human aspects and heterogeneity in...

HUMAN ASPECTS AND HETEROGENEITY IN

MISSION-ORIENTED OPPORTUNISTIC NETWORKS

Barun Kumar Saha

HUMAN ASPECTS AND HETEROGENEITY IN

MISSION-ORIENTED OPPORTUNISTIC NETWORKS

Thesis submitted to the

Indian Institute of Technology Kharagpur

for award of the degree

of

Master of Science (by Research)

by

Barun Kumar Saha

Under the guidance of

Dr. Sudip Misra and Dr. Debasis Samanta

School of Information Technology


Kharagpur - 721 302, India

January 2014

c© 2014 Barun Kumar Saha. All rights reserved.

CERTIFICATE

Date: 28/01/2014

This is to certify that the thesis entitled Human Aspects and Heterogeneity in

Mission-Oriented Opportunistic Networks, submitted by Barun Kumar Saha

to Indian Institute of Technology Kharagpur, is a record of bonafide research work under

my supervision and I consider it worthy of consideration for the award of the degree of

Master of Science (by Research) of the Institute.

Dr. Sudip Misra

Associate Professor




Dr. Debasis Samanta

Associate Professor




DECLARATION

I certify that

a. The work contained in the thesis is original and has been done by myself under

the general supervision of my supervisor.

b. The work has not been submitted to any other Institute for any degree or diploma.

c. I have followed the guidelines provided by the Institute in writing the thesis.

d. I have conformed to the norms and guidelines given in the Ethical Code of Conduct

of the Institute.

e. Whenever I have used materials (data, theoretical analysis, and text) from other

sources, I have given due credit to them by citing them in the text of the thesis

and giving their details in the references.

f. Whenever I have quoted written materials from other sources, I have put them

under quotation marks and given due credit to the sources by citing them and

giving required details in the references.

Barun Kumar Saha

Dedicated to

My parents, Dada, Boudi,

and my beloved nephews – Akash and Arush

ACKNOWLEDGMENT

“Art is — representing the beautiful. There must be Art in

everything.”

— Swami Vivekananda

It is said that life’s a journey, and that the journey is as important as the destination

itself. This leg of my journey – the stay at Indian Institute of Technology Kharagpur

for post-graduate studies – has been wonderful and would remain memorable, thanks to

several amazing people that I have came across! Indeed, I would eternally remain in-

debted to them for their endless support, help, advice, guidance, patience and tolerance.

I sincerely thank you all!

I would like to express my deepest gratitude to my supervisor, Dr. Sudip Misra,

who has constantly motivated and encouraged me all the while along this journey. Dr.

Misra’s name would remain imprinted in every aspect of my research work. He not only

guided me in identifying and defining problems, but also how to approach their solutions

and their subsequent presentation in the form of quality research articles. And above

all, he always inspired me to think big.

I would like to sincerely thank my co-supervisor, Dr. Debasis Samanta. His kind

advices on all aspects of work and life have been truly helpful. I would also like to

take this opportunity to thank Dr. Samanta, as well as Dr. Misra, for giving me the

opportunity to work in the Virtual Labs projects alongside my research work.

I am indebted to Prof. Rajib Mall and Prof. Jayanta Mukhopadhyay for their advices

and suggestions, which, indeed, have helped me to enrich and refine this work. I would

like to thank Dr. Indrajit Chakrabarty, Prof. Indranil Sen Gupta and Dr. K. S. Rao

for kindly agreeing to serve on my committee. I am grateful for their precious time and

advices, which helped in better shaping this work.

I would also like to sincerely thank Prof. Shamik Sural, Prof. Chittaranjan Mandal,

Prof. Arobinda Gupta and Dr. S. K. Ghosh for their guidance and inputs all through

these years. Thanks to Dr. Rajiv Ranjan Sahay for the enthusiasm and motivation

provided while carrying out this work.

Thanks to Mrs. Soma Paine, Mithun da, Goutam da, Malay da and Pratap da for

their continuous help and support in every aspect of my work here. I would also fondly

vii

remember the tea breaks at Utpal da’s tea shop, which often provided food for thought.

I would like to thank all the members of the Projects Lab and the students and

researchers of the School of Information Technology who have influenced my life and

to whom all I am indebted in one way or another. A few names, however, deserve

special reference. Life would not have been as easy without the guidance that I have

always received from Mr. Manas Khatua, who, indeed, may be compared to a living

encyclopedia of knowledge! Thanks to Mrs. Sujata Pal for her constant motivation.

Indeed, the way she juggles her family and research life everyday, is a true source of

inspiration. I thank her for asking me a multitude of questions – and providing critical

suggestions – which helped me in several dimensions of my research. I would like to

thank Mr. Amit Kumar Mandal for his support; I would very much miss the fun that

we had during our occasional gatherings at his place.

I express my gratitude to Mr. Sankar Narayan Das, who, on several occasions,

gave a patient hearing to the presentation of my work and provided me with valuable

suggestions. Mr. Bibudhendu Pati’s dedication to work has been inspirational, and

am thankful for his support. In these years, I have spent some quality time with Mr.

Goutam Mali, Mr. Judhistir Mahapatro and Mr. Nabiul Islam, with whom I often had

prolonged technical discussions. Thanks to Mr. Sayan Sarcar and Mr. Pradipta Kumar

Saha for their kind help and support. I would also like to express my thanks to Mr.

Tamoghna Ojha for his continuous help and support. His profound knowledge in the

matters of policy and process made achieving the milestones of this journey easier. I

would like to thank Mr. Soumalya Ghosh, who has been a source of motivation in both

life and work. Also, thanks to Mr. Tanay Chaki, with whom I have spent here some of

the most interesting times of my life.

I would not have reached at this position in life if it were not for the selfless love,

never-ending support, and sacrifices made by my parents. Although no amount of “thank

you” notes can ever be enough for their hard efforts and contributions, I would still like

to sincerely thank Maa and Baba for everything they have given. Thank you! I would

also like to thank my Dada and Boudi for their constant encouragement and motivation.

Finally, thanks to my sweet, little and beloved nephews, Akash and Arush, who remind

me to live your life everyday without any worries.

Barun Kumar Saha

viii

Abstract

Pocket Switched Networks (PSNs) are formed among mobile devices (for example,

smartphones and PDAs) carried by human beings. In the absence of any network infras-

tructure, the devices communicate opportunistically when they come within the trans-

mission ranges of one another. While the areas of Psychology and Social Sciences study

the factors affecting the different aspects related to human beings, the possible impact

of such human factors on the network dynamics has been largely overlooked. On the

other hand, the existing works on PSNs largely consider homogeneous network compo-

sition. While network heterogeneity is supposed to degrade network performance, it is

not apparent how the behavior would be when certain human aspects are considered

along with it.

To address the above issues, we propose Mission-Oriented Opportunistic Networks

(MOONs) as an extension of PSNs, by considering the mission aspects together with

human factors. The presence of human beings – the owners of the devices – as part of

the network, adds a new dimension to the MOONs, and, thus, are expected to affect the

network dynamics.

The objective of this thesis is to study the effects of human aspects and heterogeneity

on the performance of the MOONs. We define four levels of intelligence as the manifes-

tation of the intuitive decision making found in human beings. We considered the mobile

nodes – the devices with their owners – to occasionally select a destination based on their

level of intelligence, and move accordingly in order to maximize the accomplishment of

the mission objectives.

We also studied another aspect of human-network interplay in MOONs, by consid-

ering the effects of human emotions on the network operations using Meftah et al.’s

computational model of emotions. We represented the temporal variation in the traffic

generation rate and degree of “cooperation” of the users in a MOON as a function of

their contemporary emotions. Moreover, we linked the current state of emotion of the

users to the information (messages) received by them within a time period.

Finally, we considered the heterogeneities in MOONs arising due to various aspects

in hardware and software of the devices. We also studied the effects of heterogeneity

together with the scenarios described in the previous two aspects.

The results of performance evaluation show that while intelligence-induced move-

ix

ment can help in accomplishing the mission objectives, the effects of emotions in certain

scenarios (for example, post-disaster) can deteriorate the network performance. Further

degradation can be observed when heterogeneity in human aspects and the network are

considered together. Such degradation can, however, be limited by the use of throwboxes

and possibly other “bridging” devices.

Keywords: Pocket Switched Networks, Delay Tolerant Networks, Post-disaster sce-

narios, Plutchik’s circumplex model of emotions, Human emotions, Human intelligence,

Heterogeneity, Routing incompatibility

x

Contents

Certificate i

Declaration iii

Dedication v

Acknowledgment vii

Abstract ix

Contents xi

List of Figures xv

List of Tables xvii

List of Algorithms xix

List of Symbols and Abbreviations xxi

1 Introduction 1

1.1 Evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Motivation and Scope of Work . . . . . . . . . . . . . . . . . . . . . . . . 4

1.3 Problem Statement and Objectives . . . . . . . . . . . . . . . . . . . . . . 6

1.4 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.5 Organization of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Related Work 11

2.1 Mission-Oriented Networks and Participatory Sensing . . . . . . . . . . . 12

2.2 Human Aspects and Emotions . . . . . . . . . . . . . . . . . . . . . . . . 15

xi

Contents

2.3 Post-disaster Rescue Operations . . . . . . . . . . . . . . . . . . . . . . . 17

2.4 Network Heterogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3 Effects of Intelligence-induced Movement in MOONs 27

3.1 Mission Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.2 Opportunistic Communications with Intelligence . . . . . . . . . . . . . . 29

3.3 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33


4 Aspects of Human Emotion in MOONs 39

4.1 Representing the Effects of Emotion . . . . . . . . . . . . . . . . . . . . . 40

4.1.1 The Computational Model of Emotions . . . . . . . . . . . . . . . 41

4.1.2 Effects of Emotions on the Network Dynamics . . . . . . . . . . . 43

4.2 Application Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

4.2.1 Variation in Emotion . . . . . . . . . . . . . . . . . . . . . . . . . . 46

4.2.2 Variation in Traffic Load . . . . . . . . . . . . . . . . . . . . . . . . 47

4.2.3 Changes in User Cooperation . . . . . . . . . . . . . . . . . . . . . 49

4.3 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53


5 Effects of Heterogeneity 65

5.1 Aspects of Heterogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

5.1.1 Heterogeneity in Connection Dynamics . . . . . . . . . . . . . . . . 67

5.1.2 Diverse Hardware of the Devices . . . . . . . . . . . . . . . . . . . 68

5.1.3 Routing Protocols in DTN and their Compatibility . . . . . . . . . 68

5.1.4 Effects of Incompatibilities . . . . . . . . . . . . . . . . . . . . . . 70

5.2 Representation of Heterogeneous PSNs . . . . . . . . . . . . . . . . . . . . 71

5.3 Overcoming the Adverse Effects of Heterogeneity . . . . . . . . . . . . . . 72

5.3.1 Hardware Incompatibility . . . . . . . . . . . . . . . . . . . . . . . 72

5.3.2 Protocol Translation Units . . . . . . . . . . . . . . . . . . . . . . 73

5.4 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

5.4.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . 76

5.4.2 Performance Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . 78

5.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

xii

Contents

5.5.1 Effects of Heterogeneous Connection Events . . . . . . . . . . . . . 79

5.5.2 Effects of Buffer Size . . . . . . . . . . . . . . . . . . . . . . . . . . 80

5.5.3 Impact of Incompatible Networking Devices . . . . . . . . . . . . . 84

5.5.4 Effects of Heterogeneous Routing Protocols . . . . . . . . . . . . . 85

5.6 Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88


6 Conclusion 91

6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

6.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

6.3 Limitations and Future Scope of Work . . . . . . . . . . . . . . . . . . . . 95

Publications 97

References 99

xiii

List of Figures

1.1 Communication in traditional networks and DTNs . . . . . . . . . . . . . 2

1.2 Supernodes in MOONs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

3.1 A node’s behavior in a MOON . . . . . . . . . . . . . . . . . . . . . . . . 31

3.2 Encounters of mobile nodes with stationary nodes for 12 hours . . . . . . 34

3.3 Encounters of mobile nodes with stationary nodes for 24 hours . . . . . . 34

3.4 Change (%) in encounters with clustered static nodes . . . . . . . . . . . . 35

3.5 Performance using different intelligence levels . . . . . . . . . . . . . . . . 36

3.6 Performance when some nodes had the application available . . . . . . . . 36

4.1 Plutchik’s circumplex model of emotions . . . . . . . . . . . . . . . . . . . 41

4.2 Fluctuation in traffic rate based on the dominant emotion of the users . . 54

4.3 Variation in traffic rate based on amplification factor . . . . . . . . . . . . 55

4.4 Effects of emotion-induced variation in cooperation . . . . . . . . . . . . . 57

4.5 Message delivery considering variation in both traffic and cooperation . . 58

4.6 Effect of radio off probability on message delivery ratio . . . . . . . . . . . 59

4.7 Temporal fluctuations in the dominant emotions . . . . . . . . . . . . . . 60

4.8 CDF of the contacts among the devices . . . . . . . . . . . . . . . . . . . 61

4.9 Performance degradation considering emotion and heterogeneity . . . . . 62

5.1 Communication impairments in heterogeneous PSNs . . . . . . . . . . . . 67

5.2 Interactions among different types of routing protocols and PTUs . . . . . 73

5.3 Effects of (a)symmetric connection events . . . . . . . . . . . . . . . . . . 80

5.4 Effects of buffer size on the performance of SnW . . . . . . . . . . . . . . 81

5.5 Effects of buffer sizes on other routing protocols . . . . . . . . . . . . . . . 83

5.6 Effects of variable buffer sizes on different routing protocols . . . . . . . . 84

5.7 Message delivery ratio considering energy constraints . . . . . . . . . . . . 85

5.8 Effect of heterogeneous networking interfaces . . . . . . . . . . . . . . . . 86

xv

List of Figures

5.9 Message delivery ratio considering heterogeneous routing protocols . . . . 87

5.10 Communication degree of the MOON . . . . . . . . . . . . . . . . . . . . 88

xvi

List of Tables

2.1 Summary of works on sensor networks and missions . . . . . . . . . . . . 14

2.2 Summary of works on human emotion detection . . . . . . . . . . . . . . . 18

2.3 Summary of works on post-disaster mobility and communication . . . . . 21

2.4 Summary of works on network heterogeneity . . . . . . . . . . . . . . . . 23

xvii

List of Algorithms

3.1 Purposeful mobility under the L2 scheme . . . . . . . . . . . . . . . . . . . 32

4.1 Actions taken when dominant emotion is fear . . . . . . . . . . . . . . . . . 50

5.1 Interaction of the PTUs with PROPHET routers . . . . . . . . . . . . . . . 75

5.2 Interaction of the PTUs with SnW routers . . . . . . . . . . . . . . . . . . 77

xix

List of Symbols and

Abbreviations

List of Symbols

a Vector of decay coefficients of the basic emotions

B Basis vector representing the basic emotions

DE Dominant emotion

e Intensity vector of the basic emotions

G Set of mission goal(s)

I Function capturing effects of a set of messages upon a basic

emotion

if1 Network interface of type 1

k(t) Degree of user cooperation at time t

L Number of copies of a message used in Spray-and-Wait protocol

M Set of messages generated in the network

M Vector of collection of messages affecting the individual basic

emotions

M A mission

N A Mission-Oriented Opportunistic Network

P Set of mission performance parameters

pm Vector of different categories of messages generated

R(t) Traffic generation rate at time t

xxi

List of Symbols and Abbreviations

r Vector representing the effect of the received messages upon the

basic emotions

S Set of mission strategies

T̄ Average time duration between creation of two messages

Toff Maximum time duration for which a device radio can remain

switched off

T Mission time line

α Communication degree of the MOON

β Vector of amplification factors in the traffic load

γ Vector of degree of user cooperation

λ Poisson rate of message generation

τ Temporal decay function of the basic emotions

∅N Neutral emotion

List of Abbreviations

CDF Cumulative Distribution Function

DTN Delay Tolerant Network

IP Internet Protocol

MOON Mission-Oriented Opportunistic Network

PDA Personal Digital Assistant

PROPHET Probabilistic Routing Using History of Encounters and Transi-

tivity

PSN Pocket Switched Network

PTU Protocol Translation Unit

RWP Random Waypoint

SMS Short Message Service

SnW Spray-and-Wait

xxii

List of Symbols and Abbreviations

TCP Transmission Control Protocol

TTL Time to Live

TVG Time Varying Graph

WSN Wireless Sensor Network

xxiii

Chapter 1

Introduction

The ubiquitous Internet, based on the TCP/IP stack, fails to work in environments that

exhibit certain characteristics [1], for example:

• Intermittent connectivity among the nodes, which results in the lack of end-to-end

communication paths.

• High or variable latencies along the communication path.

• Asymmetric connections or link bandwidths between a source and a destination.

• High error rates along the transmission media.

To cope with such characteristics, the delay-tolerant networking architecture was

proposed [2]. Unlike the traditional Internet, Delay Tolerant Networks (DTNs) use

the store-carry-and-forward strategy for message transfer in order to account for the

intermittent connectivities. In this scenario, a source node with a message – until it meets

with the destination node of the message – replicates and forward one or more copies of

the message to the other intermediate node(s) it encounters with. These intermediate

node(s) store a copy of the message in their buffer, carry them as they move along

their way, and, in turn, replicate and forward (or deliver) to the other intermediate

1

1. Introduction

(or destination) node(s) it comes in contact with. Figure 1.1 illustrates the difference

between routing in traditional networks and the DTNs.

(a) (b)

Figure 1.1: (a) End-to-end communication paradigm in the traditional networks. On theother hand, (b) in DTNs, messages are replicated and forwarded repeatedly over time inorder to increase its chances of delivery. Here, S and D, respectively, denote the source anddestination nodes; the remaining nodes act as relays.

1.1 Evolution

Pocket Switched Networks (PSNs) [3] are a prominent subclass of the DTNs. A PSN

is formed by the portable devices such as, smart phones and PDAs, carried by human

beings. The devices in the PSNs communicate among themselves using network infras-

tructure, for example, Wi-Fi, if available. In the absence of such infrastructure, the

devices communicate opportunistically when they are in the transmission range of one

another.

Modern generation smart phones, which are equipped with one or more sensors (for

example, accelerometer and gyroscope), are increasingly becoming part of the everyday

life. Therefore, the integration of human beings in mission-oriented networks [4] with

PSNs is inevitable. This hybridization of PSNs with mission-oriented networks induces

interest in a new class of networks – Mission-Oriented Opportunistic Networks (MOONs),

which are primarily characterized by:

2

1.1. Evolution

• Mission objectives: MOONs are often associated with some mission objectives,

for example, providing prompt communication facilities in the aftermath of a large-

scale disaster.

• Opportunistic contacts: Whenever two or more nodes come in the commu-

nication range of one another, they exchange available information. In case no

communication opportunity is available at an instant of time, the nodes store the

messages in their buffer (for possibly long time).

• Human-network interplay: MOONs are not just networks where automated

devices communicate among themselves. Rather, the presence of human beings –

the owners of the devices – as part of the network, adds a new dimension. In other

words, unlike the conventional viewpoint, in these networks, the human beings are

not merely the end users who send and receive messages, and, thus, are expected

to affect the network dynamics. This combination of humans and their devices

gives rise to a supernode-like entity (Figure 1.2).

Figure 1.2: Communication in a MOON. In MOONs, the communication end pointsare not just mechanical nodes. Rather, they are the portable devices carried by humanbeings. Consequently, various human aspects could influence the network communicationmechanism.

The rest of this Chapter is organized as follows.

3

1. Introduction

1.2 Motivation and Scope of Work

The proliferation of mobile devices in the present society has opened up several new

possibilities and research opportunities – participatory [5] and human-centric [6] sensing

are some of them. This raises an interesting question – could mission objectives be

achieved through opportunistic communication among the mobile devices carried – and

possibly influenced – by their human owners?

While the areas of Psychology and Social Sciences study the factors affecting the

different aspects related to human beings, the possible impact of such human factors

on the network dynamics, and thus, mission objectives, has been largely overlooked.

