Performance evaluation of Urban Mesh Networks Vinay Sridhara Department of Electrical Engineering University of Delaware.

Performance evaluation of Performance evaluation of Urban Mesh NetworksUrban Mesh Networks

Vinay SridharaVinay Sridhara

Department of Electrical Department of Electrical EngineeringEngineering

University of DelawareUniversity of Delaware

OverviewOverview MotivationMotivation Simulation EnvironmentSimulation Environment Mobility modelsMobility models Propagation modelsPropagation models Simulation methodologySimulation methodology Simulation parametersSimulation parameters Performance evaluationPerformance evaluation

– CoverageCoverage– Achievable bit rateAchievable bit rate– Application performanceApplication performance

Web like file transfersWeb like file transfers Voice over IPVoice over IP Streaming media applicationStreaming media application

Conclusions and future workConclusions and future work ReferencesReferences

MotivationMotivation Major cities like Philadelphia, Las Vegas and Los Angeles are planning to Major cities like Philadelphia, Las Vegas and Los Angeles are planning to

have a massive wireless hotspots which is expected to span more or less have a massive wireless hotspots which is expected to span more or less the entire citythe entire city

They are aimed at supporting applications ranging from monitoring and They are aimed at supporting applications ranging from monitoring and controlling vehicular traffic to assisting emergency response and remote controlling vehicular traffic to assisting emergency response and remote situation assessment situation assessment

PhiladelphiaPhiladelphia– Plans to provide ubiquitous wireless access to residents and visitorsPlans to provide ubiquitous wireless access to residents and visitors– Area – 136 square milesArea – 136 square miles– Density – 4000 infrastructure nodes mounted on the lamppostsDensity – 4000 infrastructure nodes mounted on the lampposts– Distance between the infrastructure nodes is ~ 300mDistance between the infrastructure nodes is ~ 300m– Mobile nodes are not allowed to act as relaysMobile nodes are not allowed to act as relays

Las Vegas Las Vegas – Has already deployed a pilot test systemHas already deployed a pilot test system– Covers entire city Covers entire city – The distance between infrastructure nodes is ~ 400mThe distance between infrastructure nodes is ~ 400m– Mobile nodes are allowed to act as relaysMobile nodes are allowed to act as relays

These kind of networks have never been studied before either through These kind of networks have never been studied before either through simulations or any other meanssimulations or any other means

Things to consider when Things to consider when simulating wireless networkssimulating wireless networks

Node Mobility Propagation Model

Environment under which the network operates

UDelModelsUDelModels The important aspects for wireless The important aspects for wireless

simulationssimulations– Tool to model the environmentTool to model the environment

Should be able to specify Buildings, roads, vegetation Should be able to specify Buildings, roads, vegetation etcetc

– Realistic propagation modelRealistic propagation model Considers the effects of the environmentConsiders the effects of the environment Should be able to model reflections, diffractions and Should be able to model reflections, diffractions and

scatteringscattering– Realistic mobility modelRealistic mobility model

Should be able to give a realistic node distributionShould be able to give a realistic node distribution Should be able to depict the socio economic behavior Should be able to depict the socio economic behavior

of the peopleof the people Should be able to model different mobile nodes such Should be able to model different mobile nodes such

as pedestrians, cars, UAVsas pedestrians, cars, UAVs




UDelModels - MapBuilderUDelModels - MapBuilder In order to simulate MANETs in urban area it is In order to simulate MANETs in urban area it is

necessary to have a model of the urban areanecessary to have a model of the urban area– Random cities can be built by placing buildings randomly Random cities can be built by placing buildings randomly

and using Voronoi diagram to connect them.and using Voronoi diagram to connect them.– Such cities might lack typical characteristics of a real Such cities might lack typical characteristics of a real

city, like long thoroughfares and big intersections which city, like long thoroughfares and big intersections which play important role in mobility and propagationplay important role in mobility and propagation

MapBuilder is a software tool that helps in MapBuilder is a software tool that helps in specifying the environment for the simulationspecifying the environment for the simulation

The MapBuilder uses the 3-D GIS dataset to The MapBuilder uses the 3-D GIS dataset to construct city map skeleton with buildingsconstruct city map skeleton with buildings

It gives options to the user to draw other It gives options to the user to draw other components such as roads, traffic lights, etc.components such as roads, traffic lights, etc.

UDelModels - MapBuilderUDelModels - MapBuilder The MapBuilder does The MapBuilder does

not specify anything not specify anything with respect to the with respect to the interiors of the interiors of the buildings.buildings.

