepowerconsumptionsaving=32.3%/averagePDRdegradation=8.8% Largestenergyreductioningreatestroadtrafficdensity(60nodes,max=68.9%)
SuffersfromseveraldropinthePDR Lowestenergysavingsinscenarioswiththelargestdatarates(1024Kbps) Kruskal‐Wallis statistical test results indicate that the energy improvementscanbeconsideredstatisticallysignificant
34%26%
19%
50%
00%
10%
20%
30%
40%
50%
G1_30 G2_30 G2_45 G2_60
energyim
provem
ent
overRFC
5% 7% 5%
18%
0%
10%
20%
30%
40%
50%
G1_30 G2_30 G2_45 G2_60
PDRde
grad
a4on
overRFC
• ExperimentsDefinition• DEOptimizationResults• ValidationResults
IntroductionPower‐AwareAODV
ExperimentalAnalusisConclusionsandFutureWork
Madrid, Spain 2-6 July 2012
J.Toutouh,S.Nesmachnow,andE.Alba Evolu&onaryPower‐AwareRou&nginVANETsusingMonte‐CarloSimula&on
OPTIM 2012 HPCS 2012
4.ConclusionsandFutureWork
12/13
Automaticmethodologyforcomputingpower‐awareAODVconfigurationsforVANETs,bycouplingDEandMonte‐Carlomethod(ns‐2simulations) Maincontribution:
DEandaparallelmaster‐slaveMonte‐Carlo simulation to reduce thevariabilityoftheconfigurationevaluationduetorandomnessinNakagamimodel
Mainresults:Significant reductions in power consumption when using the DE AODV energy‐awareconfiguration(32.3%average,68.9%best);averagePDRdegradation:8.8%
PromisingmethodologyforautomaticandefficientcustomizationofVANETroutingprotocols
Futurework: improvethesearchtoreducetheQoSdegradationindensescenarios useotherfitnessfunctions(newmetrics,explicitmulti‐objectiveapproaches) parallelizethepopulationevaluationtoallowlargerpopulationsinDE improvetheresultsbyusingseveralscenariostoevaluateeachconfiguration
IntroductionPower‐AwareAODV
ExperimentalAnalusisConclusionsandFutureWork
Madrid, Spain 2-6 July 2012
J.Toutouh,S.Nesmachnow,andE.Alba Evolu&onaryPower‐AwareRou&nginVANETsusingMonte‐CarloSimula&on
OPTIM 2012 HPCS 2012
Thankyouforyourattention…
Comments [email protected]