This motivates us to study the effects, if any, of the other human aspects, for example,

emotion and intelligence, on the communication networks. The only human characteris-

tic that has received significant attention in networking research is the human mobility

pattern [7–9], which, however, is considered as an independent parameter. Thus, it re-

mains unexplored how innate human intelligence affect their movement destinations and,

consequently, the mission objectives. A study of this aspect would underscore the feasi-

bility of accomplishment of the mission objectives through intelligence-induced human

mobility.

Our motivation is further strengthened by the fact that, in today’s age, human beings

spend considerable time with their mobile devices giving rise to a form of human-device

relationship [10, 11]. Existing research works in Psychology (for example, [12–14]) have

considered the relation between human emotions and actions. Zhu and Thagard [13]

have shown that the emotions of the human beings considerably affect their actions.

Such an inference bears significance in MOONs, where the presence of human beings is

integrally inherent to the network. In particular, Tang et al. [14] reported that users’

likelihood of calling or sending SMS over phone is related to her emotional state with the

highest chance of calling while being in a positive state. This motivates us to consider

the effects of human emotions on the dynamics of MOONs, and vice versa.

4

1.2. Motivation and Scope of Work

It may be noted that traffic explosion, whether emotion-induced or not, is an issue

that may be of interest to researchers working in other types of networks such as cellular.

However, if the emotions of the users in MOONs, which already lack in end-to-end com-

munication paths, affect their engagement in cooperation, the delivery of the messages

would be further affected. This makes the problem interesting to study in the context

of MOONs.

Finally, various aspects such as routing [15–18], security [19] and congestion [20],

related to the PSNs, or DTNs in general, have been explored by the research community.

One of the core issues addressed is routing – how to send messages from the source

nodes to the destination nodes in the face of the constraints, for example, lack of end-

to-end paths, offered by the DTNs. The simplest scheme is the one where the source

node directly delivers the message to the destination node. To reduce the delivery

latency of the messages, and increase the corresponding delivery ratio, several multi-

copy-based routing protocols such as Epidemic [16], Probabilistic Routing Using History

of Encounters and Transitivity (PROPHET) [17] and Spray-and-Wait (SnW) [18] were

explored. In such a scheme, multiple (fixed or variable) copies of a given message are

spread into the network with the intention of reducing the delivery times and increasing

the chances of delivery. In the recent years, the efficiency of the proposed schemes has

been established through simulations with the connection (or location) traces obtained

from real-life experiments, rather than simulating random scenarios. The existing works,

however, have addressed scenarios where the network compositions are homogeneous.

While network heterogeneity is supposed to degrade the network performance, it

is not apparent how the behaviour would be when certain human aspect is considered

along with it. Therefore, this motivates us to study the performance of a heterogeneous

MOON. Further, the concurrent effects of heterogeneity and human aspects on the per-

formance is also investigated. This deserves importance since certain real-life occasions

(for example, a post-disaster scenario) would exhibit strong display of emotions, which,

5

1. Introduction

in turn, affects the ongoing opportunistic network communications.

To this end, we formally define a MOON, and explore how some of the human aspects

and heterogeneity affect the dynamics of such networks. The scope of this Thesis is

outlined below:

• Only two aspects related to human beings are considered here – intelligence and

emotion.

• Modeling human intelligence is very large in scope. Here, we consider only certain

aspects of it, which are largely intuitive and heuristic-based.

• We consider only a specific model of emotion (originally based on Plutchik’s [21]

model), and its effects on the network traffic generation and forwarding actions.

• The types of heterogeneity considered here are among the ones commonly men-

tioned in the related literature. Indeed, in practice, there could be other forms of

heterogeneities – for example, number of CPU cycles required – but, not considered

prominent in the concerned body of work, and thus, scoped out of this work.

1.3 Problem Statement and Objectives

The objective of this Thesis is to study the possible effects of two specific human aspects

– intelligence and emotion – of the carriers of the smart devices upon the performance of

a MOON together with heterogeneity. The problem statement of this Thesis is elicited

as follows:

While intuitive intelligence-based human movement can enhance the mission accom-

plishment in a MOON by selecting potential destinations, the effects of another human

aspect, emotion, on user cooperation and traffic generation rate degrades the network

performance in a typical post-disaster scenario, which further deteriorates due to inher-

6

1.4. Contributions

ent network heterogeneity.

The specific objectives of this work are outlined as follows:

1. Studying the aspects of human emotion in MOONs and their effect on the network

performance in the context of a post-disaster scenario.

2. Exploring the effects of different manifestations of intuitive human intelligence on

the mission objectives of MOONs by considering intelligence-induced movement of

the users.

3. Investigating the effects of heterogeneity – in terms of both hardware and software

of the devices – on the performance of PSNs, and thus, MOONs.

This study is supported by a set of extensive results obtained from simulations con-

ducted using both real-life and synthetic traces wherever suitable.

1.4 Contributions

The major contributions of this Thesis are summarized in the following points.

1. We used Meftah et al.’s [22] computational model of emotions and represented the

temporal variation in the traffic generation rate and degree of “cooperation” of

the users in a MOON as a function on their contemporary emotions. Moreover,

we linked the current state of emotion of the users to the information (messages)

received by them within a time period. Subsequently, we analyzed the performance

of a MOON formed in a post-disaster environment by considering emotions typical

to such scenarios.

2. We defined four levels of intelligence as the manifestation of the intuitive decision

making found in human beings. We considered the mobile nodes – the devices

7

1. Introduction

with their owners – to occasionally select a destination based on their level of

intelligence, and move accordingly to maximize the accomplishment of the mission

objectives.

3. We considered the heterogeneities in PSNs – and consequently, MOONs – arising

due to various aspects in hardware and software of the devices. We evaluated the

performance of such a heterogeneous network by studying the effects of diverse

buffer sizes, asymmetric connections, diverse initial energies and incompatible net-

work interfaces. Moreover, we considered the case of heterogeneous routing pro-

tocols and proposed the use of Protocol Translation Units (PTUs) to bridge two

specific incompatible routing protocols. Finally, we also considered the effects of

heterogeneity, together with the scenarios described in the two previous points.

1.5 Organization of the Thesis

The Thesis consists of 6 chapters including this introductory Chapter. The remainder

of this Thesis is organized as follows.

• Chapter 2: Literature Survey

This Chapter surveys a sample of the relevant state-of-the-art works in the do-

mains of mission-oriented networks, human emotions, post-disaster scenarios, and

network heterogeneity. We study multiple such works and identify their shortcom-

ings when considered in the context of MOONs. To address some of issues arising

due to the knowledge gap, we broadly address three problems, as already discussed

in Section 1.3 in Chapter 1.

• Chapter 3: Effects of Intelligence-induced Movement

The second Chapter formally represents a MOON, and considers the effects of

movement influenced by innate human intelligence in accomplishing the mission

8

1.5. Organization of the Thesis

objectives. To this end, we define four levels of intuitive intelligence, which helps

in deciding the destinations. Subsequently, we evaluate the performance of the

proposed scenarios considering different node densities, velocities, and other het-

erogeneous aspects.

• Chapter 4: Aspects of Human Emotions in MOONs

In this Chapter, we consider another prominent human aspect – emotions – and

study the interplay between individual’s emotions and the network dynamics.

Specifically, we consider the case where human emotions affect the traffic rate and

forwarding actions in the MOON. To complement it, we also study the scenario

in which the received messages affect the individual emotions. Relevant results of

performance evaluation are presented subsequently.

• Chapter 5: Effects of Heterogeneity

Heterogeneity in real-life can hardly be avoided. In this Chapter, we look at the

sources of heterogeneity in MOONs – and consequently, PSNs – arising due to

diverse hardware and software. We study the performance degradation in the

presence of such heterogeneity, and, in some cases, look at possible solutions also.

Incompatibility arising due to the differences in routing protocols is also studied,

and a method to account for the same is also proposed.

• Chapter 6: Summary and Conclusion

The final Chapter summarizes the outcome and contributions of this Thesis. We

also look at some drawbacks of this work when certain assumptions made here are

relaxed. Finally, we outline the different directions along which the present work

can be extended.

9

Chapter 2

Related Work

MOONs are inspired by different emerging trends of networking (for example, human-

centric sensing and PSNs). In these networks, human beings have more participation

than the traditional forms of networks. For example, it is human movement and their ac-

tivities that facilitate the operations of such networks. This Chapter presents a review of

the related works with focus on the four broad categories – 1) Mission-oriented networks

and participatory sensing, 2) Human aspects and emotions, 3) Post-disaster rescue oper-

ations and mobility, and 4) Network heterogeneity. Through these works, we primarily

look at the evolution of MOONs, and highlight the shortcomings/non-applicability of

these works in such contexts.

The rest of this Chapter is organized as follows. Section 2.1 presents an overview

of mission-oriented sensor networks and other emerging forms of sensor networks. In

Section 2.2, we discuss some of the existing works dealing with a specific human aspect

– emotions. In Section 2.3, we review different mobility models and communication

mechanisms proposed in the context of post-disaster scenarios. Section 2.4 discusses

multiple works addressing the case of heterogeneity in DTNs. In all these Sections,

the works reviewed are juxtaposed with the characteristics of MOONs, and thereby

highlighting the relevant shortcomings. Finally, Section 2.5 concludes this Chapter.

11

2. Related Work

2.1 Mission-Oriented Networks and Participatory Sensing

Mission-oriented wireless sensor networks (WSNs) [4,23], unlike other networks, require

that the sensors involved in the network coordinate with one another and align them-

selves altogether with the mission objectives [24, 25]. Such inter-device coordination

becomes essential to achieve complex missions (for example, shooting a hostile target

without causing any collateral damage).

Rao and Kesidis [23] studied mission-oriented WSNs, in which the nodes made use

of “purposeful mobility” in order to achieve the goal. The objective was to position the

nodes in such a way that the network-wide energy consumption is minimized. The nodes

in the network had multiple roles – tracking, scanning and data forwarding. The authors

considered the nodes running a distributed simulated annealing algorithm, which helped

them to position themselves in a fashion so as to minimize the network-wide transmission

cost.

Eswaran et al. [24] explored the case of marginal utility based rate adaptation in

mission-oriented WSNs and extended the well known network utility maximization

(NUM) model for the purpose. Although several works have addressed different sce-

narios using the NUM framework, the authors enumerated three key features of mission-

oriented WSNs that have been previously unexplored. Often, a single mission’s execution

is dependent on multiple sensors, which calls for the definition of a joint utility function

considering the data received from all such sources. The missions – the receivers eventu-

ally consuming the sensed data – are typically heterogeneous. While multiple receivers

can feed on data provided by a single sensor, they can have different individual utilities

defined. Moreover, when multiple missions are considered – many of which may arrive in

quick succession – the rate adaptation algorithm must have fast convergence to promptly

respond to the changes.

In the recent years, the research space of WSNs has forked into different directions

giving rise to new areas, variously identified as participatory sensing [5], people-centric

12

2.1. Mission-Oriented Networks and Participatory Sensing

sensing [6, 26], and opportunistic sensing [26]. Burke et al. [5] proposed participatory

sensing, where the mobile devices carried by human beings act as sensor nodes, and

engage in sensing their local environment in a distributed fashion. The authors noted

that the goal of participatory sensing is not only to collect data, but also to allow common

persons and professionals alike to analyze data and consequently, share knowledge.

Campbell et al. [27] presented the architecture of MetroSense, a scalable platform

for people-centric urban sensing – data about people and its immediate surroundings

in urban landscapes. MetroSense leverages opportunistic sensing, which arises due to

uncontrolled mobility of the human beings in the sensor field. Campbell et al. [26]

noted that in people-centric sensing, people act as the primary component of the sens-

ing system so formed. The authors further distinguish between two types of sensing –

participatory and opportunistic – based on the users’ role. In participatory sensing, a

user is more involved with the sensing process, for example, decision making on appli-

cation requests, data sharing and privacy issues. In contrast, under the opportunistic

sensing paradigm, most of the decision making process is delegated to the system with

predefined constraints, for example, privacy settings.

Srivastava et al.’s [6] work follows along the similar direction, where the authors

discuss several emerging sensing scenarios. In particular, they observed that in human-

centric sensing, people play three distinct roles, although not necessarily mutually ex-

clusive:

1. Sensing targets: Human beings themselves act as the sensing targets. This finds

application beyond the security domain, for example, personal health monitoring.

2. Sensor operators: People use the sensors present in their smartphones and other

portable devices – together with relevant applications – to sense the environment.

3. Data sources: In many cases (for example, social networking sites), human beings

disseminate and collect data about themselves without using any actual sensor.

13

2. Related Work

Moreover, the authors identified several challenges in the context of human-centric

sensing, ranging from energy aspects to participant selection and privacy concerns.

Synthesis

A summary of multiple works in the context of mission-oriented WSNs and human-

centric/participatory sensing is presented in Table 2.1.

Rao and Kesidis [23] presented an interesting concept of purposeful mobility in the

context of mission-oriented WSNs. The authors, however, considered a network with

an end-to-end communication paradigm. Moreover, their work assumed that any node

in the network could reach any other node. These reasons prevent the proposed model

from working in the case of MOONs, for example, a MOON formed in a post-disaster

recovery scenario. It may be noted that in MOONs – or in general, DTNs – presence

of end-to-end communication path between any source and destination node consisting

of one or more nodes is a rarity. Furthermore, scalability remains a research challenge,

since the concerned terrain size in their work was only 40 m × 40 m in area. Clearly, a

real-life MOON formed over a large terrain would not match such a scenario.

Table 2.1: A summary of different works on sensor networks and mission-oriented WSNs

Studies Aspects Shortcomings

Rao and Kesidis(2004) [23]

Purposeful mobility of the nodesto minimize the transmissionpower

Considered end-to-end commu-nications; scalability with re-spect to terrain size remains a re-search challenge

Eswaran et al. [24] Network utility maximization inmission-oriented WSNs

Considered conventional com-munication paradigm

Burke et al. [5],Srivastava et al.[6], Campbell et al.[26]

Distributed sensing based on hu-man movement and activities

Influence of human aspects otherthan movement not taken intoaccount

The new paradigms of people-centric sensing [6, 26], and opportunistic sensing [26],

in essence, acknowledge that the smartphones carried by human beings can potentially

14

2.2. Human Aspects and Emotions

act as sensors and provide diverse information. It may be noted that, while conventional

research on computer networks typically considered human beings as the end-users, in

the modern scenario, they have increased interaction with the network. This leads one

to think whether or not the purposeful mobility proposed for the machines could be

used by the human carriers as well in the PSNs, based on intelligent decision making

in order to enhance the chances of accomplishing mission objectives. In general, these

emerging paradigms of distributed sensing lacks in taking different human aspects into

consideration. Similar observation is made for other different works in mission-oriented

WSNs, for example, [24].

2.2 Human Aspects and Emotions

One of the human aspects largely considered in the existing literature is mobility. Several

real-life experiments were conducted, in which the subjects were given smart phones or

iMotes, and each such device recorded the time, apart from other information, when it

came in contact with the other devices. Based on these, the inter-contact times (ICT) –

the time duration between two successive contacts in the network – of the devices were

determined.

Chaintreau et al. [7] analyzed multiple real-life mobility traces, and concluded that

the ICT is distributed with a heavy tail, similar to the power law. Karagiannis et al. [8],

however, suggested that the distribution follows power law only until its characteristic

time, beyond which an exponential decay could be observed. Moreover these works

characterized human mobility patterns – certain pair of the devices tend to visit some

“home locations” repeatedly and regularly [8, 28]. It may be noted here that human

mobility is, indeed, the enabling factor for message delivery in PSNs.

On the other hand, Hui et al. [29] presented different algorithms for distributed

community detection based on users’ interactions using the mobility traces. Nelson [30]

proposed group management and routing schemes exploiting the structural patterns (for

15

2. Related Work

example, social connections and mobility patterns) in human-centric DTNs.

In different contexts, several research works were done with the aim of detecting

users’ emotions based on the interactions with the smart phones. LiKamWa et al. [31]

reported that the mood of the users could be detected from their smart phone usage

information such as location, phone calls made, and applications used. It was found that

the application used by a user and people with whom the user communicates deepened

on his/her mood.

Lee et al. [32], too, presented a mechanism unobtrusive emotion recognition based

on usage of smartphone. The authors identified 10 different features that are related to

human emotions. Based on this, a Bayesian Network classifier was used to recognize and

classify user emotions into seven different types. On the other hand, Kim and Choi [33]

showed that human emotions could be recognized based on as simple as the touching

patterns of smartphones. The authors argued for its wide applicability since increasing

number of devices in the modern age are enabling touch-interactions. The authors also

observed that such classification based on touch patterns can be made more accurate if

less emotions or data from other sensors are considered.

Tang et al. [14] studied how the emotions of individuals are related to their activities.

The authors also investigated the temporal correlation between the contemporary and

past emotional states of any user. Moreover, they also took social correlation of emotions

into account. Among other results, based on experiments conducted on subjects in a

university setting, the authors reported that users’ likelihood of calling or sending SMS

over phone is related to his/her emotional state, with the highest chance of calling while

being in a positive state.

On the other hand, Gao and Liu [34] studied the emotional reactions of the users

reflected in the posts in BBC and ReliefWeb, after the Japan earthquake in 2011. Neg-

ative emotion was found to be much higher than the positive one. Moreover, among the

negative emotions, sadness was found to dominate.

16

2.3. Post-disaster Rescue Operations

Synthesis

A disaster of large scale often renders the existing network communication infrastruc-

ture unusable due to disruptions, congestion and blackouts. In such scenarios, smart-

phone/featurephone based communication was found to increase sharply in contrast to

PC-based communication [35]. Since restoring cellular towers would take up to few days,

smartphone-based PSNs can serve as a “quick response” network.

While several works exist on recognizing users’ emotions – as summarized in Table 2.2

– there is lack of efforts in studying the reverse – how individual emotions could possibly

affect the network dynamics? Existing research works in Psychology, for example, [12]

and [13], considered the relation between human emotions and actions, and showed that

the nature of emotions has considerable effect on human actions [13]. This is more

relevant in MOONs, where the presence of human beings is integrally inherent to the

network.

In general, the existing works largely lack in addressing human aspects in the context

of networks. It is, however, logical, since until now, human beings were merely end-users

in the traditional networks (for example, cellular and ad hoc networks) and were not a

part of the system. To address some of the issues arising due to the existing knowledge

gap, in this work, we investigate the interaction between human emotions and network

dynamics. We restrict our scope to the study of the effects of emotions on the traffic

generation and message forwarding behavior.

2.3 Post-disaster Rescue Operations

In a post-disaster rescue operation, several groups of human rescue workers (for example,

firefighters and paramedics) are involved to locate the victims, move them to safety, and

provide medication. For a fast and efficient rescue operation, relevant information should

be exchanged among the rescue workers playing different roles. A natural disaster of large

17

2. Related Work

Table 2.2: A summary of different works on human emotion detection

Studies Aspects Shortcomings

LiKamWa etal. [31]

User’s mood detection based onsmartphone usage information

Do not consider the effects of humanemotions on smartphone usage andcommunication aspects in a network

Lee et al. [32]

Bayesian network-based de-tection and classification ofemotions

Kim andChoi [33]

Emotion recognition based ontouch patterns

Tang et al.[14]

Quantitative study of individ-ual’s emotions, their evolutionand influences

Correlation drawn between emotionand making a phone call may nothold for all scenarios; further, howindividual actions lead to network-wideeffect is not studied

scale often reduces the available communication infrastructure to zero, which, in turn,

calls for ad-hoc communications. Further, such communication should be delay-tolerant

as well – possible lack of end-to-end paths would require the collected information to

be buffered at individual nodes for long durations until a communication opportunity

is found. Moreover, human beings are equipped with their innate intelligence, which,

unlike robots, helps them in optimum decision making.

Multiple post-disaster mobility models are proposed in the literature. Nelson et

al. [36] modeled the mobility of users considering the effects of the environmental events

(for example, fire) and the role of the entities (for example, civilians and police). The

unique combinations of roles and events give rise to different movement patterns, where

the underlying mobility model is assumed to be Random Walk.