Our models assumes Our models assumes some general structures some general structures for the interiors of the for the interiors of the buildings.buildings.– OfficeOffice– Office/StoresOffice/Stores– ResidencesResidences– Residence/StoresResidence/Stores– StoresStores– Subway StationsSubway Stations

UDelModels - MapBuilderUDelModels - MapBuilder




Existing Mobility Models

Entity Mobility Models

Group Mobility Models

Random Walk

City Section

Random Waypoint

Random Direction

Boundless simulation area

Gauss-Markov

Probabilistic version of Random walk

Exponentially Correlated Random waypoint

Column Model

Nomadic Model

Pursue Model

Reference point Group Mobility

Memoryless

Past speed and direction

Urban Environments

Why do we need other Why do we need other models?models?

The existing models do not The existing models do not consider the effects of the consider the effects of the operating environmentoperating environment

Even though the empirical Even though the empirical models such as Okamura-Hata models such as Okamura-Hata are modeled for urban/suburban are modeled for urban/suburban scenarios, they are not suitable scenarios, they are not suitable for site specific modelsfor site specific models

The existing mobility models are The existing mobility models are randomrandom

They fail to incorporate the They fail to incorporate the effects of operating effects of operating environmentsenvironments

Even though some models do Even though some models do consider some effects of consider some effects of obstacles and others, they are obstacles and others, they are extremely simple and do not extremely simple and do not consider the overall social consider the overall social behavioral aspects of the mobile behavioral aspects of the mobile nodes.nodes.

UDelModels – Mobility ModelUDelModels – Mobility Model There is little doubt that the existing mobility There is little doubt that the existing mobility

models are not realisticmodels are not realistic Existing mobility models tend to follow some Existing mobility models tend to follow some

smooth distribution (e.g. Nodes are uniformly smooth distribution (e.g. Nodes are uniformly distributed)distributed)– The nodes tend to be in the center of the regionThe nodes tend to be in the center of the region

In reality the nodes are restricted to streets, In reality the nodes are restricted to streets, sidewalks outdoors and hallways and office sidewalks outdoors and hallways and office (residence, store) locations indoors(residence, store) locations indoors

Nodes tend to move in clustersNodes tend to move in clusters– This is due to the faster nodes catching up with the This is due to the faster nodes catching up with the

slower nodes and not passing themslower nodes and not passing them– Due to the traffic lightsDue to the traffic lights

UDelModels – Mobility ModelUDelModels – Mobility Model

UDel mobility model is based on three well UDel mobility model is based on three well studied fields namelystudied fields namely– Urban PlanningUrban Planning– Meeting AnalysisMeeting Analysis– Time use studyTime use study

The mobility models assumes a three The mobility models assumes a three layered approachlayered approach– Activity modelActivity model– Task modelTask model– Agent modelAgent model


Activity Model Activity Model – Describes the high level activity type of a mobile Describes the high level activity type of a mobile

nodenode– Based on the U.S. Bureau of Labor Statistics 2003 Based on the U.S. Bureau of Labor Statistics 2003

“Time Use Study”“Time Use Study”– Based on the data collected from more than 20,000 Based on the data collected from more than 20,000

people. (Of them 5000 were urban residents)people. (Of them 5000 were urban residents)– This model describes basic activities such as This model describes basic activities such as

Working Eating not at work Shopping At home etc

UDelModels – Mobility ModelUDelModels – Mobility Model Activity ModelActivity Model

– 1. Select a home and office.– 2. The arrival time at work is

determined.– 3. The duration at work is

determined.– 4. Determine if a break from

work is taken. If yes go to step 5– 5. The break start time is

determined.– 6. The number of activities

performed during a break is determined

– 7. Which activities are performed during the break is determined.

– 8. The duration of each activity is determined.

– 9. The arrival time back at work is determined and it is determined that a break is taken again. If so, steps 5-9 are repeated.


Task ModelTask Model– The task model The task model

consists of the tasks consists of the tasks performed during performed during different activitiesdifferent activities

– ExamplesExamples Meetings during Meetings during

office hoursoffice hours Activity eating Activity eating

consists of going to a consists of going to a restaurantrestaurant

Etc.Etc.

Agent ModelAgent Model– Inter-node speed Inter-node speed

distance distance relationshipsrelationships

– Lane changingLane changing




Propagation modelPropagation model

Free-space modelFree-space model2

4

dGGPP trtr

0 500 1000 15000

1

2

0 100 2000

1

2

addedattenuation

distance between transmitter and receiver (meters)

• Two-ray model – occurs when there is LOS between sender and receiver

• Other empirical models such as Okamura-Hata model

UDelModels – Propagation UDelModels – Propagation modelmodel

The main factors that affect the probability of a packet error rate The main factors that affect the probability of a packet error rate areare– Signal strength, Delay spread, Doppler spread and noise (includes interference)Signal strength, Delay spread, Doppler spread and noise (includes interference)

The signal strength at a receiver is given by PThe signal strength at a receiver is given by Preceivedreceived=P=Ptransmittedtransmitted*C*PL*C*PL A large volume of research has shown that signals can be modeled A large volume of research has shown that signals can be modeled

as raysas rays– These rays reflect of the ground and walls, diffracts around the corners and gets These rays reflect of the ground and walls, diffracts around the corners and gets

transmitted through the wallstransmitted through the walls PL = 1/dPL = 1/d22*Attenuation*Attenuation Attenuation is a complex number that depends on the effects of Attenuation is a complex number that depends on the effects of

reflection diffraction and transmissionreflection diffraction and transmission The signal strength can be determined by finding the lengths of all The signal strength can be determined by finding the lengths of all

the rays that hit the receiver and the attenuation experienced by the rays that hit the receiver and the attenuation experienced by themthem

Attenuation and change in phase due to reflection or transmission Attenuation and change in phase due to reflection or transmission requires the knowledge of the frequency, polarization of the signal, requires the knowledge of the frequency, polarization of the signal, angle of incidence, thickness and the type of material that is angle of incidence, thickness and the type of material that is reflecting or transmitting the signal throughreflecting or transmitting the signal through


The propagation model uses the The propagation model uses the technique of Beamtracing, technique of Beamtracing, vertical plane rays and other vertical plane rays and other techniquestechniques

The ground plane is divided into The ground plane is divided into uniform grid and the walls are uniform grid and the walls are divided into discrete uniform sized divided into discrete uniform sized tilestiles

The computation is divided into The computation is divided into two partstwo parts– Pre ProcessingPre Processing

In this part all the ray-neighbors of In this part all the ray-neighbors of a wall tile are founda wall tile are found

– BeamtracingBeamtracing This is carried out in a BFS manner This is carried out in a BFS manner

with each beam continued to be with each beam continued to be reflected, transmitted, diffracted reflected, transmitted, diffracted and/or sub-divided until the beam and/or sub-divided until the beam exits the modeled area or the exits the modeled area or the pathloss surpasses the thresholdpathloss surpasses the threshold

The Beamtracing can be carried The Beamtracing can be carried out indoors and outdoors (but out indoors and outdoors (but Beamtracing exceeds the Beamtracing exceeds the computational ability due to computational ability due to number of walls)number of walls)


Computational ComplexityComputational Complexity

NumberNumber CoverageCoverage TimeTime

LOSLOS 937937 5656

11 26232623 5959

22 39273927 61.561.5

33 42434243 8585

>4>4 42654265 122122

NumberNumber Reflections Reflections onlyonly

Reflections Reflections + + DiffractionsDiffractions

11 19601960 26232623

22 26162616 39273927

33 28622862 39273927

44 30653065 4326543265

AccuracyAccuracy



0 200 400 600 800 10000

20

40

60

80

100

Distance in meters

Pa

thlo

ss

in

dB

Pathloss comparison for Raytracing and Measurement

Pathloss data from RaytracingPathloss data from Measurement

A A

BG

E F

I H

DC

AAB BC

F

E

DG

trees

archway

large airconditioners

bridge

archway

H

bushesI

Distance in meters20 30 40 50 60 700

10

20

30

40

50

60

70

Pa

thlo

ss

in

dB

Pathloss comparison for Raytracing and Measurement

Pathloss data from RaytracingPathloss data from Measurement

building

buildingbuilding

building

stre

et

street

Source

Indoor Propagation ModelIndoor Propagation Model

Attenuation factor modelAttenuation factor model– The Beamtracing indoors in a large section of city The Beamtracing indoors in a large section of city

exceeds today’s computational abilitiesexceeds today’s computational abilities– It has been found that AF model can be used for It has been found that AF model can be used for

propagation indoors, with a fair accuracypropagation indoors, with a fair accuracy– It has been shown that the model can provide realistic It has been shown that the model can provide realistic

pathloss estimates with an error of less than 4dBpathloss estimates with an error of less than 4dB

1 wall1 wall 4dB4dB

1 floor1 floor 30dB30dB

2 floors2 floors 35dB35dB

3 floors3 floors 39 dB39 dB

4 or more4 or more 40dB40dB

Indoor Propagation ModelIndoor Propagation Model

1 1915105

21

23

22

20

24X

2 4 6 8 10 12 14 16 18 20 22 240

10

20

30

40

50

60

70

80

90Pathloss comparison for Model and Measurement

Measurement Points

Pat

hlos

s in

dB

Pathloss Data from ModelPathloss Data from Measurement

Big PictureBig Picture Number of cities are planning to deploy large scale Number of cities are planning to deploy large scale mesh mesh

networksnetworks to provide ubiquitous internet access to the to provide ubiquitous internet access to the residentsresidents