Uddin et al. [37] developed a mobility model for DTNs considering a scenario with

disaster forewarning. The model consisted of various centers and different mobility pat-

terns. The survivors stayed in the evacuation centers for a long time before returning

to their homes. The model, however, does not target the prevailing scenario immedi-

18

2.3. Post-disaster Rescue Operations

ately after a disaster, where opportunistic communications could be potentially used.

Aschenbruck et al. [38], on the other hand, presented a mobility model bearing close re-

semblance with the real-life rescue operations in the post-disaster scenarios. The authors

identified typical regions witnessed in such scenarios (for example, casualties treatment

area and transportation area) and included them in their model together with possible

obstacles.

Saha et al. [39] advocated a pre-planned 4-tier architecture to provide communication

facilities in the aftermath of a disaster, and showed that the optimum number of Wi-Fi

towers obtained through pre-planning provides lower latency compared to their random

placement. It may be noted that, in a large scale disaster-struck region, erecting Wi-Fi

towers and using them may not be possible until a few days. Moreover, even if such

infrastructure are deployed as preventive measures, their existence cannot be guaranteed

after a disaster has struck. Thus, such arrangements may not be useful immediately after

a disaster has occurred. In contrast, Hossman et al. [40] developed Twimight, a Twitter

client for the Android phones, which can engage in opportunistic communication in the

immediate aftermath of a disaster, until the network infrastructure can be restored.

On a different note, several synthetic mobility models are proposed to emulate the

characteristics of human mobility. Ekman et al. [41] developed the working day move-

ment (WDM) model, which attempts to capture the day-to-day movement patterns in

a typical urban scenario. Some of the characteristics of the WDM model (e.g., inter-

contact times) were found to be close to a real-life human mobility trace. The WDM

model, however, may not resemble the movements in a post-disaster scenario. A large-

scale disaster would not only destroy the available communication infrastructure, but,

perhaps, infrastructure of any kind including transportation.

19

2. Related Work

Synthesis

Table 2.3 presents a summary of multiple works on post-disaster mobility model and

communications. The mobility model in [36] is suitable for representing the events when

the disastrous incident(s) is(are) still in the effect. Rescue operations begin after such

incidents subside, and the rescue efforts in this context could be modeled according to

Aschenbruck et al. [38] modeled the mobility of the rescue workers in a post-disaster

scenario. However, apart from the rescue workers, there could be citizens potentially

unaffected by the disaster.

In general, the aftermath of a disaster can witness movement of two categories of

people:

1. Rescue workers and volunteers engaged in rescue operations (together with the

victims being transported), and

2. Unaffected citizens or victims with minor injuries.

The latter category, through their movement, provides scope for opportunistic com-

munication throughout the disaster area. The first category fails to do so because the

rescue efforts would mostly be localized in certain geographic areas.

While several mobility models were proposed, they do not focus on how the movement

of the human users are influenced by different forms of intelligence. For example, whether

or not innate human intelligence can affect their movement – and consequently, the

opportunity of communication and mission objectives – remain largely unexplored. This

is particularly relevant in the case of MOONs, where human beings are considered to

be a part of the network. This motivates us to consider the effects of another human

aspect – intuitive intelligence – on the performance of MOONs. Since intelligence defines

a broad scope, we restrict this work to the movement influenced by intuitive decision

making.

20

2.4. Network Heterogeneity

Table 2.3: A summary of different works on post-disaster mobility and communication

Studies Aspect Shortcomings

Nelson et al.[36]

Gravity-based model with rolesand events

Suitable for modeling the disastrous in-cidents

Ekman et al.[41]

Capture day-to-day movementpatterns in a typical urban sce-nario

May not resemble movements in a post-disaster scenarios

Uddin et al.[37]

Mobility model for scenarioswith disaster forewarning

Post-disaster mobility scenario may notbe the same

Aschenbrucket al. [38]

Modeled post-disaster rescue op-erations

Mobility of the other actors in the sys-tem remains unaddressed

Saha et al.[39]

Advocated pre-planned hierar-chical architecture for post-disaster communications

Disasters of large scale would affect ex-isting infrastructure; even erecting Wi-Fi towers after the disaster would takea few days

2.4 Network Heterogeneity

A multitude of existing works on DTNs have addressed the scenarios where the net-

work compositions are homogeneous. Such an assumption may not hold true in real

life. Schmohl and Baumgarten [42] noted, in the context of ubiquitous and pervasive

computing, that heterogeneity arises in mobile computing environments due to various

hardware and software composition of devices, and also the overall architecture of the

network. These issues of heterogeneity, if not addressed, would, therefore, affect the

performance of PSNs.

The issue of heterogeneity in the context of ubiquitous and pervasive computing is

widely addressed in the literature. Schmohl and Baumgarten [42] noted that heterogene-

ity arises in mobile computing environments due to the hardware and software of the

devices, and the architecture of the network. The authors proposed several approaches

for addressing heterogeneity, and presented a general architecture for implementing a

middleware to overcome the heterogeneity.

Bromberg et al. [43] proposed the Starlink framework – a middleware for run-time

bridging of heterogeneous protocols. The proposed framework can address heterogeneity

21

2. Related Work

related to the different message formats and protocol’s behavior. Stuedi and Alonso [44]

explored the integration of heterogeneous MAC protocols in mobile ad hoc networks,

with specific focus on 802.11 and Bluetooth. The authors proposed the use of software-

based virtual interface to integrate the devices with different MAC layers. The proposed

approach, although novel, is suitable for traditional networks using end-to-end commu-

nication paradigms. Moreover, the assumption of the use of such bridging software may

further lead towards heterogeneity.

The unlayered architecture of Haggle [45] was developed for the PSNs. Haggle’s focus

is on data-centric networks, and is located between the application layer and the hard-

ware interfaces. Haggle proposes forwarding data as Application Data Units (ADUs),

which have network-wide visibility, using epidemic and delegation mechanisms [45]. Petz

et al. [46] presented MaDMAN, a middleware for dynamic switching between MANET

and DTN protocols. In their proposed architecture, the network stack consisted of a

collection of different possible transport, network and link layer protocols. The network

stack could be switched with another even while the application is running, which en-

abled communications with asymmetric protocols. While no comparative performance

of the two models are available, MaDMAN supports more extensibility. However, unlike

Haggle, MaDMAN does not support data transfer with user-level naming.

Lee and Eun [47] and Tian and Li [48] considered heterogeneity in the contact pro-

cess of the mobile nodes and diversity in the pairwise contact patterns. Such factors,

however, do not reflect the heterogeneity in the composition of the concerned networks.

Li et al. [49] explored deploying defense mechanisms in PSNs to prevent malware at-

tacks. The authors considered a network of heterogeneous devices, where different types

of malware can only attack the systems they are targeted for. Manam et al. [50] pre-

sented the performance modeling of two routing protocols (two-hop and Epidemic) by

considering the nodes to have heterogeneous transmission ranges. The delivery latency

of the messages was found to decrease with the increasing transmission ranges of the

22

2.4. Network Heterogeneity

nodes.

Table 2.4: A summary of different works on network heterogeneity

Studies Aspect Shortcomings

Stuedi andAlonso(2005) [44]

Virtual interface to integrate thedevices with different MAC lay-ers

Assumption of the use of such bridgingsoftware may further lead towards het-erogeneity

Lee and Eun(2010) [47]

Heterogeneous inter-contacttimes of the nodes

Address heterogeneity arising due tothe underlying mobility; does not dealwith the diversity in devices/protocols

Tian and Li(2010) [48]

Heterogeneity in the contactpatterns of the nodes; frequentlycontacting pair of nodes exhibitExponential and Pareto inter-contact times

Petz et al.(2010) [46]

Middleware for dynamic switch-ing between MANET and DTNprotocol stacks

Useful in scenarios with low networkpartitioning; does not deal with hetero-geneity in routing protocols or how topreserve the protocol states

Li et al.(2010) [51]

Malware can only attack the sys-tems they are targeted for

Software heterogeneity of the devices isa constraint of the proposed optimiza-tion problem

Bromberg etal. (2011)[43]

Middleware for run-time bridg-ing of heterogeneous protocols

No performance evaluation in terms ofoverhead

Manam et al.(2012) [50]

Performance modeling of tworouting protocols considering thenodes with heterogeneous trans-mission ranges

Limited scope for addressing the het-erogeneities

Synthesis

A walkthrough of the existing works reveal that there is a lack of comprehensive approach

to address heterogeneity, and its impacts, in DTNs or PSNs. Besides, while works

in [47, 48, 50] focus on the reduction of communication opportunities in the network,

heterogeneity of certain aspects (e.g., incompatible network devices – in absence of any

bridging [44], and routing protocols) turn available communication opportunities useless.

Although a few works exists on the interoperability of the protocols, these do not discuss

23

2. Related Work

the overhead involved (for example, in terms of computation and processing), and their

suitability in the context of resource-constrained mobile devices. In general, the existing

works do not present any insight, quantitatively values, on how performance degrades in

heterogeneous PSNs. This motivates us to consider the different types of heterogeneities

and their effects on the performance of the MOONs, which are applicable for PSNs as

well.

On the other hand, traditional works addressing heterogeneity – be in DTNs or any

other networks – are usually limited to the consideration of hardware and software of

the communicating devices. MOONs, however, present forward a new dimension in the

domain of heterogeneity as well. As described earlier, MOONs do not merely contain

mechanical devices; rather, a node in a MOON is a combination of a device together

with its human owner. Therefore, diversity in human behavior, too, is expected to affect

the performance in MOONs. Therefore, in this work, we take into consideration certain

forms of heterogeneity in human behavior. In particular, we consider the following two

cases, in which:

1. Different movement decisions are made by different individuals based on the same

input, and

2. Different fractions of human owners in the MOON are affected by their respective

emotions.

2.5 Concluding Remarks

In this Chapter, we looked at multiple works representing some of the emerging trends

of networking where human beings play greater role. For instance, human mobility

is the enabling factor for message delivery in PSNs. However, human movement is

considered to be intrinsic to PSNs, and, therefore, acts as an independent parameter.

Apart from this, no other human factor has been considered in the context of networks.

24

2.5. Concluding Remarks

In this light, it is important to consider other human aspects as well, for example,

effects of intelligence-induced and emotion-driven actions on the mission objectives and

communication efficiency. On the other hand, while several works have addressed the

issue of heterogeneity in DTNs, their scope has been limited. Indeed, real-life networks

are expected to reflect heterogeneity of several forms, and, therefore, it is crucial to

address such issues. Moreover, when MOONs are considered, heterogeneity does not

originate only due to the hardware and software of the communicating devices, but are

ushered in as a consequence of different human aspects as well.

25

Chapter 3

Effects of Intelligence-induced

Movement in MOONs

In this Chapter, we formally define a MOON with a set of different parameters. Subse-

quently, we study the scenario where mission objectives are considered in the presence of

mission objectives. We further look at how intuitive intelligence-based decision making

helps the mission prospects in MOONs.

The specific contributions of this Chapter are summarized in the following points:

• Representing a MOON with a mission and set of nodes in the network.

• Defining four levels of intelligence, which manifest the intuitive decision making

process in human beings.

• Studying the mission performance when the mobile nodes – devices together with

their owners – engage in movement dictated by their corresponding intelligence

levels.

• Studying the effects of certain forms of heterogeneity – unique to MOONs and

otherwise – on the performance of the aforementioned MOON.

27

3. Effects of Intelligence-induced Movement in MOONs

This Chapter is organized as follows. Section 3.1 defines a MOON – a network with

mission objectives – and illustrates with an example. Section 3.2 discusses four levels

of intuitive human intelligence, and how they influence the movement of the nodes.

This Section also discusses the opportunistic communication scenario among the nodes.

In Section 3.3, a detailed setup of the simulation scenario is described. The results

of relevant performance evaluation are presented in Section 3.4. Finally, Section 3.5

summarizes the contributions and findings of this Chapter.

3.1 Mission Overview

We define a MOON as follows. Let, M = (G,T, S,P) be a mission, where G is the set of

mission goals, T is the time-line of the mission, S denotes the set of strategies applied

to execute a mission, and P is the set of performance parameters related to G. The

underlying MOON is a network N = (N,M) with a mission M and a set of nodes N . In

this work, we consider the MOONs with a single mission only.

To illustrate, let us consider a MOON formed in the aftermath of a large-scale dis-

aster. The key problem addressed here is to increase the communication opportunities

among the mobile and stationary nodes in such a scenario. In this context, increasing

the communication opportunities represents the mission goal. Whereas, the intuitive

intelligence-based movement of the mobile nodes represent the strategy of the mission.

The mobile nodes in this scenario represent the rescue workers, paramedics, fire-

fighters and unaffected or minimally affected people. On the other hand, the stationary

nodes indicate the trapped victims, rescued persons requiring immediate medical care,

various camps and so on. Therefore, the increase in communication opportunity between

a mobile and a stationary node ensures that the likelihood of availing the relevant ser-

vices is enhanced. For example, a paramedic, after coming to know about the location

of recently rescued victim, would move to attend the person.

In the remainder of this Chapter, we consider a square-shaped terrain in the context

28

3.2. Opportunistic Communications with Intelligence

of a post-disaster recovery scenario, which is divided into multiple square-shaped zones

of equal size. Each such zone is identified with (x_id, y_id), where x_id and y_id,

respectively, denotes the zone index along the x- and y- axes from a chosen origin. In

particular, the left-bottom zone is identified with (0, 0). Further, we consider that the

human rescue operators (mobile nodes) form a MOON in such a scenario, and all the

nodes are aware of their current locations.

The following definitions are provided in regards the geography of the terrain.

Definition 1. K-left zones: The K-left zones of any zone in the network represent

the K neighboring zones horizontally towards the left of the concerned zone. Thus, if

the identifier of a square zone in the network is (xid, yid), its 2-left zones would be

(xid − 1, yid) and (xid − 2, yid), provided the zones exist in the terrain.

Definition 2. 1-near zones: The 1-near zones of any zone in the network represent

the adjoining neighboring zones, which are one block away. Thus, if the identifier of a

square zone in the network is (xid, yid), then it could have up to eight possible 1-near

zones: (xid−1, yid−1), (xid−1, yid), (xid−1, yid+1), (xid, yid+1), (xid+1, yid+1),

(xid + 1, yid), (xid + 1, yid − 1), and (xid, yid − 1).

The definition of 1 -near zones could be generalized to K -near zones, where any zone

could have up to (2K + 1) × (2K + 1) − 1 neighboring zones. Other terminologies (for

example, K -top zones) can be similarly defined.

3.2 Opportunistic Communications with Intelligence

A Mission-Oriented Opportunistic Network reflects characteristics such as human be-

ings, and thus, human intelligence, which are not usually seen in a general WSN or any

other wireless network. Unlike machines, it is hypothesized that the presence of human

beings in such a network would have a positive effect on the accomplishing mission. For

instance, in the context of our post-disaster recovery operation example, when a rescue

29


worker finds a trapped victim, he/she would possibly look in the adjoining places for the

availability of any more victim(s) requiring assistance.

To depict such scenarios, we now define a set of levels of human intelligence. These

levels manifest simple intuition-based schemes that a human being could adopt when it

comes in contact with a victim. The different levels are described below.

• L1: When a mobile node learns about the location of a stationary node, it moves

towards that node. This could represent the case when a trapped person has been

recovered, and requires medical attention. In terms of network communication,

this could imply tracking of updates of one’s medical reports.

• L2: A mobile node with intelligence level L2, on learning the location of a new sta-

tionary node, visits the node as well as a randomly chosen zone from the stationary

node’s 1 -near zones.

• L3: A mobile node with intelligence level L3, on learning the location of a new

stationary node, visits the node as well as its K -left neighboring zones

• L4: A mobile node with intelligence level L4, on learning the location of a new

stationary node, visits the node as well as all of its 1 -near zones.

We note here that the L2 scheme described above is loosely based on the idea of

“Neighborhood Search Based Exploration” discussed in [52]. To summarize, in the

neighborhood search based exploration, an action in the current state is probabilistically

chosen in random from the neighborhood of the previous action state.

Figure 3.1 presents an overview of how a node behaves in a MOON. We note that in

a simulation framework we cannot distinguish between humans and devices. However,

the architecture presented here closely relates to reality. To illustrate this, in real life,

a human being would be working with the application. Therefore, the interfaces among

application, router, and database remains valid. The only difference arises in the case of

30

3.2. Opportunistic Communications with Intelligence

the mobility module, because a real human being moves by himself/herself. Nevertheless,

the proposed application could suggest to its owner the locations it should travel next.

Application

Mobilitymodule

Database

Router

(x, y)

(x, y)

(x, y)

Figure 3.1: A node’s behavior in a MOON.

When the application (Figure 3.1) running in a device receives the coordinates of a

new stationary node, the application stores the location in its database. It, then, passes

the information to the router module by creating a message (containing the newly found

coordinates) destined to all other mobile nodes. Finally, it instructs its mobility module

to go to the location. The intelligence level determines whether the node should only

visit that particular location, or any neighboring zone(s) as well.

When the above node comes in contact with a mobile node, the latter’s router re-

ceives a message containing the new coordinates, and passes upwards to the application.

The application running on this node then checks whether it has already visited that

particular stationary node (identified by location). If not, it takes steps as discussed

above.

Algorithm 3.1 presents the actions taken by the application module of a node under

intelligence level L2. For instance, line numbers 7–10 show that the application execut-

ing in the mobile node randomly selects a zone from the concerned zone’s immediate

neighbors, and subsequently visits that selected zone.

31


Algorithm 3.1: Purposeful mobility under the L2 scheme

Inputs:• s: A stationary node’s location• database: Database of locations of stationary nodes stored by the node

Output:

• Movement as per the L2 level

1 ix, iy = getZoneID(s.X, s.Y )2 neighbors = getNeighbors(ix, iy)3 if s not in database then

4 database.addLocation(s)5 moveTo(s)6 // Randomly select a neighbor

7 r = randomIntegerInRange(neighbors.size())8 zone = neighbors[r]9 c = zone.getCenter()

10 moveTo(c)

3.3 Simulation

We used the ONE simulator [53] for evaluating the effects of the proposed intelligence

model. We considered a 5×5 sq. km. simulation terrain, which was divided into multiple

square-shaped zones, each of size 10 × 10 sq. m. The stationary nodes were placed

randomly within the terrain. We considered two cases in regard to the density of nodes

– 1) 45 stationary and 20 mobile nodes, and 2) 60 stationary and 30 mobile nodes. We

considered the nodes to be moving with three different speed ranges: 0.5 m/s – 1.0 m/s,

1.0 m/s – 2.0 m/s, and 2.0 m/s – 2.5 m/s. It may be noted that the human walking speed

is about 1.5 m/s, on an average. The sub-average speed would be obtained in scenarios

with obstacles, whereas higher speed indicated movements like rushing towards a spot.

The scenarios were simulated over two simulation time durations – 12 and 24 hours. The

Epidemic [16] routing protocol was used for message dissemination by the mobile nodes.

To closely reflect the real-life scenarios, we also considered the stationary nodes to

be located in clusters around seven different points in the terrain. For example, the case

of multiple victims trapped in vicinity reflects such scenario. Moreover, we also took

32

3.4. Results

into account some heterogeneity factors that more often than less affect our daily lives.

Specifically, we considered the case when certain fraction of the mobile nodes used their

intelligence level L1, while the remaining used L4. Thereafter, we studied the scenario

where not all the mobile nodes had the relevant application available as shown in Figure

3.1. In this case, we considered homogeneous intelligence levels, i.e., all the nodes using

a single level of intelligence in order to isolate the effects of any other factor.

We contrasted our human intelligence-based schemes with the Random Waypoint

(RWP) mobility model. In case of intelligence level L2, we considered 3 -left neighboring

zones. The results for all simulation scenarios were obtained by taking an average over

30 random scenarios and the 95% confidence intervals were determined.

The following metrics were considered for performance evaluation.

• Average number of encounters between a mobile and a stationary node.