They are also aimed at supporting applications ranging They are also aimed at supporting applications ranging from monitoring and controlling vehicular traffic to assisting from monitoring and controlling vehicular traffic to assisting emergency response and remote situation assessment emergency response and remote situation assessment

Market research indicates that the mesh networks will grow Market research indicates that the mesh networks will grow to a multi-billion dollar industry with in next few yearsto a multi-billion dollar industry with in next few years

The performance of such networks have never been studied The performance of such networks have never been studied before through simulations or any other meansbefore through simulations or any other means

Past research work has focused mostly on wireless ad-hoc Past research work has focused mostly on wireless ad-hoc networks with a few mobile nodes operating in an area of networks with a few mobile nodes operating in an area of 500x500 m500x500 m22..

Also in most cases the propagation and the mobility models Also in most cases the propagation and the mobility models used are far from realityused are far from reality

Mesh Networks ?Mesh Networks ?

Wireless links

Very high bandwidth wired links

Base-station withInternet

connectivity

Fixed wirelessrelays

Performance CriteriaPerformance Criteria

CoverageCoverage– Coverage is the first issue that arises Coverage is the first issue that arises

when considering the wireless networkswhen considering the wireless networks Achievable bit rateAchievable bit rate

– 802.11b/g physical layer is used802.11b/g physical layer is used Application performanceApplication performance

– Web like file transfersWeb like file transfers– Voice over IPVoice over IP– Streaming mediaStreaming media

SimulationsSimulations

We try to evaluate the performance of the We try to evaluate the performance of the Urban mesh networks with realistic mobility Urban mesh networks with realistic mobility and propagation modelsand propagation models

We use UDelModels suite of simulation tools We use UDelModels suite of simulation tools for this performance evaluationfor this performance evaluation

Qualnet is used as the simulator for evaluating Qualnet is used as the simulator for evaluating the performance of the application protocolsthe performance of the application protocols

The propagation model and the mobility The propagation model and the mobility models are incorporated into the Qualnet models are incorporated into the Qualnet simulatorsimulator

Simulation ParametersSimulation Parameters Mobility parametersMobility parameters

– Simulations are conducted with 12,000 people with 1600 acting as Simulations are conducted with 12,000 people with 1600 acting as mobile nodesmobile nodes

This affects the way and the speeds at which the mobile nodes moveThis affects the way and the speeds at which the mobile nodes move This simulates the realistic clustering of nodesThis simulates the realistic clustering of nodes

– Simulations are conducted at different times in the daySimulations are conducted at different times in the day 11:30AM11:30AM 12:30PM12:30PM 02:00PM02:00PM

– The above times are of significant importance because most of the The above times are of significant importance because most of the people would have arrived at work and some are still arrivingpeople would have arrived at work and some are still arriving

– The people who have arrived at will be taking breaksThe people who have arrived at will be taking breaks– Simulations start 60 minutes prior to the sampling timeSimulations start 60 minutes prior to the sampling time– The sampling time duration is 20 minutesThe sampling time duration is 20 minutes– Primarily office going and taking breaks for shopping and lunch Primarily office going and taking breaks for shopping and lunch

activities are consideredactivities are considered– The speed with which a node moves is uniformly distributed between The speed with which a node moves is uniformly distributed between

1.6 and 4.0 miles per hour1.6 and 4.0 miles per hour

Simulation ParametersSimulation Parameters Simulation environmentSimulation environment

– 4 different cities with varying building densities are 4 different cities with varying building densities are modeledmodeled A section of Paddington area in LondonA section of Paddington area in London

– Area = 1000m x 1000mArea = 1000m x 1000m– Number of buildings = 130Number of buildings = 130

University of Delaware campusUniversity of Delaware campus– Area = 1700m x 1500mArea = 1700m x 1500m– Number of buildings = 140Number of buildings = 140

Idealized Grid CityIdealized Grid City– Area = 800m x 800mArea = 800m x 800m– Number of buildings = 56Number of buildings = 56

A section of city core of ChicagoA section of city core of Chicago– Area = 3000m x 3000mArea = 3000m x 3000m– Number of buildings = 270Number of buildings = 270

– Indoor structuresIndoor structures Office, Office/Stores, Residences, Residence/Stores, StoresOffice, Office/Stores, Residences, Residence/Stores, Stores

Simulation ParametersSimulation Parameters Infrastructure node densityInfrastructure node density The infrastructure nodes are assumed to mounted on The infrastructure nodes are assumed to mounted on

the lamp-posts at a height of 6 metersthe lamp-posts at a height of 6 meters

Scenario

Distance between stations

Fraction of stations that are wired

1 50m 1.0

2 50m 0.75

3 50m 0.50

4 75m 1.0

5 75m 0.75

6 75m 0.50

7 150m 1.0

8 150m 0.75

9 300m 1.0

Simulation ParametersSimulation Parameters

Propagation modelPropagation model– Realistic 3D Beamtracing is usedRealistic 3D Beamtracing is used– The pathloss trace file for the simulation The pathloss trace file for the simulation

is obtained from the mobility scenariois obtained from the mobility scenario– This pathloss trace file is used by This pathloss trace file is used by

Qualnet as a pathloss modelQualnet as a pathloss model– The transmissions are assumed to at The transmissions are assumed to at

15dBm15dBm

Simulation ScenariosSimulation Scenarios

Simulations were conducted in all the four Simulations were conducted in all the four different urban settingsdifferent urban settings

Each scenario was simulated for 100 trial Each scenario was simulated for 100 trial points and averagedpoints and averaged

The AODV protocol in Qualnet simulator The AODV protocol in Qualnet simulator was modified to operate in both wired and was modified to operate in both wired and wireless environmentswireless environments– The routing protocol was modified in such a The routing protocol was modified in such a

way that wired links always had lower cost way that wired links always had lower cost than the wireless links (both for fixed wireless than the wireless links (both for fixed wireless relays and mobile relay nodes)relays and mobile relay nodes)

CoverageCoverage We examine the fraction of the mobile nodes that are able to We examine the fraction of the mobile nodes that are able to

communicate with the wired networkcommunicate with the wired network The coverage is complicated by the fact that buildings are able to reflect, The coverage is complicated by the fact that buildings are able to reflect,

diffract and transmit the signaldiffract and transmit the signal Density of the infrastructure nodes play an important role in the coverageDensity of the infrastructure nodes play an important role in the coverage

– In this study we evaluate the coverage of the mesh networks in presence of In this study we evaluate the coverage of the mesh networks in presence of infrastructure nodes of varying densitiesinfrastructure nodes of varying densities

Another factor that affects the coverage is the fact that mobile nodes can Another factor that affects the coverage is the fact that mobile nodes can act as relays or notact as relays or not– This has a significant impact when the density of the infrastructure nodes is This has a significant impact when the density of the infrastructure nodes is

sparsesparse The physical layer model used also impacts the coverageThe physical layer model used also impacts the coverage

– The higher the bit rate, lower the coverage (higher bit rates require less losses)The higher the bit rate, lower the coverage (higher bit rates require less losses) The position of the nodes (indoor/outdoor) has a significant impact on the The position of the nodes (indoor/outdoor) has a significant impact on the

coveragecoverage– The outdoor nodes can communicate fairly easily with the infrastructure nodesThe outdoor nodes can communicate fairly easily with the infrastructure nodes– The indoor nodes cannot communicate easily with the infrastructure nodes due The indoor nodes cannot communicate easily with the infrastructure nodes due

to the fact that signal has to penetrate through the exterior wall and a lot more to the fact that signal has to penetrate through the exterior wall and a lot more interior wallsinterior walls

CoverageCoverage PerformancePerformance

50 75 150 3000

0.5

1

50 75 150 3000

0.5

1

50 75 150 3000

0.5

1

50 75 150 3000

0.5

1

50 75 150 3000

0.5

1

50 75 150 3000

0.5

1

50 75 150 3000

0.5

1

50 75 150 3000

0.5

1

50 75 150 3000

0.5

1

Outside Inside Any

Fra

ctio

n co

nnec

ted

at 5

4Mbp

s –

(-69

dBm

)

Fra

ctio

n co

nnec

ted

at 1

1Mbp

s

(-85

dBm

)

Fra

ctio

n co

nnec

ted

at 1

Mbp

s –

(-93

dB

m)

Distance between stations (m)

any number of mobile hops no mobile hops

Achievable Bit RatesAchievable Bit Rates The maximum achievable bit rate of a single The maximum achievable bit rate of a single

flow is given by the maximum achievable bit flow is given by the maximum achievable bit rate of the weakest linkrate of the weakest link

The achievable bit rate increases as the The achievable bit rate increases as the density of the infrastructure nodes increasedensity of the infrastructure nodes increase