• Number of unique encounters between a mobile and a stationary node. For ex-

ample, if a mobile node m encounters a stationary node s for say, n times, then

the total number of encounters for this pair is n, whereas the number of unique

encounters is 1.

Both these metrics help in evaluating the extent to which intelligence-induced move-

ment helps, if at all, in accomplishing the previously discussed mission objectives.

3.4 Results

Figure 3.2 shows the relative increment in encounters – with respect to the RWP mobility

model – between the mobile and the randomly located stationary nodes, on an average,

when the four different intelligence levels were used. Reduced performance is observed

using the L1 scheme, in which a mobile node only visits a new stationary node. On

the other hand, the improvement is maximum for the L4 scheme, where a mobile node

33


searches the 1 -near zones of a zone where a stationary node is located. Figures 3.2 (a)

and (b) correspond to two different node densities, as indicated. Figure 3.3 plots the

same performance results when the simulation was executed for 24 hours. Similar trends

are observed in both the Figures.

20

40

60

80

100

120

L1 L2 L3 L4

Cha

nge

in e

ncou

nter

s (%

)

Intelligence level

0.5-1.0 m/s1.0-2.0 m/s2.0-2.5 m/s

20

40

60

80

100

120

140

L1 L2 L3 L4

Cha

nge

in e

ncou

nter

s (%

)

Intelligence level

0.5-1.0 m/s1.0-2.0 m/s2.0-2.5 m/s

(a) (b)

Figure 3.2: Encounters of mobile nodes with stationary nodes with (a) 45 stationary and20 mobile nodes and (b) 60 stationary and 30 mobile nodes for 12 simulation hours.

20

40

60

80

100

120

L1 L2 L3 L4

Cha

nge

in e

ncou

nter

s (%

)

Intelligence level

0.5-1.0 m/s1.0-2.0 m/s2.0-2.5 m/s

20

40

60

80

100

120

140

160

L1 L2 L3 L4

Cha

nge

in e

ncou

nter

s (%

)

Intelligence level

0.5-1.0 m/s1.0-2.0 m/s2.0-2.5 m/s

(a) (b)

Figure 3.3: Encounters of mobile nodes with stationary nodes with (a) 45 stationary and20 mobile nodes and (b) 60 stationary and 30 mobile nodes for 24 simulation hours.

Figure 3.4 plots the result of performance evaluation when the 60 stationary nodes

were clustered around different locations in the terrain; 30 mobile nodes were considered.

From the Figure, we can observe that the number of unique encounters between a pair of

mobile and stationary node, on an average, improved to different extents for the different

intelligence levels. Indeed, when a mobile node visits a stationary node and one or more

34

3.4. Results

of its neighboring zones, the chances of encountering new stationary nodes also increases.

60

80

100

120

140

160

L1 L2 L3 L4

Cha

nge

in e

ncou

nter

s (%

)

Intelligence level

Mobile-stationaryUnique stationary

Figure 3.4: Change (%) in encounters with clustered static nodes.

We investigated some scenarios pertaining to the MOON with heterogeneous envi-

ronment. Figure 3.5 shows the case when certain fraction of the mobile nodes – shown

along the x-axis – moved according to the intelligence level L1, while the remaining

mobile nodes used the level L4. It can be observed that when 25% of the nodes used the

level L1, and the remaining used L4, the number of encounters between a mobile and a

stationary node, on an average, was highest. This is because, with more nodes visiting

the 1-near zones of the discovered stationary nodes, the chances of mobile-stationary

node encounters increased. On the other hand, reduced performance was obtained when

all the nodes used the intelligence level L1.

Finally, in Figure 3.6, we studied the performance obtained when certain percentage

of the mobile nodes – shown along the x-axis – in the MOON had the relevant application

available (Figure 3.1). The average encounters achieved for each such fraction of nodes

for the different intelligence levels are also shown. As can be intuitively explained, the

mission performance increased when the increasing number of nodes in the MOON had

the application available. The maximum performance was obtained for the 100% case,

i.e., when all the mobile nodes had the application available with them.

In this context, it can be observed that the performance under the L1, L2 and

35


0.68

0.72

0.76

0.8

0.84

25 50 75 100

Ave

rage

enc

ount

ers

Percentage of nodes

Figure 3.5: Performance when different fraction of nodes had L1 and L4 levels.

0.68

0.72

0.76

0.8

0.84

25 50 75 100

Ave

rage

enc

ount

ers

Percentage of nodes

L1L2

L3L4

Figure 3.6: Performance when different fraction of nodes had the relevant applicationavailable.

L3 schemes were quite close, although the same under the L4 level was statistically

different from the others. As such, a question regarding the necessity to have the three

levels L1, L2 and L3 may arise. We note that the primary objective here has not been

to put forward a “best” exploration scheme. Rather, it has been to underscore the

fact that movement decisions made by intuitive forms of human intelligence can help in

accomplishing the mission objectives by enhancing the communication opportunities.

36



Mission-Oriented Opportunistic Networks, with human involvement, deviate from the

traditional communication paradigm. Innate intelligence and context awareness equip

human beings to make better decisions in such a network. In this Chapter, we considered

a post-disaster rescue operation scenario with mobile and stationary nodes. We described

some algorithms to represent intuitive intelligence-based human decision making. When

a mobile node encounters a stationary node for the first time, it communicates the lo-

cation of the node to others through opportunistic contacts. The resulting purposeful

mobility of the nodes appeared to provide better communication chances with stationary

nodes as compared to the RWP model. Moreover, we considered certain forms of het-

erogeneities unique to MOONs, and studied the corresponding performance. Although

we have considered random mobilities of the nodes, the scenario can, however, be corre-

lated to real-life. For instance, consider two groups of firefighters and paramedics. The

paramedics are unlikely sweep search the entire disaster region to look for the victims.

Rather, the rescue operators are likely to pass on the information, and the paramedics

would plan their route accordingly.

37

Chapter 4

Aspects of Human Emotion in

MOONs

In this Chapter, we study an aspect of human-network interplay in MOONs, by con-

sidering the effects of human emotions on the network performance. This work differs

from the related existing works, which focus on detecting emotions of the users based on

their activities, for example, their smart phone usage patterns. The study uses Meftah’s

computational model of the emotions, which itself is based on Plutchik’s circumplex

model of emotions proposed in Psychology.

We represent the emotion of any user at any time instant as a function of the indi-

vidual’s previous state of emotion and the content of the messages received by him/her.

Based on this, we consider how the variation in individual emotions affects the traffic

generation rate in the network. Further, we investigate the scenarios in which the users

switch off their devices’ radios, and their effects on the message delivery ratio.

The specific contributions of this Chapter are summarized below:

• Using a computational model of emotions to fit scenarios typical of MOONs. In

particular, if the vector e−1 represents the previous state of emotions, and the

vector r represents the different types of messages received, then we model the

39

4. Aspects of Human Emotion in MOONs

current emotion as a function of the two parameters: e = f(e−1, r).

• Analyzing the effects of temporal variation of emotions on the traffic load in

MOONs. Specifically, if R(0) is the normal rate of message creation, then the

spike in traffic generation rate at time t of any individual i is Ri(t) = β × R(0),

where β ≥ 0, is determined by the current state of emotion.

• Representing the fluctuations in the message forwarding behavior of the users based

on their current state of emotion.

• Capturing the effects of emotion on the delivery of the messages through simula-

tions.

This Chapter consists of five sections. Section 4.1 presents a brief overview of the

computational model of emotions proposed by Meftah et al. [22]. Thereafter, we discuss

the mathematical representation that we have used in order to capture the effects of

human emotions on the network dynamics of the MOON. In Section 4.2, the effects

of the network – in the form of information received – on the individuals’ emotions is

presented. A post-disaster scenario is considered in this context. Section 4.3 discusses

in detail the simulation setup used and the metrics for performance evaluation. The

results and discussions are presented in Section 4.4. Finally, Section 4.5 summarizes this

Chapter.

4.1 Representing the Effects of Emotion

While multiple theories for modeling emotions exist, we have considered the Plutchik’s

model [21] in Psychology. Plutchik proposed a color combination-based “circumplex

model” to represent the eight basic emotions – joy, sadness, trust, disgust, fear, anger,

surprise and anticipation – and their combinations. Plutchik observed that each basic

emotion can exist with different levels of intensities. For example, a higher intensity

40

4.1. Representing the Effects of Emotion

of joy represents the emotion ecstasy. Further, two basic emotions can be combined to

derive an advanced emotion (dyad). For example, a mix of the emotions, joy and trust,

gives rise to the emotion love.

Figure 4.1: Plutchik’s circumplex model of emotions (Retrieved from:https://commons.wikimedia.org/wiki/File:Plutchik-wheel.svg on 29 November2013).

4.1.1 The Computational Model of Emotions

Meftah et al. [22] used Plutchik’s model in Psychology to develop an algebraic repre-

sentation of emotions. In [22], the basic emotions are represented with a basis vector

B =(

Joy Sadness Trust Disgust Fear Anger Surprise Anticipation)

, and an emotion

vector e =

(

e1 e2 · · · e8

)

, where ei ∈ [0, 1] indicates the intensity of the ith emotion

in B, ∀i ∈ {1, 2, · · · , 8}. Therefore, a basic emotion, for example, joy, can be represented

by the intensity vector(

0.6 0 0 0 0 0 0 0)

. In general, an intensity vector represents a

basic emotion if it contains a single non-zero coefficient [22].

Meftah et al. noted that Plutchik’s model is simple, intuitive, and includes all pos-

41


sible combinations of emotions. Their proposed algebraic model of human emotion

provides ease of computation as well. Further, the different operators proposed by them

help in combining the effects of network dynamics together with the emotional intensi-

ties. Thus, we use Meftah et al.’s model in this work. For the sake of completeness, we

present below a few definitions (Definitions 4 and 5 are adopted from [22]) and other

theoretical characterizations that help in modeling the effects of emotion.

Definition 3. Dominant Emotion: Given the intensity vector e =

(

e1 e2 · · · en

)

of an individual’s emotions at any time t, the dominant emotion (DE) of the person at

that time instant is defined as:

DE(ei(t)) =

argmaxk

(ek), ek > ej , j 6= k, ∀j

∅N , ek = ej , ∀j, k

DE(ei(t − 1)), #(max(ei(t))) ∈ (1, n),

(4.1)

∀j, k ∈ {1, 2, · · · , n}. Here, ek indicates an element of ei(t), and #(A) denotes the

count of any event A. In other words, when all the emotions have unique intensities,

the DE is the one with the maximum intensity. Here, ∅N denotes the neutral emotion,

and DE(ei(t−1)) indicates the dominant emotion of the person at a previously counted

time instant. In case, at any time instant, there exists multiple emotions with equally

highest intensities, we consider the previously dominant emotion to sustain.

Definition 4. Addition of Emotion Vectors [22]: Let, e1 =

(

e1,1 e1,2 · · · e1,n

)

and e2 =

(

e2,1 e2,2 · · · e2,n

)

be two emotion vectors. The addition of the vectors, e1

and e2, denoted by ⊕, is defined as:

e1 ⊕ e2 =(

max(e1,1, e2,1) max(e1,2, e2,2) · · · max(e1,n, e2,n))

.

Definition 5. Scalar Multiplication with Emotion Vector [22]: The scalar multi-

42

4.1. Representing the Effects of Emotion

plication of an emotion vector e =

(

e1 e2 · · · en

)

with a scalar c, denoted by ⊙, is

defined as:

c ⊙ e =

(

c × e1 c × e2 · · · c × en

)

.

Definition 6. Scalar Exponent of Emotion Vector: The scalar exponent of an

emotion vector e =

(

e1 e2 · · · en

)

with a scalar c is defined as:

c ⊛ e =

(

ce1 ce2 · · · cen

)

.

Definition 7. Diagonal Form of an Emotion Vector: Given an emotion vector

e =

(

e1 e2 · · · en

)

, its representation as a diagonal matrix is denoted by:

diag(e) =

e1 0 · · · 0

0 e2 · · · 0

......

. . . 0

0 0 · · · en

.

4.1.2 Effects of Emotions on the Network Dynamics

Let, ei(t) =

(

ei1 ei

2 · · · ein

)

denote the emotion vector of the ith person at time t.

Let, Ri(t) and ki(t), respectively, be the message generation rate and level (degree) of

cooperation of the ith person at time t. Let us consider that R(0) and k(0), respectively,

represent the message generation rate and level of cooperation under “normal conditions”.

Specifically, let us assume that k(0) = 1. Let, β =

(

β1 β2 · · · βn

)

be the coefficients

of amplification in (individual) traffic load, so that

Ri(t) = βj × R(0), (4.2)

43


where j = DE(ei(t)), is the dominant emotion for any user i at time t, βj ≥ 0, ∀j ∈

{1, 2, · · · , n}. Thus, the aggregate traffic injected at any time t in the network is given

by R(t) =∑N

i=1 Ri(t), where N denotes the total number of users. Further, let γ =(

γ1 γ2 · · · γn

)

denote the coefficients of the variation in the degrees of cooperation,

0 ≤ γj ≤ 1, ∀j ∈ {1, 2, · · · , n}, such that

ki(t) = γj × k(0). (4.3)

Thus, for any individual i at time t, Equations (4.2) and (4.3), respectively, gives

the variation in traffic generated and cooperation extended as an effect of her the then

dominant emotion j.

Further, the network dynamics – in the form of information (messages) received –

can, in turn, alter an individual’s state of emotion. Let, M denote the set of messages

generated in the MOON. Let, the vector Mi(t) =

(

Mi,1 Mi,2 · · · Mi,n

)

be a collec-

tion of the set of messages received by any node i till the time instant t, such that the

Mi,j = {m : m ∈ M} messages affect the jth emotion, ∀j ∈ {1, 2, · · · , n}.

Further, let I be a function that captures the effects of a given set of messages

(information) on a particular emotion so that I(Mi,j) ∈ [0, 1], ∀j. Then, the updated

emotion vector at time t:

ei(t) = ei(t′) ⊗ τ(t − t′) ⊕ ri(t), (4.4)

where ⊗ denotes the matrix multiplication operator and ri(t) =

(

ri,1 ri,2 · · · ri,n

)

=(

I(Mi,1) I(Mi,2) · · · I(Mi,n)

)

; t > t′.

The function τ represents the temporal decay in the intensities of the emotions

44

4.2. Application Scenario

considered to be exponential, as indicated in [54]. Therefore, with ∆t = t − t′, we have:

τ(∆t) =

e−a1∆t 0 · · · 0

0 e−a2∆t · · · 0

......

. . ....

0 0 0 e−an∆t

= diag(e ⊛ (−a ⊙ ∆t)).

Here, the constant vector a =

(

a1 a2 · · · an

)

denotes the coefficients of decay of

the basic emotions, aj ∈ (0, 1), ∀j ∈ {1, 2, · · · , n}, and e is the base of natural logarithms.

Thus, the greater the value of any aj , the quicker the decay of that basic emotion. In

other words, given an initial emotion vector e, the vector a determines which basic

emotion, if any, would be dominant in the long run when not affected by any other

external factors.

The vector addition operator is used with r in (4.4) because the effects of the received

messages can act as stimuli that could altogether alter the emotion of an individual.

If e′j represents the intensity of the jth emotion at time t′, ∀j, then, on simplification,

we get:

e(t) =(

e′1 × e−a1∆t e′

2 × e−a2∆t · · · e′n × e−an∆t

)

⊕

(

r1 r2 · · · rn

)

. (4.5)

4.2 Application Scenario

The above model is applicable in the case of a MOON formed in a post-disaster rescue

operation. In such a scenario, the survivors and the rescue workers, equipped with smart

phones, engage in opportunistic communications. The messages exchanged among the

users includes, but not limited to, information about the rescued victims and queries for

missing persons.

45


In the remaining Sections, we consider a subset of the three emotions namely, joy,

sadness, and fear, i.e., n = 3, typically occurring in the post-disaster scenario. In such a

scenario, a user feels happy if she is unhurt, and her family is safe. However, if a person is

unable to get someone’s information, she may panic. News of any injured family member

would sadden a person. We note that these assumptions could be easily generalized for

all the emotions – and could be used in other scenarios as well – by considering the

effects of the received messages upon them, if any.

4.2.1 Variation in Emotion

Let, the messages exchanged in the MOON be classified into three types – Good, Bad,

and Unknown. The Good messages represent the case when someone intends to inform

that she is safe, or, perhaps, a missing person has been located. Likewise, the Bad, and

Unknown represent the cases of dissemination of related information. Let, mG, mB, and

mU , respectively, be the counts of such messages received by any individual until time t.

To determine ri, we consider that the effect of any kind of message on a basic emotion

is proportional to the number of such messages received. Thus,

r(t) =

(

mG

mN +1mB

mN +1mU

mN +1

)

, (4.6)

where mN = mG + mB + mU , is the total number of messages received, mN ≥ 0.

While the intensity of emotions decay with time, the rate of decay may not be the

same for all. For example, one may quickly shift from a joyful state to a neutral state, but

the sadness may sustain comparatively longer. Here, we consider that a1 – the decay rate

of joy – to be greater than the decay rates of sadness and fear, i.e., a1 > a2, a3. It may

be noted here that this is in contrast to the findings in [14], where it was reported that

normal and good emotions sustained longer than the negative emotions. The apparent

deviation is explained as follows. The study in [14] was performed in a university setting,

and the distribution of the emotions exhibited there are largely different from that of the

46


aftermath of a large-scale disaster. Thus, the observations would also vary as per the

scenario considered. Indeed, it can be recalled from Section 2.2 that negative emotions

dominate in typical scenarios considered here [34].

4.2.2 Variation in Traffic Load

We consider the time distribution of message generation in the network to be Poisson,

with the rate λ messages/second. For simplicity, we further consider that, on an average,

all the users create same number of messages. Thus, if there are N users in the network,

we have X =∑N

i=1 Xi, where Xi ∼ Poisson(λi), and λi = λN

, ∀i ∈ {1, 2, · · · , N}. The

inter-arrival times of X are exponentially distributed with mean time as 1λ

. Thus, un-

der normal circumstances, the average duration between two consecutive generation of

messages for each individual is given by:

T̄i =1

λ/N. (4.7)

We consider that a user generates messages in the “normal rate” with the average

inter-message duration as T̄i when she is happy. The amplification in traffic generation

rate in this case is unity, i.e., β1 = 1. A user, however, is considered to stop generating

any message when sad so that T̄i = ∞. In this case, β2 = 0. Finally, any user is assumed

to double the message generation rate when the dominant emotion is Fear, which could

be explained by the rationale that an anxious person sends out more queries to the

different people in the MOON. This case is equivalent to halving T̄i, and is represented

with β3 = 2. Thus, β =

(

1 0 2

)

. Once again, we apply similar reasoning as earlier

to justify its divergence from [14].

Finally, let pm =

(

pm1pm2

pm3

)

denote the probabilities of different categories

of messages generated in the MOON. In other words, on an average, pm1gives the pro-

portion of Good messages generated in the MOON. Similarly, pm2and pm3

, respectively,

denote the proportion of the Bad and Unknown messages generated.

47


In this context, the following lemma and theorem are stated.

Lemma 1. In a MOON with N nodes, where the network-wide rate of traffic generation

is Poisson(λ), and the source and destination nodes are chosen uniformly at random,

the traffic arrival rate at any destination node is less than Poisson(λe), where λe =

(2N−1)6N

× λ.

Proof. Let us consider that the N nodes in the MOON form a complete graph, where the

messages could be delivered instantaneously. Also, let us assume that a particular node,

X, is selected as destination by the other (N − 1) nodes. Let, Pi denote the probability

that i nodes in the MOON select the node X as the destination. Such selection by the

source nodes is independent of each other. Then, the expected rate of traffic destined

for the node X is∑N−1

i=1 i × Pi × Poisson(λ/N) =∑N−1

i=1 i × iN−1 × Poisson(λ/N) =

2N−16 × Poisson(λ). Now, since each of the N nodes is equally likely to be chosen

as the destination, on an average, the traffic arrival rate at any destination node is

1N

× 2N−16 × Poisson(λ) = Poisson(2N−1

6N× λ) = Poisson(λe).