1 2 3 45 6 7 8 90

5

10

15

x 106

outside

1 2 3 4 5 6 7 8 9

inside

1 2 3 4 5 6 7 8 9

any

Scenario Scenario Scenario

Bit

s/S

econ

d

Application PerformanceApplication Performance The important aspect of this performance study is The important aspect of this performance study is

to see whether the coverage and achievable bit to see whether the coverage and achievable bit rates translate into application performancerates translate into application performance– Even though the channels allow for good bit rate, that Even though the channels allow for good bit rate, that

link might not be used by the routing protocollink might not be used by the routing protocol– Even if the route is found, the end-to-end performance Even if the route is found, the end-to-end performance

might still not support a particular application (e.g. TCP might still not support a particular application (e.g. TCP requires the loss probability to be less that at least 10%)requires the loss probability to be less that at least 10%)

Three popular web applications are consideredThree popular web applications are considered– Web like file transferWeb like file transfer– Voice over IPVoice over IP– Streaming MediaStreaming Media

These three applications have different These three applications have different performance requirementsperformance requirements

Web Like File TransferWeb Like File Transfer Simulations assume that the file sizes are distributed Simulations assume that the file sizes are distributed

according to the log-normal distributionaccording to the log-normal distribution

The first thing to consider as a performance criteria The first thing to consider as a performance criteria is the fraction of successful file transferis the fraction of successful file transfer

1 2 3 4 5 6 7 8 9Scenario

1 2 3 4 5 6 7 8 9Scenario

Inside Outside

0

0.5

1

Fraction of TCP connections completed

Fra

ctio

n

Web Like File TransferWeb Like File Transfer By looking at the coverage plot and the previous By looking at the coverage plot and the previous

plot, we find that the fraction of successful file plot, we find that the fraction of successful file transfer is highly correlated to the coverage transfer is highly correlated to the coverage (Correlation Coefficient = 0.84)(Correlation Coefficient = 0.84)

The fraction of nodes that can communicate with The fraction of nodes that can communicate with the wired network is 25% more that the fraction of the wired network is 25% more that the fraction of successful file transfersuccessful file transfer

Also the fraction of successful file transfer without Also the fraction of successful file transfer without the help of mobile relays is 15% more than the the help of mobile relays is 15% more than the fraction of nodes that can connect to the wired fraction of nodes that can connect to the wired networknetwork

Thus it seems that the availability of mobile nodes Thus it seems that the availability of mobile nodes acting as relay is not that helpful in this scenario acting as relay is not that helpful in this scenario

Web Like File TransferWeb Like File Transfer The next factor to consider is the relation The next factor to consider is the relation

between the average number of route changes between the average number of route changes and the number of mobile hops in the flowand the number of mobile hops in the flow

0

0.2

0.4

0.6

0.8 Inside Outside

123456 78 9Scenario

12 3 4 56 7 8 9Scenario

Num

ber

Average number of route changes

Inside

1 Hop2 Hops 3 Hops

12 3456 78 9Scenario

0

0.5

1

Fra

ctio

n

Outside

12 3456 78 9Scenario

Number of mobile hops in the route

As the number of mobile hops increase the average number As the number of mobile hops increase the average number of route changes increaseof route changes increase

The correlation coefficient was found to be -0.81The correlation coefficient was found to be -0.81 This is one of the biggest drawbacks of having the mobile This is one of the biggest drawbacks of having the mobile

nodes as relaysnodes as relays

Voice over IPVoice over IP

Hand held VoIP phones are already Hand held VoIP phones are already available in the market available in the market

Ubiquitous internet access could Ubiquitous internet access could provide low cost mobile phone service provide low cost mobile phone service provided the quality is sufficiently highprovided the quality is sufficiently high

The most important factors to consider The most important factors to consider while evaluating the VoIP performance while evaluating the VoIP performance areare– MOS : Mean Opinion ScoreMOS : Mean Opinion Score– Time to establish the connectionTime to establish the connection

Voice over IPVoice over IP

The above figure shows the fraction of calls that were good, that were bad The above figure shows the fraction of calls that were good, that were bad and calls that did not completeand calls that did not complete

The calls that took more than 10 seconds to establish were termed as failed The calls that took more than 10 seconds to establish were termed as failed callscalls

The Calls that had a delay of more than 5s or an MOS value < 3.6 were The Calls that had a delay of more than 5s or an MOS value < 3.6 were termed bad callstermed bad calls

It is interesting to note that in scenario 1 and 2, the indoor nodes tend to It is interesting to note that in scenario 1 and 2, the indoor nodes tend to perform better than the outdoor nodesperform better than the outdoor nodes– This is due to the fact that indoor nodes do not have significant mobility. Hence This is due to the fact that indoor nodes do not have significant mobility. Hence

once the call gets established it tends to have a good quality in contrast to the once the call gets established it tends to have a good quality in contrast to the outdoor nodes which are always mobileoutdoor nodes which are always mobile