In real-life MOONs, however, messages cannot be delivered instantaneously, since the

contacts among the nodes are intermittent and the messages are transferred over multiple

hops. Moreover, contact opportunities are lost due to the effects of the emotions, sadness

and fear. Assuming that the traffic arrival at the destination is still a Poisson process,

let us denote the arrival rate as λe −µ, where µ > 0, is a parameter that accounts for the

characteristics of the MOON considered. Thus, the traffic arrival rate at the destination

nodes is less than Poisson(λe).

The above reflects the maximum traffic in flow to a node, which can be useful for

the buffer management schemes.

Theorem 1. The rate of change in intensity of the ith emotion, over a time interval ∆t,

is either ei(t)×(e−ai∆t−1)∆t

orpmi

−ei(t)

∆t, where pmi

is the proportion of messages generated

48


in the MOON that affects the ith emotion. It is assumed that the proportion of the

corresponding messages received is also the same.

Proof. Let us consider equally spaced time instants tj , so that tj − tj−1 = ∆t. Then,

from Equations (4.4) and (4.6), the intensity of the ith emotion at time tj+1 = tj + ∆t

is ei(t + ∆t) = max{ei(t) × e−ai∆t, pmi}. Therefore, the time rate of change in the

intensity of an emotion ismax{ei(t)×e−ai∆t, pmi

}−ei(t)

∆t. Thus, if ei(t) × e−ai∆t ≥ pmi

, the

rate of change is ei(t)×(e−ai∆t−1)∆t

; otherwise, it ispmi

−ei(t)

∆t.

4.2.3 Changes in User Cooperation

We now elaborate on the notion of “cooperation” used in this work. Experimental evi-

dences show that the intention to use information technology (IT) is negatively related

to anxiety [55]. Beaudry and Pinsonneault [56] separately established that while hap-

piness is positively associated with the IT usage, anxiety of human beings is negatively

associated with it. Inspired by this, we consider the case in which the users deactivate

networking capabilities of their devices depending on their respective dominant emo-

tions, and, thus, loses potential opportunities of both sending and receiving messages.

We consider the scenario where a user deactivates networking capabilities of her device –

by turning her device, or just the radio, off – depending on her dominant emotion. This,

however, is not selfish behavior – a user does not look for his/her own gains. Rather, by

switching off the radio, he/she loses potential opportunities of both sending and receiving

messages.

We consider that a user keeps the radio of his/her device active when the DE is joy

or neutral. In this case, the user fully “cooperates” with the network operations, i.e.,

γ1 = 1. Any user, whose DE is sadness is assumed to switch off the device or its radio,

i.e., γ2 = 0. Finally, when the DE of any user is fear, he/she switches off the device

radio with a probability 0.5, i.e., γ3 = 0.5. In this case, the device remains inactive for a

random duration of time in between {0, Toff } seconds. However, a user is not expected

49


to turn off the radio every second. Indeed, if she does so, then no, or perhaps, immensely

less number of, messages could be transferred. Therefore, we assume that any user opts

to switch off the radio periodically after time TF ear.

Thus, in terms of user cooperation, we have γ =

(

1 0 0.5

)

. We note that, when

the DE is neutral, the individual traffic generation rate is λ, and the device radio remains

on. Algorithm 4.1 shows the different actions taken by any user when the DE is fear.

Algorithm 4.1: Actions taken when dominant emotion is fear

Inputs:• e: Emotion vector at time t• a: Decay coefficients• λ: Normal message generation rate (Poisson)• t′: Current time

Output:

• Changes in traffic load and cooperation

1 ∆t = t′ − t2 Evaluate r using Equation (4.6)3 Update e according to Equation (4.5)4 Determine dominant emotion DE5 if DE is fear then

6 λ′ = 2 × λ7 Generate messages with Poisson(λ′) process8 tL = Last time when the radio was switched off9 if t′ ≥ tL + TF ear then

10 tL = t′

11 p = Random number in the range [0, 1)12 if p < 0.5 then

13 toff = Random number in range [0, Toff )14 Switch off the radio of the device for time toff

It may be noted that the choice of β and γ are context-sensitive. For example, if one

considers winning a gold medal by a particular country during the Olympics, people in

the audience from that country would become happy and may result in traffic explosion

while trying to communicate with others. In this work, we choose the values of β and γ

such that they can represent typical post-disaster scenarios.

It may be further noted that certain effects of emotion are permanent, or at least,

50

4.3. Simulation

linger for long without being affected by other factors. For example, in the context of

the post-disaster scenario considered in this study, if a person gets news about losing

someone very close to her, the individual would become sad. In that case, no other

external messages or emotions could surpass her sadness, at lest for a long duration of

time. As an effect, the user’s device would remain inactive for a long period of time.

Proposition 1 formalizes this effect for any emotion, in general. However, we do not

consider this phenomenon further in our work.

Proposition 1. Permanent Effect on Emotions: The permanent effect of an event,

occurring at time tc, on the intensity of the jth emotion, j ∈ {1, 2, · · · , n}, of any

individual is max{ej(t), utc(t)} or min{ej(t), 1 − utc

(t)}, where utc(t) is the unit step

function, i.e., utc(t) = 0, if tc < t; otherwise, utc

(t) is 1. The choice of min or max

depends on the event and emotion considered.

4.3 Simulation

The effects of emotion on the dynamics of MOONs were evaluated using the ONE sim-

ulator [53] with real and synthetic mobility scenarios. We considered the NCSU’s [57]

KAIST real-life mobility traces for 24 hour collected in a terrain of about 33×23 sq. Km

area. We, however, treated the traces bit differently. Since there is no way to identify

which trace belonged to which device, and all these traces were generated independently

of one another, we considered the 92 traces to belong to 92 different individuals. In

other words, 92 nodes moved according to the pre-specified coordinates.

We assigned an initial emotion intensity vector to each such node. These emotions

decayed naturally in an exponential fashion in the absence of any external event. We

also assigned the initial traffic generation rate and the degree of cooperation to each

node. The proportion of different types of messages created in the MOON – Good,

Bad and Unknown – were indicated by the vector pm. Based on the messages received,

51


the individual emotions were affected. This, in turn, influenced the traffic rate and

cooperation, as discussed in Section 4.2.

We also studied at the performance of the MOON considering heterogeneity in the

network. Specifically, we investigated the scenario when all the all human owners of

the devices were not affected by their respective emotions, and, therefore, reflecting

heterogeneous actions. Moreover, we also took the heterogeneity of the devices into

consideration. We considered two types of network interfaces – if1 and if2 – that are

incompatible with one another. The transmission range of the former was taken to be

10 m, while the latter’s range was 20 m. We considered a group of devices equipped

with the interface if1, and the remaining having if2. We evaluated the performance of

such a heterogeneous MOON under different scenarios.

Additionally, we used the post-disaster mobility model developed by Nelson et al. [36].

This scenario involved 60 civilian nodes and 2 ambulances moving in a terrain of size

1000 × 1000 sq. m. with 2 hospitals for 7 hour. A single disaster event of intensity

10000–20000 unit was considered. It may be noted here that this model induces severe

partition of the nodes in the network.

Individual emotions of the nodes were updated every 300 second. Further, the nodes

checked every 900 second whether their DE is fear, with Toff = 900 second. The changes

in the traffic generation rate and switching the device radio on/off were implemented

accordingly. All the users were considered to have same initial intensities of the basic

emotions, and the rate of decay of a given basic emotion was considered to be same for

all. The message generation rate was Poisson with mean inter-message creation time as

25 second in the normal scenario. The transmission speed and range, respectively, were

2 Mbps and 10 m.

The Spray-and-Wait [18] routing protocol (in binary mode with 16 copies) was cho-

sen, since it has a upper limit on the message replication count. However, the emotion-

induced actions would affect the message delivery using other routing protocols as well.

52

4.4. Results

The outcome of our experiments was evaluated based on the following metrics:

• Count of messages generated in a given time interval, and

• Delivery ratio of the messages.

The first metric helps in quantifying the variation in traffic load in the network

depending on the dominant emotion of the users. The second metric, delivery ratio, is

measured as the ratio of the number of messages delivered to their respective destinations

to the number of the messages created in the network.

In all scenarios, wherever appropriate, average of 10 samples were considered and

the 95% confidence interval (CI) corresponding to that were computed. It may be noted

here that although the real-life movement traces considered here would always generate

a deterministic topology, in certain cases, random numbers have been used here – for

example, while implementing the effects of user degree of cooperation – which, to some

extent, results in randomization of the simulation scenarios considered.

4.4 Results

At first, we present the results obtained using the KAIST traces. Figure 4.2 shows the

variation in the average count of messages generated every hour in the network using

Poisson traffic. The effects of emotion on cooperation were not considered in this case.

The case where e =

(

0.5 0.5 0.8

)

, i.e., fear has the highest initial intensity, reflects

the situation that the users exhibited fear/anxiety after a disaster has struck. The Figure

clearly shows that there is a steep increase in the message generation rate when only the

effects of fear are considered.

In Figure 4.2 (a), the effects of the emotions fear, and fear and sadness both, appear

to be similar. This is due to the reason that the intensity of fear in this case is higher

(e3 = 0.8), and its corresponding decay coefficient is the minimum (a3 = 0.001). Thus,

fear tends to be the dominant emotion in the network in this case. However, in Figure

53


50

100

150

200

250

300

350

400

450

500

0 2 4 6 8 10 12 14

Num

ber

of m

essa

ges

Time (hours)No emotion

Fear, e = (0.5,0.5,0.8),a = (0.0020,0.0015,0.0010)

Sadness, e = (0.5,0.5,0.8),a = (0.0020,0.0015,0.0010)

Both, e = (0.5,0.5,0.8),a = (0.0020,0.0015,0.0010)

(a)

0

50

100

150

200

250

300

350

400

450

0 2 4 6 8 10 12 14

Num

ber

of m

essa

ges

Time (hours)No emotion

Fear, e = (0.5,0.8,0.5),a = (0.0020,0.0010,0.0015)

Sadness, e = (0.5,0.8,0.5),a = (0.0020,0.0010,0.0015)

Both, e = (0.5,0.8,0.5),a = (0.0020,0.0010,0.0015)

(b)

Figure 4.2: Fluctuation in traffic rate based on the dominant emotion of the users.

4.2 (b), sadness is the dominant emotion (e2 = 0.8, a2 = 0.001), and as a result of which,

traffic generation almost stops when only sadness is considered.

We also studied at how the messages created in the MOON varied when different

traffic amplifications (β) – as a consequence of fear – were considered. The effects of

54

4.4. Results

sadness on the traffic fluctuation was not considered here. Figure 4.3 plots the temporal

rate of messages generated in the MOON for different amplification factors. The initial

intensity of the emotions (pm) and their decay coefficients (a) are as indicated in the

Figure, along with the proportion of different messages created (pm).

75

150

225

300

375

450

525

0 2 4 6 8 10 12 14

Num

ber

of m

essa

ges

Time (hours)β = 1

β3 = 1.25β3 = 1.5

β3 = 1.75β3 = 2

(a)

75

150

225

300

375

450

0 2 4 6 8 10 12 14

Num

ber

of m

essa

ges

Time (hours)β = 1

β3 = 1.25β3 = 1.5

β3 = 1.75β3 = 2

(b)

Figure 4.3: Variation in traffic rate based on amplification factor when (a) e = (0.5 0.5 0.8)and (b) e = (0.5 0.8 0.5). Here, a = (0.002 0.0015 0.001).

In Figure 4.3, the result with β = 1 represents the scenario when no effects of emo-

55


tions upon the users were considered and the traffic was generated at the normal rate

following the Poisson process. The other results depict the fluctuation for different am-

plification factors (β3), as shown in the Figure. It can be observed that, with increasing

simulation time, the amplified traffic rates tend to converge to the normal rate. This

can be explained by recollecting that the messages in the network were generated for a

given time interval. Therefore, once all such messages are delivered – or lack of thereof

– the decay of individual emotions cease to be affected by the external stimuli. As a

consequence of such mostly-natural decay, after one point of time, intensities of all the

basic emotions tend to zero and thereby, making neutral emotion (∅N ) the dominant

one in most cases. As described earlier, the traffic generation rate was considered to be

invariant, as under normal scenario when the effects of joy or ∅N were considered.

We studied the delivery ratio of the messages obtained in the MOON considered here.

As noted in Chapter 3, this is considered as a typical mission performance parameter.

Figure 4.4 shows the delivery ratio of the messages, when the users switch their devices’

radios on/off as per their contemporary dominant emotion. Emotion-induced traffic

fluctuations were not considered here. We note that the delivery ratio of the messages

when no effects of emotion were considered, was 0.7249 ± 0.00423 with 95% CI.

The plots in Figures 4.4 (a) and (b), respectively, correspond to the decay rates,

a =

(

0.002 0.001 0.0015

)

and a =

(

0.002 0.0015 0.001

)

. In other words, sadness

sustains for a long time in the former case. As a consequence, when only the effects of

sadness are considered, as shown in Figure 4.4 (a), the device radios are turned off for

longer time, and the delivery of the messages become impossible.

On the other hand, in the second case (Figure 4.4 (b)), the effect of fear is prominent.

So, when only the effects of sadness were considered, moderate delivery ratio of the

messages could be observed. Further, in case of sadness, the delivery ratio is maximum

when the number of Bad messages generated – as indicated by pm – is minimum.

Figure 4.5 presents the case when the effects of emotion upon both traffic generation

56

4.4. Results

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Fear Sadness Both

Mes

sage

del

iver

y ra

tio

a = (0.002, 0.001, 0.0015)

pm = (0.1, 0.3, 0.6)pm = (0.2, 0.2, 0.6)pm = (0.2, 0.4, 0.4)

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Fear Sadness Both

Mes

sage

del

iver

y ra

tio

a = (0.002, 0.0015, 0.001)

pm = (0.1, 0.3, 0.6)pm = (0.2, 0.2, 0.6)pm = (0.2, 0.4, 0.4)

(a) (b)

Figure 4.4: Emotion-induced variation in cooperation and its effects on the message de-livery ratio when (a) a = (0.002 0.001 0.0015) and (b) a = (0.002 0.0015 0.001).

and cooperation were considered, with e =

(

0.5 0.5 0.8

)

. The case when the decay

coefficients were a =

(

0.002 0.001 0.0015

)

presents some interesting results. When

only fear was considered, although the initial intensity was high, slow decay of sadness

reduced the traffic generation rate, and, thus, the delivery ratio of the messages was

effectively increased. However, when only the effects of sadness were considered for

the same decay coefficients, the impact of switching off the device radios prevailed over

the reduced traffic generation rate. As a result, the delivery ratio of the messages was

substantially reduced. Finally, when the effects of both sadness and fear were considered

for the same a, traffic spike together with device radio switching off for longer durations

resulted in the lowest delivery ratio.

Furthermore, if we look at these three cases for the other two decay coefficients,

we find that the performance has been almost similar for both the decay coefficients.

This is due to the reason that given the initial intensities of the basic emotions and the

decay coefficients, sadness did not become the dominant emotion for most nodes in the

network.

Figure 4.6 captures the effect of turning off the device radio on the performance

of the MOON. Here, we considered pm =

(

0.2 0.4 0.4

)

, e =

(

0.5 0.5 0.8

)

, and

57


0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

Fear Sadness Both

Mes

sage

del

iver

y ra

tio

Emotion

a = (0.0020, 0.0020, 0.0020)a = (0.0020, 0.0010, 0.0015)a = (0.0020, 0.0015, 0.0010)

Figure 4.5: Effects of emotions on the message delivery ratio considering the variation inboth traffic and cooperation.

a =

(

0.002 0.0015 0.001

)

. In other words, we let fear to be the dominating emotion

among most users in the MOON. Moreover, effects of fear on the traffic amplification was

not considered – all the nodes generated messages with the normal Poisson rate. The

probability of turning off the radio is shown along the x-axis. The maximum delivery

ratio of the messages was obtained when the probability was 0.0, i.e., the case when

the users did not turn off their device radios. It can be observed that the performance

degraded steadily as the probability kept on increasing. This result bears significance in

relating the performance with the action of turning off the device radios only, since we

have deviate from several other considerations for this specific case.

Next, we studied how the individuals’ dominant emotions varied with time as a result

of the network dynamics. The heat maps in Figure 4.7 show the temporal fluctuations

in the dominant emotions of the users, sampled at every hour, for two random cases.

The indices along the color scale indicates the different emotions, starting with neutral

(= −1), joy, sadness and fear (= 2). We note that this variation is not due to the natural

decay of the emotions, but rather due to the impact of the messages received. This could

be clearly understood from Figure 4.7 (a), where, although fear has the highest initial

58

4.4. Results

0.6

0.65

0.7

0.75

0.0 0.25 0.5 0.75 1.0

Mes

sage

del

iver

y ra

tio

Radio off probability

Figure 4.6: Effect of radio off probability on message delivery ratio.

intensity and decays at the slowest rate with a3 = 0.001, the greater number of Bad

messages exchanged in the network (pm2= 0.6) result in the overall dominant emotion

as sadness. Reverse is the case shown in Figure 4.7 (b).

Figure 4.8 presents the empirical CDF of the contacts among the devices in the

network when pm =

(

0.1 0.3 0.6

)

and e =

(

0.5 0.5 0.8

)

. It could be observed

from the Figure that, while about 50% of the total contacts occurred within 5 hour,

when no emotion was considered, the corresponding fraction increased to about 75%,

when only the effects of sadness were considered. This is due to the reason that sadness

dominated in this case. Thus, the total number of contacts in the network were reduced

due to switching off the radios, and most of the contacts had occurred during the initial

time period.

Figure 4.9 (a) shows the effect of traffic explosion (due to fear) and changes in

cooperation (due to sadness) on the delivery ratio of the messages when the actions of

the different percentage of nodes were considered to be affected by their emotions. The

results indicate that the presence of few non-emotional agents (for e.g., throwboxes [58])

in such scenarios may aid the communication process.

Figure 4.9 (b) presents the delivery ratio of the messages in MOON when the de-

59


vices have heterogeneous network interfaces. The first group of the nodes – shown

along the x-axis – used the interface if1, while the remaining used if2. All the nodes

were assumed to be affected by their emotions with e =

(

0.5 0.5 0.8

)

and a =

’data.in’ index 1 matrix

0 15 30 45 60 75 90

Node number

0

5

10

15

20

25

Tim

e (

in h

ours

)

-1

0

1

2

(a)

’data.in’ index 6 matrix

0 15 30 45 60 75 90

Node number

0

5

10

15

20

25

Tim

e (

in h

ours

)

-1

0

1

2

(b)

Figure 4.7: Variation in the dominant emotions of the users when (a) pm =(0.1 0.6 0.3), e = (0.5 0.5 0.8), a = (0.002 0.0015 0.001), and (b) pm = (0.1 0.3 0.6), e =(0.5 0.8 0.5), a = (0.002 0.001 0.0015).

60

4.4. Results

10

20

30

40

50

60

70

80

90

100

0 5 10 15 20 25

% o

f co

nta

cts

Time (in hours)

No emotionSadness, a = (0.002,0.0015,0.001)

Fear, a = (0.002,0.001,0.0015)Both, a = (0.002,0.001,0.0015)

Figure 4.8: CDF of the contacts among the devices.

(

0.002 0.0015 0.001

)

. We, individually, considered the effect of fear on traffic gener-

ation rate and effect of sadness on user cooperation. It is observed that the delivery ratio

reached the maximum when 0% of the devices had if1 implying that all the devices in

the network had a higher transmission range of 20 m. The minimum is, however, reached

when half of all the devices used either network interface resulting in the least number of

message transfer opportunities. A similar “U” shaped curve was also obtained consider-

ing only the effects of sadness. We note here that different values of a were taken while

simulating the scenarios in Figures 4.9 (a) and (b). This was done in order to highlight

the impact of sadness on cooperation.

Finally, we discuss a few observations obtained using the synthetic mobility model.

Since the model results in very low contacts due to network partitioning, we only con-

sidered the effects of emotion on cooperation; traffic explosion due to emotions was not

considered. In the base case – when the nodes did not had any emotion – the deliv-

ery ratio of the messages were found to be about 13.89% with pm =

(

0.2 0.4 0.4

)

.

When only the effects of sadness were considered, the delivery ratio was nearly 0 with

a =

(

0.002 0.001 0.0015

)

; with pm =

(

0.002 0.0015 0.001

)

it was about 8.31%.

61


0

0.2

0.4

0.6

0.8

0 25 50 75 100

Mes

sage

del

iver

y ra

tio

Nodes influenced by emotion (%)

Traffic fluctuations (Fear)Cooperation (Sadness)

(a)

0

0.2

0.4

0.6

0.8

0 25 50 75 100

Mes

sage

del

iver

y ra

tio

Percentage of nodes with if1

Traffic fluctuations (Fear)Cooperation (Sadness)

(b)

Figure 4.9: Performance degradation considering the effects of traffic and cooperationwhen (a) Different percentages of the nodes were affected by emotions, and (b) Differentnetwork interfaces were used by the devices.

sadness being dominant in the former case, device radios were switched off for longer

duration resulting in almost no delivery of the messages. Considering only the effects

of fear, the message delivery ratio were about 12.43% and 10.24%, respectively, for

a =

(

0.002 0.001 0.0015

)

and

(

0.002 0.0015 0.001

)

.


We considered the effects of human emotions in MOONs by using the computational

model of emotions proposed by Meftah et al. [22]. Specifically, we considered the effects

on the offered (Poisson) traffic load and cooperation in the network. The simulation

62


results showed that when the dominant emotion of the individuals was fear, the count of

message generation heavily increased. On the contrary, when the DE was sadness and

people switched off their devices’ radio, the delivery ratio of the messages was almost

zero. While several works focused on detecting the emotions of the users, to the best

of our knowledge, this is the first work that explores the effects of emotion on the

communication in typical MOONs. This study implicates that traffic load management

schemes should be in place, especially in the typical post-disaster scenarios that we

have considered. Moreover, if the entire population is prone to be influenced by their

emotions, deployment of emotion-neutral entities may help the communication scenarios.

63

Chapter 5

Effects of Heterogeneity

In this Chapter, we report the results of the study on the impacts of nodes’ heterogeneity

on the performance of PSNs, and thus MOONs, using opportunistic communications.

We focus on the diversities reflected by the hardware and software of the devices. We

consider the asymmetric (unidirectional) connection events among the devices that can

result in due to diverse transmission ranges. Since the message transfers in MOONs, or

DTNs, in general, follow the store-carry-and-forward paradigm, we look into the impact

of buffer sizes on the network performance. We, then, investigate how incompatible

network interfaces of the mobile devices affect the delivery ratio of the messages. We

also explore the diversity in the software of the devices by considering them running

different routing protocols. To address such diversity, we propose the use of special

nodes, Protocol Translation Units (PTUs). Each PTU runs a hybrid routing protocol,

which encapsulates the functionalities of two or more routing protocols. This enables a

PTU to communicate with a node using a routing protocol that the node understands.

To address the question about the availability of the PTUs in real-life PSNs, we consider

that a certain fraction of the users already posses such devices. This is possible either

when the users purchase such devices, or are promoted by some person/organization.

Finally, we note that, independent of the different heterogeneities, the energy levels of

65

5. Effects of Heterogeneity

the participating devices plays the final role of determining how well a network would

perform.

The specific contributions of this Chapter are summarized as follows:

• Acquiring insights on – and evaluating – the effect of diversity in hardware (specifi-

cally, buffer size and network interfaces) of the devices on the network performance.

• Investigating the interaction of different routing protocols and the resulting per-

formance degradation in the network.

• Using time-varying graphs (TVGs) [59] to represent a heterogeneous PSN, and

defining communication degree to capture the effects of diverse routing protocols.

• Proposing the use of PTUs, which use hybrid routing protocols, to counter such

degradation.

The remainder of this Chapter is organized as follows. Section 5.1 discusses the dif-

ferent source of hardware and software heterogeneities and their effects on the MOON.

In Section 5.2, a TVG is used to represent a PSN and capture its heterogeneity aspect.

Section 5.3 discusses how the adverse effects of heterogeneity could be overcome. It pro-

poses the use of PTUs to bridge incompatible routing protocols. The experimental setup

to evaluate the performance of heterogeneous PSNs is discussed in Section 5.4. Real-life

connection traces were used to simulate the different scenarios. The corresponding re-

sults are presented in Section 5.5, and Section 5.6 summarizes the observations from the

experimental results. Finally, Section 5.7 concludes this Chapter.

5.1 Aspects of Heterogeneity

This Section presents the various aspects that contribute to heterogeneity in a PSN.

These include the heterogeneity in the connection establishment events, and diversities

in both hardware and software of the participating devices in the network.

66

5.1. Aspects of Heterogeneity

Figure 5.1 depicts the effects of heterogeneous node capacities in PSNs. The nodes

A and B, although within the physical transmission ranges of one other, cannot com-

municate due to the existence of diversities in their network interfaces. On the other

hand, the nodes C and D fail to communicate due to incompatibilities in their routing

protocols. Finally, the node F , which has a larger transmission range, can send messages

to node E, but not vice versa.

Figure 5.1: Communication impairments in PSNs due to heterogeneous network interfaces(A–B), incompatible routing protocols (C–D), and diverse transmission ranges (E–F).

5.1.1 Heterogeneity in Connection Dynamics

It may be noted that, the asymmetry in connections need not be limited to the data

rates, but manifested in the entire connection establishment process. Heterogeneity in

the connection dynamics of the devices arises due to one or more of the following reasons:

• Asymmetric transmission ranges and/or speeds

• Diversity in the link-layer protocols of the devices

• Asymmetric device scanning intervals

67


Diverse transmission ranges can result in one-way connectivity between a pair of

devices. Difference in data rates implies that, all other factors remaining the same, some

nodes can transfer less data in a given time than the others. Further, each device scans

for its neighbors periodically after a certain time interval. Devices with variable scan

intervals would affect the frequency of neighbor discovery, and, hence, possibly decreased

number of connection establishment events. Such issues, however, can be induced by the

underlying link-layer protocol of the devices, and their further consideration have been

scoped out in this work.

5.1.2 Diverse Hardware of the Devices

Users’ devices have a fixed buffer size, which determines how many messages can be

stored by a device at any given time. Another potential hardware related issue is the

presence of incompatible devices in the network. For example, let us consider two devices

where the first has a Bluetooth interface, while the other has a Wi-Fi interface. Such de-

vices, although may be within the transmission range of each other, cannot communicate

due to their differences in the network interfaces.

5.1.3 Routing Protocols in DTN and their Compatibility

To discount the effects of intermittent connectivities, several routing schemes based on

message replication have been proposed. In such schemes, multiple copies of a message

are forwarded to different nodes with the hope that at least one copy would reaches the

destination. The simplest scheme in this case is the Epidemic routing [16], where every

node replicates and forwards the messages they are carrying to the nodes not having

the messages’ copies. SnW [18], on the other hand, limits the maximum number of

replication possible in the network. For each message generated, SnW assigns a count

L > 1. Any node having a copy of the message forwards a copy to another node as long

as L > 1. After forwarding, it reduces its own count of copies to L/2 or L−1, depending

68

5.1. Aspects of Heterogeneity

on whether the protocol is run under binary mode or not. Accordingly, the receiving

node would have L/2 or 1 copy of the received message.

In PROPHET [17], a node forwards a replica of a message to another node only if

the other node has greater chances of encountering the destination of the message than

itself. Equations 5.1 through 5.3 [17] relate to updating delivery predictabilities on en-

counter, aging of predictabilities, and transitive predictabilities, govern the functionality

of PROPHET:

P(a,b) = P(a,b)old + (1 − P(a,b)old) × Pinit (5.1)

P(a,b) = P(a,b)old × γk (5.2)

P(a,c) = P(a,c)old + (1 − P(a,c)old) × P(a,b) × P(b,c) × β (5.3)

Here, P(a,b) and P(a,b)old, respectively, indicate the current and previous delivery

predictabilities, i.e., the likelihood that any node a would meet with another node b;

P(a,b) ∈ [0, 1]; Pinit ∈ [0, 1] is an initialization constant. The delivery predictabilities are

aged with time when two nodes do not encounter for long. The parameter γ ∈ [0, 1)

is the aging constant, and k denotes the number of time units expired since the last

update of this predictability. The scaling parameter β ∈ [0, 1] controls the extent to

which transitivity should affect the delivery predictability.

Although these routing protocols help in enhancing the message delivery ratio, most

of the protocols are not compatible with one another, primarily due to two reasons:

• The protocol-specific headers added to the messages while they are created, and

• The operation modes of the protocols, for example, single- or multi-copy routing

and message forwarding/replication criteria.

All messages exchanged in the network have headers containing general (such as, the

69


source and destination of the message) and protocol specific information. For example,

the SnW messages have a header that specifies the number of copies of a given message

is present with the current node. The header of a routing protocol can be interpreted

only by another node using the same protocol. This limits the interoperability of the

routing protocols.

Even if it is assumed that all the routing protocols use a standard header format,

which can be interpreted by all, interoperability among them is further affected by the

way each protocol works. To illustrate, consider the PROPHET protocol. For instance,

when two nodes running PROPHET come in contact with each other, they exchange

(and subsequently update) their respective delivery predictabilities of the other nodes.

This sequence of interaction would fail if, for example, a PROPHET router and a SnW

router comes in contact.

5.1.4 Effects of Incompatibilities

Lack of interoperability among the devices results in the following adverse effects in

PSNs:

• Loss of communication opportunities: Devices cannot communicate, even when

they are near to each other, i.e., communication opportunities are lost.

• Undelivered messages: Certain messages in the network always remain undelivered,

no matter what resource and time are provided.

• Increased delivery latencies: Nodes have less communication opportunities, which

affects, on an average, the time required to deliver the messages.

Thus, it is desirable that such issues are addressed to achieve better performance in

the network.

70

5.2. Representation of Heterogeneous PSNs

5.2 Representation of Heterogeneous PSNs

Casteigts et al. [59] presented the concept of TVGs as G = (V, E, τ, ρ, ζ), where V is the

set of nodes, E is the set of edges, and τ is the lifetime of the system. ρ is called the

presence function, indicating the existence of a particular edge at a given time, and is

represented as ρ : E × τ → {0, 1}. The function ζ represents the (possibly time-varying)

latency involved to travel an edge from one end point to the other.

The TVG model can be used to represent a heterogeneous PSN. In particular, ρ

accounts for multiple scenarios of heterogeneity, for example, devices with incompatible

network interfaces and asymmetric connection events. However, since ρ indicates the

temporal presence of the links, it cannot capture the scenarios when a link exists, but

no communication is possible, such as when diverse routing protocols are used.

Let, Eτ be the set of the edges that exists over the entire network lifetime, i.e.,

Eτ = {e | e ∈ E ∧ ρ(e, t) = 1, t ∈ τ}. Let us define a function φ such that φ(e ∈ Eτ ) = 1,

if the nodes at the two end points of the edge have compatible routing protocol, and 0,

otherwise. Therefore, EC = ∪e∈Eτφ(e) ⊆ Eτ gives the set of potentially communicable

edges in the network i.e., the edges through which messages can be exchanged. A measure

of α = |EC ||Eτ | indicates the communication degree of a PSN resulting due to the diverse

routing protocols, and depends on the number of nodes in the PSN, routing protocols

used by them, as well as their mobility patterns.

A logical question that arises here is – what role, if any, do the nodes play in a hetero-

geneous PSN when there is apparently a possibility of communication? By apparently,

we mean that the link layer of a device indicates that it can communicate with another

device. Even if such a link layer connectivity exists, several factors, for example, energy

levels and routing protocols can prevent the actual communication. Let us consider the

scenario when the messages sent by one node cannot be interpreted by the other due to

the difference in their protocols. In this case, however, both the nodes consume energy

during transmission/reception of the messages. Such an issue can be circumvented if the

71


link layers of the devices advertise the routing protocols used by them, and, therefore, do

not engage in further communication if the other device is not found to use a compatible

protocol.

5.3 Overcoming the Adverse Effects of Heterogeneity

This Section explores how the adverse affects of heterogeneity could be mitigated. The

approach presented here derives from the general concept of bridges discussed in [42].

5.3.1 Hardware Incompatibility

The capacity of existing PSNs with differing network interfaces can be easily augmented

in the presence of devices that are accompanied with multiple types of network interfaces.

For example, a group of devices having only Bluetooth adapters can be bridged to a

group of devices having only Wi-Fi capabilities if they come in contact with one or more

devices having both types of network interfaces.

This work considers two network interfaces – if1 and if2 – that are assumed to be

incompatible with each other. It is considered that certain devices in the PSN have

only either if1 or if2, and the remaining have both. Any device having if1 (if2 ) can

communicate with other devices having if1 (if2 ) or both. Communication is not possible

otherwise. To understand how this helps, consider a device X equipped with both if1

and if2, which comes in proximity of another device Y . If X discovers that Y has if1, it

initiates communication. However, if Y does not have if1, X checks whether Y has if2.

Assuming Y to have the latter, X and Y begins their communication. Any other device

Z, with if1, would not be able to directly communicate with Y even if they are located

close to each other.

72

5.3. Overcoming the Adverse Effects of Heterogeneity

5.3.2 Protocol Translation Units

We addresses the incompatibility issues between two specific routing protocols: PROPHET

and SnW. They are representative of two different categories of routing protocols used

with PSNs/DTNs – routing with (a) Fixed number, and (b) Unlimited copies of the mes-

sages. Moreover, while SnW maintains the state of a message (L), PROPHET considers

the state of connectivity among the nodes (P(a,b)). Thus, PROPHET uses some “intelli-

gence” in decision making. Although variations of these protocols have been proposed,

the principles described here holds good for them as well.

To overcome the communication impairments caused due to heterogeneous routing

protocols, the use of PTUs is proposed. PTUs are “special devices” that interact with

two or more routing protocols both in terms of interpretation of protocol-specific headers

and sequence of interactions. The PTUs run a hybrid routing protocol, encapsulating

the syntax and semantics of both PROPHET and SnW protocols. This enables a PTU

to communicate with both types of routers. This can be further extended to encapsulate

the logic of multiple other protocols.

SnW PTU PROPHET

routerInfo

sendMessages

sendMessages

routerInfo

sendPredictables

sendPredictables

sendMessages

sendMessages

SnW

routerInfo

Figure 5.2: Interactions among different types of routing protocols and PTUs.

73


5.3.2.1 The Use of PTUs

PTUs enable message exchanges among the heterogeneous devices that would have been

infeasible, otherwise. To understand how the PTUs handle the dynamic scenarios arising

in the PSNs, let us consider two devices X and Y using the routing protocols SnW and

PROPHET, respectively. Although X cannot successfully send a message to Y , it can

do so to a PTU device. The latter, in turn, helps in forwarding the message to Y directly

or through other intermediate nodes using the PROPHET routing protocol.

It is considered that the devices periodically emit beacon signals, which also provides

information about the routing protocol used by the respective devices. The PTUs are

assumed to advertise both the routing protocols in their beacon messages. Thus, any

device that is running PROPHET (SnW) initiates communication with the other devices

if the received beacons advertise the use of PROPHET (SnW). Figure 5.2 shows the

interaction among the different routers and the PTUs. In the Figure, the PTU identifies

the routing protocol of the other device, and behaves accordingly. The Figure also shows

the failed interaction between a SnW and a PROPHET router.

Algorithm 5.1 presents the interaction logic between a PTU and any other device

using the PROPHET routing protocol. At the beginning, all the deliverable messages

(i.e., the messages destined for the other device) are transferred. In case any such

message was received from a SnW router, the corresponding SnW headers are removed,

and the PROPHET headers are added before forwarding. Replication of the remaining

messages takes place in the following two phases:

1. In the first phase, all the messages received from the other PROPHET routers are

replicated depending on the delivery predictabilities, as shown in Equations (5.1),

(5.2) and (5.3).

2. Subsequently, any message received from the SnW routers are replicated, provided

L > 1. This ensures that the last copy of the message is directly delivered to the

74

5.3. Overcoming the Adverse Effects of Heterogeneity

destination node.

Algorithm 5.1: Interaction of the PTUs with PROPHET routers

Input:• All messages carried by the device

Output:

• Exchange new messages with the other device

1 for msg in directly deliverable message do

2 if msg has SnW or PTU header then

3 Remove the header4 Add PROPHET header

5 Forward the message

6 for msg in remaining messages do

7 // PROPHET messages

8 if msg does not have SnW header then

9 Replicate and send according to the Equations (5.1), (5.2), and (5.3)

10 for msg in messages do

11 // SnW messages

12 if msg has SnW header with L > 1 then

13 Update header with L = L/214 Replicate, remove header and send

Algorithm 5.2 presents a similar logic of interaction between the PTUs and the SnW

routers. After the directly deliverable messages, if any, are sent, messages originating

from other SnW routers are replicated depending on L. Finally, messages received from

the other PROPHET routers are replicated after heading a new header indicating L.

5.3.2.2 Time Complexity

Let us assume that n messages are generated in the concerned PSN. Thus, a PTU

can have at most n messages in its buffer. It may be noted that in Algorithm 5.1, a

PTU can identify the directly deliverable messages in O(n) time. Moreover, the actions

such as removing/updating message header and replicating/forwarding a message can be

performed in constant time. Therefore, the time complexity of the proposed algorithm

becomes O(n), which is true for the Algorithm 5.2 as well.

75


5.4 Simulation

This Section discusses the experimental setup used to simulate the effects of the above

discussed diversities in PSNs. The simulations were performed using real-life connection

traces. The metrics used to evaluate the performance of the experimental scenarios are

also presented here.

5.4.1 Experimental Setup

The effects of heterogeneity in PSNs were evaluated using the ONE simulator [53].

Real-life connection traces of 78 nodes from the Infocom’06 data set [60] were used for

a duration of 12, 18 and 24 hours, starting from the first connection setup event.

The first scenario explored the possible impacts of asymmetry in the connection

dynamics of the devices. The connection “Up” events, indicated by the device discovery

times and contact durations in [60], were considered to be unidirectional. The scenario

was contrasted with the case when such events were bi-directional.

In the second scenario, the buffer sizes of all the devices were varied from 10 MB

to 145 MB in steps of 15 MB. We studied their effects on the delivery ratio of the

messages. The SnW [18], First contact forwarding (single copy) [53,61], PROPHET [17]

and Epidemic [16] routing protocols were considered in this case. For the last three

protocols, buffer sizes up to 205 MB were considered. Further, the network performance

was investigated by dividing all the nodes into 10 groups with different buffer sizes. The

nodes belonging to each group had buffer sizes from 20 MB to 200 MB, increased in

steps of 20 MB.

Next, the effects of limited energy of the devices on the performance of PSNs were

analyzed. The typical energy consumption rates for Motorola Milestone were consid-

ered (http://www.gsmarena.com/motorola_milestone-3001.php, accessed 10 Decem-

ber 2013). In particular, the initial energy was taken to be 1400 mAh, 3.5 V, and

transmission and scanning energies as 0.7 Joule and 2 Joule, respectively.

76

5.4. Simulation

Algorithm 5.2: Interaction of the PTUs with SnW routers

Input:• All messages carried by the device

Output:

• Exchange new messages with the other device

1 for msg in directly deliverable message do

2 if msg has PROPHET or PTU header then

3 Remove the header4 Add SnW header with L = 1

5 Forward the message

6 for msg in remaining messages do

7 // SnW messages

8 if msg has SnW header with L > 1 then

9 Set L = L/2, replicate and send

10 for msg in messages do

11 // PROPHET messages

12 if msg does not have SnW header then

13 Replicate msg, add SnW header with L, and send

We investigated the effects of incompatible network interface of the devices. A frac-

tion of nodes with two network interfaces, if1 and if2, were considered. Half of the other

nodes used if1, while the remaining nodes had if2. These two types of network inter-

faces were considered to be incompatible with one another. The transmission speeds

of both the interfaces were set to 2 Mbps. The effects of the presence of nodes with

multiple network interfaces was then studied. The SnW routing protocol was used for

this scenario.

Next, we explored the interactions of two different routing protocols – PROPHET

and SnW. We divided the nodes in the MOON into two groups. The ones in the first

group used the SnW routing protocol. All the nodes in the other groups used either

1) SnW or 2) PROPHET or 3) were PTUs. Similarly, in the next scenario, we also

considered all the nodes in the first group to use PROPHET, while the composition of

the second group varied as earlier. All the nodes were considered to have infinite buffer

77


sizes in order to eliminate the effects of the buffer size, if any, on the evaluation metrics.

In the final scenario, the variation in communication degree was explored. In the first

case, we divided the 78 nodes into two groups. The first group contained the 10−50% of

the nodes incremented in steps of 10%; the other group contained the remaining nodes.

In the two other cases, we considered 5 and 10 nodes, respectively, to be the PTUs. The

remaining 68 nodes were divided into two groups in a similar way.

We considered uniform message creation events within the first 5 hour of the sim-

ulation duration. A total of 400 messages were created, with their sizes uniformly dis-

tributed between 50 KB to 1 MB. The time-to-live (TTL) for the messages were set

larger than the simulation durations. In all the scenarios, SnW was used in the binary

mode with L = 16, unless otherwise specified. For the PROPHET router, the settings

mentioned in [17] were taken. Finally, except for the first scenario described before, all

other connection events were considered to be symmetric.

5.4.2 Performance Metrics

The performance of the proposed solution was evaluated based on the following metrics:

• Delivery ratio of the messages,

• Delivery ratio versus delivery latencies, and

• Communication degree of the MOON.

The delivery ratio gives the fraction of the created messages that were delivered to

the respective destinations. A measure of this metric evaluates the effectiveness of any

message forwarding scheme. Let, M and Md, respectively, denote the number of mes-

sages created and delivered in the MOON. Then, the delivery ratio of the messages is

determined as Md/M . The delivery ratio primarily depends on communication oppor-

tunities among the nodes, and, thus, serve as an important metric in quantifying the

performance of MOONs.

78

5.5. Results

A plot of the delivery ratio of the messages versus delivery latencies provides insights

in understanding the different components of the delay associated with the delivered

messages. This is different from the empirical distribution of the delivery latencies,

which only shows the cumulative measures of the latencies.

The average delivery latency of the messages is computed by taking the average over

delivery latencies of all the messages delivered in the network. Let, tci and td

i , respectively,

denote the times when the ith message was created by the source node and delivered

to the destination node. By delivery, we mean the event when the destination node

receives the first copy of a given message – the subsequently received copies, if any, are

discarded. Thus, the average delivery latency of a message in the MOON is defined as:

δ =1

Md

Md∑

i=1

(tdi − tc

i).

Typically, this value is high in DTNs, as compared to the Internet or other networks.

We note here that since the concerned real-life connection trace would always gener-

ate a deterministic topology, and our algorithms did not have any randomness, results

from a single scenario are provided – no average, and therefore, confidence intervals –

were computed.

5.5 Results

This Section presents the results of the simulations, and related analysis.

5.5.1 Effects of Heterogeneous Connection Events

Figure 5.3 shows the impact on message delivery ratio when (all) the nodes used either

the SnW or PROPHET routing protocols. The “true” and “false” cases, respectively,

indicate the scenarios whether the connection events were considered to be symmetric

or not. It can be observed that, for the lesser durations of simulation (or low message

79


density per unit time), the asymmetry in connection among the devices reduces the

delivery ratio of the messages by 30 − 40%. When sufficient time is given (the 24-hour

case), the ratio improves significantly. This underscores the fact that, given enough time,

the messages in the network are eventually delivered.

20

40

60

80

100

12 18 24

Mes

sage

s de

liver

ed (

%)

Simulation time (hour)PROPHET (true)

PROPHET (false)SnW (true)

SnW (false)

Figure 5.3: Effects of (a)symmetric connection events on the delivery ratio of the messages.

5.5.2 Effects of Buffer Size

At first we studied the performance of SnW with different number of copies of the

messages (L) when the nodes had different buffer sized, as shown in Figure 5.4. The

devices ran the SnW routing protocol, with 8−, 16−, 24− and 32−copies, as indicated

in the Figure. It can be observed that the buffer size strongly influenced the delivery

ratio when the simulation durations were for 12 hours. This is due to the reason that

in the absence of any TTL expiry, increased buffer size enabled the nodes to exchange

greater number of messages during the available communication opportunities. When

the simulation duration was increased, the nodes in the network eventually met with

one another, and delivered the messages.

The effects of the number of copies used by the SnW protocol on the delivery ratio of

the messages is also evident. It could be seen that, as the number of copies increased, for

80

5.5. Results

a given buffer size, the delivery ratio also increased. The graphs indicate that unlimited

buffer sizes may not help in achieving higher delivery of the messages.

30

40

50

60

70

80

90

100

0 25 50 75 100 125 150

Mes

sage

s de

liver

ed (

%)

Buffer size (MB)12 hour 18 hour 24 hour

60

70

80

90

100

0 25 50 75 100 125 150

Mes

sage

s de

liver

ed (

%)


(a) (b)

60

70

80

90

100

0 25 50 75 100 125 150

Mes

sage

s de

liver

ed (

%)


60

70

80

90

100

0 25 50 75 100 125 150

Mes

sage

s de

liver

ed (

%)


(c) (d)

Figure 5.4: Delivery percentage of the messages for different buffer sizes under SnW with(a) 8-, (b) 16-, (c) 24- and (d) 32-copies.

Figure 5.4 also reveals that the rate of increase in the delivery ratio of the messages is

faster for buffer sizes up to 40 MB. Although this threshold is specific to the parameters

that were considered (i.e., message size and rate), this indicates that if more messages

are to be delivered within a short time, the nodes should have a certain minimum buffer

size.

Figure 5.5 shows the message delivery ratio with different buffer sizes when First

contact [53,61], PROPHET [17] and Epidemic [16] routing protocols were used. In First

contact routing, a node forwarded (not replication) any message it had to the first node

that it came in contact with. Figures 5.5 (b) and (c) indicate that with no fixed limit on

81


message replication, larger buffer sizes enhanced the delivery ratio. This is due to the

reason that during each communication opportunity, more number of nodes got a copy

of a message, and, thereby, increased their respective chances of delivery.

As a final case, the scenario when the devices have different buffer capacities, was

considered. Figure 5.6 presents the delivery ratio of the messages obtained after 12, 18,

and 24 hours of simulation time using the PROPHET and the SnW (16−, 24− and

32−copies) protocols. It can be observed that, the delivery ratio decreased for the 12

hour-case, as compared to the corresponding maxima in Figures 5.4 and 5.5.

Figures 5.4 and 5.5 provide some insights on how buffer size and message forwarding

schemes are related:

• With a fixed limit on message replication, consideration of excessive buffer sizes is

not useful. This is reflected by the performances of the SnW and the first contact

routing protocols.

• With no fixed limit on the number of message replicas, larger buffers enhance

the delivery ratio. This behavior, verified from the evaluation of the PROPHET

and Epidemic protocols, is due to the reason that during each communication

opportunity, more number of nodes get a copy of a message.

Figure 5.7 plots the delivery ratio of the messages versus the delivery latencies. The

devices had 200 MB of buffer space, and all the nodes used either PROPHET, SnW or

Epidemic protocol. Figure 5.7 (a) presents the base case when devices had unlimited

energy. Figure 5.7 (b) corresponds to the case when the devices had limited energy. It

can be observed that limited energy budgets significantly degraded the performance in,

otherwise, homogeneous MOONs. Further, the diversity in initial energies of the devices

worsened the delivery ratio (Figure 5.7 (b)) compared to the scenario when all the nodes

had the same initial energies (indicated by “Same” in the graph). Figure 5.7 (a) also

shows that 60% of the messages cannot be delivered within 1000 seconds time duration.

82

5.5. Results

30

40

50

60

70

80

0 25 50 75 100 125 150 175 200 225

Mes

sage

s de

liver

ed (

%)


(a)

30

40

50

60

70

80

90

100

0 25 50 75 100 125 150 175 200 225

Mes

sage

s de

liver

ed (

%)


(b)

20 30 40 50 60 70 80 90

100

0 25 50 75 100 125 150 175 200 225

Mes

sage

s de

liver

ed (

%)


(c)

Figure 5.5: Effects of buffer sizes on the message delivery ratio using (a) First contact, (b)PROPHET and (c) Epidemic routing.

83


70

75

80

85

90

95

100

12 18 24Mes

sage

s de

liver

ed (

%)

Simulation duration (hours)SnW (16 copies)SnW (24 copies)

SnW (32 copies)PROPHET

Figure 5.6: Variation in the delivery ratios of the messages under different protocols whenthe nodes have different buffer sizes.

5.5.3 Impact of Incompatible Networking Devices

Figure 5.8 (a) shows how the delivery ratio of the messages is affected when a fraction of

the devices have network interface if2, while the others used if1. The 0% case represents

the scenario when all the nodes had if1, i.e., the network was homogeneous. It can be

observed that the delivery ratio steadily decreased as long as 20% of the devices have

incompatible network interfaces. This can be explained by considering the fact that all

the nodes were partitioned into two mutually exclusive groups, based on their network

interfaces. As the size of each such group increases, more number of the nodes failed to

exchange the messages among themselves, which reduced the delivery ratio.

The impact on the message delivery ratio in the presence of the nodes with dual

network interfaces is shown in Figure 5.8 (b). The Figure suggests that the participation

of such nodes in the network highly enhanced the delivery ratio of the messages. A steady

increase can be observed till 20% presence of such nodes. This is due to the reason that

the remaining nodes with either if1 or if2 got opportunities to transfer their messages to

each other through the nodes with dual network interfaces. Beyond the 20% limit, the

84

5.5. Results

0

0.2

0.4

0.6

0.8

1

100 101 102 103 104 105

Mes

sage

del

iver

y ra

tio

Delivery latency in seconds (log scale)

PROPHETSnW

Epidemic

(a)

0

0.25

0.5

0.75

0 50 100 150 200 250

Mes

sage

del

iver

y ra

tio

Latency (in seconds) × 100

PROPHETSnW

EpidemicPROPHET (Same)

SnW (Same)Epidemic (Same)

(b)

Figure 5.7: Delivery ratio versus delivery latencies of the delivered messages (a) withoutand (b) with energy constraints.

delivery ratio stagnates and reaches the maximum level as determined by the routing

protocol.

5.5.4 Effects of Heterogeneous Routing Protocols

Figures 5.9 (a) and (b) present the delivery ratio of the messages when different routing

protocols were considered. Figure 5.9 (a) shows the case when certain fraction of the

nodes – shown along the x-axis – used SnW, while the remaining nodes used some

85


20

40

60

80

100

0 10 20 30 40 50

Mes

sage

s de

liver

ed (

%)

Percentage of incompatible devices

12 hour18 hour24 hour

(a)

20

40

60

80

100

0 10 20 30 40

Mes

sage

s de

liver

ed (

%)

Nodes with two network interfaces (%)

12 hour18 hour24 hour

(b)

Figure 5.8: Effect of different networking interfaces when: (a) Different percentage ofthe nodes had incompatible network interface if2, and (b) The nodes had dual networkinterfaces.

other routing protocol. In Figure 5.9 (a), the plots labeled with “SnW (k hour)” denote

the base case performances when all the nodes used the SnW routing protocol and the

simulation duration was k hour.

The plots with labels “PROPHET (k hour)” represent the scenarios when different

fraction of the population (shown along the x−axis) used the PROPHET protocol. It

can be observed that, in comparison to the base cases, when the fraction of the nodes

using PROPHET protocol increased, the delivery ratio drastically decreased. Finally,

the delivery ratio obtained with equivalent fraction of PTU nodes are shown. It can be

observed that, while varying the fraction of PTUs from 10% to 50% in the network, the

86

5.5. Results

20

40

60

80

100

10 20 30 40 50

Mes

sage

del

iver

ed (

%)

Percentage of other routing protocol typeSnW (12 hour)SnW (18 hour)SnW (24 hour)

PROPHET (12 hour)PROPHET (18 hour)PROPHET (24 hour)

PTU (12 hour)PTU (18 hour)PTU (24 hour)

(a)

20

40

60

80

100

10 20 30 40 50

Mes

sage

del

iver

ed (

%)

Percentage of other routing protocol typePROPHET (12 hour)PROPHET (18 hour)PROPHET (24 hour)

SnW (12 hour)SnW (18 hour)SnW (24 hour)

PTU (12 hour)PTU (18 hour)PTU (24 hour)

(b)

Figure 5.9: Percentage of messages delivered in presence of different types of routingprotocols together with (a) SnW and (b) PROPHET.

delivery ratio obtained is almost the same as the best cases considered.

Figure 5.9 (b) plots similar performance measurements when some of the nodes used

the PROPHET routing protocol, while the other nodes used SNW or was a PTU.

Finally, the variation in the communication degree (α) of the MOON is shown in

87


Figure 5.10. It shows that with the increasing group sizes, α sharply decreased. However,

in the worst, case when both the groups had equal number of nodes, the presence of 10

PTUs enhanced α by 12%.

0.5

0.6

0.7

0.8

0.9

1

10 20 30 40 50

Com

mun

icat

ion

degr

ee

Percentage of nodes

No PTUs5 PTUs

10 PTUs

Figure 5.10: Communication degree with different percentage of nodes in the first group.

5.6 Observations

The observations from Section 5.5 are summarized in the following points:

• When the time window considered is small, for multi-copy message forwarding, the

buffer size plays a significant role. However, with large enough message TTLs, the

nodes eventually meet, and deliver the messages.

• Heterogeneous connection dynamics (the simplest case due to different transmis-

sion ranges) substantially reduces the delivery ratio of the messages.

• Hardware incompatibility arising due to incompatible network interfaces is hard

to address particularly because, one may opt for software upgrade, but not for

purchasing a new phone. Therefore, any contact opportunity with devices with

multiple interfaces should be used to the best. This may require the routing

protocols to use information available from the link-layer of the devices.

88


• For approaching reality, any new protocol or mechanism proposed should take

energy consumption of the nodes into consideration.

• The performance degradation due to software-based incompatibility among the

routing protocols is severe, but could be prevented. This does not require all the

users to update their software. Rather, the presence of few “special” devices (for

example, devices with middlewares, or the PTUs as proposed here) could boost

the performance.


PSNs present an interesting communication paradigm, especially in the absence of global

network connectivities. The performance of the PSNs, however, can heavily degrade in

the face of various diversities manifested by the hardware and software of the devices. In

this work, the effects of such degradation were quantified through extensive simulations.

To counter the negative impacts of the heterogeneous routing protocols used by the

devices, the use of PTUs has been proposed. The results of performance evaluation

showed that the use of PTUs can elevate the message delivery ratio to the value obtained

in a homogeneous network. Thus, it would be desirable to deploy such devices in post-

disaster scenarios, which already shows communication impairments due to various other

reasons.

89

Chapter 6

Conclusion

In this Thesis, we proposed the notion of MOONs as an extension of the PSNs, by

considering mission objectives in the presence of opportunistic contacts. However, unlike

other communication networks, MOONs exhibit the human-network interplay – human

owners of the mobile devices in MOONs affect the network dynamics and vice versa. In

the preceding Chapters, we looked at the effects of actions driven by two specific human

aspects – intuitive intelligence and emotion – on the performance of MOONs. Moreover,

we also studied how heterogeneity in different aspects degraded the performance of the

MOONs.

In this final Chapter, we take a holistic view of our work discussed so far. We

summarize the contributions and juxtapose them with our research goals as elicited in

Section 1.3. We also look at some of the limitations, and finally discuss the directions

along which this work can be further extended.

6.1 Summary

In Chapter 3, we considered a MOON formed in the aftermath of a large-scale disaster.

The objective of the mission considered was to increase the communication opportunities

between mobile nodes (for example, paramedics and firefighters) and stationary nodes

91

6. Conclusion

(for example, just rescued victims). We defined four levels of intuitive human intelligence

that can be typically observed in such scenarios. As per these levels, when a mobile node

came to know about the location of any unvisited stationary node, it visited that new

stationary node, as well as, its vicinity. The four levels restricted the extent of the

neighborhood to be explored. The movement of the mobile (human) nodes influenced

by their respective level of intelligence were considered. It was found that, on an average,

the number of encounters between a mobile and stationary nodes increased when one

such level was employed in contrast to the simple RWP mobility model. Moreover, the

unique number of encounters between a mobile and a stationary node, on an average,

was also enhanced to different extents for the concerned levels of intelligence. However,

performance degradation was visible in the presence of heterogeneity – for example,

when different people moved according to different intelligence levels and when not all

the users had the relevant application available in their devices.

In Chapter 4, we looked at another human aspect, emotion, and studied its interplay

with the dynamics of MOON by considering its effect upon the traffic rate and the

degree of user cooperation. A computational model of emotions proposed by Meftah et

al. [22] was used in this case. We considered the scenario where traffic generation rate

increased (decreased) when the contemporary dominant emotion of the users was fear

(sadness). Moreover, the degree of user cooperation – the extent to which a user kept

her device radio on – was also varied according to the dominant emotions. While a sad

user was assumed to switch off the device radio for a certain duration of time, a user,

whose dominant emotion was fear, did so with a probability of 0.5. On the other hand,

emotions of the users varied as well based on the type of information (messages) received

over time. It was found that when only the effects of switching off the device radios were

considered, the delivery ratio of the messages drastically reduced as compared to the

case when only the effects of traffic fluctuation were considered.

In Chapter 5, we studied the sources and effects of different forms of heterogeneity on

92

6.2. Discussion

the performance of the MOONs. We considered the scenarios where devices had asym-

metric transmission ranges, diverse buffer sizes and initial energy levels, incompatible

network interfaces and heterogeneous routing protocols. For the latter case, we specifi-

cally considered two routing protocols – SnW and PROPHET – which are representative

of two broad classes of routing protocols in DTNs, in general. We used the TVGs [59]

to represent a MOON with incompatible routing protocols. To bridge such incompati-

bilities, we proposed the use of PTUs, which a re hybrid devices, and can communicate

with two or more routing protocols. The use of a few PTUs was found to considerably

lift the delivery ratio of the messages in the otherwise heterogeneous MOONs. On the

other hand, the presence of few devices with dual network interfaces was also found

to bridge the communication gap. Certain forms of diversity that can be observed in

human beings – for example, different levels of intelligence and effect of emotions – were

already studied in their relevant contexts in Chapters 3 and 4.

6.2 Discussion

Multiple works [14, 31–33, 55, 56] have studied human emotions in one form or another

– for example, exploring the effects of human emotions on their actions and detecting

emotions based on their smartphone usage. However, the broad picture lacked in explor-

ing an interesting aspect – how emotion driven action of the human owners would affect

the performance of the network formed by their devices, which we attempted to address

in Chapter 4. The case of traffic explosion considered here is unlike in other networks, as

already noted in Section 1.2. The characteristics of MOONs are not only different from

the Internet and cellular networks, but also from the well known mobile ad hoc networks.

Since MOONs lack end-to-end communication paths, message forwarding solely relies

upon the pair-wise contacts in the network. Thus, the emotion-driven action of even

a few users – whether in terms of switching off the device or its radio, or generating

excessive messages – would considerably hamper the performance. Further, not only

93

6. Conclusion

would the source-destination pair be affected, but, to different extent, the other nodes in

the network as well. This is in sharp contrast to the traditional networks. For example,

in a cellular network, if the destination does not pick up the call, only she would be

deprived of the message.

On the other hand, certain scenarios, for example, the aftermath of a large-scale

disaster as considered here, witnesses people exhibiting strong emotions. Therefore,

in such and similar scenarios, it would be helpful to deploy emotion-neutral entities

(for example, throwboxes) to prevent the communication break down. Further more,

bridging devices (for example, PTUs and devices with dual interfaces) would also help

in augmenting the performance.