It is also interesting to note that with decreasing infrastructure density, the It is also interesting to note that with decreasing infrastructure density, the outdoor nodes establish the calls, but are unable to maintain a steady outdoor nodes establish the calls, but are unable to maintain a steady connection. (Fraction of bad calls increase)connection. (Fraction of bad calls increase)

1 2 3 4 5 6 7 8 90

0.5

1Inside

Scenario

Good CallsBad CallsFailed Calls

Scenario

Outside

Performance of VoIP

Frac

tion

1 2 3 4 5 6 7 8 9

Streaming MediaStreaming Media

Another possible applicationAnother possible application Unlike the VoIP, streaming media Unlike the VoIP, streaming media

applications can tolerate significant delays applications can tolerate significant delays provided the receiver buffer is big enoughprovided the receiver buffer is big enough

Another difference is that streaming music Another difference is that streaming music requires more bandwidth than the VoIP requires more bandwidth than the VoIP applicationsapplications

Using this application we try to evaluate Using this application we try to evaluate the system with respect to the network the system with respect to the network outagesoutages

Streaming MediaStreaming Media

The above diagram shows the fraction of connections undergoing The above diagram shows the fraction of connections undergoing outagesoutages

It is interesting to note that the fraction of connections that It is interesting to note that the fraction of connections that undergo outages between 5 and 20 seconds is very smallundergo outages between 5 and 20 seconds is very small– Hence increasing the receiver buffer capacity from 5s to 20s is not Hence increasing the receiver buffer capacity from 5s to 20s is not

going to have a significant impact on the delay tolerancegoing to have a significant impact on the delay tolerance Figure also shows that a significant fraction of the connections Figure also shows that a significant fraction of the connections

underwent outages more than 20sunderwent outages more than 20s– From application perspective even these connections should be From application perspective even these connections should be

considered as failed connectionsconsidered as failed connections

Outage < 5Sec5Sec Outage < 20SecOutage 20SecFailed Connections

0

0.5

1 Inside Outside

1 2 3 4 5 6 7 8 9Scenario

1 2 3 4 5 6 7 8 9Scenario

Frac

tion

Fraction of connections undergoing outages

Time To Establish RouteTime To Establish Route

This is one of the performance measures This is one of the performance measures that affects all the applicationsthat affects all the applications

0

5

10

15

Tim

e in

Sec

onds

Inside Outside

Time taken to establish route

1 2 3 4 5 6 7 8 9Scenario

1 2 3 4 5 6 7 8 9Scenario

The call establishment time cannot be greater than 10s for VoIPThe call establishment time cannot be greater than 10s for VoIP If the connection establishment time is too high, TCP connection times outIf the connection establishment time is too high, TCP connection times out Intermittent route failures and subsequent route re-establishments taking Intermittent route failures and subsequent route re-establishments taking

long time could render a streaming connection or VoIP application as long time could render a streaming connection or VoIP application as failedfailed

ConclusionsConclusions The above results indicated some of the challenges facing the The above results indicated some of the challenges facing the

ubiquitous internet access via mesh networksubiquitous internet access via mesh networks Philadelphia has plans of having infrastructure nodes separated by Philadelphia has plans of having infrastructure nodes separated by

300m300m– The above results indicate that most of the applications wont perform well The above results indicate that most of the applications wont perform well

in such a scenarioin such a scenario The coverage and the bit rate plots most of the time contradicts the The coverage and the bit rate plots most of the time contradicts the

application performanceapplication performance– Hence coverage alone cannot be taken as a good indicator of performanceHence coverage alone cannot be taken as a good indicator of performance

Under low infrastructure densities, having the mobile nodes act as Under low infrastructure densities, having the mobile nodes act as relays increases the coverage substantially e.g. most of the cases it relays increases the coverage substantially e.g. most of the cases it increases by 100% in scenario 9increases by 100% in scenario 9

Even though 90% of the nodes outside are connected, only 50%-60% Even though 90% of the nodes outside are connected, only 50%-60% of the indoor nodes are connected in cases where the separation of the indoor nodes are connected in cases where the separation distance >= 75mdistance >= 75m

While a slightly modified version of AODV was used for the simulations, While a slightly modified version of AODV was used for the simulations, more research needs to be put into the routing protocols for these more research needs to be put into the routing protocols for these networksnetworks

MIMO, directional antennas, Co-operative networking are all promise to MIMO, directional antennas, Co-operative networking are all promise to increase the coverageincrease the coverage

ReferencesReferences [1] S. Cass, “Viva mesh vegas,” IEEE Spectrum, 2005. [2] “Zyxel 802.11 VoIP phone

"http://www.zyxel.com/product/P2000W.php".” [3] “Vipernet wi-fi VoIP phone

"http://www.vipernetworks.com/".” [4] D. Neff, “Wireless philadelphia town hall meeting,”

November 2004, available at http://207.245.67.199/philagotit/wifilive2.wmv.