While actions based on human emotions apparently have a negative effect on the

performance of the MOONs, the influence of another human aspect – intuitive human

intelligence – was found to be positive. In particular, decision on movement actions of

the mobile nodes based on such intuitive intelligence was found to increase the commu-

nication opportunities in the MOON. Although this was considered in a specific context

in this Thesis, the conclusion can be easily generalized. To understand this, consider

a set of points of interests (PoIs) – for example, shopping malls and coffee shops –

where people visit often and repeatedly over time, and therefore, increase the chances

of communication among the devices carried by them.

Finally, heterogeneity, indeed, has a negative effect upon the network performance,

which, therefore, stands as an obstacle in achieving mission objectives through oppor-

tunistic communications. This work addressed multiple measures to overcome such

adversities in real life. However, heterogeneity is not only limited to the composition

of the networking devices, as is usually considered. Diversity in human behavior and

action also bring forwards certain aspects of heterogeneity, which, however, is difficult

to address, and was beyond the scope of this Thesis.

In conclusion, while intuitive intelligence-based movement decisions are helpful, con-

94

6.3. Limitations and Future Scope of Work

sideration of certain human aspects, together, with heterogeneity, renders accomplishing

a mission with opportunistic communications a challenging task.

6.3 Limitations and Future Scope of Work

This work, while addresses multiple previously unexplored issues, can be further investi-

gated in multiple directions. The following provides a brief idea about the future scope

of work.

• Mission aspects in details: The consideration of missions has been mostly im-

plicit in the current work. In the future, this can be addressed in detail considering

each of its components, for example, distribution definition of the objectives and

cooperation among the users – based on their preferences – to accomplish the same.

• Other human factors: Apart from intuitive intelligence and emotions, there are

other human aspects (for example, human trust and expectations) that can be

taken into consideration. Moreover, other details of the aspects already considered

here, for example, emotional contagion, could be taken into account.

• Privacy and security: While the broad area of pervasive and ubiquitous com-

puting presents a promising field, the concerns of privacy and security need to

be addressed. The following presents an interesting question – given the approxi-

mate profiles of intelligence and emotions of the users, can such users’ actions be

exploited to affect the communication in a MOON?

• Further realistic analysis: One of the critical challenges faced in this work

was the lack of sufficient real-life data to be used in our experiments. It would

be interesting to device mechanisms to facilitate the collection of such data and

interpret the results presented here in such light.

95

Publications

Journal

• S. Misra and B. K. Saha, “On Emotional Aspects in Mission-Oriented Oppor-

tunistic Networks,” IET Networks, DOI: 10.1049/iet-net.2013.0080 (Accepted: Oc-

tober 2013).

• B. K. Saha and S. Misra, “Effects of Heterogeneity on the Performance of Pocket

Switched Networks,” IET Networks, DOI: 10.1049/iet-net.2013.0069 (Accepted:

July 2013).

Conference

• B. K. Saha and S. Misra, “Could Human Intelligence Enhance Communication

Opportunities in Mission-Oriented Opportunistic Networks?” in Proc. of the

1st ACM MOBICOM Workshop on Mission-Oriented Wireless Sensor Network-

ing (MiSeNet ’12), Istanbul, Turkey. ACM, Aug. 2012, pp. 15–20.

97

References

[1] F. Warthman. (2003, Mar.) Delay-tolerant networks (dtns): Atutorial v1.1. Accessed: 08 Oct. 2012. [Online]. Available:http://www.dtnrg.org/docs/tutorials/warthman-1.1.pdf

[2] K. Fall, “A delay-tolerant network architecture for challenged internets,” in Pro-ceedings of the 2003 Conference on Applications, Technologies, Architectures, andProtocols for Computer Communications (SIGCOMM ’03). New York, NY, USA:ACM, 2003, pp. 27–34.

[3] P. Hui, A. Chaintreau, J. Scott, R. Gass, J. Crowcroft, and C. Diot, “Pocketswitched networks and human mobility in conference environments,” in Proceed-ings of the 2005 ACM SIGCOMM Workshop on Delay-tolerant networking (WDTN’05). New York, NY, USA: ACM, 2005, pp. 244–251.

[4] S. Phoha, “Guest editorial: Special section on mission-oriented sensor networks,”IEEE Transactions on Mobile Computing, vol. 3, no. 3, p. 209, 2004.

[5] J. Burke, D. Estrin, M. Hansen, A. Parker, N. Ramanathan, S. Reddy, and M. B.Srivastava, “Participatory sensing,” in Workshop on World-Sensor-Web (WSW’06):Mobile Device Centric Sensor Networks and Applications, Boulder, Colorado, USA,2006, pp. 117–134.

[6] M. Srivastava, T. Abdelzaher, and B. Szymanski, “Human-centric sensing,”Philosophical Transactions of the Royal Society A: Mathematical,Physical andEngineering Sciences, vol. 370, no. 1958, pp. 176–197, Jan. 2012. [Online].Available: http://dx.doi.org/10.1098/rsta.2011.0244

[7] A. Chaintreau, P. Hui, J. Crowcroft, C. Diot, R. Gass, and J. Scott, “Impact of hu-man mobility on the design of opportunistic forwarding algorithms,” in Proceedingsof the INFOCOM 2006, Barcelona, Spain, April 2006, pp. 1–13.

[8] T. Karagiannis, J.-Y. Le Boudec, and M. Vojnović, “Power law and exponentialdecay of inter contact times between mobile devices,” in Proceedings of the 13th

Annual ACM International Conference on Mobile Computing and Networking, ser.MobiCom ’07. New York, NY, USA: ACM, 2007, pp. 183–194. [Online]. Available:http://doi.acm.org/10.1145/1287853.1287875

99

References

[9] I. Rhee, M. Shin, S. Hong, K. Lee, S. J. Kim, and S. Chong, “On the levy walk natureof human mobility: Do humans walk like monkeys?” IEEE/ACM Transactions onNetworking, vol. 19, no. 3, pp. 630–643, Jun. 2011.

[10] J. Vincent, “Emotional attachment to mobile phones: an extraordinary relation-ship,” in Mobile World, L. Hamill, A. Lasen, and D. Diaper, Eds. Springer, 2005,pp. 93–104.

[11] ——, “Emotion and the mobile phone,” Cultures of participation: Media practices,politics and literacy, pp. 95–109, 2011.

[12] F. Ferri, I. P. Stoianov, C. Gianelli, L. D’Amico, A. M. Borghi, and V. Gallese,“When action meets emotions: How facial displays of emotion influence goal-relatedbehavior,” PLoS ONE, vol. 5, no. 10, p. e13126, 10 2010. [Online]. Available:http://dx.doi.org/10.1371%2Fjournal.pone.0013126

[13] J. Zhu and P. Thagard, “Emotion and action,” Philosophical Psychology, vol. 15,no. 1, pp. 19–36, 2002.

[14] J. Tang, Y. Zhang, J. Sun, J. Rao, W. Yu, Y. Chen, and A. C. M. Fong, “Quanti-tative study of individual emotional states in social networks,” IEEE Transactionson Affective Computing, vol. 3, no. 2, pp. 132–144, 2012.

[15] C. Liu and J. Wu, “On multicopy opportunistic forwarding protocols in nondeter-ministic delay tolerant networks,” IEEE Transactions on Parallel and DistributedSystems, vol. 23, no. 6, pp. 1121 –1128, June 2012.

[16] A. Vahdat and D. Becker, “Epidemic routing for partially-connected ad hocnetworks,” Duke University, Tech Report CS-2000-06, 2000. [Online]. Available:http://issg.cs.duke.edu/epidemic/epidemic.pdf

[17] A. Lindgren, A. Doria, and O. Schelén, “Probabilistic routing in intermittentlyconnected networks,” in Proceedings of the 1st International Workshop on ServiceAssurance with Partial and Intermittent Resources (SAPIR), ser. Lecture Notesin Computer Science, P. Dini, P. Lorenz, and J. Souza, Eds., vol. 3126. Berlin,Heidelberg: Springer Berlin / Heidelberg, Aug. 2004, pp. 239–254.

[18] T. Spyropoulos, K. Psounis, and C. S. Raghavendra, “Spray and wait: an efficientrouting scheme for intermittently connected mobile networks,” in Proceedings of the2005 ACM SIGCOMM workshop on Delay-tolerant networking (WDTN ’05). NewYork, NY, USA: ACM, 2005, pp. 252–259.

[19] E. Ayday and F. Fekri, “An iterative algorithm for trust management and adversarydetection for delay-tolerant networks,” IEEE Transactions on Mobile Computing,vol. 11, no. 9, pp. 1514 –1531, Sept. 2012.

[20] N. Thompson, S. Nelson, M. Bakht, T. Abdelzaher, and R. Kravets, “Retiring repli-cants: Congestion control for intermittently-connected networks,” in Proceedings ofIEEE INFOCOM, 2010, March 2010, pp. 1 –9.

100

References

[21] R. Plutchik, Emotion, a Psychoevolutionary Synthesis. Harper and Row, 1980.

[22] I. T. Meftah, N. L. Thanh, and C. B. Amar, “Towards an algebraic modeling ofemotional states,” in 5thInternational Conference on Internet and Web Applicationsand Services (ICIW), Barcelona, Spain, May 2010, pp. 513–518.

[23] R. Rao and G. Kesidis, “Purposeful mobility for relaying and surveillance in mobilead hoc sensor networks,” IEEE Transactions on Mobile Computing, vol. 3, no. 3,pp. 225–232, 2004.

[24] S. Eswaran, A. Misra, F. Bergamaschi, and T. L. Porta, “Utility-based bandwidthadaptation in mission-oriented wireless sensor networks,” ACM Trans. Sen. Netw.,vol. 8, no. 2, pp. 17:1–17:26, Mar. 2012.

[25] C. H. Liu, K. K. Leung, C. Bisdikian, and J. W. Branch, “A newapproach to architecture of sensor networks for mission-oriented applications,”Proc. SPIE, vol. 7349, pp. 73 490L–73 490L–12, 2009. [Online]. Available:http://dx.doi.org/10.1117/12.820199

[26] A. Campbell, S. Eisenman, N. Lane, E. Miluzzo, R. Peterson, H. Lu, X. Zheng,M. Musolesi, K. Fodor, and G.-S. Ahn, “The rise of people-centric sensing,” IEEEInternet Computing, vol. 12, no. 4, pp. 12 –21, July-Aug. 2008.

[27] A. T. Campbell, S. B. Eisenman, N. D. Lane, E. Miluzzo, and R. A. Peterson,“People-centric urban sensing,” in Proceedings of the 2nd Annual InternationalWorkshop on Wireless Internet, ser. WICON ’06. New York, NY, USA: ACM,2006.

[28] L. Becchetti, A. E. F. Clementi, F. Pasquale, G. Resta, P. Santi, and R. R. Silvestri,“Information spreading in opportunistic networks is fast,” CoRR abs/1107.5241,2011.

[29] P. Hui, E. Yoneki, S. Y. Chan, and J. Crowcroft, “Distributed community detectionin delay tolerant networks,” in Proceedings of 2nd ACM/IEEE InternationalWorkshop on Mobility in the Evolving Internet Architecture, ser. MobiArch’07. New York, NY, USA: ACM, 2007, pp. 7:1–7:8. [Online]. Available:http://doi.acm.org/10.1145/1366919.1366929

[30] S. C. Nelson, “Leveraging structure for communication in human-centric dtns,”Ph.D. dissertation, University of Illinois at Urbana-Champaign, 2011.

[31] R. LiKamWa, Y. Liu, N. D. Lane, and L. Zhong, “Can your smartphone infer yourmood?” in 2nd International Workshop on Sensing Applications on Mobile Phones(PhoneSense 2011), Seattle, WA, USA, 2011.

[32] H. Lee, Y. S. Choi, S. Lee, and I. Park, “Towards unobtrusive emotion recognitionfor affective social communication,” in 2012 IEEE Consumer Communications andNetworking Conference (CCNC), Las Vegas, NV, USA, Jan. 2012, pp. 260 –264.

101

References

[33] H.-J. Kim and Y. S. Choi, “Exploring emotional preference for smartphone ap-plications,” in 2012 IEEE Consumer Communications and Networking Conference(CCNC), Las Vegas, NV, USA, Jan. 2012, pp. 245–249.

[34] C. Gao and J. Liu, “Clustering-based media analysis for understanding humanemotional reactions in an extreme event,” in Proceedings of the 20th InternationalConference on Foundations of Intelligent Systems, China, 2012, pp. 125–135.

[35] T. Sakaki, F. Toriumi, and Y. Matsuo, “Tweet trend analysis in anemergency situation,” in Proceedings of the Special Workshop on Internet andDisasters. New York, NY, USA: ACM, 2011, pp. 3:1–3:8. [Online]. Available:http://doi.acm.org/10.1145/2079360.2079363

[36] S. C. Nelson, A. F. Harris, III, and R. Kravets, “Event-driven, role-based mobilityin disaster recovery networks,” in Proceedings of CHANTS ’07. New York, NY,USA: ACM, 2007, pp. 27–34.

[37] M. Uddin, D. Nicol, T. Abdelzaher, and R. Kravets, “A post-disaster mobilitymodel for delay tolerant networking,” in Proceedings of the 2009 Winter SimulationConference (WSC), Austin, TX, USA, Dec. 2009, pp. 2785 –2796.

[38] N. Aschenbruck, E. Gerhards-Padilla, and P. Martini, “Modeling mobility indisaster area scenarios,” Perform. Eval., vol. 66, no. 12, pp. 773–790, Dec. 2009.[Online]. Available: http://dx.doi.org/10.1016/j.peva.2009.07.009

[39] S. Saha, V. K. Shah, R. Verma, R. Mandal, and S. Nandi, “Is it worth taking aplanned approach to design ad hoc infrastructure for post disaster communication?”in Proceedings of the seventh ACM international workshop on Challenged networks(CHANTS ’12). New York, NY, USA: ACM, 2012, pp. 87–90.

[40] T. Hossmann, F. Legendre, P. Carta, P. Gunningberg, and C. Rohner, “Twitterin disaster mode: opportunistic communication and distribution of sensor data inemergencies,” in Proceedings of the 3rd Extreme Conference on Communication:The Amazon Expedition, ser. ExtremeCom ’11. New York, NY, USA: ACM, 2011,pp. 1:1–1:6. [Online]. Available: http://doi.acm.org/10.1145/2414393.2414394

[41] F. Ekman, A. Keränen, J. Karvo, and J. Ott, “Working day movement model,”in Proceedings of the 1st ACM SIGMOBILE Workshop on Mobility Models. NewYork, NY, USA: ACM, 2008, pp. 33–40.

[42] R. Schmohl and U. Baumgarten, “Heterogeneity in mobile computing environmens,”in ICWN, Las Vegas, Nevada, USA, 2008, pp. 461–467.

[43] Y.-D. Bromberg, P. Grace, and L. Réveillère, “Starlink: Runtime interoperabilitybetween heterogeneous middleware protocols,” in Proceedings of the 31st Interna-tional Conference on Distributed Computing Systems (ICDCS), Minneapolis, Min-nesota, USA, Jun. 2011, pp. 446–455.

102

References

[44] P. Stuedi and G. Alonso, “Transparent heterogeneous mobile ad hoc networks,”in the 2nd Annual International Conference on Mobile and Ubiquitous Systems:Networking and Services., San Diego, California, USA, Jul. 2005, pp. 237 – 246.

[45] J. Scott, P. Hui, J. Crowcroft, and C. Diot, “Haggle: A Networking ArchitectureDesigned Around Mobile Users.” Invited paper at the 3rd Annual IFIP Conferenceon Wireless On-demand Network Systems and Services (WONS 2006), Jan. 2006.

[46] A. Petz, A. Bednarczyk, N. Paine, D. Stovall, and C. Julien, “Madman: A mid-dleware for delay-tolerant mobile ad-hoc networks,” University of Texas at Austin,Tech. Rep. TR-UTEDGE-2010-010, 2010.

[47] C.-H. Lee and D. Y. Eun, “Exploiting heterogeneity in mobile opportunistic net-works: An analytic approach,” in SECON, Boston, Massachusetts, USA, 2010, pp.502–510.

[48] Y. Tian and J. Li, “Heterogeneity of device contact process in pocket switched net-works,” in Proceedings of the 5th International Conference on Wireless Algorithms,Systems, and Applications (WASA’10). Berlin, Heidelberg: Springer-Verlag, 2010,pp. 157–166.

[49] Y. Li, P. Hui, D. Jin, L. Su, and L. Zeng, “Optimal distributed malware defense inmobile networks with heterogeneous devices,” IEEE Transactions on Mobile Com-puting, vol. 99, no. PrePrints, p. 1, Dec. 2012.

[50] V. Manam, V. Mahendran, and C. Murthy, “Performance modeling of routing indelay-tolerant networks with node heterogeneity,” in Proceedings of the 4th Interna-tional Conference on Communication Systems and Networks (COMSNETS), Ban-galore, India, Jan. 2012, pp. 1 –10.

[51] Y. Li, P. Hui, D. Jin, L. Su, and L. Zeng, “An optimal distributed malware defensesystem for mobile networks with heterogeneous devices,” in 8th Annual IEEE Com-munications Society Conference on Sensor, Mesh and Ad Hoc Communications andNetworks (SECON), Salt Lake City, Utah, USA, Jun. 2011, pp. 314 –322.

[52] N. A. Vien, N. H. Viet, S. Lee, and T. Chung, “Heuristic search based explorationin reinforcement learning,” in IWANN, San Sebastian, Spain, 2007, pp. 110–118.

[53] A. Keränen, J. Ott, and T. Kärkkäinen, “The ONE simulator for DTNprotocol evaluation,” in Proceedings of the 2nd International Conference onSimulation Tools and Techniques, ser. Simutools ’09. ICST, Brussels, Belgium,Belgium: ICST (Institute for Computer Sciences, Social-Informatics andTelecommunications Engineering), 2009, pp. 55:1–55:10. [Online]. Available:http://dx.doi.org/10.4108/ICST.SIMUTOOLS2009.5674

[54] R. Picard, Affective Computing. Cambridge: MIT Press, 1997.

103

References

[55] V. Venkatesh, M. G. Morris, G. B. Davis, and F. D. Davis, “User acceptance ofinformation technology: Toward a unified view,” MIS Quarterly, vol. 27, no. 3, pp.425–478, Sep. 2003.

[56] A. Beaudry and A. Pinsonneault, “The other side of acceptance: Studying the directand indirect effects of emotions on information technology use,” MIS Quarterly,vol. 34, no. 4, pp. 689–710, Dec. 2010.

[57] I. Rhee, M. Shin, S. Hong, K. Lee, S. Kim, and S. Chong, “CRAW-DAD data set ncsu/mobilitymodels (v. 2009-07-23),” Downloaded fromhttp://crawdad.cs.dartmouth.edu/ncsu/mobilitymodels, Jul. 2009, [Accessed: 12Dec. 2012].

[58] W. Zhao, Y. Chen, M. Ammar, M. Corner, B. Levine, and E. Zegura, “Capacityenhancement using throwboxes in DTNs,” in IEEE International Conference onMobile Adhoc and Sensor Systems (MASS), 2006, pp. 31–40.

[59] A. Casteigts, P. Flocchini, W. Quattrociocchi, and N. Santoro, “Time-varyinggraphs and dynamic networks,” International Journal of Parallel, Emergent andDistributed Systems, vol. 27, no. 5, pp. 387–408, 2012.

[60] J. Scott, R. Gass, J. Crowcroft, P. Hui, C. Diot, and A. Chaintreau, “CRAW-DAD trace cambridge/haggle/imote/infocom (v. 2006-01-31),” Downloaded fromhttp://crawdad.cs.dartmouth.edu/cambridge/haggle/imote/infocom, Jan. 2006.

[61] K. Massri, A. Vernata, and A. Vitaletti, “Routing protocols for delay tolerant net-works: a quantitative evaluation,” in Proceedings of the 7th ACM Workshop onPerformance Monitoring and Measurement of Heterogeneous Wireless and WiredNetworks. USA: ACM, 2012, pp. 107–114.

104

human aspects and heterogeneity in...

Documents