[5] A. Motley and J. Keenan, “Radio coverage in buildings,” britsh telecom technical journal, vol. 8, pp.19—24, 1990.

[6] S. Y. Seidel and T. S. Rappaport, “A ray tracing technique to predict path loss and delay spread inside buildings,” in IEEE GLOBECOM, 1992, pp. 1825—1829.

[7] J. Kim and S. Bohacek, “A survey-based mobilty model of people for simulation of urban mesh networks,” in Submitted, 2005, available at http://www.eecis.udel.edu/ bohacek.

ReferencesReferences [8] H. Bertoni, Radio Propagation for Modern Wireless

Systems. Prentice Hall, December 1999. 17 [9] G. Liang and H. L. Bertoni, “A new approach to 3-d ray

tracing for propagation prediction in citiee,” IEEE transaction on Antennas and propagation, vol. 46, pp. 853—863, 1998.

[10] V. Sridhara and S. Bohacek, “Propagation modeling for urban mesh networks,” in Submitted, 2005.

[11] D. A. McNamara, C.W. Pistorius, and J. A. G. Malherbee, Introduction to Uniform Geometrical Theory of Diffraction. Artech House, 1990.

[12] S. Ray, “Realistic mobility for MANET simulation,” Master’s thesis, The University of British Columbia, 2003.

[13] D. Batacharjee, A. Rao, C. Shah, M. Shah, and A. Helmy, “Empirical modeling of campus-wide pedestrian mobility: Observation on the USC campus,” in IEEE Vehicular Technology Conference (VTC), 2004.

ReferencesReferences [14] M. McNett and G. M. Voelker, “Access and mobility of

wireless PDA users,” UC San Diego, Tech. Rep. CS2004-0780, 2004.

[15] B. Pushkarev and J. M. Zupan, Urban Space for Pedestrians. MIT press, 1975.

[16] Transportation Research Board, 2000 Highway Capacity Manual. Washington, D.C.: National Research Council, 2000.

[17] N. C. Romano and J. F. Numamaker, “Meeting analysis: Findings from research and practice,” in Proceedings of Teh 34th Hawaii International Conference on Systems Science, 2001.

[18] A. Szalai, The Use of Time. the hague: mouton, 1972. [19] U. D. of Labor Bureau of Labor Statistics, “American time use

survey (ATUS),” 2003, http://www.bls.gov/tus/. [20] R. R. Panko and S. T. Kinney, “Meeting profiles: Size, duration,

and location,” in Proceedings of The 28th Annual Hawaii International Conference on Systems Science, 1995.

[21] R. K. Mosvick and R. B. Nelson, We’ve Got to Start Meetings Like This! Glenview, Illinois: Scott, Foresman and company, 1987.

ReferencesReferences [22] S. P. Hoogendoorn and P. H. L. Bovy, “Pedestriabn route-

choice andb activity scheduling theory and models,” transportation research B, 2003.

[23] S. P. Hoogendoorn, M. Hauser, and N. Rodrigues, “Application of microscopic pedestrian flow simulation to station design evaluation in Lisbon train stations,” in TRB 2004 Annual Meeting, 2004.18

[24] M. Wist, “Office space - how much is enough?” 2001, available at http://www.geraldeve.com/asp/uploads/uploadedfiles/1/579/oodsurvey2001.pdf.

[25] “Http://www.scalable-networks.com/,” Tech. Rep. [26] A. B. Downey, “The structural cause of file size

distributions,” in Proceedings of the Ninth International Symposium in Modeling, Analysis and Simulation of Computer and Telecommunication Systems (MASCOTS’01), 2001.

[27] A. P. Markopoulou, F. A. Tobagi, and M. J. Karam, “Assessing the quality of voice communications over Internet backbones,” IEEE/ACM transactions on networking, vol. 11, pp. 747—760, 2003.

Thank YouThank You

Performance evaluation of Urban Mesh Networks Vinay Sridhara Department of Electrical Engineering University of Delaware.

Documents

realistic propagation

realistic mobility model

infrastructure nodes

wireless simulations

urban area random cities

massive wireless hotspots

relays las vegas las

kind of